xref: /freebsd/sys/dev/nvme/nvme_private.h (revision 0bf48626aaa33768078f5872b922b1487b3a9296)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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  * $FreeBSD$
29  */
30 
31 #ifndef __NVME_PRIVATE_H__
32 #define __NVME_PRIVATE_H__
33 
34 #include <sys/param.h>
35 #include <sys/bio.h>
36 #include <sys/bus.h>
37 #include <sys/kernel.h>
38 #include <sys/lock.h>
39 #include <sys/malloc.h>
40 #include <sys/mutex.h>
41 #include <sys/rman.h>
42 #include <sys/systm.h>
43 #include <sys/taskqueue.h>
44 
45 #include <vm/uma.h>
46 
47 #include <machine/bus.h>
48 
49 #include "nvme.h"
50 
51 #define DEVICE2SOFTC(dev) ((struct nvme_controller *) device_get_softc(dev))
52 
53 MALLOC_DECLARE(M_NVME);
54 
55 #define IDT32_PCI_ID		0x80d0111d /* 32 channel board */
56 #define IDT8_PCI_ID		0x80d2111d /* 8 channel board */
57 
58 /*
59  * For commands requiring more than 2 PRP entries, one PRP will be
60  *  embedded in the command (prp1), and the rest of the PRP entries
61  *  will be in a list pointed to by the command (prp2).  This means
62  *  that real max number of PRP entries we support is 32+1, which
63  *  results in a max xfer size of 32*PAGE_SIZE.
64  */
65 #define NVME_MAX_PRP_LIST_ENTRIES	(NVME_MAX_XFER_SIZE / PAGE_SIZE)
66 
67 #define NVME_ADMIN_TRACKERS	(16)
68 #define NVME_ADMIN_ENTRIES	(128)
69 /* min and max are defined in admin queue attributes section of spec */
70 #define NVME_MIN_ADMIN_ENTRIES	(2)
71 #define NVME_MAX_ADMIN_ENTRIES	(4096)
72 
73 /*
74  * NVME_IO_ENTRIES defines the size of an I/O qpair's submission and completion
75  *  queues, while NVME_IO_TRACKERS defines the maximum number of I/O that we
76  *  will allow outstanding on an I/O qpair at any time.  The only advantage in
77  *  having IO_ENTRIES > IO_TRACKERS is for debugging purposes - when dumping
78  *  the contents of the submission and completion queues, it will show a longer
79  *  history of data.
80  */
81 #define NVME_IO_ENTRIES		(256)
82 #define NVME_IO_TRACKERS	(128)
83 #define NVME_MIN_IO_TRACKERS	(4)
84 #define NVME_MAX_IO_TRACKERS	(1024)
85 
86 /*
87  * NVME_MAX_IO_ENTRIES is not defined, since it is specified in CC.MQES
88  *  for each controller.
89  */
90 
91 #define NVME_INT_COAL_TIME	(0)	/* disabled */
92 #define NVME_INT_COAL_THRESHOLD (0)	/* 0-based */
93 
94 #define NVME_MAX_NAMESPACES	(16)
95 #define NVME_MAX_CONSUMERS	(2)
96 #define NVME_MAX_ASYNC_EVENTS	(8)
97 
98 #define NVME_DEFAULT_TIMEOUT_PERIOD	(30)    /* in seconds */
99 #define NVME_MIN_TIMEOUT_PERIOD		(5)
100 #define NVME_MAX_TIMEOUT_PERIOD		(120)
101 
102 #define NVME_DEFAULT_RETRY_COUNT	(4)
103 
104 /* Maximum log page size to fetch for AERs. */
105 #define NVME_MAX_AER_LOG_SIZE		(4096)
106 
107 /*
108  * Define CACHE_LINE_SIZE here for older FreeBSD versions that do not define
109  *  it.
