xref: /linux/drivers/infiniband/hw/qib/qib.h (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
1 #ifndef _QIB_KERNEL_H
2 #define _QIB_KERNEL_H
3 /*
4  * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
5  * All rights reserved.
6  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
7  *
8  * This software is available to you under a choice of one of two
9  * licenses.  You may choose to be licensed under the terms of the GNU
10  * General Public License (GPL) Version 2, available from the file
11  * COPYING in the main directory of this source tree, or the
12  * OpenIB.org BSD license below:
13  *
14  *     Redistribution and use in source and binary forms, with or
15  *     without modification, are permitted provided that the following
16  *     conditions are met:
17  *
18  *      - Redistributions of source code must retain the above
19  *        copyright notice, this list of conditions and the following
20  *        disclaimer.
21  *
22  *      - Redistributions in binary form must reproduce the above
23  *        copyright notice, this list of conditions and the following
24  *        disclaimer in the documentation and/or other materials
25  *        provided with the distribution.
26  *
27  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
34  * SOFTWARE.
35  */
36 
37 /*
38  * This header file is the base header file for qlogic_ib kernel code
39  * qib_user.h serves a similar purpose for user code.
40  */
41 
42 #include <linux/interrupt.h>
43 #include <linux/pci.h>
44 #include <linux/dma-mapping.h>
45 #include <linux/mutex.h>
46 #include <linux/list.h>
47 #include <linux/scatterlist.h>
48 #include <linux/slab.h>
49 #include <linux/io.h>
50 #include <linux/fs.h>
51 #include <linux/completion.h>
52 #include <linux/kref.h>
53 #include <linux/sched.h>
54 
55 #include "qib_common.h"
56 #include "qib_verbs.h"
57 
58 /* only s/w major version of QLogic_IB we can handle */
59 #define QIB_CHIP_VERS_MAJ 2U
60 
61 /* don't care about this except printing */
62 #define QIB_CHIP_VERS_MIN 0U
63 
64 /* The Organization Unique Identifier (Mfg code), and its position in GUID */
65 #define QIB_OUI 0x001175
66 #define QIB_OUI_LSB 40
67 
68 /*
69  * per driver stats, either not device nor port-specific, or
70  * summed over all of the devices and ports.
71  * They are described by name via ipathfs filesystem, so layout
72  * and number of elements can change without breaking compatibility.
73  * If members are added or deleted qib_statnames[] in qib_fs.c must
74  * change to match.
75  */
76 struct qlogic_ib_stats {
77 	__u64 sps_ints; /* number of interrupts handled */
78 	__u64 sps_errints; /* number of error interrupts */
79 	__u64 sps_txerrs; /* tx-related packet errors */
80 	__u64 sps_rcverrs; /* non-crc rcv packet errors */
81 	__u64 sps_hwerrs; /* hardware errors reported (parity, etc.) */
82 	__u64 sps_nopiobufs; /* no pio bufs avail from kernel */
83 	__u64 sps_ctxts; /* number of contexts currently open */
84 	__u64 sps_lenerrs; /* number of kernel packets where RHF != LRH len */
85 	__u64 sps_buffull;
86 	__u64 sps_hdrfull;
87 };
88 
89 extern struct qlogic_ib_stats qib_stats;
90 extern struct pci_error_handlers qib_pci_err_handler;
91 extern struct pci_driver qib_driver;
92 
93 #define QIB_CHIP_SWVERSION QIB_CHIP_VERS_MAJ
94 /*
95  * First-cut critierion for "device is active" is
96  * two thousand dwords combined Tx, Rx traffic per
97  * 5-second interval. SMA packets are 64 dwords,
98  * and occur "a few per second", presumably each way.
99  */
100 #define QIB_TRAFFIC_ACTIVE_THRESHOLD (2000)
101 
102 /*
103  * Struct used to indicate which errors are logged in each of the
104  * error-counters that are logged to EEPROM. A counter is incremented
105  * _once_ (saturating at 255) for each event with any bits set in
106  * the error or hwerror register masks below.
107  */
108 #define QIB_EEP_LOG_CNT (4)
109 struct qib_eep_log_mask {
110 	u64 errs_to_log;
111 	u64 hwerrs_to_log;
112 };
113 
114 /*
115  * Below contains all data related to a single context (formerly called port).
116  */
117 struct qib_ctxtdata {
118 	void **rcvegrbuf;
119 	dma_addr_t *rcvegrbuf_phys;
120 	/* rcvhdrq base, needs mmap before useful */
121 	void *rcvhdrq;
122 	/* kernel virtual address where hdrqtail is updated */
123 	void *rcvhdrtail_kvaddr;
124 	/*
125 	 * temp buffer for expected send setup, allocated at open, instead
126 	 * of each setup call
127 	 */
128 	void *tid_pg_list;
129 	/*
130 	 * Shared page for kernel to signal user processes that send buffers
131 	 * need disarming.  The process should call QIB_CMD_DISARM_BUFS
132 	 * or QIB_CMD_ACK_EVENT with IPATH_EVENT_DISARM_BUFS set.
133 	 */
134 	unsigned long *user_event_mask;
135 	/* when waiting for rcv or pioavail */
136 	wait_queue_head_t wait;
137 	/*
138 	 * rcvegr bufs base, physical, must fit
139 	 * in 44 bits so 32 bit programs mmap64 44 bit works)
140 	 */
141 	dma_addr_t rcvegr_phys;
142 	/* mmap of hdrq, must fit in 44 bits */
143 	dma_addr_t rcvhdrq_phys;
144 	dma_addr_t rcvhdrqtailaddr_phys;
145 
146 	/*
147 	 * number of opens (including slave sub-contexts) on this instance
148 	 * (ignoring forks, dup, etc. for now)
149 	 */
150 	int cnt;
151 	/*
152 	 * how much space to leave at start of eager TID entries for
153 	 * protocol use, on each TID
154 	 */
155 	/* instead of calculating it */
156 	unsigned ctxt;
157 	/* non-zero if ctxt is being shared. */
158 	u16 subctxt_cnt;
159 	/* non-zero if ctxt is being shared. */
160 	u16 subctxt_id;
161 	/* number of eager TID entries. */
162 	u16 rcvegrcnt;
163 	/* index of first eager TID entry. */
164 	u16 rcvegr_tid_base;
165 	/* number of pio bufs for this ctxt (all procs, if shared) */
166 	u32 piocnt;
167 	/* first pio buffer for this ctxt */
168 	u32 pio_base;
169 	/* chip offset of PIO buffers for this ctxt */
170 	u32 piobufs;
171 	/* how many alloc_pages() chunks in rcvegrbuf_pages */
172 	u32 rcvegrbuf_chunks;
173 	/* how many egrbufs per chunk */
174 	u16 rcvegrbufs_perchunk;
175 	/* ilog2 of above */
176 	u16 rcvegrbufs_perchunk_shift;
177 	/* order for rcvegrbuf_pages */
178 	size_t rcvegrbuf_size;
179 	/* rcvhdrq size (for freeing) */
180 	size_t rcvhdrq_size;
181 	/* per-context flags for fileops/intr communication */
182 	unsigned long flag;
183 	/* next expected TID to check when looking for free */
184 	u32 tidcursor;
185 	/* WAIT_RCV that timed out, no interrupt */
186 	u32 rcvwait_to;
187 	/* WAIT_PIO that timed out, no interrupt */
188 	u32 piowait_to;
189 	/* WAIT_RCV already happened, no wait */
190 	u32 rcvnowait;
191 	/* WAIT_PIO already happened, no wait */
192 	u32 pionowait;
193 	/* total number of polled urgent packets */
194 	u32 urgent;
