]>
Commit | Line | Data |
---|---|---|
65be2c79 MO |
1 | /* |
2 | * CXL Flash Device Driver | |
3 | * | |
4 | * Written by: Manoj N. Kumar <manoj@linux.vnet.ibm.com>, IBM Corporation | |
5 | * Matthew R. Ochs <mrochs@linux.vnet.ibm.com>, IBM Corporation | |
6 | * | |
7 | * Copyright (C) 2015 IBM Corporation | |
8 | * | |
9 | * This program is free software; you can redistribute it and/or | |
10 | * modify it under the terms of the GNU General Public License | |
11 | * as published by the Free Software Foundation; either version | |
12 | * 2 of the License, or (at your option) any later version. | |
13 | */ | |
14 | ||
15 | #include <linux/delay.h> | |
16 | #include <linux/file.h> | |
17 | #include <linux/syscalls.h> | |
18 | #include <misc/cxl.h> | |
19 | #include <asm/unaligned.h> | |
20 | ||
21 | #include <scsi/scsi.h> | |
22 | #include <scsi/scsi_host.h> | |
23 | #include <scsi/scsi_cmnd.h> | |
24 | #include <scsi/scsi_eh.h> | |
25 | #include <uapi/scsi/cxlflash_ioctl.h> | |
26 | ||
27 | #include "sislite.h" | |
28 | #include "common.h" | |
2cb79266 | 29 | #include "vlun.h" |
65be2c79 MO |
30 | #include "superpipe.h" |
31 | ||
32 | struct cxlflash_global global; | |
33 | ||
2cb79266 MO |
34 | /** |
35 | * marshal_rele_to_resize() - translate release to resize structure | |
36 | * @rele: Source structure from which to translate/copy. | |
37 | * @resize: Destination structure for the translate/copy. | |
38 | */ | |
39 | static void marshal_rele_to_resize(struct dk_cxlflash_release *release, | |
40 | struct dk_cxlflash_resize *resize) | |
41 | { | |
42 | resize->hdr = release->hdr; | |
43 | resize->context_id = release->context_id; | |
44 | resize->rsrc_handle = release->rsrc_handle; | |
45 | } | |
46 | ||
65be2c79 MO |
47 | /** |
48 | * marshal_det_to_rele() - translate detach to release structure | |
49 | * @detach: Destination structure for the translate/copy. | |
50 | * @rele: Source structure from which to translate/copy. | |
51 | */ | |
52 | static void marshal_det_to_rele(struct dk_cxlflash_detach *detach, | |
53 | struct dk_cxlflash_release *release) | |
54 | { | |
55 | release->hdr = detach->hdr; | |
56 | release->context_id = detach->context_id; | |
57 | } | |
58 | ||
59 | /** | |
60 | * cxlflash_free_errpage() - frees resources associated with global error page | |
61 | */ | |
62 | void cxlflash_free_errpage(void) | |
63 | { | |
64 | ||
65 | mutex_lock(&global.mutex); | |
66 | if (global.err_page) { | |
67 | __free_page(global.err_page); | |
68 | global.err_page = NULL; | |
69 | } | |
70 | mutex_unlock(&global.mutex); | |
71 | } | |
72 | ||
73 | /** | |
74 | * cxlflash_stop_term_user_contexts() - stops/terminates known user contexts | |
75 | * @cfg: Internal structure associated with the host. | |
76 | * | |
77 | * When the host needs to go down, all users must be quiesced and their | |
78 | * memory freed. This is accomplished by putting the contexts in error | |
79 | * state which will notify the user and let them 'drive' the tear-down. | |
80 | * Meanwhile, this routine camps until all user contexts have been removed. | |
81 | */ | |
82 | void cxlflash_stop_term_user_contexts(struct cxlflash_cfg *cfg) | |
83 | { | |
84 | struct device *dev = &cfg->dev->dev; | |
85 | int i, found; | |
86 | ||
87 | cxlflash_mark_contexts_error(cfg); | |
88 | ||
89 | while (true) { | |
90 | found = false; | |
91 | ||
92 | for (i = 0; i < MAX_CONTEXT; i++) | |
93 | if (cfg->ctx_tbl[i]) { | |
94 | found = true; | |
95 | break; | |
96 | } | |
97 | ||
98 | if (!found && list_empty(&cfg->ctx_err_recovery)) | |
99 | return; | |
100 | ||
101 | dev_dbg(dev, "%s: Wait for user contexts to quiesce...\n", | |
102 | __func__); | |
103 | wake_up_all(&cfg->limbo_waitq); | |
104 | ssleep(1); | |
105 | } | |
106 | } | |
107 | ||
108 | /** | |
109 | * find_error_context() - locates a context by cookie on the error recovery list | |
110 | * @cfg: Internal structure associated with the host. | |
111 | * @rctxid: Desired context by id. | |
112 | * @file: Desired context by file. | |
113 | * | |
114 | * Return: Found context on success, NULL on failure | |
115 | */ | |
116 | static struct ctx_info *find_error_context(struct cxlflash_cfg *cfg, u64 rctxid, | |
117 | struct file *file) | |
118 | { | |
119 | struct ctx_info *ctxi; | |
120 | ||
121 | list_for_each_entry(ctxi, &cfg->ctx_err_recovery, list) | |
122 | if ((ctxi->ctxid == rctxid) || (ctxi->file == file)) | |
123 | return ctxi; | |
124 | ||
125 | return NULL; | |
126 | } | |
127 | ||
128 | /** | |
129 | * get_context() - obtains a validated and locked context reference | |
130 | * @cfg: Internal structure associated with the host. | |
131 | * @rctxid: Desired context (raw, un-decoded format). | |
132 | * @arg: LUN information or file associated with request. | |
133 | * @ctx_ctrl: Control information to 'steer' desired lookup. | |
134 | * | |
135 | * NOTE: despite the name pid, in linux, current->pid actually refers | |
136 | * to the lightweight process id (tid) and can change if the process is | |
137 | * multi threaded. The tgid remains constant for the process and only changes | |
138 | * when the process of fork. For all intents and purposes, think of tgid | |
139 | * as a pid in the traditional sense. | |
140 | * | |
141 | * Return: Validated context on success, NULL on failure | |
142 | */ | |
143 | struct ctx_info *get_context(struct cxlflash_cfg *cfg, u64 rctxid, | |
144 | void *arg, enum ctx_ctrl ctx_ctrl) | |
145 | { | |
146 | struct device *dev = &cfg->dev->dev; | |
147 | struct ctx_info *ctxi = NULL; | |
148 | struct lun_access *lun_access = NULL; | |
149 | struct file *file = NULL; | |
150 | struct llun_info *lli = arg; | |
151 | u64 ctxid = DECODE_CTXID(rctxid); | |
152 | int rc; | |
153 | pid_t pid = current->tgid, ctxpid = 0; | |
154 | ||
155 | if (ctx_ctrl & CTX_CTRL_FILE) { | |
156 | lli = NULL; | |
157 | file = (struct file *)arg; | |
158 | } | |
159 | ||
160 | if (ctx_ctrl & CTX_CTRL_CLONE) | |
161 | pid = current->parent->tgid; | |
162 | ||
163 | if (likely(ctxid < MAX_CONTEXT)) { | |
164 | while (true) { | |
165 | rc = mutex_lock_interruptible(&cfg->ctx_tbl_list_mutex); | |
166 | if (rc) | |
167 | goto out; | |
168 | ||
169 | ctxi = cfg->ctx_tbl[ctxid]; | |
170 | if (ctxi) | |
171 | if ((file && (ctxi->file != file)) || | |
172 | (!file && (ctxi->ctxid != rctxid))) | |
173 | ctxi = NULL; | |
174 | ||
175 | if ((ctx_ctrl & CTX_CTRL_ERR) || | |
176 | (!ctxi && (ctx_ctrl & CTX_CTRL_ERR_FALLBACK))) | |
177 | ctxi = find_error_context(cfg, rctxid, file); | |
178 | if (!ctxi) { | |
179 | mutex_unlock(&cfg->ctx_tbl_list_mutex); | |
180 | goto out; | |
181 | } | |
182 | ||
183 | /* | |
184 | * Need to acquire ownership of the context while still | |
185 | * under the table/list lock to serialize with a remove | |
186 | * thread. Use the 'try' to avoid stalling the | |
187 | * table/list lock for a single context. | |
188 | * | |
189 | * Note that the lock order is: | |
190 | * | |
191 | * cfg->ctx_tbl_list_mutex -> ctxi->mutex | |
192 | * | |
193 | * Therefore release ctx_tbl_list_mutex before retrying. | |
194 | */ | |
195 | rc = mutex_trylock(&ctxi->mutex); | |
196 | mutex_unlock(&cfg->ctx_tbl_list_mutex); | |
197 | if (rc) | |
198 | break; /* got the context's lock! */ | |
199 | } | |
200 | ||
201 | if (ctxi->unavail) | |
202 | goto denied; | |
203 | ||
204 | ctxpid = ctxi->pid; | |
205 | if (likely(!(ctx_ctrl & CTX_CTRL_NOPID))) | |
206 | if (pid != ctxpid) | |
207 | goto denied; | |
208 | ||
209 | if (lli) { | |
210 | list_for_each_entry(lun_access, &ctxi->luns, list) | |
211 | if (lun_access->lli == lli) | |
212 | goto out; | |
213 | goto denied; | |
214 | } | |
215 | } | |
216 | ||
217 | out: | |
218 | dev_dbg(dev, "%s: rctxid=%016llX ctxinfo=%p ctxpid=%u pid=%u " | |
219 | "ctx_ctrl=%u\n", __func__, rctxid, ctxi, ctxpid, pid, | |
220 | ctx_ctrl); | |
221 | ||
222 | return ctxi; | |
223 | ||
224 | denied: | |
225 | mutex_unlock(&ctxi->mutex); | |
226 | ctxi = NULL; | |
227 | goto out; | |
228 | } | |
229 | ||
230 | /** | |
231 | * put_context() - release a context that was retrieved from get_context() | |
232 | * @ctxi: Context to release. | |
233 | * | |
234 | * For now, releasing the context equates to unlocking it's mutex. | |
235 | */ | |
236 | void put_context(struct ctx_info *ctxi) | |
237 | { | |
238 | mutex_unlock(&ctxi->mutex); | |
239 | } | |
240 | ||
241 | /** | |
242 | * afu_attach() - attach a context to the AFU | |
243 | * @cfg: Internal structure associated with the host. | |
244 | * @ctxi: Context to attach. | |
245 | * | |
246 | * Upon setting the context capabilities, they must be confirmed with | |
247 | * a read back operation as the context might have been closed since | |
248 | * the mailbox was unlocked. When this occurs, registration is failed. | |
249 | * | |
250 | * Return: 0 on success, -errno on failure | |
251 | */ | |
252 | static int afu_attach(struct cxlflash_cfg *cfg, struct ctx_info *ctxi) | |
253 | { | |
254 | struct device *dev = &cfg->dev->dev; | |
255 | struct afu *afu = cfg->afu; | |
256 | struct sisl_ctrl_map *ctrl_map = ctxi->ctrl_map; | |
257 | int rc = 0; | |
258 | u64 val; | |
259 | ||
260 | /* Unlock cap and restrict user to read/write cmds in translated mode */ | |
261 | readq_be(&ctrl_map->mbox_r); | |
262 | val = (SISL_CTX_CAP_READ_CMD | SISL_CTX_CAP_WRITE_CMD); | |
263 | writeq_be(val, &ctrl_map->ctx_cap); | |
264 | val = readq_be(&ctrl_map->ctx_cap); | |
265 | if (val != (SISL_CTX_CAP_READ_CMD | SISL_CTX_CAP_WRITE_CMD)) { | |
266 | dev_err(dev, "%s: ctx may be closed val=%016llX\n", | |
267 | __func__, val); | |
