2 * CXL Flash Device Driver
7 * Copyright (C) 2015 IBM Corporation
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.
15 #include <linux/syscalls.h>
17 #include <asm/unaligned.h>
18 #include <asm/bitsperlong.h>
20 #include <scsi/scsi_cmnd.h>
21 #include <scsi/scsi_host.h>
22 #include <uapi/scsi/cxlflash_ioctl.h>
27 #include "superpipe.h"
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.
34 static void marshal_virt_to_resize(struct dk_cxlflash_uvirtual *virt,
35 struct dk_cxlflash_resize *resize)
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;
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.
49 static void marshal_clone_to_rele(struct dk_cxlflash_clone *clone,
50 struct dk_cxlflash_release *release)
52 release->hdr = clone->hdr;
53 release->context_id = clone->context_id_dst;
57 * ba_init() - initializes a block allocator
58 * @ba_lun: Block allocator to initialize.
60 * Return: 0 on success, -errno on failure
62 static int ba_init(struct ba_lun *ba_lun)
64 struct ba_lun_info *bali = NULL;
65 int lun_size_au = 0, i = 0;
66 int last_word_underflow = 0;
69 pr_debug("%s: Initializing LUN: lun_id=%016llx "
70 "ba_lun->lsize=%lx ba_lun->au_size=%lX\n",
71 __func__, ba_lun->lun_id, ba_lun->lsize, ba_lun->au_size);
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__);
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 lun_id=%016llx\n",
84 __func__, ba_lun->lun_id);
88 bali->total_aus = lun_size_au;
89 bali->lun_bmap_size = lun_size_au / BITS_PER_LONG;
91 if (lun_size_au % BITS_PER_LONG)
92 bali->lun_bmap_size++;
94 /* Allocate bitmap space */
95 bali->lun_alloc_map = kzalloc((bali->lun_bmap_size * sizeof(u64)),
97 if (unlikely(!bali->lun_alloc_map)) {
98 pr_err("%s: Failed to allocate lun allocation map: "
99 "lun_id=%016llx\n", __func__, ba_lun->lun_id);
104 /* Initialize the bit map size and set all bits to '1' */
105 bali->free_aun_cnt = lun_size_au;
107 for (i = 0; i < bali->lun_bmap_size; i++)
108 bali->lun_alloc_map[i] = 0xFFFFFFFFFFFFFFFFULL;
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);
118 clear_bit(i, (ulong *)lam);
121 /* Initialize high elevator index, low/curr already at 0 from kzalloc */
122 bali->free_high_idx = bali->lun_bmap_size;
124 /* Allocate clone map */
125 bali->aun_clone_map = kzalloc((bali->total_aus * sizeof(u8)),
127 if (unlikely(!bali->aun_clone_map)) {
128 pr_err("%s: Failed to allocate clone map: lun_id=%016llx\n",
129 __func__, ba_lun->lun_id);
130 kfree(bali->lun_alloc_map);
135 /* Pass the allocated LUN info as a handle to the user */
136 ba_lun->ba_lun_handle = bali;
138 pr_debug("%s: Successfully initialized the LUN: "
139 "lun_id=%016llx bitmap size=%x, free_aun_cnt=%llx\n",
140 __func__, ba_lun->lun_id, bali->lun_bmap_size,
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.
152 * Return: The bit position within the passed back word, -1 on failure
154 static int find_free_range(u32 low,
156 struct ba_lun_info *bali, int *bit_word)
160 ulong *lam, num_bits;
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);
168 pr_devel("%s: Found free bit %llu in LUN "
169 "map entry %016llx at bitmap index = %d\n",
170 __func__, bit_pos, bali->lun_alloc_map[i], i);
173 bali->free_aun_cnt--;
174 clear_bit(bit_pos, lam);
182 * ba_alloc() - allocates a block from the block allocator
183 * @ba_lun: Block allocator from which to allocate a block.
185 * Return: The allocated block, -1 on failure
187 static u64 ba_alloc(struct ba_lun *ba_lun)
191 struct ba_lun_info *bali = NULL;
193 bali = ba_lun->ba_lun_handle;
195 pr_debug("%s: Received block allocation request: "
196 "lun_id=%016llx free_aun_cnt=%llx\n",
197 __func__, ba_lun->lun_id, bali->free_aun_cnt);
199 if (bali->free_aun_cnt == 0) {
200 pr_debug("%s: No space left on LUN: lun_id=%016llx\n",
201 __func__, ba_lun->lun_id);
205 /* Search to find a free entry, curr->high then low->curr */
206 bit_pos = find_free_range(bali->free_curr_idx,
207 bali->free_high_idx, bali, &bit_word);
209 bit_pos = find_free_range(bali->free_low_idx,
213 pr_debug("%s: Could not find an allocation unit on LUN:"
214 " lun_id=%016llx\n", __func__, ba_lun->lun_id);
219 /* Update the free_curr_idx */
220 if (bit_pos == HIBIT)
221 bali->free_curr_idx = bit_word + 1;
223 bali->free_curr_idx = bit_word;
225 pr_debug("%s: Allocating AU number=%llx lun_id=%016llx "
226 "free_aun_cnt=%llx\n", __func__,
227 ((bit_word * BITS_PER_LONG) + bit_pos), ba_lun->lun_id,
230 return (u64) ((bit_word * BITS_PER_LONG) + bit_pos);
234 * validate_alloc() - validates the specified block has been allocated
235 * @ba_lun_info: LUN info owning the block allocator.
