xref: /freebsd/sys/dev/nvme/nvme_private.h (revision 148a8da8)
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 
118 struct nvme_completion_poll_status {
119 
120 	struct nvme_completion	cpl;
121 	int			done;
122 };
123 
124 #define NVME_REQUEST_VADDR	1
125 #define NVME_REQUEST_NULL	2 /* For requests with no payload. */
126 #define NVME_REQUEST_UIO	3
127 #define NVME_REQUEST_BIO	4
128 #define NVME_REQUEST_CCB        5
129 
130 struct nvme_request {
131 
132 	struct nvme_command		cmd;
133 	struct nvme_qpair		*qpair;
134 	union {
135 		void			*payload;
136 		struct bio		*bio;
137 	} u;
138 	uint32_t			type;
139 	uint32_t			payload_size;
140 	boolean_t			timeout;
141 	nvme_cb_fn_t			cb_fn;
142 	void				*cb_arg;
143 	int32_t				retries;
144 	STAILQ_ENTRY(nvme_request)	stailq;
145 };
146 
147 struct nvme_async_event_request {
148 
149 	struct nvme_controller		*ctrlr;
150 	struct nvme_request		*req;
151 	struct nvme_completion		cpl;
152 	uint32_t			log_page_id;
153 	uint32_t			log_page_size;
154 	uint8_t				log_page_buffer[NVME_MAX_AER_LOG_SIZE];
155 };
156 
157 struct nvme_tracker {
158 
159 	TAILQ_ENTRY(nvme_tracker)	tailq;
160 	struct nvme_request		*req;
161 	struct nvme_qpair		*qpair;
162 	struct callout			timer;
163 	bus_dmamap_t			payload_dma_map;
164 	uint16_t			cid;
165 
166 	uint64_t			*prp;
167 	bus_addr_t			prp_bus_addr;
168 };
169 
170 struct nvme_qpair {
171 
172 	struct nvme_controller	*ctrlr;
173 	uint32_t		id;
174 	uint32_t		phase;
175 
176 	uint16_t		vector;
177 	int			rid;
178 	struct resource		*res;
179 	void 			*tag;
180 
181 	uint32_t		num_entries;
182 	uint32_t		num_trackers;
183 	uint32_t		sq_tdbl_off;
184 	uint32_t		cq_hdbl_off;
185 
186 	uint32_t		sq_head;
187 	uint32_t		sq_tail;
188 	uint32_t		cq_head;
189 
190 	int64_t			num_cmds;
191 	int64_t			num_intr_handler_calls;
192 
193 	struct nvme_command	*cmd;
194 	struct nvme_completion	*cpl;
195 
196 	bus_dma_tag_t		dma_tag;
197 	bus_dma_tag_t		dma_tag_payload;
198 
199 	bus_dmamap_t		queuemem_map;
200 	uint64_t		cmd_bus_addr;
201 	uint64_t		cpl_bus_addr;
202 
203 	TAILQ_HEAD(, nvme_tracker)	free_tr;
204 	TAILQ_HEAD(, nvme_tracker)	outstanding_tr;
205 	STAILQ_HEAD(, nvme_request)	queued_req;
206 
207 	struct nvme_tracker	**act_tr;
208 
209 	boolean_t		is_enabled;
210 
211 	struct mtx		lock __aligned(CACHE_LINE_SIZE);
212 
213 } __aligned(CACHE_LINE_SIZE);
214 
215 struct nvme_namespace {
216 
217 	struct nvme_controller		*ctrlr;
218 	struct nvme_namespace_data	data;
219 	uint32_t			id;
220 	uint32_t			flags;
221 	struct cdev			*cdev;
222 	void				*cons_cookie[NVME_MAX_CONSUMERS];
223 	uint32_t			stripesize;
224 	struct mtx			lock;
225 };
226 
227 /*
228  * One of these per allocated PCI device.
