Commit | Line | Data |
---|---|---|
2cb79266 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/syscalls.h> | |
16 | #include <misc/cxl.h> | |
17 | #include <asm/unaligned.h> | |
18 | #include <asm/bitsperlong.h> | |
19 | ||
20 | #include <scsi/scsi_cmnd.h> | |
21 | #include <scsi/scsi_host.h> | |
22 | #include <uapi/scsi/cxlflash_ioctl.h> | |
23 | ||
24 | #include "sislite.h" | |
25 | #include "common.h" | |
26 | #include "vlun.h" | |
27 | #include "superpipe.h" | |
28 | ||
29 | /** | |
30 | * marshal_virt_to_resize() - translate uvirtual to resize structure | |
31 | * @virt: Source structure from which to translate/copy. | |
32 | * @resize: Destination structure for the translate/copy. | |
33 | */ | |
34 | static void marshal_virt_to_resize(struct dk_cxlflash_uvirtual *virt, | |
35 | struct dk_cxlflash_resize *resize) | |
36 | { | |
37 | resize->hdr = virt->hdr; | |
38 | resize->context_id = virt->context_id; | |
39 | resize->rsrc_handle = virt->rsrc_handle; | |
40 | resize->req_size = virt->lun_size; | |
41 | resize->last_lba = virt->last_lba; | |
42 | } | |
43 | ||
44 | /** | |
45 | * marshal_clone_to_rele() - translate clone to release structure | |
46 | * @clone: Source structure from which to translate/copy. | |
47 | * @rele: Destination structure for the translate/copy. | |
48 | */ | |
49 | static void marshal_clone_to_rele(struct dk_cxlflash_clone *clone, | |
50 | struct dk_cxlflash_release *release) | |
51 | { | |
52 | release->hdr = clone->hdr; | |
53 | release->context_id = clone->context_id_dst; | |
54 | } | |
55 | ||
56 | /** | |
57 | * ba_init() - initializes a block allocator | |
58 | * @ba_lun: Block allocator to initialize. | |
59 | * | |
60 | * Return: 0 on success, -errno on failure | |
61 | */ | |
62 | static int ba_init(struct ba_lun *ba_lun) | |
63 | { | |
64 | struct ba_lun_info *bali = NULL; | |
65 | int lun_size_au = 0, i = 0; | |
66 | int last_word_underflow = 0; | |
67 | u64 *lam; | |
68 | ||
69 | pr_debug("%s: Initializing LUN: lun_id = %llX, " | |
70 | "ba_lun->lsize = %lX, ba_lun->au_size = %lX\n", | |
71 | __func__, ba_lun->lun_id, ba_lun->lsize, ba_lun->au_size); | |
72 | ||
73 | /* Calculate bit map size */ | |
74 | lun_size_au = ba_lun->lsize / ba_lun->au_size; | |
75 | if (lun_size_au == 0) { | |
76 | pr_debug("%s: Requested LUN size of 0!\n", __func__); | |
77 | return -EINVAL; | |
78 | } | |
79 | ||
80 | /* Allocate lun information container */ | |
81 | bali = kzalloc(sizeof(struct ba_lun_info), GFP_KERNEL); | |
82 | if (unlikely(!bali)) { | |
83 | pr_err("%s: Failed to allocate lun_info for lun_id %llX\n", | |
84 | __func__, ba_lun->lun_id); | |
85 | return -ENOMEM; | |
86 | } | |
87 | ||
88 | bali->total_aus = lun_size_au; | |
89 | bali->lun_bmap_size = lun_size_au / BITS_PER_LONG; | |
90 | ||
91 | if (lun_size_au % BITS_PER_LONG) | |
92 | bali->lun_bmap_size++; | |
93 | ||
94 | /* Allocate bitmap space */ | |
95 | bali->lun_alloc_map = kzalloc((bali->lun_bmap_size * sizeof(u64)), | |
96 | GFP_KERNEL); | |
97 | if (unlikely(!bali->lun_alloc_map)) { | |
98 | pr_err("%s: Failed to allocate lun allocation map: " | |
99 | "lun_id = %llX\n", __func__, ba_lun->lun_id); | |
100 | kfree(bali); | |
101 | return -ENOMEM; | |
102 | } | |
103 | ||
104 | /* Initialize the bit map size and set all bits to '1' */ | |
105 | bali->free_aun_cnt = lun_size_au; | |
106 | ||
107 | for (i = 0; i < bali->lun_bmap_size; i++) | |
108 | bali->lun_alloc_map[i] = 0xFFFFFFFFFFFFFFFFULL; | |
109 | ||
110 | /* If the last word not fully utilized, mark extra bits as allocated */ | |
111 | last_word_underflow = (bali->lun_bmap_size * BITS_PER_LONG); | |
112 | last_word_underflow -= bali->free_aun_cnt; | |
113 | if (last_word_underflow > 0) { | |
114 | lam = &bali->lun_alloc_map[bali->lun_bmap_size - 1]; | |
115 | for (i = (HIBIT - last_word_underflow + 1); | |
116 | i < BITS_PER_LONG; | |
117 | i++) | |
118 | clear_bit(i, (ulong *)lam); | |
119 | } | |
120 | ||
121 | /* Initialize high elevator index, low/curr already at 0 from kzalloc */ | |
122 | bali->free_high_idx = bali->lun_bmap_size; | |
123 | ||
124 | /* Allocate clone map */ | |
125 | bali->aun_clone_map = kzalloc((bali->total_aus * sizeof(u8)), | |
126 | GFP_KERNEL); | |
127 | if (unlikely(!bali->aun_clone_map)) { | |
128 | pr_err("%s: Failed to allocate clone map: lun_id = %llX\n", | |
129 | __func__, ba_lun->lun_id); | |
130 | kfree(bali->lun_alloc_map); | |
131 | kfree(bali); | |
132 | return -ENOMEM; | |
133 | } | |
134 | ||
f15fbf8d | 135 | /* Pass the allocated LUN info as a handle to the user */ |
2cb79266 MO |
136 | ba_lun->ba_lun_handle = bali; |
137 | ||
138 | pr_debug("%s: Successfully initialized the LUN: " | |
139 | "lun_id = %llX, bitmap size = %X, free_aun_cnt = %llX\n", | |
140 | __func__, ba_lun->lun_id, bali->lun_bmap_size, | |
141 | bali->free_aun_cnt); | |
142 | return 0; | |
143 | } | |
144 | ||
145 | /** | |
146 | * find_free_range() - locates a free bit within the block allocator | |
147 | * @low: First word in block allocator to start search. | |
148 | * @high: Last word in block allocator to search. | |
149 | * @bali: LUN information structure owning the block allocator to search. | |
150 | * @bit_word: Passes back the word in the block allocator owning the free bit. | |
151 | * | |
152 | * Return: The bit position within the passed back word, -1 on failure | |
153 | */ | |
154 | static int find_free_range(u32 low, | |
155 | u32 high, | |
156 | struct ba_lun_info *bali, int *bit_word) | |
157 | { | |
158 | int i; | |
159 | u64 bit_pos = -1; | |
160 | ulong *lam, num_bits; | |
161 | ||
162 | for (i = low; i < high; i++) | |
163 | if (bali->lun_alloc_map[i] != 0) { | |
164 | lam = (ulong *)&bali->lun_alloc_map[i]; | |
165 | num_bits = (sizeof(*lam) * BITS_PER_BYTE); | |