236 * @aun: Block to validate.
238 * Return: 0 on success, -1 on failure
240 static int validate_alloc(struct ba_lun_info *bali, u64 aun)
242 int idx = 0, bit_pos = 0;
244 idx = aun / BITS_PER_LONG;
245 bit_pos = aun % BITS_PER_LONG;
247 if (test_bit(bit_pos, (ulong *)&bali->lun_alloc_map[idx]))
254 * ba_free() - frees a block from the block allocator
255 * @ba_lun: Block allocator from which to allocate a block.
256 * @to_free: Block to free.
258 * Return: 0 on success, -1 on failure
260 static int ba_free(struct ba_lun *ba_lun, u64 to_free)
262 int idx = 0, bit_pos = 0;
263 struct ba_lun_info *bali = NULL;
265 bali = ba_lun->ba_lun_handle;
267 if (validate_alloc(bali, to_free)) {
268 pr_debug("%s: AUN %llx is not allocated on lun_id=%016llx\n",
269 __func__, to_free, ba_lun->lun_id);
273 pr_debug("%s: Received a request to free AU=%llx lun_id=%016llx "
274 "free_aun_cnt=%llx\n", __func__, to_free, ba_lun->lun_id,
277 if (bali->aun_clone_map[to_free] > 0) {
278 pr_debug("%s: AUN %llx lun_id=%016llx cloned. Clone count=%x\n",
279 __func__, to_free, ba_lun->lun_id,
280 bali->aun_clone_map[to_free]);
281 bali->aun_clone_map[to_free]--;
285 idx = to_free / BITS_PER_LONG;
286 bit_pos = to_free % BITS_PER_LONG;
288 set_bit(bit_pos, (ulong *)&bali->lun_alloc_map[idx]);
289 bali->free_aun_cnt++;
291 if (idx < bali->free_low_idx)
292 bali->free_low_idx = idx;
293 else if (idx > bali->free_high_idx)
294 bali->free_high_idx = idx;
296 pr_debug("%s: Successfully freed AU bit_pos=%x bit map index=%x "
297 "lun_id=%016llx free_aun_cnt=%llx\n", __func__, bit_pos, idx,
298 ba_lun->lun_id, bali->free_aun_cnt);
304 * ba_clone() - Clone a chunk of the block allocation table
305 * @ba_lun: Block allocator from which to allocate a block.
306 * @to_free: Block to free.
308 * Return: 0 on success, -1 on failure
310 static int ba_clone(struct ba_lun *ba_lun, u64 to_clone)
312 struct ba_lun_info *bali = ba_lun->ba_lun_handle;
314 if (validate_alloc(bali, to_clone)) {
315 pr_debug("%s: AUN=%llx not allocated on lun_id=%016llx\n",
316 __func__, to_clone, ba_lun->lun_id);
320 pr_debug("%s: Received a request to clone AUN %llx on lun_id=%016llx\n",
321 __func__, to_clone, ba_lun->lun_id);
323 if (bali->aun_clone_map[to_clone] == MAX_AUN_CLONE_CNT) {
324 pr_debug("%s: AUN %llx on lun_id=%016llx hit max clones already\n",
325 __func__, to_clone, ba_lun->lun_id);
329 bali->aun_clone_map[to_clone]++;
335 * ba_space() - returns the amount of free space left in the block allocator
336 * @ba_lun: Block allocator.
338 * Return: Amount of free space in block allocator
340 static u64 ba_space(struct ba_lun *ba_lun)
342 struct ba_lun_info *bali = ba_lun->ba_lun_handle;
344 return bali->free_aun_cnt;
348 * cxlflash_ba_terminate() - frees resources associated with the block allocator
349 * @ba_lun: Block allocator.
351 * Safe to call in a partially allocated state.
353 void cxlflash_ba_terminate(struct ba_lun *ba_lun)
355 struct ba_lun_info *bali = ba_lun->ba_lun_handle;
358 kfree(bali->aun_clone_map);
359 kfree(bali->lun_alloc_map);
361 ba_lun->ba_lun_handle = NULL;
366 * init_vlun() - initializes a LUN for virtual use
367 * @lun_info: LUN information structure that owns the block allocator.
369 * Return: 0 on success, -errno on failure
371 static int init_vlun(struct llun_info *lli)
374 struct glun_info *gli = lli->parent;
375 struct blka *blka = &gli->blka;
377 memset(blka, 0, sizeof(*blka));
378 mutex_init(&blka->mutex);
380 /* LUN IDs are unique per port, save the index instead */
381 blka->ba_lun.lun_id = lli->lun_index;
382 blka->ba_lun.lsize = gli->max_lba + 1;
383 blka->ba_lun.lba_size = gli->blk_len;
385 blka->ba_lun.au_size = MC_CHUNK_SIZE;
386 blka->nchunk = blka->ba_lun.lsize / MC_CHUNK_SIZE;
388 rc = ba_init(&blka->ba_lun);
390 pr_debug("%s: cannot init block_alloc, rc=%d\n", __func__, rc);
392 pr_debug("%s: returning rc=%d lli=%p\n", __func__, rc, lli);
397 * write_same16() - sends a SCSI WRITE_SAME16 (0) command to specified LUN
398 * @sdev: SCSI device associated with LUN.