268 | rc = -EAGAIN; | |
269 | goto out; | |
270 | } | |
271 | ||
272 | /* Set up MMIO registers pointing to the RHT */ | |
273 | writeq_be((u64)ctxi->rht_start, &ctrl_map->rht_start); | |
274 | val = SISL_RHT_CNT_ID((u64)MAX_RHT_PER_CONTEXT, (u64)(afu->ctx_hndl)); | |
275 | writeq_be(val, &ctrl_map->rht_cnt_id); | |
276 | out: | |
277 | dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc); | |
278 | return rc; | |
279 | } | |
280 | ||
281 | /** | |
282 | * read_cap16() - issues a SCSI READ_CAP16 command | |
283 | * @sdev: SCSI device associated with LUN. | |
284 | * @lli: LUN destined for capacity request. | |
285 | * | |
286 | * Return: 0 on success, -errno on failure | |
287 | */ | |
288 | static int read_cap16(struct scsi_device *sdev, struct llun_info *lli) | |
289 | { | |
290 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
291 | struct device *dev = &cfg->dev->dev; | |
292 | struct glun_info *gli = lli->parent; | |
293 | u8 *cmd_buf = NULL; | |
294 | u8 *scsi_cmd = NULL; | |
295 | u8 *sense_buf = NULL; | |
296 | int rc = 0; | |
297 | int result = 0; | |
298 | int retry_cnt = 0; | |
299 | u32 tout = (MC_DISCOVERY_TIMEOUT * HZ); | |
300 | ||
301 | retry: | |
302 | cmd_buf = kzalloc(CMD_BUFSIZE, GFP_KERNEL); | |
303 | scsi_cmd = kzalloc(MAX_COMMAND_SIZE, GFP_KERNEL); | |
304 | sense_buf = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL); | |
305 | if (unlikely(!cmd_buf || !scsi_cmd || !sense_buf)) { | |
306 | rc = -ENOMEM; | |
307 | goto out; | |
308 | } | |
309 | ||
310 | scsi_cmd[0] = SERVICE_ACTION_IN_16; /* read cap(16) */ | |
311 | scsi_cmd[1] = SAI_READ_CAPACITY_16; /* service action */ | |
312 | put_unaligned_be32(CMD_BUFSIZE, &scsi_cmd[10]); | |
313 | ||
314 | dev_dbg(dev, "%s: %ssending cmd(0x%x)\n", __func__, | |
315 | retry_cnt ? "re" : "", scsi_cmd[0]); | |
316 | ||
317 | result = scsi_execute(sdev, scsi_cmd, DMA_FROM_DEVICE, cmd_buf, | |
3ebf2030 MK |
318 | CMD_BUFSIZE, sense_buf, tout, CMD_RETRIES, |
319 | 0, NULL); | |
65be2c79 MO |
320 | |
321 | if (driver_byte(result) == DRIVER_SENSE) { | |
322 | result &= ~(0xFF<<24); /* DRIVER_SENSE is not an error */ | |
323 | if (result & SAM_STAT_CHECK_CONDITION) { | |
324 | struct scsi_sense_hdr sshdr; | |
325 | ||
326 | scsi_normalize_sense(sense_buf, SCSI_SENSE_BUFFERSIZE, | |
327 | &sshdr); | |
328 | switch (sshdr.sense_key) { | |
329 | case NO_SENSE: | |
330 | case RECOVERED_ERROR: | |
331 | /* fall through */ | |
332 | case NOT_READY: | |
333 | result &= ~SAM_STAT_CHECK_CONDITION; | |
334 | break; | |
335 | case UNIT_ATTENTION: | |
336 | switch (sshdr.asc) { | |
337 | case 0x29: /* Power on Reset or Device Reset */ | |
338 | /* fall through */ | |
339 | case 0x2A: /* Device capacity changed */ | |
340 | case 0x3F: /* Report LUNs changed */ | |
341 | /* Retry the command once more */ | |
342 | if (retry_cnt++ < 1) { | |
343 | kfree(cmd_buf); | |
344 | kfree(scsi_cmd); | |
345 | kfree(sense_buf); | |
346 | goto retry; | |
347 | } | |
348 | } | |
349 | break; | |
350 | default: | |
351 | break; | |
352 | } | |
353 | } | |
354 | } | |
355 | ||
356 | if (result) { | |
357 | dev_err(dev, "%s: command failed, result=0x%x\n", | |
358 | __func__, result); | |
359 | rc = -EIO; | |
360 | goto out; | |
361 | } | |
362 | ||
363 | /* | |
364 | * Read cap was successful, grab values from the buffer; | |
365 | * note that we don't need to worry about unaligned access | |
366 | * as the buffer is allocated on an aligned boundary. | |
367 | */ | |
368 | mutex_lock(&gli->mutex); | |
369 | gli->max_lba = be64_to_cpu(*((u64 *)&cmd_buf[0])); | |
370 | gli->blk_len = be32_to_cpu(*((u32 *)&cmd_buf[8])); | |
371 | mutex_unlock(&gli->mutex); | |
372 | ||
373 | out: | |
374 | kfree(cmd_buf); | |
375 | kfree(scsi_cmd); | |
376 | kfree(sense_buf); | |
377 | ||
378 | dev_dbg(dev, "%s: maxlba=%lld blklen=%d rc=%d\n", | |
379 | __func__, gli->max_lba, gli->blk_len, rc); | |
380 | return rc; | |
381 | } | |
382 | ||
383 | /** | |
384 | * get_rhte() - obtains validated resource handle table entry reference | |
385 | * @ctxi: Context owning the resource handle. | |
386 | * @rhndl: Resource handle associated with entry. | |
387 | * @lli: LUN associated with request. | |
388 | * | |
389 | * Return: Validated RHTE on success, NULL on failure | |
390 | */ | |
391 | struct sisl_rht_entry *get_rhte(struct ctx_info *ctxi, res_hndl_t rhndl, | |
392 | struct llun_info *lli) | |
393 | { | |
394 | struct sisl_rht_entry *rhte = NULL; | |
395 | ||
396 | if (unlikely(!ctxi->rht_start)) { | |
397 | pr_debug("%s: Context does not have allocated RHT!\n", | |
398 | __func__); | |
399 | goto out; | |
400 | } | |
401 | ||
402 | if (unlikely(rhndl >= MAX_RHT_PER_CONTEXT)) { | |
403 | pr_debug("%s: Bad resource handle! (%d)\n", __func__, rhndl); | |
404 | goto out; | |
405 | } | |
406 | ||
407 | if (unlikely(ctxi->rht_lun[rhndl] != lli)) { | |
408 | pr_debug("%s: Bad resource handle LUN! (%d)\n", | |
409 | __func__, rhndl); | |
410 | goto out; | |
411 | } | |
412 | ||
413 | rhte = &ctxi->rht_start[rhndl]; | |
414 | if (unlikely(rhte->nmask == 0)) { | |
415 | pr_debug("%s: Unopened resource handle! (%d)\n", | |
416 | __func__, rhndl); | |
417 | rhte = NULL; | |
418 | goto out; | |
419 | } | |
420 | ||
421 | out: | |
422 | return rhte; | |
423 | } | |
424 | ||
425 | /** | |
426 | * rhte_checkout() - obtains free/empty resource handle table entry | |
427 | * @ctxi: Context owning the resource handle. | |
428 | * @lli: LUN associated with request. | |
429 | * | |
430 | * Return: Free RHTE on success, NULL on failure | |
431 | */ | |
432 | struct sisl_rht_entry *rhte_checkout(struct ctx_info *ctxi, | |
433 | struct llun_info *lli) | |
434 | { | |
435 | struct sisl_rht_entry *rhte = NULL; | |
436 | int i; | |
437 | ||
438 | /* Find a free RHT entry */ | |
439 | for (i = 0; i < MAX_RHT_PER_CONTEXT; i++) | |
440 | if (ctxi->rht_start[i].nmask == 0) { | |
441 | rhte = &ctxi->rht_start[i]; | |
442 | ctxi->rht_out++; | |
443 | break; | |
444 | } | |
445 | ||
446 | if (likely(rhte)) | |
447 | ctxi->rht_lun[i] = lli; | |
448 | ||
449 | pr_debug("%s: returning rhte=%p (%d)\n", __func__, rhte, i); | |
450 | return rhte; | |
451 | } | |
452 | ||
453 | /** | |
454 | * rhte_checkin() - releases a resource handle table entry | |
455 | * @ctxi: Context owning the resource handle. | |
456 | * @rhte: RHTE to release. | |
457 | */ | |
458 | void rhte_checkin(struct ctx_info *ctxi, | |
459 | struct sisl_rht_entry *rhte) | |
460 | { | |
461 | u32 rsrc_handle = rhte - ctxi->rht_start; | |
462 | ||
463 | rhte->nmask = 0; | |
464 | rhte->fp = 0; | |
465 | ctxi->rht_out--; | |
466 | ctxi->rht_lun[rsrc_handle] = NULL; | |
2cb79266 | 467 | ctxi->rht_needs_ws[rsrc_handle] = false; |
65be2c79 MO |
468 | } |
469 | ||
470 | /** | |
471 | * rhte_format1() - populates a RHTE for format 1 | |
472 | * @rhte: RHTE to populate. | |
473 | * @lun_id: LUN ID of LUN associated with RHTE. | |
474 | * @perm: Desired permissions for RHTE. | |
475 | * @port_sel: Port selection mask | |
476 | */ | |
477 | static void rht_format1(struct sisl_rht_entry *rhte, u64 lun_id, u32 perm, | |
478 | u32 port_sel) | |
479 | { | |
480 | /* | |
481 | * Populate the Format 1 RHT entry for direct access (physical | |
482 | * LUN) using the synchronization sequence defined in the | |
483 | * SISLite specification. | |
484 | */ | |
485 | struct sisl_rht_entry_f1 dummy = { 0 }; | |
486 | struct sisl_rht_entry_f1 *rhte_f1 = (struct sisl_rht_entry_f1 *)rhte; | |
487 | ||
488 | memset(rhte_f1, 0, sizeof(*rhte_f1)); | |
489 | rhte_f1->fp = SISL_RHT_FP(1U, 0); | |
490 | dma_wmb(); /* Make setting of format bit visible */ | |
491 | ||
492 | rhte_f1->lun_id = lun_id; | |
493 | dma_wmb(); /* Make setting of LUN id visible */ | |
494 | ||
495 | /* | |
496 | * Use a dummy RHT Format 1 entry to build the second dword | |
497 | * of the entry that must be populated in a single write when | |
498 | * enabled (valid bit set to TRUE). | |
499 | */ | |
500 | dummy.valid = 0x80; | |
501 | dummy.fp = SISL_RHT_FP(1U, perm); | |
502 | dummy.port_sel = port_sel; | |
503 | rhte_f1->dw = dummy.dw; | |
504 | ||
505 | dma_wmb(); /* Make remaining RHT entry fields visible */ | |
506 | } | |
507 | ||
508 | /** | |
509 | * cxlflash_lun_attach() - attaches a user to a LUN and manages the LUN's mode | |
510 | * @gli: LUN to attach. | |
511 | * @mode: Desired mode of the LUN. | |
512 | * @locked: Mutex status on current thread. | |
513 | * | |
514 | * Return: 0 on success, -errno on failure | |
515 | */ | |
516 | int cxlflash_lun_attach(struct glun_info *gli, enum lun_mode mode, bool locked) | |
517 | { | |
518 | int rc = 0; | |
519 | ||
520 | if (!locked) | |
521 | mutex_lock(&gli->mutex); | |
522 | ||
523 | if (gli->mode == MODE_NONE) | |
524 | gli->mode = mode; | |
525 | else if (gli->mode != mode) { | |
526 | pr_debug("%s: LUN operating in mode %d, requested mode %d\n", | |
527 | __func__, gli->mode, mode); | |
528 | rc = -EINVAL; | |
529 | goto out; | |
530 | } | |
531 | ||
532 | gli->users++; | |
533 | WARN_ON(gli->users <= 0); | |
534 | out: | |
535 | pr_debug("%s: Returning rc=%d gli->mode=%u gli->users=%u\n", | |
536 | __func__, rc, gli->mode, gli->users); | |
537 | if (!locked) | |
538 | mutex_unlock(&gli->mutex); | |
539 | return rc; | |
540 | } | |
541 | ||
542 | /** | |
543 | * cxlflash_lun_detach() - detaches a user from a LUN and resets the LUN's mode | |
544 | * @gli: LUN to detach. | |
2cb79266 MO |
545 | * |
546 | * When resetting the mode, terminate block allocation resources as they | |
547 | * are no longer required (service is safe to call even when block allocation | |