166 | bit_pos = find_first_bit(lam, num_bits); | |
167 | ||
f15fbf8d | 168 | pr_devel("%s: Found free bit %llX in LUN " |
2cb79266 MO |
169 | "map entry %llX at bitmap index = %X\n", |
170 | __func__, bit_pos, bali->lun_alloc_map[i], | |
171 | i); | |
172 | ||
173 | *bit_word = i; | |
174 | bali->free_aun_cnt--; | |
175 | clear_bit(bit_pos, lam); | |
176 | break; | |
177 | } | |
178 | ||
179 | return bit_pos; | |
180 | } | |
181 | ||
182 | /** | |
183 | * ba_alloc() - allocates a block from the block allocator | |
184 | * @ba_lun: Block allocator from which to allocate a block. | |
185 | * | |
186 | * Return: The allocated block, -1 on failure | |
187 | */ | |
188 | static u64 ba_alloc(struct ba_lun *ba_lun) | |
189 | { | |
190 | u64 bit_pos = -1; | |
191 | int bit_word = 0; | |
192 | struct ba_lun_info *bali = NULL; | |
193 | ||
194 | bali = ba_lun->ba_lun_handle; | |
195 | ||
196 | pr_debug("%s: Received block allocation request: " | |
197 | "lun_id = %llX, free_aun_cnt = %llX\n", | |
198 | __func__, ba_lun->lun_id, bali->free_aun_cnt); | |
199 | ||
200 | if (bali->free_aun_cnt == 0) { | |
201 | pr_debug("%s: No space left on LUN: lun_id = %llX\n", | |
202 | __func__, ba_lun->lun_id); | |
203 | return -1ULL; | |
204 | } | |
205 | ||
206 | /* Search to find a free entry, curr->high then low->curr */ | |
207 | bit_pos = find_free_range(bali->free_curr_idx, | |
208 | bali->free_high_idx, bali, &bit_word); | |
209 | if (bit_pos == -1) { | |
210 | bit_pos = find_free_range(bali->free_low_idx, | |
211 | bali->free_curr_idx, | |
212 | bali, &bit_word); | |
213 | if (bit_pos == -1) { | |
214 | pr_debug("%s: Could not find an allocation unit on LUN:" | |
215 | " lun_id = %llX\n", __func__, ba_lun->lun_id); | |
216 | return -1ULL; | |
217 | } | |
218 | } | |
219 | ||
220 | /* Update the free_curr_idx */ | |
221 | if (bit_pos == HIBIT) | |
222 | bali->free_curr_idx = bit_word + 1; | |
223 | else | |
224 | bali->free_curr_idx = bit_word; | |
225 | ||
226 | pr_debug("%s: Allocating AU number %llX, on lun_id %llX, " | |
227 | "free_aun_cnt = %llX\n", __func__, | |
228 | ((bit_word * BITS_PER_LONG) + bit_pos), ba_lun->lun_id, | |
229 | bali->free_aun_cnt); | |
230 | ||
231 | return (u64) ((bit_word * BITS_PER_LONG) + bit_pos); | |
232 | } | |
233 | ||
234 | /** | |
235 | * validate_alloc() - validates the specified block has been allocated | |
236 | * @ba_lun_info: LUN info owning the block allocator. | |
237 | * @aun: Block to validate. | |
238 | * | |
239 | * Return: 0 on success, -1 on failure | |
240 | */ | |
241 | static int validate_alloc(struct ba_lun_info *bali, u64 aun) | |
242 | { | |
243 | int idx = 0, bit_pos = 0; | |
244 | ||
245 | idx = aun / BITS_PER_LONG; | |
246 | bit_pos = aun % BITS_PER_LONG; | |
247 | ||
248 | if (test_bit(bit_pos, (ulong *)&bali->lun_alloc_map[idx])) | |
249 | return -1; | |
250 | ||
251 | return 0; | |
252 | } | |
253 | ||
254 | /** | |
255 | * ba_free() - frees a block from the block allocator | |
256 | * @ba_lun: Block allocator from which to allocate a block. | |
257 | * @to_free: Block to free. | |
258 | * | |
259 | * Return: 0 on success, -1 on failure | |
260 | */ | |
261 | static int ba_free(struct ba_lun *ba_lun, u64 to_free) | |
262 | { | |
263 | int idx = 0, bit_pos = 0; | |
264 | struct ba_lun_info *bali = NULL; | |
265 | ||
266 | bali = ba_lun->ba_lun_handle; | |
267 | ||
268 | if (validate_alloc(bali, to_free)) { | |
269 | pr_debug("%s: The AUN %llX is not allocated on lun_id %llX\n", | |
270 | __func__, to_free, ba_lun->lun_id); | |
271 | return -1; | |
272 | } | |
273 | ||
274 | pr_debug("%s: Received a request to free AU %llX on lun_id %llX, " | |
275 | "free_aun_cnt = %llX\n", __func__, to_free, ba_lun->lun_id, | |
276 | bali->free_aun_cnt); | |
277 | ||
278 | if (bali->aun_clone_map[to_free] > 0) { | |
279 | pr_debug("%s: AUN %llX on lun_id %llX has been cloned. Clone " | |
280 | "count = %X\n", __func__, to_free, ba_lun->lun_id, | |
281 | bali->aun_clone_map[to_free]); | |
282 | bali->aun_clone_map[to_free]--; | |
283 | return 0; | |
284 | } | |
285 | ||
286 | idx = to_free / BITS_PER_LONG; | |
287 | bit_pos = to_free % BITS_PER_LONG; | |
288 | ||
289 | set_bit(bit_pos, (ulong *)&bali->lun_alloc_map[idx]); | |
290 | bali->free_aun_cnt++; | |
291 | ||
292 | if (idx < bali->free_low_idx) | |
293 | bali->free_low_idx = idx; | |
294 | else if (idx > bali->free_high_idx) | |
295 | bali->free_high_idx = idx; | |
296 | ||
297 | pr_debug("%s: Successfully freed AU at bit_pos %X, bit map index %X on " | |
298 | "lun_id %llX, free_aun_cnt = %llX\n", __func__, bit_pos, idx, | |
299 | ba_lun->lun_id, bali->free_aun_cnt); | |
300 | ||
301 | return 0; | |
302 | } | |
303 | ||
304 | /** | |
305 | * ba_clone() - Clone a chunk of the block allocation table | |
306 | * @ba_lun: Block allocator from which to allocate a block. | |
307 | * @to_free: Block to free. | |
308 | * | |
309 | * Return: 0 on success, -1 on failure | |
310 | */ | |
311 | static int ba_clone(struct ba_lun *ba_lun, u64 to_clone) | |
312 | { | |
313 | struct ba_lun_info *bali = ba_lun->ba_lun_handle; | |
314 | ||
315 | if (validate_alloc(bali, to_clone)) { | |
316 | pr_debug("%s: AUN %llX is not allocated on lun_id %llX\n", | |
317 | __func__, to_clone, ba_lun->lun_id); | |
318 | return -1; | |
319 | } | |
320 | ||
321 | pr_debug("%s: Received a request to clone AUN %llX on lun_id %llX\n", | |
322 | __func__, to_clone, ba_lun->lun_id); | |
323 | ||
324 | if (bali->aun_clone_map[to_clone] == MAX_AUN_CLONE_CNT) { | |
325 | pr_debug("%s: AUN %llX on lun_id %llX hit max clones already\n", | |
326 | __func__, to_clone, ba_lun->lun_id); | |
327 | return -1; | |
328 | } | |
329 | ||
330 | bali->aun_clone_map[to_clone]++; | |
331 | ||
332 | return 0; | |
333 | } | |
334 | ||
335 | /** | |
