xref: /linux/drivers/scsi/mpt3sas/mpt3sas_base.c (revision a1d1eb2f57501b2e7e2076ce89b3f3a666ddbfdd)
1 /*
2  * This is the Fusion MPT base driver providing common API layer interface
3  * for access to MPT (Message Passing Technology) firmware.
4  *
5  * This code is based on drivers/scsi/mpt3sas/mpt3sas_base.c
6  * Copyright (C) 2012-2014  LSI Corporation
7  * Copyright (C) 2013-2014 Avago Technologies
8  *  (mailto: MPT-FusionLinux.pdl@avagotech.com)
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public License
12  * as published by the Free Software Foundation; either version 2
13  * of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
18  * GNU General Public License for more details.
19  *
20  * NO WARRANTY
21  * THE PROGRAM IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OR
22  * CONDITIONS OF ANY KIND, EITHER EXPRESS OR IMPLIED INCLUDING, WITHOUT
23  * LIMITATION, ANY WARRANTIES OR CONDITIONS OF TITLE, NON-INFRINGEMENT,
24  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE. Each Recipient is
25  * solely responsible for determining the appropriateness of using and
26  * distributing the Program and assumes all risks associated with its
27  * exercise of rights under this Agreement, including but not limited to
28  * the risks and costs of program errors, damage to or loss of data,
29  * programs or equipment, and unavailability or interruption of operations.
30 
31  * DISCLAIMER OF LIABILITY
32  * NEITHER RECIPIENT NOR ANY CONTRIBUTORS SHALL HAVE ANY LIABILITY FOR ANY
33  * DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
34  * DAMAGES (INCLUDING WITHOUT LIMITATION LOST PROFITS), HOWEVER CAUSED AND
35  * ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR
36  * TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE
37  * USE OR DISTRIBUTION OF THE PROGRAM OR THE EXERCISE OF ANY RIGHTS GRANTED
38  * HEREUNDER, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGES
39 
40  * You should have received a copy of the GNU General Public License
41  * along with this program; if not, write to the Free Software
42  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301,
43  * USA.
44  */
45 
46 #include <linux/kernel.h>
47 #include <linux/module.h>
48 #include <linux/errno.h>
49 #include <linux/init.h>
50 #include <linux/slab.h>
51 #include <linux/types.h>
52 #include <linux/pci.h>
53 #include <linux/kdev_t.h>
54 #include <linux/blkdev.h>
55 #include <linux/delay.h>
56 #include <linux/interrupt.h>
57 #include <linux/dma-mapping.h>
58 #include <linux/io.h>
59 #include <linux/time.h>
60 #include <linux/ktime.h>
61 #include <linux/kthread.h>
62 #include <asm/page.h>        /* To get host page size per arch */
63 
64 
65 #include "mpt3sas_base.h"
66 
67 static MPT_CALLBACK	mpt_callbacks[MPT_MAX_CALLBACKS];
68 
69 
70 #define FAULT_POLLING_INTERVAL 1000 /* in milliseconds */
71 
72  /* maximum controller queue depth */
73 #define MAX_HBA_QUEUE_DEPTH	30000
74 #define MAX_CHAIN_DEPTH		100000
75 static int max_queue_depth = -1;
76 module_param(max_queue_depth, int, 0444);
77 MODULE_PARM_DESC(max_queue_depth, " max controller queue depth ");
78 
79 static int max_sgl_entries = -1;
80 module_param(max_sgl_entries, int, 0444);
81 MODULE_PARM_DESC(max_sgl_entries, " max sg entries ");
82 
83 static int msix_disable = -1;
84 module_param(msix_disable, int, 0444);
85 MODULE_PARM_DESC(msix_disable, " disable msix routed interrupts (default=0)");
86 
87 static int smp_affinity_enable = 1;
88 module_param(smp_affinity_enable, int, 0444);
89 MODULE_PARM_DESC(smp_affinity_enable, "SMP affinity feature enable/disable Default: enable(1)");
90 
91 static int max_msix_vectors = -1;
92 module_param(max_msix_vectors, int, 0444);
93 MODULE_PARM_DESC(max_msix_vectors,
94 	" max msix vectors");
95 
96 static int irqpoll_weight = -1;
97 module_param(irqpoll_weight, int, 0444);
98 MODULE_PARM_DESC(irqpoll_weight,
99 	"irq poll weight (default= one fourth of HBA queue depth)");
100 
101 static int mpt3sas_fwfault_debug;
102 MODULE_PARM_DESC(mpt3sas_fwfault_debug,
103 	" enable detection of firmware fault and halt firmware - (default=0)");
104 
105 static int perf_mode = -1;
106 module_param(perf_mode, int, 0444);
107 MODULE_PARM_DESC(perf_mode,
108 	"Performance mode (only for Aero/Sea Generation), options:\n\t\t"
109 	"0 - balanced: high iops mode is enabled &\n\t\t"
110 	"interrupt coalescing is enabled only on high iops queues,\n\t\t"
111 	"1 - iops: high iops mode is disabled &\n\t\t"
112 	"interrupt coalescing is enabled on all queues,\n\t\t"
113 	"2 - latency: high iops mode is disabled &\n\t\t"
114 	"interrupt coalescing is enabled on all queues with timeout value 0xA,\n"
115 	"\t\tdefault - default perf_mode is 'balanced'"
116 	);
117 
118 static int poll_queues;
119 module_param(poll_queues, int, 0444);
120 MODULE_PARM_DESC(poll_queues, "Number of queues to be use for io_uring poll mode.\n\t\t"
121 	"This parameter is effective only if host_tagset_enable=1. &\n\t\t"
122 	"when poll_queues are enabled then &\n\t\t"
123 	"perf_mode is set to latency mode. &\n\t\t"
124 	);
125 
126 enum mpt3sas_perf_mode {
127 	MPT_PERF_MODE_DEFAULT	= -1,
128 	MPT_PERF_MODE_BALANCED	= 0,
129 	MPT_PERF_MODE_IOPS	= 1,
130 	MPT_PERF_MODE_LATENCY	= 2,
131 };
132 
133 static int
134 _base_wait_on_iocstate(struct MPT3SAS_ADAPTER *ioc,
135 		u32 ioc_state, int timeout);
136 static int
137 _base_get_ioc_facts(struct MPT3SAS_ADAPTER *ioc);
138 static void
139 _base_clear_outstanding_commands(struct MPT3SAS_ADAPTER *ioc);
140 
141 static u32
142 _base_readl_ext_retry(const void __iomem *addr);
143 
144 /**
145  * mpt3sas_base_check_cmd_timeout - Function
146  *		to check timeout and command termination due
147  *		to Host reset.
148  *
149  * @ioc:	per adapter object.
150  * @status:	Status of issued command.
151  * @mpi_request:mf request pointer.
152  * @sz:		size of buffer.
153  *
154  * Return: 1/0 Reset to be done or Not
155  */
156 u8
157 mpt3sas_base_check_cmd_timeout(struct MPT3SAS_ADAPTER *ioc,
158 		u8 status, void *mpi_request, int sz)
159 {
160 	u8 issue_reset = 0;
161 
162 	if (!(status & MPT3_CMD_RESET))
163 		issue_reset = 1;
164 
165 	ioc_err(ioc, "Command %s\n",
166 		issue_reset == 0 ? "terminated due to Host Reset" : "Timeout");
167 	_debug_dump_mf(mpi_request, sz);
168 
169 	return issue_reset;
170 }
171 
172 /**
173  * _scsih_set_fwfault_debug - global setting of ioc->fwfault_debug.
174  * @val: ?
175  * @kp: ?
176  *
177  * Return: ?
178  */
179 static int
180 _scsih_set_fwfault_debug(const char *val, const struct kernel_param *kp)
181 {
182 	int ret = param_set_int(val, kp);
183 	struct MPT3SAS_ADAPTER *ioc;
184 
185 	if (ret)
186 		return ret;
187 
188 	/* global ioc spinlock to protect controller list on list operations */
189 	pr_info("setting fwfault_debug(%d)\n", mpt3sas_fwfault_debug);
190 	spin_lock(&gioc_lock);
191 	list_for_each_entry(ioc, &mpt3sas_ioc_list, list)
192 		ioc->fwfault_debug = mpt3sas_fwfault_debug;
193 	spin_unlock(&gioc_lock);
194 	return 0;
195 }
196 module_param_call(mpt3sas_fwfault_debug, _scsih_set_fwfault_debug,
197 	param_get_int, &mpt3sas_fwfault_debug, 0644);
198 
199 /**
200  * _base_readl_aero - retry readl for max three times.
201  * @addr: MPT Fusion system interface register address
202  *
203  * Retry the readl() for max three times if it gets zero value
204  * while reading the system interface register.
205  */
206 static inline u32
207 _base_readl_aero(const void __iomem *addr)
208 {
209 	u32 i = 0, ret_val;
210 
211 	do {
212 		ret_val = readl(addr);
213 		i++;
214 	} while (ret_val == 0 && i < 3);
215 
216 	return ret_val;
217 }
218 
219 static u32
220 _base_readl_ext_retry(const void __iomem *addr)
221 {
222 	u32 i, ret_val;
223 
224 	for (i = 0 ; i < 30 ; i++) {
225 		ret_val = readl(addr);
226 		if (ret_val != 0)
227 			break;
228 	}
229 
230 	return ret_val;
231 }
232 
233 static inline u32
234 _base_readl(const void __iomem *addr)
235 {
236 	return readl(addr);
237 }
238 
239 /**
240  * _base_clone_reply_to_sys_mem - copies reply to reply free iomem
241  *				  in BAR0 space.
242  *
243  * @ioc: per adapter object
244  * @reply: reply message frame(lower 32bit addr)
245  * @index: System request message index.
246  */
247 static void
248 _base_clone_reply_to_sys_mem(struct MPT3SAS_ADAPTER *ioc, u32 reply,
249 		u32 index)
250 {
251 	/*
252 	 * 256 is offset within sys register.
253 	 * 256 offset MPI frame starts. Max MPI frame supported is 32.
254 	 * 32 * 128 = 4K. From here, Clone of reply free for mcpu starts
255 	 */
256 	u16 cmd_credit = ioc->facts.RequestCredit + 1;
257 	void __iomem *reply_free_iomem = (void __iomem *)ioc->chip +
258 			MPI_FRAME_START_OFFSET +
259 			(cmd_credit * ioc->request_sz) + (index * sizeof(u32));
260 
261 	writel(reply, reply_free_iomem);
262 }
263 
264 /**
265  * _base_clone_mpi_to_sys_mem - Writes/copies MPI frames
266  *				to system/BAR0 region.
267  *
268  * @dst_iomem: Pointer to the destination location in BAR0 space.
269  * @src: Pointer to the Source data.
270  * @size: Size of data to be copied.
271  */
272 static void
273 _base_clone_mpi_to_sys_mem(void *dst_iomem, void *src, u32 size)
274 {
275 	int i;
276 	u32 *src_virt_mem = (u32 *)src;
277 
278 	for (i = 0; i < size/4; i++)
279 		writel((u32)src_virt_mem[i],
280 				(void __iomem *)dst_iomem + (i * 4));
281 }
282 
283 /**
284  * _base_clone_to_sys_mem - Writes/copies data to system/BAR0 region
285  *
286  * @dst_iomem: Pointer to the destination location in BAR0 space.
287  * @src: Pointer to the Source data.
288  * @size: Size of data to be copied.
289  */
290 static void
291 _base_clone_to_sys_mem(void __iomem *dst_iomem, void *src, u32 size)
292 {
293 	int i;
294 	u32 *src_virt_mem = (u32 *)(src);
295 
296 	for (i = 0; i < size/4; i++)
297 		writel((u32)src_virt_mem[i],
298 			(void __iomem *)dst_iomem + (i * 4));
299 }
300 
301 /**
302  * _base_get_chain - Calculates and Returns virtual chain address
303  *			 for the provided smid in BAR0 space.
304  *
305  * @ioc: per adapter object
306  * @smid: system request message index
307  * @sge_chain_count: Scatter gather chain count.
308  *
309  * Return: the chain address.
310  */
311 static inline void __iomem*
312 _base_get_chain(struct MPT3SAS_ADAPTER *ioc, u16 smid,
313 		u8 sge_chain_count)
314 {
315 	void __iomem *base_chain, *chain_virt;
316 	u16 cmd_credit = ioc->facts.RequestCredit + 1;
317 
318 	base_chain  = (void __iomem *)ioc->chip + MPI_FRAME_START_OFFSET +
319 		(cmd_credit * ioc->request_sz) +
320 		REPLY_FREE_POOL_SIZE;
321 	chain_virt = base_chain + (smid * ioc->facts.MaxChainDepth *
322 			ioc->request_sz) + (sge_chain_count * ioc->request_sz);
323 	return chain_virt;
324 }
325 
326 /**
327  * _base_get_chain_phys - Calculates and Returns physical address
328  *			in BAR0 for scatter gather chains, for
329  *			the provided smid.
330  *
331  * @ioc: per adapter object
332  * @smid: system request message index
333  * @sge_chain_count: Scatter gather chain count.
334  *
335  * Return: Physical chain address.
336  */
337 static inline phys_addr_t
338 _base_get_chain_phys(struct MPT3SAS_ADAPTER *ioc, u16 smid,
339 		u8 sge_chain_count)
340 {
341 	phys_addr_t base_chain_phys, chain_phys;
342 	u16 cmd_credit = ioc->facts.RequestCredit + 1;
343 
344 	base_chain_phys  = ioc->chip_phys + MPI_FRAME_START_OFFSET +
345 		(cmd_credit * ioc->request_sz) +
346 		REPLY_FREE_POOL_SIZE;
347 	chain_phys = base_chain_phys + (smid * ioc->facts.MaxChainDepth *
348 			ioc->request_sz) + (sge_chain_count * ioc->request_sz);
349 	return chain_phys;
350 }
351 
352 /**
353  * _base_get_buffer_bar0 - Calculates and Returns BAR0 mapped Host
354  *			buffer address for the provided smid.
355  *			(Each smid can have 64K starts from 17024)
356  *
357  * @ioc: per adapter object
358  * @smid: system request message index
359  *
360  * Return: Pointer to buffer location in BAR0.
361  */
362 
363 static void __iomem *
364 _base_get_buffer_bar0(struct MPT3SAS_ADAPTER *ioc, u16 smid)
365 {
366 	u16 cmd_credit = ioc->facts.RequestCredit + 1;
367 	// Added extra 1 to reach end of chain.
368 	void __iomem *chain_end = _base_get_chain(ioc,
369 			cmd_credit + 1,
370 			ioc->facts.MaxChainDepth);
371 	return chain_end + (smid * 64 * 1024);
372 }
373 
374 /**
375  * _base_get_buffer_phys_bar0 - Calculates and Returns BAR0 mapped
376  *		Host buffer Physical address for the provided smid.
377  *		(Each smid can have 64K starts from 17024)
378  *
379  * @ioc: per adapter object
380  * @smid: system request message index
381  *
382  * Return: Pointer to buffer location in BAR0.
383  */
384 static phys_addr_t
385 _base_get_buffer_phys_bar0(struct MPT3SAS_ADAPTER *ioc, u16 smid)
386 {
387 	u16 cmd_credit = ioc->facts.RequestCredit + 1;
388 	phys_addr_t chain_end_phys = _base_get_chain_phys(ioc,
389 			cmd_credit + 1,
390 			ioc->facts.MaxChainDepth);
391 	return chain_end_phys + (smid * 64 * 1024);
392 }
393 
394 /**
395  * _base_get_chain_buffer_dma_to_chain_buffer - Iterates chain
396  *			lookup list and Provides chain_buffer
397  *			address for the matching dma address.
398  *			(Each smid can have 64K starts from 17024)
399  *
400  * @ioc: per adapter object
401  * @chain_buffer_dma: Chain buffer dma address.
402  *
403  * Return: Pointer to chain buffer. Or Null on Failure.
404  */
405 static void *
406 _base_get_chain_buffer_dma_to_chain_buffer(struct MPT3SAS_ADAPTER *ioc,
407 		dma_addr_t chain_buffer_dma)
408 {
409 	u16 index, j;
410 	struct chain_tracker *ct;
411 
412 	for (index = 0; index < ioc->scsiio_depth; index++) {
413 		for (j = 0; j < ioc->chains_needed_per_io; j++) {
414 			ct = &ioc->chain_lookup[index].chains_per_smid[j];
415 			if (ct && ct->chain_buffer_dma == chain_buffer_dma)
416 				return ct->chain_buffer;
417 		}
418 	}
419 	ioc_info(ioc, "Provided chain_buffer_dma address is not in the lookup list\n");
420 	return NULL;
421 }
422 
423 /**
424  * _clone_sg_entries -	MPI EP's scsiio and config requests
425  *			are handled here. Base function for
426  *			double buffering, before submitting
427  *			the requests.
428  *
429  * @ioc: per adapter object.
430  * @mpi_request: mf request pointer.
431  * @smid: system request message index.
432  */
433 static void _clone_sg_entries(struct MPT3SAS_ADAPTER *ioc,
434 		void *mpi_request, u16 smid)
435 {
436 	Mpi2SGESimple32_t *sgel, *sgel_next;
437 	u32  sgl_flags, sge_chain_count = 0;
438 	bool is_write = false;
439 	u16 i = 0;
440 	void __iomem *buffer_iomem;
441 	phys_addr_t buffer_iomem_phys;
442 	void __iomem *buff_ptr;
443 	phys_addr_t buff_ptr_phys;
444 	void __iomem *dst_chain_addr[MCPU_MAX_CHAINS_PER_IO];
445 	void *src_chain_addr[MCPU_MAX_CHAINS_PER_IO];
446 	phys_addr_t dst_addr_phys;
447 	MPI2RequestHeader_t *request_hdr;
448 	struct scsi_cmnd *scmd;
449 	struct scatterlist *sg_scmd = NULL;
450 	int is_scsiio_req = 0;
451 
452 	request_hdr = (MPI2RequestHeader_t *) mpi_request;
453 
454 	if (request_hdr->Function == MPI2_FUNCTION_SCSI_IO_REQUEST) {
455 		Mpi25SCSIIORequest_t *scsiio_request =
456 			(Mpi25SCSIIORequest_t *)mpi_request;
457 		sgel = (Mpi2SGESimple32_t *) &scsiio_request->SGL;
458 		is_scsiio_req = 1;
459 	} else if (request_hdr->Function == MPI2_FUNCTION_CONFIG) {
460 		Mpi2ConfigRequest_t  *config_req =
461 			(Mpi2ConfigRequest_t *)mpi_request;
462 		sgel = (Mpi2SGESimple32_t *) &config_req->PageBufferSGE;
463 	} else
464 		return;
465 
466 	/* From smid we can get scsi_cmd, once we have sg_scmd,
467 	 * we just need to get sg_virt and sg_next to get virtual
468 	 * address associated with sgel->Address.
469 	 */
470 
471 	if (is_scsiio_req) {
472 		/* Get scsi_cmd using smid */
473 		scmd = mpt3sas_scsih_scsi_lookup_get(ioc, smid);
474 		if (scmd == NULL) {
475 			ioc_err(ioc, "scmd is NULL\n");
476 			return;
477 		}
478 
479 		/* Get sg_scmd from scmd provided */
480 		sg_scmd = scsi_sglist(scmd);
481 	}
482 
483 	/*
484 	 * 0 - 255	System register
485 	 * 256 - 4352	MPI Frame. (This is based on maxCredit 32)
486 	 * 4352 - 4864	Reply_free pool (512 byte is reserved
487 	 *		considering maxCredit 32. Reply need extra
488 	 *		room, for mCPU case kept four times of
489 	 *		maxCredit).
490 	 * 4864 - 17152	SGE chain element. (32cmd * 3 chain of
491 	 *		128 byte size = 12288)
492 	 * 17152 - x	Host buffer mapped with smid.
493 	 *		(Each smid can have 64K Max IO.)
494 	 * BAR0+Last 1K MSIX Addr and Data
495 	 * Total size in use 2113664 bytes of 4MB BAR0
496 	 */
497 
498 	buffer_iomem = _base_get_buffer_bar0(ioc, smid);
499 	buffer_iomem_phys = _base_get_buffer_phys_bar0(ioc, smid);
500 
501 	buff_ptr = buffer_iomem;
502 	buff_ptr_phys = buffer_iomem_phys;
503 	WARN_ON(buff_ptr_phys > U32_MAX);
504 
505 	if (le32_to_cpu(sgel->FlagsLength) &
506 			(MPI2_SGE_FLAGS_HOST_TO_IOC << MPI2_SGE_FLAGS_SHIFT))
507 		is_write = true;
508 
509 	for (i = 0; i < MPT_MIN_PHYS_SEGMENTS + ioc->facts.MaxChainDepth; i++) {
510 
511 		sgl_flags =
512 		    (le32_to_cpu(sgel->FlagsLength) >> MPI2_SGE_FLAGS_SHIFT);
513 
514 		switch (sgl_flags & MPI2_SGE_FLAGS_ELEMENT_MASK) {
515 		case MPI2_SGE_FLAGS_CHAIN_ELEMENT:
516 			/*
517 			 * Helper function which on passing
518 			 * chain_buffer_dma returns chain_buffer. Get
519 			 * the virtual address for sgel->Address
520 			 */
521 			sgel_next =
522 				_base_get_chain_buffer_dma_to_chain_buffer(ioc,
523 						le32_to_cpu(sgel->Address));
524 			if (sgel_next == NULL)
525 				return;
526 			/*
527 			 * This is coping 128 byte chain
528 			 * frame (not a host buffer)
529 			 */
530 			dst_chain_addr[sge_chain_count] =
531 				_base_get_chain(ioc,
532 					smid, sge_chain_count);
533 			src_chain_addr[sge_chain_count] =
534 						(void *) sgel_next;
535 			dst_addr_phys = _base_get_chain_phys(ioc,
536 						smid, sge_chain_count);
537 			WARN_ON(dst_addr_phys > U32_MAX);
538 			sgel->Address =
539 				cpu_to_le32(lower_32_bits(dst_addr_phys));
540 			sgel = sgel_next;
541 			sge_chain_count++;
542 			break;
543 		case MPI2_SGE_FLAGS_SIMPLE_ELEMENT:
544 			if (is_write) {
545 				if (is_scsiio_req) {
546 					_base_clone_to_sys_mem(buff_ptr,
547 					    sg_virt(sg_scmd),
548 					    (le32_to_cpu(sgel->FlagsLength) &
549 					    0x00ffffff));
550 					/*
551 					 * FIXME: this relies on a a zero
552 					 * PCI mem_offset.
553 					 */
554 					sgel->Address =
555 					    cpu_to_le32((u32)buff_ptr_phys);
556 				} else {
557 					_base_clone_to_sys_mem(buff_ptr,
558 					    ioc->config_vaddr,
559 					    (le32_to_cpu(sgel->FlagsLength) &
560 					    0x00ffffff));
561 					sgel->Address =
562 					    cpu_to_le32((u32)buff_ptr_phys);
563 				}
564 			}
565 			buff_ptr += (le32_to_cpu(sgel->FlagsLength) &
566 			    0x00ffffff);
567 			buff_ptr_phys += (le32_to_cpu(sgel->FlagsLength) &
568 			    0x00ffffff);
569 			if ((le32_to_cpu(sgel->FlagsLength) &
570 			    (MPI2_SGE_FLAGS_END_OF_BUFFER
571 					<< MPI2_SGE_FLAGS_SHIFT)))
572 				goto eob_clone_chain;
573 			else {
574 				/*
575 				 * Every single element in MPT will have
576 				 * associated sg_next. Better to sanity that
577 				 * sg_next is not NULL, but it will be a bug
578 				 * if it is null.
579 				 */
580 				if (is_scsiio_req) {
581 					sg_scmd = sg_next(sg_scmd);
582 					if (sg_scmd)
583 						sgel++;
584 					else
585 						goto eob_clone_chain;
586 				}
587 			}
588 			break;
589 		}
590 	}
591 
592 eob_clone_chain:
593 	for (i = 0; i < sge_chain_count; i++) {
594 		if (is_scsiio_req)
595 			_base_clone_to_sys_mem(dst_chain_addr[i],
596 				src_chain_addr[i], ioc->request_sz);
597 	}
598 }
599 
600 /**
601  *  mpt3sas_remove_dead_ioc_func - kthread context to remove dead ioc
602  * @arg: input argument, used to derive ioc
603  *
604  * Return:
605  * 0 if controller is removed from pci subsystem.
606  * -1 for other case.
607  */
608 static int mpt3sas_remove_dead_ioc_func(void *arg)
609 {
610 	struct MPT3SAS_ADAPTER *ioc = (struct MPT3SAS_ADAPTER *)arg;
611 	struct pci_dev *pdev;
612 
613 	if (!ioc)
614 		return -1;
615 
616 	pdev = ioc->pdev;
617 	if (!pdev)
618 		return -1;
619 	pci_stop_and_remove_bus_device_locked(pdev);
620 	return 0;
621 }
622 
623 /**
624  * _base_sync_drv_fw_timestamp - Sync Drive-Fw TimeStamp.
625  * @ioc: Per Adapter Object
626  *
627  * Return: nothing.
628  */
629 static void _base_sync_drv_fw_timestamp(struct MPT3SAS_ADAPTER *ioc)
630 {
631 	Mpi26IoUnitControlRequest_t *mpi_request;
632 	Mpi26IoUnitControlReply_t *mpi_reply;
633 	u16 smid;
634 	ktime_t current_time;
635 	u64 TimeStamp = 0;
636 	u8 issue_reset = 0;
637 
638 	mutex_lock(&ioc->scsih_cmds.mutex);
639 	if (ioc->scsih_cmds.status != MPT3_CMD_NOT_USED) {
640 		ioc_err(ioc, "scsih_cmd in use %s\n", __func__);
641 		goto out;
642 	}
643 	ioc->scsih_cmds.status = MPT3_CMD_PENDING;
644 	smid = mpt3sas_base_get_smid(ioc, ioc->scsih_cb_idx);
645 	if (!smid) {
646 		ioc_err(ioc, "Failed obtaining a smid %s\n", __func__);
647 		ioc->scsih_cmds.status = MPT3_CMD_NOT_USED;
648 		goto out;
649 	}
650 	mpi_request = mpt3sas_base_get_msg_frame(ioc, smid);
651 	ioc->scsih_cmds.smid = smid;
652 	memset(mpi_request, 0, sizeof(Mpi26IoUnitControlRequest_t));
653 	mpi_request->Function = MPI2_FUNCTION_IO_UNIT_CONTROL;
654 	mpi_request->Operation = MPI26_CTRL_OP_SET_IOC_PARAMETER;
655 	mpi_request->IOCParameter = MPI26_SET_IOC_PARAMETER_SYNC_TIMESTAMP;
656 	current_time = ktime_get_real();
657 	TimeStamp = ktime_to_ms(current_time);
658 	mpi_request->Reserved7 = cpu_to_le32(TimeStamp >> 32);
659 	mpi_request->IOCParameterValue = cpu_to_le32(TimeStamp & 0xFFFFFFFF);
660 	init_completion(&ioc->scsih_cmds.done);
661 	ioc->put_smid_default(ioc, smid);
662 	dinitprintk(ioc, ioc_info(ioc,
663 	    "Io Unit Control Sync TimeStamp (sending), @time %lld ms\n",
664 	    TimeStamp));
665 	wait_for_completion_timeout(&ioc->scsih_cmds.done,
666 		MPT3SAS_TIMESYNC_TIMEOUT_SECONDS*HZ);
667 	if (!(ioc->scsih_cmds.status & MPT3_CMD_COMPLETE)) {
668 		mpt3sas_check_cmd_timeout(ioc,
669 		    ioc->scsih_cmds.status, mpi_request,
670 		    sizeof(Mpi2SasIoUnitControlRequest_t)/4, issue_reset);
671 		goto issue_host_reset;
672 	}
673 	if (ioc->scsih_cmds.status & MPT3_CMD_REPLY_VALID) {
674 		mpi_reply = ioc->scsih_cmds.reply;
675 		dinitprintk(ioc, ioc_info(ioc,
676 		    "Io Unit Control sync timestamp (complete): ioc_status(0x%04x), loginfo(0x%08x)\n",
677 		    le16_to_cpu(mpi_reply->IOCStatus),
678 		    le32_to_cpu(mpi_reply->IOCLogInfo)));
679 	}
680 issue_host_reset:
681 	if (issue_reset)
682 		mpt3sas_base_hard_reset_handler(ioc, FORCE_BIG_HAMMER);
683 	ioc->scsih_cmds.status = MPT3_CMD_NOT_USED;
684 out:
685 	mutex_unlock(&ioc->scsih_cmds.mutex);
686 }
687 
688 /**
689  * _base_fault_reset_work - workq handling ioc fault conditions
690  * @work: input argument, used to derive ioc
691  *
692  * Context: sleep.
693  */
694 static void
695 _base_fault_reset_work(struct work_struct *work)
696 {
697 	struct MPT3SAS_ADAPTER *ioc =
698 	    container_of(work, struct MPT3SAS_ADAPTER, fault_reset_work.work);
699 	unsigned long	 flags;
700 	u32 doorbell;
701 	int rc;
702 	struct task_struct *p;
703 
704 
705 	spin_lock_irqsave(&ioc->ioc_reset_in_progress_lock, flags);
706 	if ((ioc->shost_recovery && (ioc->ioc_coredump_loop == 0)) ||
707 			ioc->pci_error_recovery)
708 		goto rearm_timer;
709 	spin_unlock_irqrestore(&ioc->ioc_reset_in_progress_lock, flags);
710 
711 	doorbell = mpt3sas_base_get_iocstate(ioc, 0);
712 	if ((doorbell & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_MASK) {
713 		ioc_err(ioc, "SAS host is non-operational !!!!\n");
714 
715 		/* It may be possible that EEH recovery can resolve some of
716 		 * pci bus failure issues rather removing the dead ioc function
717 		 * by considering controller is in a non-operational state. So
718 		 * here priority is given to the EEH recovery. If it doesn't
719 		 * not resolve this issue, mpt3sas driver will consider this
720 		 * controller to non-operational state and remove the dead ioc
721 		 * function.
722 		 */
723 		if (ioc->non_operational_loop++ < 5) {
724 			spin_lock_irqsave(&ioc->ioc_reset_in_progress_lock,
725 							 flags);
726 			goto rearm_timer;
727 		}
728 
729 		/*
730 		 * Call _scsih_flush_pending_cmds callback so that we flush all
731 		 * pending commands back to OS. This call is required to avoid
732 		 * deadlock at block layer. Dead IOC will fail to do diag reset,
733 		 * and this call is safe since dead ioc will never return any
734 		 * command back from HW.
735 		 */
736 		mpt3sas_base_pause_mq_polling(ioc);
737 		ioc->schedule_dead_ioc_flush_running_cmds(ioc);
738 		/*
739 		 * Set remove_host flag early since kernel thread will
740 		 * take some time to execute.
741 		 */
742 		ioc->remove_host = 1;
743 		/*Remove the Dead Host */
744 		p = kthread_run(mpt3sas_remove_dead_ioc_func, ioc,
745 		    "%s_dead_ioc_%d", ioc->driver_name, ioc->id);
746 		if (IS_ERR(p))
747 			ioc_err(ioc, "%s: Running mpt3sas_dead_ioc thread failed !!!!\n",
748 				__func__);
749 		else
750 			ioc_err(ioc, "%s: Running mpt3sas_dead_ioc thread success !!!!\n",
751 				__func__);
752 		return; /* don't rearm timer */
753 	}
754 
755 	if ((doorbell & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_COREDUMP) {
756 		u8 timeout = (ioc->manu_pg11.CoreDumpTOSec) ?
757 		    ioc->manu_pg11.CoreDumpTOSec :
758 		    MPT3SAS_DEFAULT_COREDUMP_TIMEOUT_SECONDS;
759 
760 		timeout /= (FAULT_POLLING_INTERVAL/1000);
761 
762 		if (ioc->ioc_coredump_loop == 0) {
763 			mpt3sas_print_coredump_info(ioc,
764 			    doorbell & MPI2_DOORBELL_DATA_MASK);
765 			/* do not accept any IOs and disable the interrupts */
766 			spin_lock_irqsave(
767 			    &ioc->ioc_reset_in_progress_lock, flags);
768 			ioc->shost_recovery = 1;
769 			spin_unlock_irqrestore(
770 			    &ioc->ioc_reset_in_progress_lock, flags);
771 			mpt3sas_base_mask_interrupts(ioc);
772 			mpt3sas_base_pause_mq_polling(ioc);
773 			_base_clear_outstanding_commands(ioc);
774 		}
775 
776 		ioc_info(ioc, "%s: CoreDump loop %d.",
777 		    __func__, ioc->ioc_coredump_loop);
778 
779 		/* Wait until CoreDump completes or times out */
780 		if (ioc->ioc_coredump_loop++ < timeout) {
781 			spin_lock_irqsave(
782 			    &ioc->ioc_reset_in_progress_lock, flags);
783 			goto rearm_timer;
784 		}
785 	}
786 
787 	if (ioc->ioc_coredump_loop) {
788 		if ((doorbell & MPI2_IOC_STATE_MASK) != MPI2_IOC_STATE_COREDUMP)
789 			ioc_err(ioc, "%s: CoreDump completed. LoopCount: %d",
790 			    __func__, ioc->ioc_coredump_loop);
791 		else
792 			ioc_err(ioc, "%s: CoreDump Timed out. LoopCount: %d",
793 			    __func__, ioc->ioc_coredump_loop);
794 		ioc->ioc_coredump_loop = MPT3SAS_COREDUMP_LOOP_DONE;
795 	}
796 	ioc->non_operational_loop = 0;
797 	if ((doorbell & MPI2_IOC_STATE_MASK) != MPI2_IOC_STATE_OPERATIONAL) {
798 		rc = mpt3sas_base_hard_reset_handler(ioc, FORCE_BIG_HAMMER);
799 		ioc_warn(ioc, "%s: hard reset: %s\n",
800 			 __func__, rc == 0 ? "success" : "failed");
801 		doorbell = mpt3sas_base_get_iocstate(ioc, 0);
802 		if ((doorbell & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) {
803 			mpt3sas_print_fault_code(ioc, doorbell &
804 			    MPI2_DOORBELL_DATA_MASK);
805 		} else if ((doorbell & MPI2_IOC_STATE_MASK) ==
806 		    MPI2_IOC_STATE_COREDUMP)
807 			mpt3sas_print_coredump_info(ioc, doorbell &
808 			    MPI2_DOORBELL_DATA_MASK);
809 		if (rc && (doorbell & MPI2_IOC_STATE_MASK) !=
810 		    MPI2_IOC_STATE_OPERATIONAL)
811 			return; /* don't rearm timer */
812 	}
813 	ioc->ioc_coredump_loop = 0;
814 	if (ioc->time_sync_interval &&
815 	    ++ioc->timestamp_update_count >= ioc->time_sync_interval) {
816 		ioc->timestamp_update_count = 0;
817 		_base_sync_drv_fw_timestamp(ioc);
818 	}
819 	spin_lock_irqsave(&ioc->ioc_reset_in_progress_lock, flags);
820  rearm_timer:
821 	if (ioc->fault_reset_work_q)
822 		queue_delayed_work(ioc->fault_reset_work_q,
823 		    &ioc->fault_reset_work,
824 		    msecs_to_jiffies(FAULT_POLLING_INTERVAL));
825 	spin_unlock_irqrestore(&ioc->ioc_reset_in_progress_lock, flags);
826 }
827 
828 /**
829  * mpt3sas_base_start_watchdog - start the fault_reset_work_q
830  * @ioc: per adapter object
831  *
832  * Context: sleep.
833  */
834 void
835 mpt3sas_base_start_watchdog(struct MPT3SAS_ADAPTER *ioc)
836 {
837 	unsigned long	 flags;
838 
839 	if (ioc->fault_reset_work_q)
840 		return;
841 
842 	ioc->timestamp_update_count = 0;
843 	/* initialize fault polling */
844 
845 	INIT_DELAYED_WORK(&ioc->fault_reset_work, _base_fault_reset_work);
846 	snprintf(ioc->fault_reset_work_q_name,
847 	    sizeof(ioc->fault_reset_work_q_name), "poll_%s%d_status",
848 	    ioc->driver_name, ioc->id);
849 	ioc->fault_reset_work_q = alloc_ordered_workqueue(
850 		"%s", WQ_MEM_RECLAIM, ioc->fault_reset_work_q_name);
851 	if (!ioc->fault_reset_work_q) {
852 		ioc_err(ioc, "%s: failed (line=%d)\n", __func__, __LINE__);
853 		return;
854 	}
855 	spin_lock_irqsave(&ioc->ioc_reset_in_progress_lock, flags);
856 	if (ioc->fault_reset_work_q)
857 		queue_delayed_work(ioc->fault_reset_work_q,
858 		    &ioc->fault_reset_work,
859 		    msecs_to_jiffies(FAULT_POLLING_INTERVAL));
860 	spin_unlock_irqrestore(&ioc->ioc_reset_in_progress_lock, flags);
861 }
862 
863 /**
864  * mpt3sas_base_stop_watchdog - stop the fault_reset_work_q
865  * @ioc: per adapter object
866  *
867  * Context: sleep.
868  */
869 void
870 mpt3sas_base_stop_watchdog(struct MPT3SAS_ADAPTER *ioc)
871 {
872 	unsigned long flags;
873 	struct workqueue_struct *wq;
874 
875 	spin_lock_irqsave(&ioc->ioc_reset_in_progress_lock, flags);
876 	wq = ioc->fault_reset_work_q;
877 	ioc->fault_reset_work_q = NULL;
878 	spin_unlock_irqrestore(&ioc->ioc_reset_in_progress_lock, flags);
879 	if (wq) {
880 		if (!cancel_delayed_work_sync(&ioc->fault_reset_work))
881 			flush_workqueue(wq);
882 		destroy_workqueue(wq);
883 	}
884 }
885 
886 /**
887  * mpt3sas_base_fault_info - verbose translation of firmware FAULT code
888  * @ioc: per adapter object
889  * @fault_code: fault code
890  */
891 void
892 mpt3sas_base_fault_info(struct MPT3SAS_ADAPTER *ioc, u16 fault_code)
893 {
894 	ioc_err(ioc, "fault_state(0x%04x)!\n", fault_code);
895 }
896 
897 /**
898  * mpt3sas_base_coredump_info - verbose translation of firmware CoreDump state
899  * @ioc: per adapter object
900  * @fault_code: fault code
901  *
902  * Return: nothing.
903  */
904 void
905 mpt3sas_base_coredump_info(struct MPT3SAS_ADAPTER *ioc, u16 fault_code)
906 {
907 	ioc_err(ioc, "coredump_state(0x%04x)!\n", fault_code);
908 }
909 
910 /**
911  * mpt3sas_base_wait_for_coredump_completion - Wait until coredump
912  * completes or times out
913  * @ioc: per adapter object
914  * @caller: caller function name
915  *
916  * Return: 0 for success, non-zero for failure.
917  */
918 int
919 mpt3sas_base_wait_for_coredump_completion(struct MPT3SAS_ADAPTER *ioc,
920 		const char *caller)
921 {
922 	u8 timeout = (ioc->manu_pg11.CoreDumpTOSec) ?
923 			ioc->manu_pg11.CoreDumpTOSec :
924 			MPT3SAS_DEFAULT_COREDUMP_TIMEOUT_SECONDS;
925 
926 	int ioc_state = _base_wait_on_iocstate(ioc, MPI2_IOC_STATE_FAULT,
927 					timeout);
928 
929 	if (ioc_state)
930 		ioc_err(ioc,
931 		    "%s: CoreDump timed out. (ioc_state=0x%x)\n",
932 		    caller, ioc_state);
933 	else
934 		ioc_info(ioc,
935 		    "%s: CoreDump completed. (ioc_state=0x%x)\n",
936 		    caller, ioc_state);
937 
938 	return ioc_state;
939 }
940 
941 /**
942  * mpt3sas_halt_firmware - halt's mpt controller firmware
943  * @ioc: per adapter object
944  *
945  * For debugging timeout related issues.  Writing 0xCOFFEE00
946  * to the doorbell register will halt controller firmware. With
947  * the purpose to stop both driver and firmware, the enduser can
948  * obtain a ring buffer from controller UART.
949  */
950 void
951 mpt3sas_halt_firmware(struct MPT3SAS_ADAPTER *ioc)
952 {
953 	u32 doorbell;
954 
955 	if (!ioc->fwfault_debug)
956 		return;
957 
958 	dump_stack();
959 
960 	doorbell = ioc->base_readl_ext_retry(&ioc->chip->Doorbell);
961 	if ((doorbell & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) {
962 		mpt3sas_print_fault_code(ioc, doorbell &
963 		    MPI2_DOORBELL_DATA_MASK);
964 	} else if ((doorbell & MPI2_IOC_STATE_MASK) ==
965 	    MPI2_IOC_STATE_COREDUMP) {
966 		mpt3sas_print_coredump_info(ioc, doorbell &
967 		    MPI2_DOORBELL_DATA_MASK);
968 	} else {
969 		writel(0xC0FFEE00, &ioc->chip->Doorbell);
970 		ioc_err(ioc, "Firmware is halted due to command timeout\n");
971 	}
972 
973 	if (ioc->fwfault_debug == 2)
974 		for (;;)
975 			;
976 	else
977 		panic("panic in %s\n", __func__);
978 }
979 
980 /**
981  * _base_sas_ioc_info - verbose translation of the ioc status
982  * @ioc: per adapter object
983  * @mpi_reply: reply mf payload returned from firmware
984  * @request_hdr: request mf
985  */
986 static void
987 _base_sas_ioc_info(struct MPT3SAS_ADAPTER *ioc, MPI2DefaultReply_t *mpi_reply,
988 	MPI2RequestHeader_t *request_hdr)
989 {
990 	u16 ioc_status = le16_to_cpu(mpi_reply->IOCStatus) &
991 	    MPI2_IOCSTATUS_MASK;
992 	char *desc = NULL;
993 	u16 frame_sz;
994 	char *func_str = NULL;
995 
996 	/* SCSI_IO, RAID_PASS are handled from _scsih_scsi_ioc_info */
997 	if (request_hdr->Function == MPI2_FUNCTION_SCSI_IO_REQUEST ||
998 	    request_hdr->Function == MPI2_FUNCTION_RAID_SCSI_IO_PASSTHROUGH ||
999 	    request_hdr->Function == MPI2_FUNCTION_EVENT_NOTIFICATION)
1000 		return;
1001 
1002 	if (ioc_status == MPI2_IOCSTATUS_CONFIG_INVALID_PAGE)
1003 		return;
1004 	/*
1005 	 * Older Firmware version doesn't support driver trigger pages.
1006 	 * So, skip displaying 'config invalid type' type
1007 	 * of error message.
1008 	 */
1009 	if (request_hdr->Function == MPI2_FUNCTION_CONFIG) {
1010 		Mpi2ConfigRequest_t *rqst = (Mpi2ConfigRequest_t *)request_hdr;
1011 
1012 		if ((rqst->ExtPageType ==
1013 		    MPI2_CONFIG_EXTPAGETYPE_DRIVER_PERSISTENT_TRIGGER) &&
1014 		    !(ioc->logging_level & MPT_DEBUG_CONFIG)) {
1015 			return;
1016 		}
1017 	}
1018 
1019 	switch (ioc_status) {
1020 
1021 /****************************************************************************
1022 *  Common IOCStatus values for all replies
1023 ****************************************************************************/
1024 
1025 	case MPI2_IOCSTATUS_INVALID_FUNCTION:
1026 		desc = "invalid function";
1027 		break;
1028 	case MPI2_IOCSTATUS_BUSY:
1029 		desc = "busy";
1030 		break;
1031 	case MPI2_IOCSTATUS_INVALID_SGL:
1032 		desc = "invalid sgl";
1033 		break;
1034 	case MPI2_IOCSTATUS_INTERNAL_ERROR:
1035 		desc = "internal error";
1036 		break;
1037 	case MPI2_IOCSTATUS_INVALID_VPID:
1038 		desc = "invalid vpid";
1039 		break;
1040 	case MPI2_IOCSTATUS_INSUFFICIENT_RESOURCES:
1041 		desc = "insufficient resources";
1042 		break;
1043 	case MPI2_IOCSTATUS_INSUFFICIENT_POWER:
1044 		desc = "insufficient power";
1045 		break;
1046 	case MPI2_IOCSTATUS_INVALID_FIELD:
1047 		desc = "invalid field";
1048 		break;
1049 	case MPI2_IOCSTATUS_INVALID_STATE:
1050 		desc = "invalid state";
1051 		break;
1052 	case MPI2_IOCSTATUS_OP_STATE_NOT_SUPPORTED:
1053 		desc = "op state not supported";
1054 		break;
1055 
1056 /****************************************************************************
1057 *  Config IOCStatus values
1058 ****************************************************************************/
1059 
1060 	case MPI2_IOCSTATUS_CONFIG_INVALID_ACTION:
1061 		desc = "config invalid action";
1062 		break;
1063 	case MPI2_IOCSTATUS_CONFIG_INVALID_TYPE:
1064 		desc = "config invalid type";
1065 		break;
1066 	case MPI2_IOCSTATUS_CONFIG_INVALID_PAGE:
1067 		desc = "config invalid page";
1068 		break;
1069 	case MPI2_IOCSTATUS_CONFIG_INVALID_DATA:
1070 		desc = "config invalid data";
1071 		break;
1072 	case MPI2_IOCSTATUS_CONFIG_NO_DEFAULTS:
1073 		desc = "config no defaults";
1074 		break;
1075 	case MPI2_IOCSTATUS_CONFIG_CANT_COMMIT:
1076 		desc = "config can't commit";
1077 		break;
1078 
1079 /****************************************************************************
1080 *  SCSI IO Reply
1081 ****************************************************************************/
1082 
1083 	case MPI2_IOCSTATUS_SCSI_RECOVERED_ERROR:
1084 	case MPI2_IOCSTATUS_SCSI_INVALID_DEVHANDLE:
1085 	case MPI2_IOCSTATUS_SCSI_DEVICE_NOT_THERE:
1086 	case MPI2_IOCSTATUS_SCSI_DATA_OVERRUN:
1087 	case MPI2_IOCSTATUS_SCSI_DATA_UNDERRUN:
1088 	case MPI2_IOCSTATUS_SCSI_IO_DATA_ERROR:
1089 	case MPI2_IOCSTATUS_SCSI_PROTOCOL_ERROR:
1090 	case MPI2_IOCSTATUS_SCSI_TASK_TERMINATED:
1091 	case MPI2_IOCSTATUS_SCSI_RESIDUAL_MISMATCH:
1092 	case MPI2_IOCSTATUS_SCSI_TASK_MGMT_FAILED:
1093 	case MPI2_IOCSTATUS_SCSI_IOC_TERMINATED:
1094 	case MPI2_IOCSTATUS_SCSI_EXT_TERMINATED:
1095 		break;
1096 
1097 /****************************************************************************
1098 *  For use by SCSI Initiator and SCSI Target end-to-end data protection
1099 ****************************************************************************/
1100 
1101 	case MPI2_IOCSTATUS_EEDP_GUARD_ERROR:
1102 		desc = "eedp guard error";
1103 		break;
1104 	case MPI2_IOCSTATUS_EEDP_REF_TAG_ERROR:
1105 		desc = "eedp ref tag error";
1106 		break;
1107 	case MPI2_IOCSTATUS_EEDP_APP_TAG_ERROR:
1108 		desc = "eedp app tag error";
1109 		break;
1110 
1111 /****************************************************************************
1112 *  SCSI Target values
1113 ****************************************************************************/
1114 
1115 	case MPI2_IOCSTATUS_TARGET_INVALID_IO_INDEX:
1116 		desc = "target invalid io index";
1117 		break;
1118 	case MPI2_IOCSTATUS_TARGET_ABORTED:
1119 		desc = "target aborted";
1120 		break;
1121 	case MPI2_IOCSTATUS_TARGET_NO_CONN_RETRYABLE:
1122 		desc = "target no conn retryable";
1123 		break;
1124 	case MPI2_IOCSTATUS_TARGET_NO_CONNECTION:
1125 		desc = "target no connection";
1126 		break;
1127 	case MPI2_IOCSTATUS_TARGET_XFER_COUNT_MISMATCH:
1128 		desc = "target xfer count mismatch";
1129 		break;
1130 	case MPI2_IOCSTATUS_TARGET_DATA_OFFSET_ERROR:
1131 		desc = "target data offset error";
1132 		break;
1133 	case MPI2_IOCSTATUS_TARGET_TOO_MUCH_WRITE_DATA:
1134 		desc = "target too much write data";
1135 		break;
1136 	case MPI2_IOCSTATUS_TARGET_IU_TOO_SHORT:
1137 		desc = "target iu too short";
1138 		break;
1139 	case MPI2_IOCSTATUS_TARGET_ACK_NAK_TIMEOUT:
1140 		desc = "target ack nak timeout";
1141 		break;
1142 	case MPI2_IOCSTATUS_TARGET_NAK_RECEIVED:
1143 		desc = "target nak received";
1144 		break;
1145 
1146 /****************************************************************************
1147 *  Serial Attached SCSI values
1148 ****************************************************************************/
1149 
1150 	case MPI2_IOCSTATUS_SAS_SMP_REQUEST_FAILED:
1151 		desc = "smp request failed";
1152 		break;
1153 	case MPI2_IOCSTATUS_SAS_SMP_DATA_OVERRUN:
1154 		desc = "smp data overrun";
1155 		break;
1156 
1157 /****************************************************************************
1158 *  Diagnostic Buffer Post / Diagnostic Release values
1159 ****************************************************************************/
1160 
1161 	case MPI2_IOCSTATUS_DIAGNOSTIC_RELEASED:
1162 		desc = "diagnostic released";
1163 		break;
1164 	default:
1165 		break;
1166 	}
1167 
1168 	if (!desc)
1169 		return;
1170 
1171 	switch (request_hdr->Function) {
1172 	case MPI2_FUNCTION_CONFIG:
1173 		frame_sz = sizeof(Mpi2ConfigRequest_t) + ioc->sge_size;
1174 		func_str = "config_page";
1175 		break;
1176 	case MPI2_FUNCTION_SCSI_TASK_MGMT:
1177 		frame_sz = sizeof(Mpi2SCSITaskManagementRequest_t);
1178 		func_str = "task_mgmt";
1179 		break;
1180 	case MPI2_FUNCTION_SAS_IO_UNIT_CONTROL:
1181 		frame_sz = sizeof(Mpi2SasIoUnitControlRequest_t);
1182 		func_str = "sas_iounit_ctl";
1183 		break;
1184 	case MPI2_FUNCTION_SCSI_ENCLOSURE_PROCESSOR:
1185 		frame_sz = sizeof(Mpi2SepRequest_t);
1186 		func_str = "enclosure";
1187 		break;
1188 	case MPI2_FUNCTION_IOC_INIT:
1189 		frame_sz = sizeof(Mpi2IOCInitRequest_t);
1190 		func_str = "ioc_init";
1191 		break;
1192 	case MPI2_FUNCTION_PORT_ENABLE:
1193 		frame_sz = sizeof(Mpi2PortEnableRequest_t);
1194 		func_str = "port_enable";
1195 		break;
1196 	case MPI2_FUNCTION_SMP_PASSTHROUGH:
1197 		frame_sz = sizeof(Mpi2SmpPassthroughRequest_t) + ioc->sge_size;
1198 		func_str = "smp_passthru";
1199 		break;
1200 	case MPI2_FUNCTION_NVME_ENCAPSULATED:
1201 		frame_sz = sizeof(Mpi26NVMeEncapsulatedRequest_t) +
1202 		    ioc->sge_size;
1203 		func_str = "nvme_encapsulated";
1204 		break;
1205 	default:
1206 		frame_sz = 32;
1207 		func_str = "unknown";
1208 		break;
1209 	}
1210 
1211 	ioc_warn(ioc, "ioc_status: %s(0x%04x), request(0x%p),(%s)\n",
1212 		 desc, ioc_status, request_hdr, func_str);
1213 
1214 	_debug_dump_mf(request_hdr, frame_sz/4);
1215 }
1216 
1217 /**
1218  * _base_display_event_data - verbose translation of firmware asyn events
1219  * @ioc: per adapter object
1220  * @mpi_reply: reply mf payload returned from firmware
1221  */
1222 static void
1223 _base_display_event_data(struct MPT3SAS_ADAPTER *ioc,
1224 	Mpi2EventNotificationReply_t *mpi_reply)
1225 {
1226 	char *desc = NULL;
1227 	u16 event;
1228 
1229 	if (!(ioc->logging_level & MPT_DEBUG_EVENTS))
1230 		return;
1231 
1232 	event = le16_to_cpu(mpi_reply->Event);
1233 
1234 	switch (event) {
1235 	case MPI2_EVENT_LOG_DATA:
1236 		desc = "Log Data";
1237 		break;
1238 	case MPI2_EVENT_STATE_CHANGE:
1239 		desc = "Status Change";
1240 		break;
1241 	case MPI2_EVENT_HARD_RESET_RECEIVED:
1242 		desc = "Hard Reset Received";
1243 		break;
1244 	case MPI2_EVENT_EVENT_CHANGE:
1245 		desc = "Event Change";
1246 		break;
1247 	case MPI2_EVENT_SAS_DEVICE_STATUS_CHANGE:
1248 		desc = "Device Status Change";
1249 		break;
1250 	case MPI2_EVENT_IR_OPERATION_STATUS:
1251 		if (!ioc->hide_ir_msg)
1252 			desc = "IR Operation Status";
1253 		break;
1254 	case MPI2_EVENT_SAS_DISCOVERY:
1255 	{
1256 		Mpi2EventDataSasDiscovery_t *event_data =
1257 		    (Mpi2EventDataSasDiscovery_t *)mpi_reply->EventData;
1258 		ioc_info(ioc, "Discovery: (%s)",
1259 			 event_data->ReasonCode == MPI2_EVENT_SAS_DISC_RC_STARTED ?
