xref: /freebsd/sys/dev/nvme/nvme_private.h (revision 1f4bcc459a76b7aa664f3fd557684cd0ba6da352)
1 /*-
2  * Copyright (C) 2012-2014 Intel Corporation
3  * All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  *
14  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
15  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
16  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
17  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
18  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
19  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
20  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
22  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
23  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
24  * SUCH DAMAGE.
25  *
26  * $FreeBSD$
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/kernel.h>
36 #include <sys/lock.h>
37 #include <sys/malloc.h>
38 #include <sys/mutex.h>
39 #include <sys/rman.h>
40 #include <sys/systm.h>
41 #include <sys/taskqueue.h>
42 
43 #include <vm/uma.h>
44 
45 #include <machine/bus.h>
46 
47 #include "nvme.h"
48 
49 #define DEVICE2SOFTC(dev) ((struct nvme_controller *) device_get_softc(dev))
50 
51 MALLOC_DECLARE(M_NVME);
52 
53 #define IDT32_PCI_ID		0x80d0111d /* 32 channel board */
54 #define IDT8_PCI_ID		0x80d2111d /* 8 channel board */
55 
56 /*
57  * For commands requiring more than 2 PRP entries, one PRP will be
58  *  embedded in the command (prp1), and the rest of the PRP entries
59  *  will be in a list pointed to by the command (prp2).  This means
60  *  that real max number of PRP entries we support is 32+1, which
61  *  results in a max xfer size of 32*PAGE_SIZE.
62  */
63 #define NVME_MAX_PRP_LIST_ENTRIES	(NVME_MAX_XFER_SIZE / PAGE_SIZE)
64 
65 #define NVME_ADMIN_TRACKERS	(16)
66 #define NVME_ADMIN_ENTRIES	(128)
67 /* min and max are defined in admin queue attributes section of spec */
68 #define NVME_MIN_ADMIN_ENTRIES	(2)
69 #define NVME_MAX_ADMIN_ENTRIES	(4096)
70 
71 /*
72  * NVME_IO_ENTRIES defines the size of an I/O qpair's submission and completion
73  *  queues, while NVME_IO_TRACKERS defines the maximum number of I/O that we
74  *  will allow outstanding on an I/O qpair at any time.  The only advantage in
75  *  having IO_ENTRIES > IO_TRACKERS is for debugging purposes - when dumping
76  *  the contents of the submission and completion queues, it will show a longer
77  *  history of data.
78  */
79 #define NVME_IO_ENTRIES		(256)
80 #define NVME_IO_TRACKERS	(128)
81 #define NVME_MIN_IO_TRACKERS	(4)
82 #define NVME_MAX_IO_TRACKERS	(1024)
83 
84 /*
85  * NVME_MAX_IO_ENTRIES is not defined, since it is specified in CC.MQES
86  *  for each controller.
87  */
88 
89 #define NVME_INT_COAL_TIME	(0)	/* disabled */
90 #define NVME_INT_COAL_THRESHOLD (0)	/* 0-based */
91 
92 #define NVME_MAX_NAMESPACES	(16)
93 #define NVME_MAX_CONSUMERS	(2)
94 #define NVME_MAX_ASYNC_EVENTS	(8)
95 
96 #define NVME_DEFAULT_TIMEOUT_PERIOD	(30)    /* in seconds */
97 #define NVME_MIN_TIMEOUT_PERIOD		(5)
98 #define NVME_MAX_TIMEOUT_PERIOD		(120)
99 
100 #define NVME_DEFAULT_RETRY_COUNT	(4)
101 
102 /* Maximum log page size to fetch for AERs. */
103 #define NVME_MAX_AER_LOG_SIZE		(4096)
104 
105 /*
106  * Define CACHE_LINE_SIZE here for older FreeBSD versions that do not define
107  *  it.
108  */
109 #ifndef CACHE_LINE_SIZE
110 #define CACHE_LINE_SIZE		(64)
111 #endif
112 
113 /*
114  * Use presence of the BIO_UNMAPPED flag to determine whether unmapped I/O
115  *  support and the bus_dmamap_load_bio API are available on the target
116  *  kernel.  This will ease porting back to earlier stable branches at a
117  *  later point.
