|
| 1 | +lib/sbitmap: define swap_lock as raw_spinlock_t |
| 2 | + |
| 3 | +jira LE-4034 |
| 4 | +Rebuild_History Non-Buildable kernel-4.18.0-553.72.1.el8_10 |
| 5 | +commit-author Ming Lei <ming.lei@redhat.com> |
| 6 | +commit 65f666c6203600053478ce8e34a1db269a8701c9 |
| 7 | +Empty-Commit: Cherry-Pick Conflicts during history rebuild. |
| 8 | +Will be included in final tarball splat. Ref for failed cherry-pick at: |
| 9 | +ciq/ciq_backports/kernel-4.18.0-553.72.1.el8_10/65f666c6.failed |
| 10 | + |
| 11 | +When called from sbitmap_queue_get(), sbitmap_deferred_clear() may be run |
| 12 | +with preempt disabled. In RT kernel, spin_lock() can sleep, then warning |
| 13 | +of "BUG: sleeping function called from invalid context" can be triggered. |
| 14 | + |
| 15 | +Fix it by replacing it with raw_spin_lock. |
| 16 | + |
| 17 | + Cc: Yang Yang <yang.yang@vivo.com> |
| 18 | +Fixes: 72d04bdcf3f7 ("sbitmap: fix io hung due to race on sbitmap_word::cleared") |
| 19 | + Signed-off-by: Ming Lei <ming.lei@redhat.com> |
| 20 | + Reviewed-by: Yang Yang <yang.yang@vivo.com> |
| 21 | +Link: https://lore.kernel.org/r/20240919021709.511329-1-ming.lei@redhat.com |
| 22 | + Signed-off-by: Jens Axboe <axboe@kernel.dk> |
| 23 | +(cherry picked from commit 65f666c6203600053478ce8e34a1db269a8701c9) |
| 24 | + Signed-off-by: Jonathan Maple <jmaple@ciq.com> |
| 25 | + |
| 26 | +# Conflicts: |
| 27 | +# include/linux/sbitmap.h |
| 28 | +# lib/sbitmap.c |
| 29 | +diff --cc include/linux/sbitmap.h |
| 30 | +index c5364c7c3ae5,189140bf11fc..000000000000 |
| 31 | +--- a/include/linux/sbitmap.h |
| 32 | ++++ b/include/linux/sbitmap.h |
| 33 | +@@@ -40,9 -38,9 +40,13 @@@ struct sbitmap_word |
| 34 | + unsigned long cleared ____cacheline_aligned_in_smp; |
| 35 | + |
| 36 | + /** |
| 37 | + - * @swap_lock: serializes simultaneous updates of ->word and ->cleared |
| 38 | + + * @swap_lock: Held while swapping word <-> cleared |
| 39 | + */ |
| 40 | +++<<<<<<< HEAD |
| 41 | + + RH_KABI_DEPRECATE(spinlock_t, swap_lock) |
| 42 | +++======= |
| 43 | ++ raw_spinlock_t swap_lock; |
| 44 | +++>>>>>>> 65f666c62036 (lib/sbitmap: define swap_lock as raw_spinlock_t) |
| 45 | + } ____cacheline_aligned_in_smp; |
| 46 | + |
| 47 | + /** |
| 48 | +diff --cc lib/sbitmap.c |
| 49 | +index 59f985b64b05,d3412984170c..000000000000 |
| 50 | +--- a/lib/sbitmap.c |
| 51 | ++++ b/lib/sbitmap.c |
| 52 | +@@@ -76,12 -60,30 +76,34 @@@ static inline void update_alloc_hint_af |
| 53 | + /* |
| 54 | + * See if we have deferred clears that we can batch move |
| 55 | + */ |
| 56 | + -static inline bool sbitmap_deferred_clear(struct sbitmap_word *map, |
| 57 | + - unsigned int depth, unsigned int alloc_hint, bool wrap) |
