xref: /linux/drivers/infiniband/hw/mthca/mthca_main.c (revision 0d456bad36d42d16022be045c8a53ddbb59ee478)
1 /*
2  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
3  * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4  * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
5  *
6  * This software is available to you under a choice of one of two
7  * licenses.  You may choose to be licensed under the terms of the GNU
8  * General Public License (GPL) Version 2, available from the file
9  * COPYING in the main directory of this source tree, or the
10  * OpenIB.org BSD license below:
11  *
12  *     Redistribution and use in source and binary forms, with or
13  *     without modification, are permitted provided that the following
14  *     conditions are met:
15  *
16  *      - Redistributions of source code must retain the above
17  *        copyright notice, this list of conditions and the following
18  *        disclaimer.
19  *
20  *      - Redistributions in binary form must reproduce the above
21  *        copyright notice, this list of conditions and the following
22  *        disclaimer in the documentation and/or other materials
23  *        provided with the distribution.
24  *
25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
32  * SOFTWARE.
33  */
34 
35 #include <linux/module.h>
36 #include <linux/init.h>
37 #include <linux/errno.h>
38 #include <linux/pci.h>
39 #include <linux/interrupt.h>
40 #include <linux/gfp.h>
41 
42 #include "mthca_dev.h"
43 #include "mthca_config_reg.h"
44 #include "mthca_cmd.h"
45 #include "mthca_profile.h"
46 #include "mthca_memfree.h"
47 #include "mthca_wqe.h"
48 
49 MODULE_AUTHOR("Roland Dreier");
50 MODULE_DESCRIPTION("Mellanox InfiniBand HCA low-level driver");
51 MODULE_LICENSE("Dual BSD/GPL");
52 MODULE_VERSION(DRV_VERSION);
53 
54 #ifdef CONFIG_INFINIBAND_MTHCA_DEBUG
55 
56 int mthca_debug_level = 0;
57 module_param_named(debug_level, mthca_debug_level, int, 0644);
58 MODULE_PARM_DESC(debug_level, "Enable debug tracing if > 0");
59 
60 #endif /* CONFIG_INFINIBAND_MTHCA_DEBUG */
61 
62 #ifdef CONFIG_PCI_MSI
63 
64 static int msi_x = 1;
65 module_param(msi_x, int, 0444);
66 MODULE_PARM_DESC(msi_x, "attempt to use MSI-X if nonzero");
67 
68 #else /* CONFIG_PCI_MSI */
69 
70 #define msi_x (0)
71 
72 #endif /* CONFIG_PCI_MSI */
73 
74 static int tune_pci = 0;
75 module_param(tune_pci, int, 0444);
76 MODULE_PARM_DESC(tune_pci, "increase PCI burst from the default set by BIOS if nonzero");
77 
78 DEFINE_MUTEX(mthca_device_mutex);
79 
80 #define MTHCA_DEFAULT_NUM_QP            (1 << 16)
81 #define MTHCA_DEFAULT_RDB_PER_QP        (1 << 2)
82 #define MTHCA_DEFAULT_NUM_CQ            (1 << 16)
83 #define MTHCA_DEFAULT_NUM_MCG           (1 << 13)
84 #define MTHCA_DEFAULT_NUM_MPT           (1 << 17)
85 #define MTHCA_DEFAULT_NUM_MTT           (1 << 20)
86 #define MTHCA_DEFAULT_NUM_UDAV          (1 << 15)
87 #define MTHCA_DEFAULT_NUM_RESERVED_MTTS (1 << 18)
88 #define MTHCA_DEFAULT_NUM_UARC_SIZE     (1 << 18)
89 
90 static struct mthca_profile hca_profile = {
91 	.num_qp             = MTHCA_DEFAULT_NUM_QP,
92 	.rdb_per_qp         = MTHCA_DEFAULT_RDB_PER_QP,
93 	.num_cq             = MTHCA_DEFAULT_NUM_CQ,
94 	.num_mcg            = MTHCA_DEFAULT_NUM_MCG,
95 	.num_mpt            = MTHCA_DEFAULT_NUM_MPT,
96 	.num_mtt            = MTHCA_DEFAULT_NUM_MTT,
97 	.num_udav           = MTHCA_DEFAULT_NUM_UDAV,          /* Tavor only */
98 	.fmr_reserved_mtts  = MTHCA_DEFAULT_NUM_RESERVED_MTTS, /* Tavor only */
99 	.uarc_size          = MTHCA_DEFAULT_NUM_UARC_SIZE,     /* Arbel only */
100 };
101 
102 module_param_named(num_qp, hca_profile.num_qp, int, 0444);
103 MODULE_PARM_DESC(num_qp, "maximum number of QPs per HCA");
104 
105 module_param_named(rdb_per_qp, hca_profile.rdb_per_qp, int, 0444);
106 MODULE_PARM_DESC(rdb_per_qp, "number of RDB buffers per QP");
107 
108 module_param_named(num_cq, hca_profile.num_cq, int, 0444);
109 MODULE_PARM_DESC(num_cq, "maximum number of CQs per HCA");
110 
111 module_param_named(num_mcg, hca_profile.num_mcg, int, 0444);
112 MODULE_PARM_DESC(num_mcg, "maximum number of multicast groups per HCA");
113 
114 module_param_named(num_mpt, hca_profile.num_mpt, int, 0444);
115 MODULE_PARM_DESC(num_mpt,
116 		"maximum number of memory protection table entries per HCA");
117 
118 module_param_named(num_mtt, hca_profile.num_mtt, int, 0444);
119 MODULE_PARM_DESC(num_mtt,
120 		 "maximum number of memory translation table segments per HCA");
121 
122 module_param_named(num_udav, hca_profile.num_udav, int, 0444);
123 MODULE_PARM_DESC(num_udav, "maximum number of UD address vectors per HCA");
124 
125 module_param_named(fmr_reserved_mtts, hca_profile.fmr_reserved_mtts, int, 0444);
126 MODULE_PARM_DESC(fmr_reserved_mtts,
127 		 "number of memory translation table segments reserved for FMR");
128 
129 static int log_mtts_per_seg = ilog2(MTHCA_MTT_SEG_SIZE / 8);
130 module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