548 | * resources were not present - such as when transitioning from physical mode). | |
549 | * These resources will be reallocated when needed (subsequent transition to | |
550 | * virtual mode). | |
65be2c79 MO |
551 | */ |
552 | void cxlflash_lun_detach(struct glun_info *gli) | |
553 | { | |
554 | mutex_lock(&gli->mutex); | |
555 | WARN_ON(gli->mode == MODE_NONE); | |
2cb79266 | 556 | if (--gli->users == 0) { |
65be2c79 | 557 | gli->mode = MODE_NONE; |
2cb79266 MO |
558 | cxlflash_ba_terminate(&gli->blka.ba_lun); |
559 | } | |
65be2c79 MO |
560 | pr_debug("%s: gli->users=%u\n", __func__, gli->users); |
561 | WARN_ON(gli->users < 0); | |
562 | mutex_unlock(&gli->mutex); | |
563 | } | |
564 | ||
565 | /** | |
566 | * _cxlflash_disk_release() - releases the specified resource entry | |
567 | * @sdev: SCSI device associated with LUN. | |
568 | * @ctxi: Context owning resources. | |
569 | * @release: Release ioctl data structure. | |
570 | * | |
2cb79266 MO |
571 | * For LUNs in virtual mode, the virtual LUN associated with the specified |
572 | * resource handle is resized to 0 prior to releasing the RHTE. Note that the | |
573 | * AFU sync should _not_ be performed when the context is sitting on the error | |
574 | * recovery list. A context on the error recovery list is not known to the AFU | |
575 | * due to reset. When the context is recovered, it will be reattached and made | |
576 | * known again to the AFU. | |
65be2c79 MO |
577 | * |
578 | * Return: 0 on success, -errno on failure | |
579 | */ | |
580 | int _cxlflash_disk_release(struct scsi_device *sdev, | |
581 | struct ctx_info *ctxi, | |
582 | struct dk_cxlflash_release *release) | |
583 | { | |
584 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
585 | struct device *dev = &cfg->dev->dev; | |
586 | struct llun_info *lli = sdev->hostdata; | |
587 | struct glun_info *gli = lli->parent; | |
588 | struct afu *afu = cfg->afu; | |
589 | bool put_ctx = false; | |
590 | ||
2cb79266 | 591 | struct dk_cxlflash_resize size; |
65be2c79 MO |
592 | res_hndl_t rhndl = release->rsrc_handle; |
593 | ||
594 | int rc = 0; | |
595 | u64 ctxid = DECODE_CTXID(release->context_id), | |
596 | rctxid = release->context_id; | |
597 | ||
598 | struct sisl_rht_entry *rhte; | |
599 | struct sisl_rht_entry_f1 *rhte_f1; | |
600 | ||
601 | dev_dbg(dev, "%s: ctxid=%llu rhndl=0x%llx gli->mode=%u gli->users=%u\n", | |
602 | __func__, ctxid, release->rsrc_handle, gli->mode, gli->users); | |
603 | ||
604 | if (!ctxi) { | |
605 | ctxi = get_context(cfg, rctxid, lli, CTX_CTRL_ERR_FALLBACK); | |
606 | if (unlikely(!ctxi)) { | |
607 | dev_dbg(dev, "%s: Bad context! (%llu)\n", | |
608 | __func__, ctxid); | |
609 | rc = -EINVAL; | |
610 | goto out; | |
611 | } | |
612 | ||
613 | put_ctx = true; | |
614 | } | |
615 | ||
616 | rhte = get_rhte(ctxi, rhndl, lli); | |
617 | if (unlikely(!rhte)) { | |
618 | dev_dbg(dev, "%s: Bad resource handle! (%d)\n", | |
619 | __func__, rhndl); | |
620 | rc = -EINVAL; | |
621 | goto out; | |
622 | } | |
623 | ||
2cb79266 MO |
624 | /* |
625 | * Resize to 0 for virtual LUNS by setting the size | |
626 | * to 0. This will clear LXT_START and LXT_CNT fields | |
627 | * in the RHT entry and properly sync with the AFU. | |
628 | * | |
629 | * Afterwards we clear the remaining fields. | |
630 | */ | |
65be2c79 | 631 | switch (gli->mode) { |
2cb79266 MO |
632 | case MODE_VIRTUAL: |
633 | marshal_rele_to_resize(release, &size); | |
634 | size.req_size = 0; | |
635 | rc = _cxlflash_vlun_resize(sdev, ctxi, &size); | |
636 | if (rc) { | |
637 | dev_dbg(dev, "%s: resize failed rc %d\n", __func__, rc); | |
638 | goto out; | |
639 | } | |
640 | ||
641 | break; | |
65be2c79 MO |
642 | case MODE_PHYSICAL: |
643 | /* | |
644 | * Clear the Format 1 RHT entry for direct access | |
645 | * (physical LUN) using the synchronization sequence | |
646 | * defined in the SISLite specification. | |
647 | */ | |
648 | rhte_f1 = (struct sisl_rht_entry_f1 *)rhte; | |
649 | ||
650 | rhte_f1->valid = 0; | |
651 | dma_wmb(); /* Make revocation of RHT entry visible */ | |
652 | ||
653 | rhte_f1->lun_id = 0; | |
654 | dma_wmb(); /* Make clearing of LUN id visible */ | |
655 | ||
656 | rhte_f1->dw = 0; | |
657 | dma_wmb(); /* Make RHT entry bottom-half clearing visible */ | |
658 | ||
659 | if (!ctxi->err_recovery_active) | |
660 | cxlflash_afu_sync(afu, ctxid, rhndl, AFU_HW_SYNC); | |
661 | break; | |
662 | default: | |
663 | WARN(1, "Unsupported LUN mode!"); | |
664 | goto out; | |
665 | } | |
666 | ||
667 | rhte_checkin(ctxi, rhte); | |
668 | cxlflash_lun_detach(gli); | |
669 | ||
670 | out: | |
671 | if (put_ctx) | |
672 | put_context(ctxi); | |
673 | dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc); | |
674 | return rc; | |
675 | } | |
676 | ||
677 | int cxlflash_disk_release(struct scsi_device *sdev, | |
678 | struct dk_cxlflash_release *release) | |
679 | { | |
680 | return _cxlflash_disk_release(sdev, NULL, release); | |
681 | } | |
682 | ||
683 | /** | |
684 | * destroy_context() - releases a context | |
685 | * @cfg: Internal structure associated with the host. | |
686 | * @ctxi: Context to release. | |
687 | * | |
688 | * Note that the rht_lun member of the context was cut from a single | |
689 | * allocation when the context was created and therefore does not need | |
690 | * to be explicitly freed. Also note that we conditionally check for the | |
691 | * existence of the context control map before clearing the RHT registers | |
692 | * and context capabilities because it is possible to destroy a context | |
693 | * while the context is in the error state (previous mapping was removed | |
694 | * [so we don't have to worry about clearing] and context is waiting for | |
695 | * a new mapping). | |
696 | */ | |
697 | static void destroy_context(struct cxlflash_cfg *cfg, | |
698 | struct ctx_info *ctxi) | |
699 | { | |
700 | struct afu *afu = cfg->afu; | |
701 | ||
702 | WARN_ON(!list_empty(&ctxi->luns)); | |
703 | ||
704 | /* Clear RHT registers and drop all capabilities for this context */ | |
705 | if (afu->afu_map && ctxi->ctrl_map) { | |
706 | writeq_be(0, &ctxi->ctrl_map->rht_start); | |
707 | writeq_be(0, &ctxi->ctrl_map->rht_cnt_id); | |
708 | writeq_be(0, &ctxi->ctrl_map->ctx_cap); | |
709 | } | |
710 | ||
711 | /* Free memory associated with context */ | |
712 | free_page((ulong)ctxi->rht_start); | |
2cb79266 | 713 | kfree(ctxi->rht_needs_ws); |
65be2c79 MO |
714 | kfree(ctxi->rht_lun); |
715 | kfree(ctxi); | |
716 | atomic_dec_if_positive(&cfg->num_user_contexts); | |
717 | } | |
718 | ||
719 | /** | |
720 | * create_context() - allocates and initializes a context | |
721 | * @cfg: Internal structure associated with the host. | |
722 | * @ctx: Previously obtained CXL context reference. | |
723 | * @ctxid: Previously obtained process element associated with CXL context. | |
724 | * @adap_fd: Previously obtained adapter fd associated with CXL context. | |
725 | * @file: Previously obtained file associated with CXL context. | |
726 | * @perms: User-specified permissions. | |
727 | * | |
728 | * The context's mutex is locked when an allocated context is returned. | |
729 | * | |
730 | * Return: Allocated context on success, NULL on failure | |
731 | */ | |
732 | static struct ctx_info *create_context(struct cxlflash_cfg *cfg, | |
733 | struct cxl_context *ctx, int ctxid, | |
734 | int adap_fd, struct file *file, | |
735 | u32 perms) | |
736 | { | |
737 | struct device *dev = &cfg->dev->dev; | |
738 | struct afu *afu = cfg->afu; | |
739 | struct ctx_info *ctxi = NULL; | |
740 | struct llun_info **lli = NULL; | |
2cb79266 | 741 | bool *ws = NULL; |
65be2c79 MO |
742 | struct sisl_rht_entry *rhte; |
743 | ||
744 | ctxi = kzalloc(sizeof(*ctxi), GFP_KERNEL); | |
745 | lli = kzalloc((MAX_RHT_PER_CONTEXT * sizeof(*lli)), GFP_KERNEL); | |
2cb79266 MO |
746 | ws = kzalloc((MAX_RHT_PER_CONTEXT * sizeof(*ws)), GFP_KERNEL); |
747 | if (unlikely(!ctxi || !lli || !ws)) { | |
65be2c79 MO |
748 | dev_err(dev, "%s: Unable to allocate context!\n", __func__); |
749 | goto err; | |
750 | } | |
751 | ||
752 | rhte = (struct sisl_rht_entry *)get_zeroed_page(GFP_KERNEL); | |
753 | if (unlikely(!rhte)) { | |
754 | dev_err(dev, "%s: Unable to allocate RHT!\n", __func__); | |
755 | goto err; | |
756 | } | |
757 | ||
758 | ctxi->rht_lun = lli; | |
2cb79266 | 759 | ctxi->rht_needs_ws = ws; |
65be2c79 MO |
760 | ctxi->rht_start = rhte; |
761 | ctxi->rht_perms = perms; | |
762 | ||
763 | ctxi->ctrl_map = &afu->afu_map->ctrls[ctxid].ctrl; | |
764 | ctxi->ctxid = ENCODE_CTXID(ctxi, ctxid); | |
765 | ctxi->lfd = adap_fd; | |
766 | ctxi->pid = current->tgid; /* tgid = pid */ | |
767 | ctxi->ctx = ctx; | |
768 | ctxi->file = file; | |
769 | mutex_init(&ctxi->mutex); | |
770 | INIT_LIST_HEAD(&ctxi->luns); | |
771 | INIT_LIST_HEAD(&ctxi->list); /* initialize for list_empty() */ | |
772 | ||
773 | atomic_inc(&cfg->num_user_contexts); | |
774 | mutex_lock(&ctxi->mutex); | |
775 | out: | |
776 | return ctxi; | |
777 | ||
778 | err: | |
2cb79266 | 779 | kfree(ws); |
65be2c79 MO |
780 | kfree(lli); |
781 | kfree(ctxi); | |
782 | ctxi = NULL; | |
783 | goto out; | |
784 | } | |
785 | ||
786 | /** | |
787 | * _cxlflash_disk_detach() - detaches a LUN from a context | |
788 | * @sdev: SCSI device associated with LUN. | |
789 | * @ctxi: Context owning resources. | |
790 | * @detach: Detach ioctl data structure. | |
791 | * | |
792 | * As part of the detach, all per-context resources associated with the LUN | |
793 | * are cleaned up. When detaching the last LUN for a context, the context | |
