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 */
76struct 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
89extern struct qlogic_ib_stats qib_stats;
90extern struct pci_error_handlers qib_pci_err_handler;
91extern 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)
109struct 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 */
117struct 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
233struct qib_sge_state;
234
235struct 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
245struct qib_sdma_desc {
246	__le64 qw[2];
247};
248
249struct 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#define MAX_NAME_SIZE 64
431struct qib_msix_entry {
432	struct msix_entry msix;
433	void *arg;
434	char name[MAX_NAME_SIZE];
435	cpumask_var_t mask;
436};
437
438/* Below is an opaque struct. Each chip (device) can maintain
439 * private data needed for its operation, but not germane to the
440 * rest of the driver.  For convenience, we define another that
441 * is chip-specific, per-port
442 */
443struct qib_chip_specific;
444struct qib_chipport_specific;
445
446enum qib_sdma_states {
447	qib_sdma_state_s00_hw_down,
448	qib_sdma_state_s10_hw_start_up_wait,
449	qib_sdma_state_s20_idle,
450	qib_sdma_state_s30_sw_clean_up_wait,
451	qib_sdma_state_s40_hw_clean_up_wait,
452	qib_sdma_state_s50_hw_halt_wait,
453	qib_sdma_state_s99_running,
454};
455
456enum qib_sdma_events {
457	qib_sdma_event_e00_go_hw_down,
458	qib_sdma_event_e10_go_hw_start,
459	qib_sdma_event_e20_hw_started,
460	qib_sdma_event_e30_go_running,
461	qib_sdma_event_e40_sw_cleaned,
462	qib_sdma_event_e50_hw_cleaned,
463	qib_sdma_event_e60_hw_halted,
464	qib_sdma_event_e70_go_idle,
465	qib_sdma_event_e7220_err_halted,
466	qib_sdma_event_e7322_err_halted,
467	qib_sdma_event_e90_timer_tick,
468};
469
470extern char *qib_sdma_state_names[];
471extern char *qib_sdma_event_names[];
472
473struct sdma_set_state_action {
474	unsigned op_enable:1;
475	unsigned op_intenable:1;
476	unsigned op_halt:1;
477	unsigned op_drain:1;
478	unsigned go_s99_running_tofalse:1;
479	unsigned go_s99_running_totrue:1;
480};
481
482struct qib_sdma_state {
483	struct kref          kref;
484	struct completion    comp;
485	enum qib_sdma_states current_state;
486	struct sdma_set_state_action *set_state_action;
487	unsigned             current_op;
488	unsigned             go_s99_running;
489	unsigned             first_sendbuf;
490	unsigned             last_sendbuf; /* really last +1 */
491	/* debugging/devel */
492	enum qib_sdma_states previous_state;
493	unsigned             previous_op;
494	enum qib_sdma_events last_event;
495};
496
497struct xmit_wait {
498	struct timer_list timer;
499	u64 counter;
500	u8 flags;
501	struct cache {
502		u64 psxmitdata;
503		u64 psrcvdata;
504		u64 psxmitpkts;
505		u64 psrcvpkts;
506		u64 psxmitwait;
507	} counter_cache;
508};
509
510/*
511 * The structure below encapsulates data relevant to a physical IB Port.
512 * Current chips support only one such port, but the separation
513 * clarifies things a bit. Note that to conform to IB conventions,
514 * port-numbers are one-based. The first or only port is port1.
515 */
516struct qib_pportdata {
517	struct qib_ibport ibport_data;
518
519	struct qib_devdata *dd;
520	struct qib_chippport_specific *cpspec; /* chip-specific per-port */
521	struct kobject pport_kobj;
522	struct kobject sl2vl_kobj;
523	struct kobject diagc_kobj;
524
525	/* GUID for this interface, in network order */
526	__be64 guid;
527
528	/* QIB_POLL, etc. link-state specific flags, per port */
529	u32 lflags;
530	/* qib_lflags driver is waiting for */
531	u32 state_wanted;
532	spinlock_t lflags_lock;
533	/* number of (port-specific) interrupts for this port -- saturates... */
534	u32 int_counter;
535
536	/* ref count for each pkey */
537	atomic_t pkeyrefs[4];
538
539	/*
540	 * this address is mapped readonly into user processes so they can
541	 * get status cheaply, whenever they want.  One qword of status per port
542	 */
543	u64 *statusp;
544
545	/* SendDMA related entries */
546	spinlock_t            sdma_lock;
547	struct qib_sdma_state sdma_state;
548	unsigned long         sdma_buf_jiffies;
549	struct qib_sdma_desc *sdma_descq;
550	u64                   sdma_descq_added;
551	u64                   sdma_descq_removed;
552	u16                   sdma_descq_cnt;
553	u16                   sdma_descq_tail;
554	u16                   sdma_descq_head;
555	u16                   sdma_next_intr;
556	u16                   sdma_reset_wait;
557	u8                    sdma_generation;
558	struct tasklet_struct sdma_sw_clean_up_task;
559	struct list_head      sdma_activelist;
560
561	dma_addr_t       sdma_descq_phys;
562	volatile __le64 *sdma_head_dma; /* DMA'ed by chip */
563	dma_addr_t       sdma_head_phys;
564
565	wait_queue_head_t state_wait; /* for state_wanted */
566
567	/* HoL blocking for SMP replies */
568	unsigned          hol_state;
569	struct timer_list hol_timer;
570
571	/*
572	 * Shadow copies of registers; size indicates read access size.
573	 * Most of them are readonly, but some are write-only register,
574	 * where we manipulate the bits in the shadow copy, and then write
575	 * the shadow copy to qlogic_ib.
576	 *
577	 * We deliberately make most of these 32 bits, since they have
578	 * restricted range.  For any that we read, we won't to generate 32
579	 * bit accesses, since Opteron will generate 2 separate 32 bit HT
580	 * transactions for a 64 bit read, and we want to avoid unnecessary
581	 * bus transactions.
