Lines Matching +full:no +full:- +full:memory +full:- +full:wc
1 // SPDX-License-Identifier: GPL-2.0
8 #include <linux/device-mapper.h>
13 #include <linux/dm-io.h>
14 #include <linux/dm-kcopyd.h>
30 #define MAX_AGE_UNSPECIFIED -1UL
101 #define WC_MODE_PMEM(wc) ((wc)->pmem_mode) argument
102 #define WC_MODE_FUA(wc) ((wc)->writeback_fua) argument
104 #define WC_MODE_PMEM(wc) false argument
105 #define WC_MODE_FUA(wc) false argument
107 #define WC_MODE_SORT_FREELIST(wc) (!WC_MODE_PMEM(wc)) argument
206 struct dm_writecache *wc; member
215 struct dm_writecache *wc; member
224 static void wc_lock(struct dm_writecache *wc) in wc_lock() argument
226 mutex_lock(&wc->lock); in wc_lock()
229 static void wc_unlock(struct dm_writecache *wc) in wc_unlock() argument
231 mutex_unlock(&wc->lock); in wc_unlock()
235 static int persistent_memory_claim(struct dm_writecache *wc) in persistent_memory_claim() argument
245 wc->memory_vmapped = false; in persistent_memory_claim()
247 s = wc->memory_map_size; in persistent_memory_claim()
250 r = -EINVAL; in persistent_memory_claim()
254 r = -EOVERFLOW; in persistent_memory_claim()
258 offset = get_start_sect(wc->ssd_dev->bdev); in persistent_memory_claim()
259 if (offset & (PAGE_SIZE / 512 - 1)) { in persistent_memory_claim()
260 r = -EINVAL; in persistent_memory_claim()
263 offset >>= PAGE_SHIFT - 9; in persistent_memory_claim()
267 da = dax_direct_access(wc->ssd_dev->dax_dev, offset, p, &wc->memory_map, &pfn); in persistent_memory_claim()
269 wc->memory_map = NULL; in persistent_memory_claim()
274 wc->memory_map = NULL; in persistent_memory_claim()
275 r = -EOPNOTSUPP; in persistent_memory_claim()
280 wc->memory_map = NULL; in persistent_memory_claim()
283 r = -ENOMEM; in persistent_memory_claim()
289 daa = dax_direct_access(wc->ssd_dev->dax_dev, offset + i, p - i, in persistent_memory_claim()
292 r = daa ? daa : -EINVAL; in persistent_memory_claim()
296 r = -EOPNOTSUPP; in persistent_memory_claim()
299 while (daa-- && i < p) { in persistent_memory_claim()
306 wc->memory_map = vmap(pages, p, VM_MAP, PAGE_KERNEL); in persistent_memory_claim()
307 if (!wc->memory_map) { in persistent_memory_claim()
308 r = -ENOMEM; in persistent_memory_claim()
312 wc->memory_vmapped = true; in persistent_memory_claim()
317 wc->memory_map += (size_t)wc->start_sector << SECTOR_SHIFT; in persistent_memory_claim()
318 wc->memory_map_size -= (size_t)wc->start_sector << SECTOR_SHIFT; in persistent_memory_claim()
329 static int persistent_memory_claim(struct dm_writecache *wc) in persistent_memory_claim() argument
331 return -EOPNOTSUPP; in persistent_memory_claim()
335 static void persistent_memory_release(struct dm_writecache *wc) in persistent_memory_release() argument
337 if (wc->memory_vmapped) in persistent_memory_release()
338 vunmap(wc->memory_map - ((size_t)wc->start_sector << SECTOR_SHIFT)); in persistent_memory_release()
351 return (unsigned long)addr & (PAGE_SIZE - 1); in persistent_memory_page_offset()
366 static struct wc_memory_superblock *sb(struct dm_writecache *wc) in sb() argument
368 return wc->memory_map; in sb()
371 static struct wc_memory_entry *memory_entry(struct dm_writecache *wc, struct wc_entry *e) in memory_entry() argument
373 return &sb(wc)->entries[e->index]; in memory_entry()
376 static void *memory_data(struct dm_writecache *wc, struct wc_entry *e) in memory_data() argument
378 return (char *)wc->block_start + (e->index << wc->block_size_bits); in memory_data()
381 static sector_t cache_sector(struct dm_writecache *wc, struct wc_entry *e) in cache_sector() argument
383 return wc->start_sector + wc->metadata_sectors + in cache_sector()
384 ((sector_t)e->index << (wc->block_size_bits - SECTOR_SHIFT)); in cache_sector()
387 static uint64_t read_original_sector(struct dm_writecache *wc, struct wc_entry *e) in read_original_sector() argument
390 return e->original_sector; in read_original_sector()
392 return le64_to_cpu(memory_entry(wc, e)->original_sector); in read_original_sector()
396 static uint64_t read_seq_count(struct dm_writecache *wc, struct wc_entry *e) in read_seq_count() argument
399 return e->seq_count; in read_seq_count()
401 return le64_to_cpu(memory_entry(wc, e)->seq_count); in read_seq_count()
405 static void clear_seq_count(struct dm_writecache *wc, struct wc_entry *e) in clear_seq_count() argument
408 e->seq_count = -1; in clear_seq_count()
410 pmem_assign(memory_entry(wc, e)->seq_count, cpu_to_le64(-1)); in clear_seq_count()
413 static void write_original_sector_seq_count(struct dm_writecache *wc, struct wc_entry *e, in write_original_sector_seq_count() argument
418 e->original_sector = original_sector; in write_original_sector_seq_count()
419 e->seq_count = seq_count; in write_original_sector_seq_count()
423 pmem_assign(*memory_entry(wc, e), me); in write_original_sector_seq_count()
426 #define writecache_error(wc, err, msg, arg...) \ argument
428 if (!cmpxchg(&(wc)->error, 0, err)) \
430 wake_up(&(wc)->freelist_wait); \
433 #define writecache_has_error(wc) (unlikely(READ_ONCE((wc)->error))) argument
435 static void writecache_flush_all_metadata(struct dm_writecache *wc) in writecache_flush_all_metadata() argument
437 if (!WC_MODE_PMEM(wc)) in writecache_flush_all_metadata()
438 memset(wc->dirty_bitmap, -1, wc->dirty_bitmap_size); in writecache_flush_all_metadata()
441 static void writecache_flush_region(struct dm_writecache *wc, void *ptr, size_t size) in writecache_flush_region() argument
443 if (!WC_MODE_PMEM(wc)) in writecache_flush_region()
444 __set_bit(((char *)ptr - (char *)wc->memory_map) / BITMAP_GRANULARITY, in writecache_flush_region()
445 wc->dirty_bitmap); in writecache_flush_region()
448 static void writecache_disk_flush(struct dm_writecache *wc, struct dm_dev *dev);
451 struct dm_writecache *wc; member
461 writecache_error(endio->wc, -EIO, "error writing metadata"); in writecache_notify_io()
462 BUG_ON(atomic_read(&endio->count) <= 0); in writecache_notify_io()
463 if (atomic_dec_and_test(&endio->count)) in writecache_notify_io()
464 complete(&endio->c); in writecache_notify_io()
467 static void writecache_wait_for_ios(struct dm_writecache *wc, int direction) in writecache_wait_for_ios() argument
469 wait_event(wc->bio_in_progress_wait[direction], in writecache_wait_for_ios()
470 !atomic_read(&wc->bio_in_progress[direction])); in writecache_wait_for_ios()
473 static void ssd_commit_flushed(struct dm_writecache *wc, bool wait_for_ios) in ssd_commit_flushed() argument
478 wc, in ssd_commit_flushed()
482 unsigned bitmap_bits = wc->dirty_bitmap_size * 8; in ssd_commit_flushed()
487 i = find_next_bit(wc->dirty_bitmap, bitmap_bits, i); in ssd_commit_flushed()
490 j = find_next_zero_bit(wc->dirty_bitmap, bitmap_bits, i); in ssd_commit_flushed()
492 region.bdev = wc->ssd_dev->bdev; in ssd_commit_flushed()
494 region.count = (sector_t)(j - i) * (BITMAP_GRANULARITY >> SECTOR_SHIFT); in ssd_commit_flushed()
496 if (unlikely(region.sector >= wc->metadata_sectors)) in ssd_commit_flushed()
498 if (unlikely(region.sector + region.count > wc->metadata_sectors)) in ssd_commit_flushed()
499 region.count = wc->metadata_sectors - region.sector; in ssd_commit_flushed()
501 region.sector += wc->start_sector; in ssd_commit_flushed()
506 req.mem.ptr.vma = (char *)wc->memory_map + (size_t)i * BITMAP_GRANULARITY; in ssd_commit_flushed()
507 req.client = wc->dm_io; in ssd_commit_flushed()
511 /* writing via async dm-io (implied by notify.fn above) won't return an error */ in ssd_commit_flushed()
520 writecache_wait_for_ios(wc, WRITE); in ssd_commit_flushed()
522 writecache_disk_flush(wc, wc->ssd_dev); in ssd_commit_flushed()
524 memset(wc->dirty_bitmap, 0, wc->dirty_bitmap_size); in ssd_commit_flushed()
527 static void ssd_commit_superblock(struct dm_writecache *wc) in ssd_commit_superblock() argument
533 region.bdev = wc->ssd_dev->bdev; in ssd_commit_superblock()
535 region.count = max(4096U, wc->block_size) >> SECTOR_SHIFT; in ssd_commit_superblock()
537 if (unlikely(region.sector + region.count > wc->metadata_sectors)) in ssd_commit_superblock()
538 region.count = wc->metadata_sectors - region.sector; in ssd_commit_superblock()
540 region.sector += wc->start_sector; in ssd_commit_superblock()
545 req.mem.ptr.vma = (char *)wc->memory_map; in ssd_commit_superblock()
546 req.client = wc->dm_io; in ssd_commit_superblock()
552 writecache_error(wc, r, "error writing superblock"); in ssd_commit_superblock()
