On Fri, 6 Aug 2021, Sebastian Andrzej Siewior wrote:
> Each CPU has SHMEM_INO_BATCH inodes available in `->ino_batch' which is
> per-CPU. Access here is serialized by disabling preemption. If the pool is
> empty, it gets reloaded from `->next_ino'. Access here is serialized by
> ->stat_lock which is a spinlock_t and can not be acquired with disabled
> preemption.
> One way around it would make per-CPU ino_batch struct containing the inode
> number a local_lock_t.
> Another solution is to promote ->stat_lock to a raw_spinlock_t. The critical
> sections are short. The mpol_put() must be moved outside of the critical
> section to avoid invoking the destructor with disabled preemption.
>
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@xxxxxxxxxxxxx>
Acked-by: Hugh Dickins <hughd@xxxxxxxxxx>
I was a little worried at first by the use in shmem_reconfigure(),
potentially around a percpu_counter_compare(), which in some cases does a
sum across all cpus - not as short a critical section as the rest of them.
But I see that __percpu_counter_sum() uses a raw_spinlock_t itself, and
has done so for twelve years: if that one is good, then this use here
must be good too.
Thanks.
> ---
> include/linux/shmem_fs.h | 2 +-
> mm/shmem.c | 31 +++++++++++++++++--------------
> 2 files changed, 18 insertions(+), 15 deletions(-)
>
> diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
> index 8e775ce517bb3..0a8499fb9c3cb 100644
> --- a/include/linux/shmem_fs.h
> +++ b/include/linux/shmem_fs.h
> @@ -31,7 +31,7 @@ struct shmem_sb_info {
> struct percpu_counter used_blocks; /* How many are allocated */
> unsigned long max_inodes; /* How many inodes are allowed */
> unsigned long free_inodes; /* How many are left for allocation */
> - spinlock_t stat_lock; /* Serialize shmem_sb_info changes */
> + raw_spinlock_t stat_lock; /* Serialize shmem_sb_info changes */
> umode_t mode; /* Mount mode for root directory */
> unsigned char huge; /* Whether to try for hugepages */
> kuid_t uid; /* Mount uid for root directory */
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 70d9ce294bb49..1b95afc4483c1 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -278,10 +278,10 @@ static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
> ino_t ino;
>
> if (!(sb->s_flags & SB_KERNMOUNT)) {
> - spin_lock(&sbinfo->stat_lock);
> + raw_spin_lock(&sbinfo->stat_lock);
> if (sbinfo->max_inodes) {
> if (!sbinfo->free_inodes) {
> - spin_unlock(&sbinfo->stat_lock);
> + raw_spin_unlock(&sbinfo->stat_lock);
> return -ENOSPC;
> }
> sbinfo->free_inodes--;
> @@ -304,7 +304,7 @@ static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
> }
> *inop = ino;
> }
> - spin_unlock(&sbinfo->stat_lock);
> + raw_spin_unlock(&sbinfo->stat_lock);
> } else if (inop) {
> /*
> * __shmem_file_setup, one of our callers, is lock-free: it
> @@ -319,13 +319,14 @@ static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
> * to worry about things like glibc compatibility.
> */
> ino_t *next_ino;
> +
> next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
> ino = *next_ino;
> if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
> - spin_lock(&sbinfo->stat_lock);
> + raw_spin_lock(&sbinfo->stat_lock);
> ino = sbinfo->next_ino;
> sbinfo->next_ino += SHMEM_INO_BATCH;
> - spin_unlock(&sbinfo->stat_lock);
> + raw_spin_unlock(&sbinfo->stat_lock);
> if (unlikely(is_zero_ino(ino)))
> ino++;
> }
> @@ -341,9 +342,9 @@ static void shmem_free_inode(struct super_block *sb)
> {
> struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
> if (sbinfo->max_inodes) {
> - spin_lock(&sbinfo->stat_lock);
> + raw_spin_lock(&sbinfo->stat_lock);
> sbinfo->free_inodes++;
> - spin_unlock(&sbinfo->stat_lock);
> + raw_spin_unlock(&sbinfo->stat_lock);
> }
> }
>
> @@ -1453,10 +1454,10 @@ static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
> {
> struct mempolicy *mpol = NULL;
> if (sbinfo->mpol) {
> - spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
> + raw_spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
> mpol = sbinfo->mpol;
> mpol_get(mpol);
> - spin_unlock(&sbinfo->stat_lock);
> + raw_spin_unlock(&sbinfo->stat_lock);
> }
> return mpol;
> }
> @@ -3500,9 +3501,10 @@ static int shmem_reconfigure(struct fs_context *fc)
> struct shmem_options *ctx = fc->fs_private;
> struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
> unsigned long inodes;
> + struct mempolicy *mpol = NULL;
> const char *err;
>
> - spin_lock(&sbinfo->stat_lock);
> + raw_spin_lock(&sbinfo->stat_lock);
> inodes = sbinfo->max_inodes - sbinfo->free_inodes;
> if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
> if (!sbinfo->max_blocks) {
> @@ -3547,14 +3549,15 @@ static int shmem_reconfigure(struct fs_context *fc)
> * Preserve previous mempolicy unless mpol remount option was specified.
> */
> if (ctx->mpol) {
> - mpol_put(sbinfo->mpol);
> + mpol = sbinfo->mpol;
> sbinfo->mpol = ctx->mpol; /* transfers initial ref */
> ctx->mpol = NULL;
> }
> - spin_unlock(&sbinfo->stat_lock);
> + raw_spin_unlock(&sbinfo->stat_lock);
> + mpol_put(mpol);
> return 0;
> out:
> - spin_unlock(&sbinfo->stat_lock);
> + raw_spin_unlock(&sbinfo->stat_lock);
> return invalfc(fc, "%s", err);
> }
>
> @@ -3671,7 +3674,7 @@ static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
> sbinfo->mpol = ctx->mpol;
> ctx->mpol = NULL;
>
> - spin_lock_init(&sbinfo->stat_lock);
> + raw_spin_lock_init(&sbinfo->stat_lock);
> if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
> goto failed;
> spin_lock_init(&sbinfo->shrinklist_lock);
> --
> 2.32.0
>
>