131 MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-5)");
132 
133 static char mthca_version[] __devinitdata =
134 	DRV_NAME ": Mellanox InfiniBand HCA driver v"
135 	DRV_VERSION " (" DRV_RELDATE ")\n";
136 
137 static int mthca_tune_pci(struct mthca_dev *mdev)
138 {
139 	if (!tune_pci)
140 		return 0;
141 
142 	/* First try to max out Read Byte Count */
143 	if (pci_find_capability(mdev->pdev, PCI_CAP_ID_PCIX)) {
144 		if (pcix_set_mmrbc(mdev->pdev, pcix_get_max_mmrbc(mdev->pdev))) {
145 			mthca_err(mdev, "Couldn't set PCI-X max read count, "
146 				"aborting.\n");
147 			return -ENODEV;
148 		}
149 	} else if (!(mdev->mthca_flags & MTHCA_FLAG_PCIE))
150 		mthca_info(mdev, "No PCI-X capability, not setting RBC.\n");
151 
152 	if (pci_is_pcie(mdev->pdev)) {
153 		if (pcie_set_readrq(mdev->pdev, 4096)) {
154 			mthca_err(mdev, "Couldn't write PCI Express read request, "
155 				"aborting.\n");
156 			return -ENODEV;
157 		}
158 	} else if (mdev->mthca_flags & MTHCA_FLAG_PCIE)
159 		mthca_info(mdev, "No PCI Express capability, "
160 			   "not setting Max Read Request Size.\n");
161 
162 	return 0;
163 }
164 
165 static int mthca_dev_lim(struct mthca_dev *mdev, struct mthca_dev_lim *dev_lim)
166 {
167 	int err;
168 
169 	mdev->limits.mtt_seg_size = (1 << log_mtts_per_seg) * 8;
170 	err = mthca_QUERY_DEV_LIM(mdev, dev_lim);
171 	if (err) {
172 		mthca_err(mdev, "QUERY_DEV_LIM command returned %d"
173 				", aborting.\n", err);
174 		return err;
175 	}
176 	if (dev_lim->min_page_sz > PAGE_SIZE) {
177 		mthca_err(mdev, "HCA minimum page size of %d bigger than "
178 			  "kernel PAGE_SIZE of %ld, aborting.\n",
179 			  dev_lim->min_page_sz, PAGE_SIZE);
180 		return -ENODEV;
181 	}
182 	if (dev_lim->num_ports > MTHCA_MAX_PORTS) {
183 		mthca_err(mdev, "HCA has %d ports, but we only support %d, "
184 			  "aborting.\n",
185 			  dev_lim->num_ports, MTHCA_MAX_PORTS);
186 		return -ENODEV;
187 	}
188 
189 	if (dev_lim->uar_size > pci_resource_len(mdev->pdev, 2)) {
190 		mthca_err(mdev, "HCA reported UAR size of 0x%x bigger than "
191 			  "PCI resource 2 size of 0x%llx, aborting.\n",
192 			  dev_lim->uar_size,
193 			  (unsigned long long)pci_resource_len(mdev->pdev, 2));
194 		return -ENODEV;
195 	}
196 
197 	mdev->limits.num_ports      	= dev_lim->num_ports;
198 	mdev->limits.vl_cap             = dev_lim->max_vl;
199 	mdev->limits.mtu_cap            = dev_lim->max_mtu;
200 	mdev->limits.gid_table_len  	= dev_lim->max_gids;
201 	mdev->limits.pkey_table_len 	= dev_lim->max_pkeys;
202 	mdev->limits.local_ca_ack_delay = dev_lim->local_ca_ack_delay;
203 	/*
204 	 * Need to allow for worst case send WQE overhead and check
205 	 * whether max_desc_sz imposes a lower limit than max_sg; UD
206 	 * send has the biggest overhead.
207 	 */
208 	mdev->limits.max_sg		= min_t(int, dev_lim->max_sg,
209 					      (dev_lim->max_desc_sz -
210 					       sizeof (struct mthca_next_seg) -
211 					       (mthca_is_memfree(mdev) ?
212 						sizeof (struct mthca_arbel_ud_seg) :
213 						sizeof (struct mthca_tavor_ud_seg))) /
214 						sizeof (struct mthca_data_seg));
215 	mdev->limits.max_wqes           = dev_lim->max_qp_sz;
216 	mdev->limits.max_qp_init_rdma   = dev_lim->max_requester_per_qp;
217 	mdev->limits.reserved_qps       = dev_lim->reserved_qps;
218 	mdev->limits.max_srq_wqes       = dev_lim->max_srq_sz;
219 	mdev->limits.reserved_srqs      = dev_lim->reserved_srqs;
220 	mdev->limits.reserved_eecs      = dev_lim->reserved_eecs;
221 	mdev->limits.max_desc_sz        = dev_lim->max_desc_sz;
222 	mdev->limits.max_srq_sge	= mthca_max_srq_sge(mdev);
223 	/*
224 	 * Subtract 1 from the limit because we need to allocate a
225 	 * spare CQE so the HCA HW can tell the difference between an
226 	 * empty CQ and a full CQ.
227 	 */
228 	mdev->limits.max_cqes           = dev_lim->max_cq_sz - 1;
229 	mdev->limits.reserved_cqs       = dev_lim->reserved_cqs;
230 	mdev->limits.reserved_eqs       = dev_lim->reserved_eqs;
231 	mdev->limits.reserved_mtts      = dev_lim->reserved_mtts;
232 	mdev->limits.reserved_mrws      = dev_lim->reserved_mrws;
233 	mdev->limits.reserved_uars      = dev_lim->reserved_uars;
234 	mdev->limits.reserved_pds       = dev_lim->reserved_pds;
235 	mdev->limits.port_width_cap     = dev_lim->max_port_width;
236 	mdev->limits.page_size_cap      = ~(u32) (dev_lim->min_page_sz - 1);
237 	mdev->limits.flags              = dev_lim->flags;
238 	/*
239 	 * For old FW that doesn't return static rate support, use a
240 	 * value of 0x3 (only static rate values of 0 or 1 are handled),
241 	 * except on Sinai, where even old FW can handle static rate
242 	 * values of 2 and 3.
243 	 */
244 	if (dev_lim->stat_rate_support)
245 		mdev->limits.stat_rate_support = dev_lim->stat_rate_support;
246 	else if (mdev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
247 		mdev->limits.stat_rate_support = 0xf;
248 	else
249 		mdev->limits.stat_rate_support = 0x3;
250 
251 	/* IB_DEVICE_RESIZE_MAX_WR not supported by driver.