110  */
111 #ifndef CACHE_LINE_SIZE
112 #define CACHE_LINE_SIZE		(64)
113 #endif
114 
115 extern uma_zone_t	nvme_request_zone;
116 extern int32_t		nvme_retry_count;
117 extern bool		nvme_verbose_cmd_dump;
118 
119 struct nvme_completion_poll_status {
120 
121 	struct nvme_completion	cpl;
122 	int			done;
123 };
124 
125 #define NVME_REQUEST_VADDR	1
126 #define NVME_REQUEST_NULL	2 /* For requests with no payload. */
127 #define NVME_REQUEST_UIO	3
128 #define NVME_REQUEST_BIO	4
129 #define NVME_REQUEST_CCB        5
130 
131 struct nvme_request {
132 
133 	struct nvme_command		cmd;
134 	struct nvme_qpair		*qpair;
135 	union {
136 		void			*payload;
137 		struct bio		*bio;
138 	} u;
139 	uint32_t			type;
140 	uint32_t			payload_size;
141 	boolean_t			timeout;
142 	nvme_cb_fn_t			cb_fn;
143 	void				*cb_arg;
144 	int32_t				retries;
145 	STAILQ_ENTRY(nvme_request)	stailq;
146 };
147 
148 struct nvme_async_event_request {
149 
150 	struct nvme_controller		*ctrlr;
151 	struct nvme_request		*req;
152 	struct nvme_completion		cpl;
153 	uint32_t			log_page_id;
154 	uint32_t			log_page_size;
155 	uint8_t				log_page_buffer[NVME_MAX_AER_LOG_SIZE];
156 };
157 
158 struct nvme_tracker {
159 
160 	TAILQ_ENTRY(nvme_tracker)	tailq;
161 	struct nvme_request		*req;
162 	struct nvme_qpair		*qpair;
163 	struct callout			timer;
164 	bus_dmamap_t			payload_dma_map;
165 	uint16_t			cid;
166 
167 	uint64_t			*prp;
168 	bus_addr_t			prp_bus_addr;
169 };
170 
171 struct nvme_qpair {
172 
173 	struct nvme_controller	*ctrlr;
174 	uint32_t		id;
175 	uint32_t		phase;
176 
177 	uint16_t		vector;
178 	int			rid;
179 	struct resource		*res;
180 	void 			*tag;
181 
182 	uint32_t		num_entries;
183 	uint32_t		num_trackers;
184 	uint32_t		sq_tdbl_off;
185 	uint32_t		cq_hdbl_off;
186 
187 	uint32_t		sq_head;
188 	uint32_t		sq_tail;
189 	uint32_t		cq_head;
190 
191 	int64_t			num_cmds;
192 	int64_t			num_intr_handler_calls;
193 	int64_t			num_retries;
194 	int64_t			num_failures;
195 
196 	struct nvme_command	*cmd;
197 	struct nvme_completion	*cpl;
198 
199 	bus_dma_tag_t		dma_tag;
200 	bus_dma_tag_t		dma_tag_payload;
201 
202 	bus_dmamap_t		queuemem_map;
203 	uint64_t		cmd_bus_addr;
204 	uint64_t		cpl_bus_addr;
205 
206 	TAILQ_HEAD(, nvme_tracker)	free_tr;
207 	TAILQ_HEAD(, nvme_tracker)	outstanding_tr;
208 	STAILQ_HEAD(, nvme_request)	queued_req;
209 
210 	struct nvme_tracker	**act_tr;
211 
212 	boolean_t		is_enabled;
213 
214 	struct mtx		lock __aligned(CACHE_LINE_SIZE);
215 
216 } __aligned(CACHE_LINE_SIZE);
217 
218 struct nvme_namespace {
219 
220 	struct nvme_controller		*ctrlr;
221 	struct nvme_namespace_data	data;
222 	uint32_t			id;
223 	uint32_t			flags;
224 	struct cdev			*cdev;
225 	void				*cons_cookie[NVME_MAX_CONSUMERS];
226 	uint32_t			stripesize;
227 	struct mtx			lock;
228 };
229 
230 /*
231  * One of these per allocated PCI device.