555 static void writecache_commit_flushed(struct dm_writecache *wc, bool wait_for_ios) in writecache_commit_flushed() argument
557 if (WC_MODE_PMEM(wc)) in writecache_commit_flushed()
560 ssd_commit_flushed(wc, wait_for_ios); in writecache_commit_flushed()
563 static void writecache_disk_flush(struct dm_writecache *wc, struct dm_dev *dev) in writecache_disk_flush() argument
569 region.bdev = dev->bdev; in writecache_disk_flush()
576 req.client = wc->dm_io; in writecache_disk_flush()
581 writecache_error(wc, r, "error flushing metadata: %d", r); in writecache_disk_flush()
587 static struct wc_entry *writecache_find_entry(struct dm_writecache *wc, in writecache_find_entry() argument
591 struct rb_node *node = wc->tree.rb_node; in writecache_find_entry()
598 if (read_original_sector(wc, e) == block) in writecache_find_entry()
601 node = (read_original_sector(wc, e) >= block ? in writecache_find_entry()
602 e->rb_node.rb_left : e->rb_node.rb_right); in writecache_find_entry()
606 if (read_original_sector(wc, e) >= block) { in writecache_find_entry()
609 node = rb_next(&e->rb_node); in writecache_find_entry()
621 node = rb_prev(&e->rb_node); in writecache_find_entry()
623 node = rb_next(&e->rb_node); in writecache_find_entry()
627 if (read_original_sector(wc, e2) != block) in writecache_find_entry()
633 static void writecache_insert_entry(struct dm_writecache *wc, struct wc_entry *ins) in writecache_insert_entry() argument
636 struct rb_node **node = &wc->tree.rb_node, *parent = NULL; in writecache_insert_entry()
640 parent = &e->rb_node; in writecache_insert_entry()
641 if (read_original_sector(wc, e) > read_original_sector(wc, ins)) in writecache_insert_entry()
642 node = &parent->rb_left; in writecache_insert_entry()
644 node = &parent->rb_right; in writecache_insert_entry()
646 rb_link_node(&ins->rb_node, parent, node); in writecache_insert_entry()
647 rb_insert_color(&ins->rb_node, &wc->tree); in writecache_insert_entry()
648 list_add(&ins->lru, &wc->lru); in writecache_insert_entry()
649 ins->age = jiffies; in writecache_insert_entry()
652 static void writecache_unlink(struct dm_writecache *wc, struct wc_entry *e) in writecache_unlink() argument
654 list_del(&e->lru); in writecache_unlink()
655 rb_erase(&e->rb_node, &wc->tree); in writecache_unlink()
658 static void writecache_add_to_freelist(struct dm_writecache *wc, struct wc_entry *e) in writecache_add_to_freelist() argument
660 if (WC_MODE_SORT_FREELIST(wc)) { in writecache_add_to_freelist()
661 struct rb_node **node = &wc->freetree.rb_node, *parent = NULL; in writecache_add_to_freelist()
663 wc->current_free = e; in writecache_add_to_freelist()
666 if (&e->rb_node < *node) in writecache_add_to_freelist()
667 node = &parent->rb_left; in writecache_add_to_freelist()
669 node = &parent->rb_right; in writecache_add_to_freelist()
671 rb_link_node(&e->rb_node, parent, node); in writecache_add_to_freelist()
672 rb_insert_color(&e->rb_node, &wc->freetree); in writecache_add_to_freelist()
674 list_add_tail(&e->lru, &wc->freelist); in writecache_add_to_freelist()
676 wc->freelist_size++; in writecache_add_to_freelist()
679 static inline void writecache_verify_watermark(struct dm_writecache *wc) in writecache_verify_watermark() argument
681 if (unlikely(wc->freelist_size + wc->writeback_size <= wc->freelist_high_watermark)) in writecache_verify_watermark()
682 queue_work(wc->writeback_wq, &wc->writeback_work); in writecache_verify_watermark()
687 struct dm_writecache *wc = from_timer(wc, t, max_age_timer); in writecache_max_age_timer() local
689 if (!dm_suspended(wc->ti) && !writecache_has_error(wc)) { in writecache_max_age_timer()
690 queue_work(wc->writeback_wq, &wc->writeback_work); in writecache_max_age_timer()
691 mod_timer(&wc->max_age_timer, jiffies + wc->max_age / MAX_AGE_DIV); in writecache_max_age_timer()
695 static struct wc_entry *writecache_pop_from_freelist(struct dm_writecache *wc, sector_t expected_se… in writecache_pop_from_freelist() argument
699 if (WC_MODE_SORT_FREELIST(wc)) { in writecache_pop_from_freelist()
701 if (unlikely(!wc->current_free)) in writecache_pop_from_freelist()
703 e = wc->current_free; in writecache_pop_from_freelist()
704 if (expected_sector != (sector_t)-1 && unlikely(cache_sector(wc, e) != expected_sector)) in writecache_pop_from_freelist()
706 next = rb_next(&e->rb_node); in writecache_pop_from_freelist()
707 rb_erase(&e->rb_node, &wc->freetree); in writecache_pop_from_freelist()
709 next = rb_first(&wc->freetree); in writecache_pop_from_freelist()
710 wc->current_free = next ? container_of(next, struct wc_entry, rb_node) : NULL; in writecache_pop_from_freelist()
712 if (unlikely(list_empty(&wc->freelist))) in writecache_pop_from_freelist()
714 e = container_of(wc->freelist.next, struct wc_entry, lru); in writecache_pop_from_freelist()
715 if (expected_sector != (sector_t)-1 && unlikely(cache_sector(wc, e) != expected_sector)) in writecache_pop_from_freelist()
717 list_del(&e->lru); in writecache_pop_from_freelist()
719 wc->freelist_size--; in writecache_pop_from_freelist()
721 writecache_verify_watermark(wc); in writecache_pop_from_freelist()
726 static void writecache_free_entry(struct dm_writecache *wc, struct wc_entry *e) in writecache_free_entry() argument
728 writecache_unlink(wc, e); in writecache_free_entry()
729 writecache_add_to_freelist(wc, e); in writecache_free_entry()
730 clear_seq_count(wc, e); in writecache_free_entry()
731 writecache_flush_region(wc, memory_entry(wc, e), sizeof(struct wc_memory_entry)); in writecache_free_entry()
732 if (unlikely(waitqueue_active(&wc->freelist_wait))) in writecache_free_entry()
733 wake_up(&wc->freelist_wait); in writecache_free_entry()
736 static void writecache_wait_on_freelist(struct dm_writecache *wc) in writecache_wait_on_freelist() argument
740 prepare_to_wait(&wc->freelist_wait, &wait, TASK_UNINTERRUPTIBLE); in writecache_wait_on_freelist()
741 wc_unlock(wc); in writecache_wait_on_freelist()
743 finish_wait(&wc->freelist_wait, &wait); in writecache_wait_on_freelist()
744 wc_lock(wc); in writecache_wait_on_freelist()
747 static void writecache_poison_lists(struct dm_writecache *wc) in writecache_poison_lists() argument
752 memset(&wc->tree, -1, sizeof wc->tree); in writecache_poison_lists()
753 wc->lru.next = LIST_POISON1; in writecache_poison_lists()
754 wc->lru.prev = LIST_POISON2; in writecache_poison_lists()
755 wc->freelist.next = LIST_POISON1; in writecache_poison_lists()
756 wc->freelist.prev = LIST_POISON2; in writecache_poison_lists()
759 static void writecache_flush_entry(struct dm_writecache *wc, struct wc_entry *e) in writecache_flush_entry() argument
761 writecache_flush_region(wc, memory_entry(wc, e), sizeof(struct wc_memory_entry)); in writecache_flush_entry()
762 if (WC_MODE_PMEM(wc)) in writecache_flush_entry()
763 writecache_flush_region(wc, memory_data(wc, e), wc->block_size); in writecache_flush_entry()
766 static bool writecache_entry_is_committed(struct dm_writecache *wc, struct wc_entry *e) in writecache_entry_is_committed() argument
768 return read_seq_count(wc, e) < wc->seq_count; in writecache_entry_is_committed()
771 static void writecache_flush(struct dm_writecache *wc) in writecache_flush() argument
776 wc->uncommitted_blocks = 0; in writecache_flush()
777 del_timer(&wc->autocommit_timer); in writecache_flush()
779 if (list_empty(&wc->lru)) in writecache_flush()
782 e = container_of(wc->lru.next, struct wc_entry, lru); in writecache_flush()
783 if (writecache_entry_is_committed(wc, e)) { in writecache_flush()
784 if (wc->overwrote_committed) { in writecache_flush()
785 writecache_wait_for_ios(wc, WRITE); in writecache_flush()
786 writecache_disk_flush(wc, wc->ssd_dev); in writecache_flush()
787 wc->overwrote_committed = false; in writecache_flush()
792 writecache_flush_entry(wc, e); in writecache_flush()
793 if (unlikely(e->lru.next == &wc->lru)) in writecache_flush()
795 e2 = container_of(e->lru.next, struct wc_entry, lru); in writecache_flush()
796 if (writecache_entry_is_committed(wc, e2)) in writecache_flush()
801 writecache_commit_flushed(wc, true); in writecache_flush()
803 wc->seq_count++; in writecache_flush()
804 pmem_assign(sb(wc)->seq_count, cpu_to_le64(wc->seq_count)); in writecache_flush()
805 if (WC_MODE_PMEM(wc)) in writecache_flush()
806 writecache_commit_flushed(wc, false); in writecache_flush()