252 	   May be doable since hardware supports it for SRQ.
253 
254 	   IB_DEVICE_N_NOTIFY_CQ is supported by hardware but not by driver.
255 
256 	   IB_DEVICE_SRQ_RESIZE is supported by hardware but SRQ is not
257 	   supported by driver. */
258 	mdev->device_cap_flags = IB_DEVICE_CHANGE_PHY_PORT |
259 		IB_DEVICE_PORT_ACTIVE_EVENT |
260 		IB_DEVICE_SYS_IMAGE_GUID |
261 		IB_DEVICE_RC_RNR_NAK_GEN;
262 
263 	if (dev_lim->flags & DEV_LIM_FLAG_BAD_PKEY_CNTR)
264 		mdev->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
265 
266 	if (dev_lim->flags & DEV_LIM_FLAG_BAD_QKEY_CNTR)
267 		mdev->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
268 
269 	if (dev_lim->flags & DEV_LIM_FLAG_RAW_MULTI)
270 		mdev->device_cap_flags |= IB_DEVICE_RAW_MULTI;
271 
272 	if (dev_lim->flags & DEV_LIM_FLAG_AUTO_PATH_MIG)
273 		mdev->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
274 
275 	if (dev_lim->flags & DEV_LIM_FLAG_UD_AV_PORT_ENFORCE)
276 		mdev->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
277 
278 	if (dev_lim->flags & DEV_LIM_FLAG_SRQ)
279 		mdev->mthca_flags |= MTHCA_FLAG_SRQ;
280 
281 	if (mthca_is_memfree(mdev))
282 		if (dev_lim->flags & DEV_LIM_FLAG_IPOIB_CSUM)
283 			mdev->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
284 
285 	return 0;
286 }
287 
288 static int mthca_init_tavor(struct mthca_dev *mdev)
289 {
290 	s64 size;
291 	int err;
292 	struct mthca_dev_lim        dev_lim;
293 	struct mthca_profile        profile;
294 	struct mthca_init_hca_param init_hca;
295 
296 	err = mthca_SYS_EN(mdev);
297 	if (err) {
298 		mthca_err(mdev, "SYS_EN command returned %d, aborting.\n", err);
299 		return err;
300 	}
301 
302 	err = mthca_QUERY_FW(mdev);
303 	if (err) {
304 		mthca_err(mdev, "QUERY_FW command returned %d,"
305 				" aborting.\n", err);
306 		goto err_disable;
307 	}
308 	err = mthca_QUERY_DDR(mdev);
309 	if (err) {
310 		mthca_err(mdev, "QUERY_DDR command returned %d, aborting.\n", err);
311 		goto err_disable;
312 	}
313 
314 	err = mthca_dev_lim(mdev, &dev_lim);
315 	if (err) {
316 		mthca_err(mdev, "QUERY_DEV_LIM command returned %d, aborting.\n", err);
317 		goto err_disable;
318 	}
319 
320 	profile = hca_profile;
321 	profile.num_uar   = dev_lim.uar_size / PAGE_SIZE;
322 	profile.uarc_size = 0;
323 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
324 		profile.num_srq = dev_lim.max_srqs;
325 
326 	size = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
327 	if (size < 0) {
328 		err = size;
329 		goto err_disable;
330 	}
331 
332 	err = mthca_INIT_HCA(mdev, &init_hca);
333 	if (err) {
334 		mthca_err(mdev, "INIT_HCA command returned %d, aborting.\n", err);
335 		goto err_disable;
336 	}
337 
338 	return 0;
339 
340 err_disable:
341 	mthca_SYS_DIS(mdev);
342 
343 	return err;
344 }
345 
346 static int mthca_load_fw(struct mthca_dev *mdev)
347 {
348 	int err;
349 
350 	/* FIXME: use HCA-attached memory for FW if present */
351 
352 	mdev->fw.arbel.fw_icm =
353 		mthca_alloc_icm(mdev, mdev->fw.arbel.fw_pages,
354 				GFP_HIGHUSER | __GFP_NOWARN, 0);
355 	if (!mdev->fw.arbel.fw_icm) {
356 		mthca_err(mdev, "Couldn't allocate FW area, aborting.\n");
357 		return -ENOMEM;
358 	}
359 
360 	err = mthca_MAP_FA(mdev, mdev->fw.arbel.fw_icm);
361 	if (err) {
362 		mthca_err(mdev, "MAP_FA command returned %d, aborting.\n", err);
363 		goto err_free;
364 	}
365 	err = mthca_RUN_FW(mdev);
366 	if (err) {
367 		mthca_err(mdev, "RUN_FW command returned %d, aborting.\n", err);
368 		goto err_unmap_fa;
369 	}
370 
371 	return 0;
372 
373 err_unmap_fa:
374 	mthca_UNMAP_FA(mdev);
375 
376 err_free:
377 	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
378 	return err;
379 }
380 
381 static int mthca_init_icm(struct mthca_dev *mdev,
382 			  struct mthca_dev_lim *dev_lim,
383 			  struct mthca_init_hca_param *init_hca,
384 			  u64 icm_size)
385 {
386 	u64 aux_pages;
387 	int err;
388 
389 	err = mthca_SET_ICM_SIZE(mdev, icm_size, &aux_pages);
390 	if (err) {
391 		mthca_err(mdev, "SET_ICM_SIZE command returned %d, aborting.\n", err);
392 		return err;
393 	}
394 
395 	mthca_dbg(mdev, "%lld KB of HCA context requires %lld KB aux memory.\n",
396 		  (unsigned long long) icm_size >> 10,
397 		  (unsigned long long) aux_pages << 2);
398 
399 	mdev->fw.arbel.aux_icm = mthca_alloc_icm(mdev, aux_pages,
400 						 GFP_HIGHUSER | __GFP_NOWARN, 0);
401 	if (!mdev->fw.arbel.aux_icm) {
402 		mthca_err(mdev, "Couldn't allocate aux memory, aborting.\n");
403 		return -ENOMEM;
404 	}
405 
406 	err = mthca_MAP_ICM_AUX(mdev, mdev->fw.arbel.aux_icm);
407 	if (err) {
408 		mthca_err(mdev, "MAP_ICM_AUX returned %d, aborting.\n", err);
409 		goto err_free_aux;
410 	}
411 
412 	err = mthca_map_eq_icm(mdev, init_hca->eqc_base);
413 	if (err) {
414 		mthca_err(mdev, "Failed to map EQ context memory, aborting.\n");
415 		goto err_unmap_aux;
416 	}
417 
418 	/* CPU writes to non-reserved MTTs, while HCA might DMA to reserved mtts */
419 	mdev->limits.reserved_mtts = ALIGN(mdev->limits.reserved_mtts * mdev->limits.mtt_seg_size,
420 					   dma_get_cache_alignment()) / mdev->limits.mtt_seg_size;
421 
422 	mdev->mr_table.mtt_table = mthca_alloc_icm_table(mdev, init_hca->mtt_base,
423 							 mdev->limits.mtt_seg_size,
424 							 mdev->limits.num_mtt_segs,
425 							 mdev->limits.reserved_mtts,
426 							 1, 0);
427 	if (!mdev->mr_table.mtt_table) {