336 | * ba_space() - returns the amount of free space left in the block allocator | |
337 | * @ba_lun: Block allocator. | |
338 | * | |
339 | * Return: Amount of free space in block allocator | |
340 | */ | |
341 | static u64 ba_space(struct ba_lun *ba_lun) | |
342 | { | |
343 | struct ba_lun_info *bali = ba_lun->ba_lun_handle; | |
344 | ||
345 | return bali->free_aun_cnt; | |
346 | } | |
347 | ||
348 | /** | |
349 | * cxlflash_ba_terminate() - frees resources associated with the block allocator | |
350 | * @ba_lun: Block allocator. | |
351 | * | |
352 | * Safe to call in a partially allocated state. | |
353 | */ | |
354 | void cxlflash_ba_terminate(struct ba_lun *ba_lun) | |
355 | { | |
356 | struct ba_lun_info *bali = ba_lun->ba_lun_handle; | |
357 | ||
358 | if (bali) { | |
359 | kfree(bali->aun_clone_map); | |
360 | kfree(bali->lun_alloc_map); | |
361 | kfree(bali); | |
362 | ba_lun->ba_lun_handle = NULL; | |
363 | } | |
364 | } | |
365 | ||
366 | /** | |
367 | * init_vlun() - initializes a LUN for virtual use | |
368 | * @lun_info: LUN information structure that owns the block allocator. | |
369 | * | |
370 | * Return: 0 on success, -errno on failure | |
371 | */ | |
372 | static int init_vlun(struct llun_info *lli) | |
373 | { | |
374 | int rc = 0; | |
375 | struct glun_info *gli = lli->parent; | |
376 | struct blka *blka = &gli->blka; | |
377 | ||
378 | memset(blka, 0, sizeof(*blka)); | |
379 | mutex_init(&blka->mutex); | |
380 | ||
381 | /* LUN IDs are unique per port, save the index instead */ | |
382 | blka->ba_lun.lun_id = lli->lun_index; | |
383 | blka->ba_lun.lsize = gli->max_lba + 1; | |
384 | blka->ba_lun.lba_size = gli->blk_len; | |
385 | ||
386 | blka->ba_lun.au_size = MC_CHUNK_SIZE; | |
387 | blka->nchunk = blka->ba_lun.lsize / MC_CHUNK_SIZE; | |
388 | ||
389 | rc = ba_init(&blka->ba_lun); | |
390 | if (unlikely(rc)) | |
391 | pr_debug("%s: cannot init block_alloc, rc=%d\n", __func__, rc); | |
392 | ||
393 | pr_debug("%s: returning rc=%d lli=%p\n", __func__, rc, lli); | |
394 | return rc; | |
395 | } | |
396 | ||
397 | /** | |
398 | * write_same16() - sends a SCSI WRITE_SAME16 (0) command to specified LUN | |
399 | * @sdev: SCSI device associated with LUN. | |
400 | * @lba: Logical block address to start write same. | |
401 | * @nblks: Number of logical blocks to write same. | |
402 | * | |
aacb4ff6 MO |
403 | * The SCSI WRITE_SAME16 can take quite a while to complete. Should an EEH occur |
404 | * while in scsi_execute(), the EEH handler will attempt to recover. As part of | |
405 | * the recovery, the handler drains all currently running ioctls, waiting until | |
406 | * they have completed before proceeding with a reset. As this routine is used | |
407 | * on the ioctl path, this can create a condition where the EEH handler becomes | |
408 | * stuck, infinitely waiting for this ioctl thread. To avoid this behavior, | |
409 | * temporarily unmark this thread as an ioctl thread by releasing the ioctl read | |
410 | * semaphore. This will allow the EEH handler to proceed with a recovery while | |
411 | * this thread is still running. Once the scsi_execute() returns, reacquire the | |
412 | * ioctl read semaphore and check the adapter state in case it changed while | |
413 | * inside of scsi_execute(). The state check will wait if the adapter is still | |
414 | * being recovered or return a failure if the recovery failed. In the event that | |
415 | * the adapter reset failed, simply return the failure as the ioctl would be | |
416 | * unable to continue. | |
417 | * | |
418 | * Note that the above puts a requirement on this routine to only be called on | |
419 | * an ioctl thread. | |
420 | * | |
2cb79266 MO |
421 | * Return: 0 on success, -errno on failure |
422 | */ | |
423 | static int write_same16(struct scsi_device *sdev, | |
424 | u64 lba, | |
425 | u32 nblks) | |
426 | { | |
427 | u8 *cmd_buf = NULL; | |
428 | u8 *scsi_cmd = NULL; | |
429 | u8 *sense_buf = NULL; | |
430 | int rc = 0; | |
431 | int result = 0; | |
432 | int ws_limit = SISLITE_MAX_WS_BLOCKS; | |
433 | u64 offset = lba; | |
434 | int left = nblks; | |
471a5a60 | 435 | u32 to = sdev->request_queue->rq_timeout; |
2cb79266 MO |
436 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; |
437 | struct device *dev = &cfg->dev->dev; | |
438 | ||
439 | cmd_buf = kzalloc(CMD_BUFSIZE, GFP_KERNEL); | |
440 | scsi_cmd = kzalloc(MAX_COMMAND_SIZE, GFP_KERNEL); | |
441 | sense_buf = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL); | |
442 | if (unlikely(!cmd_buf || !scsi_cmd || !sense_buf)) { | |
443 | rc = -ENOMEM; | |
444 | goto out; | |
445 | } | |
446 | ||
447 | while (left > 0) { | |
448 | ||
449 | scsi_cmd[0] = WRITE_SAME_16; | |
450 | put_unaligned_be64(offset, &scsi_cmd[2]); | |
451 | put_unaligned_be32(ws_limit < left ? ws_limit : left, | |
452 | &scsi_cmd[10]); | |
453 | ||
aacb4ff6 MO |
454 | /* Drop the ioctl read semahpore across lengthy call */ |
455 | up_read(&cfg->ioctl_rwsem); | |
2cb79266 | 456 | result = scsi_execute(sdev, scsi_cmd, DMA_TO_DEVICE, cmd_buf, |
471a5a60 | 457 | CMD_BUFSIZE, sense_buf, to, CMD_RETRIES, |
3ebf2030 | 458 | 0, NULL); |
aacb4ff6 MO |
459 | down_read(&cfg->ioctl_rwsem); |
460 | rc = check_state(cfg); | |
461 | if (rc) { | |
462 | dev_err(dev, "%s: Failed state! result=0x08%X\n", | |
463 | __func__, result); | |
464 | rc = -ENODEV; | |
465 | goto out; | |
466 | } | |
467 | ||
2cb79266 MO |
468 | if (result) { |
469 | dev_err_ratelimited(dev, "%s: command failed for " | |
470 | "offset %lld result=0x%x\n", | |
471 | __func__, offset, result); | |
472 | rc = -EIO; | |
473 | goto out; | |
474 | } | |
475 | left -= ws_limit; | |
476 | offset += ws_limit; | |
477 | } | |
478 | ||
479 | out: | |
480 | kfree(cmd_buf); | |
481 | kfree(scsi_cmd); | |
482 | kfree(sense_buf); | |