808 ssd_commit_superblock(wc); in writecache_flush()
810 wc->overwrote_committed = false; in writecache_flush()
814 /* Free another committed entry with lower seq-count */ in writecache_flush()
815 struct rb_node *rb_node = rb_prev(&e->rb_node); in writecache_flush()
819 if (read_original_sector(wc, e2) == read_original_sector(wc, e) && in writecache_flush()
820 likely(!e2->write_in_progress)) { in writecache_flush()
821 writecache_free_entry(wc, e2); in writecache_flush()
825 if (unlikely(e->lru.prev == &wc->lru)) in writecache_flush()
827 e = container_of(e->lru.prev, struct wc_entry, lru); in writecache_flush()
832 writecache_commit_flushed(wc, false); in writecache_flush()
837 struct dm_writecache *wc = container_of(work, struct dm_writecache, flush_work); in writecache_flush_work() local
839 wc_lock(wc); in writecache_flush_work()
840 writecache_flush(wc); in writecache_flush_work()
841 wc_unlock(wc); in writecache_flush_work()
846 struct dm_writecache *wc = from_timer(wc, t, autocommit_timer); in writecache_autocommit_timer() local
847 if (!writecache_has_error(wc)) in writecache_autocommit_timer()
848 queue_work(wc->writeback_wq, &wc->flush_work); in writecache_autocommit_timer()
851 static void writecache_schedule_autocommit(struct dm_writecache *wc) in writecache_schedule_autocommit() argument
853 if (!timer_pending(&wc->autocommit_timer)) in writecache_schedule_autocommit()
854 mod_timer(&wc->autocommit_timer, jiffies + wc->autocommit_jiffies); in writecache_schedule_autocommit()
857 static void writecache_discard(struct dm_writecache *wc, sector_t start, sector_t end) in writecache_discard() argument
862 e = writecache_find_entry(wc, start, WFE_RETURN_FOLLOWING | WFE_LOWEST_SEQ); in writecache_discard()
866 while (read_original_sector(wc, e) < end) { in writecache_discard()
867 struct rb_node *node = rb_next(&e->rb_node); in writecache_discard()
869 if (likely(!e->write_in_progress)) { in writecache_discard()
871 if (!WC_MODE_PMEM(wc)) { in writecache_discard()
872 writecache_wait_for_ios(wc, READ); in writecache_discard()
873 writecache_wait_for_ios(wc, WRITE); in writecache_discard()
877 if (!writecache_entry_is_committed(wc, e)) in writecache_discard()
878 wc->uncommitted_blocks--; in writecache_discard()
879 writecache_free_entry(wc, e); in writecache_discard()
889 writecache_commit_flushed(wc, false); in writecache_discard()
892 static bool writecache_wait_for_writeback(struct dm_writecache *wc) in writecache_wait_for_writeback() argument
894 if (wc->writeback_size) { in writecache_wait_for_writeback()
895 writecache_wait_on_freelist(wc); in writecache_wait_for_writeback()
903 struct dm_writecache *wc = ti->private; in writecache_suspend() local
906 del_timer_sync(&wc->autocommit_timer); in writecache_suspend()
907 del_timer_sync(&wc->max_age_timer); in writecache_suspend()
909 wc_lock(wc); in writecache_suspend()
910 writecache_flush(wc); in writecache_suspend()
911 flush_on_suspend = wc->flush_on_suspend; in writecache_suspend()
913 wc->flush_on_suspend = false; in writecache_suspend()
914 wc->writeback_all++; in writecache_suspend()
915 queue_work(wc->writeback_wq, &wc->writeback_work); in writecache_suspend()
917 wc_unlock(wc); in writecache_suspend()
919 drain_workqueue(wc->writeback_wq); in writecache_suspend()
921 wc_lock(wc); in writecache_suspend()
923 wc->writeback_all--; in writecache_suspend()
924 while (writecache_wait_for_writeback(wc)); in writecache_suspend()
926 if (WC_MODE_PMEM(wc)) in writecache_suspend()
927 persistent_memory_flush_cache(wc->memory_map, wc->memory_map_size); in writecache_suspend()
929 writecache_poison_lists(wc); in writecache_suspend()
931 wc_unlock(wc); in writecache_suspend()
934 static int writecache_alloc_entries(struct dm_writecache *wc) in writecache_alloc_entries() argument
938 if (wc->entries) in writecache_alloc_entries()
940 wc->entries = vmalloc(array_size(sizeof(struct wc_entry), wc->n_blocks)); in writecache_alloc_entries()
941 if (!wc->entries) in writecache_alloc_entries()
942 return -ENOMEM; in writecache_alloc_entries()
943 for (b = 0; b < wc->n_blocks; b++) { in writecache_alloc_entries()
944 struct wc_entry *e = &wc->entries[b]; in writecache_alloc_entries()
945 e->index = b; in writecache_alloc_entries()
946 e->write_in_progress = false; in writecache_alloc_entries()
953 static int writecache_read_metadata(struct dm_writecache *wc, sector_t n_sectors) in writecache_read_metadata() argument
958 region.bdev = wc->ssd_dev->bdev; in writecache_read_metadata()
959 region.sector = wc->start_sector; in writecache_read_metadata()
964 req.mem.ptr.vma = (char *)wc->memory_map; in writecache_read_metadata()
965 req.client = wc->dm_io; in writecache_read_metadata()
973 struct dm_writecache *wc = ti->private; in writecache_resume() local
979 wc_lock(wc); in writecache_resume()
981 wc->data_device_sectors = i_size_read(wc->dev->bdev->bd_inode) >> SECTOR_SHIFT; in writecache_resume()
983 if (WC_MODE_PMEM(wc)) { in writecache_resume()
984 persistent_memory_invalidate_cache(wc->memory_map, wc->memory_map_size); in writecache_resume()
986 r = writecache_read_metadata(wc, wc->metadata_sectors); in writecache_resume()
989 writecache_error(wc, r, "unable to read metadata: %d", r); in writecache_resume()
991 memset((char *)wc->memory_map + sb_entries_offset, -1, in writecache_resume()
992 (wc->metadata_sectors << SECTOR_SHIFT) - sb_entries_offset); in writecache_resume()
996 wc->tree = RB_ROOT; in writecache_resume()
997 INIT_LIST_HEAD(&wc->lru); in writecache_resume()
998 if (WC_MODE_SORT_FREELIST(wc)) { in writecache_resume()
999 wc->freetree = RB_ROOT; in writecache_resume()
1000 wc->current_free = NULL; in writecache_resume()
1002 INIT_LIST_HEAD(&wc->freelist); in writecache_resume()
1004 wc->freelist_size = 0; in writecache_resume()
1006 r = copy_mc_to_kernel(&sb_seq_count, &sb(wc)->seq_count, in writecache_resume()
1009 writecache_error(wc, r, "hardware memory error when reading superblock: %d", r); in writecache_resume()
1012 wc->seq_count = le64_to_cpu(sb_seq_count); in writecache_resume()
1015 for (b = 0; b < wc->n_blocks; b++) { in writecache_resume()
1016 struct wc_entry *e = &wc->entries[b]; in writecache_resume()
1018 if (writecache_has_error(wc)) { in writecache_resume()
1019 e->original_sector = -1; in writecache_resume()
1020 e->seq_count = -1; in writecache_resume()
1023 r = copy_mc_to_kernel(&wme, memory_entry(wc, e), in writecache_resume()
1026 writecache_error(wc, r, "hardware memory error when reading metadata entry %lu: %d", in writecache_resume()
1028 e->original_sector = -1; in writecache_resume()
1029 e->seq_count = -1; in writecache_resume()
1031 e->original_sector = le64_to_cpu(wme.original_sector); in writecache_resume()
1032 e->seq_count = le64_to_cpu(wme.seq_count); in writecache_resume()
1037 for (b = 0; b < wc->n_blocks; b++) { in writecache_resume()
1038 struct wc_entry *e = &wc->entries[b]; in writecache_resume()
1039 if (!writecache_entry_is_committed(wc, e)) { in writecache_resume()
1040 if (read_seq_count(wc, e) != -1) { in writecache_resume()
1042 clear_seq_count(wc, e); in writecache_resume()
1045 writecache_add_to_freelist(wc, e); in writecache_resume()
1049 old = writecache_find_entry(wc, read_original_sector(wc, e), 0); in writecache_resume()
1051 writecache_insert_entry(wc, e); in writecache_resume()
1053 if (read_seq_count(wc, old) == read_seq_count(wc, e)) { in writecache_resume()
1054 writecache_error(wc, -EINVAL, in writecache_resume()
1056 (unsigned long long)b, (unsigned long long)read_original_sector(wc, e), in writecache_resume()
1057 (unsigned long long)read_seq_count(wc, e)); in writecache_resume()
1059 if (read_seq_count(wc, old) > read_seq_count(wc, e)) { in writecache_resume()
1062 writecache_free_entry(wc, old); in writecache_resume()
1063 writecache_insert_entry(wc, e); in writecache_resume()
1072 writecache_flush_all_metadata(wc); in writecache_resume()
1073 writecache_commit_flushed(wc, false); in writecache_resume()
1076 writecache_verify_watermark(wc); in writecache_resume()
1078 if (wc->max_age != MAX_AGE_UNSPECIFIED) in writecache_resume()
1079 mod_timer(&wc->max_age_timer, jiffies + wc->max_age / MAX_AGE_DIV); in writecache_resume()
1081 wc_unlock(wc); in writecache_resume()
1084 static int process_flush_mesg(unsigned argc, char **argv, struct dm_writecache *wc) in process_flush_mesg() argument