399 * @lba: Logical block address to start write same.
400 * @nblks: Number of logical blocks to write same.
402 * The SCSI WRITE_SAME16 can take quite a while to complete. Should an EEH occur
403 * while in scsi_execute(), the EEH handler will attempt to recover. As part of
404 * the recovery, the handler drains all currently running ioctls, waiting until
405 * they have completed before proceeding with a reset. As this routine is used
406 * on the ioctl path, this can create a condition where the EEH handler becomes
407 * stuck, infinitely waiting for this ioctl thread. To avoid this behavior,
408 * temporarily unmark this thread as an ioctl thread by releasing the ioctl read
409 * semaphore. This will allow the EEH handler to proceed with a recovery while
410 * this thread is still running. Once the scsi_execute() returns, reacquire the
411 * ioctl read semaphore and check the adapter state in case it changed while
412 * inside of scsi_execute(). The state check will wait if the adapter is still
413 * being recovered or return a failure if the recovery failed. In the event that
414 * the adapter reset failed, simply return the failure as the ioctl would be
415 * unable to continue.
417 * Note that the above puts a requirement on this routine to only be called on
420 * Return: 0 on success, -errno on failure
422 static int write_same16(struct scsi_device *sdev,
428 u8 *sense_buf = NULL;
433 struct cxlflash_cfg *cfg = shost_priv(sdev->host);
434 struct device *dev = &cfg->dev->dev;
435 const u32 s = ilog2(sdev->sector_size) - 9;
436 const u32 to = sdev->request_queue->rq_timeout;
437 const u32 ws_limit = blk_queue_get_max_sectors(sdev->request_queue,
438 REQ_OP_WRITE_SAME) >> s;
440 cmd_buf = kzalloc(CMD_BUFSIZE, GFP_KERNEL);
441 scsi_cmd = kzalloc(MAX_COMMAND_SIZE, GFP_KERNEL);
442 sense_buf = kzalloc(SCSI_SENSE_BUFFERSIZE, GFP_KERNEL);
443 if (unlikely(!cmd_buf || !scsi_cmd || !sense_buf)) {
450 scsi_cmd[0] = WRITE_SAME_16;
451 scsi_cmd[1] = cfg->ws_unmap ? 0x8 : 0;
452 put_unaligned_be64(offset, &scsi_cmd[2]);
453 put_unaligned_be32(ws_limit < left ? ws_limit : left,
456 /* Drop the ioctl read semahpore across lengthy call */
457 up_read(&cfg->ioctl_rwsem);
458 result = scsi_execute(sdev, scsi_cmd, DMA_TO_DEVICE, cmd_buf,
459 CMD_BUFSIZE, sense_buf, NULL, to,
460 CMD_RETRIES, 0, 0, NULL);
461 down_read(&cfg->ioctl_rwsem);
462 rc = check_state(cfg);
464 dev_err(dev, "%s: Failed state result=%08x\n",
471 dev_err_ratelimited(dev, "%s: command failed for "
472 "offset=%lld result=%08x\n",
473 __func__, offset, result);
485 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
490 * grow_lxt() - expands the translation table associated with the specified RHTE
491 * @afu: AFU associated with the host.
492 * @sdev: SCSI device associated with LUN.
493 * @ctxid: Context ID of context owning the RHTE.
494 * @rhndl: Resource handle associated with the RHTE.
495 * @rhte: Resource handle entry (RHTE).
496 * @new_size: Number of translation entries associated with RHTE.
498 * By design, this routine employs a 'best attempt' allocation and will
499 * truncate the requested size down if there is not sufficient space in
500 * the block allocator to satisfy the request but there does exist some
501 * amount of space. The user is made aware of this by returning the size
504 * Return: 0 on success, -errno on failure
506 static int grow_lxt(struct afu *afu,
507 struct scsi_device *sdev,
510 struct sisl_rht_entry *rhte,
513 struct cxlflash_cfg *cfg = shost_priv(sdev->host);
514 struct device *dev = &cfg->dev->dev;
515 struct sisl_lxt_entry *lxt = NULL, *lxt_old = NULL;
516 struct llun_info *lli = sdev->hostdata;
517 struct glun_info *gli = lli->parent;
518 struct blka *blka = &gli->blka;
520 u32 ngrps, ngrps_old;
521 u64 aun; /* chunk# allocated by block allocator */
522 u64 delta = *new_size - rhte->lxt_cnt;
527 * Check what is available in the block allocator before re-allocating
528 * LXT array. This is done up front under the mutex which must not be
529 * released until after allocation is complete.