483 | pr_debug("%s: returning rc=%d\n", __func__, rc); | |
484 | return rc; | |
485 | } | |
486 | ||
487 | /** | |
488 | * grow_lxt() - expands the translation table associated with the specified RHTE | |
489 | * @afu: AFU associated with the host. | |
490 | * @sdev: SCSI device associated with LUN. | |
491 | * @ctxid: Context ID of context owning the RHTE. | |
492 | * @rhndl: Resource handle associated with the RHTE. | |
493 | * @rhte: Resource handle entry (RHTE). | |
494 | * @new_size: Number of translation entries associated with RHTE. | |
495 | * | |
496 | * By design, this routine employs a 'best attempt' allocation and will | |
497 | * truncate the requested size down if there is not sufficient space in | |
498 | * the block allocator to satisfy the request but there does exist some | |
499 | * amount of space. The user is made aware of this by returning the size | |
500 | * allocated. | |
501 | * | |
502 | * Return: 0 on success, -errno on failure | |
503 | */ | |
504 | static int grow_lxt(struct afu *afu, | |
505 | struct scsi_device *sdev, | |
506 | ctx_hndl_t ctxid, | |
507 | res_hndl_t rhndl, | |
508 | struct sisl_rht_entry *rhte, | |
509 | u64 *new_size) | |
510 | { | |
511 | struct sisl_lxt_entry *lxt = NULL, *lxt_old = NULL; | |
512 | struct llun_info *lli = sdev->hostdata; | |
513 | struct glun_info *gli = lli->parent; | |
514 | struct blka *blka = &gli->blka; | |
515 | u32 av_size; | |
516 | u32 ngrps, ngrps_old; | |
517 | u64 aun; /* chunk# allocated by block allocator */ | |
518 | u64 delta = *new_size - rhte->lxt_cnt; | |
519 | u64 my_new_size; | |
520 | int i, rc = 0; | |
521 | ||
522 | /* | |
523 | * Check what is available in the block allocator before re-allocating | |
524 | * LXT array. This is done up front under the mutex which must not be | |
525 | * released until after allocation is complete. | |
526 | */ | |
527 | mutex_lock(&blka->mutex); | |
528 | av_size = ba_space(&blka->ba_lun); | |
529 | if (unlikely(av_size <= 0)) { | |
530 | pr_debug("%s: ba_space error: av_size %d\n", __func__, av_size); | |
531 | mutex_unlock(&blka->mutex); | |
532 | rc = -ENOSPC; | |
533 | goto out; | |
534 | } | |
535 | ||
536 | if (av_size < delta) | |
537 | delta = av_size; | |
538 | ||
539 | lxt_old = rhte->lxt_start; | |
540 | ngrps_old = LXT_NUM_GROUPS(rhte->lxt_cnt); | |
541 | ngrps = LXT_NUM_GROUPS(rhte->lxt_cnt + delta); | |
542 | ||
543 | if (ngrps != ngrps_old) { | |
544 | /* reallocate to fit new size */ | |
545 | lxt = kzalloc((sizeof(*lxt) * LXT_GROUP_SIZE * ngrps), | |
546 | GFP_KERNEL); | |
547 | if (unlikely(!lxt)) { | |
548 | mutex_unlock(&blka->mutex); | |
549 | rc = -ENOMEM; | |
550 | goto out; | |
551 | } | |
552 | ||
553 | /* copy over all old entries */ | |
554 | memcpy(lxt, lxt_old, (sizeof(*lxt) * rhte->lxt_cnt)); | |
555 | } else | |
556 | lxt = lxt_old; | |
557 | ||
558 | /* nothing can fail from now on */ | |
559 | my_new_size = rhte->lxt_cnt + delta; | |
560 | ||
561 | /* add new entries to the end */ | |
562 | for (i = rhte->lxt_cnt; i < my_new_size; i++) { | |
563 | /* | |
564 | * Due to the earlier check of available space, ba_alloc | |
565 | * cannot fail here. If it did due to internal error, | |
566 | * leave a rlba_base of -1u which will likely be a | |
567 | * invalid LUN (too large). | |
568 | */ | |
569 | aun = ba_alloc(&blka->ba_lun); | |
570 | if ((aun == -1ULL) || (aun >= blka->nchunk)) | |
571 | pr_debug("%s: ba_alloc error: allocated chunk# %llX, " | |
572 | "max %llX\n", __func__, aun, blka->nchunk - 1); | |
573 | ||
574 | /* select both ports, use r/w perms from RHT */ | |
575 | lxt[i].rlba_base = ((aun << MC_CHUNK_SHIFT) | | |
576 | (lli->lun_index << LXT_LUNIDX_SHIFT) | | |
577 | (RHT_PERM_RW << LXT_PERM_SHIFT | | |
578 | lli->port_sel)); | |
579 | } | |
580 | ||
581 | mutex_unlock(&blka->mutex); | |
582 | ||
583 | /* | |
584 | * The following sequence is prescribed in the SISlite spec | |
585 | * for syncing up with the AFU when adding LXT entries. | |
586 | */ | |
587 | dma_wmb(); /* Make LXT updates are visible */ | |
588 | ||
589 | rhte->lxt_start = lxt; | |
590 | dma_wmb(); /* Make RHT entry's LXT table update visible */ | |
591 | ||
592 | rhte->lxt_cnt = my_new_size; | |
593 | dma_wmb(); /* Make RHT entry's LXT table size update visible */ | |
594 | ||
595 | cxlflash_afu_sync(afu, ctxid, rhndl, AFU_LW_SYNC); | |
596 | ||
597 | /* free old lxt if reallocated */ | |
598 | if (lxt != lxt_old) | |
599 | kfree(lxt_old); | |
600 | *new_size = my_new_size; | |
601 | out: | |
602 | pr_debug("%s: returning rc=%d\n", __func__, rc); | |
603 | return rc; | |
604 | } | |
605 | ||
606 | /** | |
607 | * shrink_lxt() - reduces translation table associated with the specified RHTE | |
608 | * @afu: AFU associated with the host. | |
609 | * @sdev: SCSI device associated with LUN. | |
610 | * @rhndl: Resource handle associated with the RHTE. | |
611 | * @rhte: Resource handle entry (RHTE). | |
612 | * @ctxi: Context owning resources. | |
613 | * @new_size: Number of translation entries associated with RHTE. | |
614 | * | |
615 | * Return: 0 on success, -errno on failure | |
616 | */ | |
617 | static int shrink_lxt(struct afu *afu, | |
618 | struct scsi_device *sdev, | |
619 | res_hndl_t rhndl, | |
620 | struct sisl_rht_entry *rhte, | |
621 | struct ctx_info *ctxi, | |
622 | u64 *new_size) | |
623 | { | |
624 | struct sisl_lxt_entry *lxt, *lxt_old; | |
625 | struct llun_info *lli = sdev->hostdata; | |
626 | struct glun_info *gli = lli->parent; | |
627 | struct blka *blka = &gli->blka; | |
628 | ctx_hndl_t ctxid = DECODE_CTXID(ctxi->ctxid); | |
629 | bool needs_ws = ctxi->rht_needs_ws[rhndl]; | |
630 | bool needs_sync = !ctxi->err_recovery_active; | |
631 | u32 ngrps, ngrps_old; | |
632 | u64 aun; /* chunk# allocated by block allocator */ | |
