1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (C) 2012-2014 Intel Corporation
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29 #ifndef __NVME_PRIVATE_H__
30 #define __NVME_PRIVATE_H__
31
32 #include <sys/param.h>
33 #include <sys/bio.h>
34 #include <sys/bus.h>
35 #include <sys/counter.h>
36 #include <sys/kernel.h>
37 #include <sys/lock.h>
38 #include <sys/malloc.h>
39 #include <sys/memdesc.h>
40 #include <sys/module.h>
41 #include <sys/mutex.h>
42 #include <sys/rman.h>
43 #include <sys/systm.h>
44 #include <sys/taskqueue.h>
45
46 #include <vm/uma.h>
47
48 #include <machine/bus.h>
49
50 #include "nvme.h"
51
52 #define DEVICE2SOFTC(dev) ((struct nvme_controller *) device_get_softc(dev))
53
54 MALLOC_DECLARE(M_NVME);
55
56 #define IDT32_PCI_ID 0x80d0111d /* 32 channel board */
57 #define IDT8_PCI_ID 0x80d2111d /* 8 channel board */
58
59 #define NVME_ADMIN_TRACKERS (16)
60 #define NVME_ADMIN_ENTRIES (128)
61
62 /*
63 * NVME_IO_ENTRIES defines the size of an I/O qpair's submission and completion
64 * queues, while NVME_IO_TRACKERS defines the maximum number of I/O that we
65 * will allow outstanding on an I/O qpair at any time. The only advantage in
66 * having IO_ENTRIES > IO_TRACKERS is for debugging purposes - when dumping
67 * the contents of the submission and completion queues, it will show a longer
68 * history of data.
69 */
70 #define NVME_IO_ENTRIES (256)
71 #define NVME_IO_TRACKERS (128)
72 #define NVME_MIN_IO_TRACKERS (4)
73 #define NVME_MAX_IO_TRACKERS (1024)
74
75 #define NVME_INT_COAL_TIME (0) /* disabled */
76 #define NVME_INT_COAL_THRESHOLD (0) /* 0-based */
77
78 #define NVME_MAX_NAMESPACES (16)
79 #define NVME_MAX_CONSUMERS (2)
80 #define NVME_MAX_ASYNC_EVENTS (8)
81
82 #define NVME_ADMIN_TIMEOUT_PERIOD (60) /* in seconds */
83 #define NVME_DEFAULT_TIMEOUT_PERIOD (30) /* in seconds */
84 #define NVME_MIN_TIMEOUT_PERIOD (5)
85 #define NVME_MAX_TIMEOUT_PERIOD (120)
86
87 #define NVME_DEFAULT_RETRY_COUNT (4)
88
89 /* Maximum log page size to fetch for AERs. */
90 #define NVME_MAX_AER_LOG_SIZE (4096)
91
92 /*
93 * Define CACHE_LINE_SIZE here for older FreeBSD versions that do not define
94 * it.
95 */
96 #ifndef CACHE_LINE_SIZE
97 #define CACHE_LINE_SIZE (64)
98 #endif
99
100 #define NVME_GONE 0xfffffffful
101
102 extern int32_t nvme_retry_count;
103 extern bool nvme_verbose_cmd_dump;
104
105 struct nvme_completion_poll_status {
106 struct nvme_completion cpl;
107 int done;
108 };
109
110 struct nvme_request {
111 struct nvme_command cmd;
112 struct nvme_qpair *qpair;
113 struct memdesc payload;
114 nvme_cb_fn_t cb_fn;
115 void *cb_arg;
116 int32_t retries;
117 bool payload_valid;
118 bool timeout;
119 bool spare[2]; /* Future use */
120 STAILQ_ENTRY(nvme_request) stailq;
121 };
122
123 struct nvme_async_event_request {
124 struct nvme_controller *ctrlr;
125 struct nvme_request *req;
126 struct task task;
