[PATCHSET next 0/6] Misc cleanups / optimizations

[PATCHSET next 0/6] Misc cleanups / optimizations

[Jens Axboe]

Hi,

Nothing major in here:

- Expand io_kiocb flags to 64-bits, so we can use two more bits for
  caching cancelation sequence and pollable state.
- Misc cleanups

-- 
Jens Axboe

[PATCH 1/6] io_uring: expand main struct io_kiocb flags to 64-bits

[Jens Axboe]

We're out of space here, and none of the flags are easily reclaimable.
Bump it to 64-bits and re-arrange the struct a bit to avoid gaps.

Add a specific bitwise type for the request flags, io_request_flags_t.
This will help catch violations of casting this value to a smaller type
on 32-bit archs, like unsigned int.

No functional changes intended in this patch.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/linux/io_uring_types.h  | 87 ++++++++++++++++++---------------
 include/trace/events/io_uring.h | 14 +++---
 io_uring/filetable.h            |  2 +-
 io_uring/io_uring.c             |  9 ++--
 4 files changed, 60 insertions(+), 52 deletions(-)

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 854ad67a5f70..5ac18b05d4ee 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -428,7 +428,7 @@ struct io_tw_state {
 	bool locked;
 };
 
-enum {
+enum io_req_flags {
 	REQ_F_FIXED_FILE_BIT	= IOSQE_FIXED_FILE_BIT,
 	REQ_F_IO_DRAIN_BIT	= IOSQE_IO_DRAIN_BIT,
 	REQ_F_LINK_BIT		= IOSQE_IO_LINK_BIT,
@@ -468,70 +468,73 @@ enum {
 	__REQ_F_LAST_BIT,
 };
 
+typedef enum io_req_flags __bitwise io_req_flags_t;
+#define IO_REQ_FLAG(bitno)	((__force io_req_flags_t) BIT_ULL((bitno)))
+
 enum {
 	/* ctx owns file */
-	REQ_F_FIXED_FILE	= BIT(REQ_F_FIXED_FILE_BIT),
+	REQ_F_FIXED_FILE	= IO_REQ_FLAG(REQ_F_FIXED_FILE_BIT),
 	/* drain existing IO first */
-	REQ_F_IO_DRAIN		= BIT(REQ_F_IO_DRAIN_BIT),
+	REQ_F_IO_DRAIN		= IO_REQ_FLAG(REQ_F_IO_DRAIN_BIT),
 	/* linked sqes */
-	REQ_F_LINK		= BIT(REQ_F_LINK_BIT),
+	REQ_F_LINK		= IO_REQ_FLAG(REQ_F_LINK_BIT),
 	/* doesn't sever on completion < 0 */
-	REQ_F_HARDLINK		= BIT(REQ_F_HARDLINK_BIT),
+	REQ_F_HARDLINK		= IO_REQ_FLAG(REQ_F_HARDLINK_BIT),
 	/* IOSQE_ASYNC */
-	REQ_F_FORCE_ASYNC	= BIT(REQ_F_FORCE_ASYNC_BIT),
+	REQ_F_FORCE_ASYNC	= IO_REQ_FLAG(REQ_F_FORCE_ASYNC_BIT),
 	/* IOSQE_BUFFER_SELECT */
-	REQ_F_BUFFER_SELECT	= BIT(REQ_F_BUFFER_SELECT_BIT),
+	REQ_F_BUFFER_SELECT	= IO_REQ_FLAG(REQ_F_BUFFER_SELECT_BIT),
 	/* IOSQE_CQE_SKIP_SUCCESS */
-	REQ_F_CQE_SKIP		= BIT(REQ_F_CQE_SKIP_BIT),
+	REQ_F_CQE_SKIP		= IO_REQ_FLAG(REQ_F_CQE_SKIP_BIT),
 
 	/* fail rest of links */
-	REQ_F_FAIL		= BIT(REQ_F_FAIL_BIT),
+	REQ_F_FAIL		= IO_REQ_FLAG(REQ_F_FAIL_BIT),
 	/* on inflight list, should be cancelled and waited on exit reliably */
-	REQ_F_INFLIGHT		= BIT(REQ_F_INFLIGHT_BIT),
+	REQ_F_INFLIGHT		= IO_REQ_FLAG(REQ_F_INFLIGHT_BIT),
 	/* read/write uses file position */
-	REQ_F_CUR_POS		= BIT(REQ_F_CUR_POS_BIT),
+	REQ_F_CUR_POS		= IO_REQ_FLAG(REQ_F_CUR_POS_BIT),
 	/* must not punt to workers */
-	REQ_F_NOWAIT		= BIT(REQ_F_NOWAIT_BIT),
+	REQ_F_NOWAIT		= IO_REQ_FLAG(REQ_F_NOWAIT_BIT),
 	/* has or had linked timeout */
-	REQ_F_LINK_TIMEOUT	= BIT(REQ_F_LINK_TIMEOUT_BIT),
+	REQ_F_LINK_TIMEOUT	= IO_REQ_FLAG(REQ_F_LINK_TIMEOUT_BIT),
 	/* needs cleanup */
-	REQ_F_NEED_CLEANUP	= BIT(REQ_F_NEED_CLEANUP_BIT),
+	REQ_F_NEED_CLEANUP	= IO_REQ_FLAG(REQ_F_NEED_CLEANUP_BIT),
 	/* already went through poll handler */
-	REQ_F_POLLED		= BIT(REQ_F_POLLED_BIT),
+	REQ_F_POLLED		= IO_REQ_FLAG(REQ_F_POLLED_BIT),
 	/* buffer already selected */
-	REQ_F_BUFFER_SELECTED	= BIT(REQ_F_BUFFER_SELECTED_BIT),
+	REQ_F_BUFFER_SELECTED	= IO_REQ_FLAG(REQ_F_BUFFER_SELECTED_BIT),
 	/* buffer selected from ring, needs commit */
-	REQ_F_BUFFER_RING	= BIT(REQ_F_BUFFER_RING_BIT),
+	REQ_F_BUFFER_RING	= IO_REQ_FLAG(REQ_F_BUFFER_RING_BIT),
 	/* caller should reissue async */
-	REQ_F_REISSUE		= BIT(REQ_F_REISSUE_BIT),
+	REQ_F_REISSUE		= IO_REQ_FLAG(REQ_F_REISSUE_BIT),
 	/* supports async reads/writes */
-	REQ_F_SUPPORT_NOWAIT	= BIT(REQ_F_SUPPORT_NOWAIT_BIT),
+	REQ_F_SUPPORT_NOWAIT	= IO_REQ_FLAG(REQ_F_SUPPORT_NOWAIT_BIT),
 	/* regular file */
-	REQ_F_ISREG		= BIT(REQ_F_ISREG_BIT),
+	REQ_F_ISREG		= IO_REQ_FLAG(REQ_F_ISREG_BIT),
 	/* has creds assigned */
-	REQ_F_CREDS		= BIT(REQ_F_CREDS_BIT),
+	REQ_F_CREDS		= IO_REQ_FLAG(REQ_F_CREDS_BIT),
 	/* skip refcounting if not set */
-	REQ_F_REFCOUNT		= BIT(REQ_F_REFCOUNT_BIT),
+	REQ_F_REFCOUNT		= IO_REQ_FLAG(REQ_F_REFCOUNT_BIT),
 	/* there is a linked timeout that has to be armed */
-	REQ_F_ARM_LTIMEOUT	= BIT(REQ_F_ARM_LTIMEOUT_BIT),
+	REQ_F_ARM_LTIMEOUT	= IO_REQ_FLAG(REQ_F_ARM_LTIMEOUT_BIT),
 	/* ->async_data allocated */
-	REQ_F_ASYNC_DATA	= BIT(REQ_F_ASYNC_DATA_BIT),
+	REQ_F_ASYNC_DATA	= IO_REQ_FLAG(REQ_F_ASYNC_DATA_BIT),
 	/* don't post CQEs while failing linked requests */
-	REQ_F_SKIP_LINK_CQES	= BIT(REQ_F_SKIP_LINK_CQES_BIT),
+	REQ_F_SKIP_LINK_CQES	= IO_REQ_FLAG(REQ_F_SKIP_LINK_CQES_BIT),
 	/* single poll may be active */
-	REQ_F_SINGLE_POLL	= BIT(REQ_F_SINGLE_POLL_BIT),
+	REQ_F_SINGLE_POLL	= IO_REQ_FLAG(REQ_F_SINGLE_POLL_BIT),
 	/* double poll may active */
-	REQ_F_DOUBLE_POLL	= BIT(REQ_F_DOUBLE_POLL_BIT),
+	REQ_F_DOUBLE_POLL	= IO_REQ_FLAG(REQ_F_DOUBLE_POLL_BIT),
 	/* request has already done partial IO */
-	REQ_F_PARTIAL_IO	= BIT(REQ_F_PARTIAL_IO_BIT),
+	REQ_F_PARTIAL_IO	= IO_REQ_FLAG(REQ_F_PARTIAL_IO_BIT),
 	/* fast poll multishot mode */
-	REQ_F_APOLL_MULTISHOT	= BIT(REQ_F_APOLL_MULTISHOT_BIT),
+	REQ_F_APOLL_MULTISHOT	= IO_REQ_FLAG(REQ_F_APOLL_MULTISHOT_BIT),
 	/* recvmsg special flag, clear EPOLLIN */
-	REQ_F_CLEAR_POLLIN	= BIT(REQ_F_CLEAR_POLLIN_BIT),
+	REQ_F_CLEAR_POLLIN	= IO_REQ_FLAG(REQ_F_CLEAR_POLLIN_BIT),
 	/* hashed into ->cancel_hash_locked, protected by ->uring_lock */
-	REQ_F_HASH_LOCKED	= BIT(REQ_F_HASH_LOCKED_BIT),
+	REQ_F_HASH_LOCKED	= IO_REQ_FLAG(REQ_F_HASH_LOCKED_BIT),
 	/* don't use lazy poll wake for this request */
-	REQ_F_POLL_NO_LAZY	= BIT(REQ_F_POLL_NO_LAZY_BIT),
+	REQ_F_POLL_NO_LAZY	= IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT),
 };
 
 typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
@@ -592,15 +595,14 @@ struct io_kiocb {
 	 * and after selection it points to the buffer ID itself.
 	 */
 	u16				buf_index;
-	unsigned int			flags;
 
-	struct io_cqe			cqe;
+	atomic_t			refs;
+
+	io_req_flags_t			flags;
 
 	struct io_ring_ctx		*ctx;
 	struct task_struct		*task;
 
-	struct io_rsrc_node		*rsrc_node;
-
 	union {
 		/* store used ubuf, so we can prevent reloading */
 		struct io_mapped_ubuf	*imu;
@@ -615,18 +617,23 @@ struct io_kiocb {
 		struct io_buffer_list	*buf_list;
 	};
 
+	/* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
+	struct hlist_node		hash_node;
+
 	union {
 		/* used by request caches, completion batching and iopoll */
 		struct io_wq_work_node	comp_list;
 		/* cache ->apoll->events */
 		__poll_t apoll_events;
 	};
-	atomic_t			refs;
-	atomic_t			poll_refs;
+
+	struct io_rsrc_node		*rsrc_node;
+
+	struct io_cqe			cqe;
+
 	struct io_task_work		io_task_work;
+	atomic_t			poll_refs;
 	unsigned			nr_tw;
-	/* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
-	struct hlist_node		hash_node;
 	/* internal polling, see IORING_FEAT_FAST_POLL */
 	struct async_poll		*apoll;
 	/* opcode allocated if it needs to store data for async defer */
diff --git a/include/trace/events/io_uring.h b/include/trace/events/io_uring.h
index 69454f1f98b0..3d7704a52b73 100644
--- a/include/trace/events/io_uring.h
+++ b/include/trace/events/io_uring.h
@@ -148,7 +148,7 @@ TRACE_EVENT(io_uring_queue_async_work,
 		__field(  void *,			req		)
 		__field(  u64,				user_data	)
 		__field(  u8,				opcode		)
-		__field(  unsigned int,			flags		)
+		__field(  io_req_flags_t,		flags		)
 		__field(  struct io_wq_work *,		work		)
 		__field(  int,				rw		)
 
@@ -167,10 +167,10 @@ TRACE_EVENT(io_uring_queue_async_work,
 		__assign_str(op_str, io_uring_get_opcode(req->opcode));
 	),
 
-	TP_printk("ring %p, request %p, user_data 0x%llx, opcode %s, flags 0x%x, %s queue, work %p",
+	TP_printk("ring %p, request %p, user_data 0x%llx, opcode %s, flags 0x%lx, %s queue, work %p",
 		__entry->ctx, __entry->req, __entry->user_data,
-		__get_str(op_str),
-		__entry->flags, __entry->rw ? "hashed" : "normal", __entry->work)
+		__get_str(op_str), (long) __entry->flags,
+		__entry->rw ? "hashed" : "normal", __entry->work)
 );
 
 /**
@@ -378,7 +378,7 @@ TRACE_EVENT(io_uring_submit_req,
 		__field(  void *,		req		)
 		__field(  unsigned long long,	user_data	)
 		__field(  u8,			opcode		)
-		__field(  u32,			flags		)
+		__field(  io_req_flags_t,	flags		)
 		__field(  bool,			sq_thread	)
 
 		__string( op_str, io_uring_get_opcode(req->opcode) )
@@ -395,10 +395,10 @@ TRACE_EVENT(io_uring_submit_req,
 		__assign_str(op_str, io_uring_get_opcode(req->opcode));
 	),
 
-	TP_printk("ring %p, req %p, user_data 0x%llx, opcode %s, flags 0x%x, "
+	TP_printk("ring %p, req %p, user_data 0x%llx, opcode %s, flags 0x%lx, "
 		  "sq_thread %d", __entry->ctx, __entry->req,
 		  __entry->user_data, __get_str(op_str),
-		  __entry->flags, __entry->sq_thread)
+		  (long) __entry->flags, __entry->sq_thread)
 );
 
 /*
diff --git a/io_uring/filetable.h b/io_uring/filetable.h
index b47adf170c31..b2435c4dca1f 100644
--- a/io_uring/filetable.h
+++ b/io_uring/filetable.h
@@ -17,7 +17,7 @@ int io_fixed_fd_remove(struct io_ring_ctx *ctx, unsigned int offset);
 int io_register_file_alloc_range(struct io_ring_ctx *ctx,
 				 struct io_uring_file_index_range __user *arg);
 
-unsigned int io_file_get_flags(struct file *file);
+io_req_flags_t io_file_get_flags(struct file *file);
 
 static inline void io_file_bitmap_clear(struct io_file_table *table, int bit)
 {
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index cd9a137ad6ce..360a7ee41d3a 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -1768,9 +1768,9 @@ static void io_iopoll_req_issued(struct io_kiocb *req, unsigned int issue_flags)
 	}
 }
 
-unsigned int io_file_get_flags(struct file *file)
+io_req_flags_t io_file_get_flags(struct file *file)
 {
-	unsigned int res = 0;
+	io_req_flags_t res = 0;
 
 	if (S_ISREG(file_inode(file)->i_mode))
 		res |= REQ_F_ISREG;
@@ -2171,7 +2171,8 @@ static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
 	/* req is partially pre-initialised, see io_preinit_req() */
 	req->opcode = opcode = READ_ONCE(sqe->opcode);
 	/* same numerical values with corresponding REQ_F_*, safe to copy */
-	req->flags = sqe_flags = READ_ONCE(sqe->flags);
+	sqe_flags = READ_ONCE(sqe->flags);
+	req->flags = (io_req_flags_t) sqe_flags;
 	req->cqe.user_data = READ_ONCE(sqe->user_data);
 	req->file = NULL;
 	req->rsrc_node = NULL;
@@ -4153,7 +4154,7 @@ static int __init io_uring_init(void)
 	BUILD_BUG_ON(SQE_COMMON_FLAGS >= (1 << 8));
 	BUILD_BUG_ON((SQE_VALID_FLAGS | SQE_COMMON_FLAGS) != SQE_VALID_FLAGS);
 
-	BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(int));
+	BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(u64));
 
 	BUILD_BUG_ON(sizeof(atomic_t) != sizeof(u32));
 
-- 
2.43.0

Re: [PATCH 1/6] io_uring: expand main struct io_kiocb flags to 64-bits

[Jens Axboe]

On 2/6/24 9:22 AM, Jens Axboe wrote:
> We're out of space here, and none of the flags are easily reclaimable.
> Bump it to 64-bits and re-arrange the struct a bit to avoid gaps.
> 
> Add a specific bitwise type for the request flags, io_request_flags_t.
> This will help catch violations of casting this value to a smaller type
> on 32-bit archs, like unsigned int.
> 
> No functional changes intended in this patch.

Looks like I didn't run this through testing after making the 32-bit
change, my bad. Here's v2 of it, no changes to the rest.

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 854ad67a5f70..5aef24ef467b 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -468,70 +468,73 @@ enum {
 	__REQ_F_LAST_BIT,
 };
 
+typedef u64 __bitwise io_req_flags_t;
+#define IO_REQ_FLAG(bitno)	((__force io_req_flags_t) BIT_ULL((bitno)))
+
 enum {
 	/* ctx owns file */
-	REQ_F_FIXED_FILE	= BIT(REQ_F_FIXED_FILE_BIT),
+	REQ_F_FIXED_FILE	= IO_REQ_FLAG(REQ_F_FIXED_FILE_BIT),
 	/* drain existing IO first */
-	REQ_F_IO_DRAIN		= BIT(REQ_F_IO_DRAIN_BIT),
+	REQ_F_IO_DRAIN		= IO_REQ_FLAG(REQ_F_IO_DRAIN_BIT),
 	/* linked sqes */
-	REQ_F_LINK		= BIT(REQ_F_LINK_BIT),
+	REQ_F_LINK		= IO_REQ_FLAG(REQ_F_LINK_BIT),
 	/* doesn't sever on completion < 0 */
-	REQ_F_HARDLINK		= BIT(REQ_F_HARDLINK_BIT),
+	REQ_F_HARDLINK		= IO_REQ_FLAG(REQ_F_HARDLINK_BIT),
 	/* IOSQE_ASYNC */
-	REQ_F_FORCE_ASYNC	= BIT(REQ_F_FORCE_ASYNC_BIT),
+	REQ_F_FORCE_ASYNC	= IO_REQ_FLAG(REQ_F_FORCE_ASYNC_BIT),
 	/* IOSQE_BUFFER_SELECT */
-	REQ_F_BUFFER_SELECT	= BIT(REQ_F_BUFFER_SELECT_BIT),
+	REQ_F_BUFFER_SELECT	= IO_REQ_FLAG(REQ_F_BUFFER_SELECT_BIT),
 	/* IOSQE_CQE_SKIP_SUCCESS */
-	REQ_F_CQE_SKIP		= BIT(REQ_F_CQE_SKIP_BIT),
+	REQ_F_CQE_SKIP		= IO_REQ_FLAG(REQ_F_CQE_SKIP_BIT),
 
 	/* fail rest of links */
-	REQ_F_FAIL		= BIT(REQ_F_FAIL_BIT),
+	REQ_F_FAIL		= IO_REQ_FLAG(REQ_F_FAIL_BIT),
 	/* on inflight list, should be cancelled and waited on exit reliably */
-	REQ_F_INFLIGHT		= BIT(REQ_F_INFLIGHT_BIT),
+	REQ_F_INFLIGHT		= IO_REQ_FLAG(REQ_F_INFLIGHT_BIT),
 	/* read/write uses file position */
-	REQ_F_CUR_POS		= BIT(REQ_F_CUR_POS_BIT),
+	REQ_F_CUR_POS		= IO_REQ_FLAG(REQ_F_CUR_POS_BIT),
 	/* must not punt to workers */
-	REQ_F_NOWAIT		= BIT(REQ_F_NOWAIT_BIT),
+	REQ_F_NOWAIT		= IO_REQ_FLAG(REQ_F_NOWAIT_BIT),
 	/* has or