633 | u64 delta = rhte->lxt_cnt - *new_size; | |
634 | u64 my_new_size; | |
635 | int i, rc = 0; | |
636 | ||
637 | lxt_old = rhte->lxt_start; | |
638 | ngrps_old = LXT_NUM_GROUPS(rhte->lxt_cnt); | |
639 | ngrps = LXT_NUM_GROUPS(rhte->lxt_cnt - delta); | |
640 | ||
641 | if (ngrps != ngrps_old) { | |
642 | /* Reallocate to fit new size unless new size is 0 */ | |
643 | if (ngrps) { | |
644 | lxt = kzalloc((sizeof(*lxt) * LXT_GROUP_SIZE * ngrps), | |
645 | GFP_KERNEL); | |
646 | if (unlikely(!lxt)) { | |
647 | rc = -ENOMEM; | |
648 | goto out; | |
649 | } | |
650 | ||
651 | /* Copy over old entries that will remain */ | |
652 | memcpy(lxt, lxt_old, | |
653 | (sizeof(*lxt) * (rhte->lxt_cnt - delta))); | |
654 | } else | |
655 | lxt = NULL; | |
656 | } else | |
657 | lxt = lxt_old; | |
658 | ||
659 | /* Nothing can fail from now on */ | |
660 | my_new_size = rhte->lxt_cnt - delta; | |
661 | ||
662 | /* | |
663 | * The following sequence is prescribed in the SISlite spec | |
664 | * for syncing up with the AFU when removing LXT entries. | |
665 | */ | |
666 | rhte->lxt_cnt = my_new_size; | |
667 | dma_wmb(); /* Make RHT entry's LXT table size update visible */ | |
668 | ||
669 | rhte->lxt_start = lxt; | |
670 | dma_wmb(); /* Make RHT entry's LXT table update visible */ | |
671 | ||
672 | if (needs_sync) | |
673 | cxlflash_afu_sync(afu, ctxid, rhndl, AFU_HW_SYNC); | |
674 | ||
675 | if (needs_ws) { | |
676 | /* | |
677 | * Mark the context as unavailable, so that we can release | |
678 | * the mutex safely. | |
679 | */ | |
680 | ctxi->unavail = true; | |
681 | mutex_unlock(&ctxi->mutex); | |
682 | } | |
683 | ||
684 | /* Free LBAs allocated to freed chunks */ | |
685 | mutex_lock(&blka->mutex); | |
686 | for (i = delta - 1; i >= 0; i--) { | |
687 | /* Mask the higher 48 bits before shifting, even though | |
688 | * it is a noop | |
689 | */ | |
690 | aun = (lxt_old[my_new_size + i].rlba_base & SISL_ASTATUS_MASK); | |
691 | aun = (aun >> MC_CHUNK_SHIFT); | |
692 | if (needs_ws) | |
693 | write_same16(sdev, aun, MC_CHUNK_SIZE); | |
694 | ba_free(&blka->ba_lun, aun); | |
695 | } | |
696 | mutex_unlock(&blka->mutex); | |
697 | ||
698 | if (needs_ws) { | |
699 | /* Make the context visible again */ | |
700 | mutex_lock(&ctxi->mutex); | |
701 | ctxi->unavail = false; | |
702 | } | |
703 | ||
704 | /* Free old lxt if reallocated */ | |
705 | if (lxt != lxt_old) | |
706 | kfree(lxt_old); | |
707 | *new_size = my_new_size; | |
708 | out: | |
709 | pr_debug("%s: returning rc=%d\n", __func__, rc); | |
710 | return rc; | |
711 | } | |
712 | ||
713 | /** | |
f15fbf8d | 714 | * _cxlflash_vlun_resize() - changes the size of a virtual LUN |
2cb79266 MO |
715 | * @sdev: SCSI device associated with LUN owning virtual LUN. |
716 | * @ctxi: Context owning resources. | |
717 | * @resize: Resize ioctl data structure. | |
718 | * | |
719 | * On successful return, the user is informed of the new size (in blocks) | |
f15fbf8d MO |
720 | * of the virtual LUN in last LBA format. When the size of the virtual |
721 | * LUN is zero, the last LBA is reflected as -1. See comment in the | |
2cb79266 MO |
722 | * prologue for _cxlflash_disk_release() regarding AFU syncs and contexts |
723 | * on the error recovery list. | |
724 | * | |
725 | * Return: 0 on success, -errno on failure | |
726 | */ | |
727 | int _cxlflash_vlun_resize(struct scsi_device *sdev, | |
728 | struct ctx_info *ctxi, | |
729 | struct dk_cxlflash_resize *resize) | |
730 | { | |
731 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
732 | struct llun_info *lli = sdev->hostdata; | |
733 | struct glun_info *gli = lli->parent; | |
734 | struct afu *afu = cfg->afu; | |
735 | bool put_ctx = false; | |
736 | ||
737 | res_hndl_t rhndl = resize->rsrc_handle; | |
738 | u64 new_size; | |
739 | u64 nsectors; | |
740 | u64 ctxid = DECODE_CTXID(resize->context_id), | |
741 | rctxid = resize->context_id; | |
742 | ||
743 | struct sisl_rht_entry *rhte; | |
744 | ||
745 | int rc = 0; | |
746 | ||
747 | /* | |
748 | * The requested size (req_size) is always assumed to be in 4k blocks, | |
749 | * so we have to convert it here from 4k to chunk size. | |
750 | */ | |
751 | nsectors = (resize->req_size * CXLFLASH_BLOCK_SIZE) / gli->blk_len; | |
752 | new_size = DIV_ROUND_UP(nsectors, MC_CHUNK_SIZE); | |
753 | ||
754 | pr_debug("%s: ctxid=%llu rhndl=0x%llx, req_size=0x%llx," | |
755 | "new_size=%llx\n", __func__, ctxid, resize->rsrc_handle, | |
756 | resize->req_size, new_size); | |
757 | ||
758 | if (unlikely(gli->mode != MODE_VIRTUAL)) { | |
759 | pr_debug("%s: LUN mode does not support resize! (%d)\n", | |
760 | __func__, gli->mode); | |
761 | rc = -EINVAL; | |
762 | goto out; | |
763 | ||
764 | } | |
765 | ||
766 | if (!ctxi) { | |
767 | ctxi = get_context(cfg, rctxid, lli, CTX_CTRL_ERR_FALLBACK); | |
768 | if (unlikely(!ctxi)) { | |
769 | pr_debug("%s: Bad context! (%llu)\n", __func__, ctxid); | |
770 | rc = -EINVAL; | |
771 | goto out; | |
772 | } | |
773 | ||
774 | put_ctx = true; | |
775 | } | |
776 | ||
777 | rhte = get_rhte(ctxi, rhndl, lli); | |
778 | if (unlikely(!rhte)) { | |
779 | pr_debug("%s: Bad resource handle! (%u)\n", __func__, rhndl); | |
780 | rc = -EINVAL; | |
781 | goto out; | |
782 | } | |
783 | ||
784 | if (new_size > rhte->lxt_cnt) | |
785 | rc = grow_lxt(afu, sdev, ctxid, rhndl, rhte, &new_size); | |
786 | else if (new_size < rhte->lxt_cnt) | |
787 | rc = shrink_lxt(afu, sdev, rhndl, rhte, ctxi, &new_size); | |
788 | ||
789 | resize->hdr.return_flags = 0; | |
790 | resize->last_lba = (new_size * MC_CHUNK_SIZE * gli->blk_len); | |
791 | resize->last_lba /= CXLFLASH_BLOCK_SIZE; | |
792 | resize->last_lba--; | |
793 | ||
794 | out: | |
795 | if (put_ctx) | |
796 | put_context(ctxi); | |
797 | pr_debug("%s: resized to %lld returning rc=%d\n", | |