582	 */
583
584	/* This is the 64 bit group */
585	/* last ibcstatus.  opaque outside chip-specific code */
586	u64 lastibcstat;
587
588	/* these are the "32 bit" regs */
589
590	/*
591	 * the following two are 32-bit bitmasks, but {test,clear,set}_bit
592	 * all expect bit fields to be "unsigned long"
593	 */
594	unsigned long p_rcvctrl; /* shadow per-port rcvctrl */
595	unsigned long p_sendctrl; /* shadow per-port sendctrl */
596
597	u32 ibmtu; /* The MTU programmed for this unit */
598	/*
599	 * Current max size IB packet (in bytes) including IB headers, that
600	 * we can send. Changes when ibmtu changes.
601	 */
602	u32 ibmaxlen;
603	/*
604	 * ibmaxlen at init time, limited by chip and by receive buffer
605	 * size.  Not changed after init.
606	 */
607	u32 init_ibmaxlen;
608	/* LID programmed for this instance */
609	u16 lid;
610	/* list of pkeys programmed; 0 if not set */
611	u16 pkeys[4];
612	/* LID mask control */
613	u8 lmc;
614	u8 link_width_supported;
615	u8 link_speed_supported;
616	u8 link_width_enabled;
617	u8 link_speed_enabled;
618	u8 link_width_active;
619	u8 link_speed_active;
620	u8 vls_supported;
621	u8 vls_operational;
622	/* Rx Polarity inversion (compensate for ~tx on partner) */
623	u8 rx_pol_inv;
624
625	u8 hw_pidx;     /* physical port index */
626	u8 port;        /* IB port number and index into dd->pports - 1 */
627
628	u8 delay_mult;
629
630	/* used to override LED behavior */
631	u8 led_override;  /* Substituted for normal value, if non-zero */
632	u16 led_override_timeoff; /* delta to next timer event */
633	u8 led_override_vals[2]; /* Alternates per blink-frame */
634	u8 led_override_phase; /* Just counts, LSB picks from vals[] */
635	atomic_t led_override_timer_active;
636	/* Used to flash LEDs in override mode */
637	struct timer_list led_override_timer;
638	struct xmit_wait cong_stats;
639	struct timer_list symerr_clear_timer;
640};
641
642/* Observers. Not to be taken lightly, possibly not to ship. */
643/*
644 * If a diag read or write is to (bottom <= offset <= top),
645 * the "hoook" is called, allowing, e.g. shadows to be
646 * updated in sync with the driver. struct diag_observer
647 * is the "visible" part.
648 */
649struct diag_observer;
650
651typedef int (*diag_hook) (struct qib_devdata *dd,
652	const struct diag_observer *op,
653	u32 offs, u64 *data, u64 mask, int only_32);
654
655struct diag_observer {
656	diag_hook hook;
657	u32 bottom;
658	u32 top;
659};
660
661extern int qib_register_observer(struct qib_devdata *dd,
662	const struct diag_observer *op);
663
664/* Only declared here, not defined. Private to diags */
665struct diag_observer_list_elt;
666
667/* device data struct now contains only "general per-device" info.
668 * fields related to a physical IB port are in a qib_pportdata struct,
669 * described above) while fields only used by a particular chip-type are in
670 * a qib_chipdata struct, whose contents are opaque to this file.
671 */
672struct qib_devdata {
673	struct qib_ibdev verbs_dev;     /* must be first */
674	struct list_head list;
675	/* pointers to related structs for this device */
676	/* pci access data structure */
677	struct pci_dev *pcidev;
678	struct cdev *user_cdev;
679	struct cdev *diag_cdev;
680	struct device *user_device;
681	struct device *diag_device;
682
683	/* mem-mapped pointer to base of chip regs */
684	u64 __iomem *kregbase;
685	/* end of mem-mapped chip space excluding sendbuf and user regs */
686	u64 __iomem *kregend;
687	/* physical address of chip for io_remap, etc. */
688	resource_size_t physaddr;
689	/* qib_cfgctxts pointers */
690	struct qib_ctxtdata **rcd; /* Receive Context Data */
691
692	/* qib_pportdata, points to array of (physical) port-specific
693	 * data structs, indexed by pidx (0..n-1)
694	 */
695	struct qib_pportdata *pport;
696	struct qib_chip_specific *cspec; /* chip-specific */
697
698	/* kvirt address of 1st 2k pio buffer */
699	void __iomem *pio2kbase;
700	/* kvirt address of 1st 4k pio buffer */
701	void __iomem *pio4kbase;
702	/* mem-mapped pointer to base of PIO buffers (if using WC PAT) */
703	void __iomem *piobase;
704	/* mem-mapped pointer to base of user chip regs (if using WC PAT) */
705	u64 __iomem *userbase;
706	void __iomem *piovl15base; /* base of VL15 buffers, if not WC */
707	/*
708	 * points to area where PIOavail registers will be DMA'ed.