195 	/* saved total number of polled urgent packets for poll edge trigger */
196 	u32 urgent_poll;
197 	/* pid of process using this ctxt */
198 	pid_t pid;
199 	pid_t subpid[QLOGIC_IB_MAX_SUBCTXT];
200 	/* same size as task_struct .comm[], command that opened context */
201 	char comm[16];
202 	/* pkeys set by this use of this ctxt */
203 	u16 pkeys[4];
204 	/* so file ops can get at unit */
205 	struct qib_devdata *dd;
206 	/* so funcs that need physical port can get it easily */
207 	struct qib_pportdata *ppd;
208 	/* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
209 	void *subctxt_uregbase;
210 	/* An array of pages for the eager receive buffers * N */
211 	void *subctxt_rcvegrbuf;
212 	/* An array of pages for the eager header queue entries * N */
213 	void *subctxt_rcvhdr_base;
214 	/* The version of the library which opened this ctxt */
215 	u32 userversion;
216 	/* Bitmask of active slaves */
217 	u32 active_slaves;
218 	/* Type of packets or conditions we want to poll for */
219 	u16 poll_type;
220 	/* receive packet sequence counter */
221 	u8 seq_cnt;
222 	u8 redirect_seq_cnt;
223 	/* ctxt rcvhdrq head offset */
224 	u32 head;
225 	u32 pkt_count;
226 	/* lookaside fields */
227 	struct qib_qp *lookaside_qp;
228 	u32 lookaside_qpn;
229 	/* QPs waiting for context processing */
230 	struct list_head qp_wait_list;
231 };
232 
233 struct qib_sge_state;
234 
235 struct qib_sdma_txreq {
236 	int                 flags;
237 	int                 sg_count;
238 	dma_addr_t          addr;
239 	void              (*callback)(struct qib_sdma_txreq *, int);
240 	u16                 start_idx;  /* sdma private */
241 	u16                 next_descq_idx;  /* sdma private */
242 	struct list_head    list;       /* sdma private */
243 };
244 
245 struct qib_sdma_desc {
246 	__le64 qw[2];
247 };
248 
249 struct qib_verbs_txreq {
250 	struct qib_sdma_txreq   txreq;
251 	struct qib_qp           *qp;
252 	struct qib_swqe         *wqe;
253 	u32                     dwords;
254 	u16                     hdr_dwords;
255 	u16                     hdr_inx;
256 	struct qib_pio_header	*align_buf;
257 	struct qib_mregion	*mr;
258 	struct qib_sge_state    *ss;
259 };
260 
261 #define QIB_SDMA_TXREQ_F_USELARGEBUF  0x1
262 #define QIB_SDMA_TXREQ_F_HEADTOHOST   0x2
263 #define QIB_SDMA_TXREQ_F_INTREQ       0x4
264 #define QIB_SDMA_TXREQ_F_FREEBUF      0x8
265 #define QIB_SDMA_TXREQ_F_FREEDESC     0x10
266 
267 #define QIB_SDMA_TXREQ_S_OK        0
268 #define QIB_SDMA_TXREQ_S_SENDERROR 1
269 #define QIB_SDMA_TXREQ_S_ABORTED   2
270 #define QIB_SDMA_TXREQ_S_SHUTDOWN  3
271 
272 /*
273  * Get/Set IB link-level config parameters for f_get/set_ib_cfg()
274  * Mostly for MADs that set or query link parameters, also ipath
275  * config interfaces
276  */
277 #define QIB_IB_CFG_LIDLMC 0 /* LID (LS16b) and Mask (MS16b) */
278 #define QIB_IB_CFG_LWID_ENB 2 /* allowed Link-width */
279 #define QIB_IB_CFG_LWID 3 /* currently active Link-width */
280 #define QIB_IB_CFG_SPD_ENB 4 /* allowed Link speeds */
281 #define QIB_IB_CFG_SPD 5 /* current Link spd */
282 #define QIB_IB_CFG_RXPOL_ENB 6 /* Auto-RX-polarity enable */
283 #define QIB_IB_CFG_LREV_ENB 7 /* Auto-Lane-reversal enable */
284 #define QIB_IB_CFG_LINKLATENCY 8 /* Link Latency (IB1.2 only) */
285 #define QIB_IB_CFG_HRTBT 9 /* IB heartbeat off/enable/auto; DDR/QDR only */
286 #define QIB_IB_CFG_OP_VLS 10 /* operational VLs */
287 #define QIB_IB_CFG_VL_HIGH_CAP 11 /* num of VL high priority weights */
288 #define QIB_IB_CFG_VL_LOW_CAP 12 /* num of VL low priority weights */
289 #define QIB_IB_CFG_OVERRUN_THRESH 13 /* IB overrun threshold */
290 #define QIB_IB_CFG_PHYERR_THRESH 14 /* IB PHY error threshold */
291 #define QIB_IB_CFG_LINKDEFAULT 15 /* IB link default (sleep/poll) */
292 #define QIB_IB_CFG_PKEYS 16 /* update partition keys */
293 #define QIB_IB_CFG_MTU 17 /* update MTU in IBC */
294 #define QIB_IB_CFG_LSTATE 18 /* update linkcmd and linkinitcmd in IBC */
295 #define QIB_IB_CFG_VL_HIGH_LIMIT 19
296 #define QIB_IB_CFG_PMA_TICKS 20 /* PMA sample tick resolution */
297 #define QIB_IB_CFG_PORT 21 /* switch port we are connected to */
298 
299 /*
300  * for CFG_LSTATE: LINKCMD in upper 16 bits, LINKINITCMD in lower 16
301  * IB_LINKINITCMD_POLL and SLEEP are also used as set/get values for
302  * QIB_IB_CFG_LINKDEFAULT cmd
303  */
304 #define   IB_LINKCMD_DOWN   (0 << 16)
305 #define   IB_LINKCMD_ARMED  (1 << 16)
306 #define   IB_LINKCMD_ACTIVE (2 << 16)
307 #define   IB_LINKINITCMD_NOP     0
308 #define   IB_LINKINITCMD_POLL    1
309 #define   IB_LINKINITCMD_SLEEP   2
310 #define   IB_LINKINITCMD_DISABLE 3
311 
312 /*
313  * valid states passed to qib_set_linkstate() user call
314  */
315 #define QIB_IB_LINKDOWN         0
316 #define QIB_IB_LINKARM          1
317 #define QIB_IB_LINKACTIVE       2
318 #define QIB_IB_LINKDOWN_ONLY    3
319 #define QIB_IB_LINKDOWN_SLEEP   4
320 #define QIB_IB_LINKDOWN_DISABLE 5
321 
322 /*
323  * These 7 values (SDR, DDR, and QDR may be ORed for auto-speed
324  * negotiation) are used for the 3rd argument to path_f_set_ib_cfg
325  * with cmd QIB_IB_CFG_SPD_ENB, by direct calls or via sysfs.  They
326  * are also the the possible values for qib_link_speed_enabled and active
327  * The values were chosen to match values used within the IB spec.
328  */
329 #define QIB_IB_SDR 1
330 #define QIB_IB_DDR 2
331 #define QIB_IB_QDR 4
332 
333 #define QIB_DEFAULT_MTU 4096
334 
335 /* max number of IB ports supported per HCA */
336 #define QIB_MAX_IB_PORTS 2
337 
338 /*
339  * Possible IB config parameters for f_get/set_ib_table()
340  */
341 #define QIB_IB_TBL_VL_HIGH_ARB 1 /* Get/set VL high priority weights */
342 #define QIB_IB_TBL_VL_LOW_ARB 2 /* Get/set VL low priority weights */
343 
344 /*
345  * Possible "operations" for f_rcvctrl(ppd, op, ctxt)
346  * these are bits so they can be combined, e.g.
347  * QIB_RCVCTRL_INTRAVAIL_ENB | QIB_RCVCTRL_CTXT_ENB
348  */
349 #define QIB_RCVCTRL_TAILUPD_ENB 0x01
350 #define QIB_RCVCTRL_TAILUPD_DIS 0x02
351 #define QIB_RCVCTRL_CTXT_ENB 0x04
352 #define QIB_RCVCTRL_CTXT_DIS 0x08
353 #define QIB_RCVCTRL_INTRAVAIL_ENB 0x10
354 #define QIB_RCVCTRL_INTRAVAIL_DIS 0x20
355 #define QIB_RCVCTRL_PKEY_ENB 0x40  /* Note, default is enabled */
356 #define QIB_RCVCTRL_PKEY_DIS 0x80
357 #define QIB_RCVCTRL_BP_ENB 0x0100
358 #define QIB_RCVCTRL_BP_DIS 0x0200
359 #define QIB_RCVCTRL_TIDFLOW_ENB 0x0400
360 #define QIB_RCVCTRL_TIDFLOW_DIS 0x0800
361 
362 /*
363  * Possible "operations" for f_sendctrl(ppd, op, var)
364  * these are bits so they can be combined, e.g.
365  * QIB_SENDCTRL_BUFAVAIL_ENB | QIB_SENDCTRL_ENB
366  * Some operations (e.g. DISARM, ABORT) are known to
367  * be "one-shot", so do not modify shadow.