1260 			 "start" : "stop");
1261 		if (event_data->DiscoveryStatus)
1262 			pr_cont(" discovery_status(0x%08x)",
1263 			    le32_to_cpu(event_data->DiscoveryStatus));
1264 		pr_cont("\n");
1265 		return;
1266 	}
1267 	case MPI2_EVENT_SAS_BROADCAST_PRIMITIVE:
1268 		desc = "SAS Broadcast Primitive";
1269 		break;
1270 	case MPI2_EVENT_SAS_INIT_DEVICE_STATUS_CHANGE:
1271 		desc = "SAS Init Device Status Change";
1272 		break;
1273 	case MPI2_EVENT_SAS_INIT_TABLE_OVERFLOW:
1274 		desc = "SAS Init Table Overflow";
1275 		break;
1276 	case MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST:
1277 		desc = "SAS Topology Change List";
1278 		break;
1279 	case MPI2_EVENT_SAS_ENCL_DEVICE_STATUS_CHANGE:
1280 		desc = "SAS Enclosure Device Status Change";
1281 		break;
1282 	case MPI2_EVENT_IR_VOLUME:
1283 		if (!ioc->hide_ir_msg)
1284 			desc = "IR Volume";
1285 		break;
1286 	case MPI2_EVENT_IR_PHYSICAL_DISK:
1287 		if (!ioc->hide_ir_msg)
1288 			desc = "IR Physical Disk";
1289 		break;
1290 	case MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST:
1291 		if (!ioc->hide_ir_msg)
1292 			desc = "IR Configuration Change List";
1293 		break;
1294 	case MPI2_EVENT_LOG_ENTRY_ADDED:
1295 		if (!ioc->hide_ir_msg)
1296 			desc = "Log Entry Added";
1297 		break;
1298 	case MPI2_EVENT_TEMP_THRESHOLD:
1299 		desc = "Temperature Threshold";
1300 		break;
1301 	case MPI2_EVENT_ACTIVE_CABLE_EXCEPTION:
1302 		desc = "Cable Event";
1303 		break;
1304 	case MPI2_EVENT_SAS_DEVICE_DISCOVERY_ERROR:
1305 		desc = "SAS Device Discovery Error";
1306 		break;
1307 	case MPI2_EVENT_PCIE_DEVICE_STATUS_CHANGE:
1308 		desc = "PCIE Device Status Change";
1309 		break;
1310 	case MPI2_EVENT_PCIE_ENUMERATION:
1311 	{
1312 		Mpi26EventDataPCIeEnumeration_t *event_data =
1313 			(Mpi26EventDataPCIeEnumeration_t *)mpi_reply->EventData;
1314 		ioc_info(ioc, "PCIE Enumeration: (%s)",
1315 			 event_data->ReasonCode == MPI26_EVENT_PCIE_ENUM_RC_STARTED ?
1316 			 "start" : "stop");
1317 		if (event_data->EnumerationStatus)
1318 			pr_cont("enumeration_status(0x%08x)",
1319 				le32_to_cpu(event_data->EnumerationStatus));
1320 		pr_cont("\n");
1321 		return;
1322 	}
1323 	case MPI2_EVENT_PCIE_TOPOLOGY_CHANGE_LIST:
1324 		desc = "PCIE Topology Change List";
1325 		break;
1326 	}
1327 
1328 	if (!desc)
1329 		return;
1330 
1331 	ioc_info(ioc, "%s\n", desc);
1332 }
1333 
1334 /**
1335  * _base_sas_log_info - verbose translation of firmware log info
1336  * @ioc: per adapter object
1337  * @log_info: log info
1338  */
1339 static void
1340 _base_sas_log_info(struct MPT3SAS_ADAPTER *ioc, u32 log_info)
1341 {
1342 	union loginfo_type {
1343 		u32	loginfo;
1344 		struct {
1345 			u32	subcode:16;
1346 			u32	code:8;
1347 			u32	originator:4;
1348 			u32	bus_type:4;
1349 		} dw;
1350 	};
1351 	union loginfo_type sas_loginfo;
1352 	char *originator_str = NULL;
1353 
1354 	sas_loginfo.loginfo = log_info;
1355 	if (sas_loginfo.dw.bus_type != 3 /*SAS*/)
1356 		return;
1357 
1358 	/* each nexus loss loginfo */
1359 	if (log_info == 0x31170000)
1360 		return;
1361 
1362 	/* eat the loginfos associated with task aborts */
1363 	if (ioc->ignore_loginfos && (log_info == 0x30050000 || log_info ==
1364 	    0x31140000 || log_info == 0x31130000))
1365 		return;
1366 
1367 	switch (sas_loginfo.dw.originator) {
1368 	case 0:
1369 		originator_str = "IOP";
1370 		break;
1371 	case 1:
1372 		originator_str = "PL";
1373 		break;
1374 	case 2:
1375 		if (!ioc->hide_ir_msg)
1376 			originator_str = "IR";
1377 		else
1378 			originator_str = "WarpDrive";
1379 		break;
1380 	}
1381 
1382 	ioc_warn(ioc, "log_info(0x%08x): originator(%s), code(0x%02x), sub_code(0x%04x)\n",
1383 		 log_info,
1384 		 originator_str, sas_loginfo.dw.code, sas_loginfo.dw.subcode);
1385 }
1386 
1387 /**
1388  * _base_display_reply_info - handle reply descriptors depending on IOC Status
1389  * @ioc: per adapter object
1390  * @smid: system request message index
1391  * @msix_index: MSIX table index supplied by the OS
1392  * @reply: reply message frame (lower 32bit addr)
1393  */
1394 static void
1395 _base_display_reply_info(struct MPT3SAS_ADAPTER *ioc, u16 smid, u8 msix_index,
1396 	u32 reply)
1397 {
1398 	MPI2DefaultReply_t *mpi_reply;
1399 	u16 ioc_status;
1400 	u32 loginfo = 0;
1401 
1402 	mpi_reply = mpt3sas_base_get_reply_virt_addr(ioc, reply);
1403 	if (unlikely(!mpi_reply)) {
1404 		ioc_err(ioc, "mpi_reply not valid at %s:%d/%s()!\n",
1405 			__FILE__, __LINE__, __func__);
1406 		return;
1407 	}
1408 	ioc_status = le16_to_cpu(mpi_reply->IOCStatus);
1409 
1410 	if ((ioc_status & MPI2_IOCSTATUS_MASK) &&
1411 	    (ioc->logging_level & MPT_DEBUG_REPLY)) {
1412 		_base_sas_ioc_info(ioc, mpi_reply,
1413 		   mpt3sas_base_get_msg_frame(ioc, smid));
1414 	}
1415 
1416 	if (ioc_status & MPI2_IOCSTATUS_FLAG_LOG_INFO_AVAILABLE) {
1417 		loginfo = le32_to_cpu(mpi_reply->IOCLogInfo);
1418 		_base_sas_log_info(ioc, loginfo);
1419 	}
1420 
1421 	if (ioc_status || loginfo) {
1422 		ioc_status &= MPI2_IOCSTATUS_MASK;
1423 		mpt3sas_trigger_mpi(ioc, ioc_status, loginfo);
1424 	}
1425 }
1426 
1427 /**
1428  * mpt3sas_base_done - base internal command completion routine
1429  * @ioc: per adapter object
1430  * @smid: system request message index
1431  * @msix_index: MSIX table index supplied by the OS
1432  * @reply: reply message frame(lower 32bit addr)
1433  *
1434  * Return:
1435  * 1 meaning mf should be freed from _base_interrupt
1436  * 0 means the mf is freed from this function.
1437  */
1438 u8
1439 mpt3sas_base_done(struct MPT3SAS_ADAPTER *ioc, u16 smid, u8 msix_index,
1440 	u32 reply)
1441 {
1442 	MPI2DefaultReply_t *mpi_reply;
1443 
1444 	mpi_reply = mpt3sas_base_get_reply_virt_addr(ioc, reply);
1445 	if (mpi_reply && mpi_reply->Function == MPI2_FUNCTION_EVENT_ACK)
1446 		return mpt3sas_check_for_pending_internal_cmds(ioc, smid);
1447 
1448 	if (ioc->base_cmds.status == MPT3_CMD_NOT_USED)
1449 		return 1;
1450 
1451 	ioc->base_cmds.status |= MPT3_CMD_COMPLETE;
1452 	if (mpi_reply) {
1453 		ioc->base_cmds.status |= MPT3_CMD_REPLY_VALID;
1454 		memcpy(ioc->base_cmds.reply, mpi_reply, mpi_reply->MsgLength*4);
1455 	}
1456 	ioc->base_cmds.status &= ~MPT3_CMD_PENDING;
1457 
1458 	complete(&ioc->base_cmds.done);
1459 	return 1;
1460 }
1461 
1462 /**
1463  * _base_async_event - main callback handler for firmware asyn events
1464  * @ioc: per adapter object
1465  * @msix_index: MSIX table index supplied by the OS
1466  * @reply: reply message frame(lower 32bit addr)
1467  *
1468  * Return:
1469  * 1 meaning mf should be freed from _base_interrupt
1470  * 0 means the mf is freed from this function.
1471  */
1472 static u8
1473 _base_async_event(struct MPT3SAS_ADAPTER *ioc, u8 msix_index, u32 reply)
1474 {
1475 	Mpi2EventNotificationReply_t *mpi_reply;
1476 	Mpi2EventAckRequest_t *ack_request;
1477 	u16 smid;
1478 	struct _event_ack_list *delayed_event_ack;
1479 
1480 	mpi_reply = mpt3sas_base_get_reply_virt_addr(ioc, reply);
1481 	if (!mpi_reply)
1482 		return 1;
1483 	if (mpi_reply->Function != MPI2_FUNCTION_EVENT_NOTIFICATION)
1484 		return 1;
1485 
1486 	_base_display_event_data(ioc, mpi_reply);
1487 
1488 	if (!(mpi_reply->AckRequired & MPI2_EVENT_NOTIFICATION_ACK_REQUIRED))
1489 		goto out;
1490 	smid = mpt3sas_base_get_smid(ioc, ioc->base_cb_idx);
1491 	if (!smid) {
1492 		delayed_event_ack = kzalloc(sizeof(*delayed_event_ack),
1493 					GFP_ATOMIC);
1494 		if (!delayed_event_ack)
1495 			goto out;
1496 		INIT_LIST_HEAD(&delayed_event_ack->list);
1497 		delayed_event_ack->Event = mpi_reply->Event;
1498 		delayed_event_ack->EventContext = mpi_reply->EventContext;
1499 		list_add_tail(&delayed_event_ack->list,
1500 				&ioc->delayed_event_ack_list);
1501 		dewtprintk(ioc,
1502 			   ioc_info(ioc, "DELAYED: EVENT ACK: event (0x%04x)\n",
1503 				    le16_to_cpu(mpi_reply->Event)));
1504 		goto out;
1505 	}
1506 
1507 	ack_request = mpt3sas_base_get_msg_frame(ioc, smid);
1508 	memset(ack_request, 0, sizeof(Mpi2EventAckRequest_t));
1509 	ack_request->Function = MPI2_FUNCTION_EVENT_ACK;
1510 	ack_request->Event = mpi_reply->Event;
1511 	ack_request->EventContext = mpi_reply->EventContext;
1512 	ack_request->VF_ID = 0;  /* TODO */
1513 	ack_request->VP_ID = 0;
1514 	ioc->put_smid_default(ioc, smid);
1515 
1516  out:
1517 
1518 	/* scsih callback handler */
1519 	mpt3sas_scsih_event_callback(ioc, msix_index, reply);
1520 
1521 	/* ctl callback handler */
1522 	mpt3sas_ctl_event_callback(ioc, msix_index, reply);
1523 
1524 	return 1;
1525 }
1526 
1527 static struct scsiio_tracker *
1528 _get_st_from_smid(struct MPT3SAS_ADAPTER *ioc, u16 smid)
1529 {
1530 	struct scsi_cmnd *cmd;
1531 
1532 	if (WARN_ON(!smid) ||
1533 	    WARN_ON(smid >= ioc->hi_priority_smid))
1534 		return NULL;
1535 
1536 	cmd = mpt3sas_scsih_scsi_lookup_get(ioc, smid);
1537 	if (cmd)
1538 		return scsi_cmd_priv(cmd);
1539 
1540 	return NULL;
1541 }
1542 
1543 /**
1544  * _base_get_cb_idx - obtain the callback index
1545  * @ioc: per adapter object
1546  * @smid: system request message index
1547  *
1548  * Return: callback index.
1549  */
1550 static u8
1551 _base_get_cb_idx(struct MPT3SAS_ADAPTER *ioc, u16 smid)
1552 {
1553 	int i;
1554 	u16 ctl_smid = ioc->scsiio_depth - INTERNAL_SCSIIO_CMDS_COUNT + 1;
1555 	u8 cb_idx = 0xFF;
1556 
1557 	if (smid < ioc->hi_priority_smid) {
1558 		struct scsiio_tracker *st;
1559 
1560 		if (smid < ctl_smid) {
1561 			st = _get_st_from_smid(ioc, smid);
1562 			if (st)
1563 				cb_idx = st->cb_idx;
1564 		} else if (smid == ctl_smid)
1565 			cb_idx = ioc->ctl_cb_idx;
1566 	} else if (smid < ioc->internal_smid) {
1567 		i = smid - ioc->hi_priority_smid;
1568 		cb_idx = ioc->hpr_lookup[i].cb_idx;
1569 	} else if (smid <= ioc->hba_queue_depth) {
1570 		i = smid - ioc->internal_smid;
1571 		cb_idx = ioc->internal_lookup[i].cb_idx;
1572 	}
1573 	return cb_idx;
1574 }
1575 
1576 /**
1577  * mpt3sas_base_pause_mq_polling - pause polling on the mq poll queues
1578  *				when driver is flushing out the IOs.
1579  * @ioc: per adapter object
1580  *
1581  * Pause polling on the mq poll (io uring) queues when driver is flushing
1582  * out the IOs. Otherwise we may see the race condition of completing the same
1583  * IO from two paths.
1584  *
1585  * Returns nothing.
1586  */
1587 void
1588 mpt3sas_base_pause_mq_polling(struct MPT3SAS_ADAPTER *ioc)
1589 {
1590 	int iopoll_q_count =
1591 	    ioc->reply_queue_count - ioc->iopoll_q_start_index;
1592 	int qid;
1593 
1594 	for (qid = 0; qid < iopoll_q_count; qid++)
1595 		atomic_set(&ioc->io_uring_poll_queues[qid].pause, 1);
1596 
1597 	/*
1598 	 * wait for current poll to complete.
1599 	 */
1600 	for (qid = 0; qid < iopoll_q_count; qid++) {
1601 		while (atomic_read(&ioc->io_uring_poll_queues[qid].busy)) {
1602 			cpu_relax();
1603 			udelay(500);
1604 		}
1605 	}
1606 }
1607 
1608 /**
1609  * mpt3sas_base_resume_mq_polling - Resume polling on mq poll queues.
1610  * @ioc: per adapter object
1611  *
1612  * Returns nothing.
1613  */
1614 void
1615 mpt3sas_base_resume_mq_polling(struct MPT3SAS_ADAPTER *ioc)
1616 {
1617 	int iopoll_q_count =
1618 	    ioc->reply_queue_count - ioc->iopoll_q_start_index;
1619 	int qid;
1620 
1621 	for (qid = 0; qid < iopoll_q_count; qid++)
1622 		atomic_set(&ioc->io_uring_poll_queues[qid].pause, 0);
1623 }
1624 
1625 /**
1626  * mpt3sas_base_mask_interrupts - disable interrupts
1627  * @ioc: per adapter object
1628  *
1629  * Disabling ResetIRQ, Reply and Doorbell Interrupts
1630  */
1631 void
1632 mpt3sas_base_mask_interrupts(struct MPT3SAS_ADAPTER *ioc)
1633 {
1634 	u32 him_register;
1635 
1636 	ioc->mask_interrupts = 1;
1637 	him_register = ioc->base_readl(&ioc->chip->HostInterruptMask);
1638 	him_register |= MPI2_HIM_DIM + MPI2_HIM_RIM + MPI2_HIM_RESET_IRQ_MASK;
1639 	writel(him_register, &ioc->chip->HostInterruptMask);
1640 	ioc->base_readl(&ioc->chip->HostInterruptMask);
1641 }
1642 
1643 /**
1644  * mpt3sas_base_unmask_interrupts - enable interrupts
1645  * @ioc: per adapter object
1646  *
1647  * Enabling only Reply Interrupts
1648  */
1649 void
1650 mpt3sas_base_unmask_interrupts(struct MPT3SAS_ADAPTER *ioc)
1651 {
1652 	u32 him_register;
1653 
1654 	him_register = ioc->base_readl(&ioc->chip->HostInterruptMask);
1655 	him_register &= ~MPI2_HIM_RIM;
1656 	writel(him_register, &ioc->chip->HostInterruptMask);
1657 	ioc->mask_interrupts = 0;
1658 }
1659 
1660 union reply_descriptor {
1661 	u64 word;
1662 	struct {
1663 		u32 low;
1664 		u32 high;
1665 	} u;
1666 };
1667 
1668 static u32 base_mod64(u64 dividend, u32 divisor)
1669 {
1670 	u32 remainder;
1671 
1672 	if (!divisor)
1673 		pr_err("mpt3sas: DIVISOR is zero, in div fn\n");
1674 	remainder = do_div(dividend, divisor);
1675 	return remainder;
1676 }
1677 
1678 /**
1679  * _base_process_reply_queue - Process reply descriptors from reply
1680  *		descriptor post queue.
1681  * @reply_q: per IRQ's reply queue object.
1682  *
1683  * Return: number of reply descriptors processed from reply
1684  *		descriptor queue.
1685  */
1686 static int
1687 _base_process_reply_queue(struct adapter_reply_queue *reply_q)
1688 {
1689 	union reply_descriptor rd;
1690 	u64 completed_cmds;
1691 	u8 request_descript_type;
1692 	u16 smid;
1693 	u8 cb_idx;
1694 	u32 reply;
1695 	u8 msix_index = reply_q->msix_index;
1696 	struct MPT3SAS_ADAPTER *ioc = reply_q->ioc;
1697 	Mpi2ReplyDescriptorsUnion_t *rpf;
1698 	u8 rc;
1699 
1700 	completed_cmds = 0;
1701 	if (!atomic_add_unless(&reply_q->busy, 1, 1))
1702 		return completed_cmds;
1703 
1704 	rpf = &reply_q->reply_post_free[reply_q->reply_post_host_index];
1705 	request_descript_type = rpf->Default.ReplyFlags
1706 	     & MPI2_RPY_DESCRIPT_FLAGS_TYPE_MASK;
1707 	if (request_descript_type == MPI2_RPY_DESCRIPT_FLAGS_UNUSED) {
1708 		atomic_dec(&reply_q->busy);
1709 		return completed_cmds;
1710 	}
1711 
1712 	cb_idx = 0xFF;
1713 	do {
1714 		rd.word = le64_to_cpu(rpf->Words);
1715 		if (rd.u.low == UINT_MAX || rd.u.high == UINT_MAX)
1716 			goto out;
1717 		reply = 0;
1718 		smid = le16_to_cpu(rpf->Default.DescriptorTypeDependent1);
1719 		if (request_descript_type ==
1720 		    MPI25_RPY_DESCRIPT_FLAGS_FAST_PATH_SCSI_IO_SUCCESS ||
1721 		    request_descript_type ==
1722 		    MPI2_RPY_DESCRIPT_FLAGS_SCSI_IO_SUCCESS ||
1723 		    request_descript_type ==
1724 		    MPI26_RPY_DESCRIPT_FLAGS_PCIE_ENCAPSULATED_SUCCESS) {
1725 			cb_idx = _base_get_cb_idx(ioc, smid);
1726 			if ((likely(cb_idx < MPT_MAX_CALLBACKS)) &&
1727 			    (likely(mpt_callbacks[cb_idx] != NULL))) {
1728 				rc = mpt_callbacks[cb_idx](ioc, smid,
1729 				    msix_index, 0);
1730 				if (rc)
1731 					mpt3sas_base_free_smid(ioc, smid);
1732 			}
1733 		} else if (request_descript_type ==
1734 		    MPI2_RPY_DESCRIPT_FLAGS_ADDRESS_REPLY) {
1735 			reply = le32_to_cpu(
1736 			    rpf->AddressReply.ReplyFrameAddress);
1737 			if (reply > ioc->reply_dma_max_address ||
1738 			    reply < ioc->reply_dma_min_address)
1739 				reply = 0;
1740 			if (smid) {
1741 				cb_idx = _base_get_cb_idx(ioc, smid);
1742 				if ((likely(cb_idx < MPT_MAX_CALLBACKS)) &&
1743 				    (likely(mpt_callbacks[cb_idx] != NULL))) {
1744 					rc = mpt_callbacks[cb_idx](ioc, smid,
1745 					    msix_index, reply);
1746 					if (reply)
1747 						_base_display_reply_info(ioc,
1748 						    smid, msix_index, reply);
1749 					if (rc)
1750 						mpt3sas_base_free_smid(ioc,
1751 						    smid);
1752 				}
1753 			} else {
1754 				_base_async_event(ioc, msix_index, reply);
1755 			}
1756 
1757 			/* reply free queue handling */
1758 			if (reply) {
1759 				ioc->reply_free_host_index =
1760 				    (ioc->reply_free_host_index ==
1761 				    (ioc->reply_free_queue_depth - 1)) ?
1762 				    0 : ioc->reply_free_host_index + 1;
1763 				ioc->reply_free[ioc->reply_free_host_index] =
1764 				    cpu_to_le32(reply);
1765 				if (ioc->is_mcpu_endpoint)
1766 					_base_clone_reply_to_sys_mem(ioc,
1767 						reply,
1768 						ioc->reply_free_host_index);
1769 				writel(ioc->reply_free_host_index,
1770 				    &ioc->chip->ReplyFreeHostIndex);
1771 			}
1772 		}
1773 
1774 		rpf->Words = cpu_to_le64(ULLONG_MAX);
1775 		reply_q->reply_post_host_index =
1776 		    (reply_q->reply_post_host_index ==
1777 		    (ioc->reply_post_queue_depth - 1)) ? 0 :
1778 		    reply_q->reply_post_host_index + 1;
1779 		request_descript_type =
1780 		    reply_q->reply_post_free[reply_q->reply_post_host_index].
1781 		    Default.ReplyFlags & MPI2_RPY_DESCRIPT_FLAGS_TYPE_MASK;
1782 		completed_cmds++;
1783 		/* Update the reply post host index after continuously
1784 		 * processing the threshold number of Reply Descriptors.
1785 		 * So that FW can find enough entries to post the Reply
1786 		 * Descriptors in the reply descriptor post queue.
1787 		 */
1788 		if (completed_cmds >= ioc->thresh_hold) {
1789 			if (ioc->combined_reply_queue) {
1790 				writel(reply_q->reply_post_host_index |
1791 						((msix_index  & 7) <<
1792 						 MPI2_RPHI_MSIX_INDEX_SHIFT),
1793 				    ioc->replyPostRegisterIndex[msix_index/8]);
1794 			} else {
1795 				writel(reply_q->reply_post_host_index |
1796 						(msix_index <<
1797 						 MPI2_RPHI_MSIX_INDEX_SHIFT),
1798 						&ioc->chip->ReplyPostHostIndex);
1799 			}
1800 			if (!reply_q->is_iouring_poll_q &&
1801 			    !reply_q->irq_poll_scheduled) {
1802 				reply_q->irq_poll_scheduled = true;
1803 				irq_poll_sched(&reply_q->irqpoll);
1804 			}
1805 			atomic_dec(&reply_q->busy);
1806 			return completed_cmds;
1807 		}
1808 		if (request_descript_type == MPI2_RPY_DESCRIPT_FLAGS_UNUSED)
1809 			goto out;
1810 		if (!reply_q->reply_post_host_index)
1811 			rpf = reply_q->reply_post_free;
1812 		else
1813 			rpf++;
1814 	} while (1);
1815 
1816  out:
1817 
1818 	if (!completed_cmds) {
1819 		atomic_dec(&reply_q->busy);
1820 		return completed_cmds;
1821 	}
1822 
1823 	if (ioc->is_warpdrive) {
1824 		writel(reply_q->reply_post_host_index,
1825 		ioc->reply_post_host_index[msix_index]);
1826 		atomic_dec(&reply_q->busy);
1827 		return completed_cmds;
1828 	}
1829 
1830 	/* Update Reply Post Host Index.
1831 	 * For those HBA's which support combined reply queue feature
1832 	 * 1. Get the correct Supplemental Reply Post Host Index Register.
1833 	 *    i.e. (msix_index / 8)th entry from Supplemental Reply Post Host
1834 	 *    Index Register address bank i.e replyPostRegisterIndex[],
1835 	 * 2. Then update this register with new reply host index value
1836 	 *    in ReplyPostIndex field and the MSIxIndex field with
1837 	 *    msix_index value reduced to a value between 0 and 7,
1838 	 *    using a modulo 8 operation. Since each Supplemental Reply Post
1839 	 *    Host Index Register supports 8 MSI-X vectors.
1840 	 *
1841 	 * For other HBA's just update the Reply Post Host Index register with
1842 	 * new reply host index value in ReplyPostIndex Field and msix_index
1843 	 * value in MSIxIndex field.
1844 	 */
1845 	if (ioc->combined_reply_queue)
1846 		writel(reply_q->reply_post_host_index | ((msix_index  & 7) <<
1847 			MPI2_RPHI_MSIX_INDEX_SHIFT),
1848 			ioc->replyPostRegisterIndex[msix_index/8]);
1849 	else
1850 		writel(reply_q->reply_post_host_index | (msix_index <<
1851 			MPI2_RPHI_MSIX_INDEX_SHIFT),
1852 			&ioc->chip->ReplyPostHostIndex);
1853 	atomic_dec(&reply_q->busy);
1854 	return completed_cmds;
1855 }
1856 
1857 /**
1858  * mpt3sas_blk_mq_poll - poll the blk mq poll queue
1859  * @shost: Scsi_Host object
1860  * @queue_num: hw ctx queue number
1861  *
1862  * Return number of entries that has been processed from poll queue.
1863  */
1864 int mpt3sas_blk_mq_poll(struct Scsi_Host *shost, unsigned int queue_num)
1865 {
1866 	struct MPT3SAS_ADAPTER *ioc =
1867 	    (struct MPT3SAS_ADAPTER *)shost->hostdata;
1868 	struct adapter_reply_queue *reply_q;
1869 	int num_entries = 0;
1870 	int qid = queue_num - ioc->iopoll_q_start_index;
1871 
1872 	if (atomic_read(&ioc->io_uring_poll_queues[qid].pause) ||
1873 	    !atomic_add_unless(&ioc->io_uring_poll_queues[qid].busy, 1, 1))
1874 		return 0;
1875 
1876 	reply_q = ioc->io_uring_poll_queues[qid].reply_q;
1877 
1878 	num_entries = _base_process_reply_queue(reply_q);
1879 	atomic_dec(&ioc->io_uring_poll_queues[qid].busy);
1880 
1881 	return num_entries;
1882 }
1883 
1884 /**
1885  * _base_interrupt - MPT adapter (IOC) specific interrupt handler.
1886  * @irq: irq number (not used)
1887  * @bus_id: bus identifier cookie == pointer to MPT_ADAPTER structure
1888  *
1889  * Return: IRQ_HANDLED if processed, else IRQ_NONE.
1890  */
1891 static irqreturn_t
1892 _base_interrupt(int irq, void *bus_id)
1893 {
1894 	struct adapter_reply_queue *reply_q = bus_id;
1895 	struct MPT3SAS_ADAPTER *ioc = reply_q->ioc;
1896 
1897 	if (ioc->mask_interrupts)
1898 		return IRQ_NONE;
1899 	if (reply_q->irq_poll_scheduled)
1900 		return IRQ_HANDLED;
1901 	return ((_base_process_reply_queue(reply_q) > 0) ?
1902 			IRQ_HANDLED : IRQ_NONE);
1903 }
1904 
1905 /**
1906  * _base_irqpoll - IRQ poll callback handler
1907  * @irqpoll: irq_poll object
1908  * @budget: irq poll weight
1909  *
1910  * Return: number of reply descriptors processed
1911  */
1912 static int
1913 _base_irqpoll(struct irq_poll *irqpoll, int budget)
1914 {
1915 	struct adapter_reply_queue *reply_q;
1916 	int num_entries = 0;
1917 
1918 	reply_q = container_of(irqpoll, struct adapter_reply_queue,
1919 			irqpoll);
1920 	if (reply_q->irq_line_enable) {
1921 		disable_irq_nosync(reply_q->os_irq);
1922 		reply_q->irq_line_enable = false;
1923 	}
1924 	num_entries = _base_process_reply_queue(reply_q);
1925 	if (num_entries < budget) {
1926 		irq_poll_complete(irqpoll);
1927 		reply_q->irq_poll_scheduled = false;
1928 		reply_q->irq_line_enable = true;
1929 		enable_irq(reply_q->os_irq);
1930 		/*
1931 		 * Go for one more round of processing the
1932 		 * reply descriptor post queue in case the HBA
1933 		 * Firmware has posted some reply descriptors
1934 		 * while reenabling the IRQ.
1935 		 */
1936 		_base_process_reply_queue(reply_q);
1937 	}
1938 
1939 	return num_entries;
1940 }
1941 
1942 /**
1943  * _base_init_irqpolls - initliaze IRQ polls
1944  * @ioc: per adapter object
1945  *
1946  * Return: nothing
1947  */
1948 static void
1949 _base_init_irqpolls(struct MPT3SAS_ADAPTER *ioc)
1950 {
1951 	struct adapter_reply_queue *reply_q, *next;
1952 
1953 	if (list_empty(&ioc->reply_queue_list))
1954 		return;
1955 
1956 	list_for_each_entry_safe(reply_q, next, &ioc->reply_queue_list, list) {
1957 		if (reply_q->is_iouring_poll_q)
1958 			continue;
1959 		irq_poll_init(&reply_q->irqpoll,
1960 			ioc->hba_queue_depth/4, _base_irqpoll);
1961 		reply_q->irq_poll_scheduled = false;
1962 		reply_q->irq_line_enable = true;
1963 		reply_q->os_irq = pci_irq_vector(ioc->pdev,
1964 		    reply_q->msix_index);
1965 	}
1966 }
1967 
1968 /**
1969  * _base_is_controller_msix_enabled - is controller support muli-reply queues
1970  * @ioc: per adapter object
1971  *
1972  * Return: Whether or not MSI/X is enabled.
1973  */
1974 static inline int
1975 _base_is_controller_msix_enabled(struct MPT3SAS_ADAPTER *ioc)
1976 {
1977 	return (ioc->facts.IOCCapabilities &
1978 	    MPI2_IOCFACTS_CAPABILITY_MSI_X_INDEX) && ioc->msix_enable;
1979 }
1980 
1981 /**
1982  * mpt3sas_base_sync_reply_irqs - flush pending MSIX interrupts
1983  * @ioc: per adapter object
1984  * @poll: poll over reply descriptor pools incase interrupt for
1985  *		timed-out SCSI command got delayed
1986  * Context: non-ISR context
1987  *
1988  * Called when a Task Management request has completed.
1989  */
1990 void
1991 mpt3sas_base_sync_reply_irqs(struct MPT3SAS_ADAPTER *ioc, u8 poll)
1992 {
1993 	struct adapter_reply_queue *reply_q;
1994 
1995 	/* If MSIX capability is turned off
1996 	 * then multi-queues are not enabled
1997 	 */
1998 	if (!_base_is_controller_msix_enabled(ioc))
1999 		return;
2000 
2001 	list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
2002 		if (ioc->shost_recovery || ioc->remove_host ||
2003 				ioc->pci_error_recovery)
2004 			return;
2005 		/* TMs are on msix_index == 0 */
2006 		if (reply_q->msix_index == 0)
2007 			continue;
2008 
2009 		if (reply_q->is_iouring_poll_q) {
2010 			_base_process_reply_queue(reply_q);
2011 			continue;
2012 		}
2013 
2014 		synchronize_irq(pci_irq_vector(ioc->pdev, reply_q->msix_index));
2015 		if (reply_q->irq_poll_scheduled) {
2016 			/* Calling irq_poll_disable will wait for any pending
2017 			 * callbacks to have completed.
2018 			 */
2019 			irq_poll_disable(&reply_q->irqpoll);
2020 			irq_poll_enable(&reply_q->irqpoll);
2021 			/* check how the scheduled poll has ended,
2022 			 * clean up only if necessary
2023 			 */
2024 			if (reply_q->irq_poll_scheduled) {
2025 				reply_q->irq_poll_scheduled = false;
2026 				reply_q->irq_line_enable = true;
2027 				enable_irq(reply_q->os_irq);
2028 			}
2029 		}
2030 
2031 		if (poll)
2032 			_base_process_reply_queue(reply_q);
2033 	}
2034 }
2035 
2036 /**
2037  * mpt3sas_base_release_callback_handler - clear interrupt callback handler
2038  * @cb_idx: callback index
2039  */
2040 void
2041 mpt3sas_base_release_callback_handler(u8 cb_idx)
2042 {
2043 	mpt_callbacks[cb_idx] = NULL;
2044 }
2045 
2046 /**
2047  * mpt3sas_base_register_callback_handler - obtain index for the interrupt callback handler
2048  * @cb_func: callback function
2049  *
2050  * Return: Index of @cb_func.
2051  */
2052 u8
2053 mpt3sas_base_register_callback_handler(MPT_CALLBACK cb_func)
2054 {
2055 	u8 cb_idx;
2056 
2057 	for (cb_idx = MPT_MAX_CALLBACKS-1; cb_idx; cb_idx--)
2058 		if (mpt_callbacks[cb_idx] == NULL)
2059 			break;
2060 
2061 	mpt_callbacks[cb_idx] = cb_func;
2062 	return cb_idx;
2063 }
2064 
2065 /**
2066  * mpt3sas_base_initialize_callback_handler - initialize the interrupt callback handler
2067  */
2068 void
2069 mpt3sas_base_initialize_callback_handler(void)
2070 {
2071 	u8 cb_idx;
2072 
2073 	for (cb_idx = 0; cb_idx < MPT_MAX_CALLBACKS; cb_idx++)
2074 		mpt3sas_base_release_callback_handler(cb_idx);
2075 }
2076 
2077 
2078 /**
2079  * _base_build_zero_len_sge - build zero length sg entry
2080  * @ioc: per adapter object
2081  * @paddr: virtual address for SGE
2082  *
2083  * Create a zero length scatter gather entry to insure the IOCs hardware has
2084  * something to use if the target device goes brain dead and tries
2085  * to send data even when none is asked for.
2086  */
2087 static void
2088 _base_build_zero_len_sge(struct MPT3SAS_ADAPTER *ioc, void *paddr)
2089 {
2090 	u32 flags_length = (u32)((MPI2_SGE_FLAGS_LAST_ELEMENT |
2091 	    MPI2_SGE_FLAGS_END_OF_BUFFER | MPI2_SGE_FLAGS_END_OF_LIST |
2092 	    MPI2_SGE_FLAGS_SIMPLE_ELEMENT) <<
2093 	    MPI2_SGE_FLAGS_SHIFT);
2094 	ioc->base_add_sg_single(paddr, flags_length, -1);
2095 }
2096 
2097 /**
2098  * _base_add_sg_single_32 - Place a simple 32 bit SGE at address pAddr.
2099  * @paddr: virtual address for SGE
2100  * @flags_length: SGE flags and data transfer length
2101  * @dma_addr: Physical address
2102  */
2103 static void
2104 _base_add_sg_single_32(void *paddr, u32 flags_length, dma_addr_t dma_addr)
2105 {
2106 	Mpi2SGESimple32_t *sgel = paddr;
2107 
2108 	flags_length |= (MPI2_SGE_FLAGS_32_BIT_ADDRESSING |
2109 	    MPI2_SGE_FLAGS_SYSTEM_ADDRESS) << MPI2_SGE_FLAGS_SHIFT;
2110 	sgel->FlagsLength = cpu_to_le32(flags_length);
2111 	sgel->Address = cpu_to_le32(dma_addr);
2112 }
2113 
2114 
2115 /**
2116  * _base_add_sg_single_64 - Place a simple 64 bit SGE at address pAddr.
2117  * @paddr: virtual address for SGE
2118  * @flags_length: SGE flags and data transfer length
2119  * @dma_addr: Physical address
2120  */
2121 static void
2122 _base_add_sg_single_64(void *paddr, u32 flags_length, dma_addr_t dma_addr)
2123 {
2124 	Mpi2SGESimple64_t *sgel = paddr;
2125 
2126 	flags_length |= (MPI2_SGE_FLAGS_64_BIT_ADDRESSING |
2127 	    MPI2_SGE_FLAGS_SYSTEM_ADDRESS) << MPI2_SGE_FLAGS_SHIFT;
2128 	sgel->FlagsLength = cpu_to_le32(flags_length);
2129 	sgel->Address = cpu_to_le64(dma_addr);
2130 }
2131 
2132 /**
2133  * _base_get_chain_buffer_tracker - obtain chain tracker
2134  * @ioc: per adapter object
2135  * @scmd: SCSI commands of the IO request
2136  *
2137  * Return: chain tracker from chain_lookup table using key as
2138  * smid and smid's chain_offset.
2139  */
2140 static struct chain_tracker *
2141 _base_get_chain_buffer_tracker(struct MPT3SAS_ADAPTER *ioc,
2142 			       struct scsi_cmnd *scmd)
2143 {
2144 	struct chain_tracker *chain_req;
2145 	struct scsiio_tracker *st = scsi_cmd_priv(scmd);
2146 	u16 smid = st->smid;
2147 	u8 chain_offset =
2148 	   atomic_read(&ioc->chain_lookup[smid - 1].chain_offset);
2149 
2150 	if (chain_offset == ioc->chains_needed_per_io)
2151 		return NULL;
2152 
2153 	chain_req = &ioc->chain_lookup[smid - 1].chains_per_smid[chain_offset];
2154 	atomic_inc(&ioc->chain_lookup[smid - 1].chain_offset);
2155 	return chain_req;
2156 }
2157 
2158 
2159 /**
2160  * _base_build_sg - build generic sg
2161  * @ioc: per adapter object
2162  * @psge: virtual address for SGE
2163  * @data_out_dma: physical address for WRITES
2164  * @data_out_sz: data xfer size for WRITES
2165  * @data_in_dma: physical address for READS
2166  * @data_in_sz: data xfer size for READS
2167  */
2168 static void
2169 _base_build_sg(struct MPT3SAS_ADAPTER *ioc, void *psge,
2170 	dma_addr_t data_out_dma, size_t data_out_sz, dma_addr_t data_in_dma,
2171 	size_t data_in_sz)
2172 {
2173 	u32 sgl_flags;
2174 
2175 	if (!data_out_sz && !data_in_sz) {
2176 		_base_build_zero_len_sge(ioc, psge);
2177 		return;
2178 	}
2179 
2180 	if (data_out_sz && data_in_sz) {
2181 		/* WRITE sgel first */
2182 		sgl_flags = (MPI2_SGE_FLAGS_SIMPLE_ELEMENT |
2183 		    MPI2_SGE_FLAGS_END_OF_BUFFER | MPI2_SGE_FLAGS_HOST_TO_IOC);
2184 		sgl_flags = sgl_flags << MPI2_SGE_FLAGS_SHIFT;
2185 		ioc->base_add_sg_single(psge, sgl_flags |
2186 		    data_out_sz, data_out_dma);
2187 
2188 		/* incr sgel */
2189 		psge += ioc->sge_size;
2190 
2191 		/* READ sgel last */
2192 		sgl_flags = (MPI2_SGE_FLAGS_SIMPLE_ELEMENT |
2193 		    MPI2_SGE_FLAGS_LAST_ELEMENT | MPI2_SGE_FLAGS_END_OF_BUFFER |
2194 		    MPI2_SGE_FLAGS_END_OF_LIST);
2195 		sgl_flags = sgl_flags << MPI2_SGE_FLAGS_SHIFT;
2196 		ioc->base_add_sg_single(psge, sgl_flags |
2197 		    data_in_sz, data_in_dma);
2198 	} else if (data_out_sz) /* WRITE */ {
2199 		sgl_flags = (MPI2_SGE_FLAGS_SIMPLE_ELEMENT |
2200 		    MPI2_SGE_FLAGS_LAST_ELEMENT | MPI2_SGE_FLAGS_END_OF_BUFFER |
2201 		    MPI2_SGE_FLAGS_END_OF_LIST | MPI2_SGE_FLAGS_HOST_TO_IOC);
2202 		sgl_flags = sgl_flags << MPI2_SGE_FLAGS_SHIFT;
2203 		ioc->base_add_sg_single(psge, sgl_flags |
2204 		    data_out_sz, data_out_dma);
2205 	} else if (data_in_sz) /* READ */ {
2206 		sgl_flags = (MPI2_SGE_FLAGS_SIMPLE_ELEMENT |
2207 		    MPI2_SGE_FLAGS_LAST_ELEMENT | MPI2_SGE_FLAGS_END_OF_BUFFER |
2208 		    MPI2_SGE_FLAGS_END_OF_LIST);
2209 		sgl_flags = sgl_flags << MPI2_SGE_FLAGS_SHIFT;
2210 		ioc->base_add_sg_single(psge, sgl_flags |
2211 		    data_in_sz, data_in_dma);
2212 	}
2213 }
2214 
2215 /* IEEE format sgls */
2216 
2217 /**
2218  * _base_build_nvme_prp - This function is called for NVMe end devices to build
2219  *                        a native SGL (NVMe PRP).
2220  * @ioc: per adapter object
2221  * @smid: system request message index for getting asscociated SGL
2222  * @nvme_encap_request: the NVMe request msg frame pointer
2223  * @data_out_dma: physical address for WRITES
2224  * @data_out_sz: data xfer size for WRITES
2225  * @data_in_dma: physical address for READS
2226  * @data_in_sz: data xfer size for READS
2227  *
2228  * The native SGL is built starting in the first PRP
2229  * entry of the NVMe message (PRP1).  If the data buffer is small enough to be
2230  * described entirely using PRP1, then PRP2 is not used.  If needed, PRP2 is
2231  * used to describe a larger data buffer.  If the data buffer is too large to
2232  * describe using the two PRP entriess inside the NVMe message, then PRP1
2233  * describes the first data memory segment, and PRP2 contains a pointer to a PRP
2234  * list located elsewhere in memory to describe the remaining data memory
2235  * segments.  The PRP list will be contiguous.
2236  *
2237  * The native SGL for NVMe devices is a Physical Region Page (PRP).  A PRP
2238  * consists of a list of PRP entries to describe a number of noncontigous
2239  * physical memory segments as a single memory buffer, just as a SGL does.  Note
2240  * however, that this function is only used by the IOCTL call, so the memory
2241  * given will be guaranteed to be contiguous.  There is no need to translate
2242  * non-contiguous SGL into a PRP in this case.  All PRPs will describe
2243  * contiguous space that is one page size each.
2244  *
2245  * Each NVMe message contains two PRP entries.  The first (PRP1) either contains
2246  * a PRP list pointer or a PRP element, depending upon the command.  PRP2
2247  * contains the second PRP element if the memory being described fits within 2
2248  * PRP entries, or a PRP list pointer if the PRP spans more than two entries.
2249  *
2250  * A PRP list pointer contains the address of a PRP list, structured as a linear
2251  * array of PRP entries.  Each PRP entry in this list describes a segment of
2252  * physical memory.
2253  *
2254  * Each 64-bit PRP entry comprises an address and an offset field.  The address
2255  * always points at the beginning of a 4KB physical memory page, and the offset
2256  * describes where within that 4KB page the memory segment begins.  Only the
2257  * first element in a PRP list may contain a non-zero offset, implying that all
2258  * memory segments following the first begin at the start of a 4KB page.
2259  *
2260  * Each PRP element normally describes 4KB of physical memory, with exceptions
2261  * for the first and last elements in the list.  If the memory being described
2262  * by the list begins at a non-zero offset within the first 4KB page, then the
2263  * first PRP element will contain a non-zero offset indicating where the region
2264  * begins within the 4KB page.  The last memory segment may end before the end
2265  * of the 4KB segment, depending upon the overall size of the memory being
2266  * described by the PRP list.
2267  *
2268  * Since PRP entries lack any indication of size, the overall data buffer length
2269  * is used to determine where the end of the data memory buffer is located, and
2270  * how many PRP entries are required to describe it.
2271  */
2272 static void
2273 _base_build_nvme_prp(struct MPT3SAS_ADAPTER *ioc, u16 smid,
2274 	Mpi26NVMeEncapsulatedRequest_t *nvme_encap_request,
2275 	dma_addr_t data_out_dma, size_t data_out_sz, dma_addr_t data_in_dma,
2276 	size_t data_in_sz)
2277 {
2278 	int		prp_size = NVME_PRP_SIZE;
2279 	__le64		*prp_entry, *prp1_entry, *prp2_entry;
2280 	__le64		*prp_page;
2281 	dma_addr_t	prp_entry_dma, prp_page_dma, dma_addr;
2282 	u32		offset, entry_len;
2283 	u32		page_mask_result, page_mask;
2284 	size_t		length;
2285 	struct mpt3sas_nvme_cmd *nvme_cmd =
2286 		(void *)nvme_encap_request->NVMe_Command;
2287 
2288 	/*
2289 	 * Not all commands require a data transfer. If no data, just return
2290 	 * without constructing any PRP.
2291 	 */
2292 	if (!data_in_sz && !data_out_sz)
2293 		return;
2294 	prp1_entry = &nvme_cmd->prp1;
2295 	prp2_entry = &nvme_cmd->prp2;
2296 	prp_entry = prp1_entry;
2297 	/*
2298 	 * For the PRP entries, use the specially allocated buffer of
2299 	 * contiguous memory.
2300 	 */
2301 	prp_page = (__le64 *)mpt3sas_base_get_pcie_sgl(ioc, smid);
2302 	prp_page_dma = mpt3sas_base_get_pcie_sgl_dma(ioc, smid);
2303 
2304 	/*
2305 	 * Check if we are within 1 entry of a page boundary we don't
2306 	 * want our first entry to be a PRP List entry.
2307 	 */
2308 	page_mask = ioc->page_size - 1;
2309 	page_mask_result = (uintptr_t)((u8 *)prp_page + prp_size) & page_mask;
2310 	if (!page_mask_result) {
2311 		/* Bump up to next page boundary. */
2312 		prp_page = (__le64 *)((u8 *)prp_page + prp_size);
2313 		prp_page_dma = prp_page_dma + prp_size;
2314 	}
2315 
2316 	/*
2317 	 * Set PRP physical pointer, which initially points to the current PRP
2318 	 * DMA memory page.
2319 	 */
2320 	prp_entry_dma = prp_page_dma;
2321 
2322 	/* Get physical address and length of the data buffer. */
2323 	if (data_in_sz) {
2324 		dma_addr = data_in_dma;
2325 		length = data_in_sz;
2326 	} else {
2327 		dma_addr = data_out_dma;
2328 		length = data_out_sz;
2329 	}
2330 
2331 	/* Loop while the length is not zero. */
2332 	while (length) {
2333 		/*
2334 		 * Check if we need to put a list pointer here if we are at
2335 		 * page boundary - prp_size (8 bytes).
2336 		 */
2337 		page_mask_result = (prp_entry_dma + prp_size) & page_mask;
2338 		if (!page_mask_result) {
2339 			/*
2340 			 * This is the last entry in a PRP List, so we need to
2341 			 * put a PRP list pointer here.  What this does is:
2342 			 *   - bump the current memory pointer to the next
2343 			 *     address, which will be the next full page.
2344 			 *   - set the PRP Entry to point to that page.  This
2345 			 *     is now the PRP List pointer.
2346 			 *   - bump the PRP Entry pointer the start of the
2347 			 *     next page.  Since all of this PRP memory is
2348 			 *     contiguous, no need to get a new page - it's
2349 			 *     just the next address.
2350 			 */
2351 			prp_entry_dma++;
2352 			*prp_entry = cpu_to_le64(prp_entry_dma);
2353 			prp_entry++;
2354 		}
2355 
2356 		/* Need to handle if entry will be part of a page. */
2357 		offset = dma_addr & page_mask;
2358 		entry_len = ioc->page_size - offset;
2359 
2360 		if (prp_entry == prp1_entry) {
2361 			/*
2362 			 * Must fill in the first PRP pointer (PRP1) before
2363 			 * moving on.