709	 * Has to be on a page of it's own, because the page will be
710	 * mapped into user program space.  This copy is *ONLY* ever
711	 * written by DMA, not by the driver!  Need a copy per device
712	 * when we get to multiple devices
713	 */
714	volatile __le64 *pioavailregs_dma; /* DMA'ed by chip */
715	/* physical address where updates occur */
716	dma_addr_t pioavailregs_phys;
717
718	/* device-specific implementations of functions needed by
719	 * common code. Contrary to previous consensus, we can't
720	 * really just point to a device-specific table, because we
721	 * may need to "bend", e.g. *_f_put_tid
722	 */
723	/* fallback to alternate interrupt type if possible */
724	int (*f_intr_fallback)(struct qib_devdata *);
725	/* hard reset chip */
726	int (*f_reset)(struct qib_devdata *);
727	void (*f_quiet_serdes)(struct qib_pportdata *);
728	int (*f_bringup_serdes)(struct qib_pportdata *);
729	int (*f_early_init)(struct qib_devdata *);
730	void (*f_clear_tids)(struct qib_devdata *, struct qib_ctxtdata *);
731	void (*f_put_tid)(struct qib_devdata *, u64 __iomem*,
732				u32, unsigned long);
733	void (*f_cleanup)(struct qib_devdata *);
734	void (*f_setextled)(struct qib_pportdata *, u32);
735	/* fill out chip-specific fields */
736	int (*f_get_base_info)(struct qib_ctxtdata *, struct qib_base_info *);
737	/* free irq */
738	void (*f_free_irq)(struct qib_devdata *);
739	struct qib_message_header *(*f_get_msgheader)
740					(struct qib_devdata *, __le32 *);
741	void (*f_config_ctxts)(struct qib_devdata *);
742	int (*f_get_ib_cfg)(struct qib_pportdata *, int);
743	int (*f_set_ib_cfg)(struct qib_pportdata *, int, u32);
744	int (*f_set_ib_loopback)(struct qib_pportdata *, const char *);
745	int (*f_get_ib_table)(struct qib_pportdata *, int, void *);
746	int (*f_set_ib_table)(struct qib_pportdata *, int, void *);
747	u32 (*f_iblink_state)(u64);
748	u8 (*f_ibphys_portstate)(u64);
749	void (*f_xgxs_reset)(struct qib_pportdata *);
750	/* per chip actions needed for IB Link up/down changes */
751	int (*f_ib_updown)(struct qib_pportdata *, int, u64);
752	u32 __iomem *(*f_getsendbuf)(struct qib_pportdata *, u64, u32 *);
753	/* Read/modify/write of GPIO pins (potentially chip-specific */
754	int (*f_gpio_mod)(struct qib_devdata *dd, u32 out, u32 dir,
755		u32 mask);
756	/* Enable writes to config EEPROM (if supported) */
757	int (*f_eeprom_wen)(struct qib_devdata *dd, int wen);
758	/*
759	 * modify rcvctrl shadow[s] and write to appropriate chip-regs.
760	 * see above QIB_RCVCTRL_xxx_ENB/DIS for operations.
761	 * (ctxt == -1) means "all contexts", only meaningful for
762	 * clearing. Could remove if chip_spec shutdown properly done.
763	 */
764	void (*f_rcvctrl)(struct qib_pportdata *, unsigned int op,
765		int ctxt);
766	/* Read/modify/write sendctrl appropriately for op and port. */
767	void (*f_sendctrl)(struct qib_pportdata *, u32 op);
768	void (*f_set_intr_state)(struct qib_devdata *, u32);
769	void (*f_set_armlaunch)(struct qib_devdata *, u32);
770	void (*f_wantpiobuf_intr)(struct qib_devdata *, u32);
771	int (*f_late_initreg)(struct qib_devdata *);
772	int (*f_init_sdma_regs)(struct qib_pportdata *);
773	u16 (*f_sdma_gethead)(struct qib_pportdata *);
774	int (*f_sdma_busy)(struct qib_pportdata *);
775	void (*f_sdma_update_tail)(struct qib_pportdata *, u16);
776	void (*f_sdma_set_desc_cnt)(struct qib_pportdata *, unsigned);
777	void (*f_sdma_sendctrl)(struct qib_pportdata *, unsigned);
778	void (*f_sdma_hw_clean_up)(struct qib_pportdata *);
779	void (*f_sdma_hw_start_up)(struct qib_pportdata *);
780	void (*f_sdma_init_early)(struct qib_pportdata *);
781	void (*f_set_cntr_sample)(struct qib_pportdata *, u32, u32);
782	void (*f_update_usrhead)(struct qib_ctxtdata *, u64, u32, u32, u32);
783	u32 (*f_hdrqempty)(struct qib_ctxtdata *);
784	u64 (*f_portcntr)(struct qib_pportdata *, u32);
785	u32 (*f_read_cntrs)(struct qib_devdata *, loff_t, char **,
786		u64 **);
787	u32 (*f_read_portcntrs)(struct qib_devdata *, loff_t, u32,
788		char **, u64 **);
789	u32 (*f_setpbc_control)(struct qib_pportdata *, u32, u8, u8);
790	void (*f_initvl15_bufs)(struct qib_devdata *);
791	void (*f_init_ctxt)(struct qib_ctxtdata *);
792	void (*f_txchk_change)(struct qib_devdata *, u32, u32, u32,
793		struct qib_ctxtdata *);
794	void (*f_writescratch)(struct qib_devdata *, u32);
795	int (*f_tempsense_rd)(struct qib_devdata *, int regnum);
796
797	char *boardname; /* human readable board info */
798
799	/* template for writing TIDs  */
800	u64 tidtemplate;
801	/* value to write to free TIDs */
802	u64 tidinvalid;
803
804	/* number of registers used for pioavail */
805	u32 pioavregs;
806	/* device (not port) flags, basically device capabilities */
807	u32 flags;
808	/* last buffer for user use */
809	u32 lastctxt_piobuf;
810
811	/* saturating counter of (non-port-specific) device interrupts */
812	u32 int_counter;
813
814	/* pio bufs allocated per ctxt */
815	u32 pbufsctxt;
816	/* if remainder on bufs/ctxt, ctxts < extrabuf get 1 extra */
817	u32 ctxts_extrabuf;
818	/*
819	 * number of ctxts configured as max; zero is set to number chip
820	 * supports, less gives more pio bufs/ctxt, etc.