794 | * itself is cleaned up and released. | |
795 | * | |
796 | * Return: 0 on success, -errno on failure | |
797 | */ | |
798 | static int _cxlflash_disk_detach(struct scsi_device *sdev, | |
799 | struct ctx_info *ctxi, | |
800 | struct dk_cxlflash_detach *detach) | |
801 | { | |
802 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
803 | struct device *dev = &cfg->dev->dev; | |
804 | struct llun_info *lli = sdev->hostdata; | |
805 | struct lun_access *lun_access, *t; | |
806 | struct dk_cxlflash_release rel; | |
807 | bool put_ctx = false; | |
808 | ||
809 | int i; | |
810 | int rc = 0; | |
811 | int lfd; | |
812 | u64 ctxid = DECODE_CTXID(detach->context_id), | |
813 | rctxid = detach->context_id; | |
814 | ||
815 | dev_dbg(dev, "%s: ctxid=%llu\n", __func__, ctxid); | |
816 | ||
817 | if (!ctxi) { | |
818 | ctxi = get_context(cfg, rctxid, lli, CTX_CTRL_ERR_FALLBACK); | |
819 | if (unlikely(!ctxi)) { | |
820 | dev_dbg(dev, "%s: Bad context! (%llu)\n", | |
821 | __func__, ctxid); | |
822 | rc = -EINVAL; | |
823 | goto out; | |
824 | } | |
825 | ||
826 | put_ctx = true; | |
827 | } | |
828 | ||
829 | /* Cleanup outstanding resources tied to this LUN */ | |
830 | if (ctxi->rht_out) { | |
831 | marshal_det_to_rele(detach, &rel); | |
832 | for (i = 0; i < MAX_RHT_PER_CONTEXT; i++) { | |
833 | if (ctxi->rht_lun[i] == lli) { | |
834 | rel.rsrc_handle = i; | |
835 | _cxlflash_disk_release(sdev, ctxi, &rel); | |
836 | } | |
837 | ||
838 | /* No need to loop further if we're done */ | |
839 | if (ctxi->rht_out == 0) | |
840 | break; | |
841 | } | |
842 | } | |
843 | ||
844 | /* Take our LUN out of context, free the node */ | |
845 | list_for_each_entry_safe(lun_access, t, &ctxi->luns, list) | |
846 | if (lun_access->lli == lli) { | |
847 | list_del(&lun_access->list); | |
848 | kfree(lun_access); | |
849 | lun_access = NULL; | |
850 | break; | |
851 | } | |
852 | ||
853 | /* Tear down context following last LUN cleanup */ | |
854 | if (list_empty(&ctxi->luns)) { | |
855 | ctxi->unavail = true; | |
856 | mutex_unlock(&ctxi->mutex); | |
857 | mutex_lock(&cfg->ctx_tbl_list_mutex); | |
858 | mutex_lock(&ctxi->mutex); | |
859 | ||
860 | /* Might not have been in error list so conditionally remove */ | |
861 | if (!list_empty(&ctxi->list)) | |
862 | list_del(&ctxi->list); | |
863 | cfg->ctx_tbl[ctxid] = NULL; | |
864 | mutex_unlock(&cfg->ctx_tbl_list_mutex); | |
865 | mutex_unlock(&ctxi->mutex); | |
866 | ||
867 | lfd = ctxi->lfd; | |
868 | destroy_context(cfg, ctxi); | |
869 | ctxi = NULL; | |
870 | put_ctx = false; | |
871 | ||
872 | /* | |
873 | * As a last step, clean up external resources when not | |
874 | * already on an external cleanup thread, i.e.: close(adap_fd). | |
875 | * | |
876 | * NOTE: this will free up the context from the CXL services, | |
877 | * allowing it to dole out the same context_id on a future | |
878 | * (or even currently in-flight) disk_attach operation. | |
879 | */ | |
880 | if (lfd != -1) | |
881 | sys_close(lfd); | |
882 | } | |
883 | ||
884 | out: | |
885 | if (put_ctx) | |
886 | put_context(ctxi); | |
887 | dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc); | |
888 | return rc; | |
889 | } | |
890 | ||
891 | static int cxlflash_disk_detach(struct scsi_device *sdev, | |
892 | struct dk_cxlflash_detach *detach) | |
893 | { | |
894 | return _cxlflash_disk_detach(sdev, NULL, detach); | |
895 | } | |
896 | ||
897 | /** | |
898 | * cxlflash_cxl_release() - release handler for adapter file descriptor | |
899 | * @inode: File-system inode associated with fd. | |
900 | * @file: File installed with adapter file descriptor. | |
901 | * | |
902 | * This routine is the release handler for the fops registered with | |
903 | * the CXL services on an initial attach for a context. It is called | |
904 | * when a close is performed on the adapter file descriptor returned | |
905 | * to the user. Programmatically, the user is not required to perform | |
906 | * the close, as it is handled internally via the detach ioctl when | |
907 | * a context is being removed. Note that nothing prevents the user | |
908 | * from performing a close, but the user should be aware that doing | |
909 | * so is considered catastrophic and subsequent usage of the superpipe | |
910 | * API with previously saved off tokens will fail. | |
911 | * | |
912 | * When initiated from an external close (either by the user or via | |
913 | * a process tear down), the routine derives the context reference | |
914 | * and calls detach for each LUN associated with the context. The | |
915 | * final detach operation will cause the context itself to be freed. | |
916 | * Note that the saved off lfd is reset prior to calling detach to | |
917 | * signify that the final detach should not perform a close. | |
918 | * | |
919 | * When initiated from a detach operation as part of the tear down | |
920 | * of a context, the context is first completely freed and then the | |
921 | * close is performed. This routine will fail to derive the context | |
922 | * reference (due to the context having already been freed) and then | |
923 | * call into the CXL release entry point. | |
924 | * | |
925 | * Thus, with exception to when the CXL process element (context id) | |
926 | * lookup fails (a case that should theoretically never occur), every | |
927 | * call into this routine results in a complete freeing of a context. | |
928 | * | |
929 | * As part of the detach, all per-context resources associated with the LUN | |
930 | * are cleaned up. When detaching the last LUN for a context, the context | |
931 | * itself is cleaned up and released. | |
932 | * | |
933 | * Return: 0 on success | |
934 | */ | |
935 | static int cxlflash_cxl_release(struct inode *inode, struct file *file) | |
936 | { | |
937 | struct cxl_context *ctx = cxl_fops_get_context(file); | |
938 | struct cxlflash_cfg *cfg = container_of(file->f_op, struct cxlflash_cfg, | |
939 | cxl_fops); | |
940 | struct device *dev = &cfg->dev->dev; | |
941 | struct ctx_info *ctxi = NULL; | |
942 | struct dk_cxlflash_detach detach = { { 0 }, 0 }; | |
943 | struct lun_access *lun_access, *t; | |
944 | enum ctx_ctrl ctrl = CTX_CTRL_ERR_FALLBACK | CTX_CTRL_FILE; | |
945 | int ctxid; | |
946 | ||
947 | ctxid = cxl_process_element(ctx); | |
948 | if (unlikely(ctxid < 0)) { | |
949 | dev_err(dev, "%s: Context %p was closed! (%d)\n", | |
950 | __func__, ctx, ctxid); | |
951 | goto out; | |
952 | } | |
953 | ||
954 | ctxi = get_context(cfg, ctxid, file, ctrl); | |
955 | if (unlikely(!ctxi)) { | |
956 | ctxi = get_context(cfg, ctxid, file, ctrl | CTX_CTRL_CLONE); | |
957 | if (!ctxi) { | |
958 | dev_dbg(dev, "%s: Context %d already free!\n", | |
959 | __func__, ctxid); | |
960 | goto out_release; | |
961 | } | |
962 | ||
963 | dev_dbg(dev, "%s: Another process owns context %d!\n", | |
964 | __func__, ctxid); | |
965 | put_context(ctxi); | |
966 | goto out; | |
967 | } | |
968 | ||
969 | dev_dbg(dev, "%s: close(%d) for context %d\n", | |
970 | __func__, ctxi->lfd, ctxid); | |
971 | ||
972 | /* Reset the file descriptor to indicate we're on a close() thread */ | |
973 | ctxi->lfd = -1; | |
974 | detach.context_id = ctxi->ctxid; | |
975 | list_for_each_entry_safe(lun_access, t, &ctxi->luns, list) | |
976 | _cxlflash_disk_detach(lun_access->sdev, ctxi, &detach); | |
977 | out_release: | |
978 | cxl_fd_release(inode, file); | |
979 | out: | |
980 | dev_dbg(dev, "%s: returning\n", __func__); | |
981 | return 0; | |
982 | } | |
983 | ||
984 | /** | |
985 | * unmap_context() - clears a previously established mapping | |
986 | * @ctxi: Context owning the mapping. | |
987 | * | |
988 | * This routine is used to switch between the error notification page | |
989 | * (dummy page of all 1's) and the real mapping (established by the CXL | |
990 | * fault handler). | |
991 | */ | |
992 | static void unmap_context(struct ctx_info *ctxi) | |
993 | { | |
994 | unmap_mapping_range(ctxi->file->f_mapping, 0, 0, 1); | |
995 | } | |
996 | ||
997 | /** | |
998 | * get_err_page() - obtains and allocates the error notification page | |
999 | * | |
1000 | * Return: error notification page on success, NULL on failure | |
1001 | */ | |
1002 | static struct page *get_err_page(void) | |
1003 | { | |
1004 | struct page *err_page = global.err_page; | |
1005 | ||
1006 | if (unlikely(!err_page)) { | |
1007 | err_page = alloc_page(GFP_KERNEL); | |
1008 | if (unlikely(!err_page)) { | |
1009 | pr_err("%s: Unable to allocate err_page!\n", __func__); | |
1010 | goto out; | |
1011 | } | |
1012 | ||
1013 | memset(page_address(err_page), -1, PAGE_SIZE); | |
1014 | ||
1015 | /* Serialize update w/ other threads to avoid a leak */ | |
1016 | mutex_lock(&global.mutex); | |
1017 | if (likely(!global.err_page)) | |
1018 | global.err_page = err_page; | |
1019 | else { | |
1020 | __free_page(err_page); | |
1021 | err_page = global.err_page; | |
1022 | } | |
1023 | mutex_unlock(&global.mutex); | |
1024 | } | |
1025 | ||
1026 | out: | |
1027 | pr_debug("%s: returning err_page=%p\n", __func__, err_page); | |
1028 | return err_page; | |
1029 | } | |
1030 | ||
1031 | /** | |
1032 | * cxlflash_mmap_fault() - mmap fault handler for adapter file descriptor | |
1033 | * @vma: VM area associated with mapping. | |
1034 | * @vmf: VM fault associated with current fault. | |
1035 | * | |
1036 | * To support error notification via MMIO, faults are 'caught' by this routine | |
1037 | * that was inserted before passing back the adapter file descriptor on attach. | |
1038 | * When a fault occurs, this routine evaluates if error recovery is active and | |