232  */
233 struct nvme_controller {
234 
235 	device_t		dev;
236 
237 	struct mtx		lock;
238 
239 	uint32_t		ready_timeout_in_ms;
240 	uint32_t		quirks;
241 #define	QUIRK_DELAY_B4_CHK_RDY	1		/* Can't touch MMIO on disable */
242 #define	QUIRK_DISABLE_TIMEOUT	2		/* Disable broken completion timeout feature */
243 
244 	bus_space_tag_t		bus_tag;
245 	bus_space_handle_t	bus_handle;
246 	int			resource_id;
247 	struct resource		*resource;
248 
249 	/*
250 	 * The NVMe spec allows for the MSI-X table to be placed in BAR 4/5,
251 	 *  separate from the control registers which are in BAR 0/1.  These
252 	 *  members track the mapping of BAR 4/5 for that reason.
253 	 */
254 	int			bar4_resource_id;
255 	struct resource		*bar4_resource;
256 
257 	uint32_t		msix_enabled;
258 	uint32_t		force_intx;
259 	uint32_t		enable_aborts;
260 
261 	uint32_t		num_io_queues;
262 	uint32_t		num_cpus_per_ioq;
263 	uint32_t		max_hw_pend_io;
264 
265 	/* Fields for tracking progress during controller initialization. */
266 	struct intr_config_hook	config_hook;
267 	uint32_t		ns_identified;
268 	uint32_t		queues_created;
269 
270 	struct task		reset_task;
271 	struct task		fail_req_task;
272 	struct taskqueue	*taskqueue;
273 
274 	/* For shared legacy interrupt. */
275 	int			rid;
276 	struct resource		*res;
277 	void			*tag;
278 
279 	bus_dma_tag_t		hw_desc_tag;
280 	bus_dmamap_t		hw_desc_map;
281 
282 	/** maximum i/o size in bytes */
283 	uint32_t		max_xfer_size;
284 
285 	/** minimum page size supported by this controller in bytes */
286 	uint32_t		min_page_size;
287 
288 	/** interrupt coalescing time period (in microseconds) */
289 	uint32_t		int_coal_time;
290 
291 	/** interrupt coalescing threshold */
292 	uint32_t		int_coal_threshold;
293 
294 	/** timeout period in seconds */
295 	uint32_t		timeout_period;
296 
297 	struct nvme_qpair	adminq;
298 	struct nvme_qpair	*ioq;
299 
300 	struct nvme_registers		*regs;
301 
302 	struct nvme_controller_data	cdata;
303 	struct nvme_namespace		ns[NVME_MAX_NAMESPACES];
304 
305 	struct cdev			*cdev;
306 
307 	/** bit mask of event types currently enabled for async events */
308 	uint32_t			async_event_config;
309 
310 	uint32_t			num_aers;
311 	struct nvme_async_event_request	aer[NVME_MAX_ASYNC_EVENTS];
312 
313 	void				*cons_cookie[NVME_MAX_CONSUMERS];
314 
315 	uint32_t			is_resetting;
316 	uint32_t			is_initialized;
317 	uint32_t			notification_sent;
318 
319 	boolean_t			is_failed;
320 	STAILQ_HEAD(, nvme_request)	fail_req;
321 };
322 
323 #define nvme_mmio_offsetof(reg)						       \
324 	offsetof(struct nvme_registers, reg)
325 
326 #define nvme_mmio_read_4(sc, reg)					       \
327 	bus_space_read_4((sc)->bus_tag, (sc)->bus_handle,		       \
328 	    nvme_mmio_offsetof(reg))
329 
330 #define nvme_mmio_write_4(sc, reg, val)					       \
331 	bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,		       \
332 	    nvme_mmio_offsetof(reg), val)
333 
334 #define nvme_mmio_write_8(sc, reg, val)					       \
335 	do {								       \
336 		bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,	       \
337 		    nvme_mmio_offsetof(reg), val & 0xFFFFFFFF); 	       \
338 		bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,	       \