798 | __func__, resize->last_lba, rc); | |
799 | return rc; | |
800 | } | |
801 | ||
802 | int cxlflash_vlun_resize(struct scsi_device *sdev, | |
803 | struct dk_cxlflash_resize *resize) | |
804 | { | |
805 | return _cxlflash_vlun_resize(sdev, NULL, resize); | |
806 | } | |
807 | ||
808 | /** | |
809 | * cxlflash_restore_luntable() - Restore LUN table to prior state | |
810 | * @cfg: Internal structure associated with the host. | |
811 | */ | |
812 | void cxlflash_restore_luntable(struct cxlflash_cfg *cfg) | |
813 | { | |
814 | struct llun_info *lli, *temp; | |
815 | u32 chan; | |
816 | u32 lind; | |
817 | struct afu *afu = cfg->afu; | |
1786f4a0 | 818 | struct sisl_global_map __iomem *agm = &afu->afu_map->global; |
2cb79266 MO |
819 | |
820 | mutex_lock(&global.mutex); | |
821 | ||
822 | list_for_each_entry_safe(lli, temp, &cfg->lluns, list) { | |
823 | if (!lli->in_table) | |
824 | continue; | |
825 | ||
826 | lind = lli->lun_index; | |
827 | ||
828 | if (lli->port_sel == BOTH_PORTS) { | |
829 | writeq_be(lli->lun_id[0], &agm->fc_port[0][lind]); | |
830 | writeq_be(lli->lun_id[1], &agm->fc_port[1][lind]); | |
831 | pr_debug("%s: Virtual LUN on slot %d id0=%llx, " | |
832 | "id1=%llx\n", __func__, lind, | |
833 | lli->lun_id[0], lli->lun_id[1]); | |
834 | } else { | |
835 | chan = PORT2CHAN(lli->port_sel); | |
836 | writeq_be(lli->lun_id[chan], &agm->fc_port[chan][lind]); | |
837 | pr_debug("%s: Virtual LUN on slot %d chan=%d, " | |
838 | "id=%llx\n", __func__, lind, chan, | |
839 | lli->lun_id[chan]); | |
840 | } | |
841 | } | |
842 | ||
843 | mutex_unlock(&global.mutex); | |
844 | } | |
845 | ||
846 | /** | |
847 | * init_luntable() - write an entry in the LUN table | |
848 | * @cfg: Internal structure associated with the host. | |
849 | * @lli: Per adapter LUN information structure. | |
850 | * | |
851 | * On successful return, a LUN table entry is created. | |
852 | * At the top for LUNs visible on both ports. | |
853 | * At the bottom for LUNs visible only on one port. | |
854 | * | |
855 | * Return: 0 on success, -errno on failure | |
856 | */ | |
857 | static int init_luntable(struct cxlflash_cfg *cfg, struct llun_info *lli) | |
858 | { | |
859 | u32 chan; | |
860 | u32 lind; | |
861 | int rc = 0; | |
862 | struct afu *afu = cfg->afu; | |
1786f4a0 | 863 | struct sisl_global_map __iomem *agm = &afu->afu_map->global; |
2cb79266 MO |
864 | |
865 | mutex_lock(&global.mutex); | |
866 | ||
867 | if (lli->in_table) | |
868 | goto out; | |
869 | ||
870 | if (lli->port_sel == BOTH_PORTS) { | |
871 | /* | |
872 | * If this LUN is visible from both ports, we will put | |
873 | * it in the top half of the LUN table. | |
874 | */ | |
875 | if ((cfg->promote_lun_index == cfg->last_lun_index[0]) || | |
876 | (cfg->promote_lun_index == cfg->last_lun_index[1])) { | |
877 | rc = -ENOSPC; | |
878 | goto out; | |
879 | } | |
880 | ||
881 | lind = lli->lun_index = cfg->promote_lun_index; | |
882 | writeq_be(lli->lun_id[0], &agm->fc_port[0][lind]); | |
883 | writeq_be(lli->lun_id[1], &agm->fc_port[1][lind]); | |
884 | cfg->promote_lun_index++; | |
885 | pr_debug("%s: Virtual LUN on slot %d id0=%llx, id1=%llx\n", | |
886 | __func__, lind, lli->lun_id[0], lli->lun_id[1]); | |
887 | } else { | |
888 | /* | |
889 | * If this LUN is visible only from one port, we will put | |
890 | * it in the bottom half of the LUN table. | |
891 | */ | |
892 | chan = PORT2CHAN(lli->port_sel); | |
893 | if (cfg->promote_lun_index == cfg->last_lun_index[chan]) { | |
894 | rc = -ENOSPC; | |
895 | goto out; | |
896 | } | |
897 | ||
898 | lind = lli->lun_index = cfg->last_lun_index[chan]; | |
899 | writeq_be(lli->lun_id[chan], &agm->fc_port[chan][lind]); | |
900 | cfg->last_lun_index[chan]--; | |
901 | pr_debug("%s: Virtual LUN on slot %d chan=%d, id=%llx\n", | |
902 | __func__, lind, chan, lli->lun_id[chan]); | |
903 | } | |
904 | ||
905 | lli->in_table = true; | |
906 | out: | |
907 | mutex_unlock(&global.mutex); | |
908 | pr_debug("%s: returning rc=%d\n", __func__, rc); | |
909 | return rc; | |
910 | } | |
911 | ||
912 | /** | |
913 | * cxlflash_disk_virtual_open() - open a virtual disk of specified size | |
914 | * @sdev: SCSI device associated with LUN owning virtual LUN. | |
915 | * @arg: UVirtual ioctl data structure. | |
916 | * | |
917 | * On successful return, the user is informed of the resource handle | |
f15fbf8d MO |
918 | * to be used to identify the virtual LUN and the size (in blocks) of |
919 | * the virtual LUN in last LBA format. When the size of the virtual LUN | |
2cb79266 MO |
920 | * is zero, the last LBA is reflected as -1. |
921 | * | |
922 | * Return: 0 on success, -errno on failure | |
923 | */ | |
924 | int cxlflash_disk_virtual_open(struct scsi_device *sdev, void *arg) | |
925 | { | |
926 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
927 | struct device *dev = &cfg->dev->dev; | |
928 | struct llun_info *lli = sdev->hostdata; | |
929 | struct glun_info *gli = lli->parent; | |
930 | ||
931 | struct dk_cxlflash_uvirtual *virt = (struct dk_cxlflash_uvirtual *)arg; | |
932 | struct dk_cxlflash_resize resize; | |
933 | ||
934 | u64 ctxid = DECODE_CTXID(virt->context_id), | |
935 | rctxid = virt->context_id; | |
936 | u64 lun_size = virt->lun_size; | |
937 | u64 last_lba = 0; | |
938 | u64 rsrc_handle = -1; | |
939 | ||
940 | int rc = 0; | |
941 | ||
942 | struct ctx_info *ctxi = NULL; | |
943 | struct sisl_rht_entry *rhte = NULL; | |
944 | ||
945 | pr_debug("%s: ctxid=%llu ls=0x%llx\n", __func__, ctxid, lun_size); | |
946 | ||
2843fdbd | 947 | /* Setup the LUNs block allocator on first call */ |
2cb79266 MO |
948 | mutex_lock(&gli->mutex); |
949 | if (gli->mode == MODE_NONE) { | |
2cb79266 MO |
950 | rc = init_vlun(lli); |
951 | if (rc) { | |
952 | dev_err(dev, "%s: call to init_vlun failed rc=%d!\n", | |