2364 			 */
2365 			*prp1_entry = cpu_to_le64(dma_addr);
2366 
2367 			/*
2368 			 * Now point to the second PRP entry within the
2369 			 * command (PRP2).
2370 			 */
2371 			prp_entry = prp2_entry;
2372 		} else if (prp_entry == prp2_entry) {
2373 			/*
2374 			 * Should the PRP2 entry be a PRP List pointer or just
2375 			 * a regular PRP pointer?  If there is more than one
2376 			 * more page of data, must use a PRP List pointer.
2377 			 */
2378 			if (length > ioc->page_size) {
2379 				/*
2380 				 * PRP2 will contain a PRP List pointer because
2381 				 * more PRP's are needed with this command. The
2382 				 * list will start at the beginning of the
2383 				 * contiguous buffer.
2384 				 */
2385 				*prp2_entry = cpu_to_le64(prp_entry_dma);
2386 
2387 				/*
2388 				 * The next PRP Entry will be the start of the
2389 				 * first PRP List.
2390 				 */
2391 				prp_entry = prp_page;
2392 			} else {
2393 				/*
2394 				 * After this, the PRP Entries are complete.
2395 				 * This command uses 2 PRP's and no PRP list.
2396 				 */
2397 				*prp2_entry = cpu_to_le64(dma_addr);
2398 			}
2399 		} else {
2400 			/*
2401 			 * Put entry in list and bump the addresses.
2402 			 *
2403 			 * After PRP1 and PRP2 are filled in, this will fill in
2404 			 * all remaining PRP entries in a PRP List, one per
2405 			 * each time through the loop.
2406 			 */
2407 			*prp_entry = cpu_to_le64(dma_addr);
2408 			prp_entry++;
2409 			prp_entry_dma++;
2410 		}
2411 
2412 		/*
2413 		 * Bump the phys address of the command's data buffer by the
2414 		 * entry_len.
2415 		 */
2416 		dma_addr += entry_len;
2417 
2418 		/* Decrement length accounting for last partial page. */
2419 		if (entry_len > length)
2420 			length = 0;
2421 		else
2422 			length -= entry_len;
2423 	}
2424 }
2425 
2426 /**
2427  * base_make_prp_nvme - Prepare PRPs (Physical Region Page) -
2428  *			SGLs specific to NVMe drives only
2429  *
2430  * @ioc:		per adapter object
2431  * @scmd:		SCSI command from the mid-layer
2432  * @mpi_request:	mpi request
2433  * @smid:		msg Index
2434  * @sge_count:		scatter gather element count.
2435  *
2436  * Return:		true: PRPs are built
2437  *			false: IEEE SGLs needs to be built
2438  */
2439 static void
2440 base_make_prp_nvme(struct MPT3SAS_ADAPTER *ioc,
2441 		struct scsi_cmnd *scmd,
2442 		Mpi25SCSIIORequest_t *mpi_request,
2443 		u16 smid, int sge_count)
2444 {
2445 	int sge_len, num_prp_in_chain = 0;
2446 	Mpi25IeeeSgeChain64_t *main_chain_element, *ptr_first_sgl;
2447 	__le64 *curr_buff;
2448 	dma_addr_t msg_dma, sge_addr, offset;
2449 	u32 page_mask, page_mask_result;
2450 	struct scatterlist *sg_scmd;
2451 	u32 first_prp_len;
2452 	int data_len = scsi_bufflen(scmd);
2453 	u32 nvme_pg_size;
2454 
2455 	nvme_pg_size = max_t(u32, ioc->page_size, NVME_PRP_PAGE_SIZE);
2456 	/*
2457 	 * Nvme has a very convoluted prp format.  One prp is required
2458 	 * for each page or partial page. Driver need to split up OS sg_list
2459 	 * entries if it is longer than one page or cross a page
2460 	 * boundary.  Driver also have to insert a PRP list pointer entry as
2461 	 * the last entry in each physical page of the PRP list.
2462 	 *
2463 	 * NOTE: The first PRP "entry" is actually placed in the first
2464 	 * SGL entry in the main message as IEEE 64 format.  The 2nd
2465 	 * entry in the main message is the chain element, and the rest
2466 	 * of the PRP entries are built in the contiguous pcie buffer.
2467 	 */
2468 	page_mask = nvme_pg_size - 1;
2469 
2470 	/*
2471 	 * Native SGL is needed.
2472 	 * Put a chain element in main message frame that points to the first
2473 	 * chain buffer.
2474 	 *
2475 	 * NOTE:  The ChainOffset field must be 0 when using a chain pointer to
2476 	 *        a native SGL.
2477 	 */
2478 
2479 	/* Set main message chain element pointer */
2480 	main_chain_element = (pMpi25IeeeSgeChain64_t)&mpi_request->SGL;
2481 	/*
2482 	 * For NVMe the chain element needs to be the 2nd SG entry in the main
2483 	 * message.
2484 	 */
2485 	main_chain_element = (Mpi25IeeeSgeChain64_t *)
2486 		((u8 *)main_chain_element + sizeof(MPI25_IEEE_SGE_CHAIN64));
2487 
2488 	/*
2489 	 * For the PRP entries, use the specially allocated buffer of
2490 	 * contiguous memory.  Normal chain buffers can't be used
2491 	 * because each chain buffer would need to be the size of an OS
2492 	 * page (4k).
2493 	 */
2494 	curr_buff = mpt3sas_base_get_pcie_sgl(ioc, smid);
2495 	msg_dma = mpt3sas_base_get_pcie_sgl_dma(ioc, smid);
2496 
2497 	main_chain_element->Address = cpu_to_le64(msg_dma);
2498 	main_chain_element->NextChainOffset = 0;
2499 	main_chain_element->Flags = MPI2_IEEE_SGE_FLAGS_CHAIN_ELEMENT |
2500 			MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR |
2501 			MPI26_IEEE_SGE_FLAGS_NSF_NVME_PRP;
2502 
2503 	/* Build first prp, sge need not to be page aligned*/
2504 	ptr_first_sgl = (pMpi25IeeeSgeChain64_t)&mpi_request->SGL;
2505 	sg_scmd = scsi_sglist(scmd);
2506 	sge_addr = sg_dma_address(sg_scmd);
2507 	sge_len = sg_dma_len(sg_scmd);
2508 
2509 	offset = sge_addr & page_mask;
2510 	first_prp_len = nvme_pg_size - offset;
2511 
2512 	ptr_first_sgl->Address = cpu_to_le64(sge_addr);
2513 	ptr_first_sgl->Length = cpu_to_le32(first_prp_len);
2514 
2515 	data_len -= first_prp_len;
2516 
2517 	if (sge_len > first_prp_len) {
2518 		sge_addr += first_prp_len;
2519 		sge_len -= first_prp_len;
2520 	} else if (data_len && (sge_len == first_prp_len)) {
2521 		sg_scmd = sg_next(sg_scmd);
2522 		sge_addr = sg_dma_address(sg_scmd);
2523 		sge_len = sg_dma_len(sg_scmd);
2524 	}
2525 
2526 	for (;;) {
2527 		offset = sge_addr & page_mask;
2528 
2529 		/* Put PRP pointer due to page boundary*/
2530 		page_mask_result = (uintptr_t)(curr_buff + 1) & page_mask;
2531 		if (unlikely(!page_mask_result)) {
2532 			scmd_printk(KERN_NOTICE,
2533 				scmd, "page boundary curr_buff: 0x%p\n",
2534 				curr_buff);
2535 			msg_dma += 8;
2536 			*curr_buff = cpu_to_le64(msg_dma);
2537 			curr_buff++;
2538 			num_prp_in_chain++;
2539 		}
2540 
2541 		*curr_buff = cpu_to_le64(sge_addr);
2542 		curr_buff++;
2543 		msg_dma += 8;
2544 		num_prp_in_chain++;
2545 
2546 		sge_addr += nvme_pg_size;
2547 		sge_len -= nvme_pg_size;
2548 		data_len -= nvme_pg_size;
2549 
2550 		if (data_len <= 0)
2551 			break;
2552 
2553 		if (sge_len > 0)
2554 			continue;
2555 
2556 		sg_scmd = sg_next(sg_scmd);
2557 		sge_addr = sg_dma_address(sg_scmd);
2558 		sge_len = sg_dma_len(sg_scmd);
2559 	}
2560 
2561 	main_chain_element->Length =
2562 		cpu_to_le32(num_prp_in_chain * sizeof(u64));
2563 	return;
2564 }
2565 
2566 static bool
2567 base_is_prp_possible(struct MPT3SAS_ADAPTER *ioc,
2568 	struct _pcie_device *pcie_device, struct scsi_cmnd *scmd, int sge_count)
2569 {
2570 	u32 data_length = 0;
2571 	bool build_prp = true;
2572 
2573 	data_length = scsi_bufflen(scmd);
2574 	if (pcie_device &&
2575 	    (mpt3sas_scsih_is_pcie_scsi_device(pcie_device->device_info))) {
2576 		build_prp = false;
2577 		return build_prp;
2578 	}
2579 
2580 	/* If Datalenth is <= 16K and number of SGE’s entries are <= 2
2581 	 * we built IEEE SGL
2582 	 */
2583 	if ((data_length <= NVME_PRP_PAGE_SIZE*4) && (sge_count <= 2))
2584 		build_prp = false;
2585 
2586 	return build_prp;
2587 }
2588 
2589 /**
2590  * _base_check_pcie_native_sgl - This function is called for PCIe end devices to
2591  * determine if the driver needs to build a native SGL.  If so, that native
2592  * SGL is built in the special contiguous buffers allocated especially for
2593  * PCIe SGL creation.  If the driver will not build a native SGL, return
2594  * TRUE and a normal IEEE SGL will be built.  Currently this routine
2595  * supports NVMe.
2596  * @ioc: per adapter object
2597  * @mpi_request: mf request pointer
2598  * @smid: system request message index
2599  * @scmd: scsi command
2600  * @pcie_device: points to the PCIe device's info
2601  *
2602  * Return: 0 if native SGL was built, 1 if no SGL was built
2603  */
2604 static int
2605 _base_check_pcie_native_sgl(struct MPT3SAS_ADAPTER *ioc,
2606 	Mpi25SCSIIORequest_t *mpi_request, u16 smid, struct scsi_cmnd *scmd,
2607 	struct _pcie_device *pcie_device)
2608 {
2609 	int sges_left;
2610 
2611 	/* Get the SG list pointer and info. */
2612 	sges_left = scsi_dma_map(scmd);
2613 	if (sges_left < 0)
2614 		return 1;
2615 
2616 	/* Check if we need to build a native SG list. */
2617 	if (!base_is_prp_possible(ioc, pcie_device,
2618 				scmd, sges_left)) {
2619 		/* We built a native SG list, just return. */
2620 		goto out;
2621 	}
2622 
2623 	/*
2624 	 * Build native NVMe PRP.
2625 	 */
2626 	base_make_prp_nvme(ioc, scmd, mpi_request,
2627 			smid, sges_left);
2628 
2629 	return 0;
2630 out:
2631 	scsi_dma_unmap(scmd);
2632 	return 1;
2633 }
2634 
2635 /**
2636  * _base_add_sg_single_ieee - add sg element for IEEE format
2637  * @paddr: virtual address for SGE
2638  * @flags: SGE flags
2639  * @chain_offset: number of 128 byte elements from start of segment
2640  * @length: data transfer length
2641  * @dma_addr: Physical address
2642  */
2643 static void
2644 _base_add_sg_single_ieee(void *paddr, u8 flags, u8 chain_offset, u32 length,
2645 	dma_addr_t dma_addr)
2646 {
2647 	Mpi25IeeeSgeChain64_t *sgel = paddr;
2648 
2649 	sgel->Flags = flags;
2650 	sgel->NextChainOffset = chain_offset;
2651 	sgel->Length = cpu_to_le32(length);
2652 	sgel->Address = cpu_to_le64(dma_addr);
2653 }
2654 
2655 /**
2656  * _base_build_zero_len_sge_ieee - build zero length sg entry for IEEE format
2657  * @ioc: per adapter object
2658  * @paddr: virtual address for SGE
2659  *
2660  * Create a zero length scatter gather entry to insure the IOCs hardware has
2661  * something to use if the target device goes brain dead and tries
2662  * to send data even when none is asked for.
2663  */
2664 static void
2665 _base_build_zero_len_sge_ieee(struct MPT3SAS_ADAPTER *ioc, void *paddr)
2666 {
2667 	u8 sgl_flags = (MPI2_IEEE_SGE_FLAGS_SIMPLE_ELEMENT |
2668 		MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR |
2669 		MPI25_IEEE_SGE_FLAGS_END_OF_LIST);
2670 
2671 	_base_add_sg_single_ieee(paddr, sgl_flags, 0, 0, -1);
2672 }
2673 
2674 static inline int _base_scsi_dma_map(struct scsi_cmnd *cmd)
2675 {
2676 	/*
2677 	 * Some firmware versions byte-swap the REPORT ZONES command reply from
2678 	 * ATA-ZAC devices by directly accessing in the host buffer. This does
2679 	 * not respect the default command DMA direction and causes IOMMU page
2680 	 * faults on some architectures with an IOMMU enforcing write mappings
2681 	 * (e.g. AMD hosts). Avoid such issue by making the report zones buffer
2682 	 * mapping bi-directional.
2683 	 */
2684 	if (cmd->cmnd[0] == ZBC_IN && cmd->cmnd[1] == ZI_REPORT_ZONES)
2685 		cmd->sc_data_direction = DMA_BIDIRECTIONAL;
2686 
2687 	return scsi_dma_map(cmd);
2688 }
2689 
2690 /**
2691  * _base_build_sg_scmd - main sg creation routine
2692  *		pcie_device is unused here!
2693  * @ioc: per adapter object
2694  * @scmd: scsi command
2695  * @smid: system request message index
2696  * @unused: unused pcie_device pointer
2697  * Context: none.
2698  *
2699  * The main routine that builds scatter gather table from a given
2700  * scsi request sent via the .queuecommand main handler.
2701  *
2702  * Return: 0 success, anything else error
2703  */
2704 static int
2705 _base_build_sg_scmd(struct MPT3SAS_ADAPTER *ioc,
2706 	struct scsi_cmnd *scmd, u16 smid, struct _pcie_device *unused)
2707 {
2708 	Mpi2SCSIIORequest_t *mpi_request;
2709 	dma_addr_t chain_dma;
2710 	struct scatterlist *sg_scmd;
2711 	void *sg_local, *chain;
2712 	u32 chain_offset;
2713 	u32 chain_length;
2714 	u32 chain_flags;
2715 	int sges_left;
2716 	u32 sges_in_segment;
2717 	u32 sgl_flags;
2718 	u32 sgl_flags_last_element;
2719 	u32 sgl_flags_end_buffer;
2720 	struct chain_tracker *chain_req;
2721 
2722 	mpi_request = mpt3sas_base_get_msg_frame(ioc, smid);
2723 
2724 	/* init scatter gather flags */
2725 	sgl_flags = MPI2_SGE_FLAGS_SIMPLE_ELEMENT;
2726 	if (scmd->sc_data_direction == DMA_TO_DEVICE)
2727 		sgl_flags |= MPI2_SGE_FLAGS_HOST_TO_IOC;
2728 	sgl_flags_last_element = (sgl_flags | MPI2_SGE_FLAGS_LAST_ELEMENT)
2729 	    << MPI2_SGE_FLAGS_SHIFT;
2730 	sgl_flags_end_buffer = (sgl_flags | MPI2_SGE_FLAGS_LAST_ELEMENT |
2731 	    MPI2_SGE_FLAGS_END_OF_BUFFER | MPI2_SGE_FLAGS_END_OF_LIST)
2732 	    << MPI2_SGE_FLAGS_SHIFT;
2733 	sgl_flags = sgl_flags << MPI2_SGE_FLAGS_SHIFT;
2734 
2735 	sg_scmd = scsi_sglist(scmd);
2736 	sges_left = _base_scsi_dma_map(scmd);
2737 	if (sges_left < 0)
2738 		return -ENOMEM;
2739 
2740 	sg_local = &mpi_request->SGL;
2741 	sges_in_segment = ioc->max_sges_in_main_message;
2742 	if (sges_left <= sges_in_segment)
2743 		goto fill_in_last_segment;
2744 
2745 	mpi_request->ChainOffset = (offsetof(Mpi2SCSIIORequest_t, SGL) +
2746 	    (sges_in_segment * ioc->sge_size))/4;
2747 
2748 	/* fill in main message segment when there is a chain following */
2749 	while (sges_in_segment) {
2750 		if (sges_in_segment == 1)
2751 			ioc->base_add_sg_single(sg_local,
2752 			    sgl_flags_last_element | sg_dma_len(sg_scmd),
2753 			    sg_dma_address(sg_scmd));
2754 		else
2755 			ioc->base_add_sg_single(sg_local, sgl_flags |
2756 			    sg_dma_len(sg_scmd), sg_dma_address(sg_scmd));
2757 		sg_scmd = sg_next(sg_scmd);
2758 		sg_local += ioc->sge_size;
2759 		sges_left--;
2760 		sges_in_segment--;
2761 	}
2762 
2763 	/* initializing the chain flags and pointers */
2764 	chain_flags = MPI2_SGE_FLAGS_CHAIN_ELEMENT << MPI2_SGE_FLAGS_SHIFT;
2765 	chain_req = _base_get_chain_buffer_tracker(ioc, scmd);
2766 	if (!chain_req)
2767 		return -1;
2768 	chain = chain_req->chain_buffer;
2769 	chain_dma = chain_req->chain_buffer_dma;
2770 	do {
2771 		sges_in_segment = (sges_left <=
2772 		    ioc->max_sges_in_chain_message) ? sges_left :
2773 		    ioc->max_sges_in_chain_message;
2774 		chain_offset = (sges_left == sges_in_segment) ?
2775 		    0 : (sges_in_segment * ioc->sge_size)/4;
2776 		chain_length = sges_in_segment * ioc->sge_size;
2777 		if (chain_offset) {
2778 			chain_offset = chain_offset <<
2779 			    MPI2_SGE_CHAIN_OFFSET_SHIFT;
2780 			chain_length += ioc->sge_size;
2781 		}
2782 		ioc->base_add_sg_single(sg_local, chain_flags | chain_offset |
2783 		    chain_length, chain_dma);
2784 		sg_local = chain;
2785 		if (!chain_offset)
2786 			goto fill_in_last_segment;
2787 
2788 		/* fill in chain segments */
2789 		while (sges_in_segment) {
2790 			if (sges_in_segment == 1)
2791 				ioc->base_add_sg_single(sg_local,
2792 				    sgl_flags_last_element |
2793 				    sg_dma_len(sg_scmd),
2794 				    sg_dma_address(sg_scmd));
2795 			else
2796 				ioc->base_add_sg_single(sg_local, sgl_flags |
2797 				    sg_dma_len(sg_scmd),
2798 				    sg_dma_address(sg_scmd));
2799 			sg_scmd = sg_next(sg_scmd);
2800 			sg_local += ioc->sge_size;
2801 			sges_left--;
2802 			sges_in_segment--;
2803 		}
2804 
2805 		chain_req = _base_get_chain_buffer_tracker(ioc, scmd);
2806 		if (!chain_req)
2807 			return -1;
2808 		chain = chain_req->chain_buffer;
2809 		chain_dma = chain_req->chain_buffer_dma;
2810 	} while (1);
2811 
2812 
2813  fill_in_last_segment:
2814 
2815 	/* fill the last segment */
2816 	while (sges_left) {
2817 		if (sges_left == 1)
2818 			ioc->base_add_sg_single(sg_local, sgl_flags_end_buffer |
2819 			    sg_dma_len(sg_scmd), sg_dma_address(sg_scmd));
2820 		else
2821 			ioc->base_add_sg_single(sg_local, sgl_flags |
2822 			    sg_dma_len(sg_scmd), sg_dma_address(sg_scmd));
2823 		sg_scmd = sg_next(sg_scmd);
2824 		sg_local += ioc->sge_size;
2825 		sges_left--;
2826 	}
2827 
2828 	return 0;
2829 }
2830 
2831 /**
2832  * _base_build_sg_scmd_ieee - main sg creation routine for IEEE format
2833  * @ioc: per adapter object
2834  * @scmd: scsi command
2835  * @smid: system request message index
2836  * @pcie_device: Pointer to pcie_device. If set, the pcie native sgl will be
2837  * constructed on need.
2838  * Context: none.
2839  *
2840  * The main routine that builds scatter gather table from a given
2841  * scsi request sent via the .queuecommand main handler.
2842  *
2843  * Return: 0 success, anything else error
2844  */
2845 static int
2846 _base_build_sg_scmd_ieee(struct MPT3SAS_ADAPTER *ioc,
2847 	struct scsi_cmnd *scmd, u16 smid, struct _pcie_device *pcie_device)
2848 {
2849 	Mpi25SCSIIORequest_t *mpi_request;
2850 	dma_addr_t chain_dma;
2851 	struct scatterlist *sg_scmd;
2852 	void *sg_local, *chain;
2853 	u32 chain_offset;
2854 	u32 chain_length;
2855 	int sges_left;
2856 	u32 sges_in_segment;
2857 	u8 simple_sgl_flags;
2858 	u8 simple_sgl_flags_last;
2859 	u8 chain_sgl_flags;
2860 	struct chain_tracker *chain_req;
2861 
2862 	mpi_request = mpt3sas_base_get_msg_frame(ioc, smid);
2863 
2864 	/* init scatter gather flags */
2865 	simple_sgl_flags = MPI2_IEEE_SGE_FLAGS_SIMPLE_ELEMENT |
2866 	    MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR;
2867 	simple_sgl_flags_last = simple_sgl_flags |
2868 	    MPI25_IEEE_SGE_FLAGS_END_OF_LIST;
2869 	chain_sgl_flags = MPI2_IEEE_SGE_FLAGS_CHAIN_ELEMENT |
2870 	    MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR;
2871 
2872 	/* Check if we need to build a native SG list. */
2873 	if ((pcie_device) && (_base_check_pcie_native_sgl(ioc, mpi_request,
2874 			smid, scmd, pcie_device) == 0)) {
2875 		/* We built a native SG list, just return. */
2876 		return 0;
2877 	}
2878 
2879 	sg_scmd = scsi_sglist(scmd);
2880 	sges_left = _base_scsi_dma_map(scmd);
2881 	if (sges_left < 0)
2882 		return -ENOMEM;
2883 
2884 	sg_local = &mpi_request->SGL;
2885 	sges_in_segment = (ioc->request_sz -
2886 		   offsetof(Mpi25SCSIIORequest_t, SGL))/ioc->sge_size_ieee;
2887 	if (sges_left <= sges_in_segment)
2888 		goto fill_in_last_segment;
2889 
2890 	mpi_request->ChainOffset = (sges_in_segment - 1 /* chain element */) +
2891 	    (offsetof(Mpi25SCSIIORequest_t, SGL)/ioc->sge_size_ieee);
2892 
2893 	/* fill in main message segment when there is a chain following */
2894 	while (sges_in_segment > 1) {
2895 		_base_add_sg_single_ieee(sg_local, simple_sgl_flags, 0,
2896 		    sg_dma_len(sg_scmd), sg_dma_address(sg_scmd));
2897 		sg_scmd = sg_next(sg_scmd);
2898 		sg_local += ioc->sge_size_ieee;
2899 		sges_left--;
2900 		sges_in_segment--;
2901 	}
2902 
2903 	/* initializing the pointers */
2904 	chain_req = _base_get_chain_buffer_tracker(ioc, scmd);
2905 	if (!chain_req)
2906 		return -1;
2907 	chain = chain_req->chain_buffer;
2908 	chain_dma = chain_req->chain_buffer_dma;
2909 	do {
2910 		sges_in_segment = (sges_left <=
2911 		    ioc->max_sges_in_chain_message) ? sges_left :
2912 		    ioc->max_sges_in_chain_message;
2913 		chain_offset = (sges_left == sges_in_segment) ?
2914 		    0 : sges_in_segment;
2915 		chain_length = sges_in_segment * ioc->sge_size_ieee;
2916 		if (chain_offset)
2917 			chain_length += ioc->sge_size_ieee;
2918 		_base_add_sg_single_ieee(sg_local, chain_sgl_flags,
2919 		    chain_offset, chain_length, chain_dma);
2920 
2921 		sg_local = chain;
2922 		if (!chain_offset)
2923 			goto fill_in_last_segment;
2924 
2925 		/* fill in chain segments */
2926 		while (sges_in_segment) {
2927 			_base_add_sg_single_ieee(sg_local, simple_sgl_flags, 0,
2928 			    sg_dma_len(sg_scmd), sg_dma_address(sg_scmd));
2929 			sg_scmd = sg_next(sg_scmd);
2930 			sg_local += ioc->sge_size_ieee;
2931 			sges_left--;
2932 			sges_in_segment--;
2933 		}
2934 
2935 		chain_req = _base_get_chain_buffer_tracker(ioc, scmd);
2936 		if (!chain_req)
2937 			return -1;
2938 		chain = chain_req->chain_buffer;
2939 		chain_dma = chain_req->chain_buffer_dma;
2940 	} while (1);
2941 
2942 
2943  fill_in_last_segment:
2944 
2945 	/* fill the last segment */
2946 	while (sges_left > 0) {
2947 		if (sges_left == 1)
2948 			_base_add_sg_single_ieee(sg_local,
2949 			    simple_sgl_flags_last, 0, sg_dma_len(sg_scmd),
2950 			    sg_dma_address(sg_scmd));
2951 		else
2952 			_base_add_sg_single_ieee(sg_local, simple_sgl_flags, 0,
2953 			    sg_dma_len(sg_scmd), sg_dma_address(sg_scmd));
2954 		sg_scmd = sg_next(sg_scmd);
2955 		sg_local += ioc->sge_size_ieee;
2956 		sges_left--;
2957 	}
2958 
2959 	return 0;
2960 }
2961 
2962 /**
2963  * _base_build_sg_ieee - build generic sg for IEEE format
2964  * @ioc: per adapter object
2965  * @psge: virtual address for SGE
2966  * @data_out_dma: physical address for WRITES
2967  * @data_out_sz: data xfer size for WRITES
2968  * @data_in_dma: physical address for READS
2969  * @data_in_sz: data xfer size for READS
2970  */
2971 static void
2972 _base_build_sg_ieee(struct MPT3SAS_ADAPTER *ioc, void *psge,
2973 	dma_addr_t data_out_dma, size_t data_out_sz, dma_addr_t data_in_dma,
2974 	size_t data_in_sz)
2975 {
2976 	u8 sgl_flags;
2977 
2978 	if (!data_out_sz && !data_in_sz) {
2979 		_base_build_zero_len_sge_ieee(ioc, psge);
2980 		return;
2981 	}
2982 
2983 	if (data_out_sz && data_in_sz) {
2984 		/* WRITE sgel first */
2985 		sgl_flags = MPI2_IEEE_SGE_FLAGS_SIMPLE_ELEMENT |
2986 		    MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR;
2987 		_base_add_sg_single_ieee(psge, sgl_flags, 0, data_out_sz,
2988 		    data_out_dma);
2989 
2990 		/* incr sgel */
2991 		psge += ioc->sge_size_ieee;
2992 
2993 		/* READ sgel last */
2994 		sgl_flags |= MPI25_IEEE_SGE_FLAGS_END_OF_LIST;
2995 		_base_add_sg_single_ieee(psge, sgl_flags, 0, data_in_sz,
2996 		    data_in_dma);
2997 	} else if (data_out_sz) /* WRITE */ {
2998 		sgl_flags = MPI2_IEEE_SGE_FLAGS_SIMPLE_ELEMENT |
2999 		    MPI25_IEEE_SGE_FLAGS_END_OF_LIST |
3000 		    MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR;
3001 		_base_add_sg_single_ieee(psge, sgl_flags, 0, data_out_sz,
3002 		    data_out_dma);
3003 	} else if (data_in_sz) /* READ */ {
3004 		sgl_flags = MPI2_IEEE_SGE_FLAGS_SIMPLE_ELEMENT |
3005 		    MPI25_IEEE_SGE_FLAGS_END_OF_LIST |
3006 		    MPI2_IEEE_SGE_FLAGS_SYSTEM_ADDR;
3007 		_base_add_sg_single_ieee(psge, sgl_flags, 0, data_in_sz,
3008 		    data_in_dma);
3009 	}
3010 }
3011 
3012 #define convert_to_kb(x) ((x) << (PAGE_SHIFT - 10))
3013 
3014 /**
3015  * _base_config_dma_addressing - set dma addressing
3016  * @ioc: per adapter object
3017  * @pdev: PCI device struct
3018  *
3019  * Return: 0 for success, non-zero for failure.
3020  */
3021 static int
3022 _base_config_dma_addressing(struct MPT3SAS_ADAPTER *ioc, struct pci_dev *pdev)
3023 {
3024 	struct sysinfo s;
3025 	u64 coherent_dma_mask, dma_mask;
3026 
3027 	if (ioc->is_mcpu_endpoint || sizeof(dma_addr_t) == 4) {
3028 		ioc->dma_mask = 32;
3029 		coherent_dma_mask = dma_mask = DMA_BIT_MASK(32);
3030 	/* Set 63 bit DMA mask for all SAS3 and SAS35 controllers */
3031 	} else if (ioc->hba_mpi_version_belonged > MPI2_VERSION) {
3032 		ioc->dma_mask = 63;
3033 		coherent_dma_mask = dma_mask = DMA_BIT_MASK(63);
3034 	} else {
3035 		ioc->dma_mask = 64;
3036 		coherent_dma_mask = dma_mask = DMA_BIT_MASK(64);
3037 	}
3038 
3039 	if (ioc->use_32bit_dma)
3040 		coherent_dma_mask = DMA_BIT_MASK(32);
3041 
3042 	if (dma_set_mask(&pdev->dev, dma_mask) ||
3043 	    dma_set_coherent_mask(&pdev->dev, coherent_dma_mask))
3044 		return -ENODEV;
3045 
3046 	if (ioc->dma_mask > 32) {
3047 		ioc->base_add_sg_single = &_base_add_sg_single_64;
3048 		ioc->sge_size = sizeof(Mpi2SGESimple64_t);
3049 	} else {
3050 		ioc->base_add_sg_single = &_base_add_sg_single_32;
3051 		ioc->sge_size = sizeof(Mpi2SGESimple32_t);
3052 	}
3053 
3054 	si_meminfo(&s);
3055 	ioc_info(ioc, "%d BIT PCI BUS DMA ADDRESSING SUPPORTED, total mem (%ld kB)\n",
3056 		ioc->dma_mask, convert_to_kb(s.totalram));
3057 
3058 	return 0;
3059 }
3060 
3061 /**
3062  * _base_check_enable_msix - checks MSIX capabable.
3063  * @ioc: per adapter object
3064  *
3065  * Check to see if card is capable of MSIX, and set number
3066  * of available msix vectors
3067  */
3068 static int
3069 _base_check_enable_msix(struct MPT3SAS_ADAPTER *ioc)
3070 {
3071 	int base;
3072 	u16 message_control;
3073 
3074 	/* Check whether controller SAS2008 B0 controller,
3075 	 * if it is SAS2008 B0 controller use IO-APIC instead of MSIX
3076 	 */
3077 	if (ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2008 &&
3078 	    ioc->pdev->revision == SAS2_PCI_DEVICE_B0_REVISION) {
3079 		return -EINVAL;
3080 	}
3081 
3082 	base = pci_find_capability(ioc->pdev, PCI_CAP_ID_MSIX);
3083 	if (!base) {
3084 		dfailprintk(ioc, ioc_info(ioc, "msix not supported\n"));
3085 		return -EINVAL;
3086 	}
3087 
3088 	/* get msix vector count */
3089 	/* NUMA_IO not supported for older controllers */
3090 	if (ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2004 ||
3091 	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2008 ||
3092 	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2108_1 ||
3093 	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2108_2 ||
3094 	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2108_3 ||
3095 	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2116_1 ||
3096 	    ioc->pdev->device == MPI2_MFGPAGE_DEVID_SAS2116_2)
3097 		ioc->msix_vector_count = 1;
3098 	else {
3099 		pci_read_config_word(ioc->pdev, base + 2, &message_control);
3100 		ioc->msix_vector_count = (message_control & 0x3FF) + 1;
3101 	}
3102 	dinitprintk(ioc, ioc_info(ioc, "msix is supported, vector_count(%d)\n",
3103 				  ioc->msix_vector_count));
3104 	return 0;
3105 }
3106 
3107 /**
3108  * mpt3sas_base_free_irq - free irq
3109  * @ioc: per adapter object
3110  *
3111  * Freeing respective reply_queue from the list.
3112  */
3113 void
3114 mpt3sas_base_free_irq(struct MPT3SAS_ADAPTER *ioc)
3115 {
3116 	unsigned int irq;
3117 	struct adapter_reply_queue *reply_q, *next;
3118 
3119 	if (list_empty(&ioc->reply_queue_list))
3120 		return;
3121 
3122 	list_for_each_entry_safe(reply_q, next, &ioc->reply_queue_list, list) {
3123 		list_del(&reply_q->list);
3124 		if (reply_q->is_iouring_poll_q) {
3125 			kfree(reply_q);
3126 			continue;
3127 		}
3128 
3129 		if (ioc->smp_affinity_enable) {
3130 			irq = pci_irq_vector(ioc->pdev, reply_q->msix_index);
3131 			irq_update_affinity_hint(irq, NULL);
3132 		}
3133 		free_irq(pci_irq_vector(ioc->pdev, reply_q->msix_index),
3134 			 reply_q);
3135 		kfree(reply_q);
3136 	}
3137 }
3138 
3139 /**
3140  * _base_request_irq - request irq
3141  * @ioc: per adapter object
3142  * @index: msix index into vector table
3143  *
3144  * Inserting respective reply_queue into the list.
3145  */
3146 static int
3147 _base_request_irq(struct MPT3SAS_ADAPTER *ioc, u8 index)
3148 {
3149 	struct pci_dev *pdev = ioc->pdev;
3150 	struct adapter_reply_queue *reply_q;
3151 	int r, qid;
3152 
3153 	reply_q =  kzalloc(sizeof(struct adapter_reply_queue), GFP_KERNEL);
3154 	if (!reply_q) {
3155 		ioc_err(ioc, "unable to allocate memory %zu!\n",
3156 			sizeof(struct adapter_reply_queue));
3157 		return -ENOMEM;
3158 	}
3159 	reply_q->ioc = ioc;
3160 	reply_q->msix_index = index;
3161 
3162 	atomic_set(&reply_q->busy, 0);
3163 
3164 	if (index >= ioc->iopoll_q_start_index) {
3165 		qid = index - ioc->iopoll_q_start_index;
3166 		snprintf(reply_q->name, MPT_NAME_LENGTH, "%s%d-mq-poll%d",
3167 		    ioc->driver_name, ioc->id, qid);
3168 		reply_q->is_iouring_poll_q = 1;
3169 		ioc->io_uring_poll_queues[qid].reply_q = reply_q;
3170 		goto out;
3171 	}
3172 
3173 
3174 	if (ioc->msix_enable)
3175 		snprintf(reply_q->name, MPT_NAME_LENGTH, "%s%d-msix%d",
3176 		    ioc->driver_name, ioc->id, index);
3177 	else
3178 		snprintf(reply_q->name, MPT_NAME_LENGTH, "%s%d",
3179 		    ioc->driver_name, ioc->id);
3180 	r = request_irq(pci_irq_vector(pdev, index), _base_interrupt,
3181 			IRQF_SHARED, reply_q->name, reply_q);
3182 	if (r) {
3183 		pr_err("%s: unable to allocate interrupt %d!\n",
3184 		       reply_q->name, pci_irq_vector(pdev, index));
3185 		kfree(reply_q);
3186 		return -EBUSY;
3187 	}
3188 out:
3189 	INIT_LIST_HEAD(&reply_q->list);
3190 	list_add_tail(&reply_q->list, &ioc->reply_queue_list);
3191 	return 0;
3192 }
3193 
3194 /**
3195  * _base_assign_reply_queues - assigning msix index for each cpu
3196  * @ioc: per adapter object
3197  *
3198  * The enduser would need to set the affinity via /proc/irq/#/smp_affinity
3199  */
3200 static void
3201 _base_assign_reply_queues(struct MPT3SAS_ADAPTER *ioc)
3202 {
3203 	unsigned int cpu, nr_cpus, nr_msix, index = 0, irq;
3204 	struct adapter_reply_queue *reply_q;
3205 	int iopoll_q_count = ioc->reply_queue_count -
3206 	    ioc->iopoll_q_start_index;
3207 	const struct cpumask *mask;
3208 
3209 	if (!_base_is_controller_msix_enabled(ioc))
3210 		return;
3211 
3212 	if (ioc->msix_load_balance)
3213 		return;
3214 
3215 	memset(ioc->cpu_msix_table, 0, ioc->cpu_msix_table_sz);
3216 
3217 	nr_cpus = num_online_cpus();
3218 	nr_msix = ioc->reply_queue_count = min(ioc->reply_queue_count,
3219 					       ioc->facts.MaxMSIxVectors);
3220 	if (!nr_msix)
3221 		return;
3222 
3223 	if (ioc->smp_affinity_enable) {
3224 
3225 		/*
3226 		 * set irq affinity to local numa node for those irqs
3227 		 * corresponding to high iops queues.
3228 		 */
3229 		if (ioc->high_iops_queues) {
3230 			mask = cpumask_of_node(dev_to_node(&ioc->pdev->dev));
3231 			for (index = 0; index < ioc->high_iops_queues;
3232 			    index++) {
3233 				irq = pci_irq_vector(ioc->pdev, index);
3234 				irq_set_affinity_and_hint(irq, mask);
3235 			}
3236 		}
3237 
3238 		list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
3239 			const cpumask_t *mask;
3240 
3241 			if (reply_q->msix_index < ioc->high_iops_queues ||
3242 			    reply_q->msix_index >= ioc->iopoll_q_start_index)
3243 				continue;
3244 
3245 			mask = pci_irq_get_affinity(ioc->pdev,
3246 			    reply_q->msix_index);
3247 			if (!mask) {
3248 				ioc_warn(ioc, "no affinity for msi %x\n",
3249 					 reply_q->msix_index);
3250 				goto fall_back;
3251 			}
3252 
3253 			for_each_cpu_and(cpu, mask, cpu_online_mask) {
3254 				if (cpu >= ioc->cpu_msix_table_sz)
3255 					break;
3256 				ioc->cpu_msix_table[cpu] = reply_q->msix_index;
3257 			}
3258 		}
3259 		return;
3260 	}
3261 
3262 fall_back:
3263 	cpu = cpumask_first(cpu_online_mask);
3264 	nr_msix -= (ioc->high_iops_queues - iopoll_q_count);
3265 	index = 0;
3266 
3267 	list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
3268 		unsigned int i, group = nr_cpus / nr_msix;
3269 
3270 		if (reply_q->msix_index < ioc->high_iops_queues ||
3271 		    reply_q->msix_index >= ioc->iopoll_q_start_index)
3272 			continue;
3273 
3274 		if (cpu >= nr_cpus)
3275 			break;
3276 
3277 		if (index < nr_cpus % nr_msix)
3278 			group++;
3279 
3280 		for (i = 0 ; i < group ; i++) {
3281 			ioc->cpu_msix_table[cpu] = reply_q->msix_index;
3282 			cpu = cpumask_next(cpu, cpu_online_mask);
3283 		}
3284 		index++;
3285 	}
3286 }
3287 
3288 /**
3289  * _base_check_and_enable_high_iops_queues - enable high iops mode
3290  * @ioc: per adapter object
3291  * @hba_msix_vector_count: msix vectors supported by HBA
3292  *
3293  * Enable high iops queues only if
3294  *  - HBA is a SEA/AERO controller and
3295  *  - MSI-Xs vector supported by the HBA is 128 and
3296  *  - total CPU count in the system >=16 and
3297  *  - loaded driver with default max_msix_vectors module parameter and
3298  *  - system booted in non kdump mode
3299  *
3300  * Return: nothing.
3301  */
3302 static void
3303 _base_check_and_enable_high_iops_queues(struct MPT3SAS_ADAPTER *ioc,
3304 		int hba_msix_vector_count)
3305 {
3306 	u16 lnksta, speed;
3307 
3308 	/*
3309 	 * Disable high iops queues if io uring poll queues are enabled.
3310 	 */
3311 	if (perf_mode == MPT_PERF_MODE_IOPS ||
3312 	    perf_mode == MPT_PERF_MODE_LATENCY ||
3313 	    ioc->io_uring_poll_queues) {
3314 		ioc->high_iops_queues = 0;
3315 		return;
3316 	}
3317 
3318 	if (perf_mode == MPT_PERF_MODE_DEFAULT) {
3319 
3320 		pcie_capability_read_word(ioc->pdev, PCI_EXP_LNKSTA, &lnksta);
3321 		speed = lnksta & PCI_EXP_LNKSTA_CLS;
3322 
3323 		if (speed < 0x4) {
3324 			ioc->high_iops_queues = 0;
3325 			return;
3326 		}
3327 	}
3328 
3329 	if (!reset_devices && ioc->is_aero_ioc &&
3330 	    hba_msix_vector_count == MPT3SAS_GEN35_MAX_MSIX_QUEUES &&
3331 	    num_online_cpus() >= MPT3SAS_HIGH_IOPS_REPLY_QUEUES &&
3332 	    max_msix_vectors == -1)
3333 		ioc->high_iops_queues = MPT3SAS_HIGH_IOPS_REPLY_QUEUES;
3334 	else
3335 		ioc->high_iops_queues = 0;
3336 }
3337 
3338 /**
3339  * mpt3sas_base_disable_msix - disables msix
3340  * @ioc: per adapter object
3341  *
3342  */
3343 void
3344 mpt3sas_base_disable_msix(struct MPT3SAS_ADAPTER *ioc)
3345 {
3346 	if (!ioc->msix_enable)
3347 		return;
3348 	pci_free_irq_vectors(ioc->pdev);
3349 	ioc->msix_enable = 0;
3350 	kfree(ioc->io_uring_poll_queues);
3351 }
3352 
3353 /**
3354  * _base_alloc_irq_vectors - allocate msix vectors
3355  * @ioc: per adapter object
3356  *
3357  */
3358 static int
3359 _base_alloc_irq_vectors(struct MPT3SAS_ADAPTER *ioc)
3360 {
3361 	int i, irq_flags = PCI_IRQ_MSIX;
3362 	struct irq_affinity desc = { .pre_vectors = ioc->high_iops_queues };
3363 	struct irq_affinity *descp = &desc;
3364 	/*
3365 	 * Don't allocate msix vectors for poll_queues.
3366 	 * msix_vectors is always within a range of FW supported reply queue.
3367 	 */
3368 	int nr_msix_vectors = ioc->iopoll_q_start_index;
3369 
3370 
3371 	if (ioc->smp_affinity_enable)
3372 		irq_flags |= PCI_IRQ_AFFINITY | PCI_IRQ_ALL_TYPES;
3373 	else
3374 		descp = NULL;
3375 
3376 	ioc_info(ioc, " %d %d %d\n", ioc->high_iops_queues,
3377 	    ioc->reply_queue_count, nr_msix_vectors);
3378 
3379 	i = pci_alloc_irq_vectors_affinity(ioc->pdev,
3380 	    ioc->high_iops_queues,
3381 	    nr_msix_vectors, irq_flags, descp);
3382 
3383 	return i;
3384 }
3385 
3386 /**
3387  * _base_enable_msix - enables msix, failback to io_apic
3388  * @ioc: per adapter object
3389  *
3390  */
3391 static int
3392 _base_enable_msix(struct MPT3SAS_ADAPTER *ioc)
3393 {
3394 	int r;
3395 	int i, local_max_msix_vectors;
3396 	u8 try_msix = 0;
3397 	int iopoll_q_count = 0;
3398 
3399 	ioc->msix_load_balance = false;
3400 
3401 	if (msix_disable == -1 || msix_disable == 0)
3402 		try_msix = 1;
3403 
3404 	if (!try_msix)
3405 		goto try_ioapic;
3406 
3407 	if (_base_check_enable_msix(ioc) != 0)
3408 		goto try_ioapic;
3409 
3410 	ioc_info(ioc, "MSI-X vectors supported: %d\n", ioc->msix_vector_count);
3411 	pr_info("\t no of cores: %d, max_msix_vectors: %d\n",
3412 		ioc->cpu_count, max_msix_vectors);
3413 
3414 	ioc->reply_queue_count =
3415 		min_t(int, ioc->cpu_count, ioc->msix_vector_count);
3416 
3417 	if (!ioc->rdpq_array_enable && max_msix_vectors == -1)
3418 		local_max_msix_vectors = (reset_devices) ? 1 : 8;
3419 	else
3420 		local_max_msix_vectors = max_msix_vectors;
3421 
3422 	if (local_max_msix_vectors == 0)
3423 		goto try_ioapic;
3424 
3425 	/*
3426 	 * Enable msix_load_balance only if combined reply queue mode is
3427 	 * disabled on SAS3 & above generation HBA devices.
3428 	 */
3429 	if (!ioc->combined_reply_queue &&
3430 	    ioc->hba_mpi_version_belonged != MPI2_VERSION) {
3431 		ioc_info(ioc,
3432 		    "combined ReplyQueue is off, Enabling msix load balance\n");
3433 		ioc->msix_load_balance = true;
3434 	}
3435 
3436 	/*
3437 	 * smp affinity setting is not need when msix load balance
3438 	 * is enabled.
3439 	 */
3440 	if (ioc->msix_load_balance)
3441 		ioc->smp_affinity_enable = 0;
3442 
3443 	if (!ioc->smp_affinity_enable || ioc->reply_queue_count <= 1)
3444 		ioc->shost->host_tagset = 0;
3445 
3446 	/*
3447 	 * Enable io uring poll queues only if host_tagset is enabled.
3448 	 */
3449 	if (ioc->shost->host_tagset)
3450 		iopoll_q_count = poll_queues;
3451 
3452 	if (iopoll_q_count) {
3453 		ioc->io_uring_poll_queues = kcalloc(iopoll_q_count,
3454 		    sizeof(struct io_uring_poll_queue), GFP_KERNEL);
3455 		if (!ioc->io_uring_poll_queues)
3456 			iopoll_q_count = 0;
3457 	}
3458 
3459 	if (ioc->is_aero_ioc)
3460 		_base_check_and_enable_high_iops_queues(ioc,
3461 		    ioc->msix_vector_count);
3462 
3463 	/*
3464 	 * Add high iops queues count to reply queue count if high iops queues
3465 	 * are enabled.
3466 	 */
3467 	ioc->reply_queue_count = min_t(int,
3468 	    ioc->reply_queue_count + ioc->high_iops_queues,
3469 	    ioc->msix_vector_count);
3470 
3471 	/*
3472 	 * Adjust the reply queue count incase reply queue count
3473 	 * exceeds the user provided MSIx vectors count.
3474 	 */
3475 	if (local_max_msix_vectors > 0)
3476 		ioc->reply_queue_count = min_t(int, local_max_msix_vectors,
3477 		    ioc->reply_queue_count);
3478 	/*
3479 	 * Add io uring poll queues count to reply queues count
3480 	 * if io uring is enabled in driver.
3481 	 */
3482 	if (iopoll_q_count) {
3483 		if (ioc->reply_queue_count < (iopoll_q_count + MPT3_MIN_IRQS))
3484 			iopoll_q_count = 0;
3485 		ioc->reply_queue_count = min_t(int,
3486 		    ioc->reply_queue_count + iopoll_q_count,
3487 		    ioc->msix_vector_count);
3488 	}
3489 
3490 	/*
3491 	 * Starting index of io uring poll queues in reply queue list.
3492 	 */
3493 	ioc->iopoll_q_start_index =
3494 	    ioc->reply_queue_count - iopoll_q_count;
3495 
3496 	r = _base_alloc_irq_vectors(ioc);
3497 	if (r < 0) {
3498 		ioc_info(ioc, "pci_alloc_irq_vectors failed (r=%d) !!!\n", r);
3499 		goto try_ioapic;
3500 	}
3501 
3502 	/*
3503 	 * Adjust the reply queue count if the allocated
3504 	 * MSIx vectors is less then the requested number
3505 	 * of MSIx vectors.
3506 	 */
3507 	if (r < ioc->iopoll_q_start_index) {
3508 		ioc->reply_queue_count = r + iopoll_q_count;
3509 		ioc->iopoll_q_start_index =
3510 		    ioc->reply_queue_count - iopoll_q_count;
3511 	}
3512 
3513 	ioc->msix_enable = 1;
3514 	for (i = 0; i < ioc->reply_queue_count; i++) {
3515 		r = _base_request_irq(ioc, i);
3516 		if (r) {
3517 			mpt3sas_base_free_irq(ioc);
3518 			mpt3sas_base_disable_msix(ioc);
3519 			goto try_ioapic;
3520 		}
3521 	}
3522 
3523 	ioc_info(ioc, "High IOPs queues : %s\n",
3524 			ioc->high_iops_queues ? "enabled" : "disabled");
3525 
3526 	return 0;
3527 
3528 /* failback to io_apic interrupt routing */
3529  try_ioapic:
3530 	ioc->high_iops_queues = 0;
3531 	ioc_info(ioc, "High IOPs queues : disabled\n");
3532 	ioc->reply_queue_count = 1;
3533 	ioc->iopoll_q_start_index = ioc->reply_queue_count - 0;
3534 	r = pci_alloc_irq_vectors(ioc->pdev, 1, 1, PCI_IRQ_INTX);
3535 	if (r < 0) {
3536 		dfailprintk(ioc,
3537 			    ioc_info(ioc, "pci_alloc_irq_vector(legacy) failed (r=%d) !!!\n",
3538 				     r));
3539 	} else
3540 		r = _base_request_irq(ioc, 0);
3541 
3542 	return r;
3543 }
3544 
3545 /**
3546  * mpt3sas_base_unmap_resources - free controller resources
3547  * @ioc: per adapter object
3548  */
3549 static void
3550 mpt3sas_base_unmap_resources(struct MPT3SAS_ADAPTER *ioc)
3551 {
3552 	struct pci_dev *pdev = ioc->pdev;
3553 
3554 	dexitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
3555 
3556 	mpt3sas_base_free_irq(ioc);
3557 	mpt3sas_base_disable_msix(ioc);
3558 
3559 	kfree(ioc->replyPostRegisterIndex);
3560 	ioc->replyPostRegisterIndex = NULL;
3561 
3562 
3563 	if (ioc->chip_phys) {
3564 		iounmap(ioc->chip);
3565 		ioc->chip_phys = 0;
3566 	}
3567 
3568 	if (pci_is_enabled(pdev)) {
3569 		pci_release_selected_regions(ioc->pdev, ioc->bars);
3570 		pci_disable_device(pdev);
3571 	}
3572 }
3573 
3574 static int
3575 _base_diag_reset(struct MPT3SAS_ADAPTER *ioc);
3576 
3577 /**
3578  * mpt3sas_base_check_for_fault_and_issue_reset - check if IOC is in fault state
3579  *     and if it is in fault state then issue diag reset.