1039 | * if so, installs the error page to 'notify' the user about the error state. | |
1040 | * During normal operation, the fault is simply handled by the original fault | |
1041 | * handler that was installed by CXL services as part of initializing the | |
1042 | * adapter file descriptor. The VMA's page protection bits are toggled to | |
1043 | * indicate cached/not-cached depending on the memory backing the fault. | |
1044 | * | |
1045 | * Return: 0 on success, VM_FAULT_SIGBUS on failure | |
1046 | */ | |
1047 | static int cxlflash_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) | |
1048 | { | |
1049 | struct file *file = vma->vm_file; | |
1050 | struct cxl_context *ctx = cxl_fops_get_context(file); | |
1051 | struct cxlflash_cfg *cfg = container_of(file->f_op, struct cxlflash_cfg, | |
1052 | cxl_fops); | |
1053 | struct device *dev = &cfg->dev->dev; | |
1054 | struct ctx_info *ctxi = NULL; | |
1055 | struct page *err_page = NULL; | |
1056 | enum ctx_ctrl ctrl = CTX_CTRL_ERR_FALLBACK | CTX_CTRL_FILE; | |
1057 | int rc = 0; | |
1058 | int ctxid; | |
1059 | ||
1060 | ctxid = cxl_process_element(ctx); | |
1061 | if (unlikely(ctxid < 0)) { | |
1062 | dev_err(dev, "%s: Context %p was closed! (%d)\n", | |
1063 | __func__, ctx, ctxid); | |
1064 | goto err; | |
1065 | } | |
1066 | ||
1067 | ctxi = get_context(cfg, ctxid, file, ctrl); | |
1068 | if (unlikely(!ctxi)) { | |
1069 | dev_dbg(dev, "%s: Bad context! (%d)\n", __func__, ctxid); | |
1070 | goto err; | |
1071 | } | |
1072 | ||
1073 | dev_dbg(dev, "%s: fault(%d) for context %d\n", | |
1074 | __func__, ctxi->lfd, ctxid); | |
1075 | ||
1076 | if (likely(!ctxi->err_recovery_active)) { | |
1077 | vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot); | |
1078 | rc = ctxi->cxl_mmap_vmops->fault(vma, vmf); | |
1079 | } else { | |
1080 | dev_dbg(dev, "%s: err recovery active, use err_page!\n", | |
1081 | __func__); | |
1082 | ||
1083 | err_page = get_err_page(); | |
1084 | if (unlikely(!err_page)) { | |
1085 | dev_err(dev, "%s: Could not obtain error page!\n", | |
1086 | __func__); | |
1087 | rc = VM_FAULT_RETRY; | |
1088 | goto out; | |
1089 | } | |
1090 | ||
1091 | get_page(err_page); | |
1092 | vmf->page = err_page; | |
1093 | vma->vm_page_prot = pgprot_cached(vma->vm_page_prot); | |
1094 | } | |
1095 | ||
1096 | out: | |
1097 | if (likely(ctxi)) | |
1098 | put_context(ctxi); | |
1099 | dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc); | |
1100 | return rc; | |
1101 | ||
1102 | err: | |
1103 | rc = VM_FAULT_SIGBUS; | |
1104 | goto out; | |
1105 | } | |
1106 | ||
1107 | /* | |
1108 | * Local MMAP vmops to 'catch' faults | |
1109 | */ | |
1110 | static const struct vm_operations_struct cxlflash_mmap_vmops = { | |
1111 | .fault = cxlflash_mmap_fault, | |
1112 | }; | |
1113 | ||
1114 | /** | |
1115 | * cxlflash_cxl_mmap() - mmap handler for adapter file descriptor | |
1116 | * @file: File installed with adapter file descriptor. | |
1117 | * @vma: VM area associated with mapping. | |
1118 | * | |
1119 | * Installs local mmap vmops to 'catch' faults for error notification support. | |
1120 | * | |
1121 | * Return: 0 on success, -errno on failure | |
1122 | */ | |
1123 | static int cxlflash_cxl_mmap(struct file *file, struct vm_area_struct *vma) | |
1124 | { | |
1125 | struct cxl_context *ctx = cxl_fops_get_context(file); | |
1126 | struct cxlflash_cfg *cfg = container_of(file->f_op, struct cxlflash_cfg, | |
1127 | cxl_fops); | |
1128 | struct device *dev = &cfg->dev->dev; | |
1129 | struct ctx_info *ctxi = NULL; | |
1130 | enum ctx_ctrl ctrl = CTX_CTRL_ERR_FALLBACK | CTX_CTRL_FILE; | |
1131 | int ctxid; | |
1132 | int rc = 0; | |
1133 | ||
1134 | ctxid = cxl_process_element(ctx); | |
1135 | if (unlikely(ctxid < 0)) { | |
1136 | dev_err(dev, "%s: Context %p was closed! (%d)\n", | |
1137 | __func__, ctx, ctxid); | |
1138 | rc = -EIO; | |
1139 | goto out; | |
1140 | } | |
1141 | ||
1142 | ctxi = get_context(cfg, ctxid, file, ctrl); | |
1143 | if (unlikely(!ctxi)) { | |
1144 | dev_dbg(dev, "%s: Bad context! (%d)\n", __func__, ctxid); | |
1145 | rc = -EIO; | |
1146 | goto out; | |
1147 | } | |
1148 | ||
1149 | dev_dbg(dev, "%s: mmap(%d) for context %d\n", | |
1150 | __func__, ctxi->lfd, ctxid); | |
1151 | ||
1152 | rc = cxl_fd_mmap(file, vma); | |
1153 | if (likely(!rc)) { | |
1154 | /* Insert ourself in the mmap fault handler path */ | |
1155 | ctxi->cxl_mmap_vmops = vma->vm_ops; | |
1156 | vma->vm_ops = &cxlflash_mmap_vmops; | |
1157 | } | |
1158 | ||
1159 | out: | |
1160 | if (likely(ctxi)) | |
1161 | put_context(ctxi); | |
1162 | return rc; | |
1163 | } | |
1164 | ||
1165 | /* | |
1166 | * Local fops for adapter file descriptor | |
1167 | */ | |
1168 | static const struct file_operations cxlflash_cxl_fops = { | |
1169 | .owner = THIS_MODULE, | |
1170 | .mmap = cxlflash_cxl_mmap, | |
1171 | .release = cxlflash_cxl_release, | |
1172 | }; | |
1173 | ||
1174 | /** | |
1175 | * cxlflash_mark_contexts_error() - move contexts to error state and list | |
1176 | * @cfg: Internal structure associated with the host. | |
1177 | * | |
1178 | * A context is only moved over to the error list when there are no outstanding | |
1179 | * references to it. This ensures that a running operation has completed. | |
1180 | * | |
1181 | * Return: 0 on success, -errno on failure | |
1182 | */ | |
1183 | int cxlflash_mark_contexts_error(struct cxlflash_cfg *cfg) | |
1184 | { | |
1185 | int i, rc = 0; | |
1186 | struct ctx_info *ctxi = NULL; | |
1187 | ||
1188 | mutex_lock(&cfg->ctx_tbl_list_mutex); | |
1189 | ||
1190 | for (i = 0; i < MAX_CONTEXT; i++) { | |
1191 | ctxi = cfg->ctx_tbl[i]; | |
1192 | if (ctxi) { | |
1193 | mutex_lock(&ctxi->mutex); | |
1194 | cfg->ctx_tbl[i] = NULL; | |
1195 | list_add(&ctxi->list, &cfg->ctx_err_recovery); | |
1196 | ctxi->err_recovery_active = true; | |
1197 | ctxi->ctrl_map = NULL; | |
1198 | unmap_context(ctxi); | |
1199 | mutex_unlock(&ctxi->mutex); | |
1200 | } | |
1201 | } | |
1202 | ||
1203 | mutex_unlock(&cfg->ctx_tbl_list_mutex); | |
1204 | return rc; | |
1205 | } | |
1206 | ||
1207 | /* | |
1208 | * Dummy NULL fops | |
1209 | */ | |
1210 | static const struct file_operations null_fops = { | |
1211 | .owner = THIS_MODULE, | |
1212 | }; | |
1213 | ||
1214 | /** | |
1215 | * cxlflash_disk_attach() - attach a LUN to a context | |
1216 | * @sdev: SCSI device associated with LUN. | |
1217 | * @attach: Attach ioctl data structure. | |
1218 | * | |
1219 | * Creates a context and attaches LUN to it. A LUN can only be attached | |
1220 | * one time to a context (subsequent attaches for the same context/LUN pair | |
1221 | * are not supported). Additional LUNs can be attached to a context by | |
1222 | * specifying the 'reuse' flag defined in the cxlflash_ioctl.h header. | |
1223 | * | |
1224 | * Return: 0 on success, -errno on failure | |
1225 | */ | |
1226 | static int cxlflash_disk_attach(struct scsi_device *sdev, | |
1227 | struct dk_cxlflash_attach *attach) | |
1228 | { | |
1229 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
1230 | struct device *dev = &cfg->dev->dev; | |
1231 | struct afu *afu = cfg->afu; | |
1232 | struct llun_info *lli = sdev->hostdata; | |
1233 | struct glun_info *gli = lli->parent; | |
1234 | struct cxl_ioctl_start_work *work; | |
1235 | struct ctx_info *ctxi = NULL; | |
1236 | struct lun_access *lun_access = NULL; | |
1237 | int rc = 0; | |
1238 | u32 perms; | |
1239 | int ctxid = -1; | |
1240 | u64 rctxid = 0UL; | |
1241 | struct file *file; | |
1242 | ||
1243 | struct cxl_context *ctx; | |
1244 | ||
1245 | int fd = -1; | |
1246 | ||
1247 | /* On first attach set fileops */ | |
1248 | if (atomic_read(&cfg->num_user_contexts) == 0) | |
1249 | cfg->cxl_fops = cxlflash_cxl_fops; | |
1250 | ||
1251 | if (attach->num_interrupts > 4) { | |
1252 | dev_dbg(dev, "%s: Cannot support this many interrupts %llu\n", | |
1253 | __func__, attach->num_interrupts); | |
1254 | rc = -EINVAL; | |
1255 | goto out; | |
1256 | } | |
1257 | ||
1258 | if (gli->max_lba == 0) { | |
1259 | dev_dbg(dev, "%s: No capacity info for this LUN (%016llX)\n", | |
1260 | __func__, lli->lun_id[sdev->channel]); | |
1261 | rc = read_cap16(sdev, lli); | |
1262 | if (rc) { | |
1263 | dev_err(dev, "%s: Invalid device! (%d)\n", | |
1264 | __func__, rc); | |
1265 | rc = -ENODEV; | |
1266 | goto out; | |
1267 | } | |
1268 | dev_dbg(dev, "%s: LBA = %016llX\n", __func__, gli->max_lba); | |
1269 | dev_dbg(dev, "%s: BLK_LEN = %08X\n", __func__, gli->blk_len); | |
1270 | } | |
1271 | ||
1272 | if (attach->hdr.flags & DK_CXLFLASH_ATTACH_REUSE_CONTEXT) { | |
1273 | rctxid = attach->context_id; | |
1274 | ctxi = get_context(cfg, rctxid, NULL, 0); | |
1275 | if (!ctxi) { | |
1276 | dev_dbg(dev, "%s: Bad context! (%016llX)\n", | |
1277 | __func__, rctxid); | |
1278 | rc = -EINVAL; | |
1279 | goto out; | |
1280 | } | |
1281 | ||
1282 | list_for_each_entry(lun_access, &ctxi->luns, list) | |
1283 | if (lun_access->lli == lli) { | |
1284 | dev_dbg(dev, "%s: Already attached!\n", | |
1285 | __func__); | |
1286 | rc = -EINVAL; | |
1287 | goto out; | |
1288 | } | |
1289 | } | |
1290 | ||
1291 | lun_access = kzalloc(sizeof(*lun_access), GFP_KERNEL); | |
1292 | if (unlikely(!lun_access)) { | |
1293 | dev_err(dev, "%s: Unable to allocate lun_access!\n", __func__); | |
1294 | rc = -ENOMEM; | |
1295 | goto out; | |
1296 | } | |
1297 | ||
1298 | lun_access->lli = lli; | |
1299 | lun_access->sdev = sdev; | |
1300 | ||
1301 | /* Non-NULL context indicates reuse */ | |
1302 | if (ctxi) { | |
1303 | dev_dbg(dev, "%s: Reusing context for LUN! (%016llX)\n", | |
1304 | __func__, rctxid); | |
1305 | list_add(&lun_access->list, &ctxi->luns); | |