531 mutex_lock(&blka->mutex);
532 av_size = ba_space(&blka->ba_lun);
533 if (unlikely(av_size <= 0)) {
534 dev_dbg(dev, "%s: ba_space error av_size=%d\n",
536 mutex_unlock(&blka->mutex);
544 lxt_old = rhte->lxt_start;
545 ngrps_old = LXT_NUM_GROUPS(rhte->lxt_cnt);
546 ngrps = LXT_NUM_GROUPS(rhte->lxt_cnt + delta);
548 if (ngrps != ngrps_old) {
549 /* reallocate to fit new size */
550 lxt = kzalloc((sizeof(*lxt) * LXT_GROUP_SIZE * ngrps),
552 if (unlikely(!lxt)) {
553 mutex_unlock(&blka->mutex);
558 /* copy over all old entries */
559 memcpy(lxt, lxt_old, (sizeof(*lxt) * rhte->lxt_cnt));
563 /* nothing can fail from now on */
564 my_new_size = rhte->lxt_cnt + delta;
566 /* add new entries to the end */
567 for (i = rhte->lxt_cnt; i < my_new_size; i++) {
569 * Due to the earlier check of available space, ba_alloc
570 * cannot fail here. If it did due to internal error,
571 * leave a rlba_base of -1u which will likely be a
572 * invalid LUN (too large).
574 aun = ba_alloc(&blka->ba_lun);
575 if ((aun == -1ULL) || (aun >= blka->nchunk))
576 dev_dbg(dev, "%s: ba_alloc error allocated chunk=%llu "
577 "max=%llu\n", __func__, aun, blka->nchunk - 1);
579 /* select both ports, use r/w perms from RHT */
580 lxt[i].rlba_base = ((aun << MC_CHUNK_SHIFT) |
581 (lli->lun_index << LXT_LUNIDX_SHIFT) |
582 (RHT_PERM_RW << LXT_PERM_SHIFT |
586 mutex_unlock(&blka->mutex);
589 * The following sequence is prescribed in the SISlite spec
590 * for syncing up with the AFU when adding LXT entries.
592 dma_wmb(); /* Make LXT updates are visible */
594 rhte->lxt_start = lxt;
595 dma_wmb(); /* Make RHT entry's LXT table update visible */
597 rhte->lxt_cnt = my_new_size;
598 dma_wmb(); /* Make RHT entry's LXT table size update visible */
600 rc = cxlflash_afu_sync(afu, ctxid, rhndl, AFU_LW_SYNC);
604 /* free old lxt if reallocated */
607 *new_size = my_new_size;
609 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
614 * shrink_lxt() - reduces translation table associated with the specified RHTE
615 * @afu: AFU associated with the host.
616 * @sdev: SCSI device associated with LUN.
617 * @rhndl: Resource handle associated with the RHTE.
618 * @rhte: Resource handle entry (RHTE).
619 * @ctxi: Context owning resources.
620 * @new_size: Number of translation entries associated with RHTE.
622 * Return: 0 on success, -errno on failure
624 static int shrink_lxt(struct afu *afu,
625 struct scsi_device *sdev,
627 struct sisl_rht_entry *rhte,
628 struct ctx_info *ctxi,
631 struct cxlflash_cfg *cfg = shost_priv(sdev->host);
632 struct device *dev = &cfg->dev->dev;
633 struct sisl_lxt_entry *lxt, *lxt_old;
634 struct llun_info *lli = sdev->hostdata;
635 struct glun_info *gli = lli->parent;
636 struct blka *blka = &gli->blka;
637 ctx_hndl_t ctxid = DECODE_CTXID(ctxi->ctxid);
638 bool needs_ws = ctxi->rht_needs_ws[rhndl];
639 bool needs_sync = !ctxi->err_recovery_active;
640 u32 ngrps, ngrps_old;
641 u64 aun; /* chunk# allocated by block allocator */
642 u64 delta = rhte->lxt_cnt - *new_size;
646 lxt_old = rhte->lxt_start;
647 ngrps_old = LXT_NUM_GROUPS(rhte->lxt_cnt);
648 ngrps = LXT_NUM_GROUPS(rhte->lxt_cnt - delta);
650 if (ngrps != ngrps_old) {
651 /* Reallocate to fit new size unless new size is 0 */
653 lxt = kzalloc((sizeof(*lxt) * LXT_GROUP_SIZE * ngrps),
655 if (unlikely(!lxt)) {
660 /* Copy over old entries that will remain */
662 (sizeof(*lxt) * (rhte->lxt_cnt - delta)));
668 /* Nothing can fail from now on */
669 my_new_size = rhte->lxt_cnt - delta;
672 * The following sequence is prescribed in the SISlite spec
673 * for syncing up with the AFU when removing LXT entries.