1087 return -EINVAL; in process_flush_mesg()
1089 wc_lock(wc); in process_flush_mesg()
1090 if (dm_suspended(wc->ti)) { in process_flush_mesg()
1091 wc_unlock(wc); in process_flush_mesg()
1092 return -EBUSY; in process_flush_mesg()
1094 if (writecache_has_error(wc)) { in process_flush_mesg()
1095 wc_unlock(wc); in process_flush_mesg()
1096 return -EIO; in process_flush_mesg()
1099 writecache_flush(wc); in process_flush_mesg()
1100 wc->writeback_all++; in process_flush_mesg()
1101 queue_work(wc->writeback_wq, &wc->writeback_work); in process_flush_mesg()
1102 wc_unlock(wc); in process_flush_mesg()
1104 flush_workqueue(wc->writeback_wq); in process_flush_mesg()
1106 wc_lock(wc); in process_flush_mesg()
1107 wc->writeback_all--; in process_flush_mesg()
1108 if (writecache_has_error(wc)) { in process_flush_mesg()
1109 wc_unlock(wc); in process_flush_mesg()
1110 return -EIO; in process_flush_mesg()
1112 wc_unlock(wc); in process_flush_mesg()
1117 static int process_flush_on_suspend_mesg(unsigned argc, char **argv, struct dm_writecache *wc) in process_flush_on_suspend_mesg() argument
1120 return -EINVAL; in process_flush_on_suspend_mesg()
1122 wc_lock(wc); in process_flush_on_suspend_mesg()
1123 wc->flush_on_suspend = true; in process_flush_on_suspend_mesg()
1124 wc_unlock(wc); in process_flush_on_suspend_mesg()
1129 static void activate_cleaner(struct dm_writecache *wc) in activate_cleaner() argument
1131 wc->flush_on_suspend = true; in activate_cleaner()
1132 wc->cleaner = true; in activate_cleaner()
1133 wc->freelist_high_watermark = wc->n_blocks; in activate_cleaner()
1134 wc->freelist_low_watermark = wc->n_blocks; in activate_cleaner()
1137 static int process_cleaner_mesg(unsigned argc, char **argv, struct dm_writecache *wc) in process_cleaner_mesg() argument
1140 return -EINVAL; in process_cleaner_mesg()
1142 wc_lock(wc); in process_cleaner_mesg()
1143 activate_cleaner(wc); in process_cleaner_mesg()
1144 if (!dm_suspended(wc->ti)) in process_cleaner_mesg()
1145 writecache_verify_watermark(wc); in process_cleaner_mesg()
1146 wc_unlock(wc); in process_cleaner_mesg()
1154 int r = -EINVAL; in writecache_message()
1155 struct dm_writecache *wc = ti->private; in writecache_message() local
1158 r = process_flush_mesg(argc, argv, wc); in writecache_message()
1160 r = process_flush_on_suspend_mesg(argc, argv, wc); in writecache_message()
1162 r = process_cleaner_mesg(argc, argv, wc); in writecache_message()
1173 * non-temporal stores perform better with block size 512 in memcpy_flushcache_optimized()
1179 * We see that movnti performs better for 512-byte blocks, and in memcpy_flushcache_optimized()
1180 * clflushopt performs better for 1024-byte and larger blocks. So, we in memcpy_flushcache_optimized()
1183 * NOTE: this happens to be the case now (with dm-writecache's single in memcpy_flushcache_optimized()
1184 * threaded model) but re-evaluate this once memcpy_flushcache() is in memcpy_flushcache_optimized()
1186 * advantage seen with cache-allocating-writes plus flushing. in memcpy_flushcache_optimized()
1197 size -= 64; in memcpy_flushcache_optimized()
1205 static void bio_copy_block(struct dm_writecache *wc, struct bio *bio, void *data) in bio_copy_block() argument
1211 unsigned remaining_size = wc->block_size; in bio_copy_block()
1214 struct bio_vec bv = bio_iter_iovec(bio, bio->bi_iter); in bio_copy_block()
1225 writecache_error(wc, r, "hardware memory error when reading data: %d", r); in bio_copy_block()
1226 bio->bi_status = BLK_STS_IOERR; in bio_copy_block()
1236 remaining_size -= size; in bio_copy_block()
1243 struct dm_writecache *wc = data; in writecache_flush_thread() local
1248 wc_lock(wc); in writecache_flush_thread()
1249 bio = bio_list_pop(&wc->flush_list); in writecache_flush_thread()
1252 wc_unlock(wc); in writecache_flush_thread()
1264 writecache_discard(wc, bio->bi_iter.bi_sector, in writecache_flush_thread()
1266 wc_unlock(wc); in writecache_flush_thread()
1267 bio_set_dev(bio, wc->dev->bdev); in writecache_flush_thread()
1270 writecache_flush(wc); in writecache_flush_thread()
1271 wc_unlock(wc); in writecache_flush_thread()
1272 if (writecache_has_error(wc)) in writecache_flush_thread()
1273 bio->bi_status = BLK_STS_IOERR; in writecache_flush_thread()
1281 static void writecache_offload_bio(struct dm_writecache *wc, struct bio *bio) in writecache_offload_bio() argument
1283 if (bio_list_empty(&wc->flush_list)) in writecache_offload_bio()
1284 wake_up_process(wc->flush_thread); in writecache_offload_bio()
1285 bio_list_add(&wc->flush_list, bio); in writecache_offload_bio()
1291 struct dm_writecache *wc = ti->private; in writecache_map() local
1293 bio->bi_private = NULL; in writecache_map()
1295 wc_lock(wc); in writecache_map()
1297 if (unlikely(bio->bi_opf & REQ_PREFLUSH)) { in writecache_map()
1298 if (writecache_has_error(wc)) in writecache_map()
1300 if (WC_MODE_PMEM(wc)) { in writecache_map()
1301 writecache_flush(wc); in writecache_map()
1302 if (writecache_has_error(wc)) in writecache_map()
1304 if (unlikely(wc->cleaner)) in writecache_map()
1310 writecache_offload_bio(wc, bio); in writecache_map()
1315 bio->bi_iter.bi_sector = dm_target_offset(ti, bio->bi_iter.bi_sector); in writecache_map()
1317 if (unlikely((((unsigned)bio->bi_iter.bi_sector | bio_sectors(bio)) & in writecache_map()
1318 (wc->block_size / 512 - 1)) != 0)) { in writecache_map()
1320 (unsigned long long)bio->bi_iter.bi_sector, in writecache_map()
1321 bio->bi_iter.bi_size, wc->block_size); in writecache_map()
1326 if (writecache_has_error(wc)) in writecache_map()
1328 if (WC_MODE_PMEM(wc)) { in writecache_map()
1329 writecache_discard(wc, bio->bi_iter.bi_sector, bio_end_sector(bio)); in writecache_map()
1332 writecache_offload_bio(wc, bio); in writecache_map()
1339 e = writecache_find_entry(wc, bio->bi_iter.bi_sector, WFE_RETURN_FOLLOWING); in writecache_map()
1340 if (e && read_original_sector(wc, e) == bio->bi_iter.bi_sector) { in writecache_map()
1341 if (WC_MODE_PMEM(wc)) { in writecache_map()
1342 bio_copy_block(wc, bio, memory_data(wc, e)); in writecache_map()
1343 if (bio->bi_iter.bi_size) in writecache_map()
1347 dm_accept_partial_bio(bio, wc->block_size >> SECTOR_SHIFT); in writecache_map()
1348 bio_set_dev(bio, wc->ssd_dev->bdev); in writecache_map()
1349 bio->bi_iter.bi_sector = cache_sector(wc, e); in writecache_map()
1350 if (!writecache_entry_is_committed(wc, e)) in writecache_map()
1351 writecache_wait_for_ios(wc, WRITE); in writecache_map()
1357 read_original_sector(wc, e) - bio->bi_iter.bi_sector; in writecache_map()
1358 if (next_boundary < bio->bi_iter.bi_size >> SECTOR_SHIFT) { in writecache_map()
1368 if (writecache_has_error(wc)) in writecache_map()
1370 e = writecache_find_entry(wc, bio->bi_iter.bi_sector, 0); in writecache_map()
1372 if (!writecache_entry_is_committed(wc, e)) { in writecache_map()
1376 if (!WC_MODE_PMEM(wc) && !e->write_in_progress) { in writecache_map()
1377 wc->overwrote_committed = true; in writecache_map()
1383 if (unlikely(wc->cleaner)) in writecache_map()
1386 e = writecache_pop_from_freelist(wc, (sector_t)-1); in writecache_map()
1388 if (!WC_MODE_PMEM(wc) && !found_entry) { in writecache_map()
1390 e = writecache_find_entry(wc, bio->bi_iter.bi_sector, WFE_RETURN_FOLLOWING); in writecache_map()
1392 sector_t next_boundary = read_original_sector(wc, e) - bio->bi_iter.bi_sector; in writecache_map()
1394 if (next_boundary < bio->bi_iter.bi_size >> SECTOR_SHIFT) { in writecache_map()
1400 writecache_wait_on_freelist(wc); in writecache_map()
1403 write_original_sector_seq_count(wc, e, bio->bi_iter.bi_sector, wc->seq_count); in writecache_map()
1404 writecache_insert_entry(wc, e); in writecache_map()
1405 wc->uncommitted_blocks++; in writecache_map()
1407 if (WC_MODE_PMEM(wc)) { in writecache_map()
1408 bio_copy_block(wc, bio, memory_data(wc, e)); in writecache_map()
1410 unsigned bio_size = wc->block_size; in writecache_map()
1411 sector_t start_cache_sec = cache_sector(wc, e); in writecache_map()
1414 while (bio_size < bio->bi_iter.bi_size) { in writecache_map()
1416 struct wc_entry *f = writecache_pop_from_freelist(wc, current_cache_sec); in writecache_map()
1419 write_original_sector_seq_count(wc, f, bio->bi_iter.bi_sector + in writecache_map()
1420 (bio_size >> SECTOR_SHIFT), wc->seq_count); in writecache_map()