953 | __func__, rc); | |
954 | rc = -ENOMEM; | |
955 | goto err0; | |
956 | } | |
957 | } | |
958 | ||
959 | rc = cxlflash_lun_attach(gli, MODE_VIRTUAL, true); | |
960 | if (unlikely(rc)) { | |
961 | dev_err(dev, "%s: Failed to attach to LUN! (VIRTUAL)\n", | |
962 | __func__); | |
963 | goto err0; | |
964 | } | |
965 | mutex_unlock(&gli->mutex); | |
966 | ||
2843fdbd MO |
967 | rc = init_luntable(cfg, lli); |
968 | if (rc) { | |
969 | dev_err(dev, "%s: call to init_luntable failed rc=%d!\n", | |
970 | __func__, rc); | |
971 | goto err1; | |
972 | } | |
973 | ||
2cb79266 MO |
974 | ctxi = get_context(cfg, rctxid, lli, 0); |
975 | if (unlikely(!ctxi)) { | |
976 | dev_err(dev, "%s: Bad context! (%llu)\n", __func__, ctxid); | |
977 | rc = -EINVAL; | |
978 | goto err1; | |
979 | } | |
980 | ||
981 | rhte = rhte_checkout(ctxi, lli); | |
982 | if (unlikely(!rhte)) { | |
983 | dev_err(dev, "%s: too many opens for this context\n", __func__); | |
984 | rc = -EMFILE; /* too many opens */ | |
985 | goto err1; | |
986 | } | |
987 | ||
988 | rsrc_handle = (rhte - ctxi->rht_start); | |
989 | ||
990 | /* Populate RHT format 0 */ | |
991 | rhte->nmask = MC_RHT_NMASK; | |
992 | rhte->fp = SISL_RHT_FP(0U, ctxi->rht_perms); | |
993 | ||
994 | /* Resize even if requested size is 0 */ | |
995 | marshal_virt_to_resize(virt, &resize); | |
996 | resize.rsrc_handle = rsrc_handle; | |
997 | rc = _cxlflash_vlun_resize(sdev, ctxi, &resize); | |
998 | if (rc) { | |
999 | dev_err(dev, "%s: resize failed rc %d\n", __func__, rc); | |
1000 | goto err2; | |
1001 | } | |
1002 | last_lba = resize.last_lba; | |
1003 | ||
1004 | if (virt->hdr.flags & DK_CXLFLASH_UVIRTUAL_NEED_WRITE_SAME) | |
1005 | ctxi->rht_needs_ws[rsrc_handle] = true; | |
1006 | ||
1007 | virt->hdr.return_flags = 0; | |
1008 | virt->last_lba = last_lba; | |
1009 | virt->rsrc_handle = rsrc_handle; | |
1010 | ||
d5e26bb1 MO |
1011 | if (lli->port_sel == BOTH_PORTS) |
1012 | virt->hdr.return_flags |= DK_CXLFLASH_ALL_PORTS_ACTIVE; | |
2cb79266 MO |
1013 | out: |
1014 | if (likely(ctxi)) | |
1015 | put_context(ctxi); | |
1016 | pr_debug("%s: returning handle 0x%llx rc=%d llba %lld\n", | |
1017 | __func__, rsrc_handle, rc, last_lba); | |
1018 | return rc; | |
1019 | ||
1020 | err2: | |
1021 | rhte_checkin(ctxi, rhte); | |
1022 | err1: | |
1023 | cxlflash_lun_detach(gli); | |
1024 | goto out; | |
1025 | err0: | |
1026 | /* Special common cleanup prior to successful LUN attach */ | |
1027 | cxlflash_ba_terminate(&gli->blka.ba_lun); | |
1028 | mutex_unlock(&gli->mutex); | |
1029 | goto out; | |
1030 | } | |
1031 | ||
1032 | /** | |
1033 | * clone_lxt() - copies translation tables from source to destination RHTE | |
1034 | * @afu: AFU associated with the host. | |
1035 | * @blka: Block allocator associated with LUN. | |
1036 | * @ctxid: Context ID of context owning the RHTE. | |
1037 | * @rhndl: Resource handle associated with the RHTE. | |
1038 | * @rhte: Destination resource handle entry (RHTE). | |
1039 | * @rhte_src: Source resource handle entry (RHTE). | |
1040 | * | |
1041 | * Return: 0 on success, -errno on failure | |
1042 | */ | |
1043 | static int clone_lxt(struct afu *afu, | |
1044 | struct blka *blka, | |
1045 | ctx_hndl_t ctxid, | |
1046 | res_hndl_t rhndl, | |
1047 | struct sisl_rht_entry *rhte, | |
1048 | struct sisl_rht_entry *rhte_src) | |
1049 | { | |
1050 | struct sisl_lxt_entry *lxt; | |
1051 | u32 ngrps; | |
1052 | u64 aun; /* chunk# allocated by block allocator */ | |
1053 | int i, j; | |
1054 | ||
1055 | ngrps = LXT_NUM_GROUPS(rhte_src->lxt_cnt); | |
1056 | ||
1057 | if (ngrps) { | |
1058 | /* allocate new LXTs for clone */ | |
1059 | lxt = kzalloc((sizeof(*lxt) * LXT_GROUP_SIZE * ngrps), | |
1060 | GFP_KERNEL); | |
1061 | if (unlikely(!lxt)) | |
1062 | return -ENOMEM; | |
1063 | ||
1064 | /* copy over */ | |
1065 | memcpy(lxt, rhte_src->lxt_start, | |
1066 | (sizeof(*lxt) * rhte_src->lxt_cnt)); | |
1067 | ||
1068 | /* clone the LBAs in block allocator via ref_cnt */ | |
1069 | mutex_lock(&blka->mutex); | |
1070 | for (i = 0; i < rhte_src->lxt_cnt; i++) { | |
1071 | aun = (lxt[i].rlba_base >> MC_CHUNK_SHIFT); | |
1072 | if (ba_clone(&blka->ba_lun, aun) == -1ULL) { | |
1073 | /* free the clones already made */ | |
1074 | for (j = 0; j < i; j++) { | |
1075 | aun = (lxt[j].rlba_base >> | |
1076 | MC_CHUNK_SHIFT); | |
1077 | ba_free(&blka->ba_lun, aun); | |
1078 | } | |
1079 | ||
1080 | mutex_unlock(&blka->mutex); | |
1081 | kfree(lxt); | |
1082 | return -EIO; | |
1083 | } | |
1084 | } | |
1085 | mutex_unlock(&blka->mutex); | |
1086 | } else { | |
1087 | lxt = NULL; | |
1088 | } | |
1089 | ||
1090 | /* | |
1091 | * The following sequence is prescribed in the SISlite spec | |
1092 | * for syncing up with the AFU when adding LXT entries. | |
1093 | */ | |
1094 | dma_wmb(); /* Make LXT updates are visible */ | |
1095 | ||
1096 | rhte->lxt_start = lxt; | |
1097 | dma_wmb(); /* Make RHT entry's LXT table update visible */ | |
1098 | ||
1099 | rhte->lxt_cnt = rhte_src->lxt_cnt; | |
1100 | dma_wmb(); /* Make RHT entry's LXT table size update visible */ | |
1101 | ||
1102 | cxlflash_afu_sync(afu, ctxid, rhndl, AFU_LW_SYNC); | |
1103 | ||
1104 | pr_debug("%s: returning\n", __func__); | |
1105 | return 0; | |
1106 | } | |
1107 | ||
1108 | /** | |
1109 | * cxlflash_disk_clone() - clone a context by making snapshot of another | |
1110 | * @sdev: SCSI device associated with LUN owning virtual LUN. | |
1111 | * @clone: Clone ioctl data structure. | |
1112 | * | |
1113 | * This routine effectively performs cxlflash_disk_open operation for each | |
1114 | * in-use virtual resource in the source context. Note that the destination | |
1115 | * context must be in pristine state and cannot have any resource handles | |