821	 */
822	u32 cfgctxts;
823	/*
824	 * number of ctxts available for PSM open
825	 */
826	u32 freectxts;
827
828	/*
829	 * hint that we should update pioavailshadow before
830	 * looking for a PIO buffer
831	 */
832	u32 upd_pio_shadow;
833
834	/* internal debugging stats */
835	u32 maxpkts_call;
836	u32 avgpkts_call;
837	u64 nopiobufs;
838
839	/* PCI Vendor ID (here for NodeInfo) */
840	u16 vendorid;
841	/* PCI Device ID (here for NodeInfo) */
842	u16 deviceid;
843	/* for write combining settings */
844	unsigned long wc_cookie;
845	unsigned long wc_base;
846	unsigned long wc_len;
847
848	/* shadow copy of struct page *'s for exp tid pages */
849	struct page **pageshadow;
850	/* shadow copy of dma handles for exp tid pages */
851	dma_addr_t *physshadow;
852	u64 __iomem *egrtidbase;
853	spinlock_t sendctrl_lock; /* protect changes to sendctrl shadow */
854	/* around rcd and (user ctxts) ctxt_cnt use (intr vs free) */
855	spinlock_t uctxt_lock; /* rcd and user context changes */
856	/*
857	 * per unit status, see also portdata statusp
858	 * mapped readonly into user processes so they can get unit and
859	 * IB link status cheaply
860	 */
861	u64 *devstatusp;
862	char *freezemsg; /* freeze msg if hw error put chip in freeze */
863	u32 freezelen; /* max length of freezemsg */
864	/* timer used to prevent stats overflow, error throttling, etc. */
865	struct timer_list stats_timer;
866
867	/* timer to verify interrupts work, and fallback if possible */
868	struct timer_list intrchk_timer;
869	unsigned long ureg_align; /* user register alignment */
870
871	/*
872	 * Protects pioavailshadow, pioavailkernel, pio_need_disarm, and
873	 * pio_writing.
874	 */
875	spinlock_t pioavail_lock;
876
877	/*
878	 * Shadow copies of registers; size indicates read access size.
879	 * Most of them are readonly, but some are write-only register,
880	 * where we manipulate the bits in the shadow copy, and then write
881	 * the shadow copy to qlogic_ib.
882	 *
883	 * We deliberately make most of these 32 bits, since they have
884	 * restricted range.  For any that we read, we won't to generate 32
885	 * bit accesses, since Opteron will generate 2 separate 32 bit HT
886	 * transactions for a 64 bit read, and we want to avoid unnecessary
887	 * bus transactions.
888	 */
889
890	/* This is the 64 bit group */
891
892	unsigned long pioavailshadow[6];
893	/* bitmap of send buffers available for the kernel to use with PIO. */
894	unsigned long pioavailkernel[6];
895	/* bitmap of send buffers which need to be disarmed. */
896	unsigned long pio_need_disarm[3];
897	/* bitmap of send buffers which are being written to. */
898	unsigned long pio_writing[3];
899	/* kr_revision shadow */
900	u64 revision;
901	/* Base GUID for device (from eeprom, network order) */
902	__be64 base_guid;
903
904	/*
905	 * kr_sendpiobufbase value (chip offset of pio buffers), and the
906	 * base of the 2KB buffer s(user processes only use 2K)
907	 */
908	u64 piobufbase;
909	u32 pio2k_bufbase;
910
911	/* these are the "32 bit" regs */
912
913	/* number of GUIDs in the flash for this interface */
914	u32 nguid;
915	/*
916	 * the following two are 32-bit bitmasks, but {test,clear,set}_bit
917	 * all expect bit fields to be "unsigned long"
918	 */
919	unsigned long rcvctrl; /* shadow per device rcvctrl */
920	unsigned long sendctrl; /* shadow per device sendctrl */
921
922	/* value we put in kr_rcvhdrcnt */
923	u32 rcvhdrcnt;
924	/* value we put in kr_rcvhdrsize */
925	u32 rcvhdrsize;
926	/* value we put in kr_rcvhdrentsize */
927	u32 rcvhdrentsize;
928	/* kr_ctxtcnt value */
929	u32 ctxtcnt;
930	/* kr_pagealign value */
931	u32 palign;
932	/* number of "2KB" PIO buffers */
933	u32 piobcnt2k;
934	/* size in bytes of "2KB" PIO buffers */
935	u32 piosize2k;
936	/* max usable size in dwords of a "2KB" PIO buffer before going "4KB" */
937	u32 piosize2kmax_dwords;
938	/* number of "4KB" PIO buffers */
939	u32 piobcnt4k;
940	/* size in bytes of "4KB" PIO buffers */
941	u32 piosize4k;
942	/* kr_rcvegrbase value */
943	u32 rcvegrbase;
944	/* kr_rcvtidbase value */
945	u32 rcvtidbase;
946	/* kr_rcvtidcnt value */
947	u32 rcvtidcnt;
948	/* kr_userregbase */
949	u32 uregbase;
950	/* shadow the control register contents */
951	u32 control;
952
953	/* chip address space used by 4k pio buffers */
954	u32 align4k;
955	/* size of each rcvegrbuffer */
956	u16 rcvegrbufsize;
957	/* log2 of above */
958	u16 rcvegrbufsize_shift;
959	/* localbus width (1, 2,4,8,16,32) from config space  */
960	u32 lbus_width;
961	/* localbus speed in MHz */
962	u32 lbus_speed;
963	int unit; /* unit # of this chip */
964
965	/* start of CHIP_SPEC move to chipspec, but need code changes */
966	/* low and high portions of MSI capability/vector */
967	u32 msi_lo;
968	/* saved after PCIe init for restore after reset */
969	u32 msi_hi;
970	/* MSI data (vector) saved for restore */
971	u16 msi_data;
972	/* so we can rewrite it after a chip reset */
973	u32 pcibar0;
974	/* so we can rewrite it after a chip reset */
975	u32 pcibar1;
976	u64 rhdrhead_intr_off;
977
978	/*
979	 * ASCII serial number, from flash, large enough for original
980	 * all digit strings, and longer QLogic serial number format
981	 */
982	u8 serial[16];
983	/* human readable board version */
984	u8 boardversion[96];
985	u8 lbus_info[32]; /* human readable localbus info */
986	/* chip major rev, from qib_revision */
987	u8 majrev;
988	/* chip minor rev, from qib_revision */
989	u8 minrev;
990
991	/* Misc small ints */
992	/* Number of physical ports available */
993	u8 num_pports;
994	/* Lowest context number which can be used by user processes */
995	u8 first_user_ctxt;
996	u8 n_krcv_queues;
997	u8 qpn_mask;
998	u8 skip_kctxt_mask;
999
1000	u16 rhf_offset; /* offset of RHF within receive header entry */
1001
1002	/*
1003	 * GPIO pins for twsi-connected devices, and device code for eeprom
1004	 */
1005	u8 gpio_sda_num;
1006	u8 gpio_scl_num;
1007	u8 twsi_eeprom_dev;
1008	u8 board_atten;
1009
1010	/* Support (including locks) for EEPROM logging of errors and time */
1011	/* control access to actual counters, timer */
1012	spinlock_t eep_st_lock;
1013	/* control high-level access to EEPROM */
1014	struct mutex eep_lock;
1015	uint64_t traffic_wds;
1016	/* active time is kept in seconds, but logged in hours */
1017	atomic_t active_time;
1018	/* Below are nominal shadow of EEPROM, new since last EEPROM update */
1019	uint8_t eep_st_errs[QIB_EEP_LOG_CNT];
1020	uint8_t eep_st_new_errs[QIB_EEP_LOG_CNT];
1021	uint16_t eep_hrs;
1022	/*
1023	 * masks for which bits of errs, hwerrs that cause
1024	 * each of the counters to increment.