368  */
369 #define QIB_SENDCTRL_DISARM       (0x1000)
370 #define QIB_SENDCTRL_DISARM_BUF(bufn) ((bufn) | QIB_SENDCTRL_DISARM)
371 	/* available (0x2000) */
372 #define QIB_SENDCTRL_AVAIL_DIS    (0x4000)
373 #define QIB_SENDCTRL_AVAIL_ENB    (0x8000)
374 #define QIB_SENDCTRL_AVAIL_BLIP  (0x10000)
375 #define QIB_SENDCTRL_SEND_DIS    (0x20000)
376 #define QIB_SENDCTRL_SEND_ENB    (0x40000)
377 #define QIB_SENDCTRL_FLUSH       (0x80000)
378 #define QIB_SENDCTRL_CLEAR      (0x100000)
379 #define QIB_SENDCTRL_DISARM_ALL (0x200000)
380 
381 /*
382  * These are the generic indices for requesting per-port
383  * counter values via the f_portcntr function.  They
384  * are always returned as 64 bit values, although most
385  * are 32 bit counters.
386  */
387 /* send-related counters */
388 #define QIBPORTCNTR_PKTSEND         0U
389 #define QIBPORTCNTR_WORDSEND        1U
390 #define QIBPORTCNTR_PSXMITDATA      2U
391 #define QIBPORTCNTR_PSXMITPKTS      3U
392 #define QIBPORTCNTR_PSXMITWAIT      4U
393 #define QIBPORTCNTR_SENDSTALL       5U
394 /* receive-related counters */
395 #define QIBPORTCNTR_PKTRCV          6U
396 #define QIBPORTCNTR_PSRCVDATA       7U
397 #define QIBPORTCNTR_PSRCVPKTS       8U
398 #define QIBPORTCNTR_RCVEBP          9U
399 #define QIBPORTCNTR_RCVOVFL         10U
400 #define QIBPORTCNTR_WORDRCV         11U
401 /* IB link related error counters */
402 #define QIBPORTCNTR_RXLOCALPHYERR   12U
403 #define QIBPORTCNTR_RXVLERR         13U
404 #define QIBPORTCNTR_ERRICRC         14U
405 #define QIBPORTCNTR_ERRVCRC         15U
406 #define QIBPORTCNTR_ERRLPCRC        16U
407 #define QIBPORTCNTR_BADFORMAT       17U
408 #define QIBPORTCNTR_ERR_RLEN        18U
409 #define QIBPORTCNTR_IBSYMBOLERR     19U
410 #define QIBPORTCNTR_INVALIDRLEN     20U
411 #define QIBPORTCNTR_UNSUPVL         21U
412 #define QIBPORTCNTR_EXCESSBUFOVFL   22U
413 #define QIBPORTCNTR_ERRLINK         23U
414 #define QIBPORTCNTR_IBLINKDOWN      24U
415 #define QIBPORTCNTR_IBLINKERRRECOV  25U
416 #define QIBPORTCNTR_LLI             26U
417 /* other error counters */
418 #define QIBPORTCNTR_RXDROPPKT       27U
419 #define QIBPORTCNTR_VL15PKTDROP     28U
420 #define QIBPORTCNTR_ERRPKEY         29U
421 #define QIBPORTCNTR_KHDROVFL        30U
422 /* sampling counters (these are actually control registers) */
423 #define QIBPORTCNTR_PSINTERVAL      31U
424 #define QIBPORTCNTR_PSSTART         32U
425 #define QIBPORTCNTR_PSSTAT          33U
426 
427 /* how often we check for packet activity for "power on hours (in seconds) */
428 #define ACTIVITY_TIMER 5
429 
430 /* Below is an opaque struct. Each chip (device) can maintain
431  * private data needed for its operation, but not germane to the
432  * rest of the driver.  For convenience, we define another that
433  * is chip-specific, per-port
434  */
435 struct qib_chip_specific;
436 struct qib_chipport_specific;
437 
438 enum qib_sdma_states {
439 	qib_sdma_state_s00_hw_down,
440 	qib_sdma_state_s10_hw_start_up_wait,
441 	qib_sdma_state_s20_idle,
442 	qib_sdma_state_s30_sw_clean_up_wait,
443 	qib_sdma_state_s40_hw_clean_up_wait,
444 	qib_sdma_state_s50_hw_halt_wait,
445 	qib_sdma_state_s99_running,
446 };
447 
448 enum qib_sdma_events {
449 	qib_sdma_event_e00_go_hw_down,
450 	qib_sdma_event_e10_go_hw_start,
451 	qib_sdma_event_e20_hw_started,
452 	qib_sdma_event_e30_go_running,
453 	qib_sdma_event_e40_sw_cleaned,
454 	qib_sdma_event_e50_hw_cleaned,
455 	qib_sdma_event_e60_hw_halted,
456 	qib_sdma_event_e70_go_idle,
457 	qib_sdma_event_e7220_err_halted,
458 	qib_sdma_event_e7322_err_halted,
459 	qib_sdma_event_e90_timer_tick,
460 };
461 
462 extern char *qib_sdma_state_names[];
463 extern char *qib_sdma_event_names[];
464 
465 struct sdma_set_state_action {
466 	unsigned op_enable:1;
467 	unsigned op_intenable:1;
468 	unsigned op_halt:1;
469 	unsigned op_drain:1;
470 	unsigned go_s99_running_tofalse:1;
471 	unsigned go_s99_running_totrue:1;
472 };
473 
474 struct qib_sdma_state {
475 	struct kref          kref;
476 	struct completion    comp;
477 	enum qib_sdma_states current_state;
478 	struct sdma_set_state_action *set_state_action;
479 	unsigned             current_op;
480 	unsigned             go_s99_running;
481 	unsigned             first_sendbuf;
482 	unsigned             last_sendbuf; /* really last +1 */
483 	/* debugging/devel */
484 	enum qib_sdma_states previous_state;
485 	unsigned             previous_op;
486 	enum qib_sdma_events last_event;
487 };
488 
489 struct xmit_wait {
490 	struct timer_list timer;
491 	u64 counter;
492 	u8 flags;
493 	struct cache {
494 		u64 psxmitdata;
495 		u64 psrcvdata;
496 		u64 psxmitpkts;
497 		u64 psrcvpkts;
498 		u64 psxmitwait;
499 	} counter_cache;
500 };
501 
502 /*
503  * The structure below encapsulates data relevant to a physical IB Port.
504  * Current chips support only one such port, but the separation
505  * clarifies things a bit. Note that to conform to IB conventions,
506  * port-numbers are one-based. The first or only port is port1.
507  */
508 struct qib_pportdata {
509 	struct qib_ibport ibport_data;
510 
511 	struct qib_devdata *dd;
512 	struct qib_chippport_specific *cpspec; /* chip-specific per-port */
513 	struct kobject pport_kobj;
514 	struct kobject sl2vl_kobj;
515 	struct kobject diagc_kobj;
516 
517 	/* GUID for this interface, in network order */
518 	__be64 guid;
519 
520 	/* QIB_POLL, etc. link-state specific flags, per port */
521 	u32 lflags;
522 	/* qib_lflags driver is waiting for */
523 	u32 state_wanted;
524 	spinlock_t lflags_lock;
525 	/* number of (port-specific) interrupts for this port -- saturates... */
526 	u32 int_counter;
527 
528 	/* ref count for each pkey */
529 	atomic_t pkeyrefs[4];
530 
531 	/*
532 	 * this address is mapped readonly into user processes so they can
533 	 * get status cheaply, whenever they want.  One qword of status per port
534 	 */
535 	u64 *statusp;
536 
537 	/* SendDMA related entries */
538 	spinlock_t            sdma_lock;
539 	struct qib_sdma_state sdma_state;
540 	unsigned long         sdma_buf_jiffies;
541 	struct qib_sdma_desc *sdma_descq;
542 	u64                   sdma_descq_added;
543 	u64                   sdma_descq_removed;
544 	u16                   sdma_descq_cnt;
545 	u16                   sdma_descq_tail;
546 	u16                   sdma_descq_head;
547 	u16                   sdma_next_intr;
548 	u16                   sdma_reset_wait;
549 	u8                    sdma_generation;
550 	struct tasklet_struct sdma_sw_clean_up_task;
551 	struct list_head      sdma_activelist;
552 
553 	dma_addr_t       sdma_descq_phys;
554 	volatile __le64 *sdma_head_dma; /* DMA'ed by chip */
555 	dma_addr_t       sdma_head_phys;
556 
557 	wait_queue_head_t state_wait; /* for state_wanted */
558 
559 	/* HoL blocking for SMP replies */
560 	unsigned          hol_state;
561 	struct timer_list hol_timer;
562 
563 	/*
564 	 * Shadow copies of registers; size indicates read access size.
565 	 * Most of them are readonly, but some are write-only register,
566 	 * where we manipulate the bits in the shadow copy, and then write
567 	 * the shadow copy to qlogic_ib.
