xref: /freebsd/sys/dev/nvme/nvme_private.h (revision cbd30a72ca196976c1c700400ecd424baa1b9c16)
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;
176 	bus_addr_t			prp_bus_addr;
177 };
178 
179 struct nvme_qpair {
180 
181 	struct nvme_controller	*ctrlr;
182 	uint32_t		id;
183 	uint32_t		phase;
184 
185 	uint16_t		vector;
186 	int			rid;
187 	struct resource		*res;
188 	void 			*tag;
189 
190 	uint32_t		num_entries;
191 	uint32_t		num_trackers;
192 	uint32_t		sq_tdbl_off;
193 	uint32_t		cq_hdbl_off;
194 
195 	uint32_t		sq_head;
196 	uint32_t		sq_tail;
197 	uint32_t		cq_head;
198 
199 	int64_t			num_cmds;
200 	int64_t			num_intr_handler_calls;
201 
202 	struct nvme_command	*cmd;
203 	struct nvme_completion	*cpl;
204 
205 	bus_dma_tag_t		dma_tag;
206 	bus_dma_tag_t		dma_tag_payload;
207 
208 	bus_dmamap_t		queuemem_map;
209 	uint64_t		cmd_bus_addr;
210 	uint64_t		cpl_bus_addr;
211 
212 	TAILQ_HEAD(, nvme_tracker)	free_tr;
213 	TAILQ_HEAD(, nvme_tracker)	outstanding_tr;
214 	STAILQ_HEAD(, nvme_request)	queued_req;
215 
216 	struct nvme_tracker	**act_tr;
217 
218 	boolean_t		is_enabled;
219 
220 	struct mtx		lock __aligned(CACHE_LINE_SIZE);
221 
222 } __aligned(CACHE_LINE_SIZE);
223 
224 struct nvme_namespace {
225 
226 	struct nvme_controller		*ctrlr;
227 	struct nvme_namespace_data	data;
228 	uint16_t			id;
229 	uint16_t			flags;
230 	struct cdev			*cdev;
231 	void				*cons_cookie[NVME_MAX_CONSUMERS];
232 	uint32_t			stripesize;
233 	struct mtx			lock;
234 };
235 
236 /*
237  * One of these per allocated PCI device.
238  */
239 struct nvme_controller {
240 
241 	device_t		dev;
242 
243 	struct mtx		lock;
244 
245 	uint32_t		ready_timeout_in_ms;
246 
247 	bus_space_tag_t		bus_tag;
248 	bus_space_handle_t	bus_handle;
249 	int			resource_id;
250 	struct resource		*resource;
251 
252 	/*
253 	 * The NVMe spec allows for the MSI-X table to be placed in BAR 4/5,
254 	 *  separate from the control registers which are in BAR 0/1.  These
255 	 *  members track the mapping of BAR 4/5 for that reason.