3580  * @ioc: per adapter object
3581  *
3582  * Return: 0 for success, non-zero for failure.
3583  */
3584 int
3585 mpt3sas_base_check_for_fault_and_issue_reset(struct MPT3SAS_ADAPTER *ioc)
3586 {
3587 	u32 ioc_state;
3588 	int rc = -EFAULT;
3589 
3590 	dinitprintk(ioc, pr_info("%s\n", __func__));
3591 	if (ioc->pci_error_recovery)
3592 		return 0;
3593 	ioc_state = mpt3sas_base_get_iocstate(ioc, 0);
3594 	dhsprintk(ioc, pr_info("%s: ioc_state(0x%08x)\n", __func__, ioc_state));
3595 
3596 	if ((ioc_state & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) {
3597 		mpt3sas_print_fault_code(ioc, ioc_state &
3598 		    MPI2_DOORBELL_DATA_MASK);
3599 		mpt3sas_base_mask_interrupts(ioc);
3600 		rc = _base_diag_reset(ioc);
3601 	} else if ((ioc_state & MPI2_IOC_STATE_MASK) ==
3602 	    MPI2_IOC_STATE_COREDUMP) {
3603 		mpt3sas_print_coredump_info(ioc, ioc_state &
3604 		     MPI2_DOORBELL_DATA_MASK);
3605 		mpt3sas_base_wait_for_coredump_completion(ioc, __func__);
3606 		mpt3sas_base_mask_interrupts(ioc);
3607 		rc = _base_diag_reset(ioc);
3608 	}
3609 
3610 	return rc;
3611 }
3612 
3613 /**
3614  * mpt3sas_base_map_resources - map in controller resources (io/irq/memap)
3615  * @ioc: per adapter object
3616  *
3617  * Return: 0 for success, non-zero for failure.
3618  */
3619 int
3620 mpt3sas_base_map_resources(struct MPT3SAS_ADAPTER *ioc)
3621 {
3622 	struct pci_dev *pdev = ioc->pdev;
3623 	u32 memap_sz;
3624 	u32 pio_sz;
3625 	int i, r = 0, rc;
3626 	u64 pio_chip = 0;
3627 	phys_addr_t chip_phys = 0;
3628 	struct adapter_reply_queue *reply_q;
3629 	int iopoll_q_count = 0;
3630 
3631 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
3632 
3633 	ioc->bars = pci_select_bars(pdev, IORESOURCE_MEM);
3634 	if (pci_enable_device_mem(pdev)) {
3635 		ioc_warn(ioc, "pci_enable_device_mem: failed\n");
3636 		ioc->bars = 0;
3637 		return -ENODEV;
3638 	}
3639 
3640 
3641 	if (pci_request_selected_regions(pdev, ioc->bars,
3642 	    ioc->driver_name)) {
3643 		ioc_warn(ioc, "pci_request_selected_regions: failed\n");
3644 		ioc->bars = 0;
3645 		r = -ENODEV;
3646 		goto out_fail;
3647 	}
3648 
3649 	pci_set_master(pdev);
3650 
3651 
3652 	if (_base_config_dma_addressing(ioc, pdev) != 0) {
3653 		ioc_warn(ioc, "no suitable DMA mask for %s\n", pci_name(pdev));
3654 		r = -ENODEV;
3655 		goto out_fail;
3656 	}
3657 
3658 	for (i = 0, memap_sz = 0, pio_sz = 0; (i < DEVICE_COUNT_RESOURCE) &&
3659 	     (!memap_sz || !pio_sz); i++) {
3660 		if (pci_resource_flags(pdev, i) & IORESOURCE_IO) {
3661 			if (pio_sz)
3662 				continue;
3663 			pio_chip = (u64)pci_resource_start(pdev, i);
3664 			pio_sz = pci_resource_len(pdev, i);
3665 		} else if (pci_resource_flags(pdev, i) & IORESOURCE_MEM) {
3666 			if (memap_sz)
3667 				continue;
3668 			ioc->chip_phys = pci_resource_start(pdev, i);
3669 			chip_phys = ioc->chip_phys;
3670 			memap_sz = pci_resource_len(pdev, i);
3671 			ioc->chip = ioremap(ioc->chip_phys, memap_sz);
3672 		}
3673 	}
3674 
3675 	if (ioc->chip == NULL) {
3676 		ioc_err(ioc,
3677 		    "unable to map adapter memory! or resource not found\n");
3678 		r = -EINVAL;
3679 		goto out_fail;
3680 	}
3681 
3682 	mpt3sas_base_mask_interrupts(ioc);
3683 
3684 	r = _base_get_ioc_facts(ioc);
3685 	if (r) {
3686 		rc = mpt3sas_base_check_for_fault_and_issue_reset(ioc);
3687 		if (rc || (_base_get_ioc_facts(ioc)))
3688 			goto out_fail;
3689 	}
3690 
3691 	if (!ioc->rdpq_array_enable_assigned) {
3692 		ioc->rdpq_array_enable = ioc->rdpq_array_capable;
3693 		ioc->rdpq_array_enable_assigned = 1;
3694 	}
3695 
3696 	r = _base_enable_msix(ioc);
3697 	if (r)
3698 		goto out_fail;
3699 
3700 	iopoll_q_count = ioc->reply_queue_count - ioc->iopoll_q_start_index;
3701 	for (i = 0; i < iopoll_q_count; i++) {
3702 		atomic_set(&ioc->io_uring_poll_queues[i].busy, 0);
3703 		atomic_set(&ioc->io_uring_poll_queues[i].pause, 0);
3704 	}
3705 
3706 	if (!ioc->is_driver_loading)
3707 		_base_init_irqpolls(ioc);
3708 	/* Use the Combined reply queue feature only for SAS3 C0 & higher
3709 	 * revision HBAs and also only when reply queue count is greater than 8
3710 	 */
3711 	if (ioc->combined_reply_queue) {
3712 		/* Determine the Supplemental Reply Post Host Index Registers
3713 		 * Addresse. Supplemental Reply Post Host Index Registers
3714 		 * starts at offset MPI25_SUP_REPLY_POST_HOST_INDEX_OFFSET and
3715 		 * each register is at offset bytes of
3716 		 * MPT3_SUP_REPLY_POST_HOST_INDEX_REG_OFFSET from previous one.
3717 		 */
3718 		ioc->replyPostRegisterIndex = kcalloc(
3719 		     ioc->combined_reply_index_count,
3720 		     sizeof(resource_size_t *), GFP_KERNEL);
3721 		if (!ioc->replyPostRegisterIndex) {
3722 			ioc_err(ioc,
3723 			    "allocation for replyPostRegisterIndex failed!\n");
3724 			r = -ENOMEM;
3725 			goto out_fail;
3726 		}
3727 
3728 		for (i = 0; i < ioc->combined_reply_index_count; i++) {
3729 			ioc->replyPostRegisterIndex[i] =
3730 				(resource_size_t __iomem *)
3731 				((u8 __force *)&ioc->chip->Doorbell +
3732 				 MPI25_SUP_REPLY_POST_HOST_INDEX_OFFSET +
3733 				 (i * MPT3_SUP_REPLY_POST_HOST_INDEX_REG_OFFSET));
3734 		}
3735 	}
3736 
3737 	if (ioc->is_warpdrive) {
3738 		ioc->reply_post_host_index[0] = (resource_size_t __iomem *)
3739 		    &ioc->chip->ReplyPostHostIndex;
3740 
3741 		for (i = 1; i < ioc->cpu_msix_table_sz; i++)
3742 			ioc->reply_post_host_index[i] =
3743 			(resource_size_t __iomem *)
3744 			((u8 __iomem *)&ioc->chip->Doorbell + (0x4000 + ((i - 1)
3745 			* 4)));
3746 	}
3747 
3748 	list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
3749 		if (reply_q->msix_index >= ioc->iopoll_q_start_index) {
3750 			pr_info("%s: enabled: index: %d\n",
3751 			    reply_q->name, reply_q->msix_index);
3752 			continue;
3753 		}
3754 
3755 		pr_info("%s: %s enabled: IRQ %d\n",
3756 			reply_q->name,
3757 			ioc->msix_enable ? "PCI-MSI-X" : "IO-APIC",
3758 			pci_irq_vector(ioc->pdev, reply_q->msix_index));
3759 	}
3760 
3761 	ioc_info(ioc, "iomem(%pap), mapped(0x%p), size(%d)\n",
3762 		 &chip_phys, ioc->chip, memap_sz);
3763 	ioc_info(ioc, "ioport(0x%016llx), size(%d)\n",
3764 		 (unsigned long long)pio_chip, pio_sz);
3765 
3766 	/* Save PCI configuration state for recovery from PCI AER/EEH errors */
3767 	pci_save_state(pdev);
3768 	return 0;
3769 
3770  out_fail:
3771 	mpt3sas_base_unmap_resources(ioc);
3772 	return r;
3773 }
3774 
3775 /**
3776  * mpt3sas_base_get_msg_frame - obtain request mf pointer
3777  * @ioc: per adapter object
3778  * @smid: system request message index(smid zero is invalid)
3779  *
3780  * Return: virt pointer to message frame.
3781  */
3782 void *
3783 mpt3sas_base_get_msg_frame(struct MPT3SAS_ADAPTER *ioc, u16 smid)
3784 {
3785 	return (void *)(ioc->request + (smid * ioc->request_sz));
3786 }
3787 
3788 /**
3789  * mpt3sas_base_get_sense_buffer - obtain a sense buffer virt addr
3790  * @ioc: per adapter object
3791  * @smid: system request message index
3792  *
3793  * Return: virt pointer to sense buffer.
3794  */
3795 void *
3796 mpt3sas_base_get_sense_buffer(struct MPT3SAS_ADAPTER *ioc, u16 smid)
3797 {
3798 	return (void *)(ioc->sense + ((smid - 1) * SCSI_SENSE_BUFFERSIZE));
3799 }
3800 
3801 /**
3802  * mpt3sas_base_get_sense_buffer_dma - obtain a sense buffer dma addr
3803  * @ioc: per adapter object
3804  * @smid: system request message index
3805  *
3806  * Return: phys pointer to the low 32bit address of the sense buffer.
3807  */
3808 __le32
3809 mpt3sas_base_get_sense_buffer_dma(struct MPT3SAS_ADAPTER *ioc, u16 smid)
3810 {
3811 	return cpu_to_le32(ioc->sense_dma + ((smid - 1) *
3812 	    SCSI_SENSE_BUFFERSIZE));
3813 }
3814 
3815 /**
3816  * mpt3sas_base_get_pcie_sgl - obtain a PCIe SGL virt addr
3817  * @ioc: per adapter object
3818  * @smid: system request message index
3819  *
3820  * Return: virt pointer to a PCIe SGL.
3821  */
3822 void *
3823 mpt3sas_base_get_pcie_sgl(struct MPT3SAS_ADAPTER *ioc, u16 smid)
3824 {
3825 	return (void *)(ioc->pcie_sg_lookup[smid - 1].pcie_sgl);
3826 }
3827 
3828 /**
3829  * mpt3sas_base_get_pcie_sgl_dma - obtain a PCIe SGL dma addr
3830  * @ioc: per adapter object
3831  * @smid: system request message index
3832  *
3833  * Return: phys pointer to the address of the PCIe buffer.
3834  */
3835 dma_addr_t
3836 mpt3sas_base_get_pcie_sgl_dma(struct MPT3SAS_ADAPTER *ioc, u16 smid)
3837 {
3838 	return ioc->pcie_sg_lookup[smid - 1].pcie_sgl_dma;
3839 }
3840 
3841 /**
3842  * mpt3sas_base_get_reply_virt_addr - obtain reply frames virt address
3843  * @ioc: per adapter object
3844  * @phys_addr: lower 32 physical addr of the reply
3845  *
3846  * Converts 32bit lower physical addr into a virt address.
3847  */
3848 void *
3849 mpt3sas_base_get_reply_virt_addr(struct MPT3SAS_ADAPTER *ioc, u32 phys_addr)
3850 {
3851 	if (!phys_addr)
3852 		return NULL;
3853 	return ioc->reply + (phys_addr - (u32)ioc->reply_dma);
3854 }
3855 
3856 /**
3857  * _base_get_msix_index - get the msix index
3858  * @ioc: per adapter object
3859  * @scmd: scsi_cmnd object
3860  *
3861  * Return: msix index of general reply queues,
3862  * i.e. reply queue on which IO request's reply
3863  * should be posted by the HBA firmware.
3864  */
3865 static inline u8
3866 _base_get_msix_index(struct MPT3SAS_ADAPTER *ioc,
3867 	struct scsi_cmnd *scmd)
3868 {
3869 	/* Enables reply_queue load balancing */
3870 	if (ioc->msix_load_balance)
3871 		return ioc->reply_queue_count ?
3872 		    base_mod64(atomic64_add_return(1,
3873 		    &ioc->total_io_cnt), ioc->reply_queue_count) : 0;
3874 
3875 	if (scmd && ioc->shost->nr_hw_queues > 1) {
3876 		u32 tag = blk_mq_unique_tag(scsi_cmd_to_rq(scmd));
3877 
3878 		return blk_mq_unique_tag_to_hwq(tag) +
3879 			ioc->high_iops_queues;
3880 	}
3881 
3882 	return ioc->cpu_msix_table[raw_smp_processor_id()];
3883 }
3884 
3885 /**
3886  * _base_get_high_iops_msix_index - get the msix index of
3887  *				high iops queues
3888  * @ioc: per adapter object
3889  * @scmd: scsi_cmnd object
3890  *
3891  * Return: msix index of high iops reply queues.
3892  * i.e. high iops reply queue on which IO request's
3893  * reply should be posted by the HBA firmware.
3894  */
3895 static inline u8
3896 _base_get_high_iops_msix_index(struct MPT3SAS_ADAPTER *ioc,
3897 	struct scsi_cmnd *scmd)
3898 {
3899 	/**
3900 	 * Round robin the IO interrupts among the high iops
3901 	 * reply queues in terms of batch count 16 when outstanding
3902 	 * IOs on the target device is >=8.
3903 	 */
3904 
3905 	if (scsi_device_busy(scmd->device) > MPT3SAS_DEVICE_HIGH_IOPS_DEPTH)
3906 		return base_mod64((
3907 		    atomic64_add_return(1, &ioc->high_iops_outstanding) /
3908 		    MPT3SAS_HIGH_IOPS_BATCH_COUNT),
3909 		    MPT3SAS_HIGH_IOPS_REPLY_QUEUES);
3910 
3911 	return _base_get_msix_index(ioc, scmd);
3912 }
3913 
3914 /**
3915  * mpt3sas_base_get_smid - obtain a free smid from internal queue
3916  * @ioc: per adapter object
3917  * @cb_idx: callback index
3918  *
3919  * Return: smid (zero is invalid)
3920  */
3921 u16
3922 mpt3sas_base_get_smid(struct MPT3SAS_ADAPTER *ioc, u8 cb_idx)
3923 {
3924 	unsigned long flags;
3925 	struct request_tracker *request;
3926 	u16 smid;
3927 
3928 	spin_lock_irqsave(&ioc->scsi_lookup_lock, flags);
3929 	if (list_empty(&ioc->internal_free_list)) {
3930 		spin_unlock_irqrestore(&ioc->scsi_lookup_lock, flags);
3931 		ioc_err(ioc, "%s: smid not available\n", __func__);
3932 		return 0;
3933 	}
3934 
3935 	request = list_entry(ioc->internal_free_list.next,
3936 	    struct request_tracker, tracker_list);
3937 	request->cb_idx = cb_idx;
3938 	smid = request->smid;
3939 	list_del(&request->tracker_list);
3940 	spin_unlock_irqrestore(&ioc->scsi_lookup_lock, flags);
3941 	return smid;
3942 }
3943 
3944 /**
3945  * mpt3sas_base_get_smid_scsiio - obtain a free smid from scsiio queue
3946  * @ioc: per adapter object
3947  * @cb_idx: callback index
3948  * @scmd: pointer to scsi command object
3949  *
3950  * Return: smid (zero is invalid)
3951  */
3952 u16
3953 mpt3sas_base_get_smid_scsiio(struct MPT3SAS_ADAPTER *ioc, u8 cb_idx,
3954 	struct scsi_cmnd *scmd)
3955 {
3956 	struct scsiio_tracker *request = scsi_cmd_priv(scmd);
3957 	u16 smid;
3958 	u32 tag, unique_tag;
3959 
3960 	unique_tag = blk_mq_unique_tag(scsi_cmd_to_rq(scmd));
3961 	tag = blk_mq_unique_tag_to_tag(unique_tag);
3962 
3963 	/*
3964 	 * Store hw queue number corresponding to the tag.
3965 	 * This hw queue number is used later to determine
3966 	 * the unique_tag using the logic below. This unique_tag
3967 	 * is used to retrieve the scmd pointer corresponding
3968 	 * to tag using scsi_host_find_tag() API.
3969 	 *
3970 	 * tag = smid - 1;
3971 	 * unique_tag = ioc->io_queue_num[tag] << BLK_MQ_UNIQUE_TAG_BITS | tag;
3972 	 */
3973 	ioc->io_queue_num[tag] = blk_mq_unique_tag_to_hwq(unique_tag);
3974 
3975 	smid = tag + 1;
3976 	request->cb_idx = cb_idx;
3977 	request->smid = smid;
3978 	request->scmd = scmd;
3979 	INIT_LIST_HEAD(&request->chain_list);
3980 	return smid;
3981 }
3982 
3983 /**
3984  * mpt3sas_base_get_smid_hpr - obtain a free smid from hi-priority queue
3985  * @ioc: per adapter object
3986  * @cb_idx: callback index
3987  *
3988  * Return: smid (zero is invalid)
3989  */
3990 u16
3991 mpt3sas_base_get_smid_hpr(struct MPT3SAS_ADAPTER *ioc, u8 cb_idx)
3992 {
3993 	unsigned long flags;
3994 	struct request_tracker *request;
3995 	u16 smid;
3996 
3997 	spin_lock_irqsave(&ioc->scsi_lookup_lock, flags);
3998 	if (list_empty(&ioc->hpr_free_list)) {
3999 		spin_unlock_irqrestore(&ioc->scsi_lookup_lock, flags);
4000 		return 0;
4001 	}
4002 
4003 	request = list_entry(ioc->hpr_free_list.next,
4004 	    struct request_tracker, tracker_list);
4005 	request->cb_idx = cb_idx;
4006 	smid = request->smid;
4007 	list_del(&request->tracker_list);
4008 	spin_unlock_irqrestore(&ioc->scsi_lookup_lock, flags);
4009 	return smid;
4010 }
4011 
4012 static void
4013 _base_recovery_check(struct MPT3SAS_ADAPTER *ioc)
4014 {
4015 	/*
4016 	 * See _wait_for_commands_to_complete() call with regards to this code.
4017 	 */
4018 	if (ioc->shost_recovery && ioc->pending_io_count) {
4019 		ioc->pending_io_count = scsi_host_busy(ioc->shost);
4020 		if (ioc->pending_io_count == 0)
4021 			wake_up(&ioc->reset_wq);
4022 	}
4023 }
4024 
4025 void mpt3sas_base_clear_st(struct MPT3SAS_ADAPTER *ioc,
4026 			   struct scsiio_tracker *st)
4027 {
4028 	if (WARN_ON(st->smid == 0))
4029 		return;
4030 	st->cb_idx = 0xFF;
4031 	st->direct_io = 0;
4032 	st->scmd = NULL;
4033 	atomic_set(&ioc->chain_lookup[st->smid - 1].chain_offset, 0);
4034 	st->smid = 0;
4035 }
4036 
4037 /**
4038  * mpt3sas_base_free_smid - put smid back on free_list
4039  * @ioc: per adapter object
4040  * @smid: system request message index
4041  */
4042 void
4043 mpt3sas_base_free_smid(struct MPT3SAS_ADAPTER *ioc, u16 smid)
4044 {
4045 	unsigned long flags;
4046 	int i;
4047 
4048 	if (smid < ioc->hi_priority_smid) {
4049 		struct scsiio_tracker *st;
4050 		void *request;
4051 
4052 		st = _get_st_from_smid(ioc, smid);
4053 		if (!st) {
4054 			_base_recovery_check(ioc);
4055 			return;
4056 		}
4057 
4058 		/* Clear MPI request frame */
4059 		request = mpt3sas_base_get_msg_frame(ioc, smid);
4060 		memset(request, 0, ioc->request_sz);
4061 
4062 		mpt3sas_base_clear_st(ioc, st);
4063 		_base_recovery_check(ioc);
4064 		ioc->io_queue_num[smid - 1] = 0;
4065 		return;
4066 	}
4067 
4068 	spin_lock_irqsave(&ioc->scsi_lookup_lock, flags);
4069 	if (smid < ioc->internal_smid) {
4070 		/* hi-priority */
4071 		i = smid - ioc->hi_priority_smid;
4072 		ioc->hpr_lookup[i].cb_idx = 0xFF;
4073 		list_add(&ioc->hpr_lookup[i].tracker_list, &ioc->hpr_free_list);
4074 	} else if (smid <= ioc->hba_queue_depth) {
4075 		/* internal queue */
4076 		i = smid - ioc->internal_smid;
4077 		ioc->internal_lookup[i].cb_idx = 0xFF;
4078 		list_add(&ioc->internal_lookup[i].tracker_list,
4079 		    &ioc->internal_free_list);
4080 	}
4081 	spin_unlock_irqrestore(&ioc->scsi_lookup_lock, flags);
4082 }
4083 
4084 /**
4085  * _base_mpi_ep_writeq - 32 bit write to MMIO
4086  * @b: data payload
4087  * @addr: address in MMIO space
4088  * @writeq_lock: spin lock
4089  *
4090  * This special handling for MPI EP to take care of 32 bit
4091  * environment where its not quarenteed to send the entire word
4092  * in one transfer.
4093  */
4094 static inline void
4095 _base_mpi_ep_writeq(__u64 b, volatile void __iomem *addr,
4096 					spinlock_t *writeq_lock)
4097 {
4098 	unsigned long flags;
4099 
4100 	spin_lock_irqsave(writeq_lock, flags);
4101 	__raw_writel((u32)(b), addr);
4102 	__raw_writel((u32)(b >> 32), (addr + 4));
4103 	spin_unlock_irqrestore(writeq_lock, flags);
4104 }
4105 
4106 /**
4107  * _base_writeq - 64 bit write to MMIO
4108  * @b: data payload
4109  * @addr: address in MMIO space
4110  * @writeq_lock: spin lock
4111  *
4112  * Glue for handling an atomic 64 bit word to MMIO. This special handling takes
4113  * care of 32 bit environment where its not quarenteed to send the entire word
4114  * in one transfer.
4115  */
4116 #if defined(writeq) && defined(CONFIG_64BIT)
4117 static inline void
4118 _base_writeq(__u64 b, volatile void __iomem *addr, spinlock_t *writeq_lock)
4119 {
4120 	wmb();
4121 	__raw_writeq(b, addr);
4122 	barrier();
4123 }
4124 #else
4125 static inline void
4126 _base_writeq(__u64 b, volatile void __iomem *addr, spinlock_t *writeq_lock)
4127 {
4128 	_base_mpi_ep_writeq(b, addr, writeq_lock);
4129 }
4130 #endif
4131 
4132 /**
4133  * _base_set_and_get_msix_index - get the msix index and assign to msix_io
4134  *                                variable of scsi tracker
4135  * @ioc: per adapter object
4136  * @smid: system request message index
4137  *
4138  * Return: msix index.
4139  */
4140 static u8
4141 _base_set_and_get_msix_index(struct MPT3SAS_ADAPTER *ioc, u16 smid)
4142 {
4143 	struct scsiio_tracker *st = NULL;
4144 
4145 	if (smid < ioc->hi_priority_smid)
4146 		st = _get_st_from_smid(ioc, smid);
4147 
4148 	if (st == NULL)
4149 		return  _base_get_msix_index(ioc, NULL);
4150 
4151 	st->msix_io = ioc->get_msix_index_for_smlio(ioc, st->scmd);
4152 	return st->msix_io;
4153 }
4154 
4155 /**
4156  * _base_put_smid_mpi_ep_scsi_io - send SCSI_IO request to firmware
4157  * @ioc: per adapter object
4158  * @smid: system request message index
4159  * @handle: device handle
4160  */
4161 static void
4162 _base_put_smid_mpi_ep_scsi_io(struct MPT3SAS_ADAPTER *ioc,
4163 	u16 smid, u16 handle)
4164 {
4165 	Mpi2RequestDescriptorUnion_t descriptor;
4166 	u64 *request = (u64 *)&descriptor;
4167 	void *mpi_req_iomem;
4168 	__le32 *mfp = (__le32 *)mpt3sas_base_get_msg_frame(ioc, smid);
4169 
4170 	_clone_sg_entries(ioc, (void *) mfp, smid);
4171 	mpi_req_iomem = (void __force *)ioc->chip +
4172 			MPI_FRAME_START_OFFSET + (smid * ioc->request_sz);
4173 	_base_clone_mpi_to_sys_mem(mpi_req_iomem, (void *)mfp,
4174 					ioc->request_sz);
4175 	descriptor.SCSIIO.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_SCSI_IO;
4176 	descriptor.SCSIIO.MSIxIndex = _base_set_and_get_msix_index(ioc, smid);
4177 	descriptor.SCSIIO.SMID = cpu_to_le16(smid);
4178 	descriptor.SCSIIO.DevHandle = cpu_to_le16(handle);
4179 	descriptor.SCSIIO.LMID = 0;
4180 	_base_mpi_ep_writeq(*request, &ioc->chip->RequestDescriptorPostLow,
4181 	    &ioc->scsi_lookup_lock);
4182 }
4183 
4184 /**
4185  * _base_put_smid_scsi_io - send SCSI_IO request to firmware
4186  * @ioc: per adapter object
4187  * @smid: system request message index
4188  * @handle: device handle
4189  */
4190 static void
4191 _base_put_smid_scsi_io(struct MPT3SAS_ADAPTER *ioc, u16 smid, u16 handle)
4192 {
4193 	Mpi2RequestDescriptorUnion_t descriptor;
4194 	u64 *request = (u64 *)&descriptor;
4195 
4196 
4197 	descriptor.SCSIIO.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_SCSI_IO;
4198 	descriptor.SCSIIO.MSIxIndex = _base_set_and_get_msix_index(ioc, smid);
4199 	descriptor.SCSIIO.SMID = cpu_to_le16(smid);
4200 	descriptor.SCSIIO.DevHandle = cpu_to_le16(handle);
4201 	descriptor.SCSIIO.LMID = 0;
4202 	_base_writeq(*request, &ioc->chip->RequestDescriptorPostLow,
4203 	    &ioc->scsi_lookup_lock);
4204 }
4205 
4206 /**
4207  * _base_put_smid_fast_path - send fast path request to firmware
4208  * @ioc: per adapter object
4209  * @smid: system request message index
4210  * @handle: device handle
4211  */
4212 static void
4213 _base_put_smid_fast_path(struct MPT3SAS_ADAPTER *ioc, u16 smid,
4214 	u16 handle)
4215 {
4216 	Mpi2RequestDescriptorUnion_t descriptor;
4217 	u64 *request = (u64 *)&descriptor;
4218 
4219 	descriptor.SCSIIO.RequestFlags =
4220 	    MPI25_REQ_DESCRIPT_FLAGS_FAST_PATH_SCSI_IO;
4221 	descriptor.SCSIIO.MSIxIndex = _base_set_and_get_msix_index(ioc, smid);
4222 	descriptor.SCSIIO.SMID = cpu_to_le16(smid);
4223 	descriptor.SCSIIO.DevHandle = cpu_to_le16(handle);
4224 	descriptor.SCSIIO.LMID = 0;
4225 	_base_writeq(*request, &ioc->chip->RequestDescriptorPostLow,
4226 	    &ioc->scsi_lookup_lock);
4227 }
4228 
4229 /**
4230  * _base_put_smid_hi_priority - send Task Management request to firmware
4231  * @ioc: per adapter object
4232  * @smid: system request message index
4233  * @msix_task: msix_task will be same as msix of IO in case of task abort else 0
4234  */
4235 static void
4236 _base_put_smid_hi_priority(struct MPT3SAS_ADAPTER *ioc, u16 smid,
4237 	u16 msix_task)
4238 {
4239 	Mpi2RequestDescriptorUnion_t descriptor;
4240 	void *mpi_req_iomem;
4241 	u64 *request;
4242 
4243 	if (ioc->is_mcpu_endpoint) {
4244 		__le32 *mfp = (__le32 *)mpt3sas_base_get_msg_frame(ioc, smid);
4245 
4246 		/* TBD 256 is offset within sys register. */
4247 		mpi_req_iomem = (void __force *)ioc->chip
4248 					+ MPI_FRAME_START_OFFSET
4249 					+ (smid * ioc->request_sz);
4250 		_base_clone_mpi_to_sys_mem(mpi_req_iomem, (void *)mfp,
4251 							ioc->request_sz);
4252 	}
4253 
4254 	request = (u64 *)&descriptor;
4255 
4256 	descriptor.HighPriority.RequestFlags =
4257 	    MPI2_REQ_DESCRIPT_FLAGS_HIGH_PRIORITY;
4258 	descriptor.HighPriority.MSIxIndex =  msix_task;
4259 	descriptor.HighPriority.SMID = cpu_to_le16(smid);
4260 	descriptor.HighPriority.LMID = 0;
4261 	descriptor.HighPriority.Reserved1 = 0;
4262 	if (ioc->is_mcpu_endpoint)
4263 		_base_mpi_ep_writeq(*request,
4264 				&ioc->chip->RequestDescriptorPostLow,
4265 				&ioc->scsi_lookup_lock);
4266 	else
4267 		_base_writeq(*request, &ioc->chip->RequestDescriptorPostLow,
4268 		    &ioc->scsi_lookup_lock);
4269 }
4270 
4271 /**
4272  * mpt3sas_base_put_smid_nvme_encap - send NVMe encapsulated request to
4273  *  firmware
4274  * @ioc: per adapter object
4275  * @smid: system request message index
4276  */
4277 void
4278 mpt3sas_base_put_smid_nvme_encap(struct MPT3SAS_ADAPTER *ioc, u16 smid)
4279 {
4280 	Mpi2RequestDescriptorUnion_t descriptor;
4281 	u64 *request = (u64 *)&descriptor;
4282 
4283 	descriptor.Default.RequestFlags =
4284 		MPI26_REQ_DESCRIPT_FLAGS_PCIE_ENCAPSULATED;
4285 	descriptor.Default.MSIxIndex =  _base_set_and_get_msix_index(ioc, smid);
4286 	descriptor.Default.SMID = cpu_to_le16(smid);
4287 	descriptor.Default.LMID = 0;
4288 	descriptor.Default.DescriptorTypeDependent = 0;
4289 	_base_writeq(*request, &ioc->chip->RequestDescriptorPostLow,
4290 	    &ioc->scsi_lookup_lock);
4291 }
4292 
4293 /**
4294  * _base_put_smid_default - Default, primarily used for config pages
4295  * @ioc: per adapter object
4296  * @smid: system request message index
4297  */
4298 static void
4299 _base_put_smid_default(struct MPT3SAS_ADAPTER *ioc, u16 smid)
4300 {
4301 	Mpi2RequestDescriptorUnion_t descriptor;
4302 	void *mpi_req_iomem;
4303 	u64 *request;
4304 
4305 	if (ioc->is_mcpu_endpoint) {
4306 		__le32 *mfp = (__le32 *)mpt3sas_base_get_msg_frame(ioc, smid);
4307 
4308 		_clone_sg_entries(ioc, (void *) mfp, smid);
4309 		/* TBD 256 is offset within sys register */
4310 		mpi_req_iomem = (void __force *)ioc->chip +
4311 			MPI_FRAME_START_OFFSET + (smid * ioc->request_sz);
4312 		_base_clone_mpi_to_sys_mem(mpi_req_iomem, (void *)mfp,
4313 							ioc->request_sz);
4314 	}
4315 	request = (u64 *)&descriptor;
4316 	descriptor.Default.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE;
4317 	descriptor.Default.MSIxIndex = _base_set_and_get_msix_index(ioc, smid);
4318 	descriptor.Default.SMID = cpu_to_le16(smid);
4319 	descriptor.Default.LMID = 0;
4320 	descriptor.Default.DescriptorTypeDependent = 0;
4321 	if (ioc->is_mcpu_endpoint)
4322 		_base_mpi_ep_writeq(*request,
4323 				&ioc->chip->RequestDescriptorPostLow,
4324 				&ioc->scsi_lookup_lock);
4325 	else
4326 		_base_writeq(*request, &ioc->chip->RequestDescriptorPostLow,
4327 				&ioc->scsi_lookup_lock);
4328 }
4329 
4330 /**
4331  * _base_put_smid_scsi_io_atomic - send SCSI_IO request to firmware using
4332  *   Atomic Request Descriptor
4333  * @ioc: per adapter object
4334  * @smid: system request message index
4335  * @handle: device handle, unused in this function, for function type match
4336  *
4337  * Return: nothing.
4338  */
4339 static void
4340 _base_put_smid_scsi_io_atomic(struct MPT3SAS_ADAPTER *ioc, u16 smid,
4341 	u16 handle)
4342 {
4343 	Mpi26AtomicRequestDescriptor_t descriptor;
4344 	u32 *request = (u32 *)&descriptor;
4345 
4346 	descriptor.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_SCSI_IO;
4347 	descriptor.MSIxIndex = _base_set_and_get_msix_index(ioc, smid);
4348 	descriptor.SMID = cpu_to_le16(smid);
4349 
4350 	writel(cpu_to_le32(*request), &ioc->chip->AtomicRequestDescriptorPost);
4351 }
4352 
4353 /**
4354  * _base_put_smid_fast_path_atomic - send fast path request to firmware
4355  * using Atomic Request Descriptor
4356  * @ioc: per adapter object
4357  * @smid: system request message index
4358  * @handle: device handle, unused in this function, for function type match
4359  * Return: nothing
4360  */
4361 static void
4362 _base_put_smid_fast_path_atomic(struct MPT3SAS_ADAPTER *ioc, u16 smid,
4363 	u16 handle)
4364 {
4365 	Mpi26AtomicRequestDescriptor_t descriptor;
4366 	u32 *request = (u32 *)&descriptor;
4367 
4368 	descriptor.RequestFlags = MPI25_REQ_DESCRIPT_FLAGS_FAST_PATH_SCSI_IO;
4369 	descriptor.MSIxIndex = _base_set_and_get_msix_index(ioc, smid);
4370 	descriptor.SMID = cpu_to_le16(smid);
4371 
4372 	writel(cpu_to_le32(*request), &ioc->chip->AtomicRequestDescriptorPost);
4373 }
4374 
4375 /**
4376  * _base_put_smid_hi_priority_atomic - send Task Management request to
4377  * firmware using Atomic Request Descriptor
4378  * @ioc: per adapter object
4379  * @smid: system request message index
4380  * @msix_task: msix_task will be same as msix of IO in case of task abort else 0
4381  *
4382  * Return: nothing.
4383  */
4384 static void
4385 _base_put_smid_hi_priority_atomic(struct MPT3SAS_ADAPTER *ioc, u16 smid,
4386 	u16 msix_task)
4387 {
4388 	Mpi26AtomicRequestDescriptor_t descriptor;
4389 	u32 *request = (u32 *)&descriptor;
4390 
4391 	descriptor.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_HIGH_PRIORITY;
4392 	descriptor.MSIxIndex = msix_task;
4393 	descriptor.SMID = cpu_to_le16(smid);
4394 
4395 	writel(cpu_to_le32(*request), &ioc->chip->AtomicRequestDescriptorPost);
4396 }
4397 
4398 /**
4399  * _base_put_smid_default_atomic - Default, primarily used for config pages
4400  * use Atomic Request Descriptor
4401  * @ioc: per adapter object
4402  * @smid: system request message index
4403  *
4404  * Return: nothing.
4405  */
4406 static void
4407 _base_put_smid_default_atomic(struct MPT3SAS_ADAPTER *ioc, u16 smid)
4408 {
4409 	Mpi26AtomicRequestDescriptor_t descriptor;
4410 	u32 *request = (u32 *)&descriptor;
4411 
4412 	descriptor.RequestFlags = MPI2_REQ_DESCRIPT_FLAGS_DEFAULT_TYPE;
4413 	descriptor.MSIxIndex = _base_set_and_get_msix_index(ioc, smid);
4414 	descriptor.SMID = cpu_to_le16(smid);
4415 
4416 	writel(cpu_to_le32(*request), &ioc->chip->AtomicRequestDescriptorPost);
4417 }
4418 
4419 /**
4420  * _base_display_OEMs_branding - Display branding string
4421  * @ioc: per adapter object
4422  */
4423 static void
4424 _base_display_OEMs_branding(struct MPT3SAS_ADAPTER *ioc)
4425 {
4426 	if (ioc->pdev->subsystem_vendor != PCI_VENDOR_ID_INTEL)
4427 		return;
4428 
4429 	switch (ioc->pdev->subsystem_vendor) {
4430 	case PCI_VENDOR_ID_INTEL:
4431 		switch (ioc->pdev->device) {
4432 		case MPI2_MFGPAGE_DEVID_SAS2008:
4433 			switch (ioc->pdev->subsystem_device) {
4434 			case MPT2SAS_INTEL_RMS2LL080_SSDID:
4435 				ioc_info(ioc, "%s\n",
4436 					 MPT2SAS_INTEL_RMS2LL080_BRANDING);
4437 				break;
4438 			case MPT2SAS_INTEL_RMS2LL040_SSDID:
4439 				ioc_info(ioc, "%s\n",
4440 					 MPT2SAS_INTEL_RMS2LL040_BRANDING);
4441 				break;
4442 			case MPT2SAS_INTEL_SSD910_SSDID:
4443 				ioc_info(ioc, "%s\n",
4444 					 MPT2SAS_INTEL_SSD910_BRANDING);
4445 				break;
4446 			default:
4447 				ioc_info(ioc, "Intel(R) Controller: Subsystem ID: 0x%X\n",
4448 					 ioc->pdev->subsystem_device);
4449 				break;
4450 			}
4451 			break;
4452 		case MPI2_MFGPAGE_DEVID_SAS2308_2:
4453 			switch (ioc->pdev->subsystem_device) {
4454 			case MPT2SAS_INTEL_RS25GB008_SSDID:
4455 				ioc_info(ioc, "%s\n",
4456 					 MPT2SAS_INTEL_RS25GB008_BRANDING);
4457 				break;
4458 			case MPT2SAS_INTEL_RMS25JB080_SSDID:
4459 				ioc_info(ioc, "%s\n",
4460 					 MPT2SAS_INTEL_RMS25JB080_BRANDING);
4461 				break;
4462 			case MPT2SAS_INTEL_RMS25JB040_SSDID:
4463 				ioc_info(ioc, "%s\n",
4464 					 MPT2SAS_INTEL_RMS25JB040_BRANDING);
4465 				break;
4466 			case MPT2SAS_INTEL_RMS25KB080_SSDID:
4467 				ioc_info(ioc, "%s\n",
4468 					 MPT2SAS_INTEL_RMS25KB080_BRANDING);
4469 				break;
4470 			case MPT2SAS_INTEL_RMS25KB040_SSDID:
4471 				ioc_info(ioc, "%s\n",
4472 					 MPT2SAS_INTEL_RMS25KB040_BRANDING);
4473 				break;
4474 			case MPT2SAS_INTEL_RMS25LB040_SSDID:
4475 				ioc_info(ioc, "%s\n",
4476 					 MPT2SAS_INTEL_RMS25LB040_BRANDING);
4477 				break;
4478 			case MPT2SAS_INTEL_RMS25LB080_SSDID:
4479 				ioc_info(ioc, "%s\n",
4480 					 MPT2SAS_INTEL_RMS25LB080_BRANDING);
4481 				break;
4482 			default:
4483 				ioc_info(ioc, "Intel(R) Controller: Subsystem ID: 0x%X\n",
4484 					 ioc->pdev->subsystem_device);
4485 				break;
4486 			}
4487 			break;
4488 		case MPI25_MFGPAGE_DEVID_SAS3008:
4489 			switch (ioc->pdev->subsystem_device) {
4490 			case MPT3SAS_INTEL_RMS3JC080_SSDID:
4491 				ioc_info(ioc, "%s\n",
4492 					 MPT3SAS_INTEL_RMS3JC080_BRANDING);
4493 				break;
4494 
4495 			case MPT3SAS_INTEL_RS3GC008_SSDID:
4496 				ioc_info(ioc, "%s\n",
4497 					 MPT3SAS_INTEL_RS3GC008_BRANDING);
4498 				break;
4499 			case MPT3SAS_INTEL_RS3FC044_SSDID:
4500 				ioc_info(ioc, "%s\n",
4501 					 MPT3SAS_INTEL_RS3FC044_BRANDING);
4502 				break;
4503 			case MPT3SAS_INTEL_RS3UC080_SSDID:
4504 				ioc_info(ioc, "%s\n",
4505 					 MPT3SAS_INTEL_RS3UC080_BRANDING);
4506 				break;
4507 			default:
4508 				ioc_info(ioc, "Intel(R) Controller: Subsystem ID: 0x%X\n",
4509 					 ioc->pdev->subsystem_device);
4510 				break;
4511 			}
4512 			break;
4513 		default:
4514 			ioc_info(ioc, "Intel(R) Controller: Subsystem ID: 0x%X\n",
4515 				 ioc->pdev->subsystem_device);
4516 			break;
4517 		}
4518 		break;
4519 	case PCI_VENDOR_ID_DELL:
4520 		switch (ioc->pdev->device) {
4521 		case MPI2_MFGPAGE_DEVID_SAS2008:
4522 			switch (ioc->pdev->subsystem_device) {
4523 			case MPT2SAS_DELL_6GBPS_SAS_HBA_SSDID:
4524 				ioc_info(ioc, "%s\n",
4525 					 MPT2SAS_DELL_6GBPS_SAS_HBA_BRANDING);
4526 				break;
4527 			case MPT2SAS_DELL_PERC_H200_ADAPTER_SSDID:
4528 				ioc_info(ioc, "%s\n",
4529 					 MPT2SAS_DELL_PERC_H200_ADAPTER_BRANDING);
4530 				break;
4531 			case MPT2SAS_DELL_PERC_H200_INTEGRATED_SSDID:
4532 				ioc_info(ioc, "%s\n",
4533 					 MPT2SAS_DELL_PERC_H200_INTEGRATED_BRANDING);
4534 				break;
4535 			case MPT2SAS_DELL_PERC_H200_MODULAR_SSDID:
4536 				ioc_info(ioc, "%s\n",
4537 					 MPT2SAS_DELL_PERC_H200_MODULAR_BRANDING);
4538 				break;
4539 			case MPT2SAS_DELL_PERC_H200_EMBEDDED_SSDID:
4540 				ioc_info(ioc, "%s\n",
4541 					 MPT2SAS_DELL_PERC_H200_EMBEDDED_BRANDING);
4542 				break;
4543 			case MPT2SAS_DELL_PERC_H200_SSDID:
4544 				ioc_info(ioc, "%s\n",
4545 					 MPT2SAS_DELL_PERC_H200_BRANDING);
4546 				break;
4547 			case MPT2SAS_DELL_6GBPS_SAS_SSDID:
4548 				ioc_info(ioc, "%s\n",
4549 					 MPT2SAS_DELL_6GBPS_SAS_BRANDING);
4550 				break;
4551 			default:
4552 				ioc_info(ioc, "Dell 6Gbps HBA: Subsystem ID: 0x%X\n",
4553 					 ioc->pdev->subsystem_device);
4554 				break;
4555 			}
4556 			break;
4557 		case MPI25_MFGPAGE_DEVID_SAS3008:
4558 			switch (ioc->pdev->subsystem_device) {
4559 			case MPT3SAS_DELL_12G_HBA_SSDID:
4560 				ioc_info(ioc, "%s\n",
4561 					 MPT3SAS_DELL_12G_HBA_BRANDING);
4562 				break;
4563 			default:
4564 				ioc_info(ioc, "Dell 12Gbps HBA: Subsystem ID: 0x%X\n",
4565 					 ioc->pdev->subsystem_device);
4566 				break;
4567 			}
4568 			break;
4569 		default:
4570 			ioc_info(ioc, "Dell HBA: Subsystem ID: 0x%X\n",
4571 				 ioc->pdev->subsystem_device);
4572 			break;
4573 		}
4574 		break;
4575 	case PCI_VENDOR_ID_CISCO:
4576 		switch (ioc->pdev->device) {
4577 		case MPI25_MFGPAGE_DEVID_SAS3008:
4578 			switch (ioc->pdev->subsystem_device) {
4579 			case MPT3SAS_CISCO_12G_8E_HBA_SSDID:
4580 				ioc_info(ioc, "%s\n",
4581 					 MPT3SAS_CISCO_12G_8E_HBA_BRANDING);
4582 				break;
4583 			case MPT3SAS_CISCO_12G_8I_HBA_SSDID:
4584 				ioc_info(ioc, "%s\n",
4585 					 MPT3SAS_CISCO_12G_8I_HBA_BRANDING);
4586 				break;
4587 			case MPT3SAS_CISCO_12G_AVILA_HBA_SSDID:
4588 				ioc_info(ioc, "%s\n",
4589 					 MPT3SAS_CISCO_12G_AVILA_HBA_BRANDING);
4590 				break;
4591 			default:
4592 				ioc_info(ioc, "Cisco 12Gbps SAS HBA: Subsystem ID: 0x%X\n",
4593 					 ioc->pdev->subsystem_device);
4594 				break;
4595 			}
4596 			break;
4597 		case MPI25_MFGPAGE_DEVID_SAS3108_1:
4598 			switch (ioc->pdev->subsystem_device) {
4599 			case MPT3SAS_CISCO_12G_AVILA_HBA_SSDID:
4600 				ioc_info(ioc, "%s\n",
4601 					 MPT3SAS_CISCO_12G_AVILA_HBA_BRANDING);
4602 				break;
4603 			case MPT3SAS_CISCO_12G_COLUSA_MEZZANINE_HBA_SSDID:
4604 				ioc_info(ioc, "%s\n",
4605 					 MPT3SAS_CISCO_12G_COLUSA_MEZZANINE_HBA_BRANDING);
4606 				break;
4607 			default:
4608 				ioc_info(ioc, "Cisco 12Gbps SAS HBA: Subsystem ID: 0x%X\n",
4609 					 ioc->pdev->subsystem_device);
4610 				break;
4611 			}
4612 			break;
4613 		default:
4614 			ioc_info(ioc, "Cisco SAS HBA: Subsystem ID: 0x%X\n",
4615 				 ioc->pdev->subsystem_device);
4616 			break;
4617 		}
4618 		break;
4619 	case MPT2SAS_HP_3PAR_SSVID:
4620 		switch (ioc->pdev->device) {
4621 		case MPI2_MFGPAGE_DEVID_SAS2004:
4622 			switch (ioc->pdev->subsystem_device) {
4623 			case MPT2SAS_HP_DAUGHTER_2_4_INTERNAL_SSDID:
4624 				ioc_info(ioc, "%s\n",
4625 					 MPT2SAS_HP_DAUGHTER_2_4_INTERNAL_BRANDING);
4626 				break;
4627 			default:
4628 				ioc_info(ioc, "HP 6Gbps SAS HBA: Subsystem ID: 0x%X\n",
4629 					 ioc->pdev->subsystem_device);
4630 				break;
4631 			}
4632 			break;
4633 		case MPI2_MFGPAGE_DEVID_SAS2308_2:
4634 			switch (ioc->pdev->subsystem_device) {
4635 			case MPT2SAS_HP_2_4_INTERNAL_SSDID:
4636 				ioc_info(ioc, "%s\n",
4637 					 MPT2SAS_HP_2_4_INTERNAL_BRANDING);
4638 				break;
4639 			case MPT2SAS_HP_2_4_EXTERNAL_SSDID:
4640 				ioc_info(ioc, "%s\n",
4641 					 MPT2SAS_HP_2_4_EXTERNAL_BRANDING);
4642 				break;
4643 			case MPT2SAS_HP_1_4_INTERNAL_1_4_EXTERNAL_SSDID:
4644 				ioc_info(ioc, "%s\n",
4645 					 MPT2SAS_HP_1_4_INTERNAL_1_4_EXTERNAL_BRANDING);
4646 				break;
4647 			case MPT2SAS_HP_EMBEDDED_2_4_INTERNAL_SSDID:
4648 				ioc_info(ioc, "%s\n",
4649 					 MPT2SAS_HP_EMBEDDED_2_4_INTERNAL_BRANDING);
4650 				break;
4651 			default:
4652 				ioc_info(ioc, "HP 6Gbps SAS HBA: Subsystem ID: 0x%X\n",
4653 					 ioc->pdev->subsystem_device);
4654 				break;
4655 			}
4656 			break;
4657 		default:
4658 			ioc_info(ioc, "HP SAS HBA: Subsystem ID: 0x%X\n",
4659 				 ioc->pdev->subsystem_device);
4660 			break;
4661 		}
4662 		break;
4663 	default:
4664 		break;
4665 	}
4666 }
4667 
4668 /**
4669  * _base_display_fwpkg_version - sends FWUpload request to pull FWPkg
4670  *				version from FW Image Header.
4671  * @ioc: per adapter object
4672  *
4673  * Return: 0 for success, non-zero for failure.
