xref: /openbmc/linux/drivers/infiniband/hw/qib/qib.h (revision 12eb4683)
1 #ifndef _QIB_KERNEL_H
2 #define _QIB_KERNEL_H
3 /*
4  * Copyright (c) 2012, 2013 Intel Corporation.  All rights reserved.
5  * Copyright (c) 2006 - 2012 QLogic Corporation. 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 #include <linux/kthread.h>
55 
56 #include "qib_common.h"
57 #include "qib_verbs.h"
58 
59 /* only s/w major version of QLogic_IB we can handle */
60 #define QIB_CHIP_VERS_MAJ 2U
61 
62 /* don't care about this except printing */
63 #define QIB_CHIP_VERS_MIN 0U
64 
65 /* The Organization Unique Identifier (Mfg code), and its position in GUID */
66 #define QIB_OUI 0x001175
67 #define QIB_OUI_LSB 40
68 
69 /*
70  * per driver stats, either not device nor port-specific, or
71  * summed over all of the devices and ports.
72  * They are described by name via ipathfs filesystem, so layout
73  * and number of elements can change without breaking compatibility.
74  * If members are added or deleted qib_statnames[] in qib_fs.c must
75  * change to match.
76  */
77 struct qlogic_ib_stats {
78 	__u64 sps_ints; /* number of interrupts handled */
79 	__u64 sps_errints; /* number of error interrupts */
80 	__u64 sps_txerrs; /* tx-related packet errors */
81 	__u64 sps_rcverrs; /* non-crc rcv packet errors */
82 	__u64 sps_hwerrs; /* hardware errors reported (parity, etc.) */
83 	__u64 sps_nopiobufs; /* no pio bufs avail from kernel */
84 	__u64 sps_ctxts; /* number of contexts currently open */
85 	__u64 sps_lenerrs; /* number of kernel packets where RHF != LRH len */
86 	__u64 sps_buffull;
87 	__u64 sps_hdrfull;
88 };
89 
90 extern struct qlogic_ib_stats qib_stats;
91 extern const struct pci_error_handlers qib_pci_err_handler;
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 
118 #ifdef CONFIG_DEBUG_FS
119 struct qib_opcode_stats_perctx;
120 #endif
121 
122 struct qib_ctxtdata {
123 	void **rcvegrbuf;
124 	dma_addr_t *rcvegrbuf_phys;
125 	/* rcvhdrq base, needs mmap before useful */
126 	void *rcvhdrq;
127 	/* kernel virtual address where hdrqtail is updated */
128 	void *rcvhdrtail_kvaddr;
129 	/*
130 	 * temp buffer for expected send setup, allocated at open, instead
131 	 * of each setup call
132 	 */
133 	void *tid_pg_list;
134 	/*
135 	 * Shared page for kernel to signal user processes that send buffers
136 	 * need disarming.  The process should call QIB_CMD_DISARM_BUFS
137 	 * or QIB_CMD_ACK_EVENT with IPATH_EVENT_DISARM_BUFS set.
138 	 */
139 	unsigned long *user_event_mask;
140 	/* when waiting for rcv or pioavail */
141 	wait_queue_head_t wait;
142 	/*
143 	 * rcvegr bufs base, physical, must fit
144 	 * in 44 bits so 32 bit programs mmap64 44 bit works)
145 	 */
146 	dma_addr_t rcvegr_phys;
147 	/* mmap of hdrq, must fit in 44 bits */
148 	dma_addr_t rcvhdrq_phys;
149 	dma_addr_t rcvhdrqtailaddr_phys;
150 
151 	/*
152 	 * number of opens (including slave sub-contexts) on this instance
153 	 * (ignoring forks, dup, etc. for now)
154 	 */
155 	int cnt;
156 	/*
157 	 * how much space to leave at start of eager TID entries for
158 	 * protocol use, on each TID
159 	 */
160 	/* instead of calculating it */
161 	unsigned ctxt;
162 	/* local node of context */
163 	int node_id;
164 	/* non-zero if ctxt is being shared. */
165 	u16 subctxt_cnt;
166 	/* non-zero if ctxt is being shared. */
167 	u16 subctxt_id;
168 	/* number of eager TID entries. */
169 	u16 rcvegrcnt;
170 	/* index of first eager TID entry. */
171 	u16 rcvegr_tid_base;
172 	/* number of pio bufs for this ctxt (all procs, if shared) */
173 	u32 piocnt;
174 	/* first pio buffer for this ctxt */
175 	u32 pio_base;
176 	/* chip offset of PIO buffers for this ctxt */
177 	u32 piobufs;
178 	/* how many alloc_pages() chunks in rcvegrbuf_pages */
179 	u32 rcvegrbuf_chunks;
180 	/* how many egrbufs per chunk */
181 	u16 rcvegrbufs_perchunk;
182 	/* ilog2 of above */
183 	u16 rcvegrbufs_perchunk_shift;
184 	/* order for rcvegrbuf_pages */
185 	size_t rcvegrbuf_size;
186 	/* rcvhdrq size (for freeing) */
187 	size_t rcvhdrq_size;
188 	/* per-context flags for fileops/intr communication */
189 	unsigned long flag;
190 	/* next expected TID to check when looking for free */
191 	u32 tidcursor;
192 	/* WAIT_RCV that timed out, no interrupt */
193 	u32 rcvwait_to;
194 	/* WAIT_PIO that timed out, no interrupt */
195 	u32 piowait_to;
196 	/* WAIT_RCV already happened, no wait */
197 	u32 rcvnowait;
198 	/* WAIT_PIO already happened, no wait */
199 	u32 pionowait;
200 	/* total number of polled urgent packets */
201 	u32 urgent;
202 	/* saved total number of polled urgent packets for poll edge trigger */
203 	u32 urgent_poll;
204 	/* pid of process using this ctxt */
205 	pid_t pid;
206 	pid_t subpid[QLOGIC_IB_MAX_SUBCTXT];
207 	/* same size as task_struct .comm[], command that opened context */
208 	char comm[16];
209 	/* pkeys set by this use of this ctxt */
210 	u16 pkeys[4];
211 	/* so file ops can get at unit */
212 	struct qib_devdata *dd;
213 	/* so funcs that need physical port can get it easily */
214 	struct qib_pportdata *ppd;
215 	/* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
216 	void *subctxt_uregbase;
217 	/* An array of pages for the eager receive buffers * N */
218 	void *subctxt_rcvegrbuf;
219 	/* An array of pages for the eager header queue entries * N */
220 	void *subctxt_rcvhdr_base;
221 	/* The version of the library which opened this ctxt */
222 	u32 userversion;
223 	/* Bitmask of active slaves */
224 	u32 active_slaves;
225 	/* Type of packets or conditions we want to poll for */
226 	u16 poll_type;
227 	/* receive packet sequence counter */
228 	u8 seq_cnt;
229 	u8 redirect_seq_cnt;
230 	/* ctxt rcvhdrq head offset */
231 	u32 head;
232 	/* lookaside fields */
233 	struct qib_qp *lookaside_qp;
234 	u32 lookaside_qpn;
235 	/* QPs waiting for context processing */
236 	struct list_head qp_wait_list;
237 #ifdef CONFIG_DEBUG_FS
238 	/* verbs stats per CTX */
239 	struct qib_opcode_stats_perctx *opstats;
240 #endif
241 };
242 
243 struct qib_sge_state;
244 
245 struct qib_sdma_txreq {