1025	 */
1026	struct qib_eep_log_mask eep_st_masks[QIB_EEP_LOG_CNT];
1027	struct qib_diag_client *diag_client;
1028	spinlock_t qib_diag_trans_lock; /* protect diag observer ops */
1029	struct diag_observer_list_elt *diag_observer_list;
1030
1031	u8 psxmitwait_supported;
1032	/* cycle length of PS* counters in HW (in picoseconds) */
1033	u16 psxmitwait_check_rate;
1034	/* high volume overflow errors defered to tasklet */
1035	struct tasklet_struct error_tasklet;
1036};
1037
1038/* hol_state values */
1039#define QIB_HOL_UP       0
1040#define QIB_HOL_INIT     1
1041
1042#define QIB_SDMA_SENDCTRL_OP_ENABLE    (1U << 0)
1043#define QIB_SDMA_SENDCTRL_OP_INTENABLE (1U << 1)
1044#define QIB_SDMA_SENDCTRL_OP_HALT      (1U << 2)
1045#define QIB_SDMA_SENDCTRL_OP_CLEANUP   (1U << 3)
1046#define QIB_SDMA_SENDCTRL_OP_DRAIN     (1U << 4)
1047
1048/* operation types for f_txchk_change() */
1049#define TXCHK_CHG_TYPE_DIS1  3
1050#define TXCHK_CHG_TYPE_ENAB1 2
1051#define TXCHK_CHG_TYPE_KERN  1
1052#define TXCHK_CHG_TYPE_USER  0
1053
1054#define QIB_CHASE_TIME msecs_to_jiffies(145)
1055#define QIB_CHASE_DIS_TIME msecs_to_jiffies(160)
1056
1057/* Private data for file operations */
1058struct qib_filedata {
1059	struct qib_ctxtdata *rcd;
1060	unsigned subctxt;
1061	unsigned tidcursor;
1062	struct qib_user_sdma_queue *pq;
1063	int rec_cpu_num; /* for cpu affinity; -1 if none */
1064};
1065
1066extern struct list_head qib_dev_list;
1067extern spinlock_t qib_devs_lock;
1068extern struct qib_devdata *qib_lookup(int unit);
1069extern u32 qib_cpulist_count;
1070extern unsigned long *qib_cpulist;
1071
1072extern unsigned qib_wc_pat;
1073int qib_init(struct qib_devdata *, int);
1074int init_chip_wc_pat(struct qib_devdata *dd, u32);
1075int qib_enable_wc(struct qib_devdata *dd);
1076void qib_disable_wc(struct qib_devdata *dd);
1077int qib_count_units(int *npresentp, int *nupp);
1078int qib_count_active_units(void);
1079
1080int qib_cdev_init(int minor, const char *name,
1081		  const struct file_operations *fops,
1082		  struct cdev **cdevp, struct device **devp);
1083void qib_cdev_cleanup(struct cdev **cdevp, struct device **devp);
1084int qib_dev_init(void);
1085void qib_dev_cleanup(void);
1086
1087int qib_diag_add(struct qib_devdata *);
1088void qib_diag_remove(struct qib_devdata *);
1089void qib_handle_e_ibstatuschanged(struct qib_pportdata *, u64);
1090void qib_sdma_update_tail(struct qib_pportdata *, u16); /* hold sdma_lock */
1091
1092int qib_decode_err(struct qib_devdata *dd, char *buf, size_t blen, u64 err);
1093void qib_bad_intrstatus(struct qib_devdata *);
1094void qib_handle_urcv(struct qib_devdata *, u64);
1095
1096/* clean up any per-chip chip-specific stuff */
1097void qib_chip_cleanup(struct qib_devdata *);
1098/* clean up any chip type-specific stuff */
1099void qib_chip_done(void);
1100
1101/* check to see if we have to force ordering for write combining */
1102int qib_unordered_wc(void);
1103void qib_pio_copy(void __iomem *to, const void *from, size_t count);
1104
1105void qib_disarm_piobufs(struct qib_devdata *, unsigned, unsigned);
1106int qib_disarm_piobufs_ifneeded(struct qib_ctxtdata *);
1107void qib_disarm_piobufs_set(struct qib_devdata *, unsigned long *, unsigned);
1108void qib_cancel_sends(struct qib_pportdata *);
1109
1110int qib_create_rcvhdrq(struct qib_devdata *, struct qib_ctxtdata *);
1111int qib_setup_eagerbufs(struct qib_ctxtdata *);
1112void qib_set_ctxtcnt(struct qib_devdata *);
1113int qib_create_ctxts(struct qib_devdata *dd);
1114struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *, u32);
1115void qib_init_pportdata(struct qib_pportdata *, struct qib_devdata *, u8, u8);
1116void qib_free_ctxtdata(struct qib_devdata *, struct qib_ctxtdata *);
1117
1118u32 qib_kreceive(struct qib_ctxtdata *, u32 *, u32 *);
1119int qib_reset_device(int);
1120int qib_wait_linkstate(struct qib_pportdata *, u32, int);
1121int qib_set_linkstate(struct qib_pportdata *, u8);
1122int qib_set_mtu(struct qib_pportdata *, u16);
1123int qib_set_lid(struct qib_pportdata *, u32, u8);
1124void qib_hol_down(struct qib_pportdata *);
1125void qib_hol_init(struct qib_pportdata *);
1126void qib_hol_up(struct qib_pportdata *);
1127void qib_hol_event(unsigned long);
1128void qib_disable_after_error(struct qib_devdata *);
1129int qib_set_uevent_bits(struct qib_pportdata *, const int);
1130
1131/* for use in system calls, where we want to know device type, etc. */
1132#define ctxt_fp(fp) \
1133	(((struct qib_filedata *)(fp)->private_data)->rcd)
1134#define subctxt_fp(fp) \
1135	(((struct qib_filedata *)(fp)->private_data)->subctxt)
1136#define tidcursor_fp(fp) \
1137	(((struct qib_filedata *)(fp)->private_data)->tidcursor)
1138#define user_sdma_queue_fp(fp) \
1139	(((struct qib_filedata *)(fp)->private_data)->pq)
1140