568 	 *
569 	 * We deliberately make most of these 32 bits, since they have
570 	 * restricted range.  For any that we read, we won't to generate 32
571 	 * bit accesses, since Opteron will generate 2 separate 32 bit HT
572 	 * transactions for a 64 bit read, and we want to avoid unnecessary
573 	 * bus transactions.
574 	 */
575 
576 	/* This is the 64 bit group */
577 	/* last ibcstatus.  opaque outside chip-specific code */
578 	u64 lastibcstat;
579 
580 	/* these are the "32 bit" regs */
581 
582 	/*
583 	 * the following two are 32-bit bitmasks, but {test,clear,set}_bit
584 	 * all expect bit fields to be "unsigned long"
585 	 */
586 	unsigned long p_rcvctrl; /* shadow per-port rcvctrl */
587 	unsigned long p_sendctrl; /* shadow per-port sendctrl */
588 
589 	u32 ibmtu; /* The MTU programmed for this unit */
590 	/*
591 	 * Current max size IB packet (in bytes) including IB headers, that
592 	 * we can send. Changes when ibmtu changes.
593 	 */
594 	u32 ibmaxlen;
595 	/*
596 	 * ibmaxlen at init time, limited by chip and by receive buffer
597 	 * size.  Not changed after init.
598 	 */
599 	u32 init_ibmaxlen;
600 	/* LID programmed for this instance */
601 	u16 lid;
602 	/* list of pkeys programmed; 0 if not set */
603 	u16 pkeys[4];
604 	/* LID mask control */
605 	u8 lmc;
606 	u8 link_width_supported;
607 	u8 link_speed_supported;
608 	u8 link_width_enabled;
609 	u8 link_speed_enabled;
610 	u8 link_width_active;
611 	u8 link_speed_active;
612 	u8 vls_supported;
613 	u8 vls_operational;
614 	/* Rx Polarity inversion (compensate for ~tx on partner) */
615 	u8 rx_pol_inv;
616 
617 	u8 hw_pidx;     /* physical port index */
618 	u8 port;        /* IB port number and index into dd->pports - 1 */
619 
620 	u8 delay_mult;
621 
622 	/* used to override LED behavior */
623 	u8 led_override;  /* Substituted for normal value, if non-zero */
624 	u16 led_override_timeoff; /* delta to next timer event */
625 	u8 led_override_vals[2]; /* Alternates per blink-frame */
626 	u8 led_override_phase; /* Just counts, LSB picks from vals[] */
627 	atomic_t led_override_timer_active;
628 	/* Used to flash LEDs in override mode */
629 	struct timer_list led_override_timer;
630 	struct xmit_wait cong_stats;
631 	struct timer_list symerr_clear_timer;
632 };
633 
634 /* Observers. Not to be taken lightly, possibly not to ship. */
635 /*
636  * If a diag read or write is to (bottom <= offset <= top),
637  * the "hoook" is called, allowing, e.g. shadows to be
638  * updated in sync with the driver. struct diag_observer
639  * is the "visible" part.
640  */
641 struct diag_observer;
642 
643 typedef int (*diag_hook) (struct qib_devdata *dd,
644 	const struct diag_observer *op,
645 	u32 offs, u64 *data, u64 mask, int only_32);
646 
647 struct diag_observer {
648 	diag_hook hook;
649 	u32 bottom;
650 	u32 top;
651 };
652 
653 extern int qib_register_observer(struct qib_devdata *dd,
654 	const struct diag_observer *op);
655 
656 /* Only declared here, not defined. Private to diags */
657 struct diag_observer_list_elt;
658 
659 /* device data struct now contains only "general per-device" info.
660  * fields related to a physical IB port are in a qib_pportdata struct,
661  * described above) while fields only used by a particular chip-type are in
662  * a qib_chipdata struct, whose contents are opaque to this file.
663  */
664 struct qib_devdata {
665 	struct qib_ibdev verbs_dev;     /* must be first */
666 	struct list_head list;
667 	/* pointers to related structs for this device */
668 	/* pci access data structure */
669 	struct pci_dev *pcidev;
670 	struct cdev *user_cdev;
671 	struct cdev *diag_cdev;
672 	struct device *user_device;
673 	struct device *diag_device;
674 
675 	/* mem-mapped pointer to base of chip regs */
676 	u64 __iomem *kregbase;
677 	/* end of mem-mapped chip space excluding sendbuf and user regs */
678 	u64 __iomem *kregend;
679 	/* physical address of chip for io_remap, etc. */
680 	resource_size_t physaddr;
681 	/* qib_cfgctxts pointers */
682 	struct qib_ctxtdata **rcd; /* Receive Context Data */
683 
684 	/* qib_pportdata, points to array of (physical) port-specific
685 	 * data structs, indexed by pidx (0..n-1)
686 	 */
687 	struct qib_pportdata *pport;
688 	struct qib_chip_specific *cspec; /* chip-specific */
689 
690 	/* kvirt address of 1st 2k pio buffer */
691 	void __iomem *pio2kbase;
692 	/* kvirt address of 1st 4k pio buffer */
693 	void __iomem *pio4kbase;
694 	/* mem-mapped pointer to base of PIO buffers (if using WC PAT) */
695 	void __iomem *piobase;
696 	/* mem-mapped pointer to base of user chip regs (if using WC PAT) */
697 	u64 __iomem *userbase;
698 	void __iomem *piovl15base; /* base of VL15 buffers, if not WC */
699 	/*
700 	 * points to area where PIOavail registers will be DMA'ed.
701 	 * Has to be on a page of it's own, because the page will be
702 	 * mapped into user program space.  This copy is *ONLY* ever
703 	 * written by DMA, not by the driver!  Need a copy per device
704 	 * when we get to multiple devices
705 	 */
706 	volatile __le64 *pioavailregs_dma; /* DMA'ed by chip */
707 	/* physical address where updates occur */
708 	dma_addr_t pioavailregs_phys;
709 
710 	/* device-specific implementations of functions needed by
711 	 * common code. Contrary to previous consensus, we can't
712 	 * really just point to a device-specific table, because we
713 	 * may need to "bend", e.g. *_f_put_tid
714 	 */
715 	/* fallback to alternate interrupt type if possible */
716 	int (*f_intr_fallback)(struct qib_devdata *);
717 	/* hard reset chip */
718 	int (*f_reset)(struct qib_devdata *);
719 	void (*f_quiet_serdes)(struct qib_pportdata *);
720 	int (*f_bringup_serdes)(struct qib_pportdata *);
721 	int (*f_early_init)(struct qib_devdata *);
722 	void (*f_clear_tids)(struct qib_devdata *, struct qib_ctxtdata *);
723 	void (*f_put_tid)(struct qib_devdata *, u64 __iomem*,
724 				u32, unsigned long);
725 	void (*f_cleanup)(struct qib_devdata *);
726 	void (*f_setextled)(struct qib_pportdata *, u32);
727 	/* fill out chip-specific fields */
728 	int (*f_get_base_info)(struct qib_ctxtdata *, struct qib_base_info *);
729 	/* free irq */
730 	void (*f_free_irq)(struct qib_devdata *);
731 	struct qib_message_header *(*f_get_msgheader)
732 					(struct qib_devdata *, __le32 *);
733 	void (*f_config_ctxts)(struct qib_devdata *);
734 	int (*f_get_ib_cfg)(struct qib_pportdata *, int);
735 	int (*f_set_ib_cfg)(struct qib_pportdata *, int, u32);
736 	int (*f_set_ib_loopback)(struct qib_pportdata *, const char *);
737 	int (*f_get_ib_table)(struct qib_pportdata *, int, void *);
738 	int (*f_set_ib_table)(struct qib_pportdata *, int, void *);
739 	u32 (*f_iblink_state)(u64);
740 	u8 (*f_ibphys_portstate)(u64);
741 	void (*f_xgxs_reset)(struct qib_pportdata *);
742 	/* per chip actions needed for IB Link up/down changes */
743 	int (*f_ib_updown)(struct qib_pportdata *, int, u64);
744 	u32 __iomem *(*f_getsendbuf)(struct qib_pportdata *, u64, u32 *);
745 	/* Read/modify/write of GPIO pins (potentially chip-specific */
746 	int (*f_gpio_mod)(struct qib_devdata *dd, u32 out, u32 dir,
747 		u32 mask);
748 	/* Enable writes to config EEPROM (if supported) */
749 	int (*f_eeprom_wen)(struct qib_devdata *dd, int wen);
750 	/*
751 	 * modify rcvctrl shadow[s] and write to appropriate chip-regs.
752 	 * see above QIB_RCVCTRL_xxx_ENB/DIS for operations.
753 	 * (ctxt == -1) means "all contexts", only meaningful for
754 	 * clearing. Could remove if chip_spec shutdown properly done.