256 	 */
257 	int			bar4_resource_id;
258 	struct resource		*bar4_resource;
259 
260 	uint32_t		msix_enabled;
261 	uint32_t		force_intx;
262 	uint32_t		enable_aborts;
263 
264 	uint32_t		num_io_queues;
265 	uint32_t		num_cpus_per_ioq;
266 
267 	/* Fields for tracking progress during controller initialization. */
268 	struct intr_config_hook	config_hook;
269 	uint32_t		ns_identified;
270 	uint32_t		queues_created;
271 
272 	struct task		reset_task;
273 	struct task		fail_req_task;
274 	struct taskqueue	*taskqueue;
275 
276 	/* For shared legacy interrupt. */
277 	int			rid;
278 	struct resource		*res;
279 	void			*tag;
280 
281 	bus_dma_tag_t		hw_desc_tag;
282 	bus_dmamap_t		hw_desc_map;
283 
284 	/** maximum i/o size in bytes */
285 	uint32_t		max_xfer_size;
286 
287 	/** minimum page size supported by this controller in bytes */
288 	uint32_t		min_page_size;
289 
290 	/** interrupt coalescing time period (in microseconds) */
291 	uint32_t		int_coal_time;
292 
293 	/** interrupt coalescing threshold */
294 	uint32_t		int_coal_threshold;
295 
296 	/** timeout period in seconds */
297 	uint32_t		timeout_period;
298 
299 	struct nvme_qpair	adminq;
300 	struct nvme_qpair	*ioq;
301 
302 	struct nvme_registers		*regs;
303 
304 	struct nvme_controller_data	cdata;
305 	struct nvme_namespace		ns[NVME_MAX_NAMESPACES];
306 
307 	struct cdev			*cdev;
308 
309 	/** bit mask of warning types currently enabled for async events */
310 	union nvme_critical_warning_state	async_event_config;
311 
312 	uint32_t			num_aers;
313 	struct nvme_async_event_request	aer[NVME_MAX_ASYNC_EVENTS];
314 
315 	void				*cons_cookie[NVME_MAX_CONSUMERS];
316 
317 	uint32_t			is_resetting;
318 	uint32_t			is_initialized;
319 	uint32_t			notification_sent;
320 
321 	boolean_t			is_failed;
322 	STAILQ_HEAD(, nvme_request)	fail_req;
323 };
324 
325 #define nvme_mmio_offsetof(reg)						       \
326 	offsetof(struct nvme_registers, reg)
327 
328 #define nvme_mmio_read_4(sc, reg)					       \
329 	bus_space_read_4((sc)->bus_tag, (sc)->bus_handle,		       \
330 	    nvme_mmio_offsetof(reg))
331 
332 #define nvme_mmio_write_4(sc, reg, val)					       \
333 	bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,		       \
334 	    nvme_mmio_offsetof(reg), val)
335 
336 #define nvme_mmio_write_8(sc, reg, val) \
337 	do {								       \
338 		bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,	       \
339 		    nvme_mmio_offsetof(reg), val & 0xFFFFFFFF); 	       \
340 		bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,	       \
341 		    nvme_mmio_offsetof(reg)+4,				       \
342 		    (val & 0xFFFFFFFF00000000UL) >> 32);		       \
343 	} while (0);
344 
345 #if __FreeBSD_version < 800054
346 #define wmb()	__asm volatile("sfence" ::: "memory")
347 #define mb()	__asm volatile("mfence" ::: "memory")
348 #endif
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 					  uint16_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, uint16_t vector,
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 					      union nvme_critical_warning_state 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 int	nvme_ctrlr_hw_reset(struct nvme_controller *ctrlr);
407 void	nvme_ctrlr_reset(struct nvme_controller *ctrlr);
408 /* ctrlr defined as void * to allow use with config_intrhook. */
409 void	nvme_ctrlr_start_config_hook(void *ctrlr_arg);
410 void	nvme_ctrlr_submit_admin_request(struct nvme_controller *ctrlr,
411 					struct nvme_request *req);
412 void	nvme_ctrlr_submit_io_request(struct nvme_controller *ctrlr,
413 				     struct nvme_request *req);
414 void	nvme_ctrlr_post_failed_request(struct nvme_controller *ctrlr,
415 				       struct nvme_request *req);
416 
417 int	nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t id,