4674  */
4675 	static int
4676 _base_display_fwpkg_version(struct MPT3SAS_ADAPTER *ioc)
4677 {
4678 	Mpi2FWImageHeader_t *fw_img_hdr;
4679 	Mpi26ComponentImageHeader_t *cmp_img_hdr;
4680 	Mpi25FWUploadRequest_t *mpi_request;
4681 	Mpi2FWUploadReply_t mpi_reply;
4682 	int r = 0, issue_diag_reset = 0;
4683 	u32  package_version = 0;
4684 	void *fwpkg_data = NULL;
4685 	dma_addr_t fwpkg_data_dma;
4686 	u16 smid, ioc_status;
4687 	size_t data_length;
4688 
4689 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
4690 
4691 	if (ioc->base_cmds.status & MPT3_CMD_PENDING) {
4692 		ioc_err(ioc, "%s: internal command already in use\n", __func__);
4693 		return -EAGAIN;
4694 	}
4695 
4696 	data_length = sizeof(Mpi2FWImageHeader_t);
4697 	fwpkg_data = dma_alloc_coherent(&ioc->pdev->dev, data_length,
4698 			&fwpkg_data_dma, GFP_KERNEL);
4699 	if (!fwpkg_data) {
4700 		ioc_err(ioc,
4701 		    "Memory allocation for fwpkg data failed at %s:%d/%s()!\n",
4702 			__FILE__, __LINE__, __func__);
4703 		return -ENOMEM;
4704 	}
4705 
4706 	smid = mpt3sas_base_get_smid(ioc, ioc->base_cb_idx);
4707 	if (!smid) {
4708 		ioc_err(ioc, "%s: failed obtaining a smid\n", __func__);
4709 		r = -EAGAIN;
4710 		goto out;
4711 	}
4712 
4713 	ioc->base_cmds.status = MPT3_CMD_PENDING;
4714 	mpi_request = mpt3sas_base_get_msg_frame(ioc, smid);
4715 	ioc->base_cmds.smid = smid;
4716 	memset(mpi_request, 0, sizeof(Mpi25FWUploadRequest_t));
4717 	mpi_request->Function = MPI2_FUNCTION_FW_UPLOAD;
4718 	mpi_request->ImageType = MPI2_FW_UPLOAD_ITYPE_FW_FLASH;
4719 	mpi_request->ImageSize = cpu_to_le32(data_length);
4720 	ioc->build_sg(ioc, &mpi_request->SGL, 0, 0, fwpkg_data_dma,
4721 			data_length);
4722 	init_completion(&ioc->base_cmds.done);
4723 	ioc->put_smid_default(ioc, smid);
4724 	/* Wait for 15 seconds */
4725 	wait_for_completion_timeout(&ioc->base_cmds.done,
4726 			FW_IMG_HDR_READ_TIMEOUT*HZ);
4727 	ioc_info(ioc, "%s: complete\n", __func__);
4728 	if (!(ioc->base_cmds.status & MPT3_CMD_COMPLETE)) {
4729 		ioc_err(ioc, "%s: timeout\n", __func__);
4730 		_debug_dump_mf(mpi_request,
4731 				sizeof(Mpi25FWUploadRequest_t)/4);
4732 		issue_diag_reset = 1;
4733 	} else {
4734 		memset(&mpi_reply, 0, sizeof(Mpi2FWUploadReply_t));
4735 		if (ioc->base_cmds.status & MPT3_CMD_REPLY_VALID) {
4736 			memcpy(&mpi_reply, ioc->base_cmds.reply,
4737 					sizeof(Mpi2FWUploadReply_t));
4738 			ioc_status = le16_to_cpu(mpi_reply.IOCStatus) &
4739 						MPI2_IOCSTATUS_MASK;
4740 			if (ioc_status == MPI2_IOCSTATUS_SUCCESS) {
4741 				fw_img_hdr = (Mpi2FWImageHeader_t *)fwpkg_data;
4742 				if (le32_to_cpu(fw_img_hdr->Signature) ==
4743 				    MPI26_IMAGE_HEADER_SIGNATURE0_MPI26) {
4744 					cmp_img_hdr =
4745 					    (Mpi26ComponentImageHeader_t *)
4746 					    (fwpkg_data);
4747 					package_version =
4748 					    le32_to_cpu(
4749 					    cmp_img_hdr->ApplicationSpecific);
4750 				} else
4751 					package_version =
4752 					    le32_to_cpu(
4753 					    fw_img_hdr->PackageVersion.Word);
4754 				if (package_version)
4755 					ioc_info(ioc,
4756 					"FW Package Ver(%02d.%02d.%02d.%02d)\n",
4757 					((package_version) & 0xFF000000) >> 24,
4758 					((package_version) & 0x00FF0000) >> 16,
4759 					((package_version) & 0x0000FF00) >> 8,
4760 					(package_version) & 0x000000FF);
4761 			} else {
4762 				_debug_dump_mf(&mpi_reply,
4763 						sizeof(Mpi2FWUploadReply_t)/4);
4764 			}
4765 		}
4766 	}
4767 	ioc->base_cmds.status = MPT3_CMD_NOT_USED;
4768 out:
4769 	if (fwpkg_data)
4770 		dma_free_coherent(&ioc->pdev->dev, data_length, fwpkg_data,
4771 				fwpkg_data_dma);
4772 	if (issue_diag_reset) {
4773 		if (ioc->drv_internal_flags & MPT_DRV_INTERNAL_FIRST_PE_ISSUED)
4774 			return -EFAULT;
4775 		if (mpt3sas_base_check_for_fault_and_issue_reset(ioc))
4776 			return -EFAULT;
4777 		r = -EAGAIN;
4778 	}
4779 	return r;
4780 }
4781 
4782 /**
4783  * _base_display_ioc_capabilities - Display IOC's capabilities.
4784  * @ioc: per adapter object
4785  */
4786 static void
4787 _base_display_ioc_capabilities(struct MPT3SAS_ADAPTER *ioc)
4788 {
4789 	int i = 0;
4790 	char desc[17] = {0};
4791 	u32 iounit_pg1_flags;
4792 
4793 	memtostr(desc, ioc->manu_pg0.ChipName);
4794 	ioc_info(ioc, "%s: FWVersion(%02d.%02d.%02d.%02d), ChipRevision(0x%02x)\n",
4795 		 desc,
4796 		 (ioc->facts.FWVersion.Word & 0xFF000000) >> 24,
4797 		 (ioc->facts.FWVersion.Word & 0x00FF0000) >> 16,
4798 		 (ioc->facts.FWVersion.Word & 0x0000FF00) >> 8,
4799 		 ioc->facts.FWVersion.Word & 0x000000FF,
4800 		 ioc->pdev->revision);
4801 
4802 	_base_display_OEMs_branding(ioc);
4803 
4804 	if (ioc->facts.ProtocolFlags & MPI2_IOCFACTS_PROTOCOL_NVME_DEVICES) {
4805 		pr_info("%sNVMe", i ? "," : "");
4806 		i++;
4807 	}
4808 
4809 	ioc_info(ioc, "Protocol=(");
4810 
4811 	if (ioc->facts.ProtocolFlags & MPI2_IOCFACTS_PROTOCOL_SCSI_INITIATOR) {
4812 		pr_cont("Initiator");
4813 		i++;
4814 	}
4815 
4816 	if (ioc->facts.ProtocolFlags & MPI2_IOCFACTS_PROTOCOL_SCSI_TARGET) {
4817 		pr_cont("%sTarget", i ? "," : "");
4818 		i++;
4819 	}
4820 
4821 	i = 0;
4822 	pr_cont("), Capabilities=(");
4823 
4824 	if (!ioc->hide_ir_msg) {
4825 		if (ioc->facts.IOCCapabilities &
4826 		    MPI2_IOCFACTS_CAPABILITY_INTEGRATED_RAID) {
4827 			pr_cont("Raid");
4828 			i++;
4829 		}
4830 	}
4831 
4832 	if (ioc->facts.IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_TLR) {
4833 		pr_cont("%sTLR", i ? "," : "");
4834 		i++;
4835 	}
4836 
4837 	if (ioc->facts.IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_MULTICAST) {
4838 		pr_cont("%sMulticast", i ? "," : "");
4839 		i++;
4840 	}
4841 
4842 	if (ioc->facts.IOCCapabilities &
4843 	    MPI2_IOCFACTS_CAPABILITY_BIDIRECTIONAL_TARGET) {
4844 		pr_cont("%sBIDI Target", i ? "," : "");
4845 		i++;
4846 	}
4847 
4848 	if (ioc->facts.IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_EEDP) {
4849 		pr_cont("%sEEDP", i ? "," : "");
4850 		i++;
4851 	}
4852 
4853 	if (ioc->facts.IOCCapabilities &
4854 	    MPI2_IOCFACTS_CAPABILITY_SNAPSHOT_BUFFER) {
4855 		pr_cont("%sSnapshot Buffer", i ? "," : "");
4856 		i++;
4857 	}
4858 
4859 	if (ioc->facts.IOCCapabilities &
4860 	    MPI2_IOCFACTS_CAPABILITY_DIAG_TRACE_BUFFER) {
4861 		pr_cont("%sDiag Trace Buffer", i ? "," : "");
4862 		i++;
4863 	}
4864 
4865 	if (ioc->facts.IOCCapabilities &
4866 	    MPI2_IOCFACTS_CAPABILITY_EXTENDED_BUFFER) {
4867 		pr_cont("%sDiag Extended Buffer", i ? "," : "");
4868 		i++;
4869 	}
4870 
4871 	if (ioc->facts.IOCCapabilities &
4872 	    MPI2_IOCFACTS_CAPABILITY_TASK_SET_FULL_HANDLING) {
4873 		pr_cont("%sTask Set Full", i ? "," : "");
4874 		i++;
4875 	}
4876 
4877 	iounit_pg1_flags = le32_to_cpu(ioc->iounit_pg1.Flags);
4878 	if (!(iounit_pg1_flags & MPI2_IOUNITPAGE1_NATIVE_COMMAND_Q_DISABLE)) {
4879 		pr_cont("%sNCQ", i ? "," : "");
4880 		i++;
4881 	}
4882 
4883 	pr_cont(")\n");
4884 }
4885 
4886 /**
4887  * mpt3sas_base_update_missing_delay - change the missing delay timers
4888  * @ioc: per adapter object
4889  * @device_missing_delay: amount of time till device is reported missing
4890  * @io_missing_delay: interval IO is returned when there is a missing device
4891  *
4892  * Passed on the command line, this function will modify the device missing
4893  * delay, as well as the io missing delay. This should be called at driver
4894  * load time.
4895  */
4896 void
4897 mpt3sas_base_update_missing_delay(struct MPT3SAS_ADAPTER *ioc,
4898 	u16 device_missing_delay, u8 io_missing_delay)
4899 {
4900 	u16 dmd, dmd_new, dmd_orignal;
4901 	u8 io_missing_delay_original;
4902 	u16 sz;
4903 	Mpi2SasIOUnitPage1_t *sas_iounit_pg1 = NULL;
4904 	Mpi2ConfigReply_t mpi_reply;
4905 	u8 num_phys = 0;
4906 	u16 ioc_status;
4907 
4908 	mpt3sas_config_get_number_hba_phys(ioc, &num_phys);
4909 	if (!num_phys)
4910 		return;
4911 
4912 	sz = struct_size(sas_iounit_pg1, PhyData, num_phys);
4913 	sas_iounit_pg1 = kzalloc(sz, GFP_KERNEL);
4914 	if (!sas_iounit_pg1) {
4915 		ioc_err(ioc, "failure at %s:%d/%s()!\n",
4916 			__FILE__, __LINE__, __func__);
4917 		goto out;
4918 	}
4919 	if ((mpt3sas_config_get_sas_iounit_pg1(ioc, &mpi_reply,
4920 	    sas_iounit_pg1, sz))) {
4921 		ioc_err(ioc, "failure at %s:%d/%s()!\n",
4922 			__FILE__, __LINE__, __func__);
4923 		goto out;
4924 	}
4925 	ioc_status = le16_to_cpu(mpi_reply.IOCStatus) &
4926 	    MPI2_IOCSTATUS_MASK;
4927 	if (ioc_status != MPI2_IOCSTATUS_SUCCESS) {
4928 		ioc_err(ioc, "failure at %s:%d/%s()!\n",
4929 			__FILE__, __LINE__, __func__);
4930 		goto out;
4931 	}
4932 
4933 	/* device missing delay */
4934 	dmd = sas_iounit_pg1->ReportDeviceMissingDelay;
4935 	if (dmd & MPI2_SASIOUNIT1_REPORT_MISSING_UNIT_16)
4936 		dmd = (dmd & MPI2_SASIOUNIT1_REPORT_MISSING_TIMEOUT_MASK) * 16;
4937 	else
4938 		dmd = dmd & MPI2_SASIOUNIT1_REPORT_MISSING_TIMEOUT_MASK;
4939 	dmd_orignal = dmd;
4940 	if (device_missing_delay > 0x7F) {
4941 		dmd = (device_missing_delay > 0x7F0) ? 0x7F0 :
4942 		    device_missing_delay;
4943 		dmd = dmd / 16;
4944 		dmd |= MPI2_SASIOUNIT1_REPORT_MISSING_UNIT_16;
4945 	} else
4946 		dmd = device_missing_delay;
4947 	sas_iounit_pg1->ReportDeviceMissingDelay = dmd;
4948 
4949 	/* io missing delay */
4950 	io_missing_delay_original = sas_iounit_pg1->IODeviceMissingDelay;
4951 	sas_iounit_pg1->IODeviceMissingDelay = io_missing_delay;
4952 
4953 	if (!mpt3sas_config_set_sas_iounit_pg1(ioc, &mpi_reply, sas_iounit_pg1,
4954 	    sz)) {
4955 		if (dmd & MPI2_SASIOUNIT1_REPORT_MISSING_UNIT_16)
4956 			dmd_new = (dmd &
4957 			    MPI2_SASIOUNIT1_REPORT_MISSING_TIMEOUT_MASK) * 16;
4958 		else
4959 			dmd_new =
4960 		    dmd & MPI2_SASIOUNIT1_REPORT_MISSING_TIMEOUT_MASK;
4961 		ioc_info(ioc, "device_missing_delay: old(%d), new(%d)\n",
4962 			 dmd_orignal, dmd_new);
4963 		ioc_info(ioc, "ioc_missing_delay: old(%d), new(%d)\n",
4964 			 io_missing_delay_original,
4965 			 io_missing_delay);
4966 		ioc->device_missing_delay = dmd_new;
4967 		ioc->io_missing_delay = io_missing_delay;
4968 	}
4969 
4970 out:
4971 	kfree(sas_iounit_pg1);
4972 }
4973 
4974 /**
4975  * _base_update_ioc_page1_inlinewith_perf_mode - Update IOC Page1 fields
4976  *    according to performance mode.
4977  * @ioc : per adapter object
4978  *
4979  * Return: zero on success; otherwise return EAGAIN error code asking the
4980  * caller to retry.
4981  */
4982 static int
4983 _base_update_ioc_page1_inlinewith_perf_mode(struct MPT3SAS_ADAPTER *ioc)
4984 {
4985 	Mpi2IOCPage1_t ioc_pg1;
4986 	Mpi2ConfigReply_t mpi_reply;
4987 	int rc;
4988 
4989 	rc = mpt3sas_config_get_ioc_pg1(ioc, &mpi_reply, &ioc->ioc_pg1_copy);
4990 	if (rc)
4991 		return rc;
4992 	memcpy(&ioc_pg1, &ioc->ioc_pg1_copy, sizeof(Mpi2IOCPage1_t));
4993 
4994 	switch (perf_mode) {
4995 	case MPT_PERF_MODE_DEFAULT:
4996 	case MPT_PERF_MODE_BALANCED:
4997 		if (ioc->high_iops_queues) {
4998 			ioc_info(ioc,
4999 				"Enable interrupt coalescing only for first\t"
5000 				"%d reply queues\n",
5001 				MPT3SAS_HIGH_IOPS_REPLY_QUEUES);
5002 			/*
5003 			 * If 31st bit is zero then interrupt coalescing is
5004 			 * enabled for all reply descriptor post queues.
5005 			 * If 31st bit is set to one then user can
5006 			 * enable/disable interrupt coalescing on per reply
5007 			 * descriptor post queue group(8) basis. So to enable
5008 			 * interrupt coalescing only on first reply descriptor
5009 			 * post queue group 31st bit and zero th bit is enabled.
5010 			 */
5011 			ioc_pg1.ProductSpecific = cpu_to_le32(0x80000000 |
5012 			    ((1 << MPT3SAS_HIGH_IOPS_REPLY_QUEUES/8) - 1));
5013 			rc = mpt3sas_config_set_ioc_pg1(ioc, &mpi_reply, &ioc_pg1);
5014 			if (rc)
5015 				return rc;
5016 			ioc_info(ioc, "performance mode: balanced\n");
5017 			return 0;
5018 		}
5019 		fallthrough;
5020 	case MPT_PERF_MODE_LATENCY:
5021 		/*
5022 		 * Enable interrupt coalescing on all reply queues
5023 		 * with timeout value 0xA
5024 		 */
5025 		ioc_pg1.CoalescingTimeout = cpu_to_le32(0xa);
5026 		ioc_pg1.Flags |= cpu_to_le32(MPI2_IOCPAGE1_REPLY_COALESCING);
5027 		ioc_pg1.ProductSpecific = 0;
5028 		rc = mpt3sas_config_set_ioc_pg1(ioc, &mpi_reply, &ioc_pg1);
5029 		if (rc)
5030 			return rc;
5031 		ioc_info(ioc, "performance mode: latency\n");
5032 		break;
5033 	case MPT_PERF_MODE_IOPS:
5034 		/*
5035 		 * Enable interrupt coalescing on all reply queues.
5036 		 */
5037 		ioc_info(ioc,
5038 		    "performance mode: iops with coalescing timeout: 0x%x\n",
5039 		    le32_to_cpu(ioc_pg1.CoalescingTimeout));
5040 		ioc_pg1.Flags |= cpu_to_le32(MPI2_IOCPAGE1_REPLY_COALESCING);
5041 		ioc_pg1.ProductSpecific = 0;
5042 		rc = mpt3sas_config_set_ioc_pg1(ioc, &mpi_reply, &ioc_pg1);
5043 		if (rc)
5044 			return rc;
5045 		break;
5046 	}
5047 	return 0;
5048 }
5049 
5050 /**
5051  * _base_get_event_diag_triggers - get event diag trigger values from
5052  *				persistent pages
5053  * @ioc : per adapter object
5054  *
5055  * Return: nothing.
5056  */
5057 static int
5058 _base_get_event_diag_triggers(struct MPT3SAS_ADAPTER *ioc)
5059 {
5060 	Mpi26DriverTriggerPage2_t trigger_pg2;
5061 	struct SL_WH_EVENT_TRIGGER_T *event_tg;
5062 	MPI26_DRIVER_MPI_EVENT_TRIGGER_ENTRY *mpi_event_tg;
5063 	Mpi2ConfigReply_t mpi_reply;
5064 	int r = 0, i = 0;
5065 	u16 count = 0;
5066 	u16 ioc_status;
5067 
5068 	r = mpt3sas_config_get_driver_trigger_pg2(ioc, &mpi_reply,
5069 	    &trigger_pg2);
5070 	if (r)
5071 		return r;
5072 
5073 	ioc_status = le16_to_cpu(mpi_reply.IOCStatus) &
5074 	    MPI2_IOCSTATUS_MASK;
5075 	if (ioc_status != MPI2_IOCSTATUS_SUCCESS) {
5076 		dinitprintk(ioc,
5077 		    ioc_err(ioc,
5078 		    "%s: Failed to get trigger pg2, ioc_status(0x%04x)\n",
5079 		   __func__, ioc_status));
5080 		return 0;
5081 	}
5082 
5083 	if (le16_to_cpu(trigger_pg2.NumMPIEventTrigger)) {
5084 		count = le16_to_cpu(trigger_pg2.NumMPIEventTrigger);
5085 		count = min_t(u16, NUM_VALID_ENTRIES, count);
5086 		ioc->diag_trigger_event.ValidEntries = count;
5087 
5088 		event_tg = &ioc->diag_trigger_event.EventTriggerEntry[0];
5089 		mpi_event_tg = &trigger_pg2.MPIEventTriggers[0];
5090 		for (i = 0; i < count; i++) {
5091 			event_tg->EventValue = le16_to_cpu(
5092 			    mpi_event_tg->MPIEventCode);
5093 			event_tg->LogEntryQualifier = le16_to_cpu(
5094 			    mpi_event_tg->MPIEventCodeSpecific);
5095 			event_tg++;
5096 			mpi_event_tg++;
5097 		}
5098 	}
5099 	return 0;
5100 }
5101 
5102 /**
5103  * _base_get_scsi_diag_triggers - get scsi diag trigger values from
5104  *				persistent pages
5105  * @ioc : per adapter object
5106  *
5107  * Return: 0 on success; otherwise return failure status.
5108  */
5109 static int
5110 _base_get_scsi_diag_triggers(struct MPT3SAS_ADAPTER *ioc)
5111 {
5112 	Mpi26DriverTriggerPage3_t trigger_pg3;
5113 	struct SL_WH_SCSI_TRIGGER_T *scsi_tg;
5114 	MPI26_DRIVER_SCSI_SENSE_TRIGGER_ENTRY *mpi_scsi_tg;
5115 	Mpi2ConfigReply_t mpi_reply;
5116 	int r = 0, i = 0;
5117 	u16 count = 0;
5118 	u16 ioc_status;
5119 
5120 	r = mpt3sas_config_get_driver_trigger_pg3(ioc, &mpi_reply,
5121 	    &trigger_pg3);
5122 	if (r)
5123 		return r;
5124 
5125 	ioc_status = le16_to_cpu(mpi_reply.IOCStatus) &
5126 	    MPI2_IOCSTATUS_MASK;
5127 	if (ioc_status != MPI2_IOCSTATUS_SUCCESS) {
5128 		dinitprintk(ioc,
5129 		    ioc_err(ioc,
5130 		    "%s: Failed to get trigger pg3, ioc_status(0x%04x)\n",
5131 		    __func__, ioc_status));
5132 		return 0;
5133 	}
5134 
5135 	if (le16_to_cpu(trigger_pg3.NumSCSISenseTrigger)) {
5136 		count = le16_to_cpu(trigger_pg3.NumSCSISenseTrigger);
5137 		count = min_t(u16, NUM_VALID_ENTRIES, count);
5138 		ioc->diag_trigger_scsi.ValidEntries = count;
5139 
5140 		scsi_tg = &ioc->diag_trigger_scsi.SCSITriggerEntry[0];
5141 		mpi_scsi_tg = &trigger_pg3.SCSISenseTriggers[0];
5142 		for (i = 0; i < count; i++) {
5143 			scsi_tg->ASCQ = mpi_scsi_tg->ASCQ;
5144 			scsi_tg->ASC = mpi_scsi_tg->ASC;
5145 			scsi_tg->SenseKey = mpi_scsi_tg->SenseKey;
5146 
5147 			scsi_tg++;
5148 			mpi_scsi_tg++;
5149 		}
5150 	}
5151 	return 0;
5152 }
5153 
5154 /**
5155  * _base_get_mpi_diag_triggers - get mpi diag trigger values from
5156  *				persistent pages
5157  * @ioc : per adapter object
5158  *
5159  * Return: 0 on success; otherwise return failure status.
5160  */
5161 static int
5162 _base_get_mpi_diag_triggers(struct MPT3SAS_ADAPTER *ioc)
5163 {
5164 	Mpi26DriverTriggerPage4_t trigger_pg4;
5165 	struct SL_WH_MPI_TRIGGER_T *status_tg;
5166 	MPI26_DRIVER_IOCSTATUS_LOGINFO_TRIGGER_ENTRY *mpi_status_tg;
5167 	Mpi2ConfigReply_t mpi_reply;
5168 	int r = 0, i = 0;
5169 	u16 count = 0;
5170 	u16 ioc_status;
5171 
5172 	r = mpt3sas_config_get_driver_trigger_pg4(ioc, &mpi_reply,
5173 	    &trigger_pg4);
5174 	if (r)
5175 		return r;
5176 
5177 	ioc_status = le16_to_cpu(mpi_reply.IOCStatus) &
5178 	    MPI2_IOCSTATUS_MASK;
5179 	if (ioc_status != MPI2_IOCSTATUS_SUCCESS) {
5180 		dinitprintk(ioc,
5181 		    ioc_err(ioc,
5182 		    "%s: Failed to get trigger pg4, ioc_status(0x%04x)\n",
5183 		    __func__, ioc_status));
5184 		return 0;
5185 	}
5186 
5187 	if (le16_to_cpu(trigger_pg4.NumIOCStatusLogInfoTrigger)) {
5188 		count = le16_to_cpu(trigger_pg4.NumIOCStatusLogInfoTrigger);
5189 		count = min_t(u16, NUM_VALID_ENTRIES, count);
5190 		ioc->diag_trigger_mpi.ValidEntries = count;
5191 
5192 		status_tg = &ioc->diag_trigger_mpi.MPITriggerEntry[0];
5193 		mpi_status_tg = &trigger_pg4.IOCStatusLoginfoTriggers[0];
5194 
5195 		for (i = 0; i < count; i++) {
5196 			status_tg->IOCStatus = le16_to_cpu(
5197 			    mpi_status_tg->IOCStatus);
5198 			status_tg->IocLogInfo = le32_to_cpu(
5199 			    mpi_status_tg->LogInfo);
5200 
5201 			status_tg++;
5202 			mpi_status_tg++;
5203 		}
5204 	}
5205 	return 0;
5206 }
5207 
5208 /**
5209  * _base_get_master_diag_triggers - get master diag trigger values from
5210  *				persistent pages
5211  * @ioc : per adapter object
5212  *
5213  * Return: nothing.
5214  */
5215 static int
5216 _base_get_master_diag_triggers(struct MPT3SAS_ADAPTER *ioc)
5217 {
5218 	Mpi26DriverTriggerPage1_t trigger_pg1;
5219 	Mpi2ConfigReply_t mpi_reply;
5220 	int r;
5221 	u16 ioc_status;
5222 
5223 	r = mpt3sas_config_get_driver_trigger_pg1(ioc, &mpi_reply,
5224 	    &trigger_pg1);
5225 	if (r)
5226 		return r;
5227 
5228 	ioc_status = le16_to_cpu(mpi_reply.IOCStatus) &
5229 	    MPI2_IOCSTATUS_MASK;
5230 	if (ioc_status != MPI2_IOCSTATUS_SUCCESS) {
5231 		dinitprintk(ioc,
5232 		    ioc_err(ioc,
5233 		    "%s: Failed to get trigger pg1, ioc_status(0x%04x)\n",
5234 		   __func__, ioc_status));
5235 		return 0;
5236 	}
5237 
5238 	if (le16_to_cpu(trigger_pg1.NumMasterTrigger))
5239 		ioc->diag_trigger_master.MasterData |=
5240 		    le32_to_cpu(
5241 		    trigger_pg1.MasterTriggers[0].MasterTriggerFlags);
5242 	return 0;
5243 }
5244 
5245 /**
5246  * _base_check_for_trigger_pages_support - checks whether HBA FW supports
5247  *					driver trigger pages or not
5248  * @ioc : per adapter object
5249  * @trigger_flags : address where trigger page0's TriggerFlags value is copied
5250  *
5251  * Return: trigger flags mask if HBA FW supports driver trigger pages;
5252  * otherwise returns %-EFAULT if driver trigger pages are not supported by FW or
5253  * return EAGAIN if diag reset occurred due to FW fault and asking the
5254  * caller to retry the command.
5255  *
5256  */
5257 static int
5258 _base_check_for_trigger_pages_support(struct MPT3SAS_ADAPTER *ioc, u32 *trigger_flags)
5259 {
5260 	Mpi26DriverTriggerPage0_t trigger_pg0;
5261 	int r = 0;
5262 	Mpi2ConfigReply_t mpi_reply;
5263 	u16 ioc_status;
5264 
5265 	r = mpt3sas_config_get_driver_trigger_pg0(ioc, &mpi_reply,
5266 	    &trigger_pg0);
5267 	if (r)
5268 		return r;
5269 
5270 	ioc_status = le16_to_cpu(mpi_reply.IOCStatus) &
5271 	    MPI2_IOCSTATUS_MASK;
5272 	if (ioc_status != MPI2_IOCSTATUS_SUCCESS)
5273 		return -EFAULT;
5274 
5275 	*trigger_flags = le16_to_cpu(trigger_pg0.TriggerFlags);
5276 	return 0;
5277 }
5278 
5279 /**
5280  * _base_get_diag_triggers - Retrieve diag trigger values from
5281  *				persistent pages.
5282  * @ioc : per adapter object
5283  *
5284  * Return: zero on success; otherwise return EAGAIN error codes
5285  * asking the caller to retry.
5286  */
5287 static int
5288 _base_get_diag_triggers(struct MPT3SAS_ADAPTER *ioc)
5289 {
5290 	int trigger_flags;
5291 	int r;
5292 
5293 	/*
5294 	 * Default setting of master trigger.
5295 	 */
5296 	ioc->diag_trigger_master.MasterData =
5297 	    (MASTER_TRIGGER_FW_FAULT + MASTER_TRIGGER_ADAPTER_RESET);
5298 
5299 	r = _base_check_for_trigger_pages_support(ioc, &trigger_flags);
5300 	if (r) {
5301 		if (r == -EAGAIN)
5302 			return r;
5303 		/*
5304 		 * Don't go for error handling when FW doesn't support
5305 		 * driver trigger pages.
5306 		 */
5307 		return 0;
5308 	}
5309 
5310 	ioc->supports_trigger_pages = 1;
5311 
5312 	/*
5313 	 * Retrieve master diag trigger values from driver trigger pg1
5314 	 * if master trigger bit enabled in TriggerFlags.
5315 	 */
5316 	if ((u16)trigger_flags &
5317 	    MPI26_DRIVER_TRIGGER0_FLAG_MASTER_TRIGGER_VALID) {
5318 		r = _base_get_master_diag_triggers(ioc);
5319 		if (r)
5320 			return r;
5321 	}
5322 
5323 	/*
5324 	 * Retrieve event diag trigger values from driver trigger pg2
5325 	 * if event trigger bit enabled in TriggerFlags.
5326 	 */
5327 	if ((u16)trigger_flags &
5328 	    MPI26_DRIVER_TRIGGER0_FLAG_MPI_EVENT_TRIGGER_VALID) {
5329 		r = _base_get_event_diag_triggers(ioc);
5330 		if (r)
5331 			return r;
5332 	}
5333 
5334 	/*
5335 	 * Retrieve scsi diag trigger values from driver trigger pg3
5336 	 * if scsi trigger bit enabled in TriggerFlags.
5337 	 */
5338 	if ((u16)trigger_flags &
5339 	    MPI26_DRIVER_TRIGGER0_FLAG_SCSI_SENSE_TRIGGER_VALID) {
5340 		r = _base_get_scsi_diag_triggers(ioc);
5341 		if (r)
5342 			return r;
5343 	}
5344 	/*
5345 	 * Retrieve mpi error diag trigger values from driver trigger pg4
5346 	 * if loginfo trigger bit enabled in TriggerFlags.
5347 	 */
5348 	if ((u16)trigger_flags &
5349 	    MPI26_DRIVER_TRIGGER0_FLAG_LOGINFO_TRIGGER_VALID) {
5350 		r = _base_get_mpi_diag_triggers(ioc);
5351 		if (r)
5352 			return r;
5353 	}
5354 	return 0;
5355 }
5356 
5357 /**
5358  * _base_update_diag_trigger_pages - Update the driver trigger pages after
5359  *			online FW update, in case updated FW supports driver
5360  *			trigger pages.
5361  * @ioc : per adapter object
5362  *
5363  * Return: nothing.
5364  */
5365 static void
5366 _base_update_diag_trigger_pages(struct MPT3SAS_ADAPTER *ioc)
5367 {
5368 
5369 	if (ioc->diag_trigger_master.MasterData)
5370 		mpt3sas_config_update_driver_trigger_pg1(ioc,
5371 		    &ioc->diag_trigger_master, 1);
5372 
5373 	if (ioc->diag_trigger_event.ValidEntries)
5374 		mpt3sas_config_update_driver_trigger_pg2(ioc,
5375 		    &ioc->diag_trigger_event, 1);
5376 
5377 	if (ioc->diag_trigger_scsi.ValidEntries)
5378 		mpt3sas_config_update_driver_trigger_pg3(ioc,
5379 		    &ioc->diag_trigger_scsi, 1);
5380 
5381 	if (ioc->diag_trigger_mpi.ValidEntries)
5382 		mpt3sas_config_update_driver_trigger_pg4(ioc,
5383 		    &ioc->diag_trigger_mpi, 1);
5384 }
5385 
5386 /**
5387  * _base_assign_fw_reported_qd	- Get FW reported QD for SAS/SATA devices.
5388  *				- On failure set default QD values.
5389  * @ioc : per adapter object
5390  *
5391  * Returns 0 for success, non-zero for failure.
5392  *
5393  */
5394 static int _base_assign_fw_reported_qd(struct MPT3SAS_ADAPTER *ioc)
5395 {
5396 	Mpi2ConfigReply_t mpi_reply;
5397 	Mpi2SasIOUnitPage1_t sas_iounit_pg1;
5398 	Mpi26PCIeIOUnitPage1_t pcie_iounit_pg1;
5399 	u16 depth;
5400 	int rc = 0;
5401 
5402 	ioc->max_wideport_qd = MPT3SAS_SAS_QUEUE_DEPTH;
5403 	ioc->max_narrowport_qd = MPT3SAS_SAS_QUEUE_DEPTH;
5404 	ioc->max_sata_qd = MPT3SAS_SATA_QUEUE_DEPTH;
5405 	ioc->max_nvme_qd = MPT3SAS_NVME_QUEUE_DEPTH;
5406 	if (!ioc->is_gen35_ioc)
5407 		goto out;
5408 	/* sas iounit page 1 */
5409 	rc = mpt3sas_config_get_sas_iounit_pg1(ioc, &mpi_reply,
5410 	    &sas_iounit_pg1, sizeof(Mpi2SasIOUnitPage1_t));
5411 	if (rc) {
5412 		pr_err("%s: failure at %s:%d/%s()!\n",
5413 		    ioc->name, __FILE__, __LINE__, __func__);
5414 		goto out;
5415 	}
5416 
5417 	depth = le16_to_cpu(sas_iounit_pg1.SASWideMaxQueueDepth);
5418 	ioc->max_wideport_qd = (depth ? depth : MPT3SAS_SAS_QUEUE_DEPTH);
5419 
5420 	depth = le16_to_cpu(sas_iounit_pg1.SASNarrowMaxQueueDepth);
5421 	ioc->max_narrowport_qd = (depth ? depth : MPT3SAS_SAS_QUEUE_DEPTH);
5422 
5423 	depth = sas_iounit_pg1.SATAMaxQDepth;
5424 	ioc->max_sata_qd = (depth ? depth : MPT3SAS_SATA_QUEUE_DEPTH);
5425 
5426 	/* pcie iounit page 1 */
5427 	rc = mpt3sas_config_get_pcie_iounit_pg1(ioc, &mpi_reply,
5428 	    &pcie_iounit_pg1, sizeof(Mpi26PCIeIOUnitPage1_t));
5429 	if (rc) {
5430 		pr_err("%s: failure at %s:%d/%s()!\n",
5431 		    ioc->name, __FILE__, __LINE__, __func__);
5432 		goto out;
5433 	}
5434 	ioc->max_nvme_qd = (le16_to_cpu(pcie_iounit_pg1.NVMeMaxQueueDepth)) ?
5435 	    (le16_to_cpu(pcie_iounit_pg1.NVMeMaxQueueDepth)) :
5436 	    MPT3SAS_NVME_QUEUE_DEPTH;
5437 out:
5438 	dinitprintk(ioc, pr_err(
5439 	    "MaxWidePortQD: 0x%x MaxNarrowPortQD: 0x%x MaxSataQD: 0x%x MaxNvmeQD: 0x%x\n",
5440 	    ioc->max_wideport_qd, ioc->max_narrowport_qd,
5441 	    ioc->max_sata_qd, ioc->max_nvme_qd));
5442 	return rc;
5443 }
5444 
5445 /**
5446  * mpt3sas_atto_validate_nvram - validate the ATTO nvram read from mfg pg1
5447  *
5448  * @ioc : per adapter object
5449  * @n   : ptr to the ATTO nvram structure
5450  * Return: 0 for success, non-zero for failure.
5451  */
5452 static int
5453 mpt3sas_atto_validate_nvram(struct MPT3SAS_ADAPTER *ioc,
5454 			    struct ATTO_SAS_NVRAM *n)
5455 {
5456 	int r = -EINVAL;
5457 	union ATTO_SAS_ADDRESS *s1;
5458 	u32 len;
5459 	u8 *pb;
5460 	u8 ckSum;
5461 
5462 	/* validate nvram checksum */
5463 	pb = (u8 *) n;
5464 	ckSum = ATTO_SASNVR_CKSUM_SEED;
5465 	len = sizeof(struct ATTO_SAS_NVRAM);
5466 
5467 	while (len--)
5468 		ckSum = ckSum + pb[len];
5469 
5470 	if (ckSum) {
5471 		ioc_err(ioc, "Invalid ATTO NVRAM checksum\n");
5472 		return r;
5473 	}
5474 
5475 	s1 = (union ATTO_SAS_ADDRESS *) n->SasAddr;
5476 
5477 	if (n->Signature[0] != 'E'
5478 	|| n->Signature[1] != 'S'
5479 	|| n->Signature[2] != 'A'
5480 	|| n->Signature[3] != 'S')
5481 		ioc_err(ioc, "Invalid ATTO NVRAM signature\n");
5482 	else if (n->Version > ATTO_SASNVR_VERSION)
5483 		ioc_info(ioc, "Invalid ATTO NVRAM version");
5484 	else if ((n->SasAddr[7] & (ATTO_SAS_ADDR_ALIGN - 1))
5485 			|| s1->b[0] != 0x50
5486 			|| s1->b[1] != 0x01
5487 			|| s1->b[2] != 0x08
5488 			|| (s1->b[3] & 0xF0) != 0x60
5489 			|| ((s1->b[3] & 0x0F) | le32_to_cpu(s1->d[1])) == 0) {
5490 		ioc_err(ioc, "Invalid ATTO SAS address\n");
5491 	} else
5492 		r = 0;
5493 	return r;
5494 }
5495 
5496 /**
5497  * mpt3sas_atto_get_sas_addr - get the ATTO SAS address from mfg page 1
5498  *
5499  * @ioc : per adapter object
5500  * @sas_addr : return sas address
5501  * Return: 0 for success, non-zero for failure.
5502  */
5503 static int
5504 mpt3sas_atto_get_sas_addr(struct MPT3SAS_ADAPTER *ioc, union ATTO_SAS_ADDRESS *sas_addr)
5505 {
5506 	Mpi2ManufacturingPage1_t mfg_pg1;
5507 	Mpi2ConfigReply_t mpi_reply;
5508 	struct ATTO_SAS_NVRAM *nvram;
5509 	int r;
5510 	__be64 addr;
5511 
5512 	r = mpt3sas_config_get_manufacturing_pg1(ioc, &mpi_reply, &mfg_pg1);
5513 	if (r) {
5514 		ioc_err(ioc, "Failed to read manufacturing page 1\n");
5515 		return r;
5516 	}
5517 
5518 	/* validate nvram */
5519 	nvram = (struct ATTO_SAS_NVRAM *) mfg_pg1.VPD;
5520 	r = mpt3sas_atto_validate_nvram(ioc, nvram);
5521 	if (r)
5522 		return r;
5523 
5524 	addr = *((__be64 *) nvram->SasAddr);
5525 	sas_addr->q = cpu_to_le64(be64_to_cpu(addr));
5526 	return r;
5527 }
5528 
5529 /**
5530  * mpt3sas_atto_init - perform initializaion for ATTO branded
5531  *					adapter.
5532  * @ioc : per adapter object
5533  *5
5534  * Return: 0 for success, non-zero for failure.
5535  */
5536 static int
5537 mpt3sas_atto_init(struct MPT3SAS_ADAPTER *ioc)
5538 {
5539 	int sz = 0;
5540 	Mpi2BiosPage4_t *bios_pg4 = NULL;
5541 	Mpi2ConfigReply_t mpi_reply;
5542 	int r;
5543 	int ix;
5544 	union ATTO_SAS_ADDRESS sas_addr;
5545 	union ATTO_SAS_ADDRESS temp;
5546 	union ATTO_SAS_ADDRESS bias;
5547 
5548 	r = mpt3sas_atto_get_sas_addr(ioc, &sas_addr);
5549 	if (r)
5550 		return r;
5551 
5552 	/* get header first to get size */
5553 	r = mpt3sas_config_get_bios_pg4(ioc, &mpi_reply, NULL, 0);
5554 	if (r) {
5555 		ioc_err(ioc, "Failed to read ATTO bios page 4 header.\n");
5556 		return r;
5557 	}
5558 
5559 	sz = mpi_reply.Header.PageLength * sizeof(u32);
5560 	bios_pg4 = kzalloc(sz, GFP_KERNEL);
5561 	if (!bios_pg4) {
5562 		ioc_err(ioc, "Failed to allocate memory for ATTO bios page.\n");
5563 		return -ENOMEM;
5564 	}
5565 
5566 	/* read bios page 4 */
5567 	r = mpt3sas_config_get_bios_pg4(ioc, &mpi_reply, bios_pg4, sz);
5568 	if (r) {
5569 		ioc_err(ioc, "Failed to read ATTO bios page 4\n");
5570 		goto out;
5571 	}
5572 
5573 	/* Update bios page 4 with the ATTO WWID */
5574 	bias.q = sas_addr.q;
5575 	bias.b[7] += ATTO_SAS_ADDR_DEVNAME_BIAS;
5576 
5577 	for (ix = 0; ix < bios_pg4->NumPhys; ix++) {
5578 		temp.q = sas_addr.q;
5579 		temp.b[7] += ix;
5580 		bios_pg4->Phy[ix].ReassignmentWWID = temp.q;
5581 		bios_pg4->Phy[ix].ReassignmentDeviceName = bias.q;
5582 	}
5583 	r = mpt3sas_config_set_bios_pg4(ioc, &mpi_reply, bios_pg4, sz);
5584 
5585 out:
5586 	kfree(bios_pg4);
5587 	return r;
5588 }
5589 
5590 /**
5591  * _base_static_config_pages - static start of day config pages
5592  * @ioc: per adapter object
5593  */
5594 static int
5595 _base_static_config_pages(struct MPT3SAS_ADAPTER *ioc)
5596 {
5597 	Mpi2IOUnitPage8_t iounit_pg8;
5598 	Mpi2ConfigReply_t mpi_reply;
5599 	u32 iounit_pg1_flags;
5600 	int tg_flags = 0;
5601 	int rc;
5602 	ioc->nvme_abort_timeout = 30;
5603 
5604 	rc = mpt3sas_config_get_manufacturing_pg0(ioc, &mpi_reply,
5605 	    &ioc->manu_pg0);
5606 	if (rc)
5607 		return rc;
5608 	if (ioc->ir_firmware) {
5609 		rc = mpt3sas_config_get_manufacturing_pg10(ioc, &mpi_reply,
5610 		    &ioc->manu_pg10);
5611 		if (rc)
5612 			return rc;
5613 	}
5614 
5615 	if (ioc->pdev->vendor == MPI2_MFGPAGE_VENDORID_ATTO) {
5616 		rc = mpt3sas_atto_init(ioc);
5617 		if (rc)
5618 			return rc;
5619 	}
5620 
5621 	/*
5622 	 * Ensure correct T10 PI operation if vendor left EEDPTagMode
5623 	 * flag unset in NVDATA.
5624 	 */
5625 	rc = mpt3sas_config_get_manufacturing_pg11(ioc, &mpi_reply,
5626 	    &ioc->manu_pg11);
5627 	if (rc)
5628 		return rc;
5629 	if (!ioc->is_gen35_ioc && ioc->manu_pg11.EEDPTagMode == 0) {
5630 		pr_err("%s: overriding NVDATA EEDPTagMode setting\n",
5631 		    ioc->name);
5632 		ioc->manu_pg11.EEDPTagMode &= ~0x3;
5633 		ioc->manu_pg11.EEDPTagMode |= 0x1;
5634 		mpt3sas_config_set_manufacturing_pg11(ioc, &mpi_reply,
5635 		    &ioc->manu_pg11);
5636 	}
5637 	if (ioc->manu_pg11.AddlFlags2 & NVME_TASK_MNGT_CUSTOM_MASK)
5638 		ioc->tm_custom_handling = 1;
5639 	else {
5640 		ioc->tm_custom_handling = 0;
5641 		if (ioc->manu_pg11.NVMeAbortTO < NVME_TASK_ABORT_MIN_TIMEOUT)
5642 			ioc->nvme_abort_timeout = NVME_TASK_ABORT_MIN_TIMEOUT;
5643 		else if (ioc->manu_pg11.NVMeAbortTO >
5644 					NVME_TASK_ABORT_MAX_TIMEOUT)
5645 			ioc->nvme_abort_timeout = NVME_TASK_ABORT_MAX_TIMEOUT;
5646 		else
5647 			ioc->nvme_abort_timeout = ioc->manu_pg11.NVMeAbortTO;
5648 	}
5649 	ioc->time_sync_interval =
5650 	    ioc->manu_pg11.TimeSyncInterval & MPT3SAS_TIMESYNC_MASK;
5651 	if (ioc->time_sync_interval) {
5652 		if (ioc->manu_pg11.TimeSyncInterval & MPT3SAS_TIMESYNC_UNIT_MASK)
5653 			ioc->time_sync_interval =
5654 			    ioc->time_sync_interval * SECONDS_PER_HOUR;
5655 		else
5656 			ioc->time_sync_interval =
5657 			    ioc->time_sync_interval * SECONDS_PER_MIN;
5658 		dinitprintk(ioc, ioc_info(ioc,
5659 		    "Driver-FW TimeSync interval is %d seconds. ManuPg11 TimeSync Unit is in %s\n",
5660 		    ioc->time_sync_interval, (ioc->manu_pg11.TimeSyncInterval &
5661 		    MPT3SAS_TIMESYNC_UNIT_MASK) ? "Hour" : "Minute"));
5662 	} else {
5663 		if (ioc->is_gen35_ioc)
5664 			ioc_warn(ioc,
5665 			    "TimeSync Interval in Manuf page-11 is not enabled. Periodic Time-Sync will be disabled\n");
5666 	}
5667 	rc = _base_assign_fw_reported_qd(ioc);
5668 	if (rc)
5669 		return rc;
5670 
5671 	/*
5672 	 * ATTO doesn't use bios page 2 and 3 for bios settings.
5673 	 */
5674 	if (ioc->pdev->vendor ==  MPI2_MFGPAGE_VENDORID_ATTO)
5675 		ioc->bios_pg3.BiosVersion = 0;
5676 	else {
5677 		rc = mpt3sas_config_get_bios_pg2(ioc, &mpi_reply, &ioc->bios_pg2);
5678 		if (rc)
5679 			return rc;
5680 		rc = mpt3sas_config_get_bios_pg3(ioc, &mpi_reply, &ioc->bios_pg3);
5681 		if (rc)
5682 			return rc;
5683 	}
5684 
5685 	rc = mpt3sas_config_get_ioc_pg8(ioc, &mpi_reply, &ioc->ioc_pg8);
5686 	if (rc)
5687 		return rc;
5688 	rc = mpt3sas_config_get_iounit_pg0(ioc, &mpi_reply, &ioc->iounit_pg0);
5689 	if (rc)
5690 		return rc;
5691 	rc = mpt3sas_config_get_iounit_pg1(ioc, &mpi_reply, &ioc->iounit_pg1);
5692 	if (rc)
5693 		return rc;
5694 	rc = mpt3sas_config_get_iounit_pg8(ioc, &mpi_reply, &iounit_pg8);
5695 	if (rc)
5696 		return rc;
5697 	_base_display_ioc_capabilities(ioc);
5698 
5699 	/*
5700 	 * Enable task_set_full handling in iounit_pg1 when the
5701 	 * facts capabilities indicate that its supported.
5702 	 */
5703 	iounit_pg1_flags = le32_to_cpu(ioc->iounit_pg1.Flags);
5704 	if ((ioc->facts.IOCCapabilities &
5705 	    MPI2_IOCFACTS_CAPABILITY_TASK_SET_FULL_HANDLING))
5706 		iounit_pg1_flags &=
5707 		    ~MPI2_IOUNITPAGE1_DISABLE_TASK_SET_FULL_HANDLING;
5708 	else
5709 		iounit_pg1_flags |=
5710 		    MPI2_IOUNITPAGE1_DISABLE_TASK_SET_FULL_HANDLING;
5711 	ioc->iounit_pg1.Flags = cpu_to_le32(iounit_pg1_flags);
5712 	rc = mpt3sas_config_set_iounit_pg1(ioc, &mpi_reply, &ioc->iounit_pg1);
5713 	if (rc)
5714 		return rc;
5715 
5716 	if (iounit_pg8.NumSensors)
5717 		ioc->temp_sensors_count = iounit_pg8.NumSensors;
5718 	if (ioc->is_aero_ioc) {
5719 		rc = _base_update_ioc_page1_inlinewith_perf_mode(ioc);
5720 		if (rc)
5721 			return rc;
5722 	}
5723 	if (ioc->is_gen35_ioc) {
5724 		if (ioc->is_driver_loading) {
5725 			rc = _base_get_diag_triggers(ioc);
5726 			if (rc)
5727 				return rc;
5728 		} else {
5729 			/*
5730 			 * In case of online HBA FW update operation,
5731 			 * check whether updated FW supports the driver trigger
5732 			 * pages or not.
5733 			 * - If previous FW has not supported driver trigger
5734 			 *   pages and newer FW supports them then update these
5735 			 *   pages with current diag trigger values.
5736 			 * - If previous FW has supported driver trigger pages
5737 			 *   and new FW doesn't support them then disable
5738 			 *   support_trigger_pages flag.
5739 			 */
5740 			_base_check_for_trigger_pages_support(ioc, &tg_flags);
5741 			if (!ioc->supports_trigger_pages && tg_flags != -EFAULT)
5742 				_base_update_diag_trigger_pages(ioc);
5743 			else if (ioc->supports_trigger_pages &&
5744 			    tg_flags == -EFAULT)
5745 				ioc->supports_trigger_pages = 0;
5746 		}
5747 	}
5748 	return 0;
5749 }
5750 
5751 /**
5752  * mpt3sas_free_enclosure_list - release memory
5753  * @ioc: per adapter object
5754  *
5755  * Free memory allocated during enclosure add.
5756  */
5757 void
5758 mpt3sas_free_enclosure_list(struct MPT3SAS_ADAPTER *ioc)
5759 {
5760 	struct _enclosure_node *enclosure_dev, *enclosure_dev_next;
5761 
5762 	/* Free enclosure list */
5763 	list_for_each_entry_safe(enclosure_dev,
5764 			enclosure_dev_next, &ioc->enclosure_list, list) {
5765 		list_del(&enclosure_dev->list);
5766 		kfree(enclosure_dev);
5767 	}
5768 }
5769 
5770 /**
5771  * _base_release_memory_pools - release memory
5772  * @ioc: per adapter object
5773  *
5774  * Free memory allocated from _base_allocate_memory_pools.