755 	 */
756 	void (*f_rcvctrl)(struct qib_pportdata *, unsigned int op,
757 		int ctxt);
758 	/* Read/modify/write sendctrl appropriately for op and port. */
759 	void (*f_sendctrl)(struct qib_pportdata *, u32 op);
760 	void (*f_set_intr_state)(struct qib_devdata *, u32);
761 	void (*f_set_armlaunch)(struct qib_devdata *, u32);
762 	void (*f_wantpiobuf_intr)(struct qib_devdata *, u32);
763 	int (*f_late_initreg)(struct qib_devdata *);
764 	int (*f_init_sdma_regs)(struct qib_pportdata *);
765 	u16 (*f_sdma_gethead)(struct qib_pportdata *);
766 	int (*f_sdma_busy)(struct qib_pportdata *);
767 	void (*f_sdma_update_tail)(struct qib_pportdata *, u16);
768 	void (*f_sdma_set_desc_cnt)(struct qib_pportdata *, unsigned);
769 	void (*f_sdma_sendctrl)(struct qib_pportdata *, unsigned);
770 	void (*f_sdma_hw_clean_up)(struct qib_pportdata *);
771 	void (*f_sdma_hw_start_up)(struct qib_pportdata *);
772 	void (*f_sdma_init_early)(struct qib_pportdata *);
773 	void (*f_set_cntr_sample)(struct qib_pportdata *, u32, u32);
774 	void (*f_update_usrhead)(struct qib_ctxtdata *, u64, u32, u32, u32);
775 	u32 (*f_hdrqempty)(struct qib_ctxtdata *);
776 	u64 (*f_portcntr)(struct qib_pportdata *, u32);
777 	u32 (*f_read_cntrs)(struct qib_devdata *, loff_t, char **,
778 		u64 **);
779 	u32 (*f_read_portcntrs)(struct qib_devdata *, loff_t, u32,
780 		char **, u64 **);
781 	u32 (*f_setpbc_control)(struct qib_pportdata *, u32, u8, u8);
782 	void (*f_initvl15_bufs)(struct qib_devdata *);
783 	void (*f_init_ctxt)(struct qib_ctxtdata *);
784 	void (*f_txchk_change)(struct qib_devdata *, u32, u32, u32,
785 		struct qib_ctxtdata *);
786 	void (*f_writescratch)(struct qib_devdata *, u32);
787 	int (*f_tempsense_rd)(struct qib_devdata *, int regnum);
788 
789 	char *boardname; /* human readable board info */
790 
791 	/* template for writing TIDs  */
792 	u64 tidtemplate;
793 	/* value to write to free TIDs */
794 	u64 tidinvalid;
795 
796 	/* number of registers used for pioavail */
797 	u32 pioavregs;
798 	/* device (not port) flags, basically device capabilities */
799 	u32 flags;
800 	/* last buffer for user use */
801 	u32 lastctxt_piobuf;
802 
803 	/* saturating counter of (non-port-specific) device interrupts */
804 	u32 int_counter;
805 
806 	/* pio bufs allocated per ctxt */
807 	u32 pbufsctxt;
808 	/* if remainder on bufs/ctxt, ctxts < extrabuf get 1 extra */
809 	u32 ctxts_extrabuf;
810 	/*
811 	 * number of ctxts configured as max; zero is set to number chip
812 	 * supports, less gives more pio bufs/ctxt, etc.
813 	 */
814 	u32 cfgctxts;
815 	/*
816 	 * number of ctxts available for PSM open
817 	 */
818 	u32 freectxts;
819 
820 	/*
821 	 * hint that we should update pioavailshadow before
822 	 * looking for a PIO buffer
823 	 */
824 	u32 upd_pio_shadow;
825 
826 	/* internal debugging stats */
827 	u32 maxpkts_call;
828 	u32 avgpkts_call;
829 	u64 nopiobufs;
830 
831 	/* PCI Vendor ID (here for NodeInfo) */
832 	u16 vendorid;
833 	/* PCI Device ID (here for NodeInfo) */
834 	u16 deviceid;
835 	/* for write combining settings */
836 	unsigned long wc_cookie;
837 	unsigned long wc_base;
838 	unsigned long wc_len;
839 
840 	/* shadow copy of struct page *'s for exp tid pages */
841 	struct page **pageshadow;
842 	/* shadow copy of dma handles for exp tid pages */
843 	dma_addr_t *physshadow;
844 	u64 __iomem *egrtidbase;
845 	spinlock_t sendctrl_lock; /* protect changes to sendctrl shadow */
846 	/* around rcd and (user ctxts) ctxt_cnt use (intr vs free) */
847 	spinlock_t uctxt_lock; /* rcd and user context changes */
848 	/*
849 	 * per unit status, see also portdata statusp
850 	 * mapped readonly into user processes so they can get unit and
851 	 * IB link status cheaply
852 	 */
853 	u64 *devstatusp;
854 	char *freezemsg; /* freeze msg if hw error put chip in freeze */
855 	u32 freezelen; /* max length of freezemsg */
856 	/* timer used to prevent stats overflow, error throttling, etc. */
857 	struct timer_list stats_timer;
858 
859 	/* timer to verify interrupts work, and fallback if possible */
860 	struct timer_list intrchk_timer;
861 	unsigned long ureg_align; /* user register alignment */
862 
863 	/*
864 	 * Protects pioavailshadow, pioavailkernel, pio_need_disarm, and
865 	 * pio_writing.
866 	 */
867 	spinlock_t pioavail_lock;
868 
869 	/*
870 	 * Shadow copies of registers; size indicates read access size.
871 	 * Most of them are readonly, but some are write-only register,
872 	 * where we manipulate the bits in the shadow copy, and then write
873 	 * the shadow copy to qlogic_ib.
874 	 *
875 	 * We deliberately make most of these 32 bits, since they have
876 	 * restricted range.  For any that we read, we won't to generate 32
877 	 * bit accesses, since Opteron will generate 2 separate 32 bit HT
878 	 * transactions for a 64 bit read, and we want to avoid unnecessary
879 	 * bus transactions.
880 	 */
881 
882 	/* This is the 64 bit group */
883 
884 	unsigned long pioavailshadow[6];
885 	/* bitmap of send buffers available for the kernel to use with PIO. */
886 	unsigned long pioavailkernel[6];
887 	/* bitmap of send buffers which need to be disarmed. */
888 	unsigned long pio_need_disarm[3];
889 	/* bitmap of send buffers which are being written to. */
890 	unsigned long pio_writing[3];
891 	/* kr_revision shadow */
892 	u64 revision;
893 	/* Base GUID for device (from eeprom, network order) */
894 	__be64 base_guid;
895 
896 	/*
897 	 * kr_sendpiobufbase value (chip offset of pio buffers), and the
898 	 * base of the 2KB buffer s(user processes only use 2K)
899 	 */
900 	u64 piobufbase;
901 	u32 pio2k_bufbase;
902 
903 	/* these are the "32 bit" regs */
904 
905 	/* number of GUIDs in the flash for this interface */
906 	u32 nguid;
907 	/*
908 	 * the following two are 32-bit bitmasks, but {test,clear,set}_bit
909 	 * all expect bit fields to be "unsigned long"
910 	 */
911 	unsigned long rcvctrl; /* shadow per device rcvctrl */
912 	unsigned long sendctrl; /* shadow per device sendctrl */
913 
914 	/* value we put in kr_rcvhdrcnt */
915 	u32 rcvhdrcnt;
916 	/* value we put in kr_rcvhdrsize */
917 	u32 rcvhdrsize;
918 	/* value we put in kr_rcvhdrentsize */
919 	u32 rcvhdrentsize;
920 	/* kr_ctxtcnt value */
921 	u32 ctxtcnt;
922 	/* kr_pagealign value */
923 	u32 palign;
924 	/* number of "2KB" PIO buffers */
925 	u32 piobcnt2k;
926 	/* size in bytes of "2KB" PIO buffers */
927 	u32 piosize2k;
928 	/* max usable size in dwords of a "2KB" PIO buffer before going "4KB" */
929 	u32 piosize2kmax_dwords;
930 	/* number of "4KB" PIO buffers */
931 	u32 piobcnt4k;
932 	/* size in bytes of "4KB" PIO buffers */
933 	u32 piosize4k;
934 	/* kr_rcvegrbase value */
935 	u32 rcvegrbase;
936 	/* kr_rcvtidbase value */
937 	u32 rcvtidbase;
938 	/* kr_rcvtidcnt value */
939 	u32 rcvtidcnt;
940 	/* kr_userregbase */
941 	u32 uregbase;
942 	/* shadow the control register contents */
943 	u32 control;