246 	int                 flags;
247 	int                 sg_count;
248 	dma_addr_t          addr;
249 	void              (*callback)(struct qib_sdma_txreq *, int);
250 	u16                 start_idx;  /* sdma private */
251 	u16                 next_descq_idx;  /* sdma private */
252 	struct list_head    list;       /* sdma private */
253 };
254 
255 struct qib_sdma_desc {
256 	__le64 qw[2];
257 };
258 
259 struct qib_verbs_txreq {
260 	struct qib_sdma_txreq   txreq;
261 	struct qib_qp           *qp;
262 	struct qib_swqe         *wqe;
263 	u32                     dwords;
264 	u16                     hdr_dwords;
265 	u16                     hdr_inx;
266 	struct qib_pio_header	*align_buf;
267 	struct qib_mregion	*mr;
268 	struct qib_sge_state    *ss;
269 };
270 
271 #define QIB_SDMA_TXREQ_F_USELARGEBUF  0x1
272 #define QIB_SDMA_TXREQ_F_HEADTOHOST   0x2
273 #define QIB_SDMA_TXREQ_F_INTREQ       0x4
274 #define QIB_SDMA_TXREQ_F_FREEBUF      0x8
275 #define QIB_SDMA_TXREQ_F_FREEDESC     0x10
276 
277 #define QIB_SDMA_TXREQ_S_OK        0
278 #define QIB_SDMA_TXREQ_S_SENDERROR 1
279 #define QIB_SDMA_TXREQ_S_ABORTED   2
280 #define QIB_SDMA_TXREQ_S_SHUTDOWN  3
281 
282 /*
283  * Get/Set IB link-level config parameters for f_get/set_ib_cfg()
284  * Mostly for MADs that set or query link parameters, also ipath
285  * config interfaces
286  */
287 #define QIB_IB_CFG_LIDLMC 0 /* LID (LS16b) and Mask (MS16b) */
288 #define QIB_IB_CFG_LWID_ENB 2 /* allowed Link-width */
289 #define QIB_IB_CFG_LWID 3 /* currently active Link-width */
290 #define QIB_IB_CFG_SPD_ENB 4 /* allowed Link speeds */
291 #define QIB_IB_CFG_SPD 5 /* current Link spd */
292 #define QIB_IB_CFG_RXPOL_ENB 6 /* Auto-RX-polarity enable */
293 #define QIB_IB_CFG_LREV_ENB 7 /* Auto-Lane-reversal enable */
294 #define QIB_IB_CFG_LINKLATENCY 8 /* Link Latency (IB1.2 only) */
295 #define QIB_IB_CFG_HRTBT 9 /* IB heartbeat off/enable/auto; DDR/QDR only */
296 #define QIB_IB_CFG_OP_VLS 10 /* operational VLs */
297 #define QIB_IB_CFG_VL_HIGH_CAP 11 /* num of VL high priority weights */
298 #define QIB_IB_CFG_VL_LOW_CAP 12 /* num of VL low priority weights */
299 #define QIB_IB_CFG_OVERRUN_THRESH 13 /* IB overrun threshold */
300 #define QIB_IB_CFG_PHYERR_THRESH 14 /* IB PHY error threshold */
301 #define QIB_IB_CFG_LINKDEFAULT 15 /* IB link default (sleep/poll) */
302 #define QIB_IB_CFG_PKEYS 16 /* update partition keys */
303 #define QIB_IB_CFG_MTU 17 /* update MTU in IBC */
304 #define QIB_IB_CFG_LSTATE 18 /* update linkcmd and linkinitcmd in IBC */
305 #define QIB_IB_CFG_VL_HIGH_LIMIT 19
306 #define QIB_IB_CFG_PMA_TICKS 20 /* PMA sample tick resolution */
307 #define QIB_IB_CFG_PORT 21 /* switch port we are connected to */
308 
309 /*
310  * for CFG_LSTATE: LINKCMD in upper 16 bits, LINKINITCMD in lower 16
311  * IB_LINKINITCMD_POLL and SLEEP are also used as set/get values for
312  * QIB_IB_CFG_LINKDEFAULT cmd
313  */
314 #define   IB_LINKCMD_DOWN   (0 << 16)
315 #define   IB_LINKCMD_ARMED  (1 << 16)
316 #define   IB_LINKCMD_ACTIVE (2 << 16)
317 #define   IB_LINKINITCMD_NOP     0
318 #define   IB_LINKINITCMD_POLL    1
319 #define   IB_LINKINITCMD_SLEEP   2
320 #define   IB_LINKINITCMD_DISABLE 3
321 
322 /*
323  * valid states passed to qib_set_linkstate() user call
324  */
325 #define QIB_IB_LINKDOWN         0
326 #define QIB_IB_LINKARM          1
327 #define QIB_IB_LINKACTIVE       2
328 #define QIB_IB_LINKDOWN_ONLY    3
329 #define QIB_IB_LINKDOWN_SLEEP   4
330 #define QIB_IB_LINKDOWN_DISABLE 5
331 
332 /*
333  * These 7 values (SDR, DDR, and QDR may be ORed for auto-speed
334  * negotiation) are used for the 3rd argument to path_f_set_ib_cfg
335  * with cmd QIB_IB_CFG_SPD_ENB, by direct calls or via sysfs.  They
336  * are also the the possible values for qib_link_speed_enabled and active
337  * The values were chosen to match values used within the IB spec.
338  */
339 #define QIB_IB_SDR 1
340 #define QIB_IB_DDR 2
341 #define QIB_IB_QDR 4
342 
343 #define QIB_DEFAULT_MTU 4096
344 
345 /* max number of IB ports supported per HCA */
346 #define QIB_MAX_IB_PORTS 2
347 
348 /*
349  * Possible IB config parameters for f_get/set_ib_table()
350  */
351 #define QIB_IB_TBL_VL_HIGH_ARB 1 /* Get/set VL high priority weights */
352 #define QIB_IB_TBL_VL_LOW_ARB 2 /* Get/set VL low priority weights */
353 
354 /*
355  * Possible "operations" for f_rcvctrl(ppd, op, ctxt)
356  * these are bits so they can be combined, e.g.
357  * QIB_RCVCTRL_INTRAVAIL_ENB | QIB_RCVCTRL_CTXT_ENB
358  */
359 #define QIB_RCVCTRL_TAILUPD_ENB 0x01
360 #define QIB_RCVCTRL_TAILUPD_DIS 0x02
361 #define QIB_RCVCTRL_CTXT_ENB 0x04
362 #define QIB_RCVCTRL_CTXT_DIS 0x08
363 #define QIB_RCVCTRL_INTRAVAIL_ENB 0x10
364 #define QIB_RCVCTRL_INTRAVAIL_DIS 0x20
365 #define QIB_RCVCTRL_PKEY_ENB 0x40  /* Note, default is enabled */
366 #define QIB_RCVCTRL_PKEY_DIS 0x80
367 #define QIB_RCVCTRL_BP_ENB 0x0100
368 #define QIB_RCVCTRL_BP_DIS 0x0200
369 #define QIB_RCVCTRL_TIDFLOW_ENB 0x0400
370 #define QIB_RCVCTRL_TIDFLOW_DIS 0x0800
371 
372 /*
373  * Possible "operations" for f_sendctrl(ppd, op, var)
374  * these are bits so they can be combined, e.g.
375  * QIB_SENDCTRL_BUFAVAIL_ENB | QIB_SENDCTRL_ENB
376  * Some operations (e.g. DISARM, ABORT) are known to
377  * be "one-shot", so do not modify shadow.
378  */
379 #define QIB_SENDCTRL_DISARM       (0x1000)
380 #define QIB_SENDCTRL_DISARM_BUF(bufn) ((bufn) | QIB_SENDCTRL_DISARM)
381 	/* available (0x2000) */
382 #define QIB_SENDCTRL_AVAIL_DIS    (0x4000)
383 #define QIB_SENDCTRL_AVAIL_ENB    (0x8000)
384 #define QIB_SENDCTRL_AVAIL_BLIP  (0x10000)
385 #define QIB_SENDCTRL_SEND_DIS    (0x20000)
386 #define QIB_SENDCTRL_SEND_ENB    (0x40000)
387 #define QIB_SENDCTRL_FLUSH       (0x80000)
388 #define QIB_SENDCTRL_CLEAR      (0x100000)
389 #define QIB_SENDCTRL_DISARM_ALL (0x200000)
390 
391 /*
392  * These are the generic indices for requesting per-port
393  * counter values via the f_portcntr function.  They
394  * are always returned as 64 bit values, although most
395  * are 32 bit counters.