428 		mthca_err(mdev, "Failed to map MTT context memory, aborting.\n");
429 		err = -ENOMEM;
430 		goto err_unmap_eq;
431 	}
432 
433 	mdev->mr_table.mpt_table = mthca_alloc_icm_table(mdev, init_hca->mpt_base,
434 							 dev_lim->mpt_entry_sz,
435 							 mdev->limits.num_mpts,
436 							 mdev->limits.reserved_mrws,
437 							 1, 1);
438 	if (!mdev->mr_table.mpt_table) {
439 		mthca_err(mdev, "Failed to map MPT context memory, aborting.\n");
440 		err = -ENOMEM;
441 		goto err_unmap_mtt;
442 	}
443 
444 	mdev->qp_table.qp_table = mthca_alloc_icm_table(mdev, init_hca->qpc_base,
445 							dev_lim->qpc_entry_sz,
446 							mdev->limits.num_qps,
447 							mdev->limits.reserved_qps,
448 							0, 0);
449 	if (!mdev->qp_table.qp_table) {
450 		mthca_err(mdev, "Failed to map QP context memory, aborting.\n");
451 		err = -ENOMEM;
452 		goto err_unmap_mpt;
453 	}
454 
455 	mdev->qp_table.eqp_table = mthca_alloc_icm_table(mdev, init_hca->eqpc_base,
456 							 dev_lim->eqpc_entry_sz,
457 							 mdev->limits.num_qps,
458 							 mdev->limits.reserved_qps,
459 							 0, 0);
460 	if (!mdev->qp_table.eqp_table) {
461 		mthca_err(mdev, "Failed to map EQP context memory, aborting.\n");
462 		err = -ENOMEM;
463 		goto err_unmap_qp;
464 	}
465 
466 	mdev->qp_table.rdb_table = mthca_alloc_icm_table(mdev, init_hca->rdb_base,
467 							 MTHCA_RDB_ENTRY_SIZE,
468 							 mdev->limits.num_qps <<
469 							 mdev->qp_table.rdb_shift, 0,
470 							 0, 0);
471 	if (!mdev->qp_table.rdb_table) {
472 		mthca_err(mdev, "Failed to map RDB context memory, aborting\n");
473 		err = -ENOMEM;
474 		goto err_unmap_eqp;
475 	}
476 
477        mdev->cq_table.table = mthca_alloc_icm_table(mdev, init_hca->cqc_base,
478 						    dev_lim->cqc_entry_sz,
479 						    mdev->limits.num_cqs,
480 						    mdev->limits.reserved_cqs,
481 						    0, 0);
482 	if (!mdev->cq_table.table) {
483 		mthca_err(mdev, "Failed to map CQ context memory, aborting.\n");
484 		err = -ENOMEM;
485 		goto err_unmap_rdb;
486 	}
487 
488 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ) {
489 		mdev->srq_table.table =
490 			mthca_alloc_icm_table(mdev, init_hca->srqc_base,
491 					      dev_lim->srq_entry_sz,
492 					      mdev->limits.num_srqs,
493 					      mdev->limits.reserved_srqs,
494 					      0, 0);
495 		if (!mdev->srq_table.table) {
496 			mthca_err(mdev, "Failed to map SRQ context memory, "
497 				  "aborting.\n");
498 			err = -ENOMEM;
499 			goto err_unmap_cq;
500 		}
501 	}
502 
503 	/*
504 	 * It's not strictly required, but for simplicity just map the
505 	 * whole multicast group table now.  The table isn't very big
506 	 * and it's a lot easier than trying to track ref counts.
507 	 */
508 	mdev->mcg_table.table = mthca_alloc_icm_table(mdev, init_hca->mc_base,
509 						      MTHCA_MGM_ENTRY_SIZE,
510 						      mdev->limits.num_mgms +
511 						      mdev->limits.num_amgms,
512 						      mdev->limits.num_mgms +
513 						      mdev->limits.num_amgms,
514 						      0, 0);
515 	if (!mdev->mcg_table.table) {
516 		mthca_err(mdev, "Failed to map MCG context memory, aborting.\n");
517 		err = -ENOMEM;
518 		goto err_unmap_srq;
519 	}
520 
521 	return 0;
522 
523 err_unmap_srq:
524 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
525 		mthca_free_icm_table(mdev, mdev->srq_table.table);
526 
527 err_unmap_cq:
528 	mthca_free_icm_table(mdev, mdev->cq_table.table);
529 
530 err_unmap_rdb:
531 	mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
532 
533 err_unmap_eqp:
534 	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
535 
536 err_unmap_qp:
537 	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
538 
539 err_unmap_mpt:
540 	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
541 
542 err_unmap_mtt:
543 	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
544 
545 err_unmap_eq:
546 	mthca_unmap_eq_icm(mdev);
547 
548 err_unmap_aux:
549 	mthca_UNMAP_ICM_AUX(mdev);
550 
551 err_free_aux:
552 	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm, 0);
553 
554 	return err;
555 }
556 
557 static void mthca_free_icms(struct mthca_dev *mdev)
558 {
559 
560 	mthca_free_icm_table(mdev, mdev->mcg_table.table);
561 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
562 		mthca_free_icm_table(mdev, mdev->srq_table.table);
563 	mthca_free_icm_table(mdev, mdev->cq_table.table);
564 	mthca_free_icm_table(mdev, mdev->qp_table.rdb_table);
565 	mthca_free_icm_table(mdev, mdev->qp_table.eqp_table);
566 	mthca_free_icm_table(mdev, mdev->qp_table.qp_table);
567 	mthca_free_icm_table(mdev, mdev->mr_table.mpt_table);
568 	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
569 	mthca_unmap_eq_icm(mdev);
570 
571 	mthca_UNMAP_ICM_AUX(mdev);
572 	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm, 0);
573 }
574 
575 static int mthca_init_arbel(struct mthca_dev *mdev)
576 {
577 	struct mthca_dev_lim        dev_lim;
578 	struct mthca_profile        profile;
579 	struct mthca_init_hca_param init_hca;
580 	s64 icm_size;
581 	int err;
582 
583 	err = mthca_QUERY_FW(mdev);
584 	if (err) {
585 		mthca_err(mdev, "QUERY_FW command failed %d, aborting.\n", err);
586 		return err;
587 	}
588 
589 	err = mthca_ENABLE_LAM(mdev);
590 	if (err == -EAGAIN) {
591 		mthca_dbg(mdev, "No HCA-attached memory (running in MemFree mode)\n");
592 		mdev->mthca_flags |= MTHCA_FLAG_NO_LAM;
593 	} else if (err) {
594 		mthca_err(mdev, "ENABLE_LAM returned %d, aborting.\n", err);