127 struct mtx mtx;
128 struct nvme_completion cpl;
129 uint32_t log_page_id;
130 uint32_t log_page_size;
131 uint8_t log_page_buffer[NVME_MAX_AER_LOG_SIZE];
132 };
133
134 struct nvme_tracker {
135 TAILQ_ENTRY(nvme_tracker) tailq;
136 struct nvme_request *req;
137 struct nvme_qpair *qpair;
138 sbintime_t deadline;
139 bus_dmamap_t payload_dma_map;
140 uint16_t cid;
141
142 uint64_t *prp;
143 bus_addr_t prp_bus_addr;
144 };
145
146 enum nvme_recovery {
147 RECOVERY_NONE = 0, /* Normal operations */
148 RECOVERY_WAITING, /* waiting for the reset to complete */
149 };
150 struct nvme_qpair {
151 struct nvme_controller *ctrlr;
152 uint32_t id;
153 int domain;
154 int cpu;
155
156 uint16_t vector;
157 int rid;
158 struct resource *res;
159 void *tag;
160
161 struct callout timer; /* recovery lock */
162 bool timer_armed; /* recovery lock */
163 enum nvme_recovery recovery_state; /* recovery lock */
164
165 uint32_t num_entries;
166 uint32_t num_trackers;
167 uint32_t sq_tdbl_off;
168 uint32_t cq_hdbl_off;
169
170 uint32_t phase;
171 uint32_t sq_head;
172 uint32_t sq_tail;
173 uint32_t cq_head;
174
175 int64_t num_cmds;
176 int64_t num_intr_handler_calls;
177 int64_t num_retries;
178 int64_t num_failures;
179 int64_t num_ignored;
180 int64_t num_recovery_nolock;
181
182 struct nvme_command *cmd;
183 struct nvme_completion *cpl;
184
185 bus_dma_tag_t dma_tag;
186 bus_dma_tag_t dma_tag_payload;
187
188 bus_dmamap_t queuemem_map;
189 uint64_t cmd_bus_addr;
190 uint64_t cpl_bus_addr;
191
192 TAILQ_HEAD(, nvme_tracker) free_tr;
193 TAILQ_HEAD(, nvme_tracker) outstanding_tr;
194 STAILQ_HEAD(, nvme_request) queued_req;
195
196 struct nvme_tracker **act_tr;
197
198 struct mtx_padalign lock;
199 struct mtx_padalign recovery;
200 } __aligned(CACHE_LINE_SIZE);
201
202 struct nvme_namespace {
203 struct nvme_controller *ctrlr;
204 struct nvme_namespace_data data;
205 uint32_t id;
206 uint32_t flags;
207 struct cdev *cdev;
208 void *cons_cookie[NVME_MAX_CONSUMERS];
209 uint32_t boundary;
210 struct mtx lock;
211 };
212
213 /*
214 * One of these per allocated PCI device.
215 */
216 struct nvme_controller {
217 device_t dev;
218
219 struct mtx lock;
220 int domain;
221 uint32_t ready_timeout_in_ms;
222 uint32_t quirks;
223 #define QUIRK_DELAY_B4_CHK_RDY 1 /* Can't touch MMIO on disable */
224 #define QUIRK_DISABLE_TIMEOUT 2 /* Disable broken completion timeout feature */
225 #define QUIRK_INTEL_ALIGNMENT 4 /* Pre NVMe 1.3 performance alignment */
226 #define QUIRK_AHCI 8 /* Attached via AHCI redirect */
227
228 bus_space_tag_t bus_tag;
229 bus_space_handle_t bus_handle;
230 int resource_id;
231 struct resource *resource;
232
233 /*
234 * The NVMe spec allows for the MSI-X table to be placed in BAR 4/5,
235 * separate from the control registers which are in BAR 0/1. These
236 * members track the mapping of BAR 4/5 for that reason.