5775  */
5776 static void
5777 _base_release_memory_pools(struct MPT3SAS_ADAPTER *ioc)
5778 {
5779 	int i = 0;
5780 	int j = 0;
5781 	int dma_alloc_count = 0;
5782 	struct chain_tracker *ct;
5783 	int count = ioc->rdpq_array_enable ? ioc->reply_queue_count : 1;
5784 
5785 	dexitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
5786 
5787 	if (ioc->request) {
5788 		dma_free_coherent(&ioc->pdev->dev, ioc->request_dma_sz,
5789 		    ioc->request,  ioc->request_dma);
5790 		dexitprintk(ioc,
5791 			    ioc_info(ioc, "request_pool(0x%p): free\n",
5792 				     ioc->request));
5793 		ioc->request = NULL;
5794 	}
5795 
5796 	if (ioc->sense) {
5797 		dma_pool_free(ioc->sense_dma_pool, ioc->sense, ioc->sense_dma);
5798 		dma_pool_destroy(ioc->sense_dma_pool);
5799 		dexitprintk(ioc,
5800 			    ioc_info(ioc, "sense_pool(0x%p): free\n",
5801 				     ioc->sense));
5802 		ioc->sense = NULL;
5803 	}
5804 
5805 	if (ioc->reply) {
5806 		dma_pool_free(ioc->reply_dma_pool, ioc->reply, ioc->reply_dma);
5807 		dma_pool_destroy(ioc->reply_dma_pool);
5808 		dexitprintk(ioc,
5809 			    ioc_info(ioc, "reply_pool(0x%p): free\n",
5810 				     ioc->reply));
5811 		ioc->reply = NULL;
5812 	}
5813 
5814 	if (ioc->reply_free) {
5815 		dma_pool_free(ioc->reply_free_dma_pool, ioc->reply_free,
5816 		    ioc->reply_free_dma);
5817 		dma_pool_destroy(ioc->reply_free_dma_pool);
5818 		dexitprintk(ioc,
5819 			    ioc_info(ioc, "reply_free_pool(0x%p): free\n",
5820 				     ioc->reply_free));
5821 		ioc->reply_free = NULL;
5822 	}
5823 
5824 	if (ioc->reply_post) {
5825 		dma_alloc_count = DIV_ROUND_UP(count,
5826 				RDPQ_MAX_INDEX_IN_ONE_CHUNK);
5827 		for (i = 0; i < count; i++) {
5828 			if (i % RDPQ_MAX_INDEX_IN_ONE_CHUNK == 0
5829 			    && dma_alloc_count) {
5830 				if (ioc->reply_post[i].reply_post_free) {
5831 					dma_pool_free(
5832 					    ioc->reply_post_free_dma_pool,
5833 					    ioc->reply_post[i].reply_post_free,
5834 					ioc->reply_post[i].reply_post_free_dma);
5835 					dexitprintk(ioc, ioc_info(ioc,
5836 					   "reply_post_free_pool(0x%p): free\n",
5837 					   ioc->reply_post[i].reply_post_free));
5838 					ioc->reply_post[i].reply_post_free =
5839 									NULL;
5840 				}
5841 				--dma_alloc_count;
5842 			}
5843 		}
5844 		dma_pool_destroy(ioc->reply_post_free_dma_pool);
5845 		if (ioc->reply_post_free_array &&
5846 			ioc->rdpq_array_enable) {
5847 			dma_pool_free(ioc->reply_post_free_array_dma_pool,
5848 			    ioc->reply_post_free_array,
5849 			    ioc->reply_post_free_array_dma);
5850 			ioc->reply_post_free_array = NULL;
5851 		}
5852 		dma_pool_destroy(ioc->reply_post_free_array_dma_pool);
5853 		kfree(ioc->reply_post);
5854 	}
5855 
5856 	if (ioc->pcie_sgl_dma_pool) {
5857 		for (i = 0; i < ioc->scsiio_depth; i++) {
5858 			dma_pool_free(ioc->pcie_sgl_dma_pool,
5859 					ioc->pcie_sg_lookup[i].pcie_sgl,
5860 					ioc->pcie_sg_lookup[i].pcie_sgl_dma);
5861 			ioc->pcie_sg_lookup[i].pcie_sgl = NULL;
5862 		}
5863 		dma_pool_destroy(ioc->pcie_sgl_dma_pool);
5864 	}
5865 	kfree(ioc->pcie_sg_lookup);
5866 	ioc->pcie_sg_lookup = NULL;
5867 
5868 	if (ioc->config_page) {
5869 		dexitprintk(ioc,
5870 			    ioc_info(ioc, "config_page(0x%p): free\n",
5871 				     ioc->config_page));
5872 		dma_free_coherent(&ioc->pdev->dev, ioc->config_page_sz,
5873 		    ioc->config_page, ioc->config_page_dma);
5874 	}
5875 
5876 	kfree(ioc->hpr_lookup);
5877 	ioc->hpr_lookup = NULL;
5878 	kfree(ioc->internal_lookup);
5879 	ioc->internal_lookup = NULL;
5880 	if (ioc->chain_lookup) {
5881 		for (i = 0; i < ioc->scsiio_depth; i++) {
5882 			for (j = ioc->chains_per_prp_buffer;
5883 			    j < ioc->chains_needed_per_io; j++) {
5884 				ct = &ioc->chain_lookup[i].chains_per_smid[j];
5885 				if (ct && ct->chain_buffer)
5886 					dma_pool_free(ioc->chain_dma_pool,
5887 						ct->chain_buffer,
5888 						ct->chain_buffer_dma);
5889 			}
5890 			kfree(ioc->chain_lookup[i].chains_per_smid);
5891 		}
5892 		dma_pool_destroy(ioc->chain_dma_pool);
5893 		kfree(ioc->chain_lookup);
5894 		ioc->chain_lookup = NULL;
5895 	}
5896 
5897 	kfree(ioc->io_queue_num);
5898 	ioc->io_queue_num = NULL;
5899 }
5900 
5901 /**
5902  * mpt3sas_check_same_4gb_region - checks whether all reply queues in a set are
5903  *	having same upper 32bits in their base memory address.
5904  * @start_address: Base address of a reply queue set
5905  * @pool_sz: Size of single Reply Descriptor Post Queues pool size
5906  *
5907  * Return: 1 if reply queues in a set have a same upper 32bits in their base
5908  * memory address, else 0.
5909  */
5910 static int
5911 mpt3sas_check_same_4gb_region(dma_addr_t start_address, u32 pool_sz)
5912 {
5913 	dma_addr_t end_address;
5914 
5915 	end_address = start_address + pool_sz - 1;
5916 
5917 	if (upper_32_bits(start_address) == upper_32_bits(end_address))
5918 		return 1;
5919 	else
5920 		return 0;
5921 }
5922 
5923 /**
5924  * _base_reduce_hba_queue_depth- Retry with reduced queue depth
5925  * @ioc: Adapter object
5926  *
5927  * Return: 0 for success, non-zero for failure.
5928  **/
5929 static inline int
5930 _base_reduce_hba_queue_depth(struct MPT3SAS_ADAPTER *ioc)
5931 {
5932 	int reduce_sz = 64;
5933 
5934 	if ((ioc->hba_queue_depth - reduce_sz) >
5935 	    (ioc->internal_depth + INTERNAL_SCSIIO_CMDS_COUNT)) {
5936 		ioc->hba_queue_depth -= reduce_sz;
5937 		return 0;
5938 	} else
5939 		return -ENOMEM;
5940 }
5941 
5942 /**
5943  * _base_allocate_pcie_sgl_pool - Allocating DMA'able memory
5944  *			for pcie sgl pools.
5945  * @ioc: Adapter object
5946  * @sz: DMA Pool size
5947  *
5948  * Return: 0 for success, non-zero for failure.
5949  */
5950 
5951 static int
5952 _base_allocate_pcie_sgl_pool(struct MPT3SAS_ADAPTER *ioc, u32 sz)
5953 {
5954 	int i = 0, j = 0;
5955 	struct chain_tracker *ct;
5956 
5957 	ioc->pcie_sgl_dma_pool =
5958 	    dma_pool_create("PCIe SGL pool", &ioc->pdev->dev, sz,
5959 	    ioc->page_size, 0);
5960 	if (!ioc->pcie_sgl_dma_pool) {
5961 		ioc_err(ioc, "PCIe SGL pool: dma_pool_create failed\n");
5962 		return -ENOMEM;
5963 	}
5964 
5965 	ioc->chains_per_prp_buffer = sz/ioc->chain_segment_sz;
5966 	ioc->chains_per_prp_buffer =
5967 	    min(ioc->chains_per_prp_buffer, ioc->chains_needed_per_io);
5968 	for (i = 0; i < ioc->scsiio_depth; i++) {
5969 		ioc->pcie_sg_lookup[i].pcie_sgl =
5970 		    dma_pool_alloc(ioc->pcie_sgl_dma_pool, GFP_KERNEL,
5971 		    &ioc->pcie_sg_lookup[i].pcie_sgl_dma);
5972 		if (!ioc->pcie_sg_lookup[i].pcie_sgl) {
5973 			ioc_err(ioc, "PCIe SGL pool: dma_pool_alloc failed\n");
5974 			return -EAGAIN;
5975 		}
5976 
5977 		if (!mpt3sas_check_same_4gb_region(
5978 		    ioc->pcie_sg_lookup[i].pcie_sgl_dma, sz)) {
5979 			ioc_err(ioc, "PCIE SGLs are not in same 4G !! pcie sgl (0x%p) dma = (0x%llx)\n",
5980 			    ioc->pcie_sg_lookup[i].pcie_sgl,
5981 			    (unsigned long long)
5982 			    ioc->pcie_sg_lookup[i].pcie_sgl_dma);
5983 			ioc->use_32bit_dma = true;
5984 			return -EAGAIN;
5985 		}
5986 
5987 		for (j = 0; j < ioc->chains_per_prp_buffer; j++) {
5988 			ct = &ioc->chain_lookup[i].chains_per_smid[j];
5989 			ct->chain_buffer =
5990 			    ioc->pcie_sg_lookup[i].pcie_sgl +
5991 			    (j * ioc->chain_segment_sz);
5992 			ct->chain_buffer_dma =
5993 			    ioc->pcie_sg_lookup[i].pcie_sgl_dma +
5994 			    (j * ioc->chain_segment_sz);
5995 		}
5996 	}
5997 	dinitprintk(ioc, ioc_info(ioc,
5998 	    "PCIe sgl pool depth(%d), element_size(%d), pool_size(%d kB)\n",
5999 	    ioc->scsiio_depth, sz, (sz * ioc->scsiio_depth)/1024));
6000 	dinitprintk(ioc, ioc_info(ioc,
6001 	    "Number of chains can fit in a PRP page(%d)\n",
6002 	    ioc->chains_per_prp_buffer));
6003 	return 0;
6004 }
6005 
6006 /**
6007  * _base_allocate_chain_dma_pool - Allocating DMA'able memory
6008  *			for chain dma pool.
6009  * @ioc: Adapter object
6010  * @sz: DMA Pool size
6011  *
6012  * Return: 0 for success, non-zero for failure.
6013  */
6014 static int
6015 _base_allocate_chain_dma_pool(struct MPT3SAS_ADAPTER *ioc, u32 sz)
6016 {
6017 	int i = 0, j = 0;
6018 	struct chain_tracker *ctr;
6019 
6020 	ioc->chain_dma_pool = dma_pool_create("chain pool", &ioc->pdev->dev,
6021 	    ioc->chain_segment_sz, 16, 0);
6022 	if (!ioc->chain_dma_pool)
6023 		return -ENOMEM;
6024 
6025 	for (i = 0; i < ioc->scsiio_depth; i++) {
6026 		for (j = ioc->chains_per_prp_buffer;
6027 		    j < ioc->chains_needed_per_io; j++) {
6028 			ctr = &ioc->chain_lookup[i].chains_per_smid[j];
6029 			ctr->chain_buffer = dma_pool_alloc(ioc->chain_dma_pool,
6030 			    GFP_KERNEL, &ctr->chain_buffer_dma);
6031 			if (!ctr->chain_buffer)
6032 				return -EAGAIN;
6033 			if (!mpt3sas_check_same_4gb_region(
6034 			    ctr->chain_buffer_dma, ioc->chain_segment_sz)) {
6035 				ioc_err(ioc,
6036 				    "Chain buffers are not in same 4G !!! Chain buff (0x%p) dma = (0x%llx)\n",
6037 				    ctr->chain_buffer,
6038 				    (unsigned long long)ctr->chain_buffer_dma);
6039 				ioc->use_32bit_dma = true;
6040 				return -EAGAIN;
6041 			}
6042 		}
6043 	}
6044 	dinitprintk(ioc, ioc_info(ioc,
6045 	    "chain_lookup depth (%d), frame_size(%d), pool_size(%d kB)\n",
6046 	    ioc->scsiio_depth, ioc->chain_segment_sz, ((ioc->scsiio_depth *
6047 	    (ioc->chains_needed_per_io - ioc->chains_per_prp_buffer) *
6048 	    ioc->chain_segment_sz))/1024));
6049 	return 0;
6050 }
6051 
6052 /**
6053  * _base_allocate_sense_dma_pool - Allocating DMA'able memory
6054  *			for sense dma pool.
6055  * @ioc: Adapter object
6056  * @sz: DMA Pool size
6057  * Return: 0 for success, non-zero for failure.
6058  */
6059 static int
6060 _base_allocate_sense_dma_pool(struct MPT3SAS_ADAPTER *ioc, u32 sz)
6061 {
6062 	ioc->sense_dma_pool =
6063 	    dma_pool_create("sense pool", &ioc->pdev->dev, sz, 4, 0);
6064 	if (!ioc->sense_dma_pool)
6065 		return -ENOMEM;
6066 	ioc->sense = dma_pool_alloc(ioc->sense_dma_pool,
6067 	    GFP_KERNEL, &ioc->sense_dma);
6068 	if (!ioc->sense)
6069 		return -EAGAIN;
6070 	if (!mpt3sas_check_same_4gb_region(ioc->sense_dma, sz)) {
6071 		dinitprintk(ioc, pr_err(
6072 		    "Bad Sense Pool! sense (0x%p) sense_dma = (0x%llx)\n",
6073 		    ioc->sense, (unsigned long long) ioc->sense_dma));
6074 		ioc->use_32bit_dma = true;
6075 		return -EAGAIN;
6076 	}
6077 	ioc_info(ioc,
6078 	    "sense pool(0x%p) - dma(0x%llx): depth(%d), element_size(%d), pool_size (%d kB)\n",
6079 	    ioc->sense, (unsigned long long)ioc->sense_dma,
6080 	    ioc->scsiio_depth, SCSI_SENSE_BUFFERSIZE, sz/1024);
6081 	return 0;
6082 }
6083 
6084 /**
6085  * _base_allocate_reply_pool - Allocating DMA'able memory
6086  *			for reply pool.
6087  * @ioc: Adapter object
6088  * @sz: DMA Pool size
6089  * Return: 0 for success, non-zero for failure.
6090  */
6091 static int
6092 _base_allocate_reply_pool(struct MPT3SAS_ADAPTER *ioc, u32 sz)
6093 {
6094 	/* reply pool, 4 byte align */
6095 	ioc->reply_dma_pool = dma_pool_create("reply pool",
6096 	    &ioc->pdev->dev, sz, 4, 0);
6097 	if (!ioc->reply_dma_pool)
6098 		return -ENOMEM;
6099 	ioc->reply = dma_pool_alloc(ioc->reply_dma_pool, GFP_KERNEL,
6100 	    &ioc->reply_dma);
6101 	if (!ioc->reply)
6102 		return -EAGAIN;
6103 	if (!mpt3sas_check_same_4gb_region(ioc->reply_dma, sz)) {
6104 		dinitprintk(ioc, pr_err(
6105 		    "Bad Reply Pool! Reply (0x%p) Reply dma = (0x%llx)\n",
6106 		    ioc->reply, (unsigned long long) ioc->reply_dma));
6107 		ioc->use_32bit_dma = true;
6108 		return -EAGAIN;
6109 	}
6110 	ioc->reply_dma_min_address = (u32)(ioc->reply_dma);
6111 	ioc->reply_dma_max_address = (u32)(ioc->reply_dma) + sz;
6112 	ioc_info(ioc,
6113 	    "reply pool(0x%p) - dma(0x%llx): depth(%d), frame_size(%d), pool_size(%d kB)\n",
6114 	    ioc->reply, (unsigned long long)ioc->reply_dma,
6115 	    ioc->reply_free_queue_depth, ioc->reply_sz, sz/1024);
6116 	return 0;
6117 }
6118 
6119 /**
6120  * _base_allocate_reply_free_dma_pool - Allocating DMA'able memory
6121  *			for reply free dma pool.
6122  * @ioc: Adapter object
6123  * @sz: DMA Pool size
6124  * Return: 0 for success, non-zero for failure.
6125  */
6126 static int
6127 _base_allocate_reply_free_dma_pool(struct MPT3SAS_ADAPTER *ioc, u32 sz)
6128 {
6129 	/* reply free queue, 16 byte align */
6130 	ioc->reply_free_dma_pool = dma_pool_create(
6131 	    "reply_free pool", &ioc->pdev->dev, sz, 16, 0);
6132 	if (!ioc->reply_free_dma_pool)
6133 		return -ENOMEM;
6134 	ioc->reply_free = dma_pool_alloc(ioc->reply_free_dma_pool,
6135 	    GFP_KERNEL, &ioc->reply_free_dma);
6136 	if (!ioc->reply_free)
6137 		return -EAGAIN;
6138 	if (!mpt3sas_check_same_4gb_region(ioc->reply_free_dma, sz)) {
6139 		dinitprintk(ioc,
6140 		    pr_err("Bad Reply Free Pool! Reply Free (0x%p) Reply Free dma = (0x%llx)\n",
6141 		    ioc->reply_free, (unsigned long long) ioc->reply_free_dma));
6142 		ioc->use_32bit_dma = true;
6143 		return -EAGAIN;
6144 	}
6145 	memset(ioc->reply_free, 0, sz);
6146 	dinitprintk(ioc, ioc_info(ioc,
6147 	    "reply_free pool(0x%p): depth(%d), element_size(%d), pool_size(%d kB)\n",
6148 	    ioc->reply_free, ioc->reply_free_queue_depth, 4, sz/1024));
6149 	dinitprintk(ioc, ioc_info(ioc,
6150 	    "reply_free_dma (0x%llx)\n",
6151 	    (unsigned long long)ioc->reply_free_dma));
6152 	return 0;
6153 }
6154 
6155 /**
6156  * _base_allocate_reply_post_free_array - Allocating DMA'able memory
6157  *			for reply post free array.
6158  * @ioc: Adapter object
6159  * @reply_post_free_array_sz: DMA Pool size
6160  * Return: 0 for success, non-zero for failure.
6161  */
6162 
6163 static int
6164 _base_allocate_reply_post_free_array(struct MPT3SAS_ADAPTER *ioc,
6165 	u32 reply_post_free_array_sz)
6166 {
6167 	ioc->reply_post_free_array_dma_pool =
6168 	    dma_pool_create("reply_post_free_array pool",
6169 	    &ioc->pdev->dev, reply_post_free_array_sz, 16, 0);
6170 	if (!ioc->reply_post_free_array_dma_pool)
6171 		return -ENOMEM;
6172 	ioc->reply_post_free_array =
6173 	    dma_pool_alloc(ioc->reply_post_free_array_dma_pool,
6174 	    GFP_KERNEL, &ioc->reply_post_free_array_dma);
6175 	if (!ioc->reply_post_free_array)
6176 		return -EAGAIN;
6177 	if (!mpt3sas_check_same_4gb_region(ioc->reply_post_free_array_dma,
6178 	    reply_post_free_array_sz)) {
6179 		dinitprintk(ioc, pr_err(
6180 		    "Bad Reply Free Pool! Reply Free (0x%p) Reply Free dma = (0x%llx)\n",
6181 		    ioc->reply_free,
6182 		    (unsigned long long) ioc->reply_free_dma));
6183 		ioc->use_32bit_dma = true;
6184 		return -EAGAIN;
6185 	}
6186 	return 0;
6187 }
6188 /**
6189  * base_alloc_rdpq_dma_pool - Allocating DMA'able memory
6190  *                     for reply queues.
6191  * @ioc: per adapter object
6192  * @sz: DMA Pool size
6193  * Return: 0 for success, non-zero for failure.
6194  */
6195 static int
6196 base_alloc_rdpq_dma_pool(struct MPT3SAS_ADAPTER *ioc, int sz)
6197 {
6198 	int i = 0;
6199 	u32 dma_alloc_count = 0;
6200 	int reply_post_free_sz = ioc->reply_post_queue_depth *
6201 		sizeof(Mpi2DefaultReplyDescriptor_t);
6202 	int count = ioc->rdpq_array_enable ? ioc->reply_queue_count : 1;
6203 
6204 	ioc->reply_post = kcalloc(count, sizeof(struct reply_post_struct),
6205 			GFP_KERNEL);
6206 	if (!ioc->reply_post)
6207 		return -ENOMEM;
6208 	/*
6209 	 *  For INVADER_SERIES each set of 8 reply queues(0-7, 8-15, ..) and
6210 	 *  VENTURA_SERIES each set of 16 reply queues(0-15, 16-31, ..) should
6211 	 *  be within 4GB boundary i.e reply queues in a set must have same
6212 	 *  upper 32-bits in their memory address. so here driver is allocating
6213 	 *  the DMA'able memory for reply queues according.
6214 	 *  Driver uses limitation of
6215 	 *  VENTURA_SERIES to manage INVADER_SERIES as well.
6216 	 */
6217 	dma_alloc_count = DIV_ROUND_UP(count,
6218 				RDPQ_MAX_INDEX_IN_ONE_CHUNK);
6219 	ioc->reply_post_free_dma_pool =
6220 		dma_pool_create("reply_post_free pool",
6221 		    &ioc->pdev->dev, sz, 16, 0);
6222 	if (!ioc->reply_post_free_dma_pool)
6223 		return -ENOMEM;
6224 	for (i = 0; i < count; i++) {
6225 		if ((i % RDPQ_MAX_INDEX_IN_ONE_CHUNK == 0) && dma_alloc_count) {
6226 			ioc->reply_post[i].reply_post_free =
6227 			    dma_pool_zalloc(ioc->reply_post_free_dma_pool,
6228 				GFP_KERNEL,
6229 				&ioc->reply_post[i].reply_post_free_dma);
6230 			if (!ioc->reply_post[i].reply_post_free)
6231 				return -ENOMEM;
6232 			/*
6233 			 * Each set of RDPQ pool must satisfy 4gb boundary
6234 			 * restriction.
6235 			 * 1) Check if allocated resources for RDPQ pool are in
6236 			 *	the same 4GB range.
6237 			 * 2) If #1 is true, continue with 64 bit DMA.
6238 			 * 3) If #1 is false, return 1. which means free all the
6239 			 * resources and set DMA mask to 32 and allocate.
6240 			 */
6241 			if (!mpt3sas_check_same_4gb_region(
6242 				ioc->reply_post[i].reply_post_free_dma, sz)) {
6243 				dinitprintk(ioc,
6244 				    ioc_err(ioc, "bad Replypost free pool(0x%p)"
6245 				    "reply_post_free_dma = (0x%llx)\n",
6246 				    ioc->reply_post[i].reply_post_free,
6247 				    (unsigned long long)
6248 				    ioc->reply_post[i].reply_post_free_dma));
6249 				return -EAGAIN;
6250 			}
6251 			dma_alloc_count--;
6252 
6253 		} else {
6254 			ioc->reply_post[i].reply_post_free =
6255 			    (Mpi2ReplyDescriptorsUnion_t *)
6256 			    ((long)ioc->reply_post[i-1].reply_post_free
6257 			    + reply_post_free_sz);
6258 			ioc->reply_post[i].reply_post_free_dma =
6259 			    (dma_addr_t)
6260 			    (ioc->reply_post[i-1].reply_post_free_dma +
6261 			    reply_post_free_sz);
6262 		}
6263 	}
6264 	return 0;
6265 }
6266 
6267 /**
6268  * _base_allocate_memory_pools - allocate start of day memory pools
6269  * @ioc: per adapter object
6270  *
6271  * Return: 0 success, anything else error.
6272  */
6273 static int
6274 _base_allocate_memory_pools(struct MPT3SAS_ADAPTER *ioc)
6275 {
6276 	struct mpt3sas_facts *facts;
6277 	u16 max_sge_elements;
6278 	u16 chains_needed_per_io;
6279 	u32 sz, total_sz, reply_post_free_sz, reply_post_free_array_sz;
6280 	u32 retry_sz;
6281 	u32 rdpq_sz = 0, sense_sz = 0;
6282 	u16 max_request_credit, nvme_blocks_needed;
6283 	unsigned short sg_tablesize;
6284 	u16 sge_size;
6285 	int i;
6286 	int ret = 0, rc = 0;
6287 
6288 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
6289 
6290 
6291 	retry_sz = 0;
6292 	facts = &ioc->facts;
6293 
6294 	/* command line tunables for max sgl entries */
6295 	if (max_sgl_entries != -1)
6296 		sg_tablesize = max_sgl_entries;
6297 	else {
6298 		if (ioc->hba_mpi_version_belonged == MPI2_VERSION)
6299 			sg_tablesize = MPT2SAS_SG_DEPTH;
6300 		else
6301 			sg_tablesize = MPT3SAS_SG_DEPTH;
6302 	}
6303 
6304 	/* max sgl entries <= MPT_KDUMP_MIN_PHYS_SEGMENTS in KDUMP mode */
6305 	if (reset_devices)
6306 		sg_tablesize = min_t(unsigned short, sg_tablesize,
6307 		   MPT_KDUMP_MIN_PHYS_SEGMENTS);
6308 
6309 	if (ioc->is_mcpu_endpoint)
6310 		ioc->shost->sg_tablesize = MPT_MIN_PHYS_SEGMENTS;
6311 	else {
6312 		if (sg_tablesize < MPT_MIN_PHYS_SEGMENTS)
6313 			sg_tablesize = MPT_MIN_PHYS_SEGMENTS;
6314 		else if (sg_tablesize > MPT_MAX_PHYS_SEGMENTS) {
6315 			sg_tablesize = min_t(unsigned short, sg_tablesize,
6316 					SG_MAX_SEGMENTS);
6317 			ioc_warn(ioc, "sg_tablesize(%u) is bigger than kernel defined SG_CHUNK_SIZE(%u)\n",
6318 				 sg_tablesize, MPT_MAX_PHYS_SEGMENTS);
6319 		}
6320 		ioc->shost->sg_tablesize = sg_tablesize;
6321 	}
6322 
6323 	ioc->internal_depth = min_t(int, (facts->HighPriorityCredit + (5)),
6324 		(facts->RequestCredit / 4));
6325 	if (ioc->internal_depth < INTERNAL_CMDS_COUNT) {
6326 		if (facts->RequestCredit <= (INTERNAL_CMDS_COUNT +
6327 				INTERNAL_SCSIIO_CMDS_COUNT)) {
6328 			ioc_err(ioc, "IOC doesn't have enough Request Credits, it has just %d number of credits\n",
6329 				facts->RequestCredit);
6330 			return -ENOMEM;
6331 		}
6332 		ioc->internal_depth = 10;
6333 	}
6334 
6335 	ioc->hi_priority_depth = ioc->internal_depth - (5);
6336 	/* command line tunables  for max controller queue depth */
6337 	if (max_queue_depth != -1 && max_queue_depth != 0) {
6338 		max_request_credit = min_t(u16, max_queue_depth +
6339 			ioc->internal_depth, facts->RequestCredit);
6340 		if (max_request_credit > MAX_HBA_QUEUE_DEPTH)
6341 			max_request_credit =  MAX_HBA_QUEUE_DEPTH;
6342 	} else if (reset_devices)
6343 		max_request_credit = min_t(u16, facts->RequestCredit,
6344 		    (MPT3SAS_KDUMP_SCSI_IO_DEPTH + ioc->internal_depth));
6345 	else
6346 		max_request_credit = min_t(u16, facts->RequestCredit,
6347 		    MAX_HBA_QUEUE_DEPTH);
6348 
6349 	/* Firmware maintains additional facts->HighPriorityCredit number of
6350 	 * credits for HiPriprity Request messages, so hba queue depth will be
6351 	 * sum of max_request_credit and high priority queue depth.
6352 	 */
6353 	ioc->hba_queue_depth = max_request_credit + ioc->hi_priority_depth;
6354 
6355 	/* request frame size */
6356 	ioc->request_sz = facts->IOCRequestFrameSize * 4;
6357 
6358 	/* reply frame size */
6359 	ioc->reply_sz = facts->ReplyFrameSize * 4;
6360 
6361 	/* chain segment size */
6362 	if (ioc->hba_mpi_version_belonged != MPI2_VERSION) {
6363 		if (facts->IOCMaxChainSegmentSize)
6364 			ioc->chain_segment_sz =
6365 					facts->IOCMaxChainSegmentSize *
6366 					MAX_CHAIN_ELEMT_SZ;
6367 		else
6368 		/* set to 128 bytes size if IOCMaxChainSegmentSize is zero */
6369 			ioc->chain_segment_sz = DEFAULT_NUM_FWCHAIN_ELEMTS *
6370 						    MAX_CHAIN_ELEMT_SZ;
6371 	} else
6372 		ioc->chain_segment_sz = ioc->request_sz;
6373 
6374 	/* calculate the max scatter element size */
6375 	sge_size = max_t(u16, ioc->sge_size, ioc->sge_size_ieee);
6376 
6377  retry_allocation:
6378 	total_sz = 0;
6379 	/* calculate number of sg elements left over in the 1st frame */
6380 	max_sge_elements = ioc->request_sz - ((sizeof(Mpi2SCSIIORequest_t) -
6381 	    sizeof(Mpi2SGEIOUnion_t)) + sge_size);
6382 	ioc->max_sges_in_main_message = max_sge_elements/sge_size;
6383 
6384 	/* now do the same for a chain buffer */
6385 	max_sge_elements = ioc->chain_segment_sz - sge_size;
6386 	ioc->max_sges_in_chain_message = max_sge_elements/sge_size;
6387 
6388 	/*
6389 	 *  MPT3SAS_SG_DEPTH = CONFIG_FUSION_MAX_SGE
6390 	 */
6391 	chains_needed_per_io = ((ioc->shost->sg_tablesize -
6392 	   ioc->max_sges_in_main_message)/ioc->max_sges_in_chain_message)
6393 	    + 1;
6394 	if (chains_needed_per_io > facts->MaxChainDepth) {
6395 		chains_needed_per_io = facts->MaxChainDepth;
6396 		ioc->shost->sg_tablesize = min_t(u16,
6397 		ioc->max_sges_in_main_message + (ioc->max_sges_in_chain_message
6398 		* chains_needed_per_io), ioc->shost->sg_tablesize);
6399 	}
6400 	ioc->chains_needed_per_io = chains_needed_per_io;
6401 
6402 	/* reply free queue sizing - taking into account for 64 FW events */
6403 	ioc->reply_free_queue_depth = ioc->hba_queue_depth + 64;
6404 
6405 	/* mCPU manage single counters for simplicity */
6406 	if (ioc->is_mcpu_endpoint)
6407 		ioc->reply_post_queue_depth = ioc->reply_free_queue_depth;
6408 	else {
6409 		/* calculate reply descriptor post queue depth */
6410 		ioc->reply_post_queue_depth = ioc->hba_queue_depth +
6411 			ioc->reply_free_queue_depth +  1;
6412 		/* align the reply post queue on the next 16 count boundary */
6413 		if (ioc->reply_post_queue_depth % 16)
6414 			ioc->reply_post_queue_depth += 16 -
6415 				(ioc->reply_post_queue_depth % 16);
6416 	}
6417 
6418 	if (ioc->reply_post_queue_depth >
6419 	    facts->MaxReplyDescriptorPostQueueDepth) {
6420 		ioc->reply_post_queue_depth =
6421 				facts->MaxReplyDescriptorPostQueueDepth -
6422 		    (facts->MaxReplyDescriptorPostQueueDepth % 16);
6423 		ioc->hba_queue_depth =
6424 				((ioc->reply_post_queue_depth - 64) / 2) - 1;
6425 		ioc->reply_free_queue_depth = ioc->hba_queue_depth + 64;
6426 	}
6427 
6428 	ioc_info(ioc,
6429 	    "scatter gather: sge_in_main_msg(%d), sge_per_chain(%d), "
6430 	    "sge_per_io(%d), chains_per_io(%d)\n",
6431 	    ioc->max_sges_in_main_message,
6432 	    ioc->max_sges_in_chain_message,
6433 	    ioc->shost->sg_tablesize,
6434 	    ioc->chains_needed_per_io);
6435 
6436 	/* reply post queue, 16 byte align */
6437 	reply_post_free_sz = ioc->reply_post_queue_depth *
6438 	    sizeof(Mpi2DefaultReplyDescriptor_t);
6439 	rdpq_sz = reply_post_free_sz * RDPQ_MAX_INDEX_IN_ONE_CHUNK;
6440 	if ((_base_is_controller_msix_enabled(ioc) && !ioc->rdpq_array_enable)
6441 	    || (ioc->reply_queue_count < RDPQ_MAX_INDEX_IN_ONE_CHUNK))
6442 		rdpq_sz = reply_post_free_sz * ioc->reply_queue_count;
6443 	ret = base_alloc_rdpq_dma_pool(ioc, rdpq_sz);
6444 	if (ret == -EAGAIN) {
6445 		/*
6446 		 * Free allocated bad RDPQ memory pools.
6447 		 * Change dma coherent mask to 32 bit and reallocate RDPQ
6448 		 */
6449 		_base_release_memory_pools(ioc);
6450 		ioc->use_32bit_dma = true;
6451 		if (_base_config_dma_addressing(ioc, ioc->pdev) != 0) {
6452 			ioc_err(ioc,
6453 			    "32 DMA mask failed %s\n", pci_name(ioc->pdev));
6454 			return -ENODEV;
6455 		}
6456 		if (base_alloc_rdpq_dma_pool(ioc, rdpq_sz))
6457 			return -ENOMEM;
6458 	} else if (ret == -ENOMEM)
6459 		return -ENOMEM;
6460 	total_sz = rdpq_sz * (!ioc->rdpq_array_enable ? 1 :
6461 	    DIV_ROUND_UP(ioc->reply_queue_count, RDPQ_MAX_INDEX_IN_ONE_CHUNK));
6462 	ioc->scsiio_depth = ioc->hba_queue_depth -
6463 	    ioc->hi_priority_depth - ioc->internal_depth;
6464 
6465 	/* set the scsi host can_queue depth
6466 	 * with some internal commands that could be outstanding
6467 	 */
6468 	ioc->shost->can_queue = ioc->scsiio_depth - INTERNAL_SCSIIO_CMDS_COUNT;
6469 	dinitprintk(ioc,
6470 		    ioc_info(ioc, "scsi host: can_queue depth (%d)\n",
6471 			     ioc->shost->can_queue));
6472 
6473 	/* contiguous pool for request and chains, 16 byte align, one extra "
6474 	 * "frame for smid=0
6475 	 */
6476 	ioc->chain_depth = ioc->chains_needed_per_io * ioc->scsiio_depth;
6477 	sz = ((ioc->scsiio_depth + 1) * ioc->request_sz);
6478 
6479 	/* hi-priority queue */
6480 	sz += (ioc->hi_priority_depth * ioc->request_sz);
6481 
6482 	/* internal queue */
6483 	sz += (ioc->internal_depth * ioc->request_sz);
6484 
6485 	ioc->request_dma_sz = sz;
6486 	ioc->request = dma_alloc_coherent(&ioc->pdev->dev, sz,
6487 			&ioc->request_dma, GFP_KERNEL);
6488 	if (!ioc->request) {
6489 		ioc_err(ioc, "request pool: dma_alloc_coherent failed: hba_depth(%d), chains_per_io(%d), frame_sz(%d), total(%d kB)\n",
6490 			ioc->hba_queue_depth, ioc->chains_needed_per_io,
6491 			ioc->request_sz, sz / 1024);
6492 		if (ioc->scsiio_depth < MPT3SAS_SAS_QUEUE_DEPTH)
6493 			goto out;
6494 		retry_sz = 64;
6495 		ioc->hba_queue_depth -= retry_sz;
6496 		_base_release_memory_pools(ioc);
6497 		goto retry_allocation;
6498 	}
6499 
6500 	if (retry_sz)
6501 		ioc_err(ioc, "request pool: dma_alloc_coherent succeed: hba_depth(%d), chains_per_io(%d), frame_sz(%d), total(%d kb)\n",
6502 			ioc->hba_queue_depth, ioc->chains_needed_per_io,
6503 			ioc->request_sz, sz / 1024);
6504 
6505 	/* hi-priority queue */
6506 	ioc->hi_priority = ioc->request + ((ioc->scsiio_depth + 1) *
6507 	    ioc->request_sz);
6508 	ioc->hi_priority_dma = ioc->request_dma + ((ioc->scsiio_depth + 1) *
6509 	    ioc->request_sz);
6510 
6511 	/* internal queue */
6512 	ioc->internal = ioc->hi_priority + (ioc->hi_priority_depth *
6513 	    ioc->request_sz);
6514 	ioc->internal_dma = ioc->hi_priority_dma + (ioc->hi_priority_depth *
6515 	    ioc->request_sz);
6516 
6517 	ioc_info(ioc,
6518 	    "request pool(0x%p) - dma(0x%llx): "
6519 	    "depth(%d), frame_size(%d), pool_size(%d kB)\n",
6520 	    ioc->request, (unsigned long long) ioc->request_dma,
6521 	    ioc->hba_queue_depth, ioc->request_sz,
6522 	    (ioc->hba_queue_depth * ioc->request_sz) / 1024);
6523 
6524 	total_sz += sz;
6525 
6526 	dinitprintk(ioc,
6527 		    ioc_info(ioc, "scsiio(0x%p): depth(%d)\n",
6528 			     ioc->request, ioc->scsiio_depth));
6529 
6530 	ioc->chain_depth = min_t(u32, ioc->chain_depth, MAX_CHAIN_DEPTH);
6531 	sz = ioc->scsiio_depth * sizeof(struct chain_lookup);
6532 	ioc->chain_lookup = kzalloc(sz, GFP_KERNEL);
6533 	if (!ioc->chain_lookup) {
6534 		ioc_err(ioc, "chain_lookup: __get_free_pages failed\n");
6535 		goto out;
6536 	}
6537 
6538 	sz = ioc->chains_needed_per_io * sizeof(struct chain_tracker);
6539 	for (i = 0; i < ioc->scsiio_depth; i++) {
6540 		ioc->chain_lookup[i].chains_per_smid = kzalloc(sz, GFP_KERNEL);
6541 		if (!ioc->chain_lookup[i].chains_per_smid) {
6542 			ioc_err(ioc, "chain_lookup: kzalloc failed\n");
6543 			goto out;
6544 		}
6545 	}
6546 
6547 	/* initialize hi-priority queue smid's */
6548 	ioc->hpr_lookup = kcalloc(ioc->hi_priority_depth,
6549 	    sizeof(struct request_tracker), GFP_KERNEL);
6550 	if (!ioc->hpr_lookup) {
6551 		ioc_err(ioc, "hpr_lookup: kcalloc failed\n");
6552 		goto out;
6553 	}
6554 	ioc->hi_priority_smid = ioc->scsiio_depth + 1;
6555 	dinitprintk(ioc,
6556 		    ioc_info(ioc, "hi_priority(0x%p): depth(%d), start smid(%d)\n",
6557 			     ioc->hi_priority,
6558 			     ioc->hi_priority_depth, ioc->hi_priority_smid));
6559 
6560 	/* initialize internal queue smid's */
6561 	ioc->internal_lookup = kcalloc(ioc->internal_depth,
6562 	    sizeof(struct request_tracker), GFP_KERNEL);
6563 	if (!ioc->internal_lookup) {
6564 		ioc_err(ioc, "internal_lookup: kcalloc failed\n");
6565 		goto out;
6566 	}
6567 	ioc->internal_smid = ioc->hi_priority_smid + ioc->hi_priority_depth;
6568 	dinitprintk(ioc,
6569 		    ioc_info(ioc, "internal(0x%p): depth(%d), start smid(%d)\n",
6570 			     ioc->internal,
6571 			     ioc->internal_depth, ioc->internal_smid));
6572 
6573 	ioc->io_queue_num = kcalloc(ioc->scsiio_depth,
6574 	    sizeof(u16), GFP_KERNEL);
6575 	if (!ioc->io_queue_num)
6576 		goto out;
6577 	/*
6578 	 * The number of NVMe page sized blocks needed is:
6579 	 *     (((sg_tablesize * 8) - 1) / (page_size - 8)) + 1
6580 	 * ((sg_tablesize * 8) - 1) is the max PRP's minus the first PRP entry
6581 	 * that is placed in the main message frame.  8 is the size of each PRP
6582 	 * entry or PRP list pointer entry.  8 is subtracted from page_size
6583 	 * because of the PRP list pointer entry at the end of a page, so this
6584 	 * is not counted as a PRP entry.  The 1 added page is a round up.
6585 	 *
6586 	 * To avoid allocation failures due to the amount of memory that could
6587 	 * be required for NVMe PRP's, only each set of NVMe blocks will be
6588 	 * contiguous, so a new set is allocated for each possible I/O.
6589 	 */
6590 
6591 	ioc->chains_per_prp_buffer = 0;
6592 	if (ioc->facts.ProtocolFlags & MPI2_IOCFACTS_PROTOCOL_NVME_DEVICES) {
6593 		nvme_blocks_needed =
6594 			(ioc->shost->sg_tablesize * NVME_PRP_SIZE) - 1;
6595 		nvme_blocks_needed /= (ioc->page_size - NVME_PRP_SIZE);
6596 		nvme_blocks_needed++;
6597 
6598 		sz = sizeof(struct pcie_sg_list) * ioc->scsiio_depth;
6599 		ioc->pcie_sg_lookup = kzalloc(sz, GFP_KERNEL);
6600 		if (!ioc->pcie_sg_lookup) {
6601 			ioc_info(ioc, "PCIe SGL lookup: kzalloc failed\n");
6602 			goto out;
6603 		}
6604 		sz = nvme_blocks_needed * ioc->page_size;
6605 		rc = _base_allocate_pcie_sgl_pool(ioc, sz);
6606 		if (rc == -ENOMEM)
6607 			return -ENOMEM;
6608 		else if (rc == -EAGAIN)
6609 			goto try_32bit_dma;
6610 		total_sz += sz * ioc->scsiio_depth;
6611 	}
6612 
6613 	rc = _base_allocate_chain_dma_pool(ioc, ioc->chain_segment_sz);
6614 	if (rc == -ENOMEM)
6615 		return -ENOMEM;
6616 	else if (rc == -EAGAIN)
6617 		goto try_32bit_dma;
6618 	total_sz += ioc->chain_segment_sz * ((ioc->chains_needed_per_io -
6619 		ioc->chains_per_prp_buffer) * ioc->scsiio_depth);
6620 	dinitprintk(ioc,
6621 	    ioc_info(ioc, "chain pool depth(%d), frame_size(%d), pool_size(%d kB)\n",
6622 	    ioc->chain_depth, ioc->chain_segment_sz,
6623 	    (ioc->chain_depth * ioc->chain_segment_sz) / 1024));
6624 	/* sense buffers, 4 byte align */
6625 	sense_sz = ioc->scsiio_depth * SCSI_SENSE_BUFFERSIZE;
6626 	rc = _base_allocate_sense_dma_pool(ioc, sense_sz);
6627 	if (rc  == -ENOMEM)
6628 		return -ENOMEM;
6629 	else if (rc == -EAGAIN)
6630 		goto try_32bit_dma;
6631 	total_sz += sense_sz;
6632 	/* reply pool, 4 byte align */
6633 	sz = ioc->reply_free_queue_depth * ioc->reply_sz;
6634 	rc = _base_allocate_reply_pool(ioc, sz);
6635 	if (rc == -ENOMEM)
6636 		return -ENOMEM;
6637 	else if (rc == -EAGAIN)
6638 		goto try_32bit_dma;
6639 	total_sz += sz;
6640 
6641 	/* reply free queue, 16 byte align */
6642 	sz = ioc->reply_free_queue_depth * 4;
6643 	rc = _base_allocate_reply_free_dma_pool(ioc, sz);
6644 	if (rc  == -ENOMEM)
6645 		return -ENOMEM;
6646 	else if (rc == -EAGAIN)
6647 		goto try_32bit_dma;
6648 	dinitprintk(ioc,
6649 		    ioc_info(ioc, "reply_free_dma (0x%llx)\n",
6650 			     (unsigned long long)ioc->reply_free_dma));
6651 	total_sz += sz;
6652 	if (ioc->rdpq_array_enable) {
6653 		reply_post_free_array_sz = ioc->reply_queue_count *
6654 		    sizeof(Mpi2IOCInitRDPQArrayEntry);
6655 		rc = _base_allocate_reply_post_free_array(ioc,
6656 		    reply_post_free_array_sz);
6657 		if (rc == -ENOMEM)
6658 			return -ENOMEM;
6659 		else if (rc == -EAGAIN)
6660 			goto try_32bit_dma;
6661 	}
6662 	ioc->config_page_sz = 512;
6663 	ioc->config_page = dma_alloc_coherent(&ioc->pdev->dev,
6664 			ioc->config_page_sz, &ioc->config_page_dma, GFP_KERNEL);
6665 	if (!ioc->config_page) {
6666 		ioc_err(ioc, "config page: dma_pool_alloc failed\n");
6667 		goto out;
6668 	}
6669 
6670 	ioc_info(ioc, "config page(0x%p) - dma(0x%llx): size(%d)\n",
6671 	    ioc->config_page, (unsigned long long)ioc->config_page_dma,
6672 	    ioc->config_page_sz);
6673 	total_sz += ioc->config_page_sz;
6674 
6675 	ioc_info(ioc, "Allocated physical memory: size(%d kB)\n",
6676 		 total_sz / 1024);
6677 	ioc_info(ioc, "Current Controller Queue Depth(%d),Max Controller Queue Depth(%d)\n",
6678 		 ioc->shost->can_queue, facts->RequestCredit);
6679 	ioc_info(ioc, "Scatter Gather Elements per IO(%d)\n",
6680 		 ioc->shost->sg_tablesize);
6681 	return 0;
6682 
6683 try_32bit_dma:
6684 	_base_release_memory_pools(ioc);
6685 	if (ioc->use_32bit_dma && (ioc->dma_mask > 32)) {
6686 		/* Change dma coherent mask to 32 bit and reallocate */
6687 		if (_base_config_dma_addressing(ioc, ioc->pdev) != 0) {
6688 			pr_err("Setting 32 bit coherent DMA mask Failed %s\n",
6689 			    pci_name(ioc->pdev));
6690 			return -ENODEV;
6691 		}
6692 	} else if (_base_reduce_hba_queue_depth(ioc) != 0)
6693 		return -ENOMEM;
6694 	goto retry_allocation;
6695 
6696  out:
6697 	return -ENOMEM;
6698 }
6699 
6700 /**
6701  * mpt3sas_base_get_iocstate - Get the current state of a MPT adapter.
6702  * @ioc: Pointer to MPT_ADAPTER structure
6703  * @cooked: Request raw or cooked IOC state
6704  *
6705  * Return: all IOC Doorbell register bits if cooked==0, else just the
6706  * Doorbell bits in MPI_IOC_STATE_MASK.
6707  */
6708 u32
6709 mpt3sas_base_get_iocstate(struct MPT3SAS_ADAPTER *ioc, int cooked)
6710 {
6711 	u32 s, sc;
6712 
6713 	s = ioc->base_readl_ext_retry(&ioc->chip->Doorbell);
6714 	sc = s & MPI2_IOC_STATE_MASK;
6715 	return cooked ? sc : s;
6716 }
6717 
6718 /**
6719  * _base_wait_on_iocstate - waiting on a particular ioc state
6720  * @ioc: ?
6721  * @ioc_state: controller state { READY, OPERATIONAL, or RESET }
6722  * @timeout: timeout in second
6723  *
6724  * Return: 0 for success, non-zero for failure.
6725  */
6726 static int
6727 _base_wait_on_iocstate(struct MPT3SAS_ADAPTER *ioc, u32 ioc_state, int timeout)
6728 {
6729 	u32 count, cntdn;
6730 	u32 current_state;
6731 
6732 	count = 0;
6733 	cntdn = 1000 * timeout;
6734 	do {
6735 		current_state = mpt3sas_base_get_iocstate(ioc, 1);
6736 		if (current_state == ioc_state)
6737 			return 0;
6738 		if (count && current_state == MPI2_IOC_STATE_FAULT)
6739 			break;
6740 		if (count && current_state == MPI2_IOC_STATE_COREDUMP)
6741 			break;
6742 
6743 		usleep_range(1000, 1500);
6744 		count++;
6745 	} while (--cntdn);
6746 
6747 	return current_state;
6748 }
6749 
6750 /**
6751  * _base_dump_reg_set -	This function will print hexdump of register set.
6752  * @ioc: per adapter object
6753  *
6754  * Return: nothing.
6755  */
6756 static inline void
6757 _base_dump_reg_set(struct MPT3SAS_ADAPTER *ioc)
6758 {
6759 	unsigned int i, sz = 256;
6760 	u32 __iomem *reg = (u32 __iomem *)ioc->chip;
6761 
6762 	ioc_info(ioc, "System Register set:\n");
6763 	for (i = 0; i < (sz / sizeof(u32)); i++)
6764 		pr_info("%08x: %08x\n", (i * 4), readl(&reg[i]));
6765 }
6766 
6767 /**
6768  * _base_wait_for_doorbell_int - waiting for controller interrupt(generated by
6769  * a write to the doorbell)
6770  * @ioc: per adapter object
6771  * @timeout: timeout in seconds
6772  *
6773  * Return: 0 for success, non-zero for failure.
6774  *
6775  * Notes: MPI2_HIS_IOC2SYS_DB_STATUS - set to one when IOC writes to doorbell.
6776  */
6777 
6778 static int
6779 _base_wait_for_doorbell_int(struct MPT3SAS_ADAPTER *ioc, int timeout)
6780 {
6781 	u32 cntdn, count;
6782 	u32 int_status;
6783 
6784 	count = 0;
6785 	cntdn = 1000 * timeout;
6786 	do {
6787 		int_status = ioc->base_readl(&ioc->chip->HostInterruptStatus);
6788 		if (int_status & MPI2_HIS_IOC2SYS_DB_STATUS) {
6789 			dhsprintk(ioc,
6790 				  ioc_info(ioc, "%s: successful count(%d), timeout(%d)\n",
6791 					   __func__, count, timeout));
6792 			return 0;
6793 		}
6794 
6795 		usleep_range(1000, 1500);
6796 		count++;
6797 	} while (--cntdn);
6798 
6799 	ioc_err(ioc, "%s: failed due to timeout count(%d), int_status(%x)!\n",
6800 		__func__, count, int_status);
6801 	return -EFAULT;
6802 }
6803 
6804 static int
6805 _base_spin_on_doorbell_int(struct MPT3SAS_ADAPTER *ioc, int timeout)
6806 {
6807 	u32 cntdn, count;
6808 	u32 int_status;
6809 
6810 	count = 0;
6811 	cntdn = 2000 * timeout;
6812 	do {
6813 		int_status = ioc->base_readl(&ioc->chip->HostInterruptStatus);
6814 		if (int_status & MPI2_HIS_IOC2SYS_DB_STATUS) {
6815 			dhsprintk(ioc,
6816 				  ioc_info(ioc, "%s: successful count(%d), timeout(%d)\n",
6817 					   __func__, count, timeout));
6818 			return 0;
6819 		}
6820 
6821 		udelay(500);
6822 		count++;
6823 	} while (--cntdn);
6824 
6825 	ioc_err(ioc, "%s: failed due to timeout count(%d), int_status(%x)!\n",
6826 		__func__, count, int_status);
6827 	return -EFAULT;
6828 
6829 }
6830 
6831 /**
6832  * _base_wait_for_doorbell_ack - waiting for controller to read the doorbell.
6833  * @ioc: per adapter object
6834  * @timeout: timeout in second
6835  *
6836  * Return: 0 for success, non-zero for failure.
6837  *
6838  * Notes: MPI2_HIS_SYS2IOC_DB_STATUS - set to one when host writes to
6839  * doorbell.