1421 writecache_insert_entry(wc, f); in writecache_map()
1422 wc->uncommitted_blocks++; in writecache_map()
1425 struct rb_node *next = rb_next(&e->rb_node); in writecache_map()
1431 if (read_original_sector(wc, f) != in writecache_map()
1432 read_original_sector(wc, e) + (wc->block_size >> SECTOR_SHIFT)) in writecache_map()
1434 if (unlikely(f->write_in_progress)) in writecache_map()
1436 if (writecache_entry_is_committed(wc, f)) in writecache_map()
1437 wc->overwrote_committed = true; in writecache_map()
1440 bio_size += wc->block_size; in writecache_map()
1441 current_cache_sec += wc->block_size >> SECTOR_SHIFT; in writecache_map()
1444 bio_set_dev(bio, wc->ssd_dev->bdev); in writecache_map()
1445 bio->bi_iter.bi_sector = start_cache_sec; in writecache_map()
1448 if (unlikely(wc->uncommitted_blocks >= wc->autocommit_blocks)) { in writecache_map()
1449 wc->uncommitted_blocks = 0; in writecache_map()
1450 queue_work(wc->writeback_wq, &wc->flush_work); in writecache_map()
1452 writecache_schedule_autocommit(wc); in writecache_map()
1456 } while (bio->bi_iter.bi_size); in writecache_map()
1458 if (unlikely(bio->bi_opf & REQ_FUA || in writecache_map()
1459 wc->uncommitted_blocks >= wc->autocommit_blocks)) in writecache_map()
1460 writecache_flush(wc); in writecache_map()
1462 writecache_schedule_autocommit(wc); in writecache_map()
1467 bio_set_dev(bio, wc->dev->bdev); in writecache_map()
1468 wc_unlock(wc); in writecache_map()
1473 bio->bi_private = (void *)1; in writecache_map()
1474 atomic_inc(&wc->bio_in_progress[bio_data_dir(bio)]); in writecache_map()
1475 wc_unlock(wc); in writecache_map()
1479 wc_unlock(wc); in writecache_map()
1484 wc_unlock(wc); in writecache_map()
1488 wc_unlock(wc); in writecache_map()
1495 struct dm_writecache *wc = ti->private; in writecache_end_io() local
1497 if (bio->bi_private != NULL) { in writecache_end_io()
1499 if (atomic_dec_and_test(&wc->bio_in_progress[dir])) in writecache_end_io()
1500 if (unlikely(waitqueue_active(&wc->bio_in_progress_wait[dir]))) in writecache_end_io()
1501 wake_up(&wc->bio_in_progress_wait[dir]); in writecache_end_io()
1509 struct dm_writecache *wc = ti->private; in writecache_iterate_devices() local
1511 return fn(ti, wc->dev, 0, ti->len, data); in writecache_iterate_devices()
1516 struct dm_writecache *wc = ti->private; in writecache_io_hints() local
1518 if (limits->logical_block_size < wc->block_size) in writecache_io_hints()
1519 limits->logical_block_size = wc->block_size; in writecache_io_hints()
1521 if (limits->physical_block_size < wc->block_size) in writecache_io_hints()
1522 limits->physical_block_size = wc->block_size; in writecache_io_hints()
1524 if (limits->io_min < wc->block_size) in writecache_io_hints()
1525 limits->io_min = wc->block_size; in writecache_io_hints()
1532 struct dm_writecache *wc = wb->wc; in writecache_writeback_endio() local
1535 raw_spin_lock_irqsave(&wc->endio_list_lock, flags); in writecache_writeback_endio()
1536 if (unlikely(list_empty(&wc->endio_list))) in writecache_writeback_endio()
1537 wake_up_process(wc->endio_thread); in writecache_writeback_endio()
1538 list_add_tail(&wb->endio_entry, &wc->endio_list); in writecache_writeback_endio()
1539 raw_spin_unlock_irqrestore(&wc->endio_list_lock, flags); in writecache_writeback_endio()
1545 struct dm_writecache *wc = c->wc; in writecache_copy_endio() local
1547 c->error = likely(!(read_err | write_err)) ? 0 : -EIO; in writecache_copy_endio()
1549 raw_spin_lock_irq(&wc->endio_list_lock); in writecache_copy_endio()
1550 if (unlikely(list_empty(&wc->endio_list))) in writecache_copy_endio()
1551 wake_up_process(wc->endio_thread); in writecache_copy_endio()
1552 list_add_tail(&c->endio_entry, &wc->endio_list); in writecache_copy_endio()
1553 raw_spin_unlock_irq(&wc->endio_list_lock); in writecache_copy_endio()
1556 static void __writecache_endio_pmem(struct dm_writecache *wc, struct list_head *list) in __writecache_endio_pmem() argument
1564 wb = list_entry(list->next, struct writeback_struct, endio_entry); in __writecache_endio_pmem()
1565 list_del(&wb->endio_entry); in __writecache_endio_pmem()
1567 if (unlikely(wb->bio.bi_status != BLK_STS_OK)) in __writecache_endio_pmem()
1568 writecache_error(wc, blk_status_to_errno(wb->bio.bi_status), in __writecache_endio_pmem()
1569 "write error %d", wb->bio.bi_status); in __writecache_endio_pmem()
1572 e = wb->wc_list[i]; in __writecache_endio_pmem()
1573 BUG_ON(!e->write_in_progress); in __writecache_endio_pmem()
1574 e->write_in_progress = false; in __writecache_endio_pmem()
1575 INIT_LIST_HEAD(&e->lru); in __writecache_endio_pmem()
1576 if (!writecache_has_error(wc)) in __writecache_endio_pmem()
1577 writecache_free_entry(wc, e); in __writecache_endio_pmem()
1578 BUG_ON(!wc->writeback_size); in __writecache_endio_pmem()
1579 wc->writeback_size--; in __writecache_endio_pmem()
1582 writecache_commit_flushed(wc, false); in __writecache_endio_pmem()
1583 wc_unlock(wc); in __writecache_endio_pmem()
1584 wc_lock(wc); in __writecache_endio_pmem()
1587 } while (++i < wb->wc_list_n); in __writecache_endio_pmem()
1589 if (wb->wc_list != wb->wc_list_inline) in __writecache_endio_pmem()
1590 kfree(wb->wc_list); in __writecache_endio_pmem()
1591 bio_put(&wb->bio); in __writecache_endio_pmem()
1595 static void __writecache_endio_ssd(struct dm_writecache *wc, struct list_head *list) in __writecache_endio_ssd() argument
1601 c = list_entry(list->next, struct copy_struct, endio_entry); in __writecache_endio_ssd()
1602 list_del(&c->endio_entry); in __writecache_endio_ssd()
1604 if (unlikely(c->error)) in __writecache_endio_ssd()
1605 writecache_error(wc, c->error, "copy error"); in __writecache_endio_ssd()
1607 e = c->e; in __writecache_endio_ssd()
1609 BUG_ON(!e->write_in_progress); in __writecache_endio_ssd()
1610 e->write_in_progress = false; in __writecache_endio_ssd()
1611 INIT_LIST_HEAD(&e->lru); in __writecache_endio_ssd()
1612 if (!writecache_has_error(wc)) in __writecache_endio_ssd()
1613 writecache_free_entry(wc, e); in __writecache_endio_ssd()
1615 BUG_ON(!wc->writeback_size); in __writecache_endio_ssd()
1616 wc->writeback_size--; in __writecache_endio_ssd()
1618 } while (--c->n_entries); in __writecache_endio_ssd()
1619 mempool_free(c, &wc->copy_pool); in __writecache_endio_ssd()
1625 struct dm_writecache *wc = data; in writecache_endio_thread() local
1630 raw_spin_lock_irq(&wc->endio_list_lock); in writecache_endio_thread()
1631 if (!list_empty(&wc->endio_list)) in writecache_endio_thread()
1634 raw_spin_unlock_irq(&wc->endio_list_lock); in writecache_endio_thread()
1646 list = wc->endio_list; in writecache_endio_thread()
1647 list.next->prev = list.prev->next = &list; in writecache_endio_thread()
1648 INIT_LIST_HEAD(&wc->endio_list); in writecache_endio_thread()
1649 raw_spin_unlock_irq(&wc->endio_list_lock); in writecache_endio_thread()
1651 if (!WC_MODE_FUA(wc)) in writecache_endio_thread()
1652 writecache_disk_flush(wc, wc->dev); in writecache_endio_thread()
1654 wc_lock(wc); in writecache_endio_thread()
1656 if (WC_MODE_PMEM(wc)) { in writecache_endio_thread()
1657 __writecache_endio_pmem(wc, &list); in writecache_endio_thread()
1659 __writecache_endio_ssd(wc, &list); in writecache_endio_thread()
1660 writecache_wait_for_ios(wc, READ); in writecache_endio_thread()
1663 writecache_commit_flushed(wc, false); in writecache_endio_thread()
1665 wc_unlock(wc); in writecache_endio_thread()
1673 struct dm_writecache *wc = wb->wc; in wc_add_block() local
1674 unsigned block_size = wc->block_size; in wc_add_block()
1675 void *address = memory_data(wc, e); in wc_add_block()
1679 if (unlikely(bio_end_sector(&wb->bio) >= wc->data_device_sectors)) in wc_add_block()
1682 return bio_add_page(&wb->bio, persistent_memory_page(address), in wc_add_block()
1691 static void __writeback_throttle(struct dm_writecache *wc, struct writeback_list *wbl) in __writeback_throttle() argument
1693 if (unlikely(wc->max_writeback_jobs)) { in __writeback_throttle()
1694 if (READ_ONCE(wc->writeback_size) - wbl->size >= wc->max_writeback_jobs) { in __writeback_throttle()
1695 wc_lock(wc); in __writeback_throttle()
1696 while (wc->writeback_size - wbl->size >= wc->max_writeback_jobs) in __writeback_throttle()
1697 writecache_wait_on_freelist(wc); in __writeback_throttle()
1698 wc_unlock(wc); in __writeback_throttle()