237 */
238 int bar4_resource_id;
239 struct resource *bar4_resource;
240
241 int msi_count;
242 uint32_t enable_aborts;
243
244 uint32_t num_io_queues;
245 uint32_t max_hw_pend_io;
246
247 /* Fields for tracking progress during controller initialization. */
248 struct intr_config_hook config_hook;
249 uint32_t ns_identified;
250 uint32_t queues_created;
251
252 struct task reset_task;
253 struct taskqueue *taskqueue;
254
255 /* For shared legacy interrupt. */
256 int rid;
257 struct resource *res;
258 void *tag;
259
260 /** maximum i/o size in bytes */
261 uint32_t max_xfer_size;
262
263 /** LO and HI capacity mask */
264 uint32_t cap_lo;
265 uint32_t cap_hi;
266
267 /** Page size and log2(page_size) - 12 that we're currently using */
268 uint32_t page_size;
269 uint32_t mps;
270
271 /** interrupt coalescing time period (in microseconds) */
272 uint32_t int_coal_time;
273
274 /** interrupt coalescing threshold */
275 uint32_t int_coal_threshold;
276
277 /** timeout period in seconds */
278 uint32_t admin_timeout_period;
279 uint32_t timeout_period;
280
281 /** doorbell stride */
282 uint32_t dstrd;
283
284 struct nvme_qpair adminq;
285 struct nvme_qpair *ioq;
286
287 struct nvme_registers *regs;
288
289 struct nvme_controller_data cdata;
290 struct nvme_namespace ns[NVME_MAX_NAMESPACES];
291
292 struct cdev *cdev;
293
294 /** bit mask of event types currently enabled for async events */
295 uint32_t async_event_config;
296
297 uint32_t num_aers;
298 struct nvme_async_event_request aer[NVME_MAX_ASYNC_EVENTS];
299
300 void *cons_cookie[NVME_MAX_CONSUMERS];
301
302 uint32_t is_resetting;
303 uint32_t notification_sent;
304 u_int fail_on_reset;
305
306 bool is_failed;
307 bool is_failed_admin;
308 bool is_dying;
309 bool isr_warned;
310 bool is_initialized;
311
312 /* Host Memory Buffer */
313 int hmb_nchunks;
314 size_t hmb_chunk;
315 bus_dma_tag_t hmb_tag;
316 struct nvme_hmb_chunk {
317 bus_dmamap_t hmbc_map;
318 void *hmbc_vaddr;
319 uint64_t hmbc_paddr;
320 } *hmb_chunks;
321 bus_dma_tag_t hmb_desc_tag;
322 bus_dmamap_t hmb_desc_map;
323 struct nvme_hmb_desc *hmb_desc_vaddr;
324 uint64_t hmb_desc_paddr;
325
326 /* Statistics */
327 counter_u64_t alignment_splits;
328 };
329
330 #define nvme_mmio_offsetof(reg) \
331 offsetof(struct nvme_registers, reg)
332
333 #define nvme_mmio_read_4(sc, reg) \
334 bus_space_read_4((sc)->bus_tag, (sc)->bus_handle, \
335 nvme_mmio_offsetof(reg))
336
337 #define nvme_mmio_write_4(sc, reg, val) \
338 bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \
339 nvme_mmio_offsetof(reg), val)
340
341 #define nvme_mmio_write_8(sc, reg, val) \
342 do { \
343 bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \
344 nvme_mmio_offsetof(reg), val & 0xFFFFFFFF); \
345 bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \
346 nvme_mmio_offsetof(reg)+4, \
347 (val & 0xFFFFFFFF00000000ULL) >> 32); \
348 } while (0);
349
350 #define nvme_printf(ctrlr, fmt, args...) \
351 device_printf(ctrlr->dev, fmt, ##args)
352
353 void nvme_ns_test(struct nvme_namespace *ns, u_long cmd, caddr_t arg);
354
355 void nvme_ctrlr_cmd_identify_controller(struct nvme_controller *ctrlr,
356 void *payload,
357 nvme_cb_fn_t cb_fn, void *cb_arg);
358 void nvme_ctrlr_cmd_identify_namespace(struct nvme_controller *ctrlr,
359 uint32_t nsid, void *payload,
360 nvme_cb_fn_t cb_fn, void *cb_arg);
361 void nvme_ctrlr_cmd_set_interrupt_coalescing(struct nvme_controller *ctrlr,
362 uint32_t microseconds,
363 uint32_t threshold,
364 nvme_cb_fn_t cb_fn,
365 void *cb_arg);
366 void nvme_ctrlr_cmd_get_error_page(struct nvme_controller *ctrlr,
367 struct nvme_error_information_entry *payload,
368 uint32_t num_entries, /* 0 = max */
369 nvme_cb_fn_t cb_fn,
370 void *cb_arg);
371 void nvme_ctrlr_cmd_get_health_information_page(struct nvme_controller *ctrlr,
372 uint32_t nsid,
373 struct nvme_health_information_page *payload,
374 nvme_cb_fn_t cb_fn,
375 void *cb_arg);
376 void nvme_ctrlr_cmd_get_firmware_page(struct nvme_controller *ctrlr,