1306 | fd = ctxi->lfd; | |
1307 | goto out_attach; | |
1308 | } | |
1309 | ||
1310 | ctx = cxl_dev_context_init(cfg->dev); | |
1311 | if (unlikely(IS_ERR_OR_NULL(ctx))) { | |
1312 | dev_err(dev, "%s: Could not initialize context %p\n", | |
1313 | __func__, ctx); | |
1314 | rc = -ENODEV; | |
1315 | goto err0; | |
1316 | } | |
1317 | ||
1318 | ctxid = cxl_process_element(ctx); | |
1319 | if (unlikely((ctxid > MAX_CONTEXT) || (ctxid < 0))) { | |
1320 | dev_err(dev, "%s: ctxid (%d) invalid!\n", __func__, ctxid); | |
1321 | rc = -EPERM; | |
1322 | goto err1; | |
1323 | } | |
1324 | ||
1325 | file = cxl_get_fd(ctx, &cfg->cxl_fops, &fd); | |
1326 | if (unlikely(fd < 0)) { | |
1327 | rc = -ENODEV; | |
1328 | dev_err(dev, "%s: Could not get file descriptor\n", __func__); | |
1329 | goto err1; | |
1330 | } | |
1331 | ||
1332 | /* Translate read/write O_* flags from fcntl.h to AFU permission bits */ | |
1333 | perms = SISL_RHT_PERM(attach->hdr.flags + 1); | |
1334 | ||
1335 | ctxi = create_context(cfg, ctx, ctxid, fd, file, perms); | |
1336 | if (unlikely(!ctxi)) { | |
1337 | dev_err(dev, "%s: Failed to create context! (%d)\n", | |
1338 | __func__, ctxid); | |
1339 | goto err2; | |
1340 | } | |
1341 | ||
1342 | work = &ctxi->work; | |
1343 | work->num_interrupts = attach->num_interrupts; | |
1344 | work->flags = CXL_START_WORK_NUM_IRQS; | |
1345 | ||
1346 | rc = cxl_start_work(ctx, work); | |
1347 | if (unlikely(rc)) { | |
1348 | dev_dbg(dev, "%s: Could not start context rc=%d\n", | |
1349 | __func__, rc); | |
1350 | goto err3; | |
1351 | } | |
1352 | ||
1353 | rc = afu_attach(cfg, ctxi); | |
1354 | if (unlikely(rc)) { | |
1355 | dev_err(dev, "%s: Could not attach AFU rc %d\n", __func__, rc); | |
1356 | goto err4; | |
1357 | } | |
1358 | ||
1359 | /* | |
1360 | * No error paths after this point. Once the fd is installed it's | |
1361 | * visible to user space and can't be undone safely on this thread. | |
1362 | * There is no need to worry about a deadlock here because no one | |
1363 | * knows about us yet; we can be the only one holding our mutex. | |
1364 | */ | |
1365 | list_add(&lun_access->list, &ctxi->luns); | |
1366 | mutex_unlock(&ctxi->mutex); | |
1367 | mutex_lock(&cfg->ctx_tbl_list_mutex); | |
1368 | mutex_lock(&ctxi->mutex); | |
1369 | cfg->ctx_tbl[ctxid] = ctxi; | |
1370 | mutex_unlock(&cfg->ctx_tbl_list_mutex); | |
1371 | fd_install(fd, file); | |
1372 | ||
1373 | out_attach: | |
1374 | attach->hdr.return_flags = 0; | |
1375 | attach->context_id = ctxi->ctxid; | |
1376 | attach->block_size = gli->blk_len; | |
1377 | attach->mmio_size = sizeof(afu->afu_map->hosts[0].harea); | |
1378 | attach->last_lba = gli->max_lba; | |
3ebf2030 MK |
1379 | attach->max_xfer = (sdev->host->max_sectors * MAX_SECTOR_UNIT) / |
1380 | gli->blk_len; | |
65be2c79 MO |
1381 | |
1382 | out: | |
1383 | attach->adap_fd = fd; | |
1384 | ||
1385 | if (ctxi) | |
1386 | put_context(ctxi); | |
1387 | ||
1388 | dev_dbg(dev, "%s: returning ctxid=%d fd=%d bs=%lld rc=%d llba=%lld\n", | |
1389 | __func__, ctxid, fd, attach->block_size, rc, attach->last_lba); | |
1390 | return rc; | |
1391 | ||
1392 | err4: | |
1393 | cxl_stop_context(ctx); | |
1394 | err3: | |
1395 | put_context(ctxi); | |
1396 | destroy_context(cfg, ctxi); | |
1397 | ctxi = NULL; | |
1398 | err2: | |
1399 | /* | |
1400 | * Here, we're overriding the fops with a dummy all-NULL fops because | |
1401 | * fput() calls the release fop, which will cause us to mistakenly | |
1402 | * call into the CXL code. Rather than try to add yet more complexity | |
1403 | * to that routine (cxlflash_cxl_release) we should try to fix the | |
1404 | * issue here. | |
1405 | */ | |
1406 | file->f_op = &null_fops; | |
1407 | fput(file); | |
1408 | put_unused_fd(fd); | |
1409 | fd = -1; | |
1410 | err1: | |
1411 | cxl_release_context(ctx); | |
1412 | err0: | |
1413 | kfree(lun_access); | |
1414 | goto out; | |
1415 | } | |
1416 | ||
1417 | /** | |
1418 | * recover_context() - recovers a context in error | |
1419 | * @cfg: Internal structure associated with the host. | |
1420 | * @ctxi: Context to release. | |
1421 | * | |
1422 | * Restablishes the state for a context-in-error. | |
1423 | * | |
1424 | * Return: 0 on success, -errno on failure | |
1425 | */ | |
1426 | static int recover_context(struct cxlflash_cfg *cfg, struct ctx_info *ctxi) | |
1427 | { | |
1428 | struct device *dev = &cfg->dev->dev; | |
1429 | int rc = 0; | |
1430 | int old_fd, fd = -1; | |
1431 | int ctxid = -1; | |
1432 | struct file *file; | |
1433 | struct cxl_context *ctx; | |
1434 | struct afu *afu = cfg->afu; | |
1435 | ||
1436 | ctx = cxl_dev_context_init(cfg->dev); | |
1437 | if (unlikely(IS_ERR_OR_NULL(ctx))) { | |
1438 | dev_err(dev, "%s: Could not initialize context %p\n", | |
1439 | __func__, ctx); | |
1440 | rc = -ENODEV; | |
1441 | goto out; | |
1442 | } | |
1443 | ||
1444 | ctxid = cxl_process_element(ctx); | |
1445 | if (unlikely((ctxid > MAX_CONTEXT) || (ctxid < 0))) { | |
1446 | dev_err(dev, "%s: ctxid (%d) invalid!\n", __func__, ctxid); | |
1447 | rc = -EPERM; | |
1448 | goto err1; | |
1449 | } | |
1450 | ||
1451 | file = cxl_get_fd(ctx, &cfg->cxl_fops, &fd); | |
1452 | if (unlikely(fd < 0)) { | |
1453 | rc = -ENODEV; | |
1454 | dev_err(dev, "%s: Could not get file descriptor\n", __func__); | |
1455 | goto err1; | |
1456 | } | |
1457 | ||
1458 | rc = cxl_start_work(ctx, &ctxi->work); | |
1459 | if (unlikely(rc)) { | |
1460 | dev_dbg(dev, "%s: Could not start context rc=%d\n", | |
1461 | __func__, rc); | |
1462 | goto err2; | |
1463 | } | |
1464 | ||
1465 | /* Update with new MMIO area based on updated context id */ | |
1466 | ctxi->ctrl_map = &afu->afu_map->ctrls[ctxid].ctrl; | |
1467 | ||
1468 | rc = afu_attach(cfg, ctxi); | |
1469 | if (rc) { | |
1470 | dev_err(dev, "%s: Could not attach AFU rc %d\n", __func__, rc); | |
1471 | goto err3; | |
1472 | } | |
1473 | ||
1474 | /* | |
1475 | * No error paths after this point. Once the fd is installed it's | |
1476 | * visible to user space and can't be undone safely on this thread. | |
1477 | */ | |
1478 | old_fd = ctxi->lfd; | |
1479 | ctxi->ctxid = ENCODE_CTXID(ctxi, ctxid); | |
1480 | ctxi->lfd = fd; | |
1481 | ctxi->ctx = ctx; | |
1482 | ctxi->file = file; | |
1483 | ||
1484 | /* | |
1485 | * Put context back in table (note the reinit of the context list); | |
1486 | * we must first drop the context's mutex and then acquire it in | |
1487 | * order with the table/list mutex to avoid a deadlock - safe to do | |
1488 | * here because no one can find us at this moment in time. | |
1489 | */ | |
1490 | mutex_unlock(&ctxi->mutex); | |
1491 | mutex_lock(&cfg->ctx_tbl_list_mutex); | |
1492 | mutex_lock(&ctxi->mutex); | |
1493 | list_del_init(&ctxi->list); | |
1494 | cfg->ctx_tbl[ctxid] = ctxi; | |
1495 | mutex_unlock(&cfg->ctx_tbl_list_mutex); | |
1496 | fd_install(fd, file); | |
1497 | ||
1498 | /* Release the original adapter fd and associated CXL resources */ | |
1499 | sys_close(old_fd); | |
1500 | out: | |
1501 | dev_dbg(dev, "%s: returning ctxid=%d fd=%d rc=%d\n", | |
1502 | __func__, ctxid, fd, rc); | |
1503 | return rc; | |
1504 | ||
1505 | err3: | |
1506 | cxl_stop_context(ctx); | |
1507 | err2: | |
1508 | fput(file); | |
1509 | put_unused_fd(fd); | |
1510 | err1: | |
1511 | cxl_release_context(ctx); | |
1512 | goto out; | |
1513 | } | |
1514 | ||
1515 | /** | |
1516 | * check_state() - checks and responds to the current adapter state | |
1517 | * @cfg: Internal structure associated with the host. | |
1518 | * | |
1519 | * This routine can block and should only be used on process context. | |
1520 | * Note that when waking up from waiting in limbo, the state is unknown | |
1521 | * and must be checked again before proceeding. | |
1522 | * | |
1523 | * Return: 0 on success, -errno on failure | |
1524 | */ | |
1525 | static int check_state(struct cxlflash_cfg *cfg) | |
1526 | { | |
1527 | struct device *dev = &cfg->dev->dev; | |
1528 | int rc = 0; | |
1529 | ||
1530 | retry: | |
1531 | switch (cfg->state) { | |
1532 | case STATE_LIMBO: | |
1533 | dev_dbg(dev, "%s: Limbo, going to wait...\n", __func__); | |
1534 | rc = wait_event_interruptible(cfg->limbo_waitq, | |
1535 | cfg->state != STATE_LIMBO); | |
1536 | if (unlikely(rc)) | |
1537 | break; | |
1538 | goto retry; | |
1539 | case STATE_FAILTERM: | |
1540 | dev_dbg(dev, "%s: Failed/Terminating!\n", __func__); | |
1541 | rc = -ENODEV; | |
1542 | break; | |
1543 | default: | |
1544 | break; | |
1545 | } | |
1546 | ||
1547 | return rc; | |
1548 | } | |
1549 | ||
1550 | /** | |
1551 | * cxlflash_afu_recover() - initiates AFU recovery | |
1552 | * @sdev: SCSI device associated with LUN. | |
1553 | * @recover: Recover ioctl data structure. | |
1554 | * | |
1555 | * Only a single recovery is allowed at a time to avoid exhausting CXL | |
1556 | * resources (leading to recovery failure) in the event that we're up | |
1557 | * against the maximum number of contexts limit. For similar reasons, | |
1558 | * a context recovery is retried if there are multiple recoveries taking | |
1559 | * place at the same time and the failure was due to CXL services being | |
1560 | * unable to keep up. | |
1561 | * | |
1562 | * Because a user can detect an error condition before the kernel, it is | |
1563 | * quite possible for this routine to act as the kernel's EEH detection | |
1564 | * source (MMIO read of mbox_r). Because of this, there is a window of | |
1565 | * time where an EEH might have been detected but not yet 'serviced' | |
1566 | * (callback invoked, causing the device to enter limbo state). To avoid | |