229  */
230 struct nvme_controller {
231 
232 	device_t		dev;
233 
234 	struct mtx		lock;
235 
236 	uint32_t		ready_timeout_in_ms;
237 	uint32_t		quirks;
238 #define	QUIRK_DELAY_B4_CHK_RDY	1		/* Can't touch MMIO on disable */
239 #define	QUIRK_DISABLE_TIMEOUT	2		/* Disable broken completion timeout feature */
240 
241 	bus_space_tag_t		bus_tag;
242 	bus_space_handle_t	bus_handle;
243 	int			resource_id;
244 	struct resource		*resource;
245 
246 	/*
247 	 * The NVMe spec allows for the MSI-X table to be placed in BAR 4/5,
248 	 *  separate from the control registers which are in BAR 0/1.  These
249 	 *  members track the mapping of BAR 4/5 for that reason.
250 	 */
251 	int			bar4_resource_id;
252 	struct resource		*bar4_resource;
253 
254 	uint32_t		msix_enabled;
255 	uint32_t		force_intx;
256 	uint32_t		enable_aborts;
257 
258 	uint32_t		num_io_queues;
259 	uint32_t		num_cpus_per_ioq;
260 	uint32_t		max_hw_pend_io;
261 
262 	/* Fields for tracking progress during controller initialization. */
263 	struct intr_config_hook	config_hook;
264 	uint32_t		ns_identified;
265 	uint32_t		queues_created;
266 
267 	struct task		reset_task;
268 	struct task		fail_req_task;
269 	struct taskqueue	*taskqueue;
270 
271 	/* For shared legacy interrupt. */
272 	int			rid;
273 	struct resource		*res;
274 	void			*tag;
275 
276 	bus_dma_tag_t		hw_desc_tag;
277 	bus_dmamap_t		hw_desc_map;
278 
279 	/** maximum i/o size in bytes */
280 	uint32_t		max_xfer_size;
281 
282 	/** minimum page size supported by this controller in bytes */
283 	uint32_t		min_page_size;
284 
285 	/** interrupt coalescing time period (in microseconds) */
286 	uint32_t		int_coal_time;
287 
288 	/** interrupt coalescing threshold */
289 	uint32_t		int_coal_threshold;
290 
291 	/** timeout period in seconds */
292 	uint32_t		timeout_period;
293 
294 	struct nvme_qpair	adminq;
295 	struct nvme_qpair	*ioq;
296 
297 	struct nvme_registers		*regs;
298 
299 	struct nvme_controller_data	cdata;
300 	struct nvme_namespace		ns[NVME_MAX_NAMESPACES];
301 
302 	struct cdev			*cdev;
303 
304 	/** bit mask of event types currently enabled for async events */
305 	uint32_t			async_event_config;
306 
307 	uint32_t			num_aers;
308 	struct nvme_async_event_request	aer[NVME_MAX_ASYNC_EVENTS];
309 
310 	void				*cons_cookie[NVME_MAX_CONSUMERS];
311 
312 	uint32_t			is_resetting;
313 	uint32_t			is_initialized;
314 	uint32_t			notification_sent;
315 
316 	boolean_t			is_failed;
317 	STAILQ_HEAD(, nvme_request)	fail_req;
318 };
319 
320 #define nvme_mmio_offsetof(reg)						       \
321 	offsetof(struct nvme_registers, reg)
322 
323 #define nvme_mmio_read_4(sc, reg)					       \
324 	bus_space_read_4((sc)->bus_tag, (sc)->bus_handle,		       \
325 	    nvme_mmio_offsetof(reg))
326 
327 #define nvme_mmio_write_4(sc, reg, val)					       \
328 	bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,		       \
329 	    nvme_mmio_offsetof(reg), val)
330 
331 #define nvme_mmio_write_8(sc, reg, val)					       \
332 	do {								       \
333 		bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,	       \
334 		    nvme_mmio_offsetof(reg), val & 0xFFFFFFFF); 	       \
335 		bus_space_write_4((sc)->bus_tag, (sc)->bus_handle,	       \
336 		    nvme_mmio_offsetof(reg)+4,				       \
337 		    (val & 0xFFFFFFFF00000000ULL) >> 32);		       \
338 	} while (0);
339 
340 #define nvme_printf(ctrlr, fmt, args...)	\
341     device_printf(ctrlr->dev, fmt, ##args)