396  */
397 /* send-related counters */
398 #define QIBPORTCNTR_PKTSEND         0U
399 #define QIBPORTCNTR_WORDSEND        1U
400 #define QIBPORTCNTR_PSXMITDATA      2U
401 #define QIBPORTCNTR_PSXMITPKTS      3U
402 #define QIBPORTCNTR_PSXMITWAIT      4U
403 #define QIBPORTCNTR_SENDSTALL       5U
404 /* receive-related counters */
405 #define QIBPORTCNTR_PKTRCV          6U
406 #define QIBPORTCNTR_PSRCVDATA       7U
407 #define QIBPORTCNTR_PSRCVPKTS       8U
408 #define QIBPORTCNTR_RCVEBP          9U
409 #define QIBPORTCNTR_RCVOVFL         10U
410 #define QIBPORTCNTR_WORDRCV         11U
411 /* IB link related error counters */
412 #define QIBPORTCNTR_RXLOCALPHYERR   12U
413 #define QIBPORTCNTR_RXVLERR         13U
414 #define QIBPORTCNTR_ERRICRC         14U
415 #define QIBPORTCNTR_ERRVCRC         15U
416 #define QIBPORTCNTR_ERRLPCRC        16U
417 #define QIBPORTCNTR_BADFORMAT       17U
418 #define QIBPORTCNTR_ERR_RLEN        18U
419 #define QIBPORTCNTR_IBSYMBOLERR     19U
420 #define QIBPORTCNTR_INVALIDRLEN     20U
421 #define QIBPORTCNTR_UNSUPVL         21U
422 #define QIBPORTCNTR_EXCESSBUFOVFL   22U
423 #define QIBPORTCNTR_ERRLINK         23U
424 #define QIBPORTCNTR_IBLINKDOWN      24U
425 #define QIBPORTCNTR_IBLINKERRRECOV  25U
426 #define QIBPORTCNTR_LLI             26U
427 /* other error counters */
428 #define QIBPORTCNTR_RXDROPPKT       27U
429 #define QIBPORTCNTR_VL15PKTDROP     28U
430 #define QIBPORTCNTR_ERRPKEY         29U
431 #define QIBPORTCNTR_KHDROVFL        30U
432 /* sampling counters (these are actually control registers) */
433 #define QIBPORTCNTR_PSINTERVAL      31U
434 #define QIBPORTCNTR_PSSTART         32U
435 #define QIBPORTCNTR_PSSTAT          33U
436 
437 /* how often we check for packet activity for "power on hours (in seconds) */
438 #define ACTIVITY_TIMER 5
439 
440 #define MAX_NAME_SIZE 64
441 
442 #ifdef CONFIG_INFINIBAND_QIB_DCA
443 struct qib_irq_notify;
444 #endif
445 
446 struct qib_msix_entry {
447 	struct msix_entry msix;
448 	void *arg;
449 #ifdef CONFIG_INFINIBAND_QIB_DCA
450 	int dca;
451 	int rcv;
452 	struct qib_irq_notify *notifier;
453 #endif
454 	char name[MAX_NAME_SIZE];
455 	cpumask_var_t mask;
456 };
457 
458 /* Below is an opaque struct. Each chip (device) can maintain
459  * private data needed for its operation, but not germane to the
460  * rest of the driver.  For convenience, we define another that
461  * is chip-specific, per-port
462  */
463 struct qib_chip_specific;
464 struct qib_chipport_specific;
465 
466 enum qib_sdma_states {
467 	qib_sdma_state_s00_hw_down,
468 	qib_sdma_state_s10_hw_start_up_wait,
469 	qib_sdma_state_s20_idle,
470 	qib_sdma_state_s30_sw_clean_up_wait,
471 	qib_sdma_state_s40_hw_clean_up_wait,
472 	qib_sdma_state_s50_hw_halt_wait,
473 	qib_sdma_state_s99_running,
474 };
475 
476 enum qib_sdma_events {
477 	qib_sdma_event_e00_go_hw_down,
478 	qib_sdma_event_e10_go_hw_start,
479 	qib_sdma_event_e20_hw_started,
480 	qib_sdma_event_e30_go_running,
481 	qib_sdma_event_e40_sw_cleaned,
482 	qib_sdma_event_e50_hw_cleaned,
483 	qib_sdma_event_e60_hw_halted,
484 	qib_sdma_event_e70_go_idle,
485 	qib_sdma_event_e7220_err_halted,
486 	qib_sdma_event_e7322_err_halted,
487 	qib_sdma_event_e90_timer_tick,
488 };
489 
490 extern char *qib_sdma_state_names[];
491 extern char *qib_sdma_event_names[];
492 
493 struct sdma_set_state_action {
494 	unsigned op_enable:1;
495 	unsigned op_intenable:1;
496 	unsigned op_halt:1;
497 	unsigned op_drain:1;
498 	unsigned go_s99_running_tofalse:1;
499 	unsigned go_s99_running_totrue:1;
500 };
501 
502 struct qib_sdma_state {
503 	struct kref          kref;
504 	struct completion    comp;
505 	enum qib_sdma_states current_state;
506 	struct sdma_set_state_action *set_state_action;
507 	unsigned             current_op;
508 	unsigned             go_s99_running;
509 	unsigned             first_sendbuf;
510 	unsigned             last_sendbuf; /* really last +1 */
511 	/* debugging/devel */
512 	enum qib_sdma_states previous_state;
513 	unsigned             previous_op;
514 	enum qib_sdma_events last_event;
515 };
516 
517 struct xmit_wait {
518 	struct timer_list timer;
519 	u64 counter;
520 	u8 flags;
521 	struct cache {
522 		u64 psxmitdata;
523 		u64 psrcvdata;
524 		u64 psxmitpkts;
525 		u64 psrcvpkts;
526 		u64 psxmitwait;
527 	} counter_cache;
528 };
529 
530 /*
531  * The structure below encapsulates data relevant to a physical IB Port.
532  * Current chips support only one such port, but the separation
533  * clarifies things a bit. Note that to conform to IB conventions,
534  * port-numbers are one-based. The first or only port is port1.
535  */
536 struct qib_pportdata {
537 	struct qib_ibport ibport_data;
538 
539 	struct qib_devdata *dd;
540 	struct qib_chippport_specific *cpspec; /* chip-specific per-port */
541 	struct kobject pport_kobj;
542 	struct kobject pport_cc_kobj;
543 	struct kobject sl2vl_kobj;
544 	struct kobject diagc_kobj;
545 
546 	/* GUID for this interface, in network order */
547 	__be64 guid;
548 
549 	/* QIB_POLL, etc. link-state specific flags, per port */
550 	u32 lflags;
551 	/* qib_lflags driver is waiting for */
552 	u32 state_wanted;
553 	spinlock_t lflags_lock;
554 
555 	/* ref count for each pkey */
556 	atomic_t pkeyrefs[4];
557 
558 	/*
559 	 * this address is mapped readonly into user processes so they can
560 	 * get status cheaply, whenever they want.  One qword of status per port
561 	 */
562 	u64 *statusp;
563 
564 	/* SendDMA related entries */
565 
566 	/* read mostly */
567 	struct qib_sdma_desc *sdma_descq;
568 	struct workqueue_struct *qib_wq;
569 	struct qib_sdma_state sdma_state;
570 	dma_addr_t       sdma_descq_phys;
571 	volatile __le64 *sdma_head_dma; /* DMA'ed by chip */
572 	dma_addr_t       sdma_head_phys;
573 	u16                   sdma_descq_cnt;
574 
575 	/* read/write using lock */
576 	spinlock_t            sdma_lock ____cacheline_aligned_in_smp;
577 	struct list_head      sdma_activelist;
578 	struct list_head      sdma_userpending;
579 	u64                   sdma_descq_added;
580 	u64                   sdma_descq_removed;
581 	u16                   sdma_descq_tail;
582 	u16                   sdma_descq_head;
583 	u8                    sdma_generation;
584 	u8                    sdma_intrequest;
585 
586 	struct tasklet_struct sdma_sw_clean_up_task
587 		____cacheline_aligned_in_smp;
588 
589 	wait_queue_head_t state_wait; /* for state_wanted */
590 
591 	/* HoL blocking for SMP replies */
592 	unsigned          hol_state;
593 	struct timer_list hol_timer;
594 
595 	/*
596 	 * Shadow copies of registers; size indicates read access size.
597 	 * Most of them are readonly, but some are write-only register,
598 	 * where we manipulate the bits in the shadow copy, and then write
599 	 * the shadow copy to qlogic_ib.
600 	 *
601 	 * We deliberately make most of these 32 bits, since they have
602 	 * restricted range.  For any that we read, we won't to generate 32
603 	 * bit accesses, since Opteron will generate 2 separate 32 bit HT
604 	 * transactions for a 64 bit read, and we want to avoid unnecessary
605 	 * bus transactions.
606 	 */
607 
608 	/* This is the 64 bit group */
609 	/* last ibcstatus.  opaque outside chip-specific code */
610 	u64 lastibcstat;
611 
612 	/* these are the "32 bit" regs */
613 
614 	/*
615 	 * the following two are 32-bit bitmasks, but {test,clear,set}_bit
616 	 * all expect bit fields to be "unsigned long"
617 	 */
618 	unsigned long p_rcvctrl; /* shadow per-port rcvctrl */
619 	unsigned long p_sendctrl; /* shadow per-port sendctrl */
620 
621 	u32 ibmtu; /* The MTU programmed for this unit */
622 	/*
623 	 * Current max size IB packet (in bytes) including IB headers, that
624 	 * we can send. Changes when ibmtu changes.
625 	 */
626 	u32 ibmaxlen;
627 	/*
628 	 * ibmaxlen at init time, limited by chip and by receive buffer
629 	 * size.  Not changed after init.
