Merge tag 'for-5.4/io_uring-2019-09-15' of git://git.kernel.dk/linux-block

Pull io_uring updates from Jens Axboe:

 - Allocate SQ/CQ ring together, more efficient. Expose this through a
   feature flag as well, so we can reduce the number of mmaps by 1
   (Hristo and me)

 - Fix for sequence logic with SQ thread (Jackie).

 - Add support for links with drain commands (Jackie).

 - Improved async merging (me)

 - Improved buffered async write performance (me)

 - Support SQ poll wakeup + event get in single io_uring_enter() (me)

 - Support larger SQ ring size. For epoll conversions, the 4k limit was
   too small for some prod workloads (Daniel).

 - put_user_page() usage (John)

* tag 'for-5.4/io_uring-2019-09-15' of git://git.kernel.dk/linux-block:
  io_uring: increase IORING_MAX_ENTRIES to 32K
  io_uring: make sqpoll wakeup possible with getevents
  io_uring: extend async work merging
  io_uring: limit parallelism of buffered writes
  io_uring: add io_queue_async_work() helper
  io_uring: optimize submit_and_wait API
  io_uring: add support for link with drain
  io_uring: fix wrong sequence setting logic
  io_uring: expose single mmap capability
  io_uring: allocate the two rings together
  fs/io_uring.c: convert put_page() to put_user_page*()

include/uapi/linux/io_uring.h

@@ -128,11 +128,17 @@ struct io_uring_params {
 	__u32 flags;
 	__u32 sq_thread_cpu;
 	__u32 sq_thread_idle;
-	__u32 resv[5];
+	__u32 features;
+	__u32 resv[4];
 	struct io_sqring_offsets sq_off;
 	struct io_cqring_offsets cq_off;
 };
 
+/*
+ * io_uring_params->features flags
+ */
+#define IORING_FEAT_SINGLE_MMAP		(1U << 0)
+
 /*
  * io_uring_register(2) opcodes and arguments
  */