On 1/8/20 9:50 AM, Jens Axboe wrote:
> On 1/8/20 9:46 AM, Mark Papadakis wrote:
>> Thus sounds perfect!
>
> I'll try and cook this up, if you can test it.
Something like the below.
diff --git a/fs/io_uring.c b/fs/io_uring.c
index 06fdda43163d..70478888bb16 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -206,6 +206,7 @@ struct io_ring_ctx {
bool account_mem;
bool cq_overflow_flushed;
bool drain_next;
+ bool eventfd_async;
/*
* Ring buffer of indices into array of io_uring_sqe, which is
@@ -960,13 +961,20 @@ static struct io_uring_cqe *io_get_cqring(struct io_ring_ctx *ctx)
return &rings->cqes[tail & ctx->cq_mask];
}
+static inline bool io_should_trigger_evfd(struct io_ring_ctx *ctx)
+{
+ if (!ctx->eventfd_async)
+ return true;
+ return io_wq_current_is_worker() || in_interrupt();
+}
+
static void io_cqring_ev_posted(struct io_ring_ctx *ctx)
{
if (waitqueue_active(&ctx->wait))
wake_up(&ctx->wait);
if (waitqueue_active(&ctx->sqo_wait))
wake_up(&ctx->sqo_wait);
- if (ctx->cq_ev_fd)
+ if (ctx->cq_ev_fd && io_should_trigger_evfd(ctx))
eventfd_signal(ctx->cq_ev_fd, 1);
}
@@ -6552,10 +6560,17 @@ static int __io_uring_register(struct io_ring_ctx *ctx, unsigned opcode,
ret = io_sqe_files_update(ctx, arg, nr_args);
break;
case IORING_REGISTER_EVENTFD:
+ case IORING_REGISTER_EVENTFD_ASYNC:
ret = -EINVAL;
if (nr_args != 1)
break;
ret = io_eventfd_register(ctx, arg);
+ if (ret)
+ break;
+ if (opcode == IORING_REGISTER_EVENTFD_ASYNC)
+ ctx->eventfd_async = true;
+ else
+ ctx->eventfd_async = false;
break;
case IORING_UNREGISTER_EVENTFD:
ret = -EINVAL;
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index be64c9757ff1..02e0b2d59b63 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -192,6 +192,7 @@ struct io_uring_params {
#define IORING_REGISTER_EVENTFD 4
#define IORING_UNREGISTER_EVENTFD 5
#define IORING_REGISTER_FILES_UPDATE 6
+#define IORING_REGISTER_EVENTFD_ASYNC 7
struct io_uring_files_update {
__u32 offset;
--
Jens Axboe
