xref: /freebsd-11-stable/sys/dev/nvme/nvme_private.h (revision 6119ba142ae312ce616082df64c3963c8a45db5f)
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 extern uma_zone_t	nvme_request_zone;
114 extern int32_t		nvme_retry_count;
115 
116 struct nvme_completion_poll_status {
117 
118 	struct nvme_completion	cpl;
119 	int			done;
120 };
121 
122 #define NVME_REQUEST_VADDR	1
123 #define NVME_REQUEST_NULL	2 /* For requests with no payload. */
124 #define NVME_REQUEST_UIO	3
125 #define NVME_REQUEST_BIO	4
126 #define NVME_REQUEST_CCB        5
127 
128 struct nvme_request {
129 
130 	struct nvme_command		cmd;
131 	struct nvme_qpair		*qpair;
132 	union {
133 		void			*payload;
134 		struct bio		*bio;
135 	} u;
136 	uint32_t			type;
137 	uint32_t			payload_size;
138 	boolean_t			timeout;
139 	nvme_cb_fn_t			cb_fn;
140 	void				*cb_arg;
141 	int32_t				retries;
142 	STAILQ_ENTRY(nvme_request)	stailq;
143 };
144 
145 struct nvme_async_event_request {
146 
147 	struct nvme_controller		*ctrlr;
148 	struct nvme_request		*req;
149 	struct nvme_completion		cpl;
150 	uint32_t			log_page_id;
151 	uint32_t			log_page_size;
152 	uint8_t				log_page_buffer[NVME_MAX_AER_LOG_SIZE];
153 };
154 
155 struct nvme_tracker {
156 
157 	TAILQ_ENTRY(nvme_tracker)	tailq;
158 	struct nvme_request		*req;
159 	struct nvme_qpair		*qpair;
160 	struct callout			timer;
161 	bus_dmamap_t			payload_dma_map;
162 	uint16_t			cid;
163 
164 	uint64_t			*prp;
165 	bus_addr_t			prp_bus_addr;
166 };
167 
168 struct nvme_qpair {
169 
170 	struct nvme_controller	*ctrlr;
171 	uint32_t		id;
172 	uint32_t		phase;
173 
174 	uint16_t		vector;
175 	int			rid;
176 	struct resource		*res;
177 	void 			*tag;
178 
179 	uint32_t		num_entries;
180 	uint32_t		num_trackers;
181 	uint32_t		sq_tdbl_off;
182 	uint32_t		cq_hdbl_off;
183 
184 	uint32_t		sq_head;
185 	uint32_t		sq_tail;
186 	uint32_t		cq_head;
187 
188 	int64_t			num_cmds;
189 	int64_t			num_intr_handler_calls;
190 
191 	struct nvme_command	*cmd;
192 	struct nvme_completion	*cpl;
193 
194 	bus_dma_tag_t		dma_tag;
195 	bus_dma_tag_t		dma_tag_payload;
196 
197 	bus_dmamap_t		queuemem_map;
198 	uint64_t		cmd_bus_addr;
199 	uint64_t		cpl_bus_addr;
200 
201 	TAILQ_HEAD(, nvme_tracker)	free_tr;
202 	TAILQ_HEAD(, nvme_tracker)	outstanding_tr;
203 	STAILQ_HEAD(, nvme_request)	queued_req;
204 
205 	struct nvme_tracker	**act_tr;
206 
207 	boolean_t		is_enabled;
208 
209 	struct mtx		lock __aligned(CACHE_LINE_SIZE);
210 
211 } __aligned(CACHE_LINE_SIZE);
212 
213 struct nvme_namespace {
214 
215 	struct nvme_controller		*ctrlr;
216 	struct nvme_namespace_data	data;
217 	uint32_t			id;
218 	uint32_t			flags;
219 	struct cdev			*cdev;
220 	void				*cons_cookie[NVME_MAX_CONSUMERS];
221 	uint32_t			stripesize;
222 	struct mtx			lock;
223 };
224 
225 /*
226  * One of these per allocated PCI device.
227  */
228 struct nvme_controller {
229 
230 	device_t		dev;
231 
232 	struct mtx		lock;
233 
234 	uint32_t		ready_timeout_in_ms;
235 	uint32_t		quirks;
236 #define	QUIRK_DELAY_B4_CHK_RDY	1		/* Can't touch MMIO on disable */
237 #define	QUIRK_DISABLE_TIMEOUT	2		/* Disable broken completion timeout feature */
238 
239 	bus_space_tag_t		bus_tag;
240 	bus_space_handle_t	bus_handle;
241 	int			resource_id;
242 	struct resource		*resource;
243 
244 	/*
245 	 * The NVMe spec allows for the MSI-X table to be placed in BAR 4/5,
246 	 *  separate from the control registers which are in BAR 0/1.  These
247 	 *  members track the mapping of BAR 4/5 for that reason.