342 
343 void	nvme_ns_test(struct nvme_namespace *ns, u_long cmd, caddr_t arg);
344 
345 void	nvme_ctrlr_cmd_identify_controller(struct nvme_controller *ctrlr,
346 					   void *payload,
347 					   nvme_cb_fn_t cb_fn, void *cb_arg);
348 void	nvme_ctrlr_cmd_identify_namespace(struct nvme_controller *ctrlr,
349 					  uint32_t nsid, void *payload,
350 					  nvme_cb_fn_t cb_fn, void *cb_arg);
351 void	nvme_ctrlr_cmd_set_interrupt_coalescing(struct nvme_controller *ctrlr,
352 						uint32_t microseconds,
353 						uint32_t threshold,
354 						nvme_cb_fn_t cb_fn,
355 						void *cb_arg);
356 void	nvme_ctrlr_cmd_get_error_page(struct nvme_controller *ctrlr,
357 				      struct nvme_error_information_entry *payload,
358 				      uint32_t num_entries, /* 0 = max */
359 				      nvme_cb_fn_t cb_fn,
360 				      void *cb_arg);
361 void	nvme_ctrlr_cmd_get_health_information_page(struct nvme_controller *ctrlr,
362 						   uint32_t nsid,
363 						   struct nvme_health_information_page *payload,
364 						   nvme_cb_fn_t cb_fn,
365 						   void *cb_arg);
366 void	nvme_ctrlr_cmd_get_firmware_page(struct nvme_controller *ctrlr,
367 					 struct nvme_firmware_page *payload,
368 					 nvme_cb_fn_t cb_fn,
369 					 void *cb_arg);
370 void	nvme_ctrlr_cmd_create_io_cq(struct nvme_controller *ctrlr,
371 				    struct nvme_qpair *io_que, uint16_t vector,
372 				    nvme_cb_fn_t cb_fn, void *cb_arg);
373 void	nvme_ctrlr_cmd_create_io_sq(struct nvme_controller *ctrlr,
374 				    struct nvme_qpair *io_que,
375 				    nvme_cb_fn_t cb_fn, void *cb_arg);
376 void	nvme_ctrlr_cmd_delete_io_cq(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_sq(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_set_num_queues(struct nvme_controller *ctrlr,
383 				      uint32_t num_queues, nvme_cb_fn_t cb_fn,
384 				      void *cb_arg);
385 void	nvme_ctrlr_cmd_set_async_event_config(struct nvme_controller *ctrlr,
386 					      uint32_t state,
387 					      nvme_cb_fn_t cb_fn, void *cb_arg);
388 void	nvme_ctrlr_cmd_abort(struct nvme_controller *ctrlr, uint16_t cid,
389 			     uint16_t sqid, nvme_cb_fn_t cb_fn, void *cb_arg);
390 
391 void	nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl);
392 
393 int	nvme_ctrlr_construct(struct nvme_controller *ctrlr, device_t dev);
394 void	nvme_ctrlr_destruct(struct nvme_controller *ctrlr, device_t dev);
395 void	nvme_ctrlr_shutdown(struct nvme_controller *ctrlr);
396 int	nvme_ctrlr_hw_reset(struct nvme_controller *ctrlr);
397 void	nvme_ctrlr_reset(struct nvme_controller *ctrlr);
398 /* ctrlr defined as void * to allow use with config_intrhook. */
399 void	nvme_ctrlr_start_config_hook(void *ctrlr_arg);
400 void	nvme_ctrlr_submit_admin_request(struct nvme_controller *ctrlr,
401 					struct nvme_request *req);
402 void	nvme_ctrlr_submit_io_request(struct nvme_controller *ctrlr,
403 				     struct nvme_request *req);
404 void	nvme_ctrlr_post_failed_request(struct nvme_controller *ctrlr,
405 				       struct nvme_request *req);
406 
407 int	nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t id,
408 			     uint16_t vector, uint32_t num_entries,
409 			     uint32_t num_trackers,
410 			     struct nvme_controller *ctrlr);
411 void	nvme_qpair_submit_tracker(struct nvme_qpair *qpair,
412 				  struct nvme_tracker *tr);
413 bool	nvme_qpair_process_completions(struct nvme_qpair *qpair);
414 void	nvme_qpair_submit_request(struct nvme_qpair *qpair,