595 		return err;
596 	}
597 
598 	err = mthca_load_fw(mdev);
599 	if (err) {
600 		mthca_err(mdev, "Loading FW returned %d, aborting.\n", err);
601 		goto err_disable;
602 	}
603 
604 	err = mthca_dev_lim(mdev, &dev_lim);
605 	if (err) {
606 		mthca_err(mdev, "QUERY_DEV_LIM returned %d, aborting.\n", err);
607 		goto err_stop_fw;
608 	}
609 
610 	profile = hca_profile;
611 	profile.num_uar  = dev_lim.uar_size / PAGE_SIZE;
612 	profile.num_udav = 0;
613 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
614 		profile.num_srq = dev_lim.max_srqs;
615 
616 	icm_size = mthca_make_profile(mdev, &profile, &dev_lim, &init_hca);
617 	if (icm_size < 0) {
618 		err = icm_size;
619 		goto err_stop_fw;
620 	}
621 
622 	err = mthca_init_icm(mdev, &dev_lim, &init_hca, icm_size);
623 	if (err)
624 		goto err_stop_fw;
625 
626 	err = mthca_INIT_HCA(mdev, &init_hca);
627 	if (err) {
628 		mthca_err(mdev, "INIT_HCA command returned %d, aborting.\n", err);
629 		goto err_free_icm;
630 	}
631 
632 	return 0;
633 
634 err_free_icm:
635 	mthca_free_icms(mdev);
636 
637 err_stop_fw:
638 	mthca_UNMAP_FA(mdev);
639 	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
640 
641 err_disable:
642 	if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
643 		mthca_DISABLE_LAM(mdev);
644 
645 	return err;
646 }
647 
648 static void mthca_close_hca(struct mthca_dev *mdev)
649 {
650 	mthca_CLOSE_HCA(mdev, 0);
651 
652 	if (mthca_is_memfree(mdev)) {
653 		mthca_free_icms(mdev);
654 
655 		mthca_UNMAP_FA(mdev);
656 		mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
657 
658 		if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
659 			mthca_DISABLE_LAM(mdev);
660 	} else
661 		mthca_SYS_DIS(mdev);
662 }
663 
664 static int mthca_init_hca(struct mthca_dev *mdev)
665 {
666 	int err;
667 	struct mthca_adapter adapter;
668 
669 	if (mthca_is_memfree(mdev))
670 		err = mthca_init_arbel(mdev);
671 	else
672 		err = mthca_init_tavor(mdev);
673 
674 	if (err)
675 		return err;
676 
677 	err = mthca_QUERY_ADAPTER(mdev, &adapter);
678 	if (err) {
679 		mthca_err(mdev, "QUERY_ADAPTER command returned %d, aborting.\n", err);
680 		goto err_close;
681 	}
682 
683 	mdev->eq_table.inta_pin = adapter.inta_pin;
684 	if (!mthca_is_memfree(mdev))
685 		mdev->rev_id = adapter.revision_id;
686 	memcpy(mdev->board_id, adapter.board_id, sizeof mdev->board_id);
687 
688 	return 0;
689 
690 err_close:
691 	mthca_close_hca(mdev);
692 	return err;
693 }
694 
695 static int mthca_setup_hca(struct mthca_dev *dev)
696 {
697 	int err;
698 
699 	MTHCA_INIT_DOORBELL_LOCK(&dev->doorbell_lock);
700 
701 	err = mthca_init_uar_table(dev);
702 	if (err) {
703 		mthca_err(dev, "Failed to initialize "
704 			  "user access region table, aborting.\n");
705 		return err;
706 	}
707 
708 	err = mthca_uar_alloc(dev, &dev->driver_uar);
709 	if (err) {
710 		mthca_err(dev, "Failed to allocate driver access region, "
711 			  "aborting.\n");
712 		goto err_uar_table_free;
713 	}
714 
715 	dev->kar = ioremap((phys_addr_t) dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
716 	if (!dev->kar) {
717 		mthca_err(dev, "Couldn't map kernel access region, "
718 			  "aborting.\n");
719 		err = -ENOMEM;
720 		goto err_uar_free;
721 	}
722 
723 	err = mthca_init_pd_table(dev);
724 	if (err) {
725 		mthca_err(dev, "Failed to initialize "
726 			  "protection domain table, aborting.\n");
727 		goto err_kar_unmap;
728 	}
729 
730 	err = mthca_init_mr_table(dev);
731 	if (err) {
732 		mthca_err(dev, "Failed to initialize "
733 			  "memory region table, aborting.\n");
734 		goto err_pd_table_free;
735 	}
736 
737 	err = mthca_pd_alloc(dev, 1, &dev->driver_pd);
738 	if (err) {
739 		mthca_err(dev, "Failed to create driver PD, "
740 			  "aborting.\n");
741 		goto err_mr_table_free;
742 	}
743 
744 	err = mthca_init_eq_table(dev);
745 	if (err) {
746 		mthca_err(dev, "Failed to initialize "
747 			  "event queue table, aborting.\n");
748 		goto err_pd_free;
749 	}
750 
751 	err = mthca_cmd_use_events(dev);
752 	if (err) {
753 		mthca_err(dev, "Failed to switch to event-driven "
754 			  "firmware commands, aborting.\n");
755 		goto err_eq_table_free;
756 	}
757 
758 	err = mthca_NOP(dev);
759 	if (err) {
760 		if (dev->mthca_flags & MTHCA_FLAG_MSI_X) {
761 			mthca_warn(dev, "NOP command failed to generate interrupt "
762 				   "(IRQ %d).\n",
763 				   dev->eq_table.eq[MTHCA_EQ_CMD].msi_x_vector);
764 			mthca_warn(dev, "Trying again with MSI-X disabled.\n");
765 		} else {
766 			mthca_err(dev, "NOP command failed to generate interrupt "
767 				  "(IRQ %d), aborting.\n",
768 				  dev->pdev->irq);
769 			mthca_err(dev, "BIOS or ACPI interrupt routing problem?\n");
770 		}
771 
772 		goto err_cmd_poll;
773 	}
774 
775 	mthca_dbg(dev, "NOP command IRQ test passed\n");
776 
777 	err = mthca_init_cq_table(dev);
778 	if (err) {
779 		mthca_err(dev, "Failed to initialize "
780 			  "completion queue table, aborting.\n");
781 		goto err_cmd_poll;
782 	}
783 
784 	err = mthca_init_srq_table(dev);
785 	if (err) {
786 		mthca_err(dev, "Failed to initialize "
787 			  "shared receive queue table, aborting.\n");
788 		goto err_cq_table_free;
789 	}
790 
791 	err = mthca_init_qp_table(dev);
792 	if (err) {
793 		mthca_err(dev, "Failed to initialize "
794 			  "queue pair table, aborting.\n");
795 		goto err_srq_table_free;
796 	}
797 