630 	 */
631 	u32 init_ibmaxlen;
632 	/* LID programmed for this instance */
633 	u16 lid;
634 	/* list of pkeys programmed; 0 if not set */
635 	u16 pkeys[4];
636 	/* LID mask control */
637 	u8 lmc;
638 	u8 link_width_supported;
639 	u8 link_speed_supported;
640 	u8 link_width_enabled;
641 	u8 link_speed_enabled;
642 	u8 link_width_active;
643 	u8 link_speed_active;
644 	u8 vls_supported;
645 	u8 vls_operational;
646 	/* Rx Polarity inversion (compensate for ~tx on partner) */
647 	u8 rx_pol_inv;
648 
649 	u8 hw_pidx;     /* physical port index */
650 	u8 port;        /* IB port number and index into dd->pports - 1 */
651 
652 	u8 delay_mult;
653 
654 	/* used to override LED behavior */
655 	u8 led_override;  /* Substituted for normal value, if non-zero */
656 	u16 led_override_timeoff; /* delta to next timer event */
657 	u8 led_override_vals[2]; /* Alternates per blink-frame */
658 	u8 led_override_phase; /* Just counts, LSB picks from vals[] */
659 	atomic_t led_override_timer_active;
660 	/* Used to flash LEDs in override mode */
661 	struct timer_list led_override_timer;
662 	struct xmit_wait cong_stats;
663 	struct timer_list symerr_clear_timer;
664 
665 	/* Synchronize access between driver writes and sysfs reads */
666 	spinlock_t cc_shadow_lock
667 		____cacheline_aligned_in_smp;
668 
669 	/* Shadow copy of the congestion control table */
670 	struct cc_table_shadow *ccti_entries_shadow;
671 
672 	/* Shadow copy of the congestion control entries */
673 	struct ib_cc_congestion_setting_attr_shadow *congestion_entries_shadow;
674 
675 	/* List of congestion control table entries */
676 	struct ib_cc_table_entry_shadow *ccti_entries;
677 
678 	/* 16 congestion entries with each entry corresponding to a SL */
679 	struct ib_cc_congestion_entry_shadow *congestion_entries;
680 
681 	/* Maximum number of congestion control entries that the agent expects
682 	 * the manager to send.
683 	 */
684 	u16 cc_supported_table_entries;
685 
686 	/* Total number of congestion control table entries */
687 	u16 total_cct_entry;
688 
689 	/* Bit map identifying service level */
690 	u16 cc_sl_control_map;
691 
692 	/* maximum congestion control table index */
693 	u16 ccti_limit;
694 
695 	/* CA's max number of 64 entry units in the congestion control table */
696 	u8 cc_max_table_entries;
697 };
698 
699 /* Observers. Not to be taken lightly, possibly not to ship. */
700 /*
701  * If a diag read or write is to (bottom <= offset <= top),
702  * the "hoook" is called, allowing, e.g. shadows to be
703  * updated in sync with the driver. struct diag_observer
704  * is the "visible" part.
705  */
706 struct diag_observer;
707 
708 typedef int (*diag_hook) (struct qib_devdata *dd,
709 	const struct diag_observer *op,
710 	u32 offs, u64 *data, u64 mask, int only_32);
711 
712 struct diag_observer {
713 	diag_hook hook;
714 	u32 bottom;
715 	u32 top;
716 };
717 
718 extern int qib_register_observer(struct qib_devdata *dd,
719 	const struct diag_observer *op);
720 
721 /* Only declared here, not defined. Private to diags */
722 struct diag_observer_list_elt;
723 
724 /* device data struct now contains only "general per-device" info.
725  * fields related to a physical IB port are in a qib_pportdata struct,
726  * described above) while fields only used by a particular chip-type are in
727  * a qib_chipdata struct, whose contents are opaque to this file.
728  */
729 struct qib_devdata {
730 	struct qib_ibdev verbs_dev;     /* must be first */
731 	struct list_head list;
732 	/* pointers to related structs for this device */
733 	/* pci access data structure */
734 	struct pci_dev *pcidev;
735 	struct cdev *user_cdev;
736 	struct cdev *diag_cdev;
737 	struct device *user_device;
738 	struct device *diag_device;
739 
740 	/* mem-mapped pointer to base of chip regs */
741 	u64 __iomem *kregbase;
742 	/* end of mem-mapped chip space excluding sendbuf and user regs */
743 	u64 __iomem *kregend;
744 	/* physical address of chip for io_remap, etc. */
745 	resource_size_t physaddr;
746 	/* qib_cfgctxts pointers */
747 	struct qib_ctxtdata **rcd; /* Receive Context Data */
748 
749 	/* qib_pportdata, points to array of (physical) port-specific
750 	 * data structs, indexed by pidx (0..n-1)
751 	 */
752 	struct qib_pportdata *pport;
753 	struct qib_chip_specific *cspec; /* chip-specific */
754 
755 	/* kvirt address of 1st 2k pio buffer */
756 	void __iomem *pio2kbase;
757 	/* kvirt address of 1st 4k pio buffer */
758 	void __iomem *pio4kbase;
759 	/* mem-mapped pointer to base of PIO buffers (if using WC PAT) */
760 	void __iomem *piobase;
761 	/* mem-mapped pointer to base of user chip regs (if using WC PAT) */
762 	u64 __iomem *userbase;
763 	void __iomem *piovl15base; /* base of VL15 buffers, if not WC */
764 	/*
765 	 * points to area where PIOavail registers will be DMA'ed.
766 	 * Has to be on a page of it's own, because the page will be
767 	 * mapped into user program space.  This copy is *ONLY* ever
768 	 * written by DMA, not by the driver!  Need a copy per device
769 	 * when we get to multiple devices
770 	 */
771 	volatile __le64 *pioavailregs_dma; /* DMA'ed by chip */
772 	/* physical address where updates occur */
773 	dma_addr_t pioavailregs_phys;
774 
775 	/* device-specific implementations of functions needed by
776 	 * common code. Contrary to previous consensus, we can't
777 	 * really just point to a device-specific table, because we
778 	 * may need to "bend", e.g. *_f_put_tid
779 	 */
780 	/* fallback to alternate interrupt type if possible */
781 	int (*f_intr_fallback)(struct qib_devdata *);
782 	/* hard reset chip */
783 	int (*f_reset)(struct qib_devdata *);
784 	void (*f_quiet_serdes)(struct qib_pportdata *);
785 	int (*f_bringup_serdes)(struct qib_pportdata *);
786 	int (*f_early_init)(struct qib_devdata *);
787 	void (*f_clear_tids)(struct qib_devdata *, struct qib_ctxtdata *);
788 	void (*f_put_tid)(struct qib_devdata *, u64 __iomem*,
789 				u32, unsigned long);
790 	void (*f_cleanup)(struct qib_devdata *);
791 	void (*f_setextled)(struct qib_pportdata *, u32);
792 	/* fill out chip-specific fields */
793 	int (*f_get_base_info)(struct qib_ctxtdata *, struct qib_base_info *);
794 	/* free irq */
795 	void (*f_free_irq)(struct qib_devdata *);
796 	struct qib_message_header *(*f_get_msgheader)
797 					(struct qib_devdata *, __le32 *);
798 	void (*f_config_ctxts)(struct qib_devdata *);
799 	int (*f_get_ib_cfg)(struct qib_pportdata *, int);
800 	int (*f_set_ib_cfg)(struct qib_pportdata *, int, u32);
801 	int (*f_set_ib_loopback)(struct qib_pportdata *, const char *);
802 	int (*f_get_ib_table)(struct qib_pportdata *, int, void *);
803 	int (*f_set_ib_table)(struct qib_pportdata *, int, void *);
804 	u32 (*f_iblink_state)(u64);
805 	u8 (*f_ibphys_portstate)(u64);
806 	void (*f_xgxs_reset)(struct qib_pportdata *);
807 	/* per chip actions needed for IB Link up/down changes */
808 	int (*f_ib_updown)(struct qib_pportdata *, int, u64);
809 	u32 __iomem *(*f_getsendbuf)(struct qib_pportdata *, u64, u32 *);
810 	/* Read/modify/write of GPIO pins (potentially chip-specific */
811 	int (*f_gpio_mod)(struct qib_devdata *dd, u32 out, u32 dir,
812 		u32 mask);
813 	/* Enable writes to config EEPROM (if supported) */
814 	int (*f_eeprom_wen)(struct qib_devdata *dd, int wen);
815 	/*
816 	 * modify rcvctrl shadow[s] and write to appropriate chip-regs.
817 	 * see above QIB_RCVCTRL_xxx_ENB/DIS for operations.
818 	 * (ctxt == -1) means "all contexts", only meaningful for
819 	 * clearing. Could remove if chip_spec shutdown properly done.