339 		    nvme_mmio_offsetof(reg)+4,				       \
340 		    (val & 0xFFFFFFFF00000000ULL) >> 32);		       \
341 	} while (0);
342 
343 #define nvme_printf(ctrlr, fmt, args...)	\
344     device_printf(ctrlr->dev, fmt, ##args)
345 
346 void	nvme_ns_test(struct nvme_namespace *ns, u_long cmd, caddr_t arg);
347 
348 void	nvme_ctrlr_cmd_identify_controller(struct nvme_controller *ctrlr,
349 					   void *payload,
350 					   nvme_cb_fn_t cb_fn, void *cb_arg);
351 void	nvme_ctrlr_cmd_identify_namespace(struct nvme_controller *ctrlr,
352 					  uint32_t nsid, void *payload,
353 					  nvme_cb_fn_t cb_fn, void *cb_arg);
354 void	nvme_ctrlr_cmd_set_interrupt_coalescing(struct nvme_controller *ctrlr,
355 						uint32_t microseconds,
356 						uint32_t threshold,
357 						nvme_cb_fn_t cb_fn,
358 						void *cb_arg);
359 void	nvme_ctrlr_cmd_get_error_page(struct nvme_controller *ctrlr,
360 				      struct nvme_error_information_entry *payload,
361 				      uint32_t num_entries, /* 0 = max */
362 				      nvme_cb_fn_t cb_fn,
363 				      void *cb_arg);
364 void	nvme_ctrlr_cmd_get_health_information_page(struct nvme_controller *ctrlr,
365 						   uint32_t nsid,
366 						   struct nvme_health_information_page *payload,
367 						   nvme_cb_fn_t cb_fn,
368 						   void *cb_arg);
369 void	nvme_ctrlr_cmd_get_firmware_page(struct nvme_controller *ctrlr,
370 					 struct nvme_firmware_page *payload,
371 					 nvme_cb_fn_t cb_fn,
372 					 void *cb_arg);
373 void	nvme_ctrlr_cmd_create_io_cq(struct nvme_controller *ctrlr,
374 				    struct nvme_qpair *io_que, uint16_t vector,
375 				    nvme_cb_fn_t cb_fn, void *cb_arg);
376 void	nvme_ctrlr_cmd_create_io_sq(struct nvme_controller *ctrlr,
377 				    struct nvme_qpair *io_que,
378 				    nvme_cb_fn_t cb_fn, void *cb_arg);
379 void	nvme_ctrlr_cmd_delete_io_cq(struct nvme_controller *ctrlr,
380 				    struct nvme_qpair *io_que,
381 				    nvme_cb_fn_t cb_fn, void *cb_arg);
382 void	nvme_ctrlr_cmd_delete_io_sq(struct nvme_controller *ctrlr,
383 				    struct nvme_qpair *io_que,
384 				    nvme_cb_fn_t cb_fn, void *cb_arg);
385 void	nvme_ctrlr_cmd_set_num_queues(struct nvme_controller *ctrlr,
386 				      uint32_t num_queues, nvme_cb_fn_t cb_fn,
387 				      void *cb_arg);
388 void	nvme_ctrlr_cmd_set_async_event_config(struct nvme_controller *ctrlr,
389 					      uint32_t state,
390 					      nvme_cb_fn_t cb_fn, void *cb_arg);
391 void	nvme_ctrlr_cmd_abort(struct nvme_controller *ctrlr, uint16_t cid,
392 			     uint16_t sqid, nvme_cb_fn_t cb_fn, void *cb_arg);
393 
394 void	nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl);
395 
396 int	nvme_ctrlr_construct(struct nvme_controller *ctrlr, device_t dev);
397 void	nvme_ctrlr_destruct(struct nvme_controller *ctrlr, device_t dev);
398 void	nvme_ctrlr_shutdown(struct nvme_controller *ctrlr);
399 int	nvme_ctrlr_hw_reset(struct nvme_controller *ctrlr);
400 void	nvme_ctrlr_reset(struct nvme_controller *ctrlr);
401 /* ctrlr defined as void * to allow use with config_intrhook. */
402 void	nvme_ctrlr_start_config_hook(void *ctrlr_arg);
403 void	nvme_ctrlr_submit_admin_request(struct nvme_controller *ctrlr,
404 					struct nvme_request *req);