798 	err = mthca_init_av_table(dev);
799 	if (err) {
800 		mthca_err(dev, "Failed to initialize "
801 			  "address vector table, aborting.\n");
802 		goto err_qp_table_free;
803 	}
804 
805 	err = mthca_init_mcg_table(dev);
806 	if (err) {
807 		mthca_err(dev, "Failed to initialize "
808 			  "multicast group table, aborting.\n");
809 		goto err_av_table_free;
810 	}
811 
812 	return 0;
813 
814 err_av_table_free:
815 	mthca_cleanup_av_table(dev);
816 
817 err_qp_table_free:
818 	mthca_cleanup_qp_table(dev);
819 
820 err_srq_table_free:
821 	mthca_cleanup_srq_table(dev);
822 
823 err_cq_table_free:
824 	mthca_cleanup_cq_table(dev);
825 
826 err_cmd_poll:
827 	mthca_cmd_use_polling(dev);
828 
829 err_eq_table_free:
830 	mthca_cleanup_eq_table(dev);
831 
832 err_pd_free:
833 	mthca_pd_free(dev, &dev->driver_pd);
834 
835 err_mr_table_free:
836 	mthca_cleanup_mr_table(dev);
837 
838 err_pd_table_free:
839 	mthca_cleanup_pd_table(dev);
840 
841 err_kar_unmap:
842 	iounmap(dev->kar);
843 
844 err_uar_free:
845 	mthca_uar_free(dev, &dev->driver_uar);
846 
847 err_uar_table_free:
848 	mthca_cleanup_uar_table(dev);
849 	return err;
850 }
851 
852 static int mthca_enable_msi_x(struct mthca_dev *mdev)
853 {
854 	struct msix_entry entries[3];
855 	int err;
856 
857 	entries[0].entry = 0;
858 	entries[1].entry = 1;
859 	entries[2].entry = 2;
860 
861 	err = pci_enable_msix(mdev->pdev, entries, ARRAY_SIZE(entries));
862 	if (err) {
863 		if (err > 0)
864 			mthca_info(mdev, "Only %d MSI-X vectors available, "
865 				   "not using MSI-X\n", err);
866 		return err;
867 	}
868 
869 	mdev->eq_table.eq[MTHCA_EQ_COMP ].msi_x_vector = entries[0].vector;
870 	mdev->eq_table.eq[MTHCA_EQ_ASYNC].msi_x_vector = entries[1].vector;
871 	mdev->eq_table.eq[MTHCA_EQ_CMD  ].msi_x_vector = entries[2].vector;
872 
873 	return 0;
874 }
875 
876 /* Types of supported HCA */
877 enum {
878 	TAVOR,			/* MT23108                        */
879 	ARBEL_COMPAT,		/* MT25208 in Tavor compat mode   */
880 	ARBEL_NATIVE,		/* MT25208 with extended features */
881 	SINAI			/* MT25204 */
882 };
883 
884 #define MTHCA_FW_VER(major, minor, subminor) \
885 	(((u64) (major) << 32) | ((u64) (minor) << 16) | (u64) (subminor))
886 
887 static struct {
888 	u64 latest_fw;
889 	u32 flags;
890 } mthca_hca_table[] = {
891 	[TAVOR]        = { .latest_fw = MTHCA_FW_VER(3, 5, 0),
892 			   .flags     = 0 },
893 	[ARBEL_COMPAT] = { .latest_fw = MTHCA_FW_VER(4, 8, 200),
894 			   .flags     = MTHCA_FLAG_PCIE },
895 	[ARBEL_NATIVE] = { .latest_fw = MTHCA_FW_VER(5, 3, 0),
896 			   .flags     = MTHCA_FLAG_MEMFREE |
897 					MTHCA_FLAG_PCIE },
898 	[SINAI]        = { .latest_fw = MTHCA_FW_VER(1, 2, 0),
899 			   .flags     = MTHCA_FLAG_MEMFREE |
900 					MTHCA_FLAG_PCIE    |
901 					MTHCA_FLAG_SINAI_OPT }
902 };
903 
904 static int __mthca_init_one(struct pci_dev *pdev, int hca_type)
905 {
906 	int ddr_hidden = 0;
907 	int err;
908 	struct mthca_dev *mdev;
909 
910 	printk(KERN_INFO PFX "Initializing %s\n",
911 	       pci_name(pdev));
912 
913 	err = pci_enable_device(pdev);
914 	if (err) {
915 		dev_err(&pdev->dev, "Cannot enable PCI device, "
916 			"aborting.\n");
917 		return err;
918 	}
919 
920 	/*
921 	 * Check for BARs.  We expect 0: 1MB, 2: 8MB, 4: DDR (may not
922 	 * be present)
923 	 */
924 	if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
925 	    pci_resource_len(pdev, 0) != 1 << 20) {
926 		dev_err(&pdev->dev, "Missing DCS, aborting.\n");
927 		err = -ENODEV;
928 		goto err_disable_pdev;
929 	}
930 	if (!(pci_resource_flags(pdev, 2) & IORESOURCE_MEM)) {
931 		dev_err(&pdev->dev, "Missing UAR, aborting.\n");
932 		err = -ENODEV;
933 		goto err_disable_pdev;
934 	}
935 	if (!(pci_resource_flags(pdev, 4) & IORESOURCE_MEM))
936 		ddr_hidden = 1;
937 
938 	err = pci_request_regions(pdev, DRV_NAME);
939 	if (err) {
940 		dev_err(&pdev->dev, "Cannot obtain PCI resources, "
941 			"aborting.\n");
942 		goto err_disable_pdev;
943 	}
944 
945 	pci_set_master(pdev);
946 
947 	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
948 	if (err) {
949 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
950 		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
951 		if (err) {
952 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
953 			goto err_free_res;
954 		}
955 	}
956 	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
957 	if (err) {
958 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
959 			 "consistent PCI DMA mask.\n");
960 		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
961 		if (err) {
962 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
963 				"aborting.\n");
964 			goto err_free_res;
965 		}
966 	}
967 
968 	/* We can handle large RDMA requests, so allow larger segments. */
969 	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
970 
971 	mdev = (struct mthca_dev *) ib_alloc_device(sizeof *mdev);
972 	if (!mdev) {
973 		dev_err(&pdev->dev, "Device struct alloc failed, "
974 			"aborting.\n");
975 		err = -ENOMEM;
976 		goto err_free_res;
977 	}
978 
979 	mdev->pdev = pdev;
980 
981 	mdev->mthca_flags = mthca_hca_table[hca_type].flags;
982 	if (ddr_hidden)
983 		mdev->mthca_flags |= MTHCA_FLAG_DDR_HIDDEN;
984 
985 	/*
986 	 * Now reset the HCA before we touch the PCI capabilities or
987 	 * attempt a firmware command, since a boot ROM may have left
988 	 * the HCA in an undefined state.