118  */
119 #ifdef BIO_UNMAPPED
120 #define NVME_UNMAPPED_BIO_SUPPORT
121 #endif
122 
123 extern uma_zone_t	nvme_request_zone;
124 extern int32_t		nvme_retry_count;
125 
126 struct nvme_completion_poll_status {
127 
128 	struct nvme_completion	cpl;
129 	boolean_t		done;
130 };
131 
132 #define NVME_REQUEST_VADDR	1
133 #define NVME_REQUEST_NULL	2 /* For requests with no payload. */
134 #define NVME_REQUEST_UIO	3
135 #ifdef NVME_UNMAPPED_BIO_SUPPORT
136 #define NVME_REQUEST_BIO	4
137 #endif
138 
139 struct nvme_request {
140 
141 	struct nvme_command		cmd;
142 	struct nvme_qpair		*qpair;
143 	union {
144 		void			*payload;
145 		struct bio		*bio;
146 	} u;
147 	uint32_t			type;
148 	uint32_t			payload_size;
149 	boolean_t			timeout;
150 	nvme_cb_fn_t			cb_fn;
151 	void				*cb_arg;
152 	int32_t				retries;
153 	STAILQ_ENTRY(nvme_request)	stailq;
154 };
155 
156 struct nvme_async_event_request {
157 
158 	struct nvme_controller		*ctrlr;
159 	struct nvme_request		*req;
160 	struct nvme_completion		cpl;
161 	uint32_t			log_page_id;
162 	uint32_t			log_page_size;
163 	uint8_t				log_page_buffer[NVME_MAX_AER_LOG_SIZE];
164 };
165 
166 struct nvme_tracker {
167 
168 	TAILQ_ENTRY(nvme_tracker)	tailq;
169 	struct nvme_request		*req;
170 	struct nvme_qpair		*qpair;
171 	struct callout			timer;
172 	bus_dmamap_t			payload_dma_map;
173 	uint16_t			cid;
174 
175 	uint64_t			prp[NVME_MAX_PRP_LIST_ENTRIES];
176 	bus_addr_t			prp_bus_addr;
177 	bus_dmamap_t			prp_dma_map;
178 };
179 
180 struct nvme_qpair {
181 
182 	struct nvme_controller	*ctrlr;
183 	uint32_t		id;
184 	uint32_t		phase;
185 
186 	uint16_t		vector;
187 	int			rid;
188 	struct resource		*res;
189 	void 			*tag;
190 
191 	uint32_t		num_entries;
192 	uint32_t		num_trackers;
193 	uint32_t		sq_tdbl_off;
194 	uint32_t		cq_hdbl_off;
195 
196 	uint32_t		sq_head;
197 	uint32_t		sq_tail;
198 	uint32_t		cq_head;
199 
200 	int64_t			num_cmds;
201 	int64_t			num_intr_handler_calls;
202 
203 	struct nvme_command	*cmd;
204 	struct nvme_completion	*cpl;
205 
206 	bus_dma_tag_t		dma_tag;
207 	bus_dma_tag_t		dma_tag_payload;
208 
209 	bus_dmamap_t		cmd_dma_map;
210 	uint64_t		cmd_bus_addr;
211 
212 	bus_dmamap_t		cpl_dma_map;
213 	uint64_t		cpl_bus_addr;
214 
215 	TAILQ_HEAD(, nvme_tracker)	free_tr;
216 	TAILQ_HEAD(, nvme_tracker)	outstanding_tr;
217 	STAILQ_HEAD(, nvme_request)	queued_req;
218 
219 	struct nvme_tracker	**act_tr;
220 
221 	boolean_t		is_enabled;
222 
223 	struct mtx		lock __aligned(CACHE_LINE_SIZE);
224 
225 } __aligned(CACHE_LINE_SIZE);
226 
227 struct nvme_namespace {
228 
229 	struct nvme_controller		*ctrlr;
230 	struct nvme_namespace_data	data;
231 	uint16_t			id;
232 	uint16_t			flags;
233 	struct cdev			*cdev;
234 	void				*cons_cookie[NVME_MAX_CONSUMERS];
235 	uint32_t			stripesize;
236 	struct mtx			lock;
237 };
238 
239 /*
240  * One of these per allocated PCI device.