377 struct nvme_firmware_page *payload,
378 nvme_cb_fn_t cb_fn,
379 void *cb_arg);
380 void nvme_ctrlr_cmd_create_io_cq(struct nvme_controller *ctrlr,
381 struct nvme_qpair *io_que,
382 nvme_cb_fn_t cb_fn, void *cb_arg);
383 void nvme_ctrlr_cmd_create_io_sq(struct nvme_controller *ctrlr,
384 struct nvme_qpair *io_que,
385 nvme_cb_fn_t cb_fn, void *cb_arg);
386 void nvme_ctrlr_cmd_delete_io_cq(struct nvme_controller *ctrlr,
387 struct nvme_qpair *io_que,
388 nvme_cb_fn_t cb_fn, void *cb_arg);
389 void nvme_ctrlr_cmd_delete_io_sq(struct nvme_controller *ctrlr,
390 struct nvme_qpair *io_que,
391 nvme_cb_fn_t cb_fn, void *cb_arg);
392 void nvme_ctrlr_cmd_set_num_queues(struct nvme_controller *ctrlr,
393 uint32_t num_queues, nvme_cb_fn_t cb_fn,
394 void *cb_arg);
395 void nvme_ctrlr_cmd_set_async_event_config(struct nvme_controller *ctrlr,
396 uint32_t state,
397 nvme_cb_fn_t cb_fn, void *cb_arg);
398 void nvme_ctrlr_cmd_abort(struct nvme_controller *ctrlr, uint16_t cid,
399 uint16_t sqid, nvme_cb_fn_t cb_fn, void *cb_arg);
400
401 void nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl);
402
403 int nvme_ctrlr_construct(struct nvme_controller *ctrlr, device_t dev);
404 void nvme_ctrlr_destruct(struct nvme_controller *ctrlr, device_t dev);
405 void nvme_ctrlr_shutdown(struct nvme_controller *ctrlr);
406 void nvme_ctrlr_reset(struct nvme_controller *ctrlr);
407 /* ctrlr defined as void * to allow use with config_intrhook. */
408 void nvme_ctrlr_start_config_hook(void *ctrlr_arg);
409 void nvme_ctrlr_submit_admin_request(struct nvme_controller *ctrlr,
410 struct nvme_request *req);
411 void nvme_ctrlr_submit_io_request(struct nvme_controller *ctrlr,
412 struct nvme_request *req);
413
414 int nvme_qpair_construct(struct nvme_qpair *qpair,
415 uint32_t num_entries, uint32_t num_trackers,
416 struct nvme_controller *ctrlr);
417 void nvme_qpair_submit_tracker(struct nvme_qpair *qpair,
418 struct nvme_tracker *tr);
419 bool nvme_qpair_process_completions(struct nvme_qpair *qpair);
420 void nvme_qpair_submit_request(struct nvme_qpair *qpair,
421 struct nvme_request *req);
422 void nvme_qpair_reset(struct nvme_qpair *qpair);
423 void nvme_qpair_fail(struct nvme_qpair *qpair);
424
425 void nvme_admin_qpair_enable(struct nvme_qpair *qpair);
426 void nvme_admin_qpair_disable(struct nvme_qpair *qpair);
427 void nvme_admin_qpair_destroy(struct nvme_qpair *qpair);
428
429 void nvme_io_qpair_enable(struct nvme_qpair *qpair);
430 void nvme_io_qpair_disable(struct nvme_qpair *qpair);
431 void nvme_io_qpair_destroy(struct nvme_qpair *qpair);
432
433 int nvme_ns_construct(struct nvme_namespace *ns, uint32_t id,
434 struct nvme_controller *ctrlr);
435 void nvme_ns_destruct(struct nvme_namespace *ns);
436
437 void nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr);
438
439 void nvme_qpair_print_command(struct nvme_qpair *qpair,
440 struct nvme_command *cmd);
441 void nvme_qpair_print_completion(struct nvme_qpair *qpair,
442 struct nvme_completion *cpl);
443
444 int nvme_attach(device_t dev);
445 int nvme_shutdown(device_t dev);
446 int nvme_detach(device_t dev);
447
448 /*
449 * Wait for a command to complete using the nvme_completion_poll_cb. Used in
450 * limited contexts where the caller knows it's OK to block briefly while the
451 * command runs. The ISR will run the callback which will set status->done to
452 * true, usually within microseconds. If not, then after one second timeout
453 * handler should reset the controller and abort all outstanding requests
454 * including this polled one. If still not after ten seconds, then something is
455 * wrong with the driver, and panic is the only way to recover.
456 *
457 * Most commands using this interface aren't actual I/O to the drive's media so
458 * complete within a few microseconds. Adaptively spin for one tick to catch the
459 * vast majority of these without waiting for a tick plus scheduling delays. Since
460 * these are on startup, this drastically reduces startup time.