820 	 */
821 	void (*f_rcvctrl)(struct qib_pportdata *, unsigned int op,
822 		int ctxt);
823 	/* Read/modify/write sendctrl appropriately for op and port. */
824 	void (*f_sendctrl)(struct qib_pportdata *, u32 op);
825 	void (*f_set_intr_state)(struct qib_devdata *, u32);
826 	void (*f_set_armlaunch)(struct qib_devdata *, u32);
827 	void (*f_wantpiobuf_intr)(struct qib_devdata *, u32);
828 	int (*f_late_initreg)(struct qib_devdata *);
829 	int (*f_init_sdma_regs)(struct qib_pportdata *);
830 	u16 (*f_sdma_gethead)(struct qib_pportdata *);
831 	int (*f_sdma_busy)(struct qib_pportdata *);
832 	void (*f_sdma_update_tail)(struct qib_pportdata *, u16);
833 	void (*f_sdma_set_desc_cnt)(struct qib_pportdata *, unsigned);
834 	void (*f_sdma_sendctrl)(struct qib_pportdata *, unsigned);
835 	void (*f_sdma_hw_clean_up)(struct qib_pportdata *);
836 	void (*f_sdma_hw_start_up)(struct qib_pportdata *);
837 	void (*f_sdma_init_early)(struct qib_pportdata *);
838 	void (*f_set_cntr_sample)(struct qib_pportdata *, u32, u32);
839 	void (*f_update_usrhead)(struct qib_ctxtdata *, u64, u32, u32, u32);
840 	u32 (*f_hdrqempty)(struct qib_ctxtdata *);
841 	u64 (*f_portcntr)(struct qib_pportdata *, u32);
842 	u32 (*f_read_cntrs)(struct qib_devdata *, loff_t, char **,
843 		u64 **);
844 	u32 (*f_read_portcntrs)(struct qib_devdata *, loff_t, u32,
845 		char **, u64 **);
846 	u32 (*f_setpbc_control)(struct qib_pportdata *, u32, u8, u8);
847 	void (*f_initvl15_bufs)(struct qib_devdata *);
848 	void (*f_init_ctxt)(struct qib_ctxtdata *);
849 	void (*f_txchk_change)(struct qib_devdata *, u32, u32, u32,
850 		struct qib_ctxtdata *);
851 	void (*f_writescratch)(struct qib_devdata *, u32);
852 	int (*f_tempsense_rd)(struct qib_devdata *, int regnum);
853 #ifdef CONFIG_INFINIBAND_QIB_DCA
854 	int (*f_notify_dca)(struct qib_devdata *, unsigned long event);
855 #endif
856 
857 	char *boardname; /* human readable board info */
858 
859 	/* template for writing TIDs  */
860 	u64 tidtemplate;
861 	/* value to write to free TIDs */
862 	u64 tidinvalid;
863 
864 	/* number of registers used for pioavail */
865 	u32 pioavregs;
866 	/* device (not port) flags, basically device capabilities */
867 	u32 flags;
868 	/* last buffer for user use */
869 	u32 lastctxt_piobuf;
870 
871 	/* saturating counter of (non-port-specific) device interrupts */
872 	u32 int_counter;
873 
874 	/* pio bufs allocated per ctxt */
875 	u32 pbufsctxt;
876 	/* if remainder on bufs/ctxt, ctxts < extrabuf get 1 extra */
877 	u32 ctxts_extrabuf;
878 	/*
879 	 * number of ctxts configured as max; zero is set to number chip
880 	 * supports, less gives more pio bufs/ctxt, etc.
881 	 */
882 	u32 cfgctxts;
883 	/*
884 	 * number of ctxts available for PSM open
885 	 */
886 	u32 freectxts;
887 
888 	/*
889 	 * hint that we should update pioavailshadow before
890 	 * looking for a PIO buffer
891 	 */
892 	u32 upd_pio_shadow;
893 
894 	/* internal debugging stats */
895 	u32 maxpkts_call;
896 	u32 avgpkts_call;
897 	u64 nopiobufs;
898 
899 	/* PCI Vendor ID (here for NodeInfo) */
900 	u16 vendorid;
901 	/* PCI Device ID (here for NodeInfo) */
902 	u16 deviceid;
903 	/* for write combining settings */
904 	unsigned long wc_cookie;
905 	unsigned long wc_base;
906 	unsigned long wc_len;
907 
908 	/* shadow copy of struct page *'s for exp tid pages */
909 	struct page **pageshadow;
910 	/* shadow copy of dma handles for exp tid pages */
911 	dma_addr_t *physshadow;
912 	u64 __iomem *egrtidbase;
913 	spinlock_t sendctrl_lock; /* protect changes to sendctrl shadow */
914 	/* around rcd and (user ctxts) ctxt_cnt use (intr vs free) */
915 	spinlock_t uctxt_lock; /* rcd and user context changes */
916 	/*
917 	 * per unit status, see also portdata statusp
918 	 * mapped readonly into user processes so they can get unit and
919 	 * IB link status cheaply
920 	 */
921 	u64 *devstatusp;
922 	char *freezemsg; /* freeze msg if hw error put chip in freeze */
923 	u32 freezelen; /* max length of freezemsg */
924 	/* timer used to prevent stats overflow, error throttling, etc. */
925 	struct timer_list stats_timer;
926 
927 	/* timer to verify interrupts work, and fallback if possible */
928 	struct timer_list intrchk_timer;
929 	unsigned long ureg_align; /* user register alignment */
930 
931 	/*
932 	 * Protects pioavailshadow, pioavailkernel, pio_need_disarm, and
933 	 * pio_writing.
934 	 */
935 	spinlock_t pioavail_lock;
936 	/*
937 	 * index of last buffer to optimize search for next
938 	 */
939 	u32 last_pio;
940 	/*
941 	 * min kernel pio buffer to optimize search
942 	 */
943 	u32 min_kernel_pio;
944 	/*
945 	 * Shadow copies of registers; size indicates read access size.
946 	 * Most of them are readonly, but some are write-only register,
947 	 * where we manipulate the bits in the shadow copy, and then write
948 	 * the shadow copy to qlogic_ib.
949 	 *
950 	 * We deliberately make most of these 32 bits, since they have
951 	 * restricted range.  For any that we read, we won't to generate 32
952 	 * bit accesses, since Opteron will generate 2 separate 32 bit HT
953 	 * transactions for a 64 bit read, and we want to avoid unnecessary
954 	 * bus transactions.
955 	 */
956 
957 	/* This is the 64 bit group */
958 
959 	unsigned long pioavailshadow[6];
960 	/* bitmap of send buffers available for the kernel to use with PIO. */
961 	unsigned long pioavailkernel[6];
962 	/* bitmap of send buffers which need to be disarmed. */
963 	unsigned long pio_need_disarm[3];
964 	/* bitmap of send buffers which are being written to. */
965 	unsigned long pio_writing[3];
966 	/* kr_revision shadow */
967 	u64 revision;
968 	/* Base GUID for device (from eeprom, network order) */
969 	__be64 base_guid;
970 
971 	/*
972 	 * kr_sendpiobufbase value (chip offset of pio buffers), and the
973 	 * base of the 2KB buffer s(user processes only use 2K)
974 	 */
975 	u64 piobufbase;
976 	u32 pio2k_bufbase;
977 
978 	/* these are the "32 bit" regs */
979 
980 	/* number of GUIDs in the flash for this interface */
981 	u32 nguid;
982 	/*
983 	 * the following two are 32-bit bitmasks, but {test,clear,set}_bit
984 	 * all expect bit fields to be "unsigned long"
985 	 */
986 	unsigned long rcvctrl; /* shadow per device rcvctrl */
987 	unsigned long sendctrl; /* shadow per device sendctrl */
988 
989 	/* value we put in kr_rcvhdrcnt */
990 	u32 rcvhdrcnt;
991 	/* value we put in kr_rcvhdrsize */
992 	u32 rcvhdrsize;
993 	/* value we put in kr_rcvhdrentsize */
994 	u32 rcvhdrentsize;
995 	/* kr_ctxtcnt value */
996 	u32 ctxtcnt;
997 	/* kr_pagealign value */
998 	u32 palign;
999 	/* number of "2KB" PIO buffers */
1000 	u32 piobcnt2k;
1001 	/* size in bytes of "2KB" PIO buffers */
1002 	u32 piosize2k;
1003 	/* max usable size in dwords of a "2KB" PIO buffer before going "4KB" */
1004 	u32 piosize2kmax_dwords;
1005 	/* number of "4KB" PIO buffers */
1006 	u32 piobcnt4k;
1007 	/* size in bytes of "4KB" PIO buffers */
1008 	u32 piosize4k;
1009 	/* kr_rcvegrbase value */
1010 	u32 rcvegrbase;
1011 	/* kr_rcvtidbase value */
1012 	u32 rcvtidbase;
1013 	/* kr_rcvtidcnt value */
1014 	u32 rcvtidcnt;
1015 	/* kr_userregbase */
1016 	u32 uregbase;
1017 	/* shadow the control register contents */