944 
945 	/* chip address space used by 4k pio buffers */
946 	u32 align4k;
947 	/* size of each rcvegrbuffer */
948 	u16 rcvegrbufsize;
949 	/* log2 of above */
950 	u16 rcvegrbufsize_shift;
951 	/* localbus width (1, 2,4,8,16,32) from config space  */
952 	u32 lbus_width;
953 	/* localbus speed in MHz */
954 	u32 lbus_speed;
955 	int unit; /* unit # of this chip */
956 
957 	/* start of CHIP_SPEC move to chipspec, but need code changes */
958 	/* low and high portions of MSI capability/vector */
959 	u32 msi_lo;
960 	/* saved after PCIe init for restore after reset */
961 	u32 msi_hi;
962 	/* MSI data (vector) saved for restore */
963 	u16 msi_data;
964 	/* so we can rewrite it after a chip reset */
965 	u32 pcibar0;
966 	/* so we can rewrite it after a chip reset */
967 	u32 pcibar1;
968 	u64 rhdrhead_intr_off;
969 
970 	/*
971 	 * ASCII serial number, from flash, large enough for original
972 	 * all digit strings, and longer QLogic serial number format
973 	 */
974 	u8 serial[16];
975 	/* human readable board version */
976 	u8 boardversion[96];
977 	u8 lbus_info[32]; /* human readable localbus info */
978 	/* chip major rev, from qib_revision */
979 	u8 majrev;
980 	/* chip minor rev, from qib_revision */
981 	u8 minrev;
982 
983 	/* Misc small ints */
984 	/* Number of physical ports available */
985 	u8 num_pports;
986 	/* Lowest context number which can be used by user processes */
987 	u8 first_user_ctxt;
988 	u8 n_krcv_queues;
989 	u8 qpn_mask;
990 	u8 skip_kctxt_mask;
991 
992 	u16 rhf_offset; /* offset of RHF within receive header entry */
993 
994 	/*
995 	 * GPIO pins for twsi-connected devices, and device code for eeprom
996 	 */
997 	u8 gpio_sda_num;
998 	u8 gpio_scl_num;
999 	u8 twsi_eeprom_dev;
1000 	u8 board_atten;
1001 
1002 	/* Support (including locks) for EEPROM logging of errors and time */
1003 	/* control access to actual counters, timer */
1004 	spinlock_t eep_st_lock;
1005 	/* control high-level access to EEPROM */
1006 	struct mutex eep_lock;
1007 	uint64_t traffic_wds;
1008 	/* active time is kept in seconds, but logged in hours */
1009 	atomic_t active_time;
1010 	/* Below are nominal shadow of EEPROM, new since last EEPROM update */
1011 	uint8_t eep_st_errs[QIB_EEP_LOG_CNT];
1012 	uint8_t eep_st_new_errs[QIB_EEP_LOG_CNT];
1013 	uint16_t eep_hrs;
1014 	/*
1015 	 * masks for which bits of errs, hwerrs that cause
1016 	 * each of the counters to increment.
1017 	 */
1018 	struct qib_eep_log_mask eep_st_masks[QIB_EEP_LOG_CNT];
1019 	struct qib_diag_client *diag_client;
1020 	spinlock_t qib_diag_trans_lock; /* protect diag observer ops */
1021 	struct diag_observer_list_elt *diag_observer_list;
1022 
1023 	u8 psxmitwait_supported;
1024 	/* cycle length of PS* counters in HW (in picoseconds) */
1025 	u16 psxmitwait_check_rate;
1026 	/* high volume overflow errors defered to tasklet */
1027 	struct tasklet_struct error_tasklet;
1028 };
1029 
1030 /* hol_state values */
1031 #define QIB_HOL_UP       0
1032 #define QIB_HOL_INIT     1
1033 
1034 #define QIB_SDMA_SENDCTRL_OP_ENABLE    (1U << 0)
1035 #define QIB_SDMA_SENDCTRL_OP_INTENABLE (1U << 1)
1036 #define QIB_SDMA_SENDCTRL_OP_HALT      (1U << 2)
1037 #define QIB_SDMA_SENDCTRL_OP_CLEANUP   (1U << 3)
1038 #define QIB_SDMA_SENDCTRL_OP_DRAIN     (1U << 4)
1039 
1040 /* operation types for f_txchk_change() */
1041 #define TXCHK_CHG_TYPE_DIS1  3
1042 #define TXCHK_CHG_TYPE_ENAB1 2
1043 #define TXCHK_CHG_TYPE_KERN  1
1044 #define TXCHK_CHG_TYPE_USER  0
1045 
1046 #define QIB_CHASE_TIME msecs_to_jiffies(145)
1047 #define QIB_CHASE_DIS_TIME msecs_to_jiffies(160)
1048 
1049 /* Private data for file operations */
1050 struct qib_filedata {
1051 	struct qib_ctxtdata *rcd;
1052 	unsigned subctxt;
1053 	unsigned tidcursor;
1054 	struct qib_user_sdma_queue *pq;
1055 	int rec_cpu_num; /* for cpu affinity; -1 if none */
1056 };
1057 
1058 extern struct list_head qib_dev_list;
1059 extern spinlock_t qib_devs_lock;
1060 extern struct qib_devdata *qib_lookup(int unit);
1061 extern u32 qib_cpulist_count;
1062 extern unsigned long *qib_cpulist;
1063 
1064 extern unsigned qib_wc_pat;
1065 int qib_init(struct qib_devdata *, int);
1066 int init_chip_wc_pat(struct qib_devdata *dd, u32);
1067 int qib_enable_wc(struct qib_devdata *dd);
1068 void qib_disable_wc(struct qib_devdata *dd);
1069 int qib_count_units(int *npresentp, int *nupp);
1070 int qib_count_active_units(void);
1071 
1072 int qib_cdev_init(int minor, const char *name,
1073 		  const struct file_operations *fops,
1074 		  struct cdev **cdevp, struct device **devp);
1075 void qib_cdev_cleanup(struct cdev **cdevp, struct device **devp);
1076 int qib_dev_init(void);
1077 void qib_dev_cleanup(void);
1078 
1079 int qib_diag_add(struct qib_devdata *);
1080 void qib_diag_remove(struct qib_devdata *);
1081 void qib_handle_e_ibstatuschanged(struct qib_pportdata *, u64);
1082 void qib_sdma_update_tail(struct qib_pportdata *, u16); /* hold sdma_lock */
1083 
1084 int qib_decode_err(struct qib_devdata *dd, char *buf, size_t blen, u64 err);
1085 void qib_bad_intrstatus(struct qib_devdata *);
1086 void qib_handle_urcv(struct qib_devdata *, u64);
1087 
1088 /* clean up any per-chip chip-specific stuff */
1089 void qib_chip_cleanup(struct qib_devdata *);
1090 /* clean up any chip type-specific stuff */
1091 void qib_chip_done(void);
1092 
1093 /* check to see if we have to force ordering for write combining */
1094 int qib_unordered_wc(void);
1095 void qib_pio_copy(void __iomem *to, const void *from, size_t count);
1096 
1097 void qib_disarm_piobufs(struct qib_devdata *, unsigned, unsigned);
1098 int qib_disarm_piobufs_ifneeded(struct qib_ctxtdata *);
1099 void qib_disarm_piobufs_set(struct qib_devdata *, unsigned long *, unsigned);
1100 void qib_cancel_sends(struct qib_pportdata *);
1101 
1102 int qib_create_rcvhdrq(struct qib_devdata *, struct qib_ctxtdata *);
1103 int qib_setup_eagerbufs(struct qib_ctxtdata *);
1104 void qib_set_ctxtcnt(struct qib_devdata *);
1105 int qib_create_ctxts(struct qib_devdata *dd);
1106 struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *, u32);
1107 void qib_init_pportdata(struct qib_pportdata *, struct qib_devdata *, u8, u8);
1108 void qib_free_ctxtdata(struct qib_devdata *, struct qib_ctxtdata *);
1109 
1110 u32 qib_kreceive(struct qib_ctxtdata *, u32 *, u32 *);
1111 int qib_reset_device(int);
1112 int qib_wait_linkstate(struct qib_pportdata *, u32, int);
1113 int qib_set_linkstate(struct qib_pportdata *, u8);
1114 int qib_set_mtu(struct qib_pportdata *, u16);
1115 int qib_set_lid(struct qib_pportdata *, u32, u8);
1116 void qib_hol_down(struct qib_pportdata *);
1117 void qib_hol_init(struct qib_pportdata *);
1118 void qib_hol_up(struct qib_pportdata *);
1119 void qib_hol_event(unsigned long);
1120 void qib_disable_after_error(struct qib_devdata *);
1121 int qib_set_uevent_bits(struct qib_pportdata *, const int);
1122 