1141static inline struct qib_devdata *dd_from_ppd(struct qib_pportdata *ppd)
1142{
1143	return ppd->dd;
1144}
1145
1146static inline struct qib_devdata *dd_from_dev(struct qib_ibdev *dev)
1147{
1148	return container_of(dev, struct qib_devdata, verbs_dev);
1149}
1150
1151static inline struct qib_devdata *dd_from_ibdev(struct ib_device *ibdev)
1152{
1153	return dd_from_dev(to_idev(ibdev));
1154}
1155
1156static inline struct qib_pportdata *ppd_from_ibp(struct qib_ibport *ibp)
1157{
1158	return container_of(ibp, struct qib_pportdata, ibport_data);
1159}
1160
1161static inline struct qib_ibport *to_iport(struct ib_device *ibdev, u8 port)
1162{
1163	struct qib_devdata *dd = dd_from_ibdev(ibdev);
1164	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
1165
1166	WARN_ON(pidx >= dd->num_pports);
1167	return &dd->pport[pidx].ibport_data;
1168}
1169
1170/*
1171 * values for dd->flags (_device_ related flags) and
1172 */
1173#define QIB_HAS_LINK_LATENCY  0x1 /* supports link latency (IB 1.2) */
1174#define QIB_INITTED           0x2 /* chip and driver up and initted */
1175#define QIB_DOING_RESET       0x4  /* in the middle of doing chip reset */
1176#define QIB_PRESENT           0x8  /* chip accesses can be done */
1177#define QIB_PIO_FLUSH_WC      0x10 /* Needs Write combining flush for PIO */
1178#define QIB_HAS_THRESH_UPDATE 0x40
1179#define QIB_HAS_SDMA_TIMEOUT  0x80
1180#define QIB_USE_SPCL_TRIG     0x100 /* SpecialTrigger launch enabled */
1181#define QIB_NODMA_RTAIL       0x200 /* rcvhdrtail register DMA enabled */
1182#define QIB_HAS_INTX          0x800 /* Supports INTx interrupts */
1183#define QIB_HAS_SEND_DMA      0x1000 /* Supports Send DMA */
1184#define QIB_HAS_VLSUPP        0x2000 /* Supports multiple VLs; PBC different */
1185#define QIB_HAS_HDRSUPP       0x4000 /* Supports header suppression */
1186#define QIB_BADINTR           0x8000 /* severe interrupt problems */
1187#define QIB_DCA_ENABLED       0x10000 /* Direct Cache Access enabled */
1188#define QIB_HAS_QSFP          0x20000 /* device (card instance) has QSFP */
1189
1190/*
1191 * values for ppd->lflags (_ib_port_ related flags)
1192 */
1193#define QIBL_LINKV             0x1 /* IB link state valid */
1194#define QIBL_LINKDOWN          0x8 /* IB link is down */
1195#define QIBL_LINKINIT          0x10 /* IB link level is up */
1196#define QIBL_LINKARMED         0x20 /* IB link is ARMED */
1197#define QIBL_LINKACTIVE        0x40 /* IB link is ACTIVE */
1198/* leave a gap for more IB-link state */
1199#define QIBL_IB_AUTONEG_INPROG 0x1000 /* non-IBTA DDR/QDR neg active */
1200#define QIBL_IB_AUTONEG_FAILED 0x2000 /* non-IBTA DDR/QDR neg failed */
1201#define QIBL_IB_LINK_DISABLED  0x4000 /* Linkdown-disable forced,
1202				       * Do not try to bring up */
1203#define QIBL_IB_FORCE_NOTIFY   0x8000 /* force notify on next ib change */
1204
1205/* IB dword length mask in PBC (lower 11 bits); same for all chips */
1206#define QIB_PBC_LENGTH_MASK                     ((1 << 11) - 1)
1207
1208
1209/* ctxt_flag bit offsets */
1210		/* waiting for a packet to arrive */
1211#define QIB_CTXT_WAITING_RCV   2
1212		/* master has not finished initializing */
1213#define QIB_CTXT_MASTER_UNINIT 4
1214		/* waiting for an urgent packet to arrive */
1215#define QIB_CTXT_WAITING_URG 5
1216
1217/* free up any allocated data at closes */
1218void qib_free_data(struct qib_ctxtdata *dd);
1219void qib_chg_pioavailkernel(struct qib_devdata *, unsigned, unsigned,
1220			    u32, struct qib_ctxtdata *);
1221struct qib_devdata *qib_init_iba7322_funcs(struct pci_dev *,
1222					   const struct pci_device_id *);
1223struct qib_devdata *qib_init_iba7220_funcs(struct pci_dev *,
1224					   const struct pci_device_id *);
1225struct qib_devdata *qib_init_iba6120_funcs(struct pci_dev *,
1226					   const struct pci_device_id *);
1227void qib_free_devdata(struct qib_devdata *);
1228struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra);
1229
1230#define QIB_TWSI_NO_DEV 0xFF
1231/* Below qib_twsi_ functions must be called with eep_lock held */
1232int qib_twsi_reset(struct qib_devdata *dd);
1233int qib_twsi_blk_rd(struct qib_devdata *dd, int dev, int addr, void *buffer,
1234		    int len);
1235int qib_twsi_blk_wr(struct qib_devdata *dd, int dev, int addr,
1236		    const void *buffer, int len);
1237void qib_get_eeprom_info(struct qib_devdata *);
1238int qib_update_eeprom_log(struct qib_devdata *dd);
1239void qib_inc_eeprom_err(struct qib_devdata *dd, u32 eidx, u32 incr);
1240void qib_dump_lookup_output_queue(struct qib_devdata *);
1241void qib_force_pio_avail_update(struct qib_devdata *);
1242void qib_clear_symerror_on_linkup(unsigned long opaque);
1243
1244/*
1245 * Set LED override, only the two LSBs have "public" meaning, but
1246 * any non-zero value substitutes them for the Link and LinkTrain
1247 * LED states.