989 	 */
990 	err = mthca_reset(mdev);
991 	if (err) {
992 		mthca_err(mdev, "Failed to reset HCA, aborting.\n");
993 		goto err_free_dev;
994 	}
995 
996 	if (mthca_cmd_init(mdev)) {
997 		mthca_err(mdev, "Failed to init command interface, aborting.\n");
998 		goto err_free_dev;
999 	}
1000 
1001 	err = mthca_tune_pci(mdev);
1002 	if (err)
1003 		goto err_cmd;
1004 
1005 	err = mthca_init_hca(mdev);
1006 	if (err)
1007 		goto err_cmd;
1008 
1009 	if (mdev->fw_ver < mthca_hca_table[hca_type].latest_fw) {
1010 		mthca_warn(mdev, "HCA FW version %d.%d.%03d is old (%d.%d.%03d is current).\n",
1011 			   (int) (mdev->fw_ver >> 32), (int) (mdev->fw_ver >> 16) & 0xffff,
1012 			   (int) (mdev->fw_ver & 0xffff),
1013 			   (int) (mthca_hca_table[hca_type].latest_fw >> 32),
1014 			   (int) (mthca_hca_table[hca_type].latest_fw >> 16) & 0xffff,
1015 			   (int) (mthca_hca_table[hca_type].latest_fw & 0xffff));
1016 		mthca_warn(mdev, "If you have problems, try updating your HCA FW.\n");
1017 	}
1018 
1019 	if (msi_x && !mthca_enable_msi_x(mdev))
1020 		mdev->mthca_flags |= MTHCA_FLAG_MSI_X;
1021 
1022 	err = mthca_setup_hca(mdev);
1023 	if (err == -EBUSY && (mdev->mthca_flags & MTHCA_FLAG_MSI_X)) {
1024 		if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1025 			pci_disable_msix(pdev);
1026 		mdev->mthca_flags &= ~MTHCA_FLAG_MSI_X;
1027 
1028 		err = mthca_setup_hca(mdev);
1029 	}
1030 
1031 	if (err)
1032 		goto err_close;
1033 
1034 	err = mthca_register_device(mdev);
1035 	if (err)
1036 		goto err_cleanup;
1037 
1038 	err = mthca_create_agents(mdev);
1039 	if (err)
1040 		goto err_unregister;
1041 
1042 	pci_set_drvdata(pdev, mdev);
1043 	mdev->hca_type = hca_type;
1044 
1045 	mdev->active = true;
1046 
1047 	return 0;
1048 
1049 err_unregister:
1050 	mthca_unregister_device(mdev);
1051 
1052 err_cleanup:
1053 	mthca_cleanup_mcg_table(mdev);
1054 	mthca_cleanup_av_table(mdev);
1055 	mthca_cleanup_qp_table(mdev);
1056 	mthca_cleanup_srq_table(mdev);
1057 	mthca_cleanup_cq_table(mdev);
1058 	mthca_cmd_use_polling(mdev);
1059 	mthca_cleanup_eq_table(mdev);
1060 
1061 	mthca_pd_free(mdev, &mdev->driver_pd);
1062 
1063 	mthca_cleanup_mr_table(mdev);
1064 	mthca_cleanup_pd_table(mdev);
1065 	mthca_cleanup_uar_table(mdev);
1066 
1067 err_close:
1068 	if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1069 		pci_disable_msix(pdev);
1070 
1071 	mthca_close_hca(mdev);
1072 
1073 err_cmd:
1074 	mthca_cmd_cleanup(mdev);
1075 
1076 err_free_dev:
1077 	ib_dealloc_device(&mdev->ib_dev);
1078 
1079 err_free_res:
1080 	pci_release_regions(pdev);
1081 
1082 err_disable_pdev:
1083 	pci_disable_device(pdev);
1084 	pci_set_drvdata(pdev, NULL);
1085 	return err;
1086 }
1087 
1088 static void __mthca_remove_one(struct pci_dev *pdev)
1089 {
1090 	struct mthca_dev *mdev = pci_get_drvdata(pdev);
1091 	int p;
1092 
1093 	if (mdev) {
1094 		mthca_free_agents(mdev);
1095 		mthca_unregister_device(mdev);
1096 
1097 		for (p = 1; p <= mdev->limits.num_ports; ++p)
1098 			mthca_CLOSE_IB(mdev, p);
1099 
1100 		mthca_cleanup_mcg_table(mdev);
1101 		mthca_cleanup_av_table(mdev);
1102 		mthca_cleanup_qp_table(mdev);
1103 		mthca_cleanup_srq_table(mdev);
1104 		mthca_cleanup_cq_table(mdev);
1105 		mthca_cmd_use_polling(mdev);
1106 		mthca_cleanup_eq_table(mdev);
1107 
1108 		mthca_pd_free(mdev, &mdev->driver_pd);
1109 
1110 		mthca_cleanup_mr_table(mdev);
1111 		mthca_cleanup_pd_table(mdev);
1112 
1113 		iounmap(mdev->kar);
1114 		mthca_uar_free(mdev, &mdev->driver_uar);
1115 		mthca_cleanup_uar_table(mdev);
1116 		mthca_close_hca(mdev);
1117 		mthca_cmd_cleanup(mdev);
1118 
1119 		if (mdev->mthca_flags & MTHCA_FLAG_MSI_X)
1120 			pci_disable_msix(pdev);
1121 
1122 		ib_dealloc_device(&mdev->ib_dev);
1123 		pci_release_regions(pdev);
1124 		pci_disable_device(pdev);
1125 		pci_set_drvdata(pdev, NULL);
1126 	}
1127 }
1128 
1129 int __mthca_restart_one(struct pci_dev *pdev)
1130 {
1131 	struct mthca_dev *mdev;
1132 	int hca_type;
1133 
1134 	mdev = pci_get_drvdata(pdev);
1135 	if (!mdev)
1136 		return -ENODEV;
1137 	hca_type = mdev->hca_type;
1138 	__mthca_remove_one(pdev);
1139 	return __mthca_init_one(pdev, hca_type);
1140 }
1141 
1142 static int __devinit mthca_init_one(struct pci_dev *pdev,
1143 				    const struct pci_device_id *id)
1144 {
1145 	int ret;
1146 
1147 	mutex_lock(&mthca_device_mutex);
1148 
1149 	printk_once(KERN_INFO "%s", mthca_version);
1150 
1151 	if (id->driver_data >= ARRAY_SIZE(mthca_hca_table)) {
1152 		printk(KERN_ERR PFX "%s has invalid driver data %lx\n",