415 				  struct nvme_request *req);
416 void	nvme_qpair_reset(struct nvme_qpair *qpair);
417 void	nvme_qpair_fail(struct nvme_qpair *qpair);
418 void	nvme_qpair_manual_complete_request(struct nvme_qpair *qpair,
419 					   struct nvme_request *req,
420                                            uint32_t sct, uint32_t sc);
421 
422 void	nvme_admin_qpair_enable(struct nvme_qpair *qpair);
423 void	nvme_admin_qpair_disable(struct nvme_qpair *qpair);
424 void	nvme_admin_qpair_destroy(struct nvme_qpair *qpair);
425 
426 void	nvme_io_qpair_enable(struct nvme_qpair *qpair);
427 void	nvme_io_qpair_disable(struct nvme_qpair *qpair);
428 void	nvme_io_qpair_destroy(struct nvme_qpair *qpair);
429 
430 int	nvme_ns_construct(struct nvme_namespace *ns, uint32_t id,
431 			  struct nvme_controller *ctrlr);
432 void	nvme_ns_destruct(struct nvme_namespace *ns);
433 
434 void	nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr);
435 
436 void	nvme_dump_command(struct nvme_command *cmd);
437 void	nvme_dump_completion(struct nvme_completion *cpl);
438 
439 static __inline void
440 nvme_single_map(void *arg, bus_dma_segment_t *seg, int nseg, int error)
441 {
442 	uint64_t *bus_addr = (uint64_t *)arg;
443 
444 	if (error != 0)
445 		printf("nvme_single_map err %d\n", error);
446 	*bus_addr = seg[0].ds_addr;
447 }
448 
449 static __inline struct nvme_request *
450 _nvme_allocate_request(nvme_cb_fn_t cb_fn, void *cb_arg)
451 {
452 	struct nvme_request *req;
453 
454 	req = uma_zalloc(nvme_request_zone, M_NOWAIT | M_ZERO);
455 	if (req != NULL) {
456 		req->cb_fn = cb_fn;
457 		req->cb_arg = cb_arg;
458 		req->timeout = TRUE;
459 	}
460 	return (req);
461 }
462 
463 static __inline struct nvme_request *
464 nvme_allocate_request_vaddr(void *payload, uint32_t payload_size,
465     nvme_cb_fn_t cb_fn, void *cb_arg)
466 {
467 	struct nvme_request *req;
468 
469 	req = _nvme_allocate_request(cb_fn, cb_arg);
470 	if (req != NULL) {
471 		req->type = NVME_REQUEST_VADDR;
472 		req->u.payload = payload;
473 		req->payload_size = payload_size;
474 	}
475 	return (req);
476 }
477 
478 static __inline struct nvme_request *
479 nvme_allocate_request_null(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_NULL;
486 	return (req);
487 }
488 
489 static __inline struct nvme_request *
490 nvme_allocate_request_bio(struct bio *bio, 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_BIO;
497 		req->u.bio = bio;
498 	}
499 	return (req);
500 }
501 
502 static __inline struct nvme_request *
503 nvme_allocate_request_ccb(union ccb *ccb, nvme_cb_fn_t cb_fn, void *cb_arg)
504 {
505 	struct nvme_request *req;
506 
507 	req = _nvme_allocate_request(cb_fn, cb_arg);
508 	if (req != NULL) {
509 		req->type = NVME_REQUEST_CCB;
510 		req->u.payload = ccb;
511 	}
512 
513 	return (req);
514 }
515 
516 #define nvme_free_request(req)	uma_zfree(nvme_request_zone, req)
517 
518 void	nvme_notify_async_consumers(struct nvme_controller *ctrlr,
519 				    const struct nvme_completion *async_cpl,
520 				    uint32_t log_page_id, void *log_page_buffer,
521 				    uint32_t log_page_size);
522 void	nvme_notify_fail_consumers(struct nvme_controller *ctrlr);
523 void	nvme_notify_new_controller(struct nvme_controller *ctrlr);
524 void	nvme_notify_ns(struct nvme_controller *ctrlr, int nsid);
525 
526 void	nvme_ctrlr_intx_handler(void *arg);
527 void	nvme_ctrlr_poll(struct nvme_controller *ctrlr);
528 
529 #endif /* __NVME_PRIVATE_H__ */
530