1248 */
1249#define QIB_LED_PHYS 1 /* Physical (linktraining) GREEN LED */
1250#define QIB_LED_LOG 2  /* Logical (link) YELLOW LED */
1251void qib_set_led_override(struct qib_pportdata *ppd, unsigned int val);
1252
1253/* send dma routines */
1254int qib_setup_sdma(struct qib_pportdata *);
1255void qib_teardown_sdma(struct qib_pportdata *);
1256void __qib_sdma_intr(struct qib_pportdata *);
1257void qib_sdma_intr(struct qib_pportdata *);
1258int qib_sdma_verbs_send(struct qib_pportdata *, struct qib_sge_state *,
1259			u32, struct qib_verbs_txreq *);
1260/* ppd->sdma_lock should be locked before calling this. */
1261int qib_sdma_make_progress(struct qib_pportdata *dd);
1262
1263/* must be called under qib_sdma_lock */
1264static inline u16 qib_sdma_descq_freecnt(const struct qib_pportdata *ppd)
1265{
1266	return ppd->sdma_descq_cnt -
1267		(ppd->sdma_descq_added - ppd->sdma_descq_removed) - 1;
1268}
1269
1270static inline int __qib_sdma_running(struct qib_pportdata *ppd)
1271{
1272	return ppd->sdma_state.current_state == qib_sdma_state_s99_running;
1273}
1274int qib_sdma_running(struct qib_pportdata *);
1275
1276void __qib_sdma_process_event(struct qib_pportdata *, enum qib_sdma_events);
1277void qib_sdma_process_event(struct qib_pportdata *, enum qib_sdma_events);
1278
1279/*
1280 * number of words used for protocol header if not set by qib_userinit();
1281 */
1282#define QIB_DFLT_RCVHDRSIZE 9
1283
1284/*
1285 * We need to be able to handle an IB header of at least 24 dwords.
1286 * We need the rcvhdrq large enough to handle largest IB header, but
1287 * still have room for a 2KB MTU standard IB packet.
1288 * Additionally, some processor/memory controller combinations
1289 * benefit quite strongly from having the DMA'ed data be cacheline
1290 * aligned and a cacheline multiple, so we set the size to 32 dwords
1291 * (2 64-byte primary cachelines for pretty much all processors of
1292 * interest).  The alignment hurts nothing, other than using somewhat
1293 * more memory.
1294 */
1295#define QIB_RCVHDR_ENTSIZE 32
1296
1297int qib_get_user_pages(unsigned long, size_t, struct page **);
1298void qib_release_user_pages(struct page **, size_t);
1299int qib_eeprom_read(struct qib_devdata *, u8, void *, int);
1300int qib_eeprom_write(struct qib_devdata *, u8, const void *, int);
1301u32 __iomem *qib_getsendbuf_range(struct qib_devdata *, u32 *, u32, u32);
1302void qib_sendbuf_done(struct qib_devdata *, unsigned);
1303
1304static inline void qib_clear_rcvhdrtail(const struct qib_ctxtdata *rcd)
1305{
1306	*((u64 *) rcd->rcvhdrtail_kvaddr) = 0ULL;
1307}
1308
1309static inline u32 qib_get_rcvhdrtail(const struct qib_ctxtdata *rcd)
1310{
1311	/*
1312	 * volatile because it's a DMA target from the chip, routine is
1313	 * inlined, and don't want register caching or reordering.
1314	 */
1315	return (u32) le64_to_cpu(
1316		*((volatile __le64 *)rcd->rcvhdrtail_kvaddr)); /* DMA'ed */
1317}
1318
1319static inline u32 qib_get_hdrqtail(const struct qib_ctxtdata *rcd)
1320{
1321	const struct qib_devdata *dd = rcd->dd;
1322	u32 hdrqtail;
1323
1324	if (dd->flags & QIB_NODMA_RTAIL) {
1325		__le32 *rhf_addr;
1326		u32 seq;
1327
1328		rhf_addr = (__le32 *) rcd->rcvhdrq +
1329			rcd->head + dd->rhf_offset;
1330		seq = qib_hdrget_seq(rhf_addr);
1331		hdrqtail = rcd->head;
1332		if (seq == rcd->seq_cnt)
1333			hdrqtail++;
1334	} else
1335		hdrqtail = qib_get_rcvhdrtail(rcd);
1336
1337	return hdrqtail;
1338}
1339
1340/*
1341 * sysfs interface.