675 rhte->lxt_cnt = my_new_size;
676 dma_wmb(); /* Make RHT entry's LXT table size update visible */
678 rhte->lxt_start = lxt;
679 dma_wmb(); /* Make RHT entry's LXT table update visible */
682 rc = cxlflash_afu_sync(afu, ctxid, rhndl, AFU_HW_SYNC);
689 * Mark the context as unavailable, so that we can release
692 ctxi->unavail = true;
693 mutex_unlock(&ctxi->mutex);
696 /* Free LBAs allocated to freed chunks */
697 mutex_lock(&blka->mutex);
698 for (i = delta - 1; i >= 0; i--) {
699 aun = lxt_old[my_new_size + i].rlba_base >> MC_CHUNK_SHIFT;
701 write_same16(sdev, aun, MC_CHUNK_SIZE);
702 ba_free(&blka->ba_lun, aun);
704 mutex_unlock(&blka->mutex);
707 /* Make the context visible again */
708 mutex_lock(&ctxi->mutex);
709 ctxi->unavail = false;
712 /* Free old lxt if reallocated */
715 *new_size = my_new_size;
717 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
722 * _cxlflash_vlun_resize() - changes the size of a virtual LUN
723 * @sdev: SCSI device associated with LUN owning virtual LUN.
724 * @ctxi: Context owning resources.
725 * @resize: Resize ioctl data structure.
727 * On successful return, the user is informed of the new size (in blocks)
728 * of the virtual LUN in last LBA format. When the size of the virtual
729 * LUN is zero, the last LBA is reflected as -1. See comment in the
730 * prologue for _cxlflash_disk_release() regarding AFU syncs and contexts
731 * on the error recovery list.
733 * Return: 0 on success, -errno on failure
735 int _cxlflash_vlun_resize(struct scsi_device *sdev,
736 struct ctx_info *ctxi,
737 struct dk_cxlflash_resize *resize)
739 struct cxlflash_cfg *cfg = shost_priv(sdev->host);
740 struct device *dev = &cfg->dev->dev;
741 struct llun_info *lli = sdev->hostdata;
742 struct glun_info *gli = lli->parent;
743 struct afu *afu = cfg->afu;
744 bool put_ctx = false;
746 res_hndl_t rhndl = resize->rsrc_handle;
749 u64 ctxid = DECODE_CTXID(resize->context_id),
750 rctxid = resize->context_id;
752 struct sisl_rht_entry *rhte;
757 * The requested size (req_size) is always assumed to be in 4k blocks,
758 * so we have to convert it here from 4k to chunk size.
760 nsectors = (resize->req_size * CXLFLASH_BLOCK_SIZE) / gli->blk_len;
761 new_size = DIV_ROUND_UP(nsectors, MC_CHUNK_SIZE);
763 dev_dbg(dev, "%s: ctxid=%llu rhndl=%llu req_size=%llu new_size=%llu\n",
764 __func__, ctxid, resize->rsrc_handle, resize->req_size,
767 if (unlikely(gli->mode != MODE_VIRTUAL)) {
768 dev_dbg(dev, "%s: LUN mode does not support resize mode=%d\n",
769 __func__, gli->mode);
776 ctxi = get_context(cfg, rctxid, lli, CTX_CTRL_ERR_FALLBACK);
777 if (unlikely(!ctxi)) {
778 dev_dbg(dev, "%s: Bad context ctxid=%llu\n",
787 rhte = get_rhte(ctxi, rhndl, lli);
788 if (unlikely(!rhte)) {
789 dev_dbg(dev, "%s: Bad resource handle rhndl=%u\n",
795 if (new_size > rhte->lxt_cnt)
796 rc = grow_lxt(afu, sdev, ctxid, rhndl, rhte, &new_size);
797 else if (new_size < rhte->lxt_cnt)
798 rc = shrink_lxt(afu, sdev, rhndl, rhte, ctxi, &new_size);
801 * Rare case where there is already sufficient space, just
802 * need to perform a translation sync with the AFU. This
803 * scenario likely follows a previous sync failure during
804 * a resize operation. Accordingly, perform the heavyweight
805 * form of translation sync as it is unknown which type of
806 * resize failed previously.
808 rc = cxlflash_afu_sync(afu, ctxid, rhndl, AFU_HW_SYNC);
815 resize->hdr.return_flags = 0;
816 resize->last_lba = (new_size * MC_CHUNK_SIZE * gli->blk_len);
817 resize->last_lba /= CXLFLASH_BLOCK_SIZE;
823 dev_dbg(dev, "%s: resized to %llu returning rc=%d\n",
824 __func__, resize->last_lba, rc);
828 int cxlflash_vlun_resize(struct scsi_device *sdev,
829 struct dk_cxlflash_resize *resize)
831 return _cxlflash_vlun_resize(sdev, NULL, resize);
835 * cxlflash_restore_luntable() - Restore LUN table to prior state
836 * @cfg: Internal structure associated with the host.
838 void cxlflash_restore_luntable(struct cxlflash_cfg *cfg)
840 struct llun_info *lli, *temp;
843 struct device *dev = &cfg->dev->dev;
844 __be64 __iomem *fc_port_luns;
846 mutex_lock(&global.mutex);
848 list_for_each_entry_safe(lli, temp, &cfg->lluns, list) {
852 lind = lli->lun_index;
853 dev_dbg(dev, "%s: Virtual LUNs on slot %d:\n", __func__, lind);
855 for (k = 0; k < cfg->num_fc_ports; k++)
856 if (lli->port_sel & (1 << k)) {
857 fc_port_luns = get_fc_port_luns(cfg, k);
858 writeq_be(lli->lun_id[k], &fc_port_luns[lind]);
859 dev_dbg(dev, "\t%d=%llx\n", k, lli->lun_id[k]);
863 mutex_unlock(&global.mutex);
867 * get_num_ports() - compute number of ports from port selection mask
868 * @psm: Port selection mask.