1018 	u32 control;
1019 
1020 	/* chip address space used by 4k pio buffers */
1021 	u32 align4k;
1022 	/* size of each rcvegrbuffer */
1023 	u16 rcvegrbufsize;
1024 	/* log2 of above */
1025 	u16 rcvegrbufsize_shift;
1026 	/* localbus width (1, 2,4,8,16,32) from config space  */
1027 	u32 lbus_width;
1028 	/* localbus speed in MHz */
1029 	u32 lbus_speed;
1030 	int unit; /* unit # of this chip */
1031 
1032 	/* start of CHIP_SPEC move to chipspec, but need code changes */
1033 	/* low and high portions of MSI capability/vector */
1034 	u32 msi_lo;
1035 	/* saved after PCIe init for restore after reset */
1036 	u32 msi_hi;
1037 	/* MSI data (vector) saved for restore */
1038 	u16 msi_data;
1039 	/* so we can rewrite it after a chip reset */
1040 	u32 pcibar0;
1041 	/* so we can rewrite it after a chip reset */
1042 	u32 pcibar1;
1043 	u64 rhdrhead_intr_off;
1044 
1045 	/*
1046 	 * ASCII serial number, from flash, large enough for original
1047 	 * all digit strings, and longer QLogic serial number format
1048 	 */
1049 	u8 serial[16];
1050 	/* human readable board version */
1051 	u8 boardversion[96];
1052 	u8 lbus_info[32]; /* human readable localbus info */
1053 	/* chip major rev, from qib_revision */
1054 	u8 majrev;
1055 	/* chip minor rev, from qib_revision */
1056 	u8 minrev;
1057 
1058 	/* Misc small ints */
1059 	/* Number of physical ports available */
1060 	u8 num_pports;
1061 	/* Lowest context number which can be used by user processes */
1062 	u8 first_user_ctxt;
1063 	u8 n_krcv_queues;
1064 	u8 qpn_mask;
1065 	u8 skip_kctxt_mask;
1066 
1067 	u16 rhf_offset; /* offset of RHF within receive header entry */
1068 
1069 	/*
1070 	 * GPIO pins for twsi-connected devices, and device code for eeprom
1071 	 */
1072 	u8 gpio_sda_num;
1073 	u8 gpio_scl_num;
1074 	u8 twsi_eeprom_dev;
1075 	u8 board_atten;
1076 
1077 	/* Support (including locks) for EEPROM logging of errors and time */
1078 	/* control access to actual counters, timer */
1079 	spinlock_t eep_st_lock;
1080 	/* control high-level access to EEPROM */
1081 	struct mutex eep_lock;
1082 	uint64_t traffic_wds;
1083 	/* active time is kept in seconds, but logged in hours */
1084 	atomic_t active_time;
1085 	/* Below are nominal shadow of EEPROM, new since last EEPROM update */
1086 	uint8_t eep_st_errs[QIB_EEP_LOG_CNT];
1087 	uint8_t eep_st_new_errs[QIB_EEP_LOG_CNT];
1088 	uint16_t eep_hrs;
1089 	/*
1090 	 * masks for which bits of errs, hwerrs that cause
1091 	 * each of the counters to increment.
1092 	 */
1093 	struct qib_eep_log_mask eep_st_masks[QIB_EEP_LOG_CNT];
1094 	struct qib_diag_client *diag_client;
1095 	spinlock_t qib_diag_trans_lock; /* protect diag observer ops */
1096 	struct diag_observer_list_elt *diag_observer_list;
1097 
1098 	u8 psxmitwait_supported;
1099 	/* cycle length of PS* counters in HW (in picoseconds) */
1100 	u16 psxmitwait_check_rate;
1101 	/* high volume overflow errors defered to tasklet */
1102 	struct tasklet_struct error_tasklet;
1103 	/* per device cq worker */
1104 	struct kthread_worker *worker;
1105 
1106 	int assigned_node_id; /* NUMA node closest to HCA */
1107 };
1108 
1109 /* hol_state values */
1110 #define QIB_HOL_UP       0
1111 #define QIB_HOL_INIT     1
1112 
1113 #define QIB_SDMA_SENDCTRL_OP_ENABLE    (1U << 0)
1114 #define QIB_SDMA_SENDCTRL_OP_INTENABLE (1U << 1)
1115 #define QIB_SDMA_SENDCTRL_OP_HALT      (1U << 2)
1116 #define QIB_SDMA_SENDCTRL_OP_CLEANUP   (1U << 3)
1117 #define QIB_SDMA_SENDCTRL_OP_DRAIN     (1U << 4)
1118 
1119 /* operation types for f_txchk_change() */
1120 #define TXCHK_CHG_TYPE_DIS1  3
1121 #define TXCHK_CHG_TYPE_ENAB1 2
1122 #define TXCHK_CHG_TYPE_KERN  1
1123 #define TXCHK_CHG_TYPE_USER  0
1124 
1125 #define QIB_CHASE_TIME msecs_to_jiffies(145)
1126 #define QIB_CHASE_DIS_TIME msecs_to_jiffies(160)
1127 
1128 /* Private data for file operations */
1129 struct qib_filedata {
1130 	struct qib_ctxtdata *rcd;
1131 	unsigned subctxt;
1132 	unsigned tidcursor;
1133 	struct qib_user_sdma_queue *pq;
1134 	int rec_cpu_num; /* for cpu affinity; -1 if none */
1135 };
1136 
1137 extern struct list_head qib_dev_list;
1138 extern spinlock_t qib_devs_lock;
1139 extern struct qib_devdata *qib_lookup(int unit);
1140 extern u32 qib_cpulist_count;
1141 extern unsigned long *qib_cpulist;
1142 
1143 extern unsigned qib_wc_pat;
1144 extern unsigned qib_cc_table_size;
1145 int qib_init(struct qib_devdata *, int);
1146 int init_chip_wc_pat(struct qib_devdata *dd, u32);
1147 int qib_enable_wc(struct qib_devdata *dd);
1148 void qib_disable_wc(struct qib_devdata *dd);
1149 int qib_count_units(int *npresentp, int *nupp);
1150 int qib_count_active_units(void);
1151 
1152 int qib_cdev_init(int minor, const char *name,
1153 		  const struct file_operations *fops,
1154 		  struct cdev **cdevp, struct device **devp);
1155 void qib_cdev_cleanup(struct cdev **cdevp, struct device **devp);
1156 int qib_dev_init(void);
1157 void qib_dev_cleanup(void);
1158 
1159 int qib_diag_add(struct qib_devdata *);
1160 void qib_diag_remove(struct qib_devdata *);
1161 void qib_handle_e_ibstatuschanged(struct qib_pportdata *, u64);
1162 void qib_sdma_update_tail(struct qib_pportdata *, u16); /* hold sdma_lock */
1163 
1164 int qib_decode_err(struct qib_devdata *dd, char *buf, size_t blen, u64 err);
1165 void qib_bad_intrstatus(struct qib_devdata *);
1166 void qib_handle_urcv(struct qib_devdata *, u64);
1167 
1168 /* clean up any per-chip chip-specific stuff */
1169 void qib_chip_cleanup(struct qib_devdata *);
1170 /* clean up any chip type-specific stuff */
1171 void qib_chip_done(void);
1172 
1173 /* check to see if we have to force ordering for write combining */
1174 int qib_unordered_wc(void);
1175 void qib_pio_copy(void __iomem *to, const void *from, size_t count);
1176 
1177 void qib_disarm_piobufs(struct qib_devdata *, unsigned, unsigned);
1178 int qib_disarm_piobufs_ifneeded(struct qib_ctxtdata *);
1179 void qib_disarm_piobufs_set(struct qib_devdata *, unsigned long *, unsigned);
1180 void qib_cancel_sends(struct qib_pportdata *);
1181 
1182 int qib_create_rcvhdrq(struct qib_devdata *, struct qib_ctxtdata *);
1183 int qib_setup_eagerbufs(struct qib_ctxtdata *);
1184 void qib_set_ctxtcnt(struct qib_devdata *);
1185 int qib_create_ctxts(struct qib_devdata *dd);
1186 struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *, u32, int);
1187 void qib_init_pportdata(struct qib_pportdata *, struct qib_devdata *, u8, u8);
1188 void qib_free_ctxtdata(struct qib_devdata *, struct qib_ctxtdata *);
1189 
1190 u32 qib_kreceive(struct qib_ctxtdata *, u32 *, u32 *);
1191 int qib_reset_device(int);
1192 int qib_wait_linkstate(struct qib_pportdata *, u32, int);
1193 int qib_set_linkstate(struct qib_pportdata *, u8);
1194 int qib_set_mtu(struct qib_pportdata *, u16);
1195 int qib_set_lid(struct qib_pportdata *, u32, u8);
1196 void qib_hol_down(struct qib_pportdata *);
1197 void qib_hol_init(struct qib_pportdata *);
1198 void qib_hol_up(struct qib_pportdata *);
1199 void qib_hol_event(unsigned long);
1200 void qib_disable_after_error(struct qib_devdata *);
1201 int qib_set_uevent_bits(struct qib_pportdata *, const int);
1202 