1123 /* for use in system calls, where we want to know device type, etc. */
1124 #define ctxt_fp(fp) \
1125 	(((struct qib_filedata *)(fp)->private_data)->rcd)
1126 #define subctxt_fp(fp) \
1127 	(((struct qib_filedata *)(fp)->private_data)->subctxt)
1128 #define tidcursor_fp(fp) \
1129 	(((struct qib_filedata *)(fp)->private_data)->tidcursor)
1130 #define user_sdma_queue_fp(fp) \
1131 	(((struct qib_filedata *)(fp)->private_data)->pq)
1132 
1133 static inline struct qib_devdata *dd_from_ppd(struct qib_pportdata *ppd)
1134 {
1135 	return ppd->dd;
1136 }
1137 
1138 static inline struct qib_devdata *dd_from_dev(struct qib_ibdev *dev)
1139 {
1140 	return container_of(dev, struct qib_devdata, verbs_dev);
1141 }
1142 
1143 static inline struct qib_devdata *dd_from_ibdev(struct ib_device *ibdev)
1144 {
1145 	return dd_from_dev(to_idev(ibdev));
1146 }
1147 
1148 static inline struct qib_pportdata *ppd_from_ibp(struct qib_ibport *ibp)
1149 {
1150 	return container_of(ibp, struct qib_pportdata, ibport_data);
1151 }
1152 
1153 static inline struct qib_ibport *to_iport(struct ib_device *ibdev, u8 port)
1154 {
1155 	struct qib_devdata *dd = dd_from_ibdev(ibdev);
1156 	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
1157 
1158 	WARN_ON(pidx >= dd->num_pports);
1159 	return &dd->pport[pidx].ibport_data;
1160 }
1161 
1162 /*
1163  * values for dd->flags (_device_ related flags) and
1164  */
1165 #define QIB_HAS_LINK_LATENCY  0x1 /* supports link latency (IB 1.2) */
1166 #define QIB_INITTED           0x2 /* chip and driver up and initted */
1167 #define QIB_DOING_RESET       0x4  /* in the middle of doing chip reset */
1168 #define QIB_PRESENT           0x8  /* chip accesses can be done */
1169 #define QIB_PIO_FLUSH_WC      0x10 /* Needs Write combining flush for PIO */
1170 #define QIB_HAS_THRESH_UPDATE 0x40
1171 #define QIB_HAS_SDMA_TIMEOUT  0x80
1172 #define QIB_USE_SPCL_TRIG     0x100 /* SpecialTrigger launch enabled */
1173 #define QIB_NODMA_RTAIL       0x200 /* rcvhdrtail register DMA enabled */
1174 #define QIB_HAS_INTX          0x800 /* Supports INTx interrupts */
1175 #define QIB_HAS_SEND_DMA      0x1000 /* Supports Send DMA */
1176 #define QIB_HAS_VLSUPP        0x2000 /* Supports multiple VLs; PBC different */
1177 #define QIB_HAS_HDRSUPP       0x4000 /* Supports header suppression */
1178 #define QIB_BADINTR           0x8000 /* severe interrupt problems */
1179 #define QIB_DCA_ENABLED       0x10000 /* Direct Cache Access enabled */
1180 #define QIB_HAS_QSFP          0x20000 /* device (card instance) has QSFP */
1181 
1182 /*
1183  * values for ppd->lflags (_ib_port_ related flags)
1184  */
1185 #define QIBL_LINKV             0x1 /* IB link state valid */
1186 #define QIBL_LINKDOWN          0x8 /* IB link is down */
1187 #define QIBL_LINKINIT          0x10 /* IB link level is up */
1188 #define QIBL_LINKARMED         0x20 /* IB link is ARMED */
1189 #define QIBL_LINKACTIVE        0x40 /* IB link is ACTIVE */
1190 /* leave a gap for more IB-link state */
1191 #define QIBL_IB_AUTONEG_INPROG 0x1000 /* non-IBTA DDR/QDR neg active */
1192 #define QIBL_IB_AUTONEG_FAILED 0x2000 /* non-IBTA DDR/QDR neg failed */
1193 #define QIBL_IB_LINK_DISABLED  0x4000 /* Linkdown-disable forced,
1194 				       * Do not try to bring up */
1195 #define QIBL_IB_FORCE_NOTIFY   0x8000 /* force notify on next ib change */
1196 
1197 /* IB dword length mask in PBC (lower 11 bits); same for all chips */
1198 #define QIB_PBC_LENGTH_MASK                     ((1 << 11) - 1)
1199 
1200 
1201 /* ctxt_flag bit offsets */
1202 		/* waiting for a packet to arrive */
1203 #define QIB_CTXT_WAITING_RCV   2
1204 		/* master has not finished initializing */
1205 #define QIB_CTXT_MASTER_UNINIT 4
1206 		/* waiting for an urgent packet to arrive */
1207 #define QIB_CTXT_WAITING_URG 5
1208 
1209 /* free up any allocated data at closes */
1210 void qib_free_data(struct qib_ctxtdata *dd);
1211 void qib_chg_pioavailkernel(struct qib_devdata *, unsigned, unsigned,
1212 			    u32, struct qib_ctxtdata *);
1213 struct qib_devdata *qib_init_iba7322_funcs(struct pci_dev *,
1214 					   const struct pci_device_id *);
1215 struct qib_devdata *qib_init_iba7220_funcs(struct pci_dev *,
1216 					   const struct pci_device_id *);
1217 struct qib_devdata *qib_init_iba6120_funcs(struct pci_dev *,
1218 					   const struct pci_device_id *);
1219 void qib_free_devdata(struct qib_devdata *);
1220 struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra);
1221 
1222 #define QIB_TWSI_NO_DEV 0xFF
1223 /* Below qib_twsi_ functions must be called with eep_lock held */
1224 int qib_twsi_reset(struct qib_devdata *dd);
1225 int qib_twsi_blk_rd(struct qib_devdata *dd, int dev, int addr, void *buffer,
1226 		    int len);
1227 int qib_twsi_blk_wr(struct qib_devdata *dd, int dev, int addr,
1228 		    const void *buffer, int len);
1229 void qib_get_eeprom_info(struct qib_devdata *);
1230 int qib_update_eeprom_log(struct qib_devdata *dd);
1231 void qib_inc_eeprom_err(struct qib_devdata *dd, u32 eidx, u32 incr);
1232 void qib_dump_lookup_output_queue(struct qib_devdata *);
1233 void qib_force_pio_avail_update(struct qib_devdata *);
1234 void qib_clear_symerror_on_linkup(unsigned long opaque);
1235 
1236 /*
1237  * Set LED override, only the two LSBs have "public" meaning, but
1238  * any non-zero value substitutes them for the Link and LinkTrain
1239  * LED states.
1240  */
1241 #define QIB_LED_PHYS 1 /* Physical (linktraining) GREEN LED */
1242 #define QIB_LED_LOG 2  /* Logical (link) YELLOW LED */
1243 void qib_set_led_override(struct qib_pportdata *ppd, unsigned int val);
1244 
1245 /* send dma routines */
1246 int qib_setup_sdma(struct qib_pportdata *);
1247 void qib_teardown_sdma(struct qib_pportdata *);
1248 void __qib_sdma_intr(struct qib_pportdata *);
1249 void qib_sdma_intr(struct qib_pportdata *);
1250 int qib_sdma_verbs_send(struct qib_pportdata *, struct qib_sge_state *,
1251 			u32, struct qib_verbs_txreq *);
1252 /* ppd->sdma_lock should be locked before calling this. */
1253 int qib_sdma_make_progress(struct qib_pportdata *dd);
1254 
1255 /* must be called under qib_sdma_lock */
1256 static inline u16 qib_sdma_descq_freecnt(const struct qib_pportdata *ppd)
1257 {
1258 	return ppd->sdma_descq_cnt -
1259 		(ppd->sdma_descq_added - ppd->sdma_descq_removed) - 1;
1260 }
1261 
1262 static inline int __qib_sdma_running(struct qib_pportdata *ppd)
1263 {
1264 	return ppd->sdma_state.current_state == qib_sdma_state_s99_running;
1265 }
1266 int qib_sdma_running(struct qib_pportdata *);
1267 
1268 void __qib_sdma_process_event(struct qib_pportdata *, enum qib_sdma_events);
1269 void qib_sdma_process_event(struct qib_pportdata *, enum qib_sdma_events);
1270 
1271 /*
1272  * number of words used for protocol header if not set by qib_userinit();
1273  */
1274 #define QIB_DFLT_RCVHDRSIZE 9
1275 
1276 /*
1277  * We need to be able to handle an IB header of at least 24 dwords.