1342 */
1343
1344extern const char ib_qib_version[];
1345
1346int qib_device_create(struct qib_devdata *);
1347void qib_device_remove(struct qib_devdata *);
1348
1349int qib_create_port_files(struct ib_device *ibdev, u8 port_num,
1350			  struct kobject *kobj);
1351int qib_verbs_register_sysfs(struct qib_devdata *);
1352void qib_verbs_unregister_sysfs(struct qib_devdata *);
1353/* Hook for sysfs read of QSFP */
1354extern int qib_qsfp_dump(struct qib_pportdata *ppd, char *buf, int len);
1355
1356int __init qib_init_qibfs(void);
1357int __exit qib_exit_qibfs(void);
1358
1359int qibfs_add(struct qib_devdata *);
1360int qibfs_remove(struct qib_devdata *);
1361
1362int qib_pcie_init(struct pci_dev *, const struct pci_device_id *);
1363int qib_pcie_ddinit(struct qib_devdata *, struct pci_dev *,
1364		    const struct pci_device_id *);
1365void qib_pcie_ddcleanup(struct qib_devdata *);
1366int qib_pcie_params(struct qib_devdata *, u32, u32 *, struct qib_msix_entry *);
1367int qib_reinit_intr(struct qib_devdata *);
1368void qib_enable_intx(struct pci_dev *);
1369void qib_nomsi(struct qib_devdata *);
1370void qib_nomsix(struct qib_devdata *);
1371void qib_pcie_getcmd(struct qib_devdata *, u16 *, u8 *, u8 *);
1372void qib_pcie_reenable(struct qib_devdata *, u16, u8, u8);
1373
1374/*
1375 * dma_addr wrappers - all 0's invalid for hw
1376 */
1377dma_addr_t qib_map_page(struct pci_dev *, struct page *, unsigned long,
1378			  size_t, int);
1379const char *qib_get_unit_name(int unit);
1380
1381/*
1382 * Flush write combining store buffers (if present) and perform a write
1383 * barrier.
1384 */
1385#if defined(CONFIG_X86_64)
1386#define qib_flush_wc() asm volatile("sfence" : : : "memory")
1387#else
1388#define qib_flush_wc() wmb() /* no reorder around wc flush */
1389#endif
1390
1391/* global module parameter variables */
1392extern unsigned qib_ibmtu;
1393extern ushort qib_cfgctxts;
1394extern ushort qib_num_cfg_vls;
1395extern ushort qib_mini_init; /* If set, do few (ideally 0) writes to chip */
1396extern unsigned qib_n_krcv_queues;
1397extern unsigned qib_sdma_fetch_arb;
1398extern unsigned qib_compat_ddr_negotiate;
1399extern int qib_special_trigger;
1400
1401extern struct mutex qib_mutex;
1402
1403/* Number of seconds before our card status check...  */
1404#define STATUS_TIMEOUT 60
1405
1406#define QIB_DRV_NAME            "ib_qib"
1407#define QIB_USER_MINOR_BASE     0
1408#define QIB_TRACE_MINOR         127
1409#define QIB_DIAGPKT_MINOR       128
1410#define QIB_DIAG_MINOR_BASE     129
1411#define QIB_NMINORS             255
1412
1413#define PCI_VENDOR_ID_PATHSCALE 0x1fc1
1414#define PCI_VENDOR_ID_QLOGIC 0x1077
1415#define PCI_DEVICE_ID_QLOGIC_IB_6120 0x10
1416#define PCI_DEVICE_ID_QLOGIC_IB_7220 0x7220
1417#define PCI_DEVICE_ID_QLOGIC_IB_7322 0x7322
1418
1419/*
1420 * qib_early_err is used (only!) to print early errors before devdata is
1421 * allocated, or when dd->pcidev may not be valid, and at the tail end of
1422 * cleanup when devdata may have been freed, etc.  qib_dev_porterr is
1423 * the same as qib_dev_err, but is used when the message really needs
1424 * the IB port# to be definitive as to what's happening..
1425 * All of these go to the trace log, and the trace log entry is done
1426 * first to avoid possible serial port delays from printk.
1427 */
1428#define qib_early_err(dev, fmt, ...) \
1429	do { \
1430		dev_err(dev, fmt, ##__VA_ARGS__); \
1431	} while (0)
1432
1433#define qib_dev_err(dd, fmt, ...) \
1434	do { \
1435		dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
1436			qib_get_unit_name((dd)->unit), ##__VA_ARGS__); \
1437	} while (0)
1438
1439#define qib_dev_porterr(dd, port, fmt, ...) \
1440	do { \
1441		dev_err(&(dd)->pcidev->dev, "%s: IB%u:%u " fmt, \
1442			qib_get_unit_name((dd)->unit), (dd)->unit, (port), \
1443			##__VA_ARGS__); \
1444	} while (0)
1445
1446#define qib_devinfo(pcidev, fmt, ...) \
1447	do { \
1448		dev_info(&(pcidev)->dev, fmt, ##__VA_ARGS__); \
1449	} while (0)
1450
1451/*
1452 * this is used for formatting hw error messages...
1453 */
1454struct qib_hwerror_msgs {
1455	u64 mask;
1456	const char *msg;
1457	size_t sz;
1458};
1459
1460#define QLOGIC_IB_HWE_MSG(a, b) { .mask = a, .msg = b }
1461
1462/* in qib_intr.c... */
1463void qib_format_hwerrors(u64 hwerrs,
1464			 const struct qib_hwerror_msgs *hwerrmsgs,
1465			 size_t nhwerrmsgs, char *msg, size_t lmsg);
1466#endif                          /* _QIB_KERNEL_H */
1467