870 * Return: Population count of port selection mask
872 static inline u8 get_num_ports(u32 psm)
874 static const u8 bits[16] = { 0, 1, 1, 2, 1, 2, 2, 3,
875 1, 2, 2, 3, 2, 3, 3, 4 };
877 return bits[psm & 0xf];
881 * init_luntable() - write an entry in the LUN table
882 * @cfg: Internal structure associated with the host.
883 * @lli: Per adapter LUN information structure.
885 * On successful return, a LUN table entry is created:
886 * - at the top for LUNs visible on multiple ports.
887 * - at the bottom for LUNs visible only on one port.
889 * Return: 0 on success, -errno on failure
891 static int init_luntable(struct cxlflash_cfg *cfg, struct llun_info *lli)
898 struct device *dev = &cfg->dev->dev;
899 __be64 __iomem *fc_port_luns;
901 mutex_lock(&global.mutex);
906 nports = get_num_ports(lli->port_sel);
907 if (nports == 0 || nports > cfg->num_fc_ports) {
908 WARN(1, "Unsupported port configuration nports=%u", nports);
915 * When LUN is visible from multiple ports, we will put
916 * it in the top half of the LUN table.
918 for (k = 0; k < cfg->num_fc_ports; k++) {
919 if (!(lli->port_sel & (1 << k)))
922 if (cfg->promote_lun_index == cfg->last_lun_index[k]) {
928 lind = lli->lun_index = cfg->promote_lun_index;
929 dev_dbg(dev, "%s: Virtual LUNs on slot %d:\n", __func__, lind);
931 for (k = 0; k < cfg->num_fc_ports; k++) {
932 if (!(lli->port_sel & (1 << k)))
935 fc_port_luns = get_fc_port_luns(cfg, k);
936 writeq_be(lli->lun_id[k], &fc_port_luns[lind]);
937 dev_dbg(dev, "\t%d=%llx\n", k, lli->lun_id[k]);
940 cfg->promote_lun_index++;
943 * When LUN is visible only from one port, we will put
944 * it in the bottom half of the LUN table.
946 chan = PORTMASK2CHAN(lli->port_sel);
947 if (cfg->promote_lun_index == cfg->last_lun_index[chan]) {
952 lind = lli->lun_index = cfg->last_lun_index[chan];
953 fc_port_luns = get_fc_port_luns(cfg, chan);
954 writeq_be(lli->lun_id[chan], &fc_port_luns[lind]);
955 cfg->last_lun_index[chan]--;
956 dev_dbg(dev, "%s: Virtual LUNs on slot %d:\n\t%d=%llx\n",
957 __func__, lind, chan, lli->lun_id[chan]);
960 lli->in_table = true;
962 mutex_unlock(&global.mutex);
963 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
968 * cxlflash_disk_virtual_open() - open a virtual disk of specified size
969 * @sdev: SCSI device associated with LUN owning virtual LUN.
970 * @arg: UVirtual ioctl data structure.
972 * On successful return, the user is informed of the resource handle
973 * to be used to identify the virtual LUN and the size (in blocks) of
974 * the virtual LUN in last LBA format. When the size of the virtual LUN
975 * is zero, the last LBA is reflected as -1.
977 * Return: 0 on success, -errno on failure
979 int cxlflash_disk_virtual_open(struct scsi_device *sdev, void *arg)
981 struct cxlflash_cfg *cfg = shost_priv(sdev->host);
982 struct device *dev = &cfg->dev->dev;
983 struct llun_info *lli = sdev->hostdata;
984 struct glun_info *gli = lli->parent;
986 struct dk_cxlflash_uvirtual *virt = (struct dk_cxlflash_uvirtual *)arg;
987 struct dk_cxlflash_resize resize;
989 u64 ctxid = DECODE_CTXID(virt->context_id),
990 rctxid = virt->context_id;
991 u64 lun_size = virt->lun_size;
993 u64 rsrc_handle = -1;
997 struct ctx_info *ctxi = NULL;
998 struct sisl_rht_entry *rhte = NULL;
1000 dev_dbg(dev, "%s: ctxid=%llu ls=%llu\n", __func__, ctxid, lun_size);
1002 /* Setup the LUNs block allocator on first call */
1003 mutex_lock(&gli->mutex);
1004 if (gli->mode == MODE_NONE) {
1005 rc = init_vlun(lli);
1007 dev_err(dev, "%s: init_vlun failed rc=%d\n",
1014 rc = cxlflash_lun_attach(gli, MODE_VIRTUAL, true);
1016 dev_err(dev, "%s: Failed attach to LUN (VIRTUAL)\n", __func__);
1019 mutex_unlock(&gli->mutex);
1021 rc = init_luntable(cfg, lli);
1023 dev_err(dev, "%s: init_luntable failed rc=%d\n", __func__, rc);
1027 ctxi = get_context(cfg, rctxid, lli, 0);
1028 if (unlikely(!ctxi)) {