1153 		       pci_name(pdev), id->driver_data);
1154 		mutex_unlock(&mthca_device_mutex);
1155 		return -ENODEV;
1156 	}
1157 
1158 	ret = __mthca_init_one(pdev, id->driver_data);
1159 
1160 	mutex_unlock(&mthca_device_mutex);
1161 
1162 	return ret;
1163 }
1164 
1165 static void __devexit mthca_remove_one(struct pci_dev *pdev)
1166 {
1167 	mutex_lock(&mthca_device_mutex);
1168 	__mthca_remove_one(pdev);
1169 	mutex_unlock(&mthca_device_mutex);
1170 }
1171 
1172 static struct pci_device_id mthca_pci_table[] = {
1173 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_TAVOR),
1174 	  .driver_data = TAVOR },
1175 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_TAVOR),
1176 	  .driver_data = TAVOR },
1177 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1178 	  .driver_data = ARBEL_COMPAT },
1179 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT),
1180 	  .driver_data = ARBEL_COMPAT },
1181 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_ARBEL),
1182 	  .driver_data = ARBEL_NATIVE },
1183 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_ARBEL),
1184 	  .driver_data = ARBEL_NATIVE },
1185 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_SINAI),
1186 	  .driver_data = SINAI },
1187 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_SINAI),
1188 	  .driver_data = SINAI },
1189 	{ PCI_DEVICE(PCI_VENDOR_ID_MELLANOX, PCI_DEVICE_ID_MELLANOX_SINAI_OLD),
1190 	  .driver_data = SINAI },
1191 	{ PCI_DEVICE(PCI_VENDOR_ID_TOPSPIN, PCI_DEVICE_ID_MELLANOX_SINAI_OLD),
1192 	  .driver_data = SINAI },
1193 	{ 0, }
1194 };
1195 
1196 MODULE_DEVICE_TABLE(pci, mthca_pci_table);
1197 
1198 static struct pci_driver mthca_driver = {
1199 	.name		= DRV_NAME,
1200 	.id_table	= mthca_pci_table,
1201 	.probe		= mthca_init_one,
1202 	.remove		= __devexit_p(mthca_remove_one)
1203 };
1204 
1205 static void __init __mthca_check_profile_val(const char *name, int *pval,
1206 					     int pval_default)
1207 {
1208 	/* value must be positive and power of 2 */
1209 	int old_pval = *pval;
1210 
1211 	if (old_pval <= 0)
1212 		*pval = pval_default;
1213 	else
1214 		*pval = roundup_pow_of_two(old_pval);
1215 
1216 	if (old_pval != *pval) {
1217 		printk(KERN_WARNING PFX "Invalid value %d for %s in module parameter.\n",
1218 		       old_pval, name);
1219 		printk(KERN_WARNING PFX "Corrected %s to %d.\n", name, *pval);
1220 	}
1221 }
1222 
1223 #define mthca_check_profile_val(name, default)				\
1224 	__mthca_check_profile_val(#name, &hca_profile.name, default)
1225 
1226 static void __init mthca_validate_profile(void)
1227 {
1228 	mthca_check_profile_val(num_qp,            MTHCA_DEFAULT_NUM_QP);
1229 	mthca_check_profile_val(rdb_per_qp,        MTHCA_DEFAULT_RDB_PER_QP);
1230 	mthca_check_profile_val(num_cq,            MTHCA_DEFAULT_NUM_CQ);
1231 	mthca_check_profile_val(num_mcg, 	   MTHCA_DEFAULT_NUM_MCG);
1232 	mthca_check_profile_val(num_mpt, 	   MTHCA_DEFAULT_NUM_MPT);
1233 	mthca_check_profile_val(num_mtt, 	   MTHCA_DEFAULT_NUM_MTT);
1234 	mthca_check_profile_val(num_udav,          MTHCA_DEFAULT_NUM_UDAV);
1235 	mthca_check_profile_val(fmr_reserved_mtts, MTHCA_DEFAULT_NUM_RESERVED_MTTS);
1236 
1237 	if (hca_profile.fmr_reserved_mtts >= hca_profile.num_mtt) {
1238 		printk(KERN_WARNING PFX "Invalid fmr_reserved_mtts module parameter %d.\n",
1239 		       hca_profile.fmr_reserved_mtts);
1240 		printk(KERN_WARNING PFX "(Must be smaller than num_mtt %d)\n",
1241 		       hca_profile.num_mtt);
1242 		hca_profile.fmr_reserved_mtts = hca_profile.num_mtt / 2;
1243 		printk(KERN_WARNING PFX "Corrected fmr_reserved_mtts to %d.\n",
1244 		       hca_profile.fmr_reserved_mtts);
1245 	}
1246 
1247 	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 5)) {
1248 		printk(KERN_WARNING PFX "bad log_mtts_per_seg (%d). Using default - %d\n",
1249 		       log_mtts_per_seg, ilog2(MTHCA_MTT_SEG_SIZE / 8));
1250 		log_mtts_per_seg = ilog2(MTHCA_MTT_SEG_SIZE / 8);
1251 	}
1252 }
1253 
1254 static int __init mthca_init(void)
1255 {
1256 	int ret;
1257 
1258 	mthca_validate_profile();
1259 
1260 	ret = mthca_catas_init();
1261 	if (ret)
1262 		return ret;
1263 
1264 	ret = pci_register_driver(&mthca_driver);
1265 	if (ret < 0) {
1266 		mthca_catas_cleanup();
1267 		return ret;
1268 	}
1269 
1270 	return 0;
1271 }
1272 
1273 static void __exit mthca_cleanup(void)
1274 {
1275 	pci_unregister_driver(&mthca_driver);
1276 	mthca_catas_cleanup();
1277 }
1278 
1279 module_init(mthca_init);
1280 module_exit(mthca_cleanup);
1281