248 	 */
249 	int			bar4_resource_id;
250 	struct resource		*bar4_resource;
251 
252 	uint32_t		msix_enabled;
253 	uint32_t		force_intx;
254 	uint32_t		enable_aborts;
255 
256 	uint32_t		num_io_queues;
257 	uint32_t		num_cpus_per_ioq;
258 	uint32_t		max_hw_pend_io;
259 
260 	/* Fields for tracking progress during controller initialization. */
261 	struct intr_config_hook	config_hook;
262 	uint32_t		ns_identified;
263 	uint32_t		queues_created;
264 
265 	struct task		reset_task;
266 	struct task		fail_req_task;
267 	struct taskqueue	*taskqueue;
268 
269 	/* For shared legacy interrupt. */
270 	int			rid;
271 	struct resource		*res;
272 	void			*tag;
273 
274 	bus_dma_tag_t		hw_desc_tag;
275 	bus_dmamap_t		hw_desc_map;
276 
277 	/** maximum i/o size in bytes */
278 	uint32_t		max_xfer_size;
279 
280 	/** minimum page size supported by this controller in bytes */
281 	uint32_t		min_page_size;
282 
283 	/** interrupt coalescing time period (in microseconds) */
284 	uint32_t		int_coal_time;
285 
286 	/** interrupt coalescing threshold */
287 	uint32_t		int_coal_threshold;
288 
289 	/** timeout period in seconds */
290 	uint32_t		timeout_period;
291 
292 	struct nvme_qpair	adminq;
293 	struct nvme_qpair	*ioq;
294 
295 	struct nvme_registers		*regs;
296 
297 	struct nvme_controller_data	cdata;
298 	struct nvme_namespace		ns[NVME_MAX_NAMESPACES];
299 
300 	struct cdev			*cdev;
301 
302 	/** bit mask of warning types currently enabled for async events */
303 	union nvme_critical_warning_state	async_event_config;
304 
305 	uint32_t			num_aers;
306 	struct nvme_async_event_request	aer[NVME_MAX_ASYNC_EVENTS];
307 
308 	void				*cons_cookie[NVME_MAX_CONSUMERS];
309 
310 	uint32_t			is_resetting;
311 	uint32_t			is_initialized;
312 	uint32_t			notification_sent;
313 
314 	boolean_t			is_failed;
315 	STAILQ_HEAD(, nvme_request)	fail_req;
316 };
317 
318 #define nvme_mmio_offsetof(reg)						       \
319 	offsetof(struct nvme_registers, reg)
320 
321 #define nvme_mmio_read_4(sc, reg)					       \
322 	bus_space_read_4((sc)->bus_tag, (sc)->bus_handle,		       \
323 	    nvme_mmio_offsetof(reg))
324 
325 #define nvme_mmio_write_4(sc, reg, val)					       \
326 	bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,		       \
327 	    nvme_mmio_offsetof(reg), val)
328 
329 #define nvme_mmio_write_8(sc, reg, val) \
330 	do {								       \
331 		bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,	       \
332 		    nvme_mmio_offsetof(reg), val & 0xFFFFFFFF); 	       \
333 		bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,	       \
334 		    nvme_mmio_offsetof(reg)+4,				       \
335 		    (val & 0xFFFFFFFF00000000UL) >> 32);		       \
336 	} while (0);
337 
338 #define nvme_printf(ctrlr, fmt, args...)	\
339     device_printf(ctrlr->dev, fmt, ##args)
340 
341 void	nvme_ns_test(struct nvme_namespace *ns, u_long cmd, caddr_t arg);
342 
343 void	nvme_ctrlr_cmd_identify_controller(struct nvme_controller *ctrlr,
344 					   void *payload,
345 					   nvme_cb_fn_t cb_fn, void *cb_arg);
346 void	nvme_ctrlr_cmd_identify_namespace(struct nvme_controller *ctrlr,
347 					  uint32_t nsid, void *payload,
348 					  nvme_cb_fn_t cb_fn, void *cb_arg);
349 void	nvme_ctrlr_cmd_set_interrupt_coalescing(struct nvme_controller *ctrlr,
350 						uint32_t microseconds,
351 						uint32_t threshold,
352 						nvme_cb_fn_t cb_fn,
353 						void *cb_arg);
354 void	nvme_ctrlr_cmd_get_error_page(struct nvme_controller *ctrlr,