1203 /* for use in system calls, where we want to know device type, etc. */
1204 #define ctxt_fp(fp) \
1205 	(((struct qib_filedata *)(fp)->private_data)->rcd)
1206 #define subctxt_fp(fp) \
1207 	(((struct qib_filedata *)(fp)->private_data)->subctxt)
1208 #define tidcursor_fp(fp) \
1209 	(((struct qib_filedata *)(fp)->private_data)->tidcursor)
1210 #define user_sdma_queue_fp(fp) \
1211 	(((struct qib_filedata *)(fp)->private_data)->pq)
1212 
1213 static inline struct qib_devdata *dd_from_ppd(struct qib_pportdata *ppd)
1214 {
1215 	return ppd->dd;
1216 }
1217 
1218 static inline struct qib_devdata *dd_from_dev(struct qib_ibdev *dev)
1219 {
1220 	return container_of(dev, struct qib_devdata, verbs_dev);
1221 }
1222 
1223 static inline struct qib_devdata *dd_from_ibdev(struct ib_device *ibdev)
1224 {
1225 	return dd_from_dev(to_idev(ibdev));
1226 }
1227 
1228 static inline struct qib_pportdata *ppd_from_ibp(struct qib_ibport *ibp)
1229 {
1230 	return container_of(ibp, struct qib_pportdata, ibport_data);
1231 }
1232 
1233 static inline struct qib_ibport *to_iport(struct ib_device *ibdev, u8 port)
1234 {
1235 	struct qib_devdata *dd = dd_from_ibdev(ibdev);
1236 	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
1237 
1238 	WARN_ON(pidx >= dd->num_pports);
1239 	return &dd->pport[pidx].ibport_data;
1240 }
1241 
1242 /*
1243  * values for dd->flags (_device_ related flags) and
1244  */
1245 #define QIB_HAS_LINK_LATENCY  0x1 /* supports link latency (IB 1.2) */
1246 #define QIB_INITTED           0x2 /* chip and driver up and initted */
1247 #define QIB_DOING_RESET       0x4  /* in the middle of doing chip reset */
1248 #define QIB_PRESENT           0x8  /* chip accesses can be done */
1249 #define QIB_PIO_FLUSH_WC      0x10 /* Needs Write combining flush for PIO */
1250 #define QIB_HAS_THRESH_UPDATE 0x40
1251 #define QIB_HAS_SDMA_TIMEOUT  0x80
1252 #define QIB_USE_SPCL_TRIG     0x100 /* SpecialTrigger launch enabled */
1253 #define QIB_NODMA_RTAIL       0x200 /* rcvhdrtail register DMA enabled */
1254 #define QIB_HAS_INTX          0x800 /* Supports INTx interrupts */
1255 #define QIB_HAS_SEND_DMA      0x1000 /* Supports Send DMA */
1256 #define QIB_HAS_VLSUPP        0x2000 /* Supports multiple VLs; PBC different */
1257 #define QIB_HAS_HDRSUPP       0x4000 /* Supports header suppression */
1258 #define QIB_BADINTR           0x8000 /* severe interrupt problems */
1259 #define QIB_DCA_ENABLED       0x10000 /* Direct Cache Access enabled */
1260 #define QIB_HAS_QSFP          0x20000 /* device (card instance) has QSFP */
1261 
1262 /*
1263  * values for ppd->lflags (_ib_port_ related flags)
1264  */
1265 #define QIBL_LINKV             0x1 /* IB link state valid */
1266 #define QIBL_LINKDOWN          0x8 /* IB link is down */
1267 #define QIBL_LINKINIT          0x10 /* IB link level is up */
1268 #define QIBL_LINKARMED         0x20 /* IB link is ARMED */
1269 #define QIBL_LINKACTIVE        0x40 /* IB link is ACTIVE */
1270 /* leave a gap for more IB-link state */
1271 #define QIBL_IB_AUTONEG_INPROG 0x1000 /* non-IBTA DDR/QDR neg active */
1272 #define QIBL_IB_AUTONEG_FAILED 0x2000 /* non-IBTA DDR/QDR neg failed */
1273 #define QIBL_IB_LINK_DISABLED  0x4000 /* Linkdown-disable forced,
1274 				       * Do not try to bring up */
1275 #define QIBL_IB_FORCE_NOTIFY   0x8000 /* force notify on next ib change */
1276 
1277 /* IB dword length mask in PBC (lower 11 bits); same for all chips */
1278 #define QIB_PBC_LENGTH_MASK                     ((1 << 11) - 1)
1279 
1280 
1281 /* ctxt_flag bit offsets */
1282 		/* waiting for a packet to arrive */
1283 #define QIB_CTXT_WAITING_RCV   2
1284 		/* master has not finished initializing */
1285 #define QIB_CTXT_MASTER_UNINIT 4
1286 		/* waiting for an urgent packet to arrive */
1287 #define QIB_CTXT_WAITING_URG 5
1288 
1289 /* free up any allocated data at closes */
1290 void qib_free_data(struct qib_ctxtdata *dd);
1291 void qib_chg_pioavailkernel(struct qib_devdata *, unsigned, unsigned,
1292 			    u32, struct qib_ctxtdata *);
1293 struct qib_devdata *qib_init_iba7322_funcs(struct pci_dev *,
1294 					   const struct pci_device_id *);
1295 struct qib_devdata *qib_init_iba7220_funcs(struct pci_dev *,
1296 					   const struct pci_device_id *);
1297 struct qib_devdata *qib_init_iba6120_funcs(struct pci_dev *,
1298 					   const struct pci_device_id *);
1299 void qib_free_devdata(struct qib_devdata *);
1300 struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra);
1301 
1302 #define QIB_TWSI_NO_DEV 0xFF
1303 /* Below qib_twsi_ functions must be called with eep_lock held */
1304 int qib_twsi_reset(struct qib_devdata *dd);
1305 int qib_twsi_blk_rd(struct qib_devdata *dd, int dev, int addr, void *buffer,
1306 		    int len);
1307 int qib_twsi_blk_wr(struct qib_devdata *dd, int dev, int addr,
1308 		    const void *buffer, int len);
1309 void qib_get_eeprom_info(struct qib_devdata *);
1310 int qib_update_eeprom_log(struct qib_devdata *dd);
1311 void qib_inc_eeprom_err(struct qib_devdata *dd, u32 eidx, u32 incr);
1312 void qib_dump_lookup_output_queue(struct qib_devdata *);
1313 void qib_force_pio_avail_update(struct qib_devdata *);
1314 void qib_clear_symerror_on_linkup(unsigned long opaque);
1315 
1316 /*
1317  * Set LED override, only the two LSBs have "public" meaning, but
1318  * any non-zero value substitutes them for the Link and LinkTrain
1319  * LED states.
1320  */
1321 #define QIB_LED_PHYS 1 /* Physical (linktraining) GREEN LED */
1322 #define QIB_LED_LOG 2  /* Logical (link) YELLOW LED */
1323 void qib_set_led_override(struct qib_pportdata *ppd, unsigned int val);
1324 
1325 /* send dma routines */
1326 int qib_setup_sdma(struct qib_pportdata *);
1327 void qib_teardown_sdma(struct qib_pportdata *);
1328 void __qib_sdma_intr(struct qib_pportdata *);
1329 void qib_sdma_intr(struct qib_pportdata *);
1330 void qib_user_sdma_send_desc(struct qib_pportdata *dd,
1331 			struct list_head *pktlist);
1332 int qib_sdma_verbs_send(struct qib_pportdata *, struct qib_sge_state *,
1333 			u32, struct qib_verbs_txreq *);
1334 /* ppd->sdma_lock should be locked before calling this. */
1335 int qib_sdma_make_progress(struct qib_pportdata *dd);
1336 
1337 static inline int qib_sdma_empty(const struct qib_pportdata *ppd)
1338 {
1339 	return ppd->sdma_descq_added == ppd->sdma_descq_removed;
1340 }
1341 
1342 /* must be called under qib_sdma_lock */
1343 static inline u16 qib_sdma_descq_freecnt(const struct qib_pportdata *ppd)
1344 {
1345 	return ppd->sdma_descq_cnt -
1346 		(ppd->sdma_descq_added - ppd->sdma_descq_removed) - 1;
1347 }
1348 
1349 static inline int __qib_sdma_running(struct qib_pportdata *ppd)
1350 {
1351 	return ppd->sdma_state.current_state == qib_sdma_state_s99_running;
1352 }
1353 int qib_sdma_running(struct qib_pportdata *);
1354 void dump_sdma_state(struct qib_pportdata *ppd);
1355 void __qib_sdma_process_event(struct qib_pportdata *, enum qib_sdma_events);
1356 void qib_sdma_process_event(struct qib_pportdata *, enum qib_sdma_events);
1357 
1358 /*
1359  * number of words used for protocol header if not set by qib_userinit();
1360  */
1361 #define QIB_DFLT_RCVHDRSIZE 9
1362 
1363 /*
1364  * We need to be able to handle an IB header of at least 24 dwords.