1704 static void __writecache_writeback_pmem(struct dm_writecache *wc, struct writeback_list *wbl) in __writecache_writeback_pmem() argument
1711 while (wbl->size) { in __writecache_writeback_pmem()
1712 wbl->size--; in __writecache_writeback_pmem()
1713 e = container_of(wbl->list.prev, struct wc_entry, lru); in __writecache_writeback_pmem()
1714 list_del(&e->lru); in __writecache_writeback_pmem()
1716 max_pages = e->wc_list_contiguous; in __writecache_writeback_pmem()
1718 bio = bio_alloc_bioset(GFP_NOIO, max_pages, &wc->bio_set); in __writecache_writeback_pmem()
1720 wb->wc = wc; in __writecache_writeback_pmem()
1721 bio->bi_end_io = writecache_writeback_endio; in __writecache_writeback_pmem()
1722 bio_set_dev(bio, wc->dev->bdev); in __writecache_writeback_pmem()
1723 bio->bi_iter.bi_sector = read_original_sector(wc, e); in __writecache_writeback_pmem()
1725 unlikely(!(wb->wc_list = kmalloc_array(max_pages, sizeof(struct wc_entry *), in __writecache_writeback_pmem()
1728 wb->wc_list = wb->wc_list_inline; in __writecache_writeback_pmem()
1734 wb->wc_list[0] = e; in __writecache_writeback_pmem()
1735 wb->wc_list_n = 1; in __writecache_writeback_pmem()
1737 while (wbl->size && wb->wc_list_n < max_pages) { in __writecache_writeback_pmem()
1738 f = container_of(wbl->list.prev, struct wc_entry, lru); in __writecache_writeback_pmem()
1739 if (read_original_sector(wc, f) != in __writecache_writeback_pmem()
1740 read_original_sector(wc, e) + (wc->block_size >> SECTOR_SHIFT)) in __writecache_writeback_pmem()
1744 wbl->size--; in __writecache_writeback_pmem()
1745 list_del(&f->lru); in __writecache_writeback_pmem()
1746 wb->wc_list[wb->wc_list_n++] = f; in __writecache_writeback_pmem()
1749 bio_set_op_attrs(bio, REQ_OP_WRITE, WC_MODE_FUA(wc) * REQ_FUA); in __writecache_writeback_pmem()
1750 if (writecache_has_error(wc)) { in __writecache_writeback_pmem()
1751 bio->bi_status = BLK_STS_IOERR; in __writecache_writeback_pmem()
1754 bio->bi_status = BLK_STS_OK; in __writecache_writeback_pmem()
1760 __writeback_throttle(wc, wbl); in __writecache_writeback_pmem()
1764 static void __writecache_writeback_ssd(struct dm_writecache *wc, struct writeback_list *wbl) in __writecache_writeback_ssd() argument
1770 while (wbl->size) { in __writecache_writeback_ssd()
1773 wbl->size--; in __writecache_writeback_ssd()
1774 e = container_of(wbl->list.prev, struct wc_entry, lru); in __writecache_writeback_ssd()
1775 list_del(&e->lru); in __writecache_writeback_ssd()
1777 n_sectors = e->wc_list_contiguous << (wc->block_size_bits - SECTOR_SHIFT); in __writecache_writeback_ssd()
1779 from.bdev = wc->ssd_dev->bdev; in __writecache_writeback_ssd()
1780 from.sector = cache_sector(wc, e); in __writecache_writeback_ssd()
1782 to.bdev = wc->dev->bdev; in __writecache_writeback_ssd()
1783 to.sector = read_original_sector(wc, e); in __writecache_writeback_ssd()
1786 c = mempool_alloc(&wc->copy_pool, GFP_NOIO); in __writecache_writeback_ssd()
1787 c->wc = wc; in __writecache_writeback_ssd()
1788 c->e = e; in __writecache_writeback_ssd()
1789 c->n_entries = e->wc_list_contiguous; in __writecache_writeback_ssd()
1791 while ((n_sectors -= wc->block_size >> SECTOR_SHIFT)) { in __writecache_writeback_ssd()
1792 wbl->size--; in __writecache_writeback_ssd()
1793 f = container_of(wbl->list.prev, struct wc_entry, lru); in __writecache_writeback_ssd()
1795 list_del(&f->lru); in __writecache_writeback_ssd()
1799 if (unlikely(to.sector + to.count > wc->data_device_sectors)) { in __writecache_writeback_ssd()
1800 if (to.sector >= wc->data_device_sectors) { in __writecache_writeback_ssd()
1804 from.count = to.count = wc->data_device_sectors - to.sector; in __writecache_writeback_ssd()
1807 dm_kcopyd_copy(wc->dm_kcopyd, &from, 1, &to, 0, writecache_copy_endio, c); in __writecache_writeback_ssd()
1809 __writeback_throttle(wc, wbl); in __writecache_writeback_ssd()
1815 struct dm_writecache *wc = container_of(work, struct dm_writecache, writeback_work); in writecache_writeback() local
1823 wc_lock(wc); in writecache_writeback()
1825 if (writecache_has_error(wc)) { in writecache_writeback()
1826 wc_unlock(wc); in writecache_writeback()
1830 if (unlikely(wc->writeback_all)) { in writecache_writeback()
1831 if (writecache_wait_for_writeback(wc)) in writecache_writeback()
1835 if (wc->overwrote_committed) { in writecache_writeback()
1836 writecache_wait_for_ios(wc, WRITE); in writecache_writeback()
1843 while (!list_empty(&wc->lru) && in writecache_writeback()
1844 (wc->writeback_all || in writecache_writeback()
1845 wc->freelist_size + wc->writeback_size <= wc->freelist_low_watermark || in writecache_writeback()
1846 (jiffies - container_of(wc->lru.prev, struct wc_entry, lru)->age >= in writecache_writeback()
1847 wc->max_age - wc->max_age / MAX_AGE_DIV))) { in writecache_writeback()
1851 likely(!wc->writeback_all) && likely(!dm_suspended(wc->ti))) { in writecache_writeback()
1852 queue_work(wc->writeback_wq, &wc->writeback_work); in writecache_writeback()
1856 if (unlikely(wc->writeback_all)) { in writecache_writeback()
1858 writecache_flush(wc); in writecache_writeback()
1859 e = container_of(rb_first(&wc->tree), struct wc_entry, rb_node); in writecache_writeback()
1863 e = container_of(wc->lru.prev, struct wc_entry, lru); in writecache_writeback()
1864 BUG_ON(e->write_in_progress); in writecache_writeback()
1865 if (unlikely(!writecache_entry_is_committed(wc, e))) { in writecache_writeback()
1866 writecache_flush(wc); in writecache_writeback()
1868 node = rb_prev(&e->rb_node); in writecache_writeback()
1871 if (unlikely(read_original_sector(wc, f) == in writecache_writeback()
1872 read_original_sector(wc, e))) { in writecache_writeback()
1873 BUG_ON(!f->write_in_progress); in writecache_writeback()
1874 list_del(&e->lru); in writecache_writeback()
1875 list_add(&e->lru, &skipped); in writecache_writeback()
1880 wc->writeback_size++; in writecache_writeback()
1881 list_del(&e->lru); in writecache_writeback()
1882 list_add(&e->lru, &wbl.list); in writecache_writeback()
1884 e->write_in_progress = true; in writecache_writeback()
1885 e->wc_list_contiguous = 1; in writecache_writeback()
1890 next_node = rb_next(&f->rb_node); in writecache_writeback()
1894 if (unlikely(read_original_sector(wc, g) == in writecache_writeback()
1895 read_original_sector(wc, f))) { in writecache_writeback()
1899 if (read_original_sector(wc, g) != in writecache_writeback()
1900 read_original_sector(wc, f) + (wc->block_size >> SECTOR_SHIFT)) in writecache_writeback()
1902 if (unlikely(g->write_in_progress)) in writecache_writeback()
1904 if (unlikely(!writecache_entry_is_committed(wc, g))) in writecache_writeback()
1907 if (!WC_MODE_PMEM(wc)) { in writecache_writeback()
1913 //if (unlikely(n_walked > WRITEBACK_LATENCY) && likely(!wc->writeback_all)) in writecache_writeback()
1916 wc->writeback_size++; in writecache_writeback()
1917 list_del(&g->lru); in writecache_writeback()
1918 list_add(&g->lru, &wbl.list); in writecache_writeback()
1920 g->write_in_progress = true; in writecache_writeback()
1921 g->wc_list_contiguous = BIO_MAX_PAGES; in writecache_writeback()
1923 e->wc_list_contiguous++; in writecache_writeback()
1924 if (unlikely(e->wc_list_contiguous == BIO_MAX_PAGES)) { in writecache_writeback()
1925 if (unlikely(wc->writeback_all)) { in writecache_writeback()
1926 next_node = rb_next(&f->rb_node); in writecache_writeback()
1937 list_splice_tail(&skipped, &wc->lru); in writecache_writeback()
1943 writecache_wait_for_writeback(wc); in writecache_writeback()
1946 wc_unlock(wc); in writecache_writeback()
1950 if (WC_MODE_PMEM(wc)) in writecache_writeback()
1951 __writecache_writeback_pmem(wc, &wbl); in writecache_writeback()
1953 __writecache_writeback_ssd(wc, &wbl); in writecache_writeback()
1957 if (unlikely(wc->writeback_all)) { in writecache_writeback()
1958 wc_lock(wc); in writecache_writeback()
1959 while (writecache_wait_for_writeback(wc)); in writecache_writeback()
1960 wc_unlock(wc); in writecache_writeback()
1975 return -ENOSPC; in calculate_memory_size()
1977 if (n_blocks >= ((size_t)-sizeof(struct wc_memory_superblock) / in calculate_memory_size()
1979 return -EFBIG; in calculate_memory_size()
1981 offset = (offset + block_size - 1) & ~(uint64_t)(block_size - 1); in calculate_memory_size()
1984 n_blocks--; in calculate_memory_size()
1990 return -EFBIG; in calculate_memory_size()
1999 static int init_memory(struct dm_writecache *wc) in init_memory() argument