6840  */
6841 static int
6842 _base_wait_for_doorbell_ack(struct MPT3SAS_ADAPTER *ioc, int timeout)
6843 {
6844 	u32 cntdn, count;
6845 	u32 int_status;
6846 	u32 doorbell;
6847 
6848 	count = 0;
6849 	cntdn = 1000 * timeout;
6850 	do {
6851 		int_status = ioc->base_readl(&ioc->chip->HostInterruptStatus);
6852 		if (!(int_status & MPI2_HIS_SYS2IOC_DB_STATUS)) {
6853 			dhsprintk(ioc,
6854 				  ioc_info(ioc, "%s: successful count(%d), timeout(%d)\n",
6855 					   __func__, count, timeout));
6856 			return 0;
6857 		} else if (int_status & MPI2_HIS_IOC2SYS_DB_STATUS) {
6858 			doorbell = ioc->base_readl_ext_retry(&ioc->chip->Doorbell);
6859 			if ((doorbell & MPI2_IOC_STATE_MASK) ==
6860 			    MPI2_IOC_STATE_FAULT) {
6861 				mpt3sas_print_fault_code(ioc, doorbell);
6862 				return -EFAULT;
6863 			}
6864 			if ((doorbell & MPI2_IOC_STATE_MASK) ==
6865 			    MPI2_IOC_STATE_COREDUMP) {
6866 				mpt3sas_print_coredump_info(ioc, doorbell);
6867 				return -EFAULT;
6868 			}
6869 		} else if (int_status == 0xFFFFFFFF)
6870 			goto out;
6871 
6872 		usleep_range(1000, 1500);
6873 		count++;
6874 	} while (--cntdn);
6875 
6876  out:
6877 	ioc_err(ioc, "%s: failed due to timeout count(%d), int_status(%x)!\n",
6878 		__func__, count, int_status);
6879 	return -EFAULT;
6880 }
6881 
6882 /**
6883  * _base_wait_for_doorbell_not_used - waiting for doorbell to not be in use
6884  * @ioc: per adapter object
6885  * @timeout: timeout in second
6886  *
6887  * Return: 0 for success, non-zero for failure.
6888  */
6889 static int
6890 _base_wait_for_doorbell_not_used(struct MPT3SAS_ADAPTER *ioc, int timeout)
6891 {
6892 	u32 cntdn, count;
6893 	u32 doorbell_reg;
6894 
6895 	count = 0;
6896 	cntdn = 1000 * timeout;
6897 	do {
6898 		doorbell_reg = ioc->base_readl_ext_retry(&ioc->chip->Doorbell);
6899 		if (!(doorbell_reg & MPI2_DOORBELL_USED)) {
6900 			dhsprintk(ioc,
6901 				  ioc_info(ioc, "%s: successful count(%d), timeout(%d)\n",
6902 					   __func__, count, timeout));
6903 			return 0;
6904 		}
6905 
6906 		usleep_range(1000, 1500);
6907 		count++;
6908 	} while (--cntdn);
6909 
6910 	ioc_err(ioc, "%s: failed due to timeout count(%d), doorbell_reg(%x)!\n",
6911 		__func__, count, doorbell_reg);
6912 	return -EFAULT;
6913 }
6914 
6915 /**
6916  * _base_send_ioc_reset - send doorbell reset
6917  * @ioc: per adapter object
6918  * @reset_type: currently only supports: MPI2_FUNCTION_IOC_MESSAGE_UNIT_RESET
6919  * @timeout: timeout in second
6920  *
6921  * Return: 0 for success, non-zero for failure.
6922  */
6923 static int
6924 _base_send_ioc_reset(struct MPT3SAS_ADAPTER *ioc, u8 reset_type, int timeout)
6925 {
6926 	u32 ioc_state;
6927 	int r = 0;
6928 	unsigned long flags;
6929 
6930 	if (reset_type != MPI2_FUNCTION_IOC_MESSAGE_UNIT_RESET) {
6931 		ioc_err(ioc, "%s: unknown reset_type\n", __func__);
6932 		return -EFAULT;
6933 	}
6934 
6935 	if (!(ioc->facts.IOCCapabilities &
6936 	   MPI2_IOCFACTS_CAPABILITY_EVENT_REPLAY))
6937 		return -EFAULT;
6938 
6939 	ioc_info(ioc, "sending message unit reset !!\n");
6940 
6941 	writel(reset_type << MPI2_DOORBELL_FUNCTION_SHIFT,
6942 	    &ioc->chip->Doorbell);
6943 	if ((_base_wait_for_doorbell_ack(ioc, 15))) {
6944 		r = -EFAULT;
6945 		goto out;
6946 	}
6947 
6948 	ioc_state = _base_wait_on_iocstate(ioc, MPI2_IOC_STATE_READY, timeout);
6949 	if (ioc_state) {
6950 		ioc_err(ioc, "%s: failed going to ready state (ioc_state=0x%x)\n",
6951 			__func__, ioc_state);
6952 		r = -EFAULT;
6953 		goto out;
6954 	}
6955  out:
6956 	if (r != 0) {
6957 		ioc_state = mpt3sas_base_get_iocstate(ioc, 0);
6958 		spin_lock_irqsave(&ioc->ioc_reset_in_progress_lock, flags);
6959 		/*
6960 		 * Wait for IOC state CoreDump to clear only during
6961 		 * HBA initialization & release time.
6962 		 */
6963 		if ((ioc_state & MPI2_IOC_STATE_MASK) ==
6964 		    MPI2_IOC_STATE_COREDUMP && (ioc->is_driver_loading == 1 ||
6965 		    ioc->fault_reset_work_q == NULL)) {
6966 			spin_unlock_irqrestore(
6967 			    &ioc->ioc_reset_in_progress_lock, flags);
6968 			mpt3sas_print_coredump_info(ioc, ioc_state);
6969 			mpt3sas_base_wait_for_coredump_completion(ioc,
6970 			    __func__);
6971 			spin_lock_irqsave(
6972 			    &ioc->ioc_reset_in_progress_lock, flags);
6973 		}
6974 		spin_unlock_irqrestore(&ioc->ioc_reset_in_progress_lock, flags);
6975 	}
6976 	ioc_info(ioc, "message unit reset: %s\n",
6977 		 r == 0 ? "SUCCESS" : "FAILED");
6978 	return r;
6979 }
6980 
6981 /**
6982  * mpt3sas_wait_for_ioc - IOC's operational state is checked here.
6983  * @ioc: per adapter object
6984  * @timeout: timeout in seconds
6985  *
6986  * Return: Waits up to timeout seconds for the IOC to
6987  * become operational. Returns 0 if IOC is present
6988  * and operational; otherwise returns %-EFAULT.
6989  */
6990 
6991 int
6992 mpt3sas_wait_for_ioc(struct MPT3SAS_ADAPTER *ioc, int timeout)
6993 {
6994 	int wait_state_count = 0;
6995 	u32 ioc_state;
6996 
6997 	do {
6998 		ioc_state = mpt3sas_base_get_iocstate(ioc, 1);
6999 		if (ioc_state == MPI2_IOC_STATE_OPERATIONAL)
7000 			break;
7001 
7002 		/*
7003 		 * Watchdog thread will be started after IOC Initialization, so
7004 		 * no need to wait here for IOC state to become operational
7005 		 * when IOC Initialization is on. Instead the driver will
7006 		 * return ETIME status, so that calling function can issue
7007 		 * diag reset operation and retry the command.
7008 		 */
7009 		if (ioc->is_driver_loading)
7010 			return -ETIME;
7011 
7012 		ssleep(1);
7013 		ioc_info(ioc, "%s: waiting for operational state(count=%d)\n",
7014 				__func__, ++wait_state_count);
7015 	} while (--timeout);
7016 	if (!timeout) {
7017 		ioc_err(ioc, "%s: failed due to ioc not operational\n", __func__);
7018 		return -EFAULT;
7019 	}
7020 	if (wait_state_count)
7021 		ioc_info(ioc, "ioc is operational\n");
7022 	return 0;
7023 }
7024 
7025 /**
7026  * _base_handshake_req_reply_wait - send request thru doorbell interface
7027  * @ioc: per adapter object
7028  * @request_bytes: request length
7029  * @request: pointer having request payload
7030  * @reply_bytes: reply length
7031  * @reply: pointer to reply payload
7032  * @timeout: timeout in second
7033  *
7034  * Return: 0 for success, non-zero for failure.
7035  */
7036 static int
7037 _base_handshake_req_reply_wait(struct MPT3SAS_ADAPTER *ioc, int request_bytes,
7038 	u32 *request, int reply_bytes, u16 *reply, int timeout)
7039 {
7040 	MPI2DefaultReply_t *default_reply = (MPI2DefaultReply_t *)reply;
7041 	int i;
7042 	u8 failed;
7043 	__le32 *mfp;
7044 
7045 	/* make sure doorbell is not in use */
7046 	if ((ioc->base_readl_ext_retry(&ioc->chip->Doorbell) & MPI2_DOORBELL_USED)) {
7047 		ioc_err(ioc, "doorbell is in use (line=%d)\n", __LINE__);
7048 		return -EFAULT;
7049 	}
7050 
7051 	/* clear pending doorbell interrupts from previous state changes */
7052 	if (ioc->base_readl(&ioc->chip->HostInterruptStatus) &
7053 	    MPI2_HIS_IOC2SYS_DB_STATUS)
7054 		writel(0, &ioc->chip->HostInterruptStatus);
7055 
7056 	/* send message to ioc */
7057 	writel(((MPI2_FUNCTION_HANDSHAKE<<MPI2_DOORBELL_FUNCTION_SHIFT) |
7058 	    ((request_bytes/4)<<MPI2_DOORBELL_ADD_DWORDS_SHIFT)),
7059 	    &ioc->chip->Doorbell);
7060 
7061 	if ((_base_spin_on_doorbell_int(ioc, 5))) {
7062 		ioc_err(ioc, "doorbell handshake int failed (line=%d)\n",
7063 			__LINE__);
7064 		return -EFAULT;
7065 	}
7066 	writel(0, &ioc->chip->HostInterruptStatus);
7067 
7068 	if ((_base_wait_for_doorbell_ack(ioc, 5))) {
7069 		ioc_err(ioc, "doorbell handshake ack failed (line=%d)\n",
7070 			__LINE__);
7071 		return -EFAULT;
7072 	}
7073 
7074 	/* send message 32-bits at a time */
7075 	for (i = 0, failed = 0; i < request_bytes/4 && !failed; i++) {
7076 		writel(cpu_to_le32(request[i]), &ioc->chip->Doorbell);
7077 		if ((_base_wait_for_doorbell_ack(ioc, 5)))
7078 			failed = 1;
7079 	}
7080 
7081 	if (failed) {
7082 		ioc_err(ioc, "doorbell handshake sending request failed (line=%d)\n",
7083 			__LINE__);
7084 		return -EFAULT;
7085 	}
7086 
7087 	/* now wait for the reply */
7088 	if ((_base_wait_for_doorbell_int(ioc, timeout))) {
7089 		ioc_err(ioc, "doorbell handshake int failed (line=%d)\n",
7090 			__LINE__);
7091 		return -EFAULT;
7092 	}
7093 
7094 	/* read the first two 16-bits, it gives the total length of the reply */
7095 	reply[0] = le16_to_cpu(ioc->base_readl_ext_retry(&ioc->chip->Doorbell)
7096 	    & MPI2_DOORBELL_DATA_MASK);
7097 	writel(0, &ioc->chip->HostInterruptStatus);
7098 	if ((_base_wait_for_doorbell_int(ioc, 5))) {
7099 		ioc_err(ioc, "doorbell handshake int failed (line=%d)\n",
7100 			__LINE__);
7101 		return -EFAULT;
7102 	}
7103 	reply[1] = le16_to_cpu(ioc->base_readl_ext_retry(&ioc->chip->Doorbell)
7104 	    & MPI2_DOORBELL_DATA_MASK);
7105 	writel(0, &ioc->chip->HostInterruptStatus);
7106 
7107 	for (i = 2; i < default_reply->MsgLength * 2; i++)  {
7108 		if ((_base_wait_for_doorbell_int(ioc, 5))) {
7109 			ioc_err(ioc, "doorbell handshake int failed (line=%d)\n",
7110 				__LINE__);
7111 			return -EFAULT;
7112 		}
7113 		if (i >=  reply_bytes/2) /* overflow case */
7114 			ioc->base_readl_ext_retry(&ioc->chip->Doorbell);
7115 		else
7116 			reply[i] = le16_to_cpu(
7117 			    ioc->base_readl_ext_retry(&ioc->chip->Doorbell)
7118 			    & MPI2_DOORBELL_DATA_MASK);
7119 		writel(0, &ioc->chip->HostInterruptStatus);
7120 	}
7121 
7122 	_base_wait_for_doorbell_int(ioc, 5);
7123 	if (_base_wait_for_doorbell_not_used(ioc, 5) != 0) {
7124 		dhsprintk(ioc,
7125 			  ioc_info(ioc, "doorbell is in use (line=%d)\n",
7126 				   __LINE__));
7127 	}
7128 	writel(0, &ioc->chip->HostInterruptStatus);
7129 
7130 	if (ioc->logging_level & MPT_DEBUG_INIT) {
7131 		mfp = (__le32 *)reply;
7132 		pr_info("\toffset:data\n");
7133 		for (i = 0; i < reply_bytes/4; i++)
7134 			ioc_info(ioc, "\t[0x%02x]:%08x\n", i*4,
7135 			    le32_to_cpu(mfp[i]));
7136 	}
7137 	return 0;
7138 }
7139 
7140 /**
7141  * mpt3sas_base_sas_iounit_control - send sas iounit control to FW
7142  * @ioc: per adapter object
7143  * @mpi_reply: the reply payload from FW
7144  * @mpi_request: the request payload sent to FW
7145  *
7146  * The SAS IO Unit Control Request message allows the host to perform low-level
7147  * operations, such as resets on the PHYs of the IO Unit, also allows the host
7148  * to obtain the IOC assigned device handles for a device if it has other
7149  * identifying information about the device, in addition allows the host to
7150  * remove IOC resources associated with the device.
7151  *
7152  * Return: 0 for success, non-zero for failure.
7153  */
7154 int
7155 mpt3sas_base_sas_iounit_control(struct MPT3SAS_ADAPTER *ioc,
7156 	Mpi2SasIoUnitControlReply_t *mpi_reply,
7157 	Mpi2SasIoUnitControlRequest_t *mpi_request)
7158 {
7159 	u16 smid;
7160 	u8 issue_reset = 0;
7161 	int rc;
7162 	void *request;
7163 
7164 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
7165 
7166 	mutex_lock(&ioc->base_cmds.mutex);
7167 
7168 	if (ioc->base_cmds.status != MPT3_CMD_NOT_USED) {
7169 		ioc_err(ioc, "%s: base_cmd in use\n", __func__);
7170 		rc = -EAGAIN;
7171 		goto out;
7172 	}
7173 
7174 	rc = mpt3sas_wait_for_ioc(ioc, IOC_OPERATIONAL_WAIT_COUNT);
7175 	if (rc)
7176 		goto out;
7177 
7178 	smid = mpt3sas_base_get_smid(ioc, ioc->base_cb_idx);
7179 	if (!smid) {
7180 		ioc_err(ioc, "%s: failed obtaining a smid\n", __func__);
7181 		rc = -EAGAIN;
7182 		goto out;
7183 	}
7184 
7185 	rc = 0;
7186 	ioc->base_cmds.status = MPT3_CMD_PENDING;
7187 	request = mpt3sas_base_get_msg_frame(ioc, smid);
7188 	ioc->base_cmds.smid = smid;
7189 	memcpy(request, mpi_request, sizeof(Mpi2SasIoUnitControlRequest_t));
7190 	if (mpi_request->Operation == MPI2_SAS_OP_PHY_HARD_RESET ||
7191 	    mpi_request->Operation == MPI2_SAS_OP_PHY_LINK_RESET)
7192 		ioc->ioc_link_reset_in_progress = 1;
7193 	init_completion(&ioc->base_cmds.done);
7194 	ioc->put_smid_default(ioc, smid);
7195 	wait_for_completion_timeout(&ioc->base_cmds.done,
7196 	    msecs_to_jiffies(10000));
7197 	if ((mpi_request->Operation == MPI2_SAS_OP_PHY_HARD_RESET ||
7198 	    mpi_request->Operation == MPI2_SAS_OP_PHY_LINK_RESET) &&
7199 	    ioc->ioc_link_reset_in_progress)
7200 		ioc->ioc_link_reset_in_progress = 0;
7201 	if (!(ioc->base_cmds.status & MPT3_CMD_COMPLETE)) {
7202 		mpt3sas_check_cmd_timeout(ioc, ioc->base_cmds.status,
7203 		    mpi_request, sizeof(Mpi2SasIoUnitControlRequest_t)/4,
7204 		    issue_reset);
7205 		goto issue_host_reset;
7206 	}
7207 	if (ioc->base_cmds.status & MPT3_CMD_REPLY_VALID)
7208 		memcpy(mpi_reply, ioc->base_cmds.reply,
7209 		    sizeof(Mpi2SasIoUnitControlReply_t));
7210 	else
7211 		memset(mpi_reply, 0, sizeof(Mpi2SasIoUnitControlReply_t));
7212 	ioc->base_cmds.status = MPT3_CMD_NOT_USED;
7213 	goto out;
7214 
7215  issue_host_reset:
7216 	if (issue_reset)
7217 		mpt3sas_base_hard_reset_handler(ioc, FORCE_BIG_HAMMER);
7218 	ioc->base_cmds.status = MPT3_CMD_NOT_USED;
7219 	rc = -EFAULT;
7220  out:
7221 	mutex_unlock(&ioc->base_cmds.mutex);
7222 	return rc;
7223 }
7224 
7225 /**
7226  * mpt3sas_base_scsi_enclosure_processor - sending request to sep device
7227  * @ioc: per adapter object
7228  * @mpi_reply: the reply payload from FW
7229  * @mpi_request: the request payload sent to FW
7230  *
7231  * The SCSI Enclosure Processor request message causes the IOC to
7232  * communicate with SES devices to control LED status signals.
7233  *
7234  * Return: 0 for success, non-zero for failure.
7235  */
7236 int
7237 mpt3sas_base_scsi_enclosure_processor(struct MPT3SAS_ADAPTER *ioc,
7238 	Mpi2SepReply_t *mpi_reply, Mpi2SepRequest_t *mpi_request)
7239 {
7240 	u16 smid;
7241 	u8 issue_reset = 0;
7242 	int rc;
7243 	void *request;
7244 
7245 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
7246 
7247 	mutex_lock(&ioc->base_cmds.mutex);
7248 
7249 	if (ioc->base_cmds.status != MPT3_CMD_NOT_USED) {
7250 		ioc_err(ioc, "%s: base_cmd in use\n", __func__);
7251 		rc = -EAGAIN;
7252 		goto out;
7253 	}
7254 
7255 	rc = mpt3sas_wait_for_ioc(ioc, IOC_OPERATIONAL_WAIT_COUNT);
7256 	if (rc)
7257 		goto out;
7258 
7259 	smid = mpt3sas_base_get_smid(ioc, ioc->base_cb_idx);
7260 	if (!smid) {
7261 		ioc_err(ioc, "%s: failed obtaining a smid\n", __func__);
7262 		rc = -EAGAIN;
7263 		goto out;
7264 	}
7265 
7266 	rc = 0;
7267 	ioc->base_cmds.status = MPT3_CMD_PENDING;
7268 	request = mpt3sas_base_get_msg_frame(ioc, smid);
7269 	ioc->base_cmds.smid = smid;
7270 	memset(request, 0, ioc->request_sz);
7271 	memcpy(request, mpi_request, sizeof(Mpi2SepReply_t));
7272 	init_completion(&ioc->base_cmds.done);
7273 	ioc->put_smid_default(ioc, smid);
7274 	wait_for_completion_timeout(&ioc->base_cmds.done,
7275 	    msecs_to_jiffies(10000));
7276 	if (!(ioc->base_cmds.status & MPT3_CMD_COMPLETE)) {
7277 		mpt3sas_check_cmd_timeout(ioc,
7278 		    ioc->base_cmds.status, mpi_request,
7279 		    sizeof(Mpi2SepRequest_t)/4, issue_reset);
7280 		goto issue_host_reset;
7281 	}
7282 	if (ioc->base_cmds.status & MPT3_CMD_REPLY_VALID)
7283 		memcpy(mpi_reply, ioc->base_cmds.reply,
7284 		    sizeof(Mpi2SepReply_t));
7285 	else
7286 		memset(mpi_reply, 0, sizeof(Mpi2SepReply_t));
7287 	ioc->base_cmds.status = MPT3_CMD_NOT_USED;
7288 	goto out;
7289 
7290  issue_host_reset:
7291 	if (issue_reset)
7292 		mpt3sas_base_hard_reset_handler(ioc, FORCE_BIG_HAMMER);
7293 	ioc->base_cmds.status = MPT3_CMD_NOT_USED;
7294 	rc = -EFAULT;
7295  out:
7296 	mutex_unlock(&ioc->base_cmds.mutex);
7297 	return rc;
7298 }
7299 
7300 /**
7301  * _base_get_port_facts - obtain port facts reply and save in ioc
7302  * @ioc: per adapter object
7303  * @port: ?
7304  *
7305  * Return: 0 for success, non-zero for failure.
7306  */
7307 static int
7308 _base_get_port_facts(struct MPT3SAS_ADAPTER *ioc, int port)
7309 {
7310 	Mpi2PortFactsRequest_t mpi_request;
7311 	Mpi2PortFactsReply_t mpi_reply;
7312 	struct mpt3sas_port_facts *pfacts;
7313 	int mpi_reply_sz, mpi_request_sz, r;
7314 
7315 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
7316 
7317 	mpi_reply_sz = sizeof(Mpi2PortFactsReply_t);
7318 	mpi_request_sz = sizeof(Mpi2PortFactsRequest_t);
7319 	memset(&mpi_request, 0, mpi_request_sz);
7320 	mpi_request.Function = MPI2_FUNCTION_PORT_FACTS;
7321 	mpi_request.PortNumber = port;
7322 	r = _base_handshake_req_reply_wait(ioc, mpi_request_sz,
7323 	    (u32 *)&mpi_request, mpi_reply_sz, (u16 *)&mpi_reply, 5);
7324 
7325 	if (r != 0) {
7326 		ioc_err(ioc, "%s: handshake failed (r=%d)\n", __func__, r);
7327 		return r;
7328 	}
7329 
7330 	pfacts = &ioc->pfacts[port];
7331 	memset(pfacts, 0, sizeof(struct mpt3sas_port_facts));
7332 	pfacts->PortNumber = mpi_reply.PortNumber;
7333 	pfacts->VP_ID = mpi_reply.VP_ID;
7334 	pfacts->VF_ID = mpi_reply.VF_ID;
7335 	pfacts->MaxPostedCmdBuffers =
7336 	    le16_to_cpu(mpi_reply.MaxPostedCmdBuffers);
7337 
7338 	return 0;
7339 }
7340 
7341 /**
7342  * _base_wait_for_iocstate - Wait until the card is in READY or OPERATIONAL
7343  * @ioc: per adapter object
7344  * @timeout:
7345  *
7346  * Return: 0 for success, non-zero for failure.
7347  */
7348 static int
7349 _base_wait_for_iocstate(struct MPT3SAS_ADAPTER *ioc, int timeout)
7350 {
7351 	u32 ioc_state;
7352 	int rc;
7353 
7354 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
7355 
7356 	if (ioc->pci_error_recovery) {
7357 		dfailprintk(ioc,
7358 			    ioc_info(ioc, "%s: host in pci error recovery\n",
7359 				     __func__));
7360 		return -EFAULT;
7361 	}
7362 
7363 	ioc_state = mpt3sas_base_get_iocstate(ioc, 0);
7364 	dhsprintk(ioc,
7365 		  ioc_info(ioc, "%s: ioc_state(0x%08x)\n",
7366 			   __func__, ioc_state));
7367 
7368 	if (((ioc_state & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_READY) ||
7369 	    (ioc_state & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_OPERATIONAL)
7370 		return 0;
7371 
7372 	if (ioc_state & MPI2_DOORBELL_USED) {
7373 		dhsprintk(ioc, ioc_info(ioc, "unexpected doorbell active!\n"));
7374 		goto issue_diag_reset;
7375 	}
7376 
7377 	if ((ioc_state & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) {
7378 		mpt3sas_print_fault_code(ioc, ioc_state &
7379 		    MPI2_DOORBELL_DATA_MASK);
7380 		goto issue_diag_reset;
7381 	} else if ((ioc_state & MPI2_IOC_STATE_MASK) ==
7382 	    MPI2_IOC_STATE_COREDUMP) {
7383 		ioc_info(ioc,
7384 		    "%s: Skipping the diag reset here. (ioc_state=0x%x)\n",
7385 		    __func__, ioc_state);
7386 		return -EFAULT;
7387 	}
7388 
7389 	ioc_state = _base_wait_on_iocstate(ioc, MPI2_IOC_STATE_READY, timeout);
7390 	if (ioc_state) {
7391 		dfailprintk(ioc,
7392 			    ioc_info(ioc, "%s: failed going to ready state (ioc_state=0x%x)\n",
7393 				     __func__, ioc_state));
7394 		return -EFAULT;
7395 	}
7396 
7397 	return 0;
7398 
7399 issue_diag_reset:
7400 	rc = _base_diag_reset(ioc);
7401 	return rc;
7402 }
7403 
7404 /**
7405  * _base_get_ioc_facts - obtain ioc facts reply and save in ioc
7406  * @ioc: per adapter object
7407  *
7408  * Return: 0 for success, non-zero for failure.
7409  */
7410 static int
7411 _base_get_ioc_facts(struct MPT3SAS_ADAPTER *ioc)
7412 {
7413 	Mpi2IOCFactsRequest_t mpi_request;
7414 	Mpi2IOCFactsReply_t mpi_reply;
7415 	struct mpt3sas_facts *facts;
7416 	int mpi_reply_sz, mpi_request_sz, r;
7417 
7418 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
7419 
7420 	r = _base_wait_for_iocstate(ioc, 10);
7421 	if (r) {
7422 		dfailprintk(ioc,
7423 			    ioc_info(ioc, "%s: failed getting to correct state\n",
7424 				     __func__));
7425 		return r;
7426 	}
7427 	mpi_reply_sz = sizeof(Mpi2IOCFactsReply_t);
7428 	mpi_request_sz = sizeof(Mpi2IOCFactsRequest_t);
7429 	memset(&mpi_request, 0, mpi_request_sz);
7430 	mpi_request.Function = MPI2_FUNCTION_IOC_FACTS;
7431 	r = _base_handshake_req_reply_wait(ioc, mpi_request_sz,
7432 	    (u32 *)&mpi_request, mpi_reply_sz, (u16 *)&mpi_reply, 5);
7433 
7434 	if (r != 0) {
7435 		ioc_err(ioc, "%s: handshake failed (r=%d)\n", __func__, r);
7436 		return r;
7437 	}
7438 
7439 	facts = &ioc->facts;
7440 	memset(facts, 0, sizeof(struct mpt3sas_facts));
7441 	facts->MsgVersion = le16_to_cpu(mpi_reply.MsgVersion);
7442 	facts->HeaderVersion = le16_to_cpu(mpi_reply.HeaderVersion);
7443 	facts->VP_ID = mpi_reply.VP_ID;
7444 	facts->VF_ID = mpi_reply.VF_ID;
7445 	facts->IOCExceptions = le16_to_cpu(mpi_reply.IOCExceptions);
7446 	facts->MaxChainDepth = mpi_reply.MaxChainDepth;
7447 	facts->WhoInit = mpi_reply.WhoInit;
7448 	facts->NumberOfPorts = mpi_reply.NumberOfPorts;
7449 	facts->MaxMSIxVectors = mpi_reply.MaxMSIxVectors;
7450 	if (ioc->msix_enable && (facts->MaxMSIxVectors <=
7451 	    MAX_COMBINED_MSIX_VECTORS(ioc->is_gen35_ioc)))
7452 		ioc->combined_reply_queue = 0;
7453 	facts->RequestCredit = le16_to_cpu(mpi_reply.RequestCredit);
7454 	facts->MaxReplyDescriptorPostQueueDepth =
7455 	    le16_to_cpu(mpi_reply.MaxReplyDescriptorPostQueueDepth);
7456 	facts->ProductID = le16_to_cpu(mpi_reply.ProductID);
7457 	facts->IOCCapabilities = le32_to_cpu(mpi_reply.IOCCapabilities);
7458 	if ((facts->IOCCapabilities & MPI2_IOCFACTS_CAPABILITY_INTEGRATED_RAID))
7459 		ioc->ir_firmware = 1;
7460 	if ((facts->IOCCapabilities &
7461 	      MPI2_IOCFACTS_CAPABILITY_RDPQ_ARRAY_CAPABLE) && (!reset_devices))
7462 		ioc->rdpq_array_capable = 1;
7463 	if ((facts->IOCCapabilities & MPI26_IOCFACTS_CAPABILITY_ATOMIC_REQ)
7464 	    && ioc->is_aero_ioc)
7465 		ioc->atomic_desc_capable = 1;
7466 	facts->FWVersion.Word = le32_to_cpu(mpi_reply.FWVersion.Word);
7467 	facts->IOCRequestFrameSize =
7468 	    le16_to_cpu(mpi_reply.IOCRequestFrameSize);
7469 	if (ioc->hba_mpi_version_belonged != MPI2_VERSION) {
7470 		facts->IOCMaxChainSegmentSize =
7471 			le16_to_cpu(mpi_reply.IOCMaxChainSegmentSize);
7472 	}
7473 	facts->MaxInitiators = le16_to_cpu(mpi_reply.MaxInitiators);
7474 	facts->MaxTargets = le16_to_cpu(mpi_reply.MaxTargets);
7475 	ioc->shost->max_id = -1;
7476 	facts->MaxSasExpanders = le16_to_cpu(mpi_reply.MaxSasExpanders);
7477 	facts->MaxEnclosures = le16_to_cpu(mpi_reply.MaxEnclosures);
7478 	facts->ProtocolFlags = le16_to_cpu(mpi_reply.ProtocolFlags);
7479 	facts->HighPriorityCredit =
7480 	    le16_to_cpu(mpi_reply.HighPriorityCredit);
7481 	facts->ReplyFrameSize = mpi_reply.ReplyFrameSize;
7482 	facts->MaxDevHandle = le16_to_cpu(mpi_reply.MaxDevHandle);
7483 	facts->CurrentHostPageSize = mpi_reply.CurrentHostPageSize;
7484 
7485 	/*
7486 	 * Get the Page Size from IOC Facts. If it's 0, default to 4k.
7487 	 */
7488 	ioc->page_size = 1 << facts->CurrentHostPageSize;
7489 	if (ioc->page_size == 1) {
7490 		ioc_info(ioc, "CurrentHostPageSize is 0: Setting default host page size to 4k\n");
7491 		ioc->page_size = 1 << MPT3SAS_HOST_PAGE_SIZE_4K;
7492 	}
7493 	dinitprintk(ioc,
7494 		    ioc_info(ioc, "CurrentHostPageSize(%d)\n",
7495 			     facts->CurrentHostPageSize));
7496 
7497 	dinitprintk(ioc,
7498 		    ioc_info(ioc, "hba queue depth(%d), max chains per io(%d)\n",
7499 			     facts->RequestCredit, facts->MaxChainDepth));
7500 	dinitprintk(ioc,
7501 		    ioc_info(ioc, "request frame size(%d), reply frame size(%d)\n",
7502 			     facts->IOCRequestFrameSize * 4,
7503 			     facts->ReplyFrameSize * 4));
7504 	return 0;
7505 }
7506 
7507 /**
7508  * _base_send_ioc_init - send ioc_init to firmware
7509  * @ioc: per adapter object
7510  *
7511  * Return: 0 for success, non-zero for failure.
7512  */
7513 static int
7514 _base_send_ioc_init(struct MPT3SAS_ADAPTER *ioc)
7515 {
7516 	Mpi2IOCInitRequest_t mpi_request;
7517 	Mpi2IOCInitReply_t mpi_reply;
7518 	int i, r = 0;
7519 	ktime_t current_time;
7520 	u16 ioc_status;
7521 	u32 reply_post_free_array_sz = 0;
7522 
7523 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
7524 
7525 	memset(&mpi_request, 0, sizeof(Mpi2IOCInitRequest_t));
7526 	mpi_request.Function = MPI2_FUNCTION_IOC_INIT;
7527 	mpi_request.WhoInit = MPI2_WHOINIT_HOST_DRIVER;
7528 	mpi_request.VF_ID = 0; /* TODO */
7529 	mpi_request.VP_ID = 0;
7530 	mpi_request.MsgVersion = cpu_to_le16(ioc->hba_mpi_version_belonged);
7531 	mpi_request.HeaderVersion = cpu_to_le16(MPI2_HEADER_VERSION);
7532 	mpi_request.HostPageSize = MPT3SAS_HOST_PAGE_SIZE_4K;
7533 
7534 	if (_base_is_controller_msix_enabled(ioc))
7535 		mpi_request.HostMSIxVectors = ioc->reply_queue_count;
7536 	mpi_request.SystemRequestFrameSize = cpu_to_le16(ioc->request_sz/4);
7537 	mpi_request.ReplyDescriptorPostQueueDepth =
7538 	    cpu_to_le16(ioc->reply_post_queue_depth);
7539 	mpi_request.ReplyFreeQueueDepth =
7540 	    cpu_to_le16(ioc->reply_free_queue_depth);
7541 
7542 	mpi_request.SenseBufferAddressHigh =
7543 	    cpu_to_le32((u64)ioc->sense_dma >> 32);
7544 	mpi_request.SystemReplyAddressHigh =
7545 	    cpu_to_le32((u64)ioc->reply_dma >> 32);
7546 	mpi_request.SystemRequestFrameBaseAddress =
7547 	    cpu_to_le64((u64)ioc->request_dma);
7548 	mpi_request.ReplyFreeQueueAddress =
7549 	    cpu_to_le64((u64)ioc->reply_free_dma);
7550 
7551 	if (ioc->rdpq_array_enable) {
7552 		reply_post_free_array_sz = ioc->reply_queue_count *
7553 		    sizeof(Mpi2IOCInitRDPQArrayEntry);
7554 		memset(ioc->reply_post_free_array, 0, reply_post_free_array_sz);
7555 		for (i = 0; i < ioc->reply_queue_count; i++)
7556 			ioc->reply_post_free_array[i].RDPQBaseAddress =
7557 			    cpu_to_le64(
7558 				(u64)ioc->reply_post[i].reply_post_free_dma);
7559 		mpi_request.MsgFlags = MPI2_IOCINIT_MSGFLAG_RDPQ_ARRAY_MODE;
7560 		mpi_request.ReplyDescriptorPostQueueAddress =
7561 		    cpu_to_le64((u64)ioc->reply_post_free_array_dma);
7562 	} else {
7563 		mpi_request.ReplyDescriptorPostQueueAddress =
7564 		    cpu_to_le64((u64)ioc->reply_post[0].reply_post_free_dma);
7565 	}
7566 
7567 	/*
7568 	 * Set the flag to enable CoreDump state feature in IOC firmware.
7569 	 */
7570 	mpi_request.ConfigurationFlags |=
7571 	    cpu_to_le16(MPI26_IOCINIT_CFGFLAGS_COREDUMP_ENABLE);
7572 
7573 	/* This time stamp specifies number of milliseconds
7574 	 * since epoch ~ midnight January 1, 1970.
7575 	 */
7576 	current_time = ktime_get_real();
7577 	mpi_request.TimeStamp = cpu_to_le64(ktime_to_ms(current_time));
7578 
7579 	if (ioc->logging_level & MPT_DEBUG_INIT) {
7580 		__le32 *mfp;
7581 		int i;
7582 
7583 		mfp = (__le32 *)&mpi_request;
7584 		ioc_info(ioc, "\toffset:data\n");
7585 		for (i = 0; i < sizeof(Mpi2IOCInitRequest_t)/4; i++)
7586 			ioc_info(ioc, "\t[0x%02x]:%08x\n", i*4,
7587 			    le32_to_cpu(mfp[i]));
7588 	}
7589 
7590 	r = _base_handshake_req_reply_wait(ioc,
7591 	    sizeof(Mpi2IOCInitRequest_t), (u32 *)&mpi_request,
7592 	    sizeof(Mpi2IOCInitReply_t), (u16 *)&mpi_reply, 30);
7593 
7594 	if (r != 0) {
7595 		ioc_err(ioc, "%s: handshake failed (r=%d)\n", __func__, r);
7596 		return r;
7597 	}
7598 
7599 	ioc_status = le16_to_cpu(mpi_reply.IOCStatus) & MPI2_IOCSTATUS_MASK;
7600 	if (ioc_status != MPI2_IOCSTATUS_SUCCESS ||
7601 	    mpi_reply.IOCLogInfo) {
7602 		ioc_err(ioc, "%s: failed\n", __func__);
7603 		r = -EIO;
7604 	}
7605 
7606 	/* Reset TimeSync Counter*/
7607 	ioc->timestamp_update_count = 0;
7608 	return r;
7609 }
7610 
7611 /**
7612  * mpt3sas_port_enable_done - command completion routine for port enable
7613  * @ioc: per adapter object
7614  * @smid: system request message index
7615  * @msix_index: MSIX table index supplied by the OS
7616  * @reply: reply message frame(lower 32bit addr)
7617  *
7618  * Return: 1 meaning mf should be freed from _base_interrupt
7619  *          0 means the mf is freed from this function.
7620  */
7621 u8
7622 mpt3sas_port_enable_done(struct MPT3SAS_ADAPTER *ioc, u16 smid, u8 msix_index,
7623 	u32 reply)
7624 {
7625 	MPI2DefaultReply_t *mpi_reply;
7626 	u16 ioc_status;
7627 
7628 	if (ioc->port_enable_cmds.status == MPT3_CMD_NOT_USED)
7629 		return 1;
7630 
7631 	mpi_reply = mpt3sas_base_get_reply_virt_addr(ioc, reply);
7632 	if (!mpi_reply)
7633 		return 1;
7634 
7635 	if (mpi_reply->Function != MPI2_FUNCTION_PORT_ENABLE)
7636 		return 1;
7637 
7638 	ioc->port_enable_cmds.status &= ~MPT3_CMD_PENDING;
7639 	ioc->port_enable_cmds.status |= MPT3_CMD_COMPLETE;
7640 	ioc->port_enable_cmds.status |= MPT3_CMD_REPLY_VALID;
7641 	memcpy(ioc->port_enable_cmds.reply, mpi_reply, mpi_reply->MsgLength*4);
7642 	ioc_status = le16_to_cpu(mpi_reply->IOCStatus) & MPI2_IOCSTATUS_MASK;
7643 	if (ioc_status != MPI2_IOCSTATUS_SUCCESS)
7644 		ioc->port_enable_failed = 1;
7645 
7646 	if (ioc->port_enable_cmds.status & MPT3_CMD_COMPLETE_ASYNC) {
7647 		ioc->port_enable_cmds.status &= ~MPT3_CMD_COMPLETE_ASYNC;
7648 		if (ioc_status == MPI2_IOCSTATUS_SUCCESS) {
7649 			mpt3sas_port_enable_complete(ioc);
7650 			return 1;
7651 		} else {
7652 			ioc->start_scan_failed = ioc_status;
7653 			ioc->start_scan = 0;
7654 			return 1;
7655 		}
7656 	}
7657 	complete(&ioc->port_enable_cmds.done);
7658 	return 1;
7659 }
7660 
7661 /**
7662  * _base_send_port_enable - send port_enable(discovery stuff) to firmware
7663  * @ioc: per adapter object
7664  *
7665  * Return: 0 for success, non-zero for failure.
7666  */
7667 static int
7668 _base_send_port_enable(struct MPT3SAS_ADAPTER *ioc)
7669 {
7670 	Mpi2PortEnableRequest_t *mpi_request;
7671 	Mpi2PortEnableReply_t *mpi_reply;
7672 	int r = 0;
7673 	u16 smid;
7674 	u16 ioc_status;
7675 
7676 	ioc_info(ioc, "sending port enable !!\n");
7677 
7678 	if (ioc->port_enable_cmds.status & MPT3_CMD_PENDING) {
7679 		ioc_err(ioc, "%s: internal command already in use\n", __func__);
7680 		return -EAGAIN;
7681 	}
7682 
7683 	smid = mpt3sas_base_get_smid(ioc, ioc->port_enable_cb_idx);
7684 	if (!smid) {
7685 		ioc_err(ioc, "%s: failed obtaining a smid\n", __func__);
7686 		return -EAGAIN;
7687 	}
7688 
7689 	ioc->port_enable_cmds.status = MPT3_CMD_PENDING;
7690 	mpi_request = mpt3sas_base_get_msg_frame(ioc, smid);
7691 	ioc->port_enable_cmds.smid = smid;
7692 	memset(mpi_request, 0, sizeof(Mpi2PortEnableRequest_t));
7693 	mpi_request->Function = MPI2_FUNCTION_PORT_ENABLE;
7694 
7695 	init_completion(&ioc->port_enable_cmds.done);
7696 	ioc->put_smid_default(ioc, smid);
7697 	wait_for_completion_timeout(&ioc->port_enable_cmds.done, 300*HZ);
7698 	if (!(ioc->port_enable_cmds.status & MPT3_CMD_COMPLETE)) {
7699 		ioc_err(ioc, "%s: timeout\n", __func__);
7700 		_debug_dump_mf(mpi_request,
7701 		    sizeof(Mpi2PortEnableRequest_t)/4);
7702 		if (ioc->port_enable_cmds.status & MPT3_CMD_RESET)
7703 			r = -EFAULT;
7704 		else
7705 			r = -ETIME;
7706 		goto out;
7707 	}
7708 
7709 	mpi_reply = ioc->port_enable_cmds.reply;
7710 	ioc_status = le16_to_cpu(mpi_reply->IOCStatus) & MPI2_IOCSTATUS_MASK;
7711 	if (ioc_status != MPI2_IOCSTATUS_SUCCESS) {
7712 		ioc_err(ioc, "%s: failed with (ioc_status=0x%08x)\n",
7713 			__func__, ioc_status);
7714 		r = -EFAULT;
7715 		goto out;
7716 	}
7717 
7718  out:
7719 	ioc->port_enable_cmds.status = MPT3_CMD_NOT_USED;
7720 	ioc_info(ioc, "port enable: %s\n", r == 0 ? "SUCCESS" : "FAILED");
7721 	return r;
7722 }
7723 
7724 /**
7725  * mpt3sas_port_enable - initiate firmware discovery (don't wait for reply)
7726  * @ioc: per adapter object
7727  *
7728  * Return: 0 for success, non-zero for failure.
7729  */
7730 int
7731 mpt3sas_port_enable(struct MPT3SAS_ADAPTER *ioc)
7732 {
7733 	Mpi2PortEnableRequest_t *mpi_request;
7734 	u16 smid;
7735 
7736 	ioc_info(ioc, "sending port enable !!\n");
7737 
7738 	if (ioc->port_enable_cmds.status & MPT3_CMD_PENDING) {
7739 		ioc_err(ioc, "%s: internal command already in use\n", __func__);
7740 		return -EAGAIN;
7741 	}
7742 
7743 	smid = mpt3sas_base_get_smid(ioc, ioc->port_enable_cb_idx);
7744 	if (!smid) {
7745 		ioc_err(ioc, "%s: failed obtaining a smid\n", __func__);
7746 		return -EAGAIN;
7747 	}
7748 	ioc->drv_internal_flags |= MPT_DRV_INTERNAL_FIRST_PE_ISSUED;
7749 	ioc->port_enable_cmds.status = MPT3_CMD_PENDING;
7750 	ioc->port_enable_cmds.status |= MPT3_CMD_COMPLETE_ASYNC;
7751 	mpi_request = mpt3sas_base_get_msg_frame(ioc, smid);
7752 	ioc->port_enable_cmds.smid = smid;
7753 	memset(mpi_request, 0, sizeof(Mpi2PortEnableRequest_t));
7754 	mpi_request->Function = MPI2_FUNCTION_PORT_ENABLE;
7755 
7756 	ioc->put_smid_default(ioc, smid);
7757 	return 0;
7758 }
7759 
7760 /**
7761  * _base_determine_wait_on_discovery - desposition
7762  * @ioc: per adapter object
7763  *
7764  * Decide whether to wait on discovery to complete. Used to either
7765  * locate boot device, or report volumes ahead of physical devices.
7766  *
7767  * Return: 1 for wait, 0 for don't wait.
7768  */
7769 static int
7770 _base_determine_wait_on_discovery(struct MPT3SAS_ADAPTER *ioc)
7771 {
7772 	/* We wait for discovery to complete if IR firmware is loaded.
7773 	 * The sas topology events arrive before PD events, so we need time to
7774 	 * turn on the bit in ioc->pd_handles to indicate PD
7775 	 * Also, it maybe required to report Volumes ahead of physical
7776 	 * devices when MPI2_IOCPAGE8_IRFLAGS_LOW_VOLUME_MAPPING is set.
7777 	 */
7778 	if (ioc->ir_firmware)
7779 		return 1;
7780 
7781 	/* if no Bios, then we don't need to wait */
7782 	if (!ioc->bios_pg3.BiosVersion)
7783 		return 0;
7784 
7785 	/* Bios is present, then we drop down here.
7786 	 *
7787 	 * If there any entries in the Bios Page 2, then we wait
7788 	 * for discovery to complete.
7789 	 */
7790 
7791 	/* Current Boot Device */
7792 	if ((ioc->bios_pg2.CurrentBootDeviceForm &
7793 	    MPI2_BIOSPAGE2_FORM_MASK) ==
7794 	    MPI2_BIOSPAGE2_FORM_NO_DEVICE_SPECIFIED &&
7795 	/* Request Boot Device */
7796 	   (ioc->bios_pg2.ReqBootDeviceForm &
7797 	    MPI2_BIOSPAGE2_FORM_MASK) ==
7798 	    MPI2_BIOSPAGE2_FORM_NO_DEVICE_SPECIFIED &&
7799 	/* Alternate Request Boot Device */
7800 	   (ioc->bios_pg2.ReqAltBootDeviceForm &
7801 	    MPI2_BIOSPAGE2_FORM_MASK) ==
7802 	    MPI2_BIOSPAGE2_FORM_NO_DEVICE_SPECIFIED)
7803 		return 0;
7804 
7805 	return 1;
7806 }
7807 
7808 /**
7809  * _base_unmask_events - turn on notification for this event
7810  * @ioc: per adapter object
7811  * @event: firmware event
7812  *
7813  * The mask is stored in ioc->event_masks.
7814  */
7815 static void
7816 _base_unmask_events(struct MPT3SAS_ADAPTER *ioc, u16 event)
7817 {
7818 	u32 desired_event;
7819 
7820 	if (event >= 128)
7821 		return;
7822 
7823 	desired_event = (1 << (event % 32));
7824 
7825 	if (event < 32)
7826 		ioc->event_masks[0] &= ~desired_event;
7827 	else if (event < 64)
7828 		ioc->event_masks[1] &= ~desired_event;
7829 	else if (event < 96)
7830 		ioc->event_masks[2] &= ~desired_event;
7831 	else if (event < 128)
7832 		ioc->event_masks[3] &= ~desired_event;
7833 }
7834 
7835 /**
7836  * _base_event_notification - send event notification
7837  * @ioc: per adapter object
7838  *
7839  * Return: 0 for success, non-zero for failure.
7840  */
7841 static int
7842 _base_event_notification(struct MPT3SAS_ADAPTER *ioc)
7843 {
7844 	Mpi2EventNotificationRequest_t *mpi_request;
7845 	u16 smid;
7846 	int r = 0;
7847 	int i, issue_diag_reset = 0;
7848 
7849 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
7850 
7851 	if (ioc->base_cmds.status & MPT3_CMD_PENDING) {
7852 		ioc_err(ioc, "%s: internal command already in use\n", __func__);
7853 		return -EAGAIN;
7854 	}
7855 
7856 	smid = mpt3sas_base_get_smid(ioc, ioc->base_cb_idx);
7857 	if (!smid) {
7858 		ioc_err(ioc, "%s: failed obtaining a smid\n", __func__);
7859 		return -EAGAIN;
7860 	}
7861 	ioc->base_cmds.status = MPT3_CMD_PENDING;
7862 	mpi_request = mpt3sas_base_get_msg_frame(ioc, smid);
7863 	ioc->base_cmds.smid = smid;
7864 	memset(mpi_request, 0, sizeof(Mpi2EventNotificationRequest_t));
7865 	mpi_request->Function = MPI2_FUNCTION_EVENT_NOTIFICATION;
7866 	mpi_request->VF_ID = 0; /* TODO */
7867 	mpi_request->VP_ID = 0;
7868 	for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++)
7869 		mpi_request->EventMasks[i] =
7870 		    cpu_to_le32(ioc->event_masks[i]);
7871 	init_completion(&ioc->base_cmds.done);
7872 	ioc->put_smid_default(ioc, smid);
7873 	wait_for_completion_timeout(&ioc->base_cmds.done, 30*HZ);
7874 	if (!(ioc->base_cmds.status & MPT3_CMD_COMPLETE)) {
7875 		ioc_err(ioc, "%s: timeout\n", __func__);
7876 		_debug_dump_mf(mpi_request,
7877 		    sizeof(Mpi2EventNotificationRequest_t)/4);
7878 		if (ioc->base_cmds.status & MPT3_CMD_RESET)
7879 			r = -EFAULT;
7880 		else
7881 			issue_diag_reset = 1;
7882 
7883 	} else
7884 		dinitprintk(ioc, ioc_info(ioc, "%s: complete\n", __func__));
7885 	ioc->base_cmds.status = MPT3_CMD_NOT_USED;
7886 
7887 	if (issue_diag_reset) {
7888 		if (ioc->drv_internal_flags & MPT_DRV_INTERNAL_FIRST_PE_ISSUED)
7889 			return -EFAULT;
7890 		if (mpt3sas_base_check_for_fault_and_issue_reset(ioc))
7891 			return -EFAULT;
7892 		r = -EAGAIN;
7893 	}
7894 	return r;
7895 }
7896 
7897 /**
7898  * mpt3sas_base_validate_event_type - validating event types
7899  * @ioc: per adapter object
7900  * @event_type: firmware event
7901  *
7902  * This will turn on firmware event notification when application
7903  * ask for that event. We don't mask events that are already enabled.
7904  */
7905 void
7906 mpt3sas_base_validate_event_type(struct MPT3SAS_ADAPTER *ioc, u32 *event_type)
7907 {
7908 	int i, j;
7909 	u32 event_mask, desired_event;
7910 	u8 send_update_to_fw;
7911 
7912 	for (i = 0, send_update_to_fw = 0; i <
7913 	    MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++) {
7914 		event_mask = ~event_type[i];
7915 		desired_event = 1;
7916 		for (j = 0; j < 32; j++) {
7917 			if (!(event_mask & desired_event) &&
7918 			    (ioc->event_masks[i] & desired_event)) {
7919 				ioc->event_masks[i] &= ~desired_event;
7920 				send_update_to_fw = 1;
7921 			}
7922 			desired_event = (desired_event << 1);
7923 		}
7924 	}
7925 
7926 	if (!send_update_to_fw)
7927 		return;
7928 
7929 	mutex_lock(&ioc->base_cmds.mutex);
7930 	_base_event_notification(ioc);
7931 	mutex_unlock(&ioc->base_cmds.mutex);
7932 }
7933 
7934 /**
7935 * mpt3sas_base_unlock_and_get_host_diagnostic- enable Host Diagnostic Register writes
7936 * @ioc: per adapter object
7937 * @host_diagnostic: host diagnostic register content
7938 *
7939 * Return: 0 for success, non-zero for failure.