1365  * We need the rcvhdrq large enough to handle largest IB header, but
1366  * still have room for a 2KB MTU standard IB packet.
1367  * Additionally, some processor/memory controller combinations
1368  * benefit quite strongly from having the DMA'ed data be cacheline
1369  * aligned and a cacheline multiple, so we set the size to 32 dwords
1370  * (2 64-byte primary cachelines for pretty much all processors of
1371  * interest).  The alignment hurts nothing, other than using somewhat
1372  * more memory.
1373  */
1374 #define QIB_RCVHDR_ENTSIZE 32
1375 
1376 int qib_get_user_pages(unsigned long, size_t, struct page **);
1377 void qib_release_user_pages(struct page **, size_t);
1378 int qib_eeprom_read(struct qib_devdata *, u8, void *, int);
1379 int qib_eeprom_write(struct qib_devdata *, u8, const void *, int);
1380 u32 __iomem *qib_getsendbuf_range(struct qib_devdata *, u32 *, u32, u32);
1381 void qib_sendbuf_done(struct qib_devdata *, unsigned);
1382 
1383 static inline void qib_clear_rcvhdrtail(const struct qib_ctxtdata *rcd)
1384 {
1385 	*((u64 *) rcd->rcvhdrtail_kvaddr) = 0ULL;
1386 }
1387 
1388 static inline u32 qib_get_rcvhdrtail(const struct qib_ctxtdata *rcd)
1389 {
1390 	/*
1391 	 * volatile because it's a DMA target from the chip, routine is
1392 	 * inlined, and don't want register caching or reordering.
1393 	 */
1394 	return (u32) le64_to_cpu(
1395 		*((volatile __le64 *)rcd->rcvhdrtail_kvaddr)); /* DMA'ed */
1396 }
1397 
1398 static inline u32 qib_get_hdrqtail(const struct qib_ctxtdata *rcd)
1399 {
1400 	const struct qib_devdata *dd = rcd->dd;
1401 	u32 hdrqtail;
1402 
1403 	if (dd->flags & QIB_NODMA_RTAIL) {
1404 		__le32 *rhf_addr;
1405 		u32 seq;
1406 
1407 		rhf_addr = (__le32 *) rcd->rcvhdrq +
1408 			rcd->head + dd->rhf_offset;
1409 		seq = qib_hdrget_seq(rhf_addr);
1410 		hdrqtail = rcd->head;
1411 		if (seq == rcd->seq_cnt)
1412 			hdrqtail++;
1413 	} else
1414 		hdrqtail = qib_get_rcvhdrtail(rcd);
1415 
1416 	return hdrqtail;
1417 }
1418 
1419 /*
1420  * sysfs interface.
1421  */
1422 
1423 extern const char ib_qib_version[];
1424 
1425 int qib_device_create(struct qib_devdata *);
1426 void qib_device_remove(struct qib_devdata *);
1427 
1428 int qib_create_port_files(struct ib_device *ibdev, u8 port_num,
1429 			  struct kobject *kobj);
1430 int qib_verbs_register_sysfs(struct qib_devdata *);
1431 void qib_verbs_unregister_sysfs(struct qib_devdata *);
1432 /* Hook for sysfs read of QSFP */
1433 extern int qib_qsfp_dump(struct qib_pportdata *ppd, char *buf, int len);
1434 
1435 int __init qib_init_qibfs(void);
1436 int __exit qib_exit_qibfs(void);
1437 
1438 int qibfs_add(struct qib_devdata *);
1439 int qibfs_remove(struct qib_devdata *);
1440 
1441 int qib_pcie_init(struct pci_dev *, const struct pci_device_id *);
1442 int qib_pcie_ddinit(struct qib_devdata *, struct pci_dev *,
1443 		    const struct pci_device_id *);
1444 void qib_pcie_ddcleanup(struct qib_devdata *);
1445 int qib_pcie_params(struct qib_devdata *, u32, u32 *, struct qib_msix_entry *);
1446 int qib_reinit_intr(struct qib_devdata *);
1447 void qib_enable_intx(struct pci_dev *);
1448 void qib_nomsi(struct qib_devdata *);
1449 void qib_nomsix(struct qib_devdata *);
1450 void qib_pcie_getcmd(struct qib_devdata *, u16 *, u8 *, u8 *);
1451 void qib_pcie_reenable(struct qib_devdata *, u16, u8, u8);
1452 
1453 /*
1454  * dma_addr wrappers - all 0's invalid for hw
1455  */
1456 dma_addr_t qib_map_page(struct pci_dev *, struct page *, unsigned long,
1457 			  size_t, int);
1458 const char *qib_get_unit_name(int unit);
1459 
1460 /*
1461  * Flush write combining store buffers (if present) and perform a write
1462  * barrier.
1463  */
1464 #if defined(CONFIG_X86_64)
1465 #define qib_flush_wc() asm volatile("sfence" : : : "memory")
1466 #else
1467 #define qib_flush_wc() wmb() /* no reorder around wc flush */
1468 #endif
1469 
1470 /* global module parameter variables */
1471 extern unsigned qib_ibmtu;
1472 extern ushort qib_cfgctxts;
1473 extern ushort qib_num_cfg_vls;
1474 extern ushort qib_mini_init; /* If set, do few (ideally 0) writes to chip */
1475 extern unsigned qib_n_krcv_queues;
1476 extern unsigned qib_sdma_fetch_arb;
1477 extern unsigned qib_compat_ddr_negotiate;
1478 extern int qib_special_trigger;
1479 extern unsigned qib_numa_aware;
1480 
1481 extern struct mutex qib_mutex;
1482 
1483 /* Number of seconds before our card status check...  */
1484 #define STATUS_TIMEOUT 60
1485 
1486 #define QIB_DRV_NAME            "ib_qib"
1487 #define QIB_USER_MINOR_BASE     0
1488 #define QIB_TRACE_MINOR         127
1489 #define QIB_DIAGPKT_MINOR       128
1490 #define QIB_DIAG_MINOR_BASE     129
1491 #define QIB_NMINORS             255
1492 
1493 #define PCI_VENDOR_ID_PATHSCALE 0x1fc1
1494 #define PCI_VENDOR_ID_QLOGIC 0x1077
1495 #define PCI_DEVICE_ID_QLOGIC_IB_6120 0x10
1496 #define PCI_DEVICE_ID_QLOGIC_IB_7220 0x7220
1497 #define PCI_DEVICE_ID_QLOGIC_IB_7322 0x7322
1498 
1499 /*
1500  * qib_early_err is used (only!) to print early errors before devdata is
1501  * allocated, or when dd->pcidev may not be valid, and at the tail end of
1502  * cleanup when devdata may have been freed, etc.  qib_dev_porterr is
1503  * the same as qib_dev_err, but is used when the message really needs
1504  * the IB port# to be definitive as to what's happening..
1505  * All of these go to the trace log, and the trace log entry is done
1506  * first to avoid possible serial port delays from printk.
1507  */
1508 #define qib_early_err(dev, fmt, ...) \
1509 	dev_err(dev, fmt, ##__VA_ARGS__)
1510 
1511 #define qib_dev_err(dd, fmt, ...) \
1512 	dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
1513 		qib_get_unit_name((dd)->unit), ##__VA_ARGS__)
1514 
1515 #define qib_dev_warn(dd, fmt, ...) \
1516 	dev_warn(&(dd)->pcidev->dev, "%s: " fmt, \
1517 		qib_get_unit_name((dd)->unit), ##__VA_ARGS__)
1518 
1519 #define qib_dev_porterr(dd, port, fmt, ...) \
1520 	dev_err(&(dd)->pcidev->dev, "%s: IB%u:%u " fmt, \
1521 		qib_get_unit_name((dd)->unit), (dd)->unit, (port), \
1522 		##__VA_ARGS__)
1523 
1524 #define qib_devinfo(pcidev, fmt, ...) \
1525 	dev_info(&(pcidev)->dev, fmt, ##__VA_ARGS__)
1526 
1527 /*
1528  * this is used for formatting hw error messages...
1529  */
1530 struct qib_hwerror_msgs {
1531 	u64 mask;
1532 	const char *msg;
1533 	size_t sz;
1534 };
1535 
1536 #define QLOGIC_IB_HWE_MSG(a, b) { .mask = a, .msg = b }
1537 
1538 /* in qib_intr.c... */
1539 void qib_format_hwerrors(u64 hwerrs,
1540 			 const struct qib_hwerror_msgs *hwerrmsgs,
1541 			 size_t nhwerrmsgs, char *msg, size_t lmsg);
1542 #endif                          /* _QIB_KERNEL_H */
1543