2004 r = calculate_memory_size(wc->memory_map_size, wc->block_size, &wc->n_blocks, NULL); in init_memory()
2008 r = writecache_alloc_entries(wc); in init_memory()
2012 for (b = 0; b < ARRAY_SIZE(sb(wc)->padding); b++) in init_memory()
2013 pmem_assign(sb(wc)->padding[b], cpu_to_le64(0)); in init_memory()
2014 pmem_assign(sb(wc)->version, cpu_to_le32(MEMORY_SUPERBLOCK_VERSION)); in init_memory()
2015 pmem_assign(sb(wc)->block_size, cpu_to_le32(wc->block_size)); in init_memory()
2016 pmem_assign(sb(wc)->n_blocks, cpu_to_le64(wc->n_blocks)); in init_memory()
2017 pmem_assign(sb(wc)->seq_count, cpu_to_le64(0)); in init_memory()
2019 for (b = 0; b < wc->n_blocks; b++) { in init_memory()
2020 write_original_sector_seq_count(wc, &wc->entries[b], -1, -1); in init_memory()
2024 writecache_flush_all_metadata(wc); in init_memory()
2025 writecache_commit_flushed(wc, false); in init_memory()
2026 pmem_assign(sb(wc)->magic, cpu_to_le32(MEMORY_SUPERBLOCK_MAGIC)); in init_memory()
2027 writecache_flush_region(wc, &sb(wc)->magic, sizeof sb(wc)->magic); in init_memory()
2028 writecache_commit_flushed(wc, false); in init_memory()
2035 struct dm_writecache *wc = ti->private; in writecache_dtr() local
2037 if (!wc) in writecache_dtr()
2040 if (wc->endio_thread) in writecache_dtr()
2041 kthread_stop(wc->endio_thread); in writecache_dtr()
2043 if (wc->flush_thread) in writecache_dtr()
2044 kthread_stop(wc->flush_thread); in writecache_dtr()
2046 bioset_exit(&wc->bio_set); in writecache_dtr()
2048 mempool_exit(&wc->copy_pool); in writecache_dtr()
2050 if (wc->writeback_wq) in writecache_dtr()
2051 destroy_workqueue(wc->writeback_wq); in writecache_dtr()
2053 if (wc->dev) in writecache_dtr()
2054 dm_put_device(ti, wc->dev); in writecache_dtr()
2056 if (wc->ssd_dev) in writecache_dtr()
2057 dm_put_device(ti, wc->ssd_dev); in writecache_dtr()
2059 if (wc->entries) in writecache_dtr()
2060 vfree(wc->entries); in writecache_dtr()
2062 if (wc->memory_map) { in writecache_dtr()
2063 if (WC_MODE_PMEM(wc)) in writecache_dtr()
2064 persistent_memory_release(wc); in writecache_dtr()
2066 vfree(wc->memory_map); in writecache_dtr()
2069 if (wc->dm_kcopyd) in writecache_dtr()
2070 dm_kcopyd_client_destroy(wc->dm_kcopyd); in writecache_dtr()
2072 if (wc->dm_io) in writecache_dtr()
2073 dm_io_client_destroy(wc->dm_io); in writecache_dtr()
2075 if (wc->dirty_bitmap) in writecache_dtr()
2076 vfree(wc->dirty_bitmap); in writecache_dtr()
2078 kfree(wc); in writecache_dtr()
2083 struct dm_writecache *wc; in writecache_ctr() local
2102 wc = kzalloc(sizeof(struct dm_writecache), GFP_KERNEL); in writecache_ctr()
2103 if (!wc) { in writecache_ctr()
2104 ti->error = "Cannot allocate writecache structure"; in writecache_ctr()
2105 r = -ENOMEM; in writecache_ctr()
2108 ti->private = wc; in writecache_ctr()
2109 wc->ti = ti; in writecache_ctr()
2111 mutex_init(&wc->lock); in writecache_ctr()
2112 wc->max_age = MAX_AGE_UNSPECIFIED; in writecache_ctr()
2113 writecache_poison_lists(wc); in writecache_ctr()
2114 init_waitqueue_head(&wc->freelist_wait); in writecache_ctr()
2115 timer_setup(&wc->autocommit_timer, writecache_autocommit_timer, 0); in writecache_ctr()
2116 timer_setup(&wc->max_age_timer, writecache_max_age_timer, 0); in writecache_ctr()
2119 atomic_set(&wc->bio_in_progress[i], 0); in writecache_ctr()
2120 init_waitqueue_head(&wc->bio_in_progress_wait[i]); in writecache_ctr()
2123 wc->dm_io = dm_io_client_create(); in writecache_ctr()
2124 if (IS_ERR(wc->dm_io)) { in writecache_ctr()
2125 r = PTR_ERR(wc->dm_io); in writecache_ctr()
2126 ti->error = "Unable to allocate dm-io client"; in writecache_ctr()
2127 wc->dm_io = NULL; in writecache_ctr()
2131 wc->writeback_wq = alloc_workqueue("writecache-writeback", WQ_MEM_RECLAIM, 1); in writecache_ctr()
2132 if (!wc->writeback_wq) { in writecache_ctr()
2133 r = -ENOMEM; in writecache_ctr()
2134 ti->error = "Could not allocate writeback workqueue"; in writecache_ctr()
2137 INIT_WORK(&wc->writeback_work, writecache_writeback); in writecache_ctr()
2138 INIT_WORK(&wc->flush_work, writecache_flush_work); in writecache_ctr()
2140 raw_spin_lock_init(&wc->endio_list_lock); in writecache_ctr()
2141 INIT_LIST_HEAD(&wc->endio_list); in writecache_ctr()
2142 wc->endio_thread = kthread_create(writecache_endio_thread, wc, "writecache_endio"); in writecache_ctr()
2143 if (IS_ERR(wc->endio_thread)) { in writecache_ctr()
2144 r = PTR_ERR(wc->endio_thread); in writecache_ctr()
2145 wc->endio_thread = NULL; in writecache_ctr()
2146 ti->error = "Couldn't spawn endio thread"; in writecache_ctr()
2149 wake_up_process(wc->endio_thread); in writecache_ctr()
2159 wc->pmem_mode = false; in writecache_ctr()
2162 wc->pmem_mode = true; in writecache_ctr()
2163 wc->writeback_fua = true; in writecache_ctr()
2166 * If the architecture doesn't support persistent memory or in writecache_ctr()
2168 * only be used in SSD-only mode. in writecache_ctr()
2170 r = -EOPNOTSUPP; in writecache_ctr()
2171 ti->error = "Persistent memory or DAX not supported on this system"; in writecache_ctr()
2178 if (WC_MODE_PMEM(wc)) { in writecache_ctr()
2179 r = bioset_init(&wc->bio_set, BIO_POOL_SIZE, in writecache_ctr()
2183 ti->error = "Could not allocate bio set"; in writecache_ctr()
2187 r = mempool_init_kmalloc_pool(&wc->copy_pool, 1, sizeof(struct copy_struct)); in writecache_ctr()
2189 ti->error = "Could not allocate mempool"; in writecache_ctr()
2200 r = dm_get_device(ti, string, dm_table_get_mode(ti->table), &wc->dev); in writecache_ctr()
2202 ti->error = "Origin data device lookup failed"; in writecache_ctr()
2213 r = dm_get_device(ti, string, dm_table_get_mode(ti->table), &wc->ssd_dev); in writecache_ctr()
2215 ti->error = "Cache data device lookup failed"; in writecache_ctr()
2218 wc->memory_map_size = i_size_read(wc->ssd_dev->bdev->bd_inode); in writecache_ctr()
2226 if (sscanf(string, "%u%c", &wc->block_size, &dummy) != 1 || in writecache_ctr()
2227 wc->block_size < 512 || wc->block_size > PAGE_SIZE || in writecache_ctr()
2228 (wc->block_size & (wc->block_size - 1))) { in writecache_ctr()
2229 r = -EINVAL; in writecache_ctr()
2230 ti->error = "Invalid block size"; in writecache_ctr()
2233 if (wc->block_size < bdev_logical_block_size(wc->dev->bdev) || in writecache_ctr()
2234 wc->block_size < bdev_logical_block_size(wc->ssd_dev->bdev)) { in writecache_ctr()
2235 r = -EINVAL; in writecache_ctr()
2236 ti->error = "Block size is smaller than device logical block size"; in writecache_ctr()
2239 wc->block_size_bits = __ffs(wc->block_size); in writecache_ctr()
2241 wc->max_writeback_jobs = MAX_WRITEBACK_JOBS; in writecache_ctr()
2242 wc->autocommit_blocks = !WC_MODE_PMEM(wc) ? AUTOCOMMIT_BLOCKS_SSD : AUTOCOMMIT_BLOCKS_PMEM; in writecache_ctr()
2243 wc->autocommit_jiffies = msecs_to_jiffies(AUTOCOMMIT_MSEC); in writecache_ctr()
2248 r = dm_read_arg_group(_args, &as, &opt_params, &ti->error); in writecache_ctr()
2253 string = dm_shift_arg(&as), opt_params--; in writecache_ctr()
2256 string = dm_shift_arg(&as), opt_params--; in writecache_ctr()
2259 wc->start_sector = start_sector; in writecache_ctr()
2260 wc->start_sector_set = true; in writecache_ctr()
2261 if (wc->start_sector != start_sector || in writecache_ctr()
2262 wc->start_sector >= wc->memory_map_size >> SECTOR_SHIFT) in writecache_ctr()
2265 string = dm_shift_arg(&as), opt_params--; in writecache_ctr()
2270 wc->high_wm_percent_value = high_wm_percent; in writecache_ctr()
2271 wc->high_wm_percent_set = true; in writecache_ctr()
2273 string = dm_shift_arg(&as), opt_params--; in writecache_ctr()
2278 wc->low_wm_percent_value = low_wm_percent; in writecache_ctr()
2279 wc->low_wm_percent_set = true; in writecache_ctr()
2281 string = dm_shift_arg(&as), opt_params--; in writecache_ctr()
2282 if (sscanf(string, "%u%c", &wc->max_writeback_jobs, &dummy) != 1) in writecache_ctr()
2284 wc->max_writeback_jobs_set = true; in writecache_ctr()
2286 string = dm_shift_arg(&as), opt_params--; in writecache_ctr()
2287 if (sscanf(string, "%u%c", &wc->autocommit_blocks, &dummy) != 1) in writecache_ctr()
2289 wc->autocommit_blocks_set = true; in writecache_ctr()
2292 string = dm_shift_arg(&as), opt_params--; in writecache_ctr()
2297 wc->autocommit_jiffies = msecs_to_jiffies(autocommit_msecs); in writecache_ctr()
2298 wc->autocommit_time_value = autocommit_msecs; in writecache_ctr()