7940 */
7941 
7942 int
7943 mpt3sas_base_unlock_and_get_host_diagnostic(struct MPT3SAS_ADAPTER *ioc,
7944 	u32 *host_diagnostic)
7945 {
7946 
7947 	u32 count;
7948 	*host_diagnostic = 0;
7949 	count = 0;
7950 
7951 	do {
7952 		/* Write magic sequence to WriteSequence register
7953 		 * Loop until in diagnostic mode
7954 		 */
7955 		drsprintk(ioc, ioc_info(ioc, "write magic sequence\n"));
7956 		writel(MPI2_WRSEQ_FLUSH_KEY_VALUE, &ioc->chip->WriteSequence);
7957 		writel(MPI2_WRSEQ_1ST_KEY_VALUE, &ioc->chip->WriteSequence);
7958 		writel(MPI2_WRSEQ_2ND_KEY_VALUE, &ioc->chip->WriteSequence);
7959 		writel(MPI2_WRSEQ_3RD_KEY_VALUE, &ioc->chip->WriteSequence);
7960 		writel(MPI2_WRSEQ_4TH_KEY_VALUE, &ioc->chip->WriteSequence);
7961 		writel(MPI2_WRSEQ_5TH_KEY_VALUE, &ioc->chip->WriteSequence);
7962 		writel(MPI2_WRSEQ_6TH_KEY_VALUE, &ioc->chip->WriteSequence);
7963 
7964 		/* wait 100 msec */
7965 		msleep(100);
7966 
7967 		if (count++ > 20) {
7968 			ioc_info(ioc,
7969 				    "Stop writing magic sequence after 20 retries\n");
7970 			_base_dump_reg_set(ioc);
7971 			return -EFAULT;
7972 		}
7973 
7974 		*host_diagnostic = ioc->base_readl_ext_retry(&ioc->chip->HostDiagnostic);
7975 		drsprintk(ioc,
7976 			     ioc_info(ioc, "wrote magic sequence: count(%d), host_diagnostic(0x%08x)\n",
7977 				     count, *host_diagnostic));
7978 
7979 	} while ((*host_diagnostic & MPI2_DIAG_DIAG_WRITE_ENABLE) == 0);
7980 	return 0;
7981 }
7982 
7983 /**
7984  * mpt3sas_base_lock_host_diagnostic: Disable Host Diagnostic Register writes
7985  * @ioc: per adapter object
7986  */
7987 
7988 void
7989 mpt3sas_base_lock_host_diagnostic(struct MPT3SAS_ADAPTER *ioc)
7990 {
7991 	drsprintk(ioc, ioc_info(ioc, "disable writes to the diagnostic register\n"));
7992 	writel(MPI2_WRSEQ_FLUSH_KEY_VALUE, &ioc->chip->WriteSequence);
7993 }
7994 
7995 /**
7996  * _base_diag_reset - the "big hammer" start of day reset
7997  * @ioc: per adapter object
7998  *
7999  * Return: 0 for success, non-zero for failure.
8000  */
8001 static int
8002 _base_diag_reset(struct MPT3SAS_ADAPTER *ioc)
8003 {
8004 	u32 host_diagnostic;
8005 	u32 ioc_state;
8006 	u32 count;
8007 	u32 hcb_size;
8008 
8009 	ioc_info(ioc, "sending diag reset !!\n");
8010 
8011 	pci_cfg_access_lock(ioc->pdev);
8012 
8013 	drsprintk(ioc, ioc_info(ioc, "clear interrupts\n"));
8014 
8015 	mutex_lock(&ioc->hostdiag_unlock_mutex);
8016 	if (mpt3sas_base_unlock_and_get_host_diagnostic(ioc, &host_diagnostic))
8017 		goto out;
8018 
8019 	hcb_size = ioc->base_readl(&ioc->chip->HCBSize);
8020 	drsprintk(ioc, ioc_info(ioc, "diag reset: issued\n"));
8021 	writel(host_diagnostic | MPI2_DIAG_RESET_ADAPTER,
8022 	     &ioc->chip->HostDiagnostic);
8023 
8024 	/* This delay allows the chip PCIe hardware time to finish reset tasks */
8025 	msleep(MPI2_HARD_RESET_PCIE_FIRST_READ_DELAY_MICRO_SEC/1000);
8026 
8027 	/* Approximately 300 second max wait */
8028 	for (count = 0; count < (300000000 /
8029 	    MPI2_HARD_RESET_PCIE_SECOND_READ_DELAY_MICRO_SEC); count++) {
8030 
8031 		host_diagnostic = ioc->base_readl_ext_retry(&ioc->chip->HostDiagnostic);
8032 
8033 		if (host_diagnostic == 0xFFFFFFFF) {
8034 			ioc_info(ioc,
8035 			    "Invalid host diagnostic register value\n");
8036 			_base_dump_reg_set(ioc);
8037 			goto out;
8038 		}
8039 		if (!(host_diagnostic & MPI2_DIAG_RESET_ADAPTER))
8040 			break;
8041 
8042 		/* Wait to pass the second read delay window */
8043 		msleep(MPI2_HARD_RESET_PCIE_SECOND_READ_DELAY_MICRO_SEC/1000);
8044 	}
8045 
8046 	if (host_diagnostic & MPI2_DIAG_HCB_MODE) {
8047 
8048 		drsprintk(ioc,
8049 			ioc_info(ioc, "restart the adapter assuming the\n"
8050 					"HCB Address points to good F/W\n"));
8051 		host_diagnostic &= ~MPI2_DIAG_BOOT_DEVICE_SELECT_MASK;
8052 		host_diagnostic |= MPI2_DIAG_BOOT_DEVICE_SELECT_HCDW;
8053 		writel(host_diagnostic, &ioc->chip->HostDiagnostic);
8054 
8055 		drsprintk(ioc, ioc_info(ioc, "re-enable the HCDW\n"));
8056 		writel(hcb_size | MPI2_HCB_SIZE_HCB_ENABLE,
8057 		    &ioc->chip->HCBSize);
8058 	}
8059 
8060 	drsprintk(ioc, ioc_info(ioc, "restart the adapter\n"));
8061 	writel(host_diagnostic & ~MPI2_DIAG_HOLD_IOC_RESET,
8062 	    &ioc->chip->HostDiagnostic);
8063 
8064 	mpt3sas_base_lock_host_diagnostic(ioc);
8065 	mutex_unlock(&ioc->hostdiag_unlock_mutex);
8066 
8067 	drsprintk(ioc, ioc_info(ioc, "Wait for FW to go to the READY state\n"));
8068 	ioc_state = _base_wait_on_iocstate(ioc, MPI2_IOC_STATE_READY, 20);
8069 	if (ioc_state) {
8070 		ioc_err(ioc, "%s: failed going to ready state (ioc_state=0x%x)\n",
8071 			__func__, ioc_state);
8072 		_base_dump_reg_set(ioc);
8073 		goto out;
8074 	}
8075 
8076 	pci_cfg_access_unlock(ioc->pdev);
8077 	ioc_info(ioc, "diag reset: SUCCESS\n");
8078 	return 0;
8079 
8080  out:
8081 	pci_cfg_access_unlock(ioc->pdev);
8082 	ioc_err(ioc, "diag reset: FAILED\n");
8083 	mutex_unlock(&ioc->hostdiag_unlock_mutex);
8084 	return -EFAULT;
8085 }
8086 
8087 /**
8088  * mpt3sas_base_make_ioc_ready - put controller in READY state
8089  * @ioc: per adapter object
8090  * @type: FORCE_BIG_HAMMER or SOFT_RESET
8091  *
8092  * Return: 0 for success, non-zero for failure.
8093  */
8094 int
8095 mpt3sas_base_make_ioc_ready(struct MPT3SAS_ADAPTER *ioc, enum reset_type type)
8096 {
8097 	u32 ioc_state;
8098 	int rc;
8099 	int count;
8100 
8101 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
8102 
8103 	if (ioc->pci_error_recovery)
8104 		return 0;
8105 
8106 	ioc_state = mpt3sas_base_get_iocstate(ioc, 0);
8107 	dhsprintk(ioc,
8108 		  ioc_info(ioc, "%s: ioc_state(0x%08x)\n",
8109 			   __func__, ioc_state));
8110 
8111 	/* if in RESET state, it should move to READY state shortly */
8112 	count = 0;
8113 	if ((ioc_state & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_RESET) {
8114 		while ((ioc_state & MPI2_IOC_STATE_MASK) !=
8115 		    MPI2_IOC_STATE_READY) {
8116 			if (count++ == 10) {
8117 				ioc_err(ioc, "%s: failed going to ready state (ioc_state=0x%x)\n",
8118 					__func__, ioc_state);
8119 				return -EFAULT;
8120 			}
8121 			ssleep(1);
8122 			ioc_state = mpt3sas_base_get_iocstate(ioc, 0);
8123 		}
8124 	}
8125 
8126 	if ((ioc_state & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_READY)
8127 		return 0;
8128 
8129 	if (ioc_state & MPI2_DOORBELL_USED) {
8130 		ioc_info(ioc, "unexpected doorbell active!\n");
8131 		goto issue_diag_reset;
8132 	}
8133 
8134 	if ((ioc_state & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT) {
8135 		mpt3sas_print_fault_code(ioc, ioc_state &
8136 		    MPI2_DOORBELL_DATA_MASK);
8137 		goto issue_diag_reset;
8138 	}
8139 
8140 	if ((ioc_state & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_COREDUMP) {
8141 		/*
8142 		 * if host reset is invoked while watch dog thread is waiting
8143 		 * for IOC state to be changed to Fault state then driver has
8144 		 * to wait here for CoreDump state to clear otherwise reset
8145 		 * will be issued to the FW and FW move the IOC state to
8146 		 * reset state without copying the FW logs to coredump region.
8147 		 */
8148 		if (ioc->ioc_coredump_loop != MPT3SAS_COREDUMP_LOOP_DONE) {
8149 			mpt3sas_print_coredump_info(ioc, ioc_state &
8150 			    MPI2_DOORBELL_DATA_MASK);
8151 			mpt3sas_base_wait_for_coredump_completion(ioc,
8152 			    __func__);
8153 		}
8154 		goto issue_diag_reset;
8155 	}
8156 
8157 	if (type == FORCE_BIG_HAMMER)
8158 		goto issue_diag_reset;
8159 
8160 	if ((ioc_state & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_OPERATIONAL)
8161 		if (!(_base_send_ioc_reset(ioc,
8162 		    MPI2_FUNCTION_IOC_MESSAGE_UNIT_RESET, 15))) {
8163 			return 0;
8164 	}
8165 
8166  issue_diag_reset:
8167 	rc = _base_diag_reset(ioc);
8168 	return rc;
8169 }
8170 
8171 /**
8172  * _base_make_ioc_operational - put controller in OPERATIONAL state
8173  * @ioc: per adapter object
8174  *
8175  * Return: 0 for success, non-zero for failure.
8176  */
8177 static int
8178 _base_make_ioc_operational(struct MPT3SAS_ADAPTER *ioc)
8179 {
8180 	int r, i, index, rc;
8181 	unsigned long	flags;
8182 	u32 reply_address;
8183 	u16 smid;
8184 	struct _tr_list *delayed_tr, *delayed_tr_next;
8185 	struct _sc_list *delayed_sc, *delayed_sc_next;
8186 	struct _event_ack_list *delayed_event_ack, *delayed_event_ack_next;
8187 	u8 hide_flag;
8188 	struct adapter_reply_queue *reply_q;
8189 	Mpi2ReplyDescriptorsUnion_t *reply_post_free_contig;
8190 
8191 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
8192 
8193 	/* clean the delayed target reset list */
8194 	list_for_each_entry_safe(delayed_tr, delayed_tr_next,
8195 	    &ioc->delayed_tr_list, list) {
8196 		list_del(&delayed_tr->list);
8197 		kfree(delayed_tr);
8198 	}
8199 
8200 
8201 	list_for_each_entry_safe(delayed_tr, delayed_tr_next,
8202 	    &ioc->delayed_tr_volume_list, list) {
8203 		list_del(&delayed_tr->list);
8204 		kfree(delayed_tr);
8205 	}
8206 
8207 	list_for_each_entry_safe(delayed_sc, delayed_sc_next,
8208 	    &ioc->delayed_sc_list, list) {
8209 		list_del(&delayed_sc->list);
8210 		kfree(delayed_sc);
8211 	}
8212 
8213 	list_for_each_entry_safe(delayed_event_ack, delayed_event_ack_next,
8214 	    &ioc->delayed_event_ack_list, list) {
8215 		list_del(&delayed_event_ack->list);
8216 		kfree(delayed_event_ack);
8217 	}
8218 
8219 	spin_lock_irqsave(&ioc->scsi_lookup_lock, flags);
8220 
8221 	/* hi-priority queue */
8222 	INIT_LIST_HEAD(&ioc->hpr_free_list);
8223 	smid = ioc->hi_priority_smid;
8224 	for (i = 0; i < ioc->hi_priority_depth; i++, smid++) {
8225 		ioc->hpr_lookup[i].cb_idx = 0xFF;
8226 		ioc->hpr_lookup[i].smid = smid;
8227 		list_add_tail(&ioc->hpr_lookup[i].tracker_list,
8228 		    &ioc->hpr_free_list);
8229 	}
8230 
8231 	/* internal queue */
8232 	INIT_LIST_HEAD(&ioc->internal_free_list);
8233 	smid = ioc->internal_smid;
8234 	for (i = 0; i < ioc->internal_depth; i++, smid++) {
8235 		ioc->internal_lookup[i].cb_idx = 0xFF;
8236 		ioc->internal_lookup[i].smid = smid;
8237 		list_add_tail(&ioc->internal_lookup[i].tracker_list,
8238 		    &ioc->internal_free_list);
8239 	}
8240 
8241 	spin_unlock_irqrestore(&ioc->scsi_lookup_lock, flags);
8242 
8243 	/* initialize Reply Free Queue */
8244 	for (i = 0, reply_address = (u32)ioc->reply_dma ;
8245 	    i < ioc->reply_free_queue_depth ; i++, reply_address +=
8246 	    ioc->reply_sz) {
8247 		ioc->reply_free[i] = cpu_to_le32(reply_address);
8248 		if (ioc->is_mcpu_endpoint)
8249 			_base_clone_reply_to_sys_mem(ioc,
8250 					reply_address, i);
8251 	}
8252 
8253 	/* initialize reply queues */
8254 	if (ioc->is_driver_loading)
8255 		_base_assign_reply_queues(ioc);
8256 
8257 	/* initialize Reply Post Free Queue */
8258 	index = 0;
8259 	reply_post_free_contig = ioc->reply_post[0].reply_post_free;
8260 	list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
8261 		/*
8262 		 * If RDPQ is enabled, switch to the next allocation.
8263 		 * Otherwise advance within the contiguous region.
8264 		 */
8265 		if (ioc->rdpq_array_enable) {
8266 			reply_q->reply_post_free =
8267 				ioc->reply_post[index++].reply_post_free;
8268 		} else {
8269 			reply_q->reply_post_free = reply_post_free_contig;
8270 			reply_post_free_contig += ioc->reply_post_queue_depth;
8271 		}
8272 
8273 		reply_q->reply_post_host_index = 0;
8274 		for (i = 0; i < ioc->reply_post_queue_depth; i++)
8275 			reply_q->reply_post_free[i].Words =
8276 			    cpu_to_le64(ULLONG_MAX);
8277 		if (!_base_is_controller_msix_enabled(ioc))
8278 			goto skip_init_reply_post_free_queue;
8279 	}
8280  skip_init_reply_post_free_queue:
8281 
8282 	r = _base_send_ioc_init(ioc);
8283 	if (r) {
8284 		/*
8285 		 * No need to check IOC state for fault state & issue
8286 		 * diag reset during host reset. This check is need
8287 		 * only during driver load time.
8288 		 */
8289 		if (!ioc->is_driver_loading)
8290 			return r;
8291 
8292 		rc = mpt3sas_base_check_for_fault_and_issue_reset(ioc);
8293 		if (rc || (_base_send_ioc_init(ioc)))
8294 			return r;
8295 	}
8296 
8297 	/* initialize reply free host index */
8298 	ioc->reply_free_host_index = ioc->reply_free_queue_depth - 1;
8299 	writel(ioc->reply_free_host_index, &ioc->chip->ReplyFreeHostIndex);
8300 
8301 	/* initialize reply post host index */
8302 	list_for_each_entry(reply_q, &ioc->reply_queue_list, list) {
8303 		if (ioc->combined_reply_queue)
8304 			writel((reply_q->msix_index & 7)<<
8305 			   MPI2_RPHI_MSIX_INDEX_SHIFT,
8306 			   ioc->replyPostRegisterIndex[reply_q->msix_index/8]);
8307 		else
8308 			writel(reply_q->msix_index <<
8309 				MPI2_RPHI_MSIX_INDEX_SHIFT,
8310 				&ioc->chip->ReplyPostHostIndex);
8311 
8312 		if (!_base_is_controller_msix_enabled(ioc))
8313 			goto skip_init_reply_post_host_index;
8314 	}
8315 
8316  skip_init_reply_post_host_index:
8317 
8318 	mpt3sas_base_unmask_interrupts(ioc);
8319 
8320 	if (ioc->hba_mpi_version_belonged != MPI2_VERSION) {
8321 		r = _base_display_fwpkg_version(ioc);
8322 		if (r)
8323 			return r;
8324 	}
8325 
8326 	r = _base_static_config_pages(ioc);
8327 	if (r)
8328 		return r;
8329 
8330 	r = _base_event_notification(ioc);
8331 	if (r)
8332 		return r;
8333 
8334 	if (!ioc->shost_recovery) {
8335 
8336 		if (ioc->is_warpdrive && ioc->manu_pg10.OEMIdentifier
8337 		    == 0x80) {
8338 			hide_flag = (u8) (
8339 			    le32_to_cpu(ioc->manu_pg10.OEMSpecificFlags0) &
8340 			    MFG_PAGE10_HIDE_SSDS_MASK);
8341 			if (hide_flag != MFG_PAGE10_HIDE_SSDS_MASK)
8342 				ioc->mfg_pg10_hide_flag = hide_flag;
8343 		}
8344 
8345 		ioc->wait_for_discovery_to_complete =
8346 		    _base_determine_wait_on_discovery(ioc);
8347 
8348 		return r; /* scan_start and scan_finished support */
8349 	}
8350 
8351 	r = _base_send_port_enable(ioc);
8352 	if (r)
8353 		return r;
8354 
8355 	return r;
8356 }
8357 
8358 /**
8359  * mpt3sas_base_free_resources - free resources controller resources
8360  * @ioc: per adapter object
8361  */
8362 void
8363 mpt3sas_base_free_resources(struct MPT3SAS_ADAPTER *ioc)
8364 {
8365 	dexitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
8366 
8367 	/* synchronizing freeing resource with pci_access_mutex lock */
8368 	mutex_lock(&ioc->pci_access_mutex);
8369 	if (ioc->chip_phys && ioc->chip) {
8370 		mpt3sas_base_mask_interrupts(ioc);
8371 		ioc->shost_recovery = 1;
8372 		mpt3sas_base_make_ioc_ready(ioc, SOFT_RESET);
8373 		ioc->shost_recovery = 0;
8374 	}
8375 
8376 	mpt3sas_base_unmap_resources(ioc);
8377 	mutex_unlock(&ioc->pci_access_mutex);
8378 	return;
8379 }
8380 
8381 /**
8382  * mpt3sas_base_attach - attach controller instance
8383  * @ioc: per adapter object
8384  *
8385  * Return: 0 for success, non-zero for failure.
8386  */
8387 int
8388 mpt3sas_base_attach(struct MPT3SAS_ADAPTER *ioc)
8389 {
8390 	int r, i, rc;
8391 	int cpu_id, last_cpu_id = 0;
8392 
8393 	dinitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
8394 
8395 	/* setup cpu_msix_table */
8396 	ioc->cpu_count = num_online_cpus();
8397 	for_each_online_cpu(cpu_id)
8398 		last_cpu_id = cpu_id;
8399 	ioc->cpu_msix_table_sz = last_cpu_id + 1;
8400 	ioc->cpu_msix_table = kzalloc(ioc->cpu_msix_table_sz, GFP_KERNEL);
8401 	ioc->reply_queue_count = 1;
8402 	if (!ioc->cpu_msix_table) {
8403 		ioc_info(ioc, "Allocation for cpu_msix_table failed!!!\n");
8404 		r = -ENOMEM;
8405 		goto out_free_resources;
8406 	}
8407 
8408 	if (ioc->is_warpdrive) {
8409 		ioc->reply_post_host_index = kcalloc(ioc->cpu_msix_table_sz,
8410 		    sizeof(resource_size_t *), GFP_KERNEL);
8411 		if (!ioc->reply_post_host_index) {
8412 			ioc_info(ioc, "Allocation for reply_post_host_index failed!!!\n");
8413 			r = -ENOMEM;
8414 			goto out_free_resources;
8415 		}
8416 	}
8417 
8418 	ioc->smp_affinity_enable = smp_affinity_enable;
8419 
8420 	ioc->rdpq_array_enable_assigned = 0;
8421 	ioc->use_32bit_dma = false;
8422 	ioc->dma_mask = 64;
8423 	if (ioc->is_aero_ioc) {
8424 		ioc->base_readl = &_base_readl_aero;
8425 		ioc->base_readl_ext_retry = &_base_readl_ext_retry;
8426 	} else {
8427 		ioc->base_readl = &_base_readl;
8428 		ioc->base_readl_ext_retry = &_base_readl;
8429 	}
8430 	r = mpt3sas_base_map_resources(ioc);
8431 	if (r)
8432 		goto out_free_resources;
8433 
8434 	pci_set_drvdata(ioc->pdev, ioc->shost);
8435 	r = _base_get_ioc_facts(ioc);
8436 	if (r) {
8437 		rc = mpt3sas_base_check_for_fault_and_issue_reset(ioc);
8438 		if (rc || (_base_get_ioc_facts(ioc)))
8439 			goto out_free_resources;
8440 	}
8441 
8442 	switch (ioc->hba_mpi_version_belonged) {
8443 	case MPI2_VERSION:
8444 		ioc->build_sg_scmd = &_base_build_sg_scmd;
8445 		ioc->build_sg = &_base_build_sg;
8446 		ioc->build_zero_len_sge = &_base_build_zero_len_sge;
8447 		ioc->get_msix_index_for_smlio = &_base_get_msix_index;
8448 		break;
8449 	case MPI25_VERSION:
8450 	case MPI26_VERSION:
8451 		/*
8452 		 * In SAS3.0,
8453 		 * SCSI_IO, SMP_PASSTHRU, SATA_PASSTHRU, Target Assist, and
8454 		 * Target Status - all require the IEEE formatted scatter gather
8455 		 * elements.
8456 		 */
8457 		ioc->build_sg_scmd = &_base_build_sg_scmd_ieee;
8458 		ioc->build_sg = &_base_build_sg_ieee;
8459 		ioc->build_nvme_prp = &_base_build_nvme_prp;
8460 		ioc->build_zero_len_sge = &_base_build_zero_len_sge_ieee;
8461 		ioc->sge_size_ieee = sizeof(Mpi2IeeeSgeSimple64_t);
8462 		if (ioc->high_iops_queues)
8463 			ioc->get_msix_index_for_smlio =
8464 					&_base_get_high_iops_msix_index;
8465 		else
8466 			ioc->get_msix_index_for_smlio = &_base_get_msix_index;
8467 		break;
8468 	}
8469 	if (ioc->atomic_desc_capable) {
8470 		ioc->put_smid_default = &_base_put_smid_default_atomic;
8471 		ioc->put_smid_scsi_io = &_base_put_smid_scsi_io_atomic;
8472 		ioc->put_smid_fast_path =
8473 				&_base_put_smid_fast_path_atomic;
8474 		ioc->put_smid_hi_priority =
8475 				&_base_put_smid_hi_priority_atomic;
8476 	} else {
8477 		ioc->put_smid_default = &_base_put_smid_default;
8478 		ioc->put_smid_fast_path = &_base_put_smid_fast_path;
8479 		ioc->put_smid_hi_priority = &_base_put_smid_hi_priority;
8480 		if (ioc->is_mcpu_endpoint)
8481 			ioc->put_smid_scsi_io =
8482 				&_base_put_smid_mpi_ep_scsi_io;
8483 		else
8484 			ioc->put_smid_scsi_io = &_base_put_smid_scsi_io;
8485 	}
8486 	/*
8487 	 * These function pointers for other requests that don't
8488 	 * the require IEEE scatter gather elements.
8489 	 *
8490 	 * For example Configuration Pages and SAS IOUNIT Control don't.
8491 	 */
8492 	ioc->build_sg_mpi = &_base_build_sg;
8493 	ioc->build_zero_len_sge_mpi = &_base_build_zero_len_sge;
8494 
8495 	r = mpt3sas_base_make_ioc_ready(ioc, SOFT_RESET);
8496 	if (r)
8497 		goto out_free_resources;
8498 
8499 	ioc->pfacts = kcalloc(ioc->facts.NumberOfPorts,
8500 	    sizeof(struct mpt3sas_port_facts), GFP_KERNEL);
8501 	if (!ioc->pfacts) {
8502 		r = -ENOMEM;
8503 		goto out_free_resources;
8504 	}
8505 
8506 	for (i = 0 ; i < ioc->facts.NumberOfPorts; i++) {
8507 		r = _base_get_port_facts(ioc, i);
8508 		if (r) {
8509 			rc = mpt3sas_base_check_for_fault_and_issue_reset(ioc);
8510 			if (rc || (_base_get_port_facts(ioc, i)))
8511 				goto out_free_resources;
8512 		}
8513 	}
8514 
8515 	r = _base_allocate_memory_pools(ioc);
8516 	if (r)
8517 		goto out_free_resources;
8518 
8519 	if (irqpoll_weight > 0)
8520 		ioc->thresh_hold = irqpoll_weight;
8521 	else
8522 		ioc->thresh_hold = ioc->hba_queue_depth/4;
8523 
8524 	_base_init_irqpolls(ioc);
8525 	init_waitqueue_head(&ioc->reset_wq);
8526 
8527 	/* allocate memory pd handle bitmask list */
8528 	ioc->pd_handles_sz = (ioc->facts.MaxDevHandle / 8);
8529 	if (ioc->facts.MaxDevHandle % 8)
8530 		ioc->pd_handles_sz++;
8531 	/*
8532 	 * pd_handles_sz should have, at least, the minimal room for
8533 	 * set_bit()/test_bit(), otherwise out-of-memory touch may occur.
8534 	 */
8535 	ioc->pd_handles_sz = ALIGN(ioc->pd_handles_sz, sizeof(unsigned long));
8536 
8537 	ioc->pd_handles = kzalloc(ioc->pd_handles_sz,
8538 	    GFP_KERNEL);
8539 	if (!ioc->pd_handles) {
8540 		r = -ENOMEM;
8541 		goto out_free_resources;
8542 	}
8543 	ioc->blocking_handles = kzalloc(ioc->pd_handles_sz,
8544 	    GFP_KERNEL);
8545 	if (!ioc->blocking_handles) {
8546 		r = -ENOMEM;
8547 		goto out_free_resources;
8548 	}
8549 
8550 	/* allocate memory for pending OS device add list */
8551 	ioc->pend_os_device_add_sz = (ioc->facts.MaxDevHandle / 8);
8552 	if (ioc->facts.MaxDevHandle % 8)
8553 		ioc->pend_os_device_add_sz++;
8554 
8555 	/*
8556 	 * pend_os_device_add_sz should have, at least, the minimal room for
8557 	 * set_bit()/test_bit(), otherwise out-of-memory may occur.
8558 	 */
8559 	ioc->pend_os_device_add_sz = ALIGN(ioc->pend_os_device_add_sz,
8560 					   sizeof(unsigned long));
8561 	ioc->pend_os_device_add = kzalloc(ioc->pend_os_device_add_sz,
8562 	    GFP_KERNEL);
8563 	if (!ioc->pend_os_device_add) {
8564 		r = -ENOMEM;
8565 		goto out_free_resources;
8566 	}
8567 
8568 	ioc->device_remove_in_progress_sz = ioc->pend_os_device_add_sz;
8569 	ioc->device_remove_in_progress =
8570 		kzalloc(ioc->device_remove_in_progress_sz, GFP_KERNEL);
8571 	if (!ioc->device_remove_in_progress) {
8572 		r = -ENOMEM;
8573 		goto out_free_resources;
8574 	}
8575 
8576 	ioc->fwfault_debug = mpt3sas_fwfault_debug;
8577 
8578 	/* base internal command bits */
8579 	mutex_init(&ioc->base_cmds.mutex);
8580 	ioc->base_cmds.reply = kzalloc(ioc->reply_sz, GFP_KERNEL);
8581 	ioc->base_cmds.status = MPT3_CMD_NOT_USED;
8582 
8583 	/* port_enable command bits */
8584 	ioc->port_enable_cmds.reply = kzalloc(ioc->reply_sz, GFP_KERNEL);
8585 	ioc->port_enable_cmds.status = MPT3_CMD_NOT_USED;
8586 
8587 	/* transport internal command bits */
8588 	ioc->transport_cmds.reply = kzalloc(ioc->reply_sz, GFP_KERNEL);
8589 	ioc->transport_cmds.status = MPT3_CMD_NOT_USED;
8590 	mutex_init(&ioc->transport_cmds.mutex);
8591 
8592 	/* scsih internal command bits */
8593 	ioc->scsih_cmds.reply = kzalloc(ioc->reply_sz, GFP_KERNEL);
8594 	ioc->scsih_cmds.status = MPT3_CMD_NOT_USED;
8595 	mutex_init(&ioc->scsih_cmds.mutex);
8596 
8597 	/* task management internal command bits */
8598 	ioc->tm_cmds.reply = kzalloc(ioc->reply_sz, GFP_KERNEL);
8599 	ioc->tm_cmds.status = MPT3_CMD_NOT_USED;
8600 	mutex_init(&ioc->tm_cmds.mutex);
8601 
8602 	/* config page internal command bits */
8603 	ioc->config_cmds.reply = kzalloc(ioc->reply_sz, GFP_KERNEL);
8604 	ioc->config_cmds.status = MPT3_CMD_NOT_USED;
8605 	mutex_init(&ioc->config_cmds.mutex);
8606 
8607 	/* ctl module internal command bits */
8608 	ioc->ctl_cmds.reply = kzalloc(ioc->reply_sz, GFP_KERNEL);
8609 	ioc->ctl_cmds.sense = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL);
8610 	ioc->ctl_cmds.status = MPT3_CMD_NOT_USED;
8611 	mutex_init(&ioc->ctl_cmds.mutex);
8612 
8613 	if (!ioc->base_cmds.reply || !ioc->port_enable_cmds.reply ||
8614 	    !ioc->transport_cmds.reply || !ioc->scsih_cmds.reply ||
8615 	    !ioc->tm_cmds.reply || !ioc->config_cmds.reply ||
8616 	    !ioc->ctl_cmds.reply || !ioc->ctl_cmds.sense) {
8617 		r = -ENOMEM;
8618 		goto out_free_resources;
8619 	}
8620 
8621 	for (i = 0; i < MPI2_EVENT_NOTIFY_EVENTMASK_WORDS; i++)
8622 		ioc->event_masks[i] = -1;
8623 
8624 	/* here we enable the events we care about */
8625 	_base_unmask_events(ioc, MPI2_EVENT_SAS_DISCOVERY);
8626 	_base_unmask_events(ioc, MPI2_EVENT_SAS_BROADCAST_PRIMITIVE);
8627 	_base_unmask_events(ioc, MPI2_EVENT_SAS_TOPOLOGY_CHANGE_LIST);
8628 	_base_unmask_events(ioc, MPI2_EVENT_SAS_DEVICE_STATUS_CHANGE);
8629 	_base_unmask_events(ioc, MPI2_EVENT_SAS_ENCL_DEVICE_STATUS_CHANGE);
8630 	_base_unmask_events(ioc, MPI2_EVENT_IR_CONFIGURATION_CHANGE_LIST);
8631 	_base_unmask_events(ioc, MPI2_EVENT_IR_VOLUME);
8632 	_base_unmask_events(ioc, MPI2_EVENT_IR_PHYSICAL_DISK);
8633 	_base_unmask_events(ioc, MPI2_EVENT_IR_OPERATION_STATUS);
8634 	_base_unmask_events(ioc, MPI2_EVENT_LOG_ENTRY_ADDED);
8635 	_base_unmask_events(ioc, MPI2_EVENT_TEMP_THRESHOLD);
8636 	_base_unmask_events(ioc, MPI2_EVENT_ACTIVE_CABLE_EXCEPTION);
8637 	_base_unmask_events(ioc, MPI2_EVENT_SAS_DEVICE_DISCOVERY_ERROR);
8638 	if (ioc->hba_mpi_version_belonged == MPI26_VERSION) {
8639 		if (ioc->is_gen35_ioc) {
8640 			_base_unmask_events(ioc,
8641 				MPI2_EVENT_PCIE_DEVICE_STATUS_CHANGE);
8642 			_base_unmask_events(ioc, MPI2_EVENT_PCIE_ENUMERATION);
8643 			_base_unmask_events(ioc,
8644 				MPI2_EVENT_PCIE_TOPOLOGY_CHANGE_LIST);
8645 		}
8646 	}
8647 	r = _base_make_ioc_operational(ioc);
8648 	if (r == -EAGAIN) {
8649 		r = _base_make_ioc_operational(ioc);
8650 		if (r)
8651 			goto out_free_resources;
8652 	}
8653 
8654 	/*
8655 	 * Copy current copy of IOCFacts in prev_fw_facts
8656 	 * and it will be used during online firmware upgrade.
8657 	 */
8658 	memcpy(&ioc->prev_fw_facts, &ioc->facts,
8659 	    sizeof(struct mpt3sas_facts));
8660 
8661 	ioc->non_operational_loop = 0;
8662 	ioc->ioc_coredump_loop = 0;
8663 	ioc->got_task_abort_from_ioctl = 0;
8664 	return 0;
8665 
8666  out_free_resources:
8667 
8668 	ioc->remove_host = 1;
8669 
8670 	mpt3sas_base_free_resources(ioc);
8671 	_base_release_memory_pools(ioc);
8672 	pci_set_drvdata(ioc->pdev, NULL);
8673 	kfree(ioc->cpu_msix_table);
8674 	if (ioc->is_warpdrive)
8675 		kfree(ioc->reply_post_host_index);
8676 	kfree(ioc->pd_handles);
8677 	kfree(ioc->blocking_handles);
8678 	kfree(ioc->device_remove_in_progress);
8679 	kfree(ioc->pend_os_device_add);
8680 	kfree(ioc->tm_cmds.reply);
8681 	kfree(ioc->transport_cmds.reply);
8682 	kfree(ioc->scsih_cmds.reply);
8683 	kfree(ioc->config_cmds.reply);
8684 	kfree(ioc->base_cmds.reply);
8685 	kfree(ioc->port_enable_cmds.reply);
8686 	kfree(ioc->ctl_cmds.reply);
8687 	kfree(ioc->ctl_cmds.sense);
8688 	kfree(ioc->pfacts);
8689 	ioc->ctl_cmds.reply = NULL;
8690 	ioc->base_cmds.reply = NULL;
8691 	ioc->tm_cmds.reply = NULL;
8692 	ioc->scsih_cmds.reply = NULL;
8693 	ioc->transport_cmds.reply = NULL;
8694 	ioc->config_cmds.reply = NULL;
8695 	ioc->pfacts = NULL;
8696 	return r;
8697 }
8698 
8699 
8700 /**
8701  * mpt3sas_base_detach - remove controller instance
8702  * @ioc: per adapter object
8703  */
8704 void
8705 mpt3sas_base_detach(struct MPT3SAS_ADAPTER *ioc)
8706 {
8707 	dexitprintk(ioc, ioc_info(ioc, "%s\n", __func__));
8708 
8709 	mpt3sas_base_stop_watchdog(ioc);
8710 	mpt3sas_base_free_resources(ioc);
8711 	_base_release_memory_pools(ioc);
8712 	mpt3sas_free_enclosure_list(ioc);
8713 	pci_set_drvdata(ioc->pdev, NULL);
8714 	kfree(ioc->cpu_msix_table);
8715 	if (ioc->is_warpdrive)
8716 		kfree(ioc->reply_post_host_index);
8717 	kfree(ioc->pd_handles);
8718 	kfree(ioc->blocking_handles);
8719 	kfree(ioc->device_remove_in_progress);
8720 	kfree(ioc->pend_os_device_add);
8721 	kfree(ioc->pfacts);
8722 	kfree(ioc->ctl_cmds.reply);
8723 	kfree(ioc->ctl_cmds.sense);
8724 	kfree(ioc->base_cmds.reply);
8725 	kfree(ioc->port_enable_cmds.reply);
8726 	kfree(ioc->tm_cmds.reply);
8727 	kfree(ioc->transport_cmds.reply);
8728 	kfree(ioc->scsih_cmds.reply);
8729 	kfree(ioc->config_cmds.reply);
8730 }
8731 
8732 /**
8733  * _base_pre_reset_handler - pre reset handler
8734  * @ioc: per adapter object
8735  */
8736 static void _base_pre_reset_handler(struct MPT3SAS_ADAPTER *ioc)
8737 {
8738 	mpt3sas_scsih_pre_reset_handler(ioc);
8739 	mpt3sas_ctl_pre_reset_handler(ioc);
8740 	dtmprintk(ioc, ioc_info(ioc, "%s: MPT3_IOC_PRE_RESET\n", __func__));
8741 }
8742 
8743 /**
8744  * _base_clear_outstanding_mpt_commands - clears outstanding mpt commands
8745  * @ioc: per adapter object
8746  */
8747 static void
8748 _base_clear_outstanding_mpt_commands(struct MPT3SAS_ADAPTER *ioc)
8749 {
8750 	dtmprintk(ioc,
8751 	    ioc_info(ioc, "%s: clear outstanding mpt cmds\n", __func__));
8752 	if (ioc->transport_cmds.status & MPT3_CMD_PENDING) {
8753 		ioc->transport_cmds.status |= MPT3_CMD_RESET;
8754 		mpt3sas_base_free_smid(ioc, ioc->transport_cmds.smid);
8755 		complete(&ioc->transport_cmds.done);
8756 	}
8757 	if (ioc->base_cmds.status & MPT3_CMD_PENDING) {
8758 		ioc->base_cmds.status |= MPT3_CMD_RESET;
8759 		mpt3sas_base_free_smid(ioc, ioc->base_cmds.smid);
8760 		complete(&ioc->base_cmds.done);
8761 	}
8762 	if (ioc->port_enable_cmds.status & MPT3_CMD_PENDING) {
8763 		ioc->port_enable_failed = 1;
8764 		ioc->port_enable_cmds.status |= MPT3_CMD_RESET;
8765 		mpt3sas_base_free_smid(ioc, ioc->port_enable_cmds.smid);
8766 		if (ioc->is_driver_loading) {
8767 			ioc->start_scan_failed =
8768 				MPI2_IOCSTATUS_INTERNAL_ERROR;
8769 			ioc->start_scan = 0;
8770 		} else {
8771 			complete(&ioc->port_enable_cmds.done);
8772 		}
8773 	}
8774 	if (ioc->config_cmds.status & MPT3_CMD_PENDING) {
8775 		ioc->config_cmds.status |= MPT3_CMD_RESET;
8776 		mpt3sas_base_free_smid(ioc, ioc->config_cmds.smid);
8777 		ioc->config_cmds.smid = USHRT_MAX;
8778 		complete(&ioc->config_cmds.done);
8779 	}
8780 }
8781 
8782 /**
8783  * _base_clear_outstanding_commands - clear all outstanding commands
8784  * @ioc: per adapter object
8785  */
8786 static void _base_clear_outstanding_commands(struct MPT3SAS_ADAPTER *ioc)
8787 {
8788 	mpt3sas_scsih_clear_outstanding_scsi_tm_commands(ioc);
8789 	mpt3sas_ctl_clear_outstanding_ioctls(ioc);
8790 	_base_clear_outstanding_mpt_commands(ioc);
8791 }
8792 
8793 /**
8794  * _base_reset_done_handler - reset done handler
8795  * @ioc: per adapter object
8796  */
8797 static void _base_reset_done_handler(struct MPT3SAS_ADAPTER *ioc)
8798 {
8799 	mpt3sas_scsih_reset_done_handler(ioc);
8800 	mpt3sas_ctl_reset_done_handler(ioc);
8801 	dtmprintk(ioc, ioc_info(ioc, "%s: MPT3_IOC_DONE_RESET\n", __func__));
8802 }
8803 
8804 /**
8805  * mpt3sas_wait_for_commands_to_complete - reset controller
8806  * @ioc: Pointer to MPT_ADAPTER structure
8807  *
8808  * This function is waiting 10s for all pending commands to complete
8809  * prior to putting controller in reset.
8810  */
8811 void
8812 mpt3sas_wait_for_commands_to_complete(struct MPT3SAS_ADAPTER *ioc)
8813 {
8814 	u32 ioc_state;
8815 
8816 	ioc->pending_io_count = 0;
8817 
8818 	ioc_state = mpt3sas_base_get_iocstate(ioc, 0);
8819 	if ((ioc_state & MPI2_IOC_STATE_MASK) != MPI2_IOC_STATE_OPERATIONAL)
8820 		return;
8821 
8822 	/* pending command count */
8823 	ioc->pending_io_count = scsi_host_busy(ioc->shost);
8824 
8825 	if (!ioc->pending_io_count)
8826 		return;
8827 
8828 	/* wait for pending commands to complete */
8829 	wait_event_timeout(ioc->reset_wq, ioc->pending_io_count == 0, 10 * HZ);
8830 }
8831 
8832 /**
8833  * _base_check_ioc_facts_changes - Look for increase/decrease of IOCFacts
8834  *     attributes during online firmware upgrade and update the corresponding
8835  *     IOC variables accordingly.
8836  *
8837  * @ioc: Pointer to MPT_ADAPTER structure
8838  */
8839 static int
8840 _base_check_ioc_facts_changes(struct MPT3SAS_ADAPTER *ioc)
8841 {
8842 	u16 pd_handles_sz;
8843 	void *pd_handles = NULL, *blocking_handles = NULL;
8844 	void *pend_os_device_add = NULL, *device_remove_in_progress = NULL;
8845 	struct mpt3sas_facts *old_facts = &ioc->prev_fw_facts;
8846 
8847 	if (ioc->facts.MaxDevHandle > old_facts->MaxDevHandle) {
8848 		pd_handles_sz = (ioc->facts.MaxDevHandle / 8);
8849 		if (ioc->facts.MaxDevHandle % 8)
8850 			pd_handles_sz++;
8851 
8852 		/*
8853 		 * pd_handles should have, at least, the minimal room for
8854 		 * set_bit()/test_bit(), otherwise out-of-memory touch may
8855 		 * occur.
8856 		 */
8857 		pd_handles_sz = ALIGN(pd_handles_sz, sizeof(unsigned long));
8858 		pd_handles = krealloc(ioc->pd_handles, pd_handles_sz,
8859 		    GFP_KERNEL);
8860 		if (!pd_handles) {
8861 			ioc_info(ioc,
8862 			    "Unable to allocate the memory for pd_handles of sz: %d\n",
8863 			    pd_handles_sz);
8864 			return -ENOMEM;
8865 		}
8866 		memset(pd_handles + ioc->pd_handles_sz, 0,
8867 		    (pd_handles_sz - ioc->pd_handles_sz));
8868 		ioc->pd_handles = pd_handles;
8869 
8870 		blocking_handles = krealloc(ioc->blocking_handles,
8871 		    pd_handles_sz, GFP_KERNEL);
8872 		if (!blocking_handles) {
8873 			ioc_info(ioc,
8874 			    "Unable to allocate the memory for "
8875 			    "blocking_handles of sz: %d\n",
8876 			    pd_handles_sz);
8877 			return -ENOMEM;
8878 		}
8879 		memset(blocking_handles + ioc->pd_handles_sz, 0,
8880 		    (pd_handles_sz - ioc->pd_handles_sz));
8881 		ioc->blocking_handles = blocking_handles;
8882 		ioc->pd_handles_sz = pd_handles_sz;
8883 
8884 		pend_os_device_add = krealloc(ioc->pend_os_device_add,
8885 		    pd_handles_sz, GFP_KERNEL);
8886 		if (!pend_os_device_add) {
8887 			ioc_info(ioc,
8888 			    "Unable to allocate the memory for pend_os_device_add of sz: %d\n",
8889 			    pd_handles_sz);
8890 			return -ENOMEM;
8891 		}
8892 		memset(pend_os_device_add + ioc->pend_os_device_add_sz, 0,
8893 		    (pd_handles_sz - ioc->pend_os_device_add_sz));
8894 		ioc->pend_os_device_add = pend_os_device_add;
8895 		ioc->pend_os_device_add_sz = pd_handles_sz;
8896 
8897 		device_remove_in_progress = krealloc(
8898 		    ioc->device_remove_in_progress, pd_handles_sz, GFP_KERNEL);
8899 		if (!device_remove_in_progress) {
8900 			ioc_info(ioc,
8901 			    "Unable to allocate the memory for "
8902 			    "device_remove_in_progress of sz: %d\n "
8903 			    , pd_handles_sz);
8904 			return -ENOMEM;
8905 		}
8906 		memset(device_remove_in_progress +
8907 		    ioc->device_remove_in_progress_sz, 0,
8908 		    (pd_handles_sz - ioc->device_remove_in_progress_sz));
8909 		ioc->device_remove_in_progress = device_remove_in_progress;
8910 		ioc->device_remove_in_progress_sz = pd_handles_sz;
8911 	}
8912 
8913 	memcpy(&ioc->prev_fw_facts, &ioc->facts, sizeof(struct mpt3sas_facts));
8914 	return 0;
8915 }
8916 
8917 /**
8918  * mpt3sas_base_hard_reset_handler - reset controller
8919  * @ioc: Pointer to MPT_ADAPTER structure
8920  * @type: FORCE_BIG_HAMMER or SOFT_RESET
8921  *
8922  * Return: 0 for success, non-zero for failure.
8923  */
8924 int
8925 mpt3sas_base_hard_reset_handler(struct MPT3SAS_ADAPTER *ioc,
8926 	enum reset_type type)
8927 {
8928 	int r;
8929 	unsigned long flags;
8930 	u32 ioc_state;
8931 	u8 is_fault = 0, is_trigger = 0;
8932 
8933 	dtmprintk(ioc, ioc_info(ioc, "%s: enter\n", __func__));
8934 
8935 	if (ioc->pci_error_recovery) {
8936 		ioc_err(ioc, "%s: pci error recovery reset\n", __func__);
8937 		r = 0;
8938 		goto out_unlocked;
8939 	}
8940 
8941 	if (mpt3sas_fwfault_debug)
8942 		mpt3sas_halt_firmware(ioc);
8943 
8944 	/* wait for an active reset in progress to complete */
8945 	mutex_lock(&ioc->reset_in_progress_mutex);
8946 
8947 	spin_lock_irqsave(&ioc->ioc_reset_in_progress_lock, flags);
8948 	ioc->shost_recovery = 1;
8949 	spin_unlock_irqrestore(&ioc->ioc_reset_in_progress_lock, flags);
8950 
8951 	if ((ioc->diag_buffer_status[MPI2_DIAG_BUF_TYPE_TRACE] &
8952 	    MPT3_DIAG_BUFFER_IS_REGISTERED) &&
8953 	    (!(ioc->diag_buffer_status[MPI2_DIAG_BUF_TYPE_TRACE] &
8954 	    MPT3_DIAG_BUFFER_IS_RELEASED))) {
8955 		is_trigger = 1;
8956 		ioc_state = mpt3sas_base_get_iocstate(ioc, 0);
8957 		if ((ioc_state & MPI2_IOC_STATE_MASK) == MPI2_IOC_STATE_FAULT ||
8958 		    (ioc_state & MPI2_IOC_STATE_MASK) ==
8959 		    MPI2_IOC_STATE_COREDUMP) {
8960 			is_fault = 1;
8961 			ioc->htb_rel.trigger_info_dwords[1] =
8962 			    (ioc_state & MPI2_DOORBELL_DATA_MASK);
8963 		}
8964 	}
8965 	_base_pre_reset_handler(ioc);
8966 	mpt3sas_wait_for_commands_to_complete(ioc);
8967 	mpt3sas_base_mask_interrupts(ioc);
8968 	mpt3sas_base_pause_mq_polling(ioc);
8969 	r = mpt3sas_base_make_ioc_ready(ioc, type);
8970 	if (r)
8971 		goto out;
8972 	_base_clear_outstanding_commands(ioc);
8973 
8974 	/* If this hard reset is called while port enable is active, then
8975 	 * there is no reason to call make_ioc_operational
8976 	 */
8977 	if (ioc->is_driver_loading && ioc->port_enable_failed) {
8978 		ioc->remove_host = 1;
8979 		r = -EFAULT;
8980 		goto out;
8981 	}
8982 	r = _base_get_ioc_facts(ioc);
8983 	if (r)
8984 		goto out;
8985 
8986 	r = _base_check_ioc_facts_changes(ioc);
8987 	if (r) {
8988 		ioc_info(ioc,
8989 		    "Some of the parameters got changed in this new firmware"
8990 		    " image and it requires system reboot\n");
8991 		goto out;
8992 	}
8993 	if (ioc->rdpq_array_enable && !ioc->rdpq_array_capable)
8994 		panic("%s: Issue occurred with flashing controller firmware."
8995 		      "Please reboot the system and ensure that the correct"
8996 		      " firmware version is running\n", ioc->name);
8997 
8998 	r = _base_make_ioc_operational(ioc);
8999 	if (!r)
9000 		_base_reset_done_handler(ioc);
9001 
9002  out:
9003 	ioc_info(ioc, "%s: %s\n", __func__, r == 0 ? "SUCCESS" : "FAILED");
9004 
9005 	spin_lock_irqsave(&ioc->ioc_reset_in_progress_lock, flags);
9006 	ioc->shost_recovery = 0;
9007 	spin_unlock_irqrestore(&ioc->ioc_reset_in_progress_lock, flags);
9008 	ioc->ioc_reset_count++;
9009 	mutex_unlock(&ioc->reset_in_progress_mutex);
9010 	mpt3sas_base_resume_mq_polling(ioc);
9011 
9012  out_unlocked:
9013 	if ((r == 0) && is_trigger) {
9014 		if (is_fault)
9015 			mpt3sas_trigger_master(ioc, MASTER_TRIGGER_FW_FAULT);
9016 		else
9017 			mpt3sas_trigger_master(ioc,
9018 			    MASTER_TRIGGER_ADAPTER_RESET);
9019 	}
9020 	dtmprintk(ioc, ioc_info(ioc, "%s: exit\n", __func__));
9021 	return r;
9022 }
9023