1278  * We need the rcvhdrq large enough to handle largest IB header, but
1279  * still have room for a 2KB MTU standard IB packet.
1280  * Additionally, some processor/memory controller combinations
1281  * benefit quite strongly from having the DMA'ed data be cacheline
1282  * aligned and a cacheline multiple, so we set the size to 32 dwords
1283  * (2 64-byte primary cachelines for pretty much all processors of
1284  * interest).  The alignment hurts nothing, other than using somewhat
1285  * more memory.
1286  */
1287 #define QIB_RCVHDR_ENTSIZE 32
1288 
1289 int qib_get_user_pages(unsigned long, size_t, struct page **);
1290 void qib_release_user_pages(struct page **, size_t);
1291 int qib_eeprom_read(struct qib_devdata *, u8, void *, int);
1292 int qib_eeprom_write(struct qib_devdata *, u8, const void *, int);
1293 u32 __iomem *qib_getsendbuf_range(struct qib_devdata *, u32 *, u32, u32);
1294 void qib_sendbuf_done(struct qib_devdata *, unsigned);
1295 
1296 static inline void qib_clear_rcvhdrtail(const struct qib_ctxtdata *rcd)
1297 {
1298 	*((u64 *) rcd->rcvhdrtail_kvaddr) = 0ULL;
1299 }
1300 
1301 static inline u32 qib_get_rcvhdrtail(const struct qib_ctxtdata *rcd)
1302 {
1303 	/*
1304 	 * volatile because it's a DMA target from the chip, routine is
1305 	 * inlined, and don't want register caching or reordering.
1306 	 */
1307 	return (u32) le64_to_cpu(
1308 		*((volatile __le64 *)rcd->rcvhdrtail_kvaddr)); /* DMA'ed */
1309 }
1310 
1311 static inline u32 qib_get_hdrqtail(const struct qib_ctxtdata *rcd)
1312 {
1313 	const struct qib_devdata *dd = rcd->dd;
1314 	u32 hdrqtail;
1315 
1316 	if (dd->flags & QIB_NODMA_RTAIL) {
1317 		__le32 *rhf_addr;
1318 		u32 seq;
1319 
1320 		rhf_addr = (__le32 *) rcd->rcvhdrq +
1321 			rcd->head + dd->rhf_offset;
1322 		seq = qib_hdrget_seq(rhf_addr);
1323 		hdrqtail = rcd->head;
1324 		if (seq == rcd->seq_cnt)
1325 			hdrqtail++;
1326 	} else
1327 		hdrqtail = qib_get_rcvhdrtail(rcd);
1328 
1329 	return hdrqtail;
1330 }
1331 
1332 /*
1333  * sysfs interface.
1334  */
1335 
1336 extern const char ib_qib_version[];
1337 
1338 int qib_device_create(struct qib_devdata *);
1339 void qib_device_remove(struct qib_devdata *);
1340 
1341 int qib_create_port_files(struct ib_device *ibdev, u8 port_num,
1342 			  struct kobject *kobj);
1343 int qib_verbs_register_sysfs(struct qib_devdata *);
1344 void qib_verbs_unregister_sysfs(struct qib_devdata *);
1345 /* Hook for sysfs read of QSFP */
1346 extern int qib_qsfp_dump(struct qib_pportdata *ppd, char *buf, int len);
1347 
1348 int __init qib_init_qibfs(void);
1349 int __exit qib_exit_qibfs(void);
1350 
1351 int qibfs_add(struct qib_devdata *);
1352 int qibfs_remove(struct qib_devdata *);
1353 
1354 int qib_pcie_init(struct pci_dev *, const struct pci_device_id *);
1355 int qib_pcie_ddinit(struct qib_devdata *, struct pci_dev *,
1356 		    const struct pci_device_id *);
1357 void qib_pcie_ddcleanup(struct qib_devdata *);
1358 int qib_pcie_params(struct qib_devdata *, u32, u32 *, struct msix_entry *);
1359 int qib_reinit_intr(struct qib_devdata *);
1360 void qib_enable_intx(struct pci_dev *);
1361 void qib_nomsi(struct qib_devdata *);
1362 void qib_nomsix(struct qib_devdata *);
1363 void qib_pcie_getcmd(struct qib_devdata *, u16 *, u8 *, u8 *);
1364 void qib_pcie_reenable(struct qib_devdata *, u16, u8, u8);
1365 
1366 /*
1367  * dma_addr wrappers - all 0's invalid for hw
1368  */
1369 dma_addr_t qib_map_page(struct pci_dev *, struct page *, unsigned long,
1370 			  size_t, int);
1371 const char *qib_get_unit_name(int unit);
1372 
1373 /*
1374  * Flush write combining store buffers (if present) and perform a write
1375  * barrier.
1376  */
1377 #if defined(CONFIG_X86_64)
1378 #define qib_flush_wc() asm volatile("sfence" : : : "memory")
1379 #else
1380 #define qib_flush_wc() wmb() /* no reorder around wc flush */
1381 #endif
1382 
1383 /* global module parameter variables */
1384 extern unsigned qib_ibmtu;
1385 extern ushort qib_cfgctxts;
1386 extern ushort qib_num_cfg_vls;
1387 extern ushort qib_mini_init; /* If set, do few (ideally 0) writes to chip */
1388 extern unsigned qib_n_krcv_queues;
1389 extern unsigned qib_sdma_fetch_arb;
1390 extern unsigned qib_compat_ddr_negotiate;
1391 extern int qib_special_trigger;
1392 
1393 extern struct mutex qib_mutex;
1394 
1395 /* Number of seconds before our card status check...  */
1396 #define STATUS_TIMEOUT 60
1397 
1398 #define QIB_DRV_NAME            "ib_qib"
1399 #define QIB_USER_MINOR_BASE     0
1400 #define QIB_TRACE_MINOR         127
1401 #define QIB_DIAGPKT_MINOR       128
1402 #define QIB_DIAG_MINOR_BASE     129
1403 #define QIB_NMINORS             255
1404 
1405 #define PCI_VENDOR_ID_PATHSCALE 0x1fc1
1406 #define PCI_VENDOR_ID_QLOGIC 0x1077
1407 #define PCI_DEVICE_ID_QLOGIC_IB_6120 0x10
1408 #define PCI_DEVICE_ID_QLOGIC_IB_7220 0x7220
1409 #define PCI_DEVICE_ID_QLOGIC_IB_7322 0x7322
1410 
1411 /*
1412  * qib_early_err is used (only!) to print early errors before devdata is
1413  * allocated, or when dd->pcidev may not be valid, and at the tail end of
1414  * cleanup when devdata may have been freed, etc.  qib_dev_porterr is
1415  * the same as qib_dev_err, but is used when the message really needs
1416  * the IB port# to be definitive as to what's happening..
1417  * All of these go to the trace log, and the trace log entry is done
1418  * first to avoid possible serial port delays from printk.
1419  */
1420 #define qib_early_err(dev, fmt, ...) \
1421 	do { \
1422 		dev_err(dev, fmt, ##__VA_ARGS__); \
1423 	} while (0)
1424 
1425 #define qib_dev_err(dd, fmt, ...) \
1426 	do { \
1427 		dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
1428 			qib_get_unit_name((dd)->unit), ##__VA_ARGS__); \
1429 	} while (0)
1430 
1431 #define qib_dev_porterr(dd, port, fmt, ...) \
1432 	do { \
1433 		dev_err(&(dd)->pcidev->dev, "%s: IB%u:%u " fmt, \
1434 			qib_get_unit_name((dd)->unit), (dd)->unit, (port), \
1435 			##__VA_ARGS__); \
1436 	} while (0)
1437 
1438 #define qib_devinfo(pcidev, fmt, ...) \
1439 	do { \
1440 		dev_info(&(pcidev)->dev, fmt, ##__VA_ARGS__); \
1441 	} while (0)
1442 
1443 /*
1444  * this is used for formatting hw error messages...
1445  */
1446 struct qib_hwerror_msgs {
1447 	u64 mask;
1448 	const char *msg;
1449 	size_t sz;
1450 };
1451 
1452 #define QLOGIC_IB_HWE_MSG(a, b) { .mask = a, .msg = b }
1453 
1454 /* in qib_intr.c... */
1455 void qib_format_hwerrors(u64 hwerrs,
1456 			 const struct qib_hwerror_msgs *hwerrmsgs,
1457 			 size_t nhwerrmsgs, char *msg, size_t lmsg);
1458 #endif                          /* _QIB_KERNEL_H */
1459