355 				      struct nvme_error_information_entry *payload,
356 				      uint32_t num_entries, /* 0 = max */
357 				      nvme_cb_fn_t cb_fn,
358 				      void *cb_arg);
359 void	nvme_ctrlr_cmd_get_health_information_page(struct nvme_controller *ctrlr,
360 						   uint32_t nsid,
361 						   struct nvme_health_information_page *payload,
362 						   nvme_cb_fn_t cb_fn,
363 						   void *cb_arg);
364 void	nvme_ctrlr_cmd_get_firmware_page(struct nvme_controller *ctrlr,
365 					 struct nvme_firmware_page *payload,
366 					 nvme_cb_fn_t cb_fn,
367 					 void *cb_arg);
368 void	nvme_ctrlr_cmd_create_io_cq(struct nvme_controller *ctrlr,
369 				    struct nvme_qpair *io_que, uint16_t vector,
370 				    nvme_cb_fn_t cb_fn, void *cb_arg);
371 void	nvme_ctrlr_cmd_create_io_sq(struct nvme_controller *ctrlr,
372 				    struct nvme_qpair *io_que,
373 				    nvme_cb_fn_t cb_fn, void *cb_arg);
374 void	nvme_ctrlr_cmd_delete_io_cq(struct nvme_controller *ctrlr,
375 				    struct nvme_qpair *io_que,
376 				    nvme_cb_fn_t cb_fn, void *cb_arg);
377 void	nvme_ctrlr_cmd_delete_io_sq(struct nvme_controller *ctrlr,
378 				    struct nvme_qpair *io_que,
379 				    nvme_cb_fn_t cb_fn, void *cb_arg);
380 void	nvme_ctrlr_cmd_set_num_queues(struct nvme_controller *ctrlr,
381 				      uint32_t num_queues, nvme_cb_fn_t cb_fn,
382 				      void *cb_arg);
383 void	nvme_ctrlr_cmd_set_async_event_config(struct nvme_controller *ctrlr,
384 					      union nvme_critical_warning_state state,
385 					      nvme_cb_fn_t cb_fn, void *cb_arg);
386 void	nvme_ctrlr_cmd_abort(struct nvme_controller *ctrlr, uint16_t cid,
387 			     uint16_t sqid, nvme_cb_fn_t cb_fn, void *cb_arg);
388 
389 void	nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl);
390 
391 int	nvme_ctrlr_construct(struct nvme_controller *ctrlr, device_t dev);
392 void	nvme_ctrlr_destruct(struct nvme_controller *ctrlr, device_t dev);
393 void	nvme_ctrlr_shutdown(struct nvme_controller *ctrlr);
394 int	nvme_ctrlr_hw_reset(struct nvme_controller *ctrlr);
395 void	nvme_ctrlr_reset(struct nvme_controller *ctrlr);
396 /* ctrlr defined as void * to allow use with config_intrhook. */
397 void	nvme_ctrlr_start_config_hook(void *ctrlr_arg);
398 void	nvme_ctrlr_submit_admin_request(struct nvme_controller *ctrlr,
399 					struct nvme_request *req);
400 void	nvme_ctrlr_submit_io_request(struct nvme_controller *ctrlr,
401 				     struct nvme_request *req);
402 void	nvme_ctrlr_post_failed_request(struct nvme_controller *ctrlr,
403 				       struct nvme_request *req);
404 
405 int	nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t id,
406 			     uint16_t vector, uint32_t num_entries,
407 			     uint32_t num_trackers,
408 			     struct nvme_controller *ctrlr);
409 void	nvme_qpair_submit_tracker(struct nvme_qpair *qpair,
410 				  struct nvme_tracker *tr);
411 bool	nvme_qpair_process_completions(struct nvme_qpair *qpair);
412 void	nvme_qpair_submit_request(struct nvme_qpair *qpair,
413 				  struct nvme_request *req);
414 void	nvme_qpair_reset(struct nvme_qpair *qpair);
415 void	nvme_qpair_fail(struct nvme_qpair *qpair);
416 void	nvme_qpair_manual_complete_request(struct nvme_qpair *qpair,
417 					   struct nvme_request *req,
418                                            uint32_t sct, uint32_t sc);
419 
420 void	nvme_admin_qpair_enable(struct nvme_qpair *qpair);
421 void	nvme_admin_qpair_disable(struct nvme_qpair *qpair);
422 void	nvme_admin_qpair_destroy(struct nvme_qpair *qpair);