241  */
242 struct nvme_controller {
243 
244 	device_t		dev;
245 
246 	struct mtx		lock;
247 
248 	uint32_t		ready_timeout_in_ms;
249 
250 	bus_space_tag_t		bus_tag;
251 	bus_space_handle_t	bus_handle;
252 	int			resource_id;
253 	struct resource		*resource;
254 
255 	/*
256 	 * The NVMe spec allows for the MSI-X table to be placed in BAR 4/5,
257 	 *  separate from the control registers which are in BAR 0/1.  These
258 	 *  members track the mapping of BAR 4/5 for that reason.
259 	 */
260 	int			bar4_resource_id;
261 	struct resource		*bar4_resource;
262 
263 	uint32_t		msix_enabled;
264 	uint32_t		force_intx;
265 	uint32_t		enable_aborts;
266 
267 	uint32_t		num_io_queues;
268 	uint32_t		num_cpus_per_ioq;
269 
270 	/* Fields for tracking progress during controller initialization. */
271 	struct intr_config_hook	config_hook;
272 	uint32_t		ns_identified;
273 	uint32_t		queues_created;
274 
275 	struct task		reset_task;
276 	struct task		fail_req_task;
277 	struct taskqueue	*taskqueue;
278 
279 	/* For shared legacy interrupt. */
280 	int			rid;
281 	struct resource		*res;
282 	void			*tag;
283 
284 	bus_dma_tag_t		hw_desc_tag;
285 	bus_dmamap_t		hw_desc_map;
286 
287 	/** maximum i/o size in bytes */
288 	uint32_t		max_xfer_size;
289 
290 	/** minimum page size supported by this controller in bytes */
291 	uint32_t		min_page_size;
292 
293 	/** interrupt coalescing time period (in microseconds) */
294 	uint32_t		int_coal_time;
295 
296 	/** interrupt coalescing threshold */
297 	uint32_t		int_coal_threshold;
298 
299 	/** timeout period in seconds */
300 	uint32_t		timeout_period;
301 
302 	struct nvme_qpair	adminq;
303 	struct nvme_qpair	*ioq;
304 
305 	struct nvme_registers		*regs;
306 
307 	struct nvme_controller_data	cdata;
308 	struct nvme_namespace		ns[NVME_MAX_NAMESPACES];
309 
310 	struct cdev			*cdev;
311 
312 	/** bit mask of warning types currently enabled for async events */
313 	union nvme_critical_warning_state	async_event_config;
314 
315 	uint32_t			num_aers;
316 	struct nvme_async_event_request	aer[NVME_MAX_ASYNC_EVENTS];
317 
318 	void				*cons_cookie[NVME_MAX_CONSUMERS];
319 
320 	uint32_t			is_resetting;
321 	uint32_t			is_initialized;
322 	uint32_t			notification_sent;
323 
324 	boolean_t			is_failed;
325 	STAILQ_HEAD(, nvme_request)	fail_req;
326 };
327 
328 #define nvme_mmio_offsetof(reg)						       \
329 	offsetof(struct nvme_registers, reg)
330 
331 #define nvme_mmio_read_4(sc, reg)					       \
332 	bus_space_read_4((sc)->bus_tag, (sc)->bus_handle,		       \
333 	    nvme_mmio_offsetof(reg))
334 
335 #define nvme_mmio_write_4(sc, reg, val)					       \
336 	bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,		       \
337 	    nvme_mmio_offsetof(reg), val)
338 
339 #define nvme_mmio_write_8(sc, reg, val) \
340 	do {								       \
341 		bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,	       \
342 		    nvme_mmio_offsetof(reg), val & 0xFFFFFFFF); 	       \
343 		bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,	       \
344 		    nvme_mmio_offsetof(reg)+4,				       \
345 		    (val & 0xFFFFFFFF00000000UL) >> 32);		       \
346 	} while (0);
347 
348 #if __FreeBSD_version < 800054
349 #define wmb()	__asm volatile("sfence" ::: "memory")
350 #define mb()	__asm volatile("mfence" ::: "memory")
351 #endif
352 
353 #define nvme_printf(ctrlr, fmt, args...)	\
354     device_printf(ctrlr->dev, fmt, ##args)