2299 wc->autocommit_time_set = true; in writecache_ctr()
2302 string = dm_shift_arg(&as), opt_params--; in writecache_ctr()
2307 wc->max_age = msecs_to_jiffies(max_age_msecs); in writecache_ctr()
2308 wc->max_age_set = true; in writecache_ctr()
2309 wc->max_age_value = max_age_msecs; in writecache_ctr()
2311 wc->cleaner_set = true; in writecache_ctr()
2312 wc->cleaner = true; in writecache_ctr()
2314 if (WC_MODE_PMEM(wc)) { in writecache_ctr()
2315 wc->writeback_fua = true; in writecache_ctr()
2316 wc->writeback_fua_set = true; in writecache_ctr()
2319 if (WC_MODE_PMEM(wc)) { in writecache_ctr()
2320 wc->writeback_fua = false; in writecache_ctr()
2321 wc->writeback_fua_set = true; in writecache_ctr()
2325 r = -EINVAL; in writecache_ctr()
2326 ti->error = "Invalid optional argument"; in writecache_ctr()
2332 r = -EINVAL; in writecache_ctr()
2333 ti->error = "High watermark must be greater than or equal to low watermark"; in writecache_ctr()
2337 if (WC_MODE_PMEM(wc)) { in writecache_ctr()
2338 if (!dax_synchronous(wc->ssd_dev->dax_dev)) { in writecache_ctr()
2339 r = -EOPNOTSUPP; in writecache_ctr()
2340 ti->error = "Asynchronous persistent memory not supported as pmem cache"; in writecache_ctr()
2344 r = persistent_memory_claim(wc); in writecache_ctr()
2346 ti->error = "Unable to map persistent memory for cache"; in writecache_ctr()
2353 wc->memory_map_size -= (uint64_t)wc->start_sector << SECTOR_SHIFT; in writecache_ctr()
2355 bio_list_init(&wc->flush_list); in writecache_ctr()
2356 wc->flush_thread = kthread_create(writecache_flush_thread, wc, "dm_writecache_flush"); in writecache_ctr()
2357 if (IS_ERR(wc->flush_thread)) { in writecache_ctr()
2358 r = PTR_ERR(wc->flush_thread); in writecache_ctr()
2359 wc->flush_thread = NULL; in writecache_ctr()
2360 ti->error = "Couldn't spawn flush thread"; in writecache_ctr()
2363 wake_up_process(wc->flush_thread); in writecache_ctr()
2365 r = calculate_memory_size(wc->memory_map_size, wc->block_size, in writecache_ctr()
2368 ti->error = "Invalid device size"; in writecache_ctr()
2372 n_bitmap_bits = (((uint64_t)n_metadata_blocks << wc->block_size_bits) + in writecache_ctr()
2373 BITMAP_GRANULARITY - 1) / BITMAP_GRANULARITY; in writecache_ctr()
2376 r = -EFBIG; in writecache_ctr()
2377 ti->error = "Invalid device size"; in writecache_ctr()
2381 wc->memory_map = vmalloc(n_metadata_blocks << wc->block_size_bits); in writecache_ctr()
2382 if (!wc->memory_map) { in writecache_ctr()
2383 r = -ENOMEM; in writecache_ctr()
2384 ti->error = "Unable to allocate memory for metadata"; in writecache_ctr()
2388 wc->dm_kcopyd = dm_kcopyd_client_create(&dm_kcopyd_throttle); in writecache_ctr()
2389 if (IS_ERR(wc->dm_kcopyd)) { in writecache_ctr()
2390 r = PTR_ERR(wc->dm_kcopyd); in writecache_ctr()
2391 ti->error = "Unable to allocate dm-kcopyd client"; in writecache_ctr()
2392 wc->dm_kcopyd = NULL; in writecache_ctr()
2396 wc->metadata_sectors = n_metadata_blocks << (wc->block_size_bits - SECTOR_SHIFT); in writecache_ctr()
2397 wc->dirty_bitmap_size = (n_bitmap_bits + BITS_PER_LONG - 1) / in writecache_ctr()
2399 wc->dirty_bitmap = vzalloc(wc->dirty_bitmap_size); in writecache_ctr()
2400 if (!wc->dirty_bitmap) { in writecache_ctr()
2401 r = -ENOMEM; in writecache_ctr()
2402 ti->error = "Unable to allocate dirty bitmap"; in writecache_ctr()
2406 r = writecache_read_metadata(wc, wc->block_size >> SECTOR_SHIFT); in writecache_ctr()
2408 ti->error = "Unable to read first block of metadata"; in writecache_ctr()
2413 r = copy_mc_to_kernel(&s, sb(wc), sizeof(struct wc_memory_superblock)); in writecache_ctr()
2415 ti->error = "Hardware memory error when reading superblock"; in writecache_ctr()
2419 r = init_memory(wc); in writecache_ctr()
2421 ti->error = "Unable to initialize device"; in writecache_ctr()
2424 r = copy_mc_to_kernel(&s, sb(wc), in writecache_ctr()
2427 ti->error = "Hardware memory error when reading superblock"; in writecache_ctr()
2433 ti->error = "Invalid magic in the superblock"; in writecache_ctr()
2434 r = -EINVAL; in writecache_ctr()
2439 ti->error = "Invalid version in the superblock"; in writecache_ctr()
2440 r = -EINVAL; in writecache_ctr()
2444 if (le32_to_cpu(s.block_size) != wc->block_size) { in writecache_ctr()
2445 ti->error = "Block size does not match superblock"; in writecache_ctr()
2446 r = -EINVAL; in writecache_ctr()
2450 wc->n_blocks = le64_to_cpu(s.n_blocks); in writecache_ctr()
2452 offset = wc->n_blocks * sizeof(struct wc_memory_entry); in writecache_ctr()
2453 if (offset / sizeof(struct wc_memory_entry) != le64_to_cpu(sb(wc)->n_blocks)) { in writecache_ctr()
2455 ti->error = "Overflow in size calculation"; in writecache_ctr()
2456 r = -EINVAL; in writecache_ctr()
2462 offset = (offset + wc->block_size - 1) & ~(size_t)(wc->block_size - 1); in writecache_ctr()
2463 data_size = wc->n_blocks * (size_t)wc->block_size; in writecache_ctr()
2464 if (!offset || (data_size / wc->block_size != wc->n_blocks) || in writecache_ctr()
2467 if (offset + data_size > wc->memory_map_size) { in writecache_ctr()
2468 ti->error = "Memory area is too small"; in writecache_ctr()
2469 r = -EINVAL; in writecache_ctr()
2473 wc->metadata_sectors = offset >> SECTOR_SHIFT; in writecache_ctr()
2474 wc->block_start = (char *)sb(wc) + offset; in writecache_ctr()
2476 x = (uint64_t)wc->n_blocks * (100 - high_wm_percent); in writecache_ctr()
2479 wc->freelist_high_watermark = x; in writecache_ctr()
2480 x = (uint64_t)wc->n_blocks * (100 - low_wm_percent); in writecache_ctr()
2483 wc->freelist_low_watermark = x; in writecache_ctr()
2485 if (wc->cleaner) in writecache_ctr()
2486 activate_cleaner(wc); in writecache_ctr()
2488 r = writecache_alloc_entries(wc); in writecache_ctr()
2490 ti->error = "Cannot allocate memory"; in writecache_ctr()
2494 ti->num_flush_bios = WC_MODE_PMEM(wc) ? 1 : 2; in writecache_ctr()
2495 ti->flush_supported = true; in writecache_ctr()
2496 ti->num_discard_bios = 1; in writecache_ctr()
2498 if (WC_MODE_PMEM(wc)) in writecache_ctr()
2499 persistent_memory_flush_cache(wc->memory_map, wc->memory_map_size); in writecache_ctr()
2504 r = -EINVAL; in writecache_ctr()
2505 ti->error = "Bad arguments"; in writecache_ctr()
2514 struct dm_writecache *wc = ti->private; in writecache_status() local
2520 DMEMIT("%ld %llu %llu %llu", writecache_has_error(wc), in writecache_status()
2521 (unsigned long long)wc->n_blocks, (unsigned long long)wc->freelist_size, in writecache_status()
2522 (unsigned long long)wc->writeback_size); in writecache_status()
2525 DMEMIT("%c %s %s %u ", WC_MODE_PMEM(wc) ? 'p' : 's', in writecache_status()
2526 wc->dev->name, wc->ssd_dev->name, wc->block_size); in writecache_status()
2528 if (wc->start_sector_set) in writecache_status()
2530 if (wc->high_wm_percent_set) in writecache_status()
2532 if (wc->low_wm_percent_set) in writecache_status()
2534 if (wc->max_writeback_jobs_set) in writecache_status()
2536 if (wc->autocommit_blocks_set) in writecache_status()
2538 if (wc->autocommit_time_set) in writecache_status()
2540 if (wc->max_age_set) in writecache_status()
2542 if (wc->cleaner_set) in writecache_status()
2544 if (wc->writeback_fua_set) in writecache_status()
2548 if (wc->start_sector_set) in writecache_status()
2549 DMEMIT(" start_sector %llu", (unsigned long long)wc->start_sector); in writecache_status()
2550 if (wc->high_wm_percent_set) in writecache_status()
2551 DMEMIT(" high_watermark %u", wc->high_wm_percent_value); in writecache_status()
2552 if (wc->low_wm_percent_set) in writecache_status()
2553 DMEMIT(" low_watermark %u", wc->low_wm_percent_value); in writecache_status()
2554 if (wc->max_writeback_jobs_set) in writecache_status()
2555 DMEMIT(" writeback_jobs %u", wc->max_writeback_jobs); in writecache_status()
2556 if (wc->autocommit_blocks_set) in writecache_status()
2557 DMEMIT(" autocommit_blocks %u", wc->autocommit_blocks); in writecache_status()
2558 if (wc->autocommit_time_set) in writecache_status()
2559 DMEMIT(" autocommit_time %u", wc->autocommit_time_value); in writecache_status()
2560 if (wc->max_age_set) in writecache_status()
2561 DMEMIT(" max_age %u", wc->max_age_value); in writecache_status()
2562 if (wc->cleaner_set) in writecache_status()
2564 if (wc->writeback_fua_set) in writecache_status()
2565 DMEMIT(" %sfua", wc->writeback_fua ? "" : "no"); in writecache_status()
2608 MODULE_AUTHOR("Mikulas Patocka <dm-devel@redhat.com>");