461 */
462 static __inline
463 void
nvme_completion_poll(struct nvme_completion_poll_status * status)464 nvme_completion_poll(struct nvme_completion_poll_status *status)
465 {
466 int timeout = ticks + 10 * hz;
467 sbintime_t delta_t = SBT_1US;
468
469 while (!atomic_load_acq_int(&status->done)) {
470 if (timeout - ticks < 0)
471 panic("NVME polled command failed to complete within 10s.");
472 pause_sbt("nvme", delta_t, 0, C_PREL(1));
473 delta_t = min(SBT_1MS, delta_t * 3 / 2);
474 }
475 }
476
477 static __inline void
nvme_single_map(void * arg,bus_dma_segment_t * seg,int nseg,int error)478 nvme_single_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
479 {
480 uint64_t *bus_addr = (uint64_t *)arg;
481
482 KASSERT(nseg == 1, ("number of segments (%d) is not 1", nseg));
483 if (error != 0)
484 printf("nvme_single_map err %d\n", error);
485 *bus_addr = seg[0].ds_addr;
486 }
487
488 static __inline struct nvme_request *
_nvme_allocate_request(const int how,nvme_cb_fn_t cb_fn,void * cb_arg)489 _nvme_allocate_request(const int how, nvme_cb_fn_t cb_fn, void *cb_arg)
490 {
491 struct nvme_request *req;
492
493 KASSERT(how == M_WAITOK || how == M_NOWAIT,
494 ("nvme_allocate_request: invalid how %d", how));
495
496 req = malloc(sizeof(*req), M_NVME, how | M_ZERO);
497 if (req != NULL) {
498 req->cb_fn = cb_fn;
499 req->cb_arg = cb_arg;
500 req->timeout = true;
501 }
502 return (req);
503 }
504
505 static __inline struct nvme_request *
nvme_allocate_request_vaddr(void * payload,uint32_t payload_size,const int how,nvme_cb_fn_t cb_fn,void * cb_arg)506 nvme_allocate_request_vaddr(void *payload, uint32_t payload_size,
507 const int how, nvme_cb_fn_t cb_fn, void *cb_arg)
508 {
509 struct nvme_request *req;
510
511 req = _nvme_allocate_request(how, cb_fn, cb_arg);
512 if (req != NULL) {
513 req->payload = memdesc_vaddr(payload, payload_size);
514 req->payload_valid = true;
515 }
516 return (req);
517 }
518
519 static __inline struct nvme_request *
nvme_allocate_request_null(const int how,nvme_cb_fn_t cb_fn,void * cb_arg)520 nvme_allocate_request_null(const int how, nvme_cb_fn_t cb_fn, void *cb_arg)
521 {
522 struct nvme_request *req;
523
524 req = _nvme_allocate_request(how, cb_fn, cb_arg);
525 return (req);
526 }
527
528 static __inline struct nvme_request *
nvme_allocate_request_bio(struct bio * bio,const int how,nvme_cb_fn_t cb_fn,void * cb_arg)529 nvme_allocate_request_bio(struct bio *bio, const int how, nvme_cb_fn_t cb_fn,
530 void *cb_arg)
531 {
532 struct nvme_request *req;
533
534 req = _nvme_allocate_request(how, cb_fn, cb_arg);
535 if (req != NULL) {
536 req->payload = memdesc_bio(bio);
537 req->payload_valid = true;
538 }
539 return (req);
540 }
541
542 static __inline struct nvme_request *
nvme_allocate_request_ccb(union ccb * ccb,const int how,nvme_cb_fn_t cb_fn,void * cb_arg)543 nvme_allocate_request_ccb(union ccb *ccb, const int how, nvme_cb_fn_t cb_fn,
544 void *cb_arg)
545 {
546 struct nvme_request *req;
547
548 req = _nvme_allocate_request(how, cb_fn, cb_arg);
549 if (req != NULL) {
550 req->payload = memdesc_ccb(ccb);
551 req->payload_valid = true;
552 }
553 return (req);
554 }
555
556 #define nvme_free_request(req) free(req, M_NVME)
557
558 void nvme_notify_async_consumers(struct nvme_controller *ctrlr,
559 const struct nvme_completion *async_cpl,
560 uint32_t log_page_id, void *log_page_buffer,
561 uint32_t log_page_size);
562 void nvme_notify_fail_consumers(struct nvme_controller *ctrlr);
563 void nvme_notify_new_controller(struct nvme_controller *ctrlr);
564 void nvme_notify_ns(struct nvme_controller *ctrlr, int nsid);
565
566 void nvme_ctrlr_shared_handler(void *arg);
567 void nvme_ctrlr_poll(struct nvme_controller *ctrlr);
568
569 int nvme_ctrlr_suspend(struct nvme_controller *ctrlr);
570 int nvme_ctrlr_resume(struct nvme_controller *ctrlr);
571
572 #endif /* __NVME_PRIVATE_H__ */
573