418 			     uint16_t vector, uint32_t num_entries,
419 			     uint32_t num_trackers,
420 			     struct nvme_controller *ctrlr);
421 void	nvme_qpair_submit_tracker(struct nvme_qpair *qpair,
422 				  struct nvme_tracker *tr);
423 void	nvme_qpair_process_completions(struct nvme_qpair *qpair);
424 void	nvme_qpair_submit_request(struct nvme_qpair *qpair,
425 				  struct nvme_request *req);
426 void	nvme_qpair_reset(struct nvme_qpair *qpair);
427 void	nvme_qpair_fail(struct nvme_qpair *qpair);
428 void	nvme_qpair_manual_complete_request(struct nvme_qpair *qpair,
429 					   struct nvme_request *req,
430 					   uint32_t sct, uint32_t sc,
431 					   boolean_t print_on_error);
432 
433 void	nvme_admin_qpair_enable(struct nvme_qpair *qpair);
434 void	nvme_admin_qpair_disable(struct nvme_qpair *qpair);
435 void	nvme_admin_qpair_destroy(struct nvme_qpair *qpair);
436 
437 void	nvme_io_qpair_enable(struct nvme_qpair *qpair);
438 void	nvme_io_qpair_disable(struct nvme_qpair *qpair);
439 void	nvme_io_qpair_destroy(struct nvme_qpair *qpair);
440 
441 int	nvme_ns_construct(struct nvme_namespace *ns, uint16_t id,
442 			  struct nvme_controller *ctrlr);
443 void	nvme_ns_destruct(struct nvme_namespace *ns);
444 
445 void	nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr);
446 
447 void	nvme_dump_command(struct nvme_command *cmd);
448 void	nvme_dump_completion(struct nvme_completion *cpl);
449 
450 static __inline void
451 nvme_single_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
452 {
453 	uint64_t *bus_addr = (uint64_t *)arg;
454 
455 	if (error != 0)
456 		printf("nvme_single_map err %d\n", error);
457 	*bus_addr = seg[0].ds_addr;
458 }
459 
460 static __inline struct nvme_request *
461 _nvme_allocate_request(nvme_cb_fn_t cb_fn, void *cb_arg)
462 {
463 	struct nvme_request *req;
464 
465 	req = uma_zalloc(nvme_request_zone, M_NOWAIT | M_ZERO);
466 	if (req != NULL) {
467 		req->cb_fn = cb_fn;
468 		req->cb_arg = cb_arg;
469 		req->timeout = TRUE;
470 	}
471 	return (req);
472 }
473 
474 static __inline struct nvme_request *
475 nvme_allocate_request_vaddr(void *payload, uint32_t payload_size,
476     nvme_cb_fn_t cb_fn, void *cb_arg)
477 {
478 	struct nvme_request *req;
479 
480 	req = _nvme_allocate_request(cb_fn, cb_arg);
481 	if (req != NULL) {
482 		req->type = NVME_REQUEST_VADDR;
483 		req->u.payload = payload;
484 		req->payload_size = payload_size;
485 	}
486 	return (req);
487 }
488 
489 static __inline struct nvme_request *
490 nvme_allocate_request_null(nvme_cb_fn_t cb_fn, void *cb_arg)
491 {
492 	struct nvme_request *req;
493 
494 	req = _nvme_allocate_request(cb_fn, cb_arg);
495 	if (req != NULL)
496 		req->type = NVME_REQUEST_NULL;
497 	return (req);
498 }
499 
500 static __inline struct nvme_request *
501 nvme_allocate_request_bio(struct bio *bio, nvme_cb_fn_t cb_fn, void *cb_arg)
502 {
503 	struct nvme_request *req;
504 
505 	req = _nvme_allocate_request(cb_fn, cb_arg);
506 	if (req != NULL) {
507 #ifdef NVME_UNMAPPED_BIO_SUPPORT
508 		req->type = NVME_REQUEST_BIO;
509 		req->u.bio = bio;
510 #else
511 		req->type = NVME_REQUEST_VADDR;
512 		req->u.payload = bio->bio_data;
513 		req->payload_size = bio->bio_bcount;
514 #endif
515 	}
516 	return (req);
517 }
518 
519 #define nvme_free_request(req)	uma_zfree(nvme_request_zone, req)
520 
521 void	nvme_notify_async_consumers(struct nvme_controller *ctrlr,
522 				    const struct nvme_completion *async_cpl,
523 				    uint32_t log_page_id, void *log_page_buffer,
524 				    uint32_t log_page_size);
525 void	nvme_notify_fail_consumers(struct nvme_controller *ctrlr);
526 void	nvme_notify_new_controller(struct nvme_controller *ctrlr);
527 
528 void	nvme_ctrlr_intx_handler(void *arg);
529 
530 #endif /* __NVME_PRIVATE_H__ */
531