405 void	nvme_ctrlr_submit_io_request(struct nvme_controller *ctrlr,
406 				     struct nvme_request *req);
407 void	nvme_ctrlr_post_failed_request(struct nvme_controller *ctrlr,
408 				       struct nvme_request *req);
409 
410 int	nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t id,
411 			     uint16_t vector, uint32_t num_entries,
412 			     uint32_t num_trackers,
413 			     struct nvme_controller *ctrlr);
414 void	nvme_qpair_submit_tracker(struct nvme_qpair *qpair,
415 				  struct nvme_tracker *tr);
416 bool	nvme_qpair_process_completions(struct nvme_qpair *qpair);
417 void	nvme_qpair_submit_request(struct nvme_qpair *qpair,
418 				  struct nvme_request *req);
419 void	nvme_qpair_reset(struct nvme_qpair *qpair);
420 void	nvme_qpair_fail(struct nvme_qpair *qpair);
421 void	nvme_qpair_manual_complete_request(struct nvme_qpair *qpair,
422 					   struct nvme_request *req,
423                                            uint32_t sct, uint32_t sc);
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_dump_command(struct nvme_command *cmd);
440 void	nvme_dump_completion(struct nvme_completion *cpl);
441 
442 static __inline void
443 nvme_single_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
444 {
445 	uint64_t *bus_addr = (uint64_t *)arg;
446 
447 	if (error != 0)
448 		printf("nvme_single_map err %d\n", error);
449 	*bus_addr = seg[0].ds_addr;
450 }
451 
452 static __inline struct nvme_request *
453 _nvme_allocate_request(nvme_cb_fn_t cb_fn, void *cb_arg)
454 {
455 	struct nvme_request *req;
456 
457 	req = uma_zalloc(nvme_request_zone, M_NOWAIT | M_ZERO);
458 	if (req != NULL) {
459 		req->cb_fn = cb_fn;
460 		req->cb_arg = cb_arg;
461 		req->timeout = TRUE;
462 	}
463 	return (req);
464 }
465 
466 static __inline struct nvme_request *
467 nvme_allocate_request_vaddr(void *payload, uint32_t payload_size,
468     nvme_cb_fn_t cb_fn, void *cb_arg)
469 {
470 	struct nvme_request *req;
471 
472 	req = _nvme_allocate_request(cb_fn, cb_arg);
473 	if (req != NULL) {
474 		req->type = NVME_REQUEST_VADDR;
475 		req->u.payload = payload;
476 		req->payload_size = payload_size;
477 	}
478 	return (req);
479 }
480 
481 static __inline struct nvme_request *
482 nvme_allocate_request_null(nvme_cb_fn_t cb_fn, void *cb_arg)
483 {
484 	struct nvme_request *req;
485 
486 	req = _nvme_allocate_request(cb_fn, cb_arg);
487 	if (req != NULL)
488 		req->type = NVME_REQUEST_NULL;
489 	return (req);
490 }
491 
492 static __inline struct nvme_request *
493 nvme_allocate_request_bio(struct bio *bio, 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_BIO;
500 		req->u.bio = bio;
501 	}
502 	return (req);
503 }
504 
505 static __inline struct nvme_request *
506 nvme_allocate_request_ccb(union ccb *ccb, nvme_cb_fn_t cb_fn, void *cb_arg)
507 {
508 	struct nvme_request *req;
509 
510 	req = _nvme_allocate_request(cb_fn, cb_arg);
511 	if (req != NULL) {
512 		req->type = NVME_REQUEST_CCB;
513 		req->u.payload = ccb;
514 	}
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 void	nvme_notify_ns(struct nvme_controller *ctrlr, int nsid);
528 
529 void	nvme_ctrlr_intx_handler(void *arg);
530 void	nvme_ctrlr_poll(struct nvme_controller *ctrlr);
531 
532 #endif /* __NVME_PRIVATE_H__ */
533