355 
356 void	nvme_ns_test(struct nvme_namespace *ns, u_long cmd, caddr_t arg);
357 
358 void	nvme_ctrlr_cmd_identify_controller(struct nvme_controller *ctrlr,
359 					   void *payload,
360 					   nvme_cb_fn_t cb_fn, void *cb_arg);
361 void	nvme_ctrlr_cmd_identify_namespace(struct nvme_controller *ctrlr,
362 					  uint16_t nsid, void *payload,
363 					  nvme_cb_fn_t cb_fn, void *cb_arg);
364 void	nvme_ctrlr_cmd_set_interrupt_coalescing(struct nvme_controller *ctrlr,
365 						uint32_t microseconds,
366 						uint32_t threshold,
367 						nvme_cb_fn_t cb_fn,
368 						void *cb_arg);
369 void	nvme_ctrlr_cmd_get_error_page(struct nvme_controller *ctrlr,
370 				      struct nvme_error_information_entry *payload,
371 				      uint32_t num_entries, /* 0 = max */
372 				      nvme_cb_fn_t cb_fn,
373 				      void *cb_arg);
374 void	nvme_ctrlr_cmd_get_health_information_page(struct nvme_controller *ctrlr,
375 						   uint32_t nsid,
376 						   struct nvme_health_information_page *payload,
377 						   nvme_cb_fn_t cb_fn,
378 						   void *cb_arg);
379 void	nvme_ctrlr_cmd_get_firmware_page(struct nvme_controller *ctrlr,
380 					 struct nvme_firmware_page *payload,
381 					 nvme_cb_fn_t cb_fn,
382 					 void *cb_arg);
383 void	nvme_ctrlr_cmd_create_io_cq(struct nvme_controller *ctrlr,
384 				    struct nvme_qpair *io_que, uint16_t vector,
385 				    nvme_cb_fn_t cb_fn, void *cb_arg);
386 void	nvme_ctrlr_cmd_create_io_sq(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_cq(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_delete_io_sq(struct nvme_controller *ctrlr,
393 				    struct nvme_qpair *io_que,
394 				    nvme_cb_fn_t cb_fn, void *cb_arg);
395 void	nvme_ctrlr_cmd_set_num_queues(struct nvme_controller *ctrlr,
396 				      uint32_t num_queues, nvme_cb_fn_t cb_fn,
397 				      void *cb_arg);
398 void	nvme_ctrlr_cmd_set_async_event_config(struct nvme_controller *ctrlr,
399 					      union nvme_critical_warning_state state,
400 					      nvme_cb_fn_t cb_fn, void *cb_arg);
401 void	nvme_ctrlr_cmd_abort(struct nvme_controller *ctrlr, uint16_t cid,
402 			     uint16_t sqid, nvme_cb_fn_t cb_fn, void *cb_arg);
403 
404 void	nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl);
405 
406 int	nvme_ctrlr_construct(struct nvme_controller *ctrlr, device_t dev);
407 void	nvme_ctrlr_destruct(struct nvme_controller *ctrlr, device_t dev);
408 void	nvme_ctrlr_shutdown(struct nvme_controller *ctrlr);
409 int	nvme_ctrlr_hw_reset(struct nvme_controller *ctrlr);
410 void	nvme_ctrlr_reset(struct nvme_controller *ctrlr);
411 /* ctrlr defined as void * to allow use with config_intrhook. */
412 void	nvme_ctrlr_start_config_hook(void *ctrlr_arg);
413 void	nvme_ctrlr_submit_admin_request(struct nvme_controller *ctrlr,
414 					struct nvme_request *req);
415 void	nvme_ctrlr_submit_io_request(struct nvme_controller *ctrlr,
416 				     struct nvme_request *req);
417 void	nvme_ctrlr_post_failed_request(struct nvme_controller *ctrlr,
418 				       struct nvme_request *req);
419 
420 void	nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t id,
421 			     uint16_t vector, uint32_t num_entries,
422 			     uint32_t num_trackers,
423 			     struct nvme_controller *ctrlr);
424 void	nvme_qpair_submit_tracker(struct nvme_qpair *qpair,
425 				  struct nvme_tracker *tr);
426 void	nvme_qpair_process_completions(struct nvme_qpair *qpair);