| 58 | + +static inline bool sbitmap_deferred_clear(struct sbitmap_word *map) |
| 59 | + { |
| 60 | + - unsigned long mask, word_mask; |
| 61 | + + unsigned long mask; |
| 62 | + |
| 63 | +++<<<<<<< HEAD |
| 64 | + + if (!READ_ONCE(map->cleared)) |
| 65 | + + return false; |
| 66 | +++======= |
| 67 | ++ guard(raw_spinlock_irqsave)(&map->swap_lock); |
| 68 | ++ |
| 69 | ++ if (!map->cleared) { |
| 70 | ++ if (depth == 0) |
| 71 | ++ return false; |
| 72 | ++ |
| 73 | ++ word_mask = (~0UL) >> (BITS_PER_LONG - depth); |
| 74 | ++ /* |
| 75 | ++ * The current behavior is to always retry after moving |
| 76 | ++ * ->cleared to word, and we change it to retry in case |
| 77 | ++ * of any free bits. To avoid an infinite loop, we need |
| 78 | ++ * to take wrap & alloc_hint into account, otherwise a |
| 79 | ++ * soft lockup may occur. |
| 80 | ++ */ |
| 81 | ++ if (!wrap && alloc_hint) |
| 82 | ++ word_mask &= ~((1UL << alloc_hint) - 1); |
| 83 | ++ |
| 84 | ++ return (READ_ONCE(map->word) & word_mask) != word_mask; |
| 85 | ++ } |
| 86 | +++>>>>>>> 65f666c62036 (lib/sbitmap: define swap_lock as raw_spinlock_t) |
| 87 | + |
| 88 | + /* |
| 89 | + * First get a stable cleared mask, setting the old mask to 0. |
| 90 | +@@@ -119,24 -122,21 +141,29 @@@ int sbitmap_init_node(struct sbitmap *s |
| 91 | + return 0; |
| 92 | + } |
| 93 | + |
| 94 | + - if (alloc_hint) { |
| 95 | + - if (init_alloc_hint(sb, flags)) |
| 96 | + - return -ENOMEM; |
| 97 | + - } else { |
| 98 | + - sb->alloc_hint = NULL; |
| 99 | + - } |
| 100 | + - |
| 101 | + - sb->map = kvzalloc_node(sb->map_nr * sizeof(*sb->map), flags, node); |
| 102 | + + /* |
| 103 | + + * RHEL8: sb->alloc_hint is kept in an additional sb->map array element prior |
| 104 | + + * to the actual sb->map due to KABI (it was not possible to add it to struct sbitmap) |
| 105 | + + */ |
| 106 | + + sb->map = kvzalloc_node((sb->map_nr + 1) * sizeof(*sb->map), flags, node); |
| 107 | + if (!sb->map) { |
| 108 | + - free_percpu(sb->alloc_hint); |
| 109 | + return -ENOMEM; |
| 110 | + } |
| 111 | + + sb->map++; |
| 112 | + |
| 113 | +++<<<<<<< HEAD |
| 114 | + + if (alloc_hint) { |
| 115 | + + if (init_alloc_hint(sb, flags)) { |
| 116 | + + kfree(--sb->map); |
| 117 | + + return -ENOMEM; |
| 118 | + + } |
| 119 | + + } else { |
| 120 | + + *SB_ALLOC_HINT_PTR(sb) = NULL; |
| 121 | + + } |
| 122 | +++======= |
| 123 | ++ for (i = 0; i < sb->map_nr; i++) |
| 124 | ++ raw_spin_lock_init(&sb->map[i].swap_lock); |
| 125 | +++>>>>>>> 65f666c62036 (lib/sbitmap: define swap_lock as raw_spinlock_t) |
| 126 | + |
| 127 | + return 0; |
| 128 | + } |
| 129 | +* Unmerged path include/linux/sbitmap.h |
| 130 | +* Unmerged path lib/sbitmap.c |
0 commit comments