1567 | * looping in this routine during that window, a 1 second sleep is in place | |
1568 | * between the time the MMIO failure is detected and the time a wait on the | |
1569 | * limbo wait queue is attempted via check_state(). | |
1570 | * | |
1571 | * Return: 0 on success, -errno on failure | |
1572 | */ | |
1573 | static int cxlflash_afu_recover(struct scsi_device *sdev, | |
1574 | struct dk_cxlflash_recover_afu *recover) | |
1575 | { | |
1576 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
1577 | struct device *dev = &cfg->dev->dev; | |
1578 | struct llun_info *lli = sdev->hostdata; | |
1579 | struct afu *afu = cfg->afu; | |
1580 | struct ctx_info *ctxi = NULL; | |
1581 | struct mutex *mutex = &cfg->ctx_recovery_mutex; | |
1582 | u64 ctxid = DECODE_CTXID(recover->context_id), | |
1583 | rctxid = recover->context_id; | |
1584 | long reg; | |
1585 | int lretry = 20; /* up to 2 seconds */ | |
1586 | int rc = 0; | |
1587 | ||
1588 | atomic_inc(&cfg->recovery_threads); | |
1589 | rc = mutex_lock_interruptible(mutex); | |
1590 | if (rc) | |
1591 | goto out; | |
1592 | ||
1593 | dev_dbg(dev, "%s: reason 0x%016llX rctxid=%016llX\n", | |
1594 | __func__, recover->reason, rctxid); | |
1595 | ||
1596 | retry: | |
1597 | /* Ensure that this process is attached to the context */ | |
1598 | ctxi = get_context(cfg, rctxid, lli, CTX_CTRL_ERR_FALLBACK); | |
1599 | if (unlikely(!ctxi)) { | |
1600 | dev_dbg(dev, "%s: Bad context! (%llu)\n", __func__, ctxid); | |
1601 | rc = -EINVAL; | |
1602 | goto out; | |
1603 | } | |
1604 | ||
1605 | if (ctxi->err_recovery_active) { | |
1606 | retry_recover: | |
1607 | rc = recover_context(cfg, ctxi); | |
1608 | if (unlikely(rc)) { | |
1609 | dev_err(dev, "%s: Recovery failed for context %llu (rc=%d)\n", | |
1610 | __func__, ctxid, rc); | |
1611 | if ((rc == -ENODEV) && | |
1612 | ((atomic_read(&cfg->recovery_threads) > 1) || | |
1613 | (lretry--))) { | |
1614 | dev_dbg(dev, "%s: Going to try again!\n", | |
1615 | __func__); | |
1616 | mutex_unlock(mutex); | |
1617 | msleep(100); | |
1618 | rc = mutex_lock_interruptible(mutex); | |
1619 | if (rc) | |
1620 | goto out; | |
1621 | goto retry_recover; | |
1622 | } | |
1623 | ||
1624 | goto out; | |
1625 | } | |
1626 | ||
1627 | ctxi->err_recovery_active = false; | |
1628 | recover->context_id = ctxi->ctxid; | |
1629 | recover->adap_fd = ctxi->lfd; | |
1630 | recover->mmio_size = sizeof(afu->afu_map->hosts[0].harea); | |
1631 | recover->hdr.return_flags |= | |
1632 | DK_CXLFLASH_RECOVER_AFU_CONTEXT_RESET; | |
1633 | goto out; | |
1634 | } | |
1635 | ||
1636 | /* Test if in error state */ | |
1637 | reg = readq_be(&afu->ctrl_map->mbox_r); | |
1638 | if (reg == -1) { | |
1639 | dev_dbg(dev, "%s: MMIO read fail! Wait for recovery...\n", | |
1640 | __func__); | |
1641 | mutex_unlock(&ctxi->mutex); | |
1642 | ctxi = NULL; | |
1643 | ssleep(1); | |
1644 | rc = check_state(cfg); | |
1645 | if (unlikely(rc)) | |
1646 | goto out; | |
1647 | goto retry; | |
1648 | } | |
1649 | ||
1650 | dev_dbg(dev, "%s: MMIO working, no recovery required!\n", __func__); | |
1651 | out: | |
1652 | if (likely(ctxi)) | |
1653 | put_context(ctxi); | |
1654 | mutex_unlock(mutex); | |
1655 | atomic_dec_if_positive(&cfg->recovery_threads); | |
1656 | return rc; | |
1657 | } | |
1658 | ||
1659 | /** | |
1660 | * process_sense() - evaluates and processes sense data | |
1661 | * @sdev: SCSI device associated with LUN. | |
1662 | * @verify: Verify ioctl data structure. | |
1663 | * | |
1664 | * Return: 0 on success, -errno on failure | |
1665 | */ | |
1666 | static int process_sense(struct scsi_device *sdev, | |
1667 | struct dk_cxlflash_verify *verify) | |
1668 | { | |
1669 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
1670 | struct device *dev = &cfg->dev->dev; | |
1671 | struct llun_info *lli = sdev->hostdata; | |
1672 | struct glun_info *gli = lli->parent; | |
1673 | u64 prev_lba = gli->max_lba; | |
1674 | struct scsi_sense_hdr sshdr = { 0 }; | |
1675 | int rc = 0; | |
1676 | ||
1677 | rc = scsi_normalize_sense((const u8 *)&verify->sense_data, | |
1678 | DK_CXLFLASH_VERIFY_SENSE_LEN, &sshdr); | |
1679 | if (!rc) { | |
1680 | dev_err(dev, "%s: Failed to normalize sense data!\n", __func__); | |
1681 | rc = -EINVAL; | |
1682 | goto out; | |
1683 | } | |
1684 | ||
1685 | switch (sshdr.sense_key) { | |
1686 | case NO_SENSE: | |
1687 | case RECOVERED_ERROR: | |
1688 | /* fall through */ | |
1689 | case NOT_READY: | |
1690 | break; | |
1691 | case UNIT_ATTENTION: | |
1692 | switch (sshdr.asc) { | |
1693 | case 0x29: /* Power on Reset or Device Reset */ | |
1694 | /* fall through */ | |
1695 | case 0x2A: /* Device settings/capacity changed */ | |
1696 | rc = read_cap16(sdev, lli); | |
1697 | if (rc) { | |
1698 | rc = -ENODEV; | |
1699 | break; | |
1700 | } | |
1701 | if (prev_lba != gli->max_lba) | |
1702 | dev_dbg(dev, "%s: Capacity changed old=%lld " | |
1703 | "new=%lld\n", __func__, prev_lba, | |
1704 | gli->max_lba); | |
1705 | break; | |
1706 | case 0x3F: /* Report LUNs changed, Rescan. */ | |
1707 | scsi_scan_host(cfg->host); | |
1708 | break; | |
1709 | default: | |
1710 | rc = -EIO; | |
1711 | break; | |
1712 | } | |
1713 | break; | |
1714 | default: | |
1715 | rc = -EIO; | |
1716 | break; | |
1717 | } | |
1718 | out: | |
1719 | dev_dbg(dev, "%s: sense_key %x asc %x ascq %x rc %d\n", __func__, | |
1720 | sshdr.sense_key, sshdr.asc, sshdr.ascq, rc); | |
1721 | return rc; | |
1722 | } | |
1723 | ||
1724 | /** | |
1725 | * cxlflash_disk_verify() - verifies a LUN is the same and handle size changes | |
1726 | * @sdev: SCSI device associated with LUN. | |
1727 | * @verify: Verify ioctl data structure. | |
1728 | * | |
1729 | * Return: 0 on success, -errno on failure | |
1730 | */ | |
1731 | static int cxlflash_disk_verify(struct scsi_device *sdev, | |
1732 | struct dk_cxlflash_verify *verify) | |
1733 | { | |
1734 | int rc = 0; | |
1735 | struct ctx_info *ctxi = NULL; | |
1736 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
1737 | struct device *dev = &cfg->dev->dev; | |
1738 | struct llun_info *lli = sdev->hostdata; | |
1739 | struct glun_info *gli = lli->parent; | |
1740 | struct sisl_rht_entry *rhte = NULL; | |
1741 | res_hndl_t rhndl = verify->rsrc_handle; | |
1742 | u64 ctxid = DECODE_CTXID(verify->context_id), | |
1743 | rctxid = verify->context_id; | |
1744 | u64 last_lba = 0; | |
1745 | ||
1746 | dev_dbg(dev, "%s: ctxid=%llu rhndl=%016llX, hint=%016llX, " | |
1747 | "flags=%016llX\n", __func__, ctxid, verify->rsrc_handle, | |
1748 | verify->hint, verify->hdr.flags); | |
1749 | ||
1750 | ctxi = get_context(cfg, rctxid, lli, 0); | |
1751 | if (unlikely(!ctxi)) { | |
1752 | dev_dbg(dev, "%s: Bad context! (%llu)\n", __func__, ctxid); | |
1753 | rc = -EINVAL; | |
1754 | goto out; | |
1755 | } | |
1756 | ||
1757 | rhte = get_rhte(ctxi, rhndl, lli); | |
1758 | if (unlikely(!rhte)) { | |
1759 | dev_dbg(dev, "%s: Bad resource handle! (%d)\n", | |
1760 | __func__, rhndl); | |
1761 | rc = -EINVAL; | |
1762 | goto out; | |
1763 | } | |
1764 | ||
1765 | /* | |
1766 | * Look at the hint/sense to see if it requires us to redrive | |
1767 | * inquiry (i.e. the Unit attention is due to the WWN changing). | |
1768 | */ | |
1769 | if (verify->hint & DK_CXLFLASH_VERIFY_HINT_SENSE) { | |
1770 | rc = process_sense(sdev, verify); | |
1771 | if (unlikely(rc)) { | |
1772 | dev_err(dev, "%s: Failed to validate sense data (%d)\n", | |
1773 | __func__, rc); | |
1774 | goto out; | |
1775 | } | |
1776 | } | |
1777 | ||
1778 | switch (gli->mode) { | |
1779 | case MODE_PHYSICAL: | |
1780 | last_lba = gli->max_lba; | |
1781 | break; | |
2cb79266 MO |
1782 | case MODE_VIRTUAL: |
1783 | /* Cast lxt_cnt to u64 for multiply to be treated as 64bit op */ | |
1784 | last_lba = ((u64)rhte->lxt_cnt * MC_CHUNK_SIZE * gli->blk_len); | |
1785 | last_lba /= CXLFLASH_BLOCK_SIZE; | |
1786 | last_lba--; | |
1787 | break; | |
65be2c79 MO |
1788 | default: |
1789 | WARN(1, "Unsupported LUN mode!"); | |
1790 | } | |
1791 | ||
1792 | verify->last_lba = last_lba; | |
1793 | ||
1794 | out: | |
1795 | if (likely(ctxi)) | |
1796 | put_context(ctxi); | |
1797 | dev_dbg(dev, "%s: returning rc=%d llba=%llX\n", | |
1798 | __func__, rc, verify->last_lba); | |
1799 | return rc; | |
1800 | } | |
1801 | ||
1802 | /** | |
1803 | * decode_ioctl() - translates an encoded ioctl to an easily identifiable string | |
1804 | * @cmd: The ioctl command to decode. | |
1805 | * | |
1806 | * Return: A string identifying the decoded ioctl. | |
1807 | */ | |
1808 | static char *decode_ioctl(int cmd) | |
1809 | { | |
1810 | switch (cmd) { | |
1811 | case DK_CXLFLASH_ATTACH: | |
1812 | return __stringify_1(DK_CXLFLASH_ATTACH); | |
1813 | case DK_CXLFLASH_USER_DIRECT: | |
1814 | return __stringify_1(DK_CXLFLASH_USER_DIRECT); | |
2cb79266 MO |
1815 | case DK_CXLFLASH_USER_VIRTUAL: |
1816 | return __stringify_1(DK_CXLFLASH_USER_VIRTUAL); | |
1817 | case DK_CXLFLASH_VLUN_RESIZE: | |
1818 | return __stringify_1(DK_CXLFLASH_VLUN_RESIZE); | |
65be2c79 MO |
1819 | case DK_CXLFLASH_RELEASE: |
1820 | return __stringify_1(DK_CXLFLASH_RELEASE); | |
1821 | case DK_CXLFLASH_DETACH: | |
1822 | return __stringify_1(DK_CXLFLASH_DETACH); | |
1823 | case DK_CXLFLASH_VERIFY: | |
1824 | return __stringify_1(DK_CXLFLASH_VERIFY); | |
2cb79266 MO |
1825 | case DK_CXLFLASH_VLUN_CLONE: |
1826 | return __stringify_1(DK_CXLFLASH_VLUN_CLONE); | |
65be2c79 MO |
1827 | case DK_CXLFLASH_RECOVER_AFU: |
1828 | return __stringify_1(DK_CXLFLASH_RECOVER_AFU); | |
1829 | case DK_CXLFLASH_MANAGE_LUN: | |
1830 | return __stringify_1(DK_CXLFLASH_MANAGE_LUN); | |
1831 | } | |
1832 | ||