1029 dev_err(dev, "%s: Bad context ctxid=%llu\n", __func__, ctxid);
1034 rhte = rhte_checkout(ctxi, lli);
1035 if (unlikely(!rhte)) {
1036 dev_err(dev, "%s: too many opens ctxid=%llu\n",
1038 rc = -EMFILE; /* too many opens */
1042 rsrc_handle = (rhte - ctxi->rht_start);
1044 /* Populate RHT format 0 */
1045 rhte->nmask = MC_RHT_NMASK;
1046 rhte->fp = SISL_RHT_FP(0U, ctxi->rht_perms);
1048 /* Resize even if requested size is 0 */
1049 marshal_virt_to_resize(virt, &resize);
1050 resize.rsrc_handle = rsrc_handle;
1051 rc = _cxlflash_vlun_resize(sdev, ctxi, &resize);
1053 dev_err(dev, "%s: resize failed rc=%d\n", __func__, rc);
1056 last_lba = resize.last_lba;
1058 if (virt->hdr.flags & DK_CXLFLASH_UVIRTUAL_NEED_WRITE_SAME)
1059 ctxi->rht_needs_ws[rsrc_handle] = true;
1061 virt->hdr.return_flags = 0;
1062 virt->last_lba = last_lba;
1063 virt->rsrc_handle = rsrc_handle;
1065 if (get_num_ports(lli->port_sel) > 1)
1066 virt->hdr.return_flags |= DK_CXLFLASH_ALL_PORTS_ACTIVE;
1070 dev_dbg(dev, "%s: returning handle=%llu rc=%d llba=%llu\n",
1071 __func__, rsrc_handle, rc, last_lba);
1075 rhte_checkin(ctxi, rhte);
1077 cxlflash_lun_detach(gli);
1080 /* Special common cleanup prior to successful LUN attach */
1081 cxlflash_ba_terminate(&gli->blka.ba_lun);
1082 mutex_unlock(&gli->mutex);
1087 * clone_lxt() - copies translation tables from source to destination RHTE
1088 * @afu: AFU associated with the host.
1089 * @blka: Block allocator associated with LUN.
1090 * @ctxid: Context ID of context owning the RHTE.
1091 * @rhndl: Resource handle associated with the RHTE.
1092 * @rhte: Destination resource handle entry (RHTE).
1093 * @rhte_src: Source resource handle entry (RHTE).
1095 * Return: 0 on success, -errno on failure
1097 static int clone_lxt(struct afu *afu,
1101 struct sisl_rht_entry *rhte,
1102 struct sisl_rht_entry *rhte_src)
1104 struct cxlflash_cfg *cfg = afu->parent;
1105 struct device *dev = &cfg->dev->dev;
1106 struct sisl_lxt_entry *lxt = NULL;
1107 bool locked = false;
1109 u64 aun; /* chunk# allocated by block allocator */
1114 ngrps = LXT_NUM_GROUPS(rhte_src->lxt_cnt);
1117 /* allocate new LXTs for clone */
1118 lxt = kzalloc((sizeof(*lxt) * LXT_GROUP_SIZE * ngrps),
1120 if (unlikely(!lxt)) {
1126 memcpy(lxt, rhte_src->lxt_start,
1127 (sizeof(*lxt) * rhte_src->lxt_cnt));
1129 /* clone the LBAs in block allocator via ref_cnt, note that the
1130 * block allocator mutex must be held until it is established
1131 * that this routine will complete without the need for a
1134 mutex_lock(&blka->mutex);
1136 for (i = 0; i < rhte_src->lxt_cnt; i++) {
1137 aun = (lxt[i].rlba_base >> MC_CHUNK_SHIFT);
1138 if (ba_clone(&blka->ba_lun, aun) == -1ULL) {
1146 * The following sequence is prescribed in the SISlite spec
1147 * for syncing up with the AFU when adding LXT entries.
1149 dma_wmb(); /* Make LXT updates are visible */
1151 rhte->lxt_start = lxt;
1152 dma_wmb(); /* Make RHT entry's LXT table update visible */
1154 rhte->lxt_cnt = rhte_src->lxt_cnt;
1155 dma_wmb(); /* Make RHT entry's LXT table size update visible */
1157 rc = cxlflash_afu_sync(afu, ctxid, rhndl, AFU_LW_SYNC);
1165 mutex_unlock(&blka->mutex);
1166 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1169 /* Reset the RHTE */
1172 rhte->lxt_start = NULL;
1175 /* free the clones already made */
1176 for (j = 0; j < i; j++) {
1177 aun = (lxt[j].rlba_base >> MC_CHUNK_SHIFT);
1178 ba_free(&blka->ba_lun, aun);
1185 * cxlflash_disk_clone() - clone a context by making snapshot of another
1186 * @sdev: SCSI device associated with LUN owning virtual LUN.
1187 * @clone: Clone ioctl data structure.
1189 * This routine effectively performs cxlflash_disk_open operation for each
1190 * in-use virtual resource in the source context. Note that the destination
1191 * context must be in pristine state and cannot have any resource handles
1192 * open at the time of the clone.