423 
424 void	nvme_io_qpair_enable(struct nvme_qpair *qpair);
425 void	nvme_io_qpair_disable(struct nvme_qpair *qpair);
426 void	nvme_io_qpair_destroy(struct nvme_qpair *qpair);
427 
428 int	nvme_ns_construct(struct nvme_namespace *ns, uint32_t id,
429 			  struct nvme_controller *ctrlr);
430 void	nvme_ns_destruct(struct nvme_namespace *ns);
431 
432 void	nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr);
433 
434 void	nvme_dump_command(struct nvme_command *cmd);
435 void	nvme_dump_completion(struct nvme_completion *cpl);
436 
437 static __inline void
nvme_single_map(void * arg,bus_dma_segment_t * seg,int nseg,int error)438 nvme_single_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
439 {
440 	uint64_t *bus_addr = (uint64_t *)arg;
441 
442 	if (error != 0)
443 		printf("nvme_single_map err %d\n", error);
444 	*bus_addr = seg[0].ds_addr;
445 }
446 
447 static __inline struct nvme_request *
_nvme_allocate_request(nvme_cb_fn_t cb_fn,void * cb_arg)448 _nvme_allocate_request(nvme_cb_fn_t cb_fn, void *cb_arg)
449 {
450 	struct nvme_request *req;
451 
452 	req = uma_zalloc(nvme_request_zone, M_NOWAIT | M_ZERO);
453 	if (req != NULL) {
454 		req->cb_fn = cb_fn;
455 		req->cb_arg = cb_arg;
456 		req->timeout = TRUE;
457 	}
458 	return (req);
459 }
460 
461 static __inline struct nvme_request *
nvme_allocate_request_vaddr(void * payload,uint32_t payload_size,nvme_cb_fn_t cb_fn,void * cb_arg)462 nvme_allocate_request_vaddr(void *payload, uint32_t payload_size,
463     nvme_cb_fn_t cb_fn, void *cb_arg)
464 {
465 	struct nvme_request *req;
466 
467 	req = _nvme_allocate_request(cb_fn, cb_arg);
468 	if (req != NULL) {
469 		req->type = NVME_REQUEST_VADDR;
470 		req->u.payload = payload;
471 		req->payload_size = payload_size;
472 	}
473 	return (req);
474 }
475 
476 static __inline struct nvme_request *
nvme_allocate_request_null(nvme_cb_fn_t cb_fn,void * cb_arg)477 nvme_allocate_request_null(nvme_cb_fn_t cb_fn, void *cb_arg)
478 {
479 	struct nvme_request *req;
480 
481 	req = _nvme_allocate_request(cb_fn, cb_arg);
482 	if (req != NULL)
483 		req->type = NVME_REQUEST_NULL;
484 	return (req);
485 }
486 
487 static __inline struct nvme_request *
nvme_allocate_request_bio(struct bio * bio,nvme_cb_fn_t cb_fn,void * cb_arg)488 nvme_allocate_request_bio(struct bio *bio, nvme_cb_fn_t cb_fn, void *cb_arg)
489 {
490 	struct nvme_request *req;
491 
492 	req = _nvme_allocate_request(cb_fn, cb_arg);
493 	if (req != NULL) {
494 		req->type = NVME_REQUEST_BIO;
495 		req->u.bio = bio;
496 	}
497 	return (req);
498 }
499 
500 static __inline struct nvme_request *
nvme_allocate_request_ccb(union ccb * ccb,nvme_cb_fn_t cb_fn,void * cb_arg)501 nvme_allocate_request_ccb(union ccb *ccb, 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 		req->type = NVME_REQUEST_CCB;
508 		req->u.payload = ccb;
509 	}
510 
511 	return (req);
512 }
513 
514 #define nvme_free_request(req)	uma_zfree(nvme_request_zone, req)
515 
516 void	nvme_notify_async_consumers(struct nvme_controller *ctrlr,
517 				    const struct nvme_completion *async_cpl,
518 				    uint32_t log_page_id, void *log_page_buffer,
519 				    uint32_t log_page_size);
520 void	nvme_notify_fail_consumers(struct nvme_controller *ctrlr);
521 void	nvme_notify_new_controller(struct nvme_controller *ctrlr);
522 
523 void	nvme_ctrlr_intx_handler(void *arg);
524 void	nvme_ctrlr_poll(struct nvme_controller *ctrlr);
525 
526 #endif /* __NVME_PRIVATE_H__ */
527