On 9/21/24 1:59 AM, Jens Axboe wrote:> Everything else about the io_uring eventfd support is nicely kept
> private to that code, except the cached_cq_tail tracking. With
> everything else in place, move io_eventfd_flush_signal() to using
> the ev_fd grab+release helpers, which then enables the direct use of
> io_ev_fd for this tracking too.
Of course forgot to refresh io_uring_types.h as well, so this patch is
missing the hunk that deletes it from io_ring_ctx. Refreshed version
below, that's the only change.
diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 4b9ba523978d..0deec302595a 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -413,10 +413,6 @@ struct io_ring_ctx {
DECLARE_HASHTABLE(napi_ht, 4);
#endif
-
- /* protected by ->completion_lock */
- unsigned evfd_last_cq_tail;
-
/*
* If IORING_SETUP_NO_MMAP is used, then the below holds
* the gup'ed pages for the two rings, and the sqes.
diff --git a/io_uring/eventfd.c b/io_uring/eventfd.c
index d1fdecd0c458..fab936d31ba8 100644
--- a/io_uring/eventfd.c
+++ b/io_uring/eventfd.c
@@ -13,10 +13,12 @@
struct io_ev_fd {
struct eventfd_ctx *cq_ev_fd;
- unsigned int eventfd_async: 1;
- struct rcu_head rcu;
+ unsigned int eventfd_async;
+ /* protected by ->completion_lock */
+ unsigned last_cq_tail;
refcount_t refs;
atomic_t ops;
+ struct rcu_head rcu;
};
enum {
@@ -123,25 +125,31 @@ void io_eventfd_signal(struct io_ring_ctx *ctx)
void io_eventfd_flush_signal(struct io_ring_ctx *ctx)
{
- bool skip;
-
- spin_lock(&ctx->completion_lock);
-
- /*
- * Eventfd should only get triggered when at least one event has been
- * posted. Some applications rely on the eventfd notification count
- * only changing IFF a new CQE has been added to the CQ ring. There's
- * no depedency on 1:1 relationship between how many times this
- * function is called (and hence the eventfd count) and number of CQEs
- * posted to the CQ ring.
- */
- skip = ctx->cached_cq_tail == ctx->evfd_last_cq_tail;
- ctx->evfd_last_cq_tail = ctx->cached_cq_tail;
- spin_unlock(&ctx->completion_lock);
- if (skip)
- return;
+ struct io_ev_fd *ev_fd;
- io_eventfd_signal(ctx);
+ ev_fd = io_eventfd_grab(ctx);
+ if (ev_fd) {
+ bool skip, put_ref = true;
+
+ /*
+ * Eventfd should only get triggered when at least one event
+ * has been posted. Some applications rely on the eventfd
+ * notification count only changing IFF a new CQE has been
+ * added to the CQ ring. There's no dependency on 1:1
+ * relationship between how many times this function is called
+ * (and hence the eventfd count) and number of CQEs posted to
+ * the CQ ring.
+ */
+ spin_lock(&ctx->completion_lock);
+ skip = ctx->cached_cq_tail == ev_fd->last_cq_tail;
+ ev_fd->last_cq_tail = ctx->cached_cq_tail;
+ spin_unlock(&ctx->completion_lock);
+
+ if (!skip)
+ put_ref = __io_eventfd_signal(ev_fd);
+
+ io_eventfd_release(ev_fd, put_ref);
+ }
}
int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg,
@@ -172,7 +180,7 @@ int io_eventfd_register(struct io_ring_ctx *ctx, void __user *arg,
}
spin_lock(&ctx->completion_lock);
- ctx->evfd_last_cq_tail = ctx->cached_cq_tail;
+ ev_fd->last_cq_tail = ctx->cached_cq_tail;
spin_unlock(&ctx->completion_lock);
ev_fd->eventfd_async = eventfd_async;
--
Jens Axboe