1194 * Return: 0 on success, -errno on failure
1196 int cxlflash_disk_clone(struct scsi_device *sdev,
1197 struct dk_cxlflash_clone *clone)
1199 struct cxlflash_cfg *cfg = shost_priv(sdev->host);
1200 struct device *dev = &cfg->dev->dev;
1201 struct llun_info *lli = sdev->hostdata;
1202 struct glun_info *gli = lli->parent;
1203 struct blka *blka = &gli->blka;
1204 struct afu *afu = cfg->afu;
1205 struct dk_cxlflash_release release = { { 0 }, 0 };
1207 struct ctx_info *ctxi_src = NULL,
1209 struct lun_access *lun_access_src, *lun_access_dst;
1211 u64 ctxid_src = DECODE_CTXID(clone->context_id_src),
1212 ctxid_dst = DECODE_CTXID(clone->context_id_dst),
1213 rctxid_src = clone->context_id_src,
1214 rctxid_dst = clone->context_id_dst;
1220 dev_dbg(dev, "%s: ctxid_src=%llu ctxid_dst=%llu\n",
1221 __func__, ctxid_src, ctxid_dst);
1223 /* Do not clone yourself */
1224 if (unlikely(rctxid_src == rctxid_dst)) {
1229 if (unlikely(gli->mode != MODE_VIRTUAL)) {
1231 dev_dbg(dev, "%s: Only supported on virtual LUNs mode=%u\n",
1232 __func__, gli->mode);
1236 ctxi_src = get_context(cfg, rctxid_src, lli, CTX_CTRL_CLONE);
1237 ctxi_dst = get_context(cfg, rctxid_dst, lli, 0);
1238 if (unlikely(!ctxi_src || !ctxi_dst)) {
1239 dev_dbg(dev, "%s: Bad context ctxid_src=%llu ctxid_dst=%llu\n",
1240 __func__, ctxid_src, ctxid_dst);
1245 /* Verify there is no open resource handle in the destination context */
1246 for (i = 0; i < MAX_RHT_PER_CONTEXT; i++)
1247 if (ctxi_dst->rht_start[i].nmask != 0) {
1252 /* Clone LUN access list */
1253 list_for_each_entry(lun_access_src, &ctxi_src->luns, list) {
1255 list_for_each_entry(lun_access_dst, &ctxi_dst->luns, list)
1256 if (lun_access_dst->sdev == lun_access_src->sdev) {
1262 lun_access_dst = kzalloc(sizeof(*lun_access_dst),
1264 if (unlikely(!lun_access_dst)) {
1265 dev_err(dev, "%s: lun_access allocation fail\n",
1271 *lun_access_dst = *lun_access_src;
1272 list_add(&lun_access_dst->list, &sidecar);
1276 if (unlikely(!ctxi_src->rht_out)) {
1277 dev_dbg(dev, "%s: Nothing to clone\n", __func__);
1281 /* User specified permission on attach */
1282 perms = ctxi_dst->rht_perms;
1285 * Copy over checked-out RHT (and their associated LXT) entries by
1286 * hand, stopping after we've copied all outstanding entries and
1287 * cleaning up if the clone fails.
1289 * Note: This loop is equivalent to performing cxlflash_disk_open and
1290 * cxlflash_vlun_resize. As such, LUN accounting needs to be taken into
1291 * account by attaching after each successful RHT entry clone. In the
1292 * event that a clone failure is experienced, the LUN detach is handled
1293 * via the cleanup performed by _cxlflash_disk_release.
1295 for (i = 0; i < MAX_RHT_PER_CONTEXT; i++) {
1296 if (ctxi_src->rht_out == ctxi_dst->rht_out)
1298 if (ctxi_src->rht_start[i].nmask == 0)
1301 /* Consume a destination RHT entry */
1302 ctxi_dst->rht_out++;
1303 ctxi_dst->rht_start[i].nmask = ctxi_src->rht_start[i].nmask;
1304 ctxi_dst->rht_start[i].fp =
1305 SISL_RHT_FP_CLONE(ctxi_src->rht_start[i].fp, perms);
1306 ctxi_dst->rht_lun[i] = ctxi_src->rht_lun[i];
1308 rc = clone_lxt(afu, blka, ctxid_dst, i,
1309 &ctxi_dst->rht_start[i],
1310 &ctxi_src->rht_start[i]);
1312 marshal_clone_to_rele(clone, &release);
1313 for (j = 0; j < i; j++) {
1314 release.rsrc_handle = j;
1315 _cxlflash_disk_release(sdev, ctxi_dst,
1319 /* Put back the one we failed on */
1320 rhte_checkin(ctxi_dst, &ctxi_dst->rht_start[i]);
1324 cxlflash_lun_attach(gli, gli->mode, false);
1328 list_splice(&sidecar, &ctxi_dst->luns);
1333 put_context(ctxi_src);
1335 put_context(ctxi_dst);
1336 dev_dbg(dev, "%s: returning rc=%d\n", __func__, rc);
1340 list_for_each_entry_safe(lun_access_src, lun_access_dst, &sidecar, list)
1341 kfree(lun_access_src);