1116 | * open at the time of the clone. | |
1117 | * | |
1118 | * Return: 0 on success, -errno on failure | |
1119 | */ | |
1120 | int cxlflash_disk_clone(struct scsi_device *sdev, | |
1121 | struct dk_cxlflash_clone *clone) | |
1122 | { | |
1123 | struct cxlflash_cfg *cfg = (struct cxlflash_cfg *)sdev->host->hostdata; | |
1124 | struct llun_info *lli = sdev->hostdata; | |
1125 | struct glun_info *gli = lli->parent; | |
1126 | struct blka *blka = &gli->blka; | |
1127 | struct afu *afu = cfg->afu; | |
1128 | struct dk_cxlflash_release release = { { 0 }, 0 }; | |
1129 | ||
1130 | struct ctx_info *ctxi_src = NULL, | |
1131 | *ctxi_dst = NULL; | |
1132 | struct lun_access *lun_access_src, *lun_access_dst; | |
1133 | u32 perms; | |
1134 | u64 ctxid_src = DECODE_CTXID(clone->context_id_src), | |
1135 | ctxid_dst = DECODE_CTXID(clone->context_id_dst), | |
1136 | rctxid_src = clone->context_id_src, | |
1137 | rctxid_dst = clone->context_id_dst; | |
1138 | int adap_fd_src = clone->adap_fd_src; | |
1139 | int i, j; | |
1140 | int rc = 0; | |
1141 | bool found; | |
1142 | LIST_HEAD(sidecar); | |
1143 | ||
1144 | pr_debug("%s: ctxid_src=%llu ctxid_dst=%llu adap_fd_src=%d\n", | |
1145 | __func__, ctxid_src, ctxid_dst, adap_fd_src); | |
1146 | ||
1147 | /* Do not clone yourself */ | |
1148 | if (unlikely(rctxid_src == rctxid_dst)) { | |
1149 | rc = -EINVAL; | |
1150 | goto out; | |
1151 | } | |
1152 | ||
1153 | if (unlikely(gli->mode != MODE_VIRTUAL)) { | |
1154 | rc = -EINVAL; | |
1155 | pr_debug("%s: Clone not supported on physical LUNs! (%d)\n", | |
1156 | __func__, gli->mode); | |
1157 | goto out; | |
1158 | } | |
1159 | ||
1160 | ctxi_src = get_context(cfg, rctxid_src, lli, CTX_CTRL_CLONE); | |
1161 | ctxi_dst = get_context(cfg, rctxid_dst, lli, 0); | |
1162 | if (unlikely(!ctxi_src || !ctxi_dst)) { | |
1163 | pr_debug("%s: Bad context! (%llu,%llu)\n", __func__, | |
1164 | ctxid_src, ctxid_dst); | |
1165 | rc = -EINVAL; | |
1166 | goto out; | |
1167 | } | |
1168 | ||
1169 | if (unlikely(adap_fd_src != ctxi_src->lfd)) { | |
1170 | pr_debug("%s: Invalid source adapter fd! (%d)\n", | |
1171 | __func__, adap_fd_src); | |
1172 | rc = -EINVAL; | |
1173 | goto out; | |
1174 | } | |
1175 | ||
1176 | /* Verify there is no open resource handle in the destination context */ | |
1177 | for (i = 0; i < MAX_RHT_PER_CONTEXT; i++) | |
1178 | if (ctxi_dst->rht_start[i].nmask != 0) { | |
1179 | rc = -EINVAL; | |
1180 | goto out; | |
1181 | } | |
1182 | ||
1183 | /* Clone LUN access list */ | |
1184 | list_for_each_entry(lun_access_src, &ctxi_src->luns, list) { | |
1185 | found = false; | |
1186 | list_for_each_entry(lun_access_dst, &ctxi_dst->luns, list) | |
1187 | if (lun_access_dst->sdev == lun_access_src->sdev) { | |
1188 | found = true; | |
1189 | break; | |
1190 | } | |
1191 | ||
1192 | if (!found) { | |
1193 | lun_access_dst = kzalloc(sizeof(*lun_access_dst), | |
1194 | GFP_KERNEL); | |
1195 | if (unlikely(!lun_access_dst)) { | |
1196 | pr_err("%s: Unable to allocate lun_access!\n", | |
1197 | __func__); | |
1198 | rc = -ENOMEM; | |
1199 | goto out; | |
1200 | } | |
1201 | ||
1202 | *lun_access_dst = *lun_access_src; | |
1203 | list_add(&lun_access_dst->list, &sidecar); | |
1204 | } | |
1205 | } | |
1206 | ||
1207 | if (unlikely(!ctxi_src->rht_out)) { | |
1208 | pr_debug("%s: Nothing to clone!\n", __func__); | |
1209 | goto out_success; | |
1210 | } | |
1211 | ||
1212 | /* User specified permission on attach */ | |
1213 | perms = ctxi_dst->rht_perms; | |
1214 | ||
1215 | /* | |
1216 | * Copy over checked-out RHT (and their associated LXT) entries by | |
1217 | * hand, stopping after we've copied all outstanding entries and | |
1218 | * cleaning up if the clone fails. | |
1219 | * | |
1220 | * Note: This loop is equivalent to performing cxlflash_disk_open and | |
1221 | * cxlflash_vlun_resize. As such, LUN accounting needs to be taken into | |
1222 | * account by attaching after each successful RHT entry clone. In the | |
1223 | * event that a clone failure is experienced, the LUN detach is handled | |
1224 | * via the cleanup performed by _cxlflash_disk_release. | |
1225 | */ | |
1226 | for (i = 0; i < MAX_RHT_PER_CONTEXT; i++) { | |
1227 | if (ctxi_src->rht_out == ctxi_dst->rht_out) | |
1228 | break; | |
1229 | if (ctxi_src->rht_start[i].nmask == 0) | |
1230 | continue; | |
1231 | ||
1232 | /* Consume a destination RHT entry */ | |
1233 | ctxi_dst->rht_out++; | |
1234 | ctxi_dst->rht_start[i].nmask = ctxi_src->rht_start[i].nmask; | |
1235 | ctxi_dst->rht_start[i].fp = | |
1236 | SISL_RHT_FP_CLONE(ctxi_src->rht_start[i].fp, perms); | |
1237 | ctxi_dst->rht_lun[i] = ctxi_src->rht_lun[i]; | |
1238 | ||
1239 | rc = clone_lxt(afu, blka, ctxid_dst, i, | |
1240 | &ctxi_dst->rht_start[i], | |
1241 | &ctxi_src->rht_start[i]); | |
1242 | if (rc) { | |
1243 | marshal_clone_to_rele(clone, &release); | |
1244 | for (j = 0; j < i; j++) { | |
1245 | release.rsrc_handle = j; | |
1246 | _cxlflash_disk_release(sdev, ctxi_dst, | |
1247 | &release); | |
1248 | } | |
1249 | ||
1250 | /* Put back the one we failed on */ | |
1251 | rhte_checkin(ctxi_dst, &ctxi_dst->rht_start[i]); | |
1252 | goto err; | |
1253 | } | |
1254 | ||
1255 | cxlflash_lun_attach(gli, gli->mode, false); | |
1256 | } | |
1257 | ||
1258 | out_success: | |
1259 | list_splice(&sidecar, &ctxi_dst->luns); | |
1260 | sys_close(adap_fd_src); | |
1261 | ||
1262 | /* fall through */ | |
1263 | out: | |
1264 | if (ctxi_src) | |
1265 | put_context(ctxi_src); | |
1266 | if (ctxi_dst) | |
1267 | put_context(ctxi_dst); | |
1268 | pr_debug("%s: returning rc=%d\n", __func__, rc); | |
1269 | return rc; | |
1270 | ||
1271 | err: | |
1272 | list_for_each_entry_safe(lun_access_src, lun_access_dst, &sidecar, list) | |
1273 | kfree(lun_access_src); | |
1274 | goto out; | |
1275 | } |