427 void	nvme_qpair_submit_request(struct nvme_qpair *qpair,
428 				  struct nvme_request *req);
429 void	nvme_qpair_reset(struct nvme_qpair *qpair);
430 void	nvme_qpair_fail(struct nvme_qpair *qpair);
431 void	nvme_qpair_manual_complete_request(struct nvme_qpair *qpair,
432 					   struct nvme_request *req,
433 					   uint32_t sct, uint32_t sc,
434 					   boolean_t print_on_error);
435 
436 void	nvme_admin_qpair_enable(struct nvme_qpair *qpair);
437 void	nvme_admin_qpair_disable(struct nvme_qpair *qpair);
438 void	nvme_admin_qpair_destroy(struct nvme_qpair *qpair);
439 
440 void	nvme_io_qpair_enable(struct nvme_qpair *qpair);
441 void	nvme_io_qpair_disable(struct nvme_qpair *qpair);
442 void	nvme_io_qpair_destroy(struct nvme_qpair *qpair);
443 
444 int	nvme_ns_construct(struct nvme_namespace *ns, uint16_t id,
445 			  struct nvme_controller *ctrlr);
446 void	nvme_ns_destruct(struct nvme_namespace *ns);
447 
448 void	nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr);
449 
450 void	nvme_dump_command(struct nvme_command *cmd);
451 void	nvme_dump_completion(struct nvme_completion *cpl);
452 
453 static __inline void
454 nvme_single_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
455 {
456 	uint64_t *bus_addr = (uint64_t *)arg;
457 
458 	if (error != 0)
459 		printf("nvme_single_map err %d\n", error);
460 	*bus_addr = seg[0].ds_addr;
461 }
462 
463 static __inline struct nvme_request *
464 _nvme_allocate_request(nvme_cb_fn_t cb_fn, void *cb_arg)
465 {
466 	struct nvme_request *req;
467 
468 	req = uma_zalloc(nvme_request_zone, M_NOWAIT | M_ZERO);
469 	if (req != NULL) {
470 		req->cb_fn = cb_fn;
471 		req->cb_arg = cb_arg;
472 		req->timeout = TRUE;
473 	}
474 	return (req);
475 }
476 
477 static __inline struct nvme_request *
478 nvme_allocate_request_vaddr(void *payload, uint32_t payload_size,
479     nvme_cb_fn_t cb_fn, void *cb_arg)
480 {
481 	struct nvme_request *req;
482 
483 	req = _nvme_allocate_request(cb_fn, cb_arg);
484 	if (req != NULL) {
485 		req->type = NVME_REQUEST_VADDR;
486 		req->u.payload = payload;
487 		req->payload_size = payload_size;
488 	}
489 	return (req);
490 }
491 
492 static __inline struct nvme_request *
493 nvme_allocate_request_null(nvme_cb_fn_t cb_fn, void *cb_arg)
494 {
495 	struct nvme_request *req;
496 
497 	req = _nvme_allocate_request(cb_fn, cb_arg);
498 	if (req != NULL)
499 		req->type = NVME_REQUEST_NULL;
500 	return (req);
501 }
502 
503 static __inline struct nvme_request *
504 nvme_allocate_request_bio(struct bio *bio, nvme_cb_fn_t cb_fn, void *cb_arg)
505 {
506 	struct nvme_request *req;
507 
508 	req = _nvme_allocate_request(cb_fn, cb_arg);
509 	if (req != NULL) {
510 #ifdef NVME_UNMAPPED_BIO_SUPPORT
511 		req->type = NVME_REQUEST_BIO;
512 		req->u.bio = bio;
513 #else
514 		req->type = NVME_REQUEST_VADDR;
515 		req->u.payload = bio->bio_data;
516 		req->payload_size = bio->bio_bcount;
517 #endif
518 	}
519 	return (req);
520 }
521 
522 #define nvme_free_request(req)	uma_zfree(nvme_request_zone, req)
523 
524 void	nvme_notify_async_consumers(struct nvme_controller *ctrlr,
525 				    const struct nvme_completion *async_cpl,
526 				    uint32_t log_page_id, void *log_page_buffer,
527 				    uint32_t log_page_size);
528 void	nvme_notify_fail_consumers(struct nvme_controller *ctrlr);
529 void	nvme_notify_new_controller(struct nvme_controller *ctrlr);
530 
531 #endif /* __NVME_PRIVATE_H__ */
532