This is a note to let you know that I've just added the patch titled
io_uring/kbuf: recycle freed mapped buffer ring entries
to the 6.6-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
io_uring-kbuf-recycle-freed-mapped-buffer-ring-entries.patch
and it can be found in the queue-6.6 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@xxxxxxxxxxxxxxx> know about it.
>From b10b73c102a2eab91e1cd62a03d6446f1dfecc64 Mon Sep 17 00:00:00 2001
From: Jens Axboe <axboe@xxxxxxxxx>
Date: Tue, 28 Nov 2023 11:17:25 -0700
Subject: io_uring/kbuf: recycle freed mapped buffer ring entries
From: Jens Axboe <axboe@xxxxxxxxx>
commit b10b73c102a2eab91e1cd62a03d6446f1dfecc64 upstream.
Right now we stash any potentially mmap'ed provided ring buffer range
for freeing at release time, regardless of when they get unregistered.
Since we're keeping track of these ranges anyway, keep track of their
registration state as well, and use that to recycle ranges when
appropriate rather than always allocate new ones.
The lookup is a basic scan of entries, checking for the best matching
free entry.
Fixes: c392cbecd8ec ("io_uring/kbuf: defer release of mapped buffer rings")
Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
Signed-off-by: Greg Kroah-Hartman <gregkh@xxxxxxxxxxxxxxxxxxx>
---
io_uring/kbuf.c | 77 ++++++++++++++++++++++++++++++++++++++++++++++++--------
1 file changed, 66 insertions(+), 11 deletions(-)
--- a/io_uring/kbuf.c
+++ b/io_uring/kbuf.c
@@ -44,6 +44,8 @@ static struct io_buffer_list *__io_buffe
struct io_buf_free {
struct hlist_node list;
void *mem;
+ size_t size;
+ int inuse;
};
static inline struct io_buffer_list *io_buffer_get_list(struct io_ring_ctx *ctx,
@@ -231,6 +233,24 @@ static __cold int io_init_bl_list(struct
return 0;
}
+/*
+ * Mark the given mapped range as free for reuse
+ */
+static void io_kbuf_mark_free(struct io_ring_ctx *ctx, struct io_buffer_list *bl)
+{
+ struct io_buf_free *ibf;
+
+ hlist_for_each_entry(ibf, &ctx->io_buf_list, list) {
+ if (bl->buf_ring == ibf->mem) {
+ ibf->inuse = 0;
+ return;
+ }
+ }
+
+ /* can't happen... */
+ WARN_ON_ONCE(1);
+}
+
static int __io_remove_buffers(struct io_ring_ctx *ctx,
struct io_buffer_list *bl, unsigned nbufs)
{
@@ -247,6 +267,7 @@ static int __io_remove_buffers(struct io
* io_kbuf_list_free() will free the page(s) at
* ->release() time.
*/
+ io_kbuf_mark_free(ctx, bl);
bl->buf_ring = NULL;
bl->is_mmap = 0;
} else if (bl->buf_nr_pages) {
@@ -560,6 +581,34 @@ error_unpin:
return -EINVAL;
}
+/*
+ * See if we have a suitable region that we can reuse, rather than allocate
+ * both a new io_buf_free and mem region again. We leave it on the list as
+ * even a reused entry will need freeing at ring release.
+ */
+static struct io_buf_free *io_lookup_buf_free_entry(struct io_ring_ctx *ctx,
+ size_t ring_size)
+{
+ struct io_buf_free *ibf, *best = NULL;
+ size_t best_dist;
+
+ hlist_for_each_entry(ibf, &ctx->io_buf_list, list) {
+ size_t dist;
+
+ if (ibf->inuse || ibf->size < ring_size)
+ continue;
+ dist = ibf->size - ring_size;
+ if (!best || dist < best_dist) {
+ best = ibf;
+ if (!dist)
+ break;
+ best_dist = dist;
+ }
+ }
+
+ return best;
+}
+
static int io_alloc_pbuf_ring(struct io_ring_ctx *ctx,
struct io_uring_buf_reg *reg,
struct io_buffer_list *bl)
@@ -569,20 +618,26 @@ static int io_alloc_pbuf_ring(struct io_
void *ptr;
ring_size = reg->ring_entries * sizeof(struct io_uring_buf_ring);
- ptr = io_mem_alloc(ring_size);
- if (!ptr)
- return -ENOMEM;
- /* Allocate and store deferred free entry */
- ibf = kmalloc(sizeof(*ibf), GFP_KERNEL_ACCOUNT);
+ /* Reuse existing entry, if we can */
+ ibf = io_lookup_buf_free_entry(ctx, ring_size);
if (!ibf) {
- io_mem_free(ptr);
- return -ENOMEM;
+ ptr = io_mem_alloc(ring_size);
+ if (!ptr)
+ return -ENOMEM;
+
+ /* Allocate and store deferred free entry */
+ ibf = kmalloc(sizeof(*ibf), GFP_KERNEL_ACCOUNT);
+ if (!ibf) {
+ io_mem_free(ptr);
+ return -ENOMEM;
+ }
+ ibf->mem = ptr;
+ ibf->size = ring_size;
+ hlist_add_head(&ibf->list, &ctx->io_buf_list);
}
- ibf->mem = ptr;
- hlist_add_head(&ibf->list, &ctx->io_buf_list);
-
- bl->buf_ring = ptr;
+ ibf->inuse = 1;
+ bl->buf_ring = ibf->mem;
bl->is_mapped = 1;
bl->is_mmap = 1;
return 0;
Patches currently in stable-queue which might be from axboe@xxxxxxxxx are
queue-6.6/io_uring-kbuf-recycle-freed-mapped-buffer-ring-entries.patch
queue-6.6/io_uring-enable-io_mem_alloc-free-to-be-used-in-other-parts.patch
queue-6.6/io_uring-kbuf-defer-release-of-mapped-buffer-rings.patch
queue-6.6/io_uring-don-t-allow-discontig-pages-for-ioring_setup_no_mmap.patch
queue-6.6/bcache-revert-replacing-is_err_or_null-with-is_err.patch
queue-6.6/powerpc-don-t-clobber-f0-vs0-during-fp-altivec-register-save.patch
queue-6.6/io_uring-free-io_buffer_list-entries-via-rcu.patch
queue-6.6/io_uring-don-t-guard-ioring_off_pbuf_ring-with-setup_no_mmap.patch