1833 | return "UNKNOWN"; | |
1834 | } | |
1835 | ||
1836 | /** | |
1837 | * cxlflash_disk_direct_open() - opens a direct (physical) disk | |
1838 | * @sdev: SCSI device associated with LUN. | |
1839 | * @arg: UDirect ioctl data structure. | |
1840 | * | |
1841 | * On successful return, the user is informed of the resource handle | |
1842 | * to be used to identify the direct lun and the size (in blocks) of | |
1843 | * the direct lun in last LBA format. | |
1844 | * | |
1845 | * Return: 0 on success, -errno on failure | |
1846 | */ | |
1847 | static int cxlflash_disk_direct_open(struct scsi_device *sdev, void *arg) | |
1848 | { | |
1849 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
1850 | struct device *dev = &cfg->dev->dev; | |
1851 | struct afu *afu = cfg->afu; | |
1852 | struct llun_info *lli = sdev->hostdata; | |
1853 | struct glun_info *gli = lli->parent; | |
1854 | ||
1855 | struct dk_cxlflash_udirect *pphys = (struct dk_cxlflash_udirect *)arg; | |
1856 | ||
1857 | u64 ctxid = DECODE_CTXID(pphys->context_id), | |
1858 | rctxid = pphys->context_id; | |
1859 | u64 lun_size = 0; | |
1860 | u64 last_lba = 0; | |
1861 | u64 rsrc_handle = -1; | |
1862 | u32 port = CHAN2PORT(sdev->channel); | |
1863 | ||
1864 | int rc = 0; | |
1865 | ||
1866 | struct ctx_info *ctxi = NULL; | |
1867 | struct sisl_rht_entry *rhte = NULL; | |
1868 | ||
1869 | pr_debug("%s: ctxid=%llu ls=0x%llx\n", __func__, ctxid, lun_size); | |
1870 | ||
1871 | rc = cxlflash_lun_attach(gli, MODE_PHYSICAL, false); | |
1872 | if (unlikely(rc)) { | |
1873 | dev_dbg(dev, "%s: Failed to attach to LUN! (PHYSICAL)\n", | |
1874 | __func__); | |
1875 | goto out; | |
1876 | } | |
1877 | ||
1878 | ctxi = get_context(cfg, rctxid, lli, 0); | |
1879 | if (unlikely(!ctxi)) { | |
1880 | dev_dbg(dev, "%s: Bad context! (%llu)\n", __func__, ctxid); | |
1881 | rc = -EINVAL; | |
1882 | goto err1; | |
1883 | } | |
1884 | ||
1885 | rhte = rhte_checkout(ctxi, lli); | |
1886 | if (unlikely(!rhte)) { | |
1887 | dev_dbg(dev, "%s: too many opens for this context\n", __func__); | |
1888 | rc = -EMFILE; /* too many opens */ | |
1889 | goto err1; | |
1890 | } | |
1891 | ||
1892 | rsrc_handle = (rhte - ctxi->rht_start); | |
1893 | ||
1894 | rht_format1(rhte, lli->lun_id[sdev->channel], ctxi->rht_perms, port); | |
1895 | cxlflash_afu_sync(afu, ctxid, rsrc_handle, AFU_LW_SYNC); | |
1896 | ||
1897 | last_lba = gli->max_lba; | |
1898 | pphys->hdr.return_flags = 0; | |
1899 | pphys->last_lba = last_lba; | |
1900 | pphys->rsrc_handle = rsrc_handle; | |
1901 | ||
1902 | out: | |
1903 | if (likely(ctxi)) | |
1904 | put_context(ctxi); | |
1905 | dev_dbg(dev, "%s: returning handle 0x%llx rc=%d llba %lld\n", | |
1906 | __func__, rsrc_handle, rc, last_lba); | |
1907 | return rc; | |
1908 | ||
1909 | err1: | |
1910 | cxlflash_lun_detach(gli); | |
1911 | goto out; | |
1912 | } | |
1913 | ||
1914 | /** | |
1915 | * ioctl_common() - common IOCTL handler for driver | |
1916 | * @sdev: SCSI device associated with LUN. | |
1917 | * @cmd: IOCTL command. | |
1918 | * | |
1919 | * Handles common fencing operations that are valid for multiple ioctls. Always | |
1920 | * allow through ioctls that are cleanup oriented in nature, even when operating | |
1921 | * in a failed/terminating state. | |
1922 | * | |
1923 | * Return: 0 on success, -errno on failure | |
1924 | */ | |
1925 | static int ioctl_common(struct scsi_device *sdev, int cmd) | |
1926 | { | |
1927 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
1928 | struct device *dev = &cfg->dev->dev; | |
1929 | struct llun_info *lli = sdev->hostdata; | |
1930 | int rc = 0; | |
1931 | ||
1932 | if (unlikely(!lli)) { | |
1933 | dev_dbg(dev, "%s: Unknown LUN\n", __func__); | |
1934 | rc = -EINVAL; | |
1935 | goto out; | |
1936 | } | |
1937 | ||
1938 | rc = check_state(cfg); | |
1939 | if (unlikely(rc) && (cfg->state == STATE_FAILTERM)) { | |
1940 | switch (cmd) { | |
2cb79266 | 1941 | case DK_CXLFLASH_VLUN_RESIZE: |
65be2c79 MO |
1942 | case DK_CXLFLASH_RELEASE: |
1943 | case DK_CXLFLASH_DETACH: | |
1944 | dev_dbg(dev, "%s: Command override! (%d)\n", | |
1945 | __func__, rc); | |
1946 | rc = 0; | |
1947 | break; | |
1948 | } | |
1949 | } | |
1950 | out: | |
1951 | return rc; | |
1952 | } | |
1953 | ||
1954 | /** | |
1955 | * cxlflash_ioctl() - IOCTL handler for driver | |
1956 | * @sdev: SCSI device associated with LUN. | |
1957 | * @cmd: IOCTL command. | |
1958 | * @arg: Userspace ioctl data structure. | |
1959 | * | |
1960 | * Return: 0 on success, -errno on failure | |
1961 | */ | |
1962 | int cxlflash_ioctl(struct scsi_device *sdev, int cmd, void __user *arg) | |
1963 | { | |
1964 | typedef int (*sioctl) (struct scsi_device *, void *); | |
1965 | ||
1966 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
1967 | struct device *dev = &cfg->dev->dev; | |
1968 | struct afu *afu = cfg->afu; | |
1969 | struct dk_cxlflash_hdr *hdr; | |
1970 | char buf[sizeof(union cxlflash_ioctls)]; | |
1971 | size_t size = 0; | |
1972 | bool known_ioctl = false; | |
1973 | int idx; | |
1974 | int rc = 0; | |
1975 | struct Scsi_Host *shost = sdev->host; | |
1976 | sioctl do_ioctl = NULL; | |
1977 | ||
1978 | static const struct { | |
1979 | size_t size; | |
1980 | sioctl ioctl; | |
1981 | } ioctl_tbl[] = { /* NOTE: order matters here */ | |
1982 | {sizeof(struct dk_cxlflash_attach), (sioctl)cxlflash_disk_attach}, | |
1983 | {sizeof(struct dk_cxlflash_udirect), cxlflash_disk_direct_open}, | |
1984 | {sizeof(struct dk_cxlflash_release), (sioctl)cxlflash_disk_release}, | |
1985 | {sizeof(struct dk_cxlflash_detach), (sioctl)cxlflash_disk_detach}, | |
1986 | {sizeof(struct dk_cxlflash_verify), (sioctl)cxlflash_disk_verify}, | |
1987 | {sizeof(struct dk_cxlflash_recover_afu), (sioctl)cxlflash_afu_recover}, | |
1988 | {sizeof(struct dk_cxlflash_manage_lun), (sioctl)cxlflash_manage_lun}, | |
2cb79266 MO |
1989 | {sizeof(struct dk_cxlflash_uvirtual), cxlflash_disk_virtual_open}, |
1990 | {sizeof(struct dk_cxlflash_resize), (sioctl)cxlflash_vlun_resize}, | |
1991 | {sizeof(struct dk_cxlflash_clone), (sioctl)cxlflash_disk_clone}, | |
65be2c79 MO |
1992 | }; |
1993 | ||
1994 | /* Restrict command set to physical support only for internal LUN */ | |
1995 | if (afu->internal_lun) | |
1996 | switch (cmd) { | |
1997 | case DK_CXLFLASH_RELEASE: | |
2cb79266 MO |
1998 | case DK_CXLFLASH_USER_VIRTUAL: |
1999 | case DK_CXLFLASH_VLUN_RESIZE: | |
2000 | case DK_CXLFLASH_VLUN_CLONE: | |
65be2c79 MO |
2001 | dev_dbg(dev, "%s: %s not supported for lun_mode=%d\n", |
2002 | __func__, decode_ioctl(cmd), afu->internal_lun); | |
2003 | rc = -EINVAL; | |
2004 | goto cxlflash_ioctl_exit; | |
2005 | } | |
2006 | ||
2007 | switch (cmd) { | |
2008 | case DK_CXLFLASH_ATTACH: | |
2009 | case DK_CXLFLASH_USER_DIRECT: | |
2010 | case DK_CXLFLASH_RELEASE: | |
2011 | case DK_CXLFLASH_DETACH: | |
2012 | case DK_CXLFLASH_VERIFY: | |
2013 | case DK_CXLFLASH_RECOVER_AFU: | |
2cb79266 MO |
2014 | case DK_CXLFLASH_USER_VIRTUAL: |
2015 | case DK_CXLFLASH_VLUN_RESIZE: | |
2016 | case DK_CXLFLASH_VLUN_CLONE: | |
65be2c79 MO |
2017 | dev_dbg(dev, "%s: %s (%08X) on dev(%d/%d/%d/%llu)\n", |
2018 | __func__, decode_ioctl(cmd), cmd, shost->host_no, | |
2019 | sdev->channel, sdev->id, sdev->lun); | |
2020 | rc = ioctl_common(sdev, cmd); | |
2021 | if (unlikely(rc)) | |
2022 | goto cxlflash_ioctl_exit; | |
2023 | ||
2024 | /* fall through */ | |
2025 | ||
2026 | case DK_CXLFLASH_MANAGE_LUN: | |
2027 | known_ioctl = true; | |
2028 | idx = _IOC_NR(cmd) - _IOC_NR(DK_CXLFLASH_ATTACH); | |
2029 | size = ioctl_tbl[idx].size; | |
2030 | do_ioctl = ioctl_tbl[idx].ioctl; | |
2031 | ||
2032 | if (likely(do_ioctl)) | |
2033 | break; | |
2034 | ||
2035 | /* fall through */ | |
2036 | default: | |
2037 | rc = -EINVAL; | |
2038 | goto cxlflash_ioctl_exit; | |
2039 | } | |
2040 | ||
2041 | if (unlikely(copy_from_user(&buf, arg, size))) { | |
2042 | dev_err(dev, "%s: copy_from_user() fail! " | |
2043 | "size=%lu cmd=%d (%s) arg=%p\n", | |
2044 | __func__, size, cmd, decode_ioctl(cmd), arg); | |
2045 | rc = -EFAULT; | |
2046 | goto cxlflash_ioctl_exit; | |
2047 | } | |
2048 | ||
2049 | hdr = (struct dk_cxlflash_hdr *)&buf; | |
2050 | if (hdr->version != DK_CXLFLASH_VERSION_0) { | |
2051 | dev_dbg(dev, "%s: Version %u not supported for %s\n", | |
2052 | __func__, hdr->version, decode_ioctl(cmd)); | |
2053 | rc = -EINVAL; | |
2054 | goto cxlflash_ioctl_exit; | |
2055 | } | |
2056 | ||
2057 | if (hdr->rsvd[0] || hdr->rsvd[1] || hdr->rsvd[2] || hdr->return_flags) { | |
2058 | dev_dbg(dev, "%s: Reserved/rflags populated!\n", __func__); | |
2059 | rc = -EINVAL; | |
2060 | goto cxlflash_ioctl_exit; | |
2061 | } | |
2062 | ||
2063 | rc = do_ioctl(sdev, (void *)&buf); | |
2064 | if (likely(!rc)) | |
2065 | if (unlikely(copy_to_user(arg, &buf, size))) { | |
2066 | dev_err(dev, "%s: copy_to_user() fail! " | |
2067 | "size=%lu cmd=%d (%s) arg=%p\n", | |
2068 | __func__, size, cmd, decode_ioctl(cmd), arg); | |
2069 | rc = -EFAULT; | |
2070 | } | |
2071 | ||
2072 | /* fall through to exit */ | |
2073 | ||
2074 | cxlflash_ioctl_exit: | |
2075 | if (unlikely(rc && known_ioctl)) | |
2076 | dev_err(dev, "%s: ioctl %s (%08X) on dev(%d/%d/%d/%llu) " | |
2077 | "returned rc %d\n", __func__, | |
2078 | decode_ioctl(cmd), cmd, shost->host_no, | |
2079 | sdev->channel, sdev->id, sdev->lun, rc); | |
2080 | else | |
2081 | dev_dbg(dev, "%s: ioctl %s (%08X) on dev(%d/%d/%d/%llu) " | |
2082 | "returned rc %d\n", __func__, decode_ioctl(cmd), | |
2083 | cmd, shost->host_no, sdev->channel, sdev->id, | |
2084 | sdev->lun, rc); | |
2085 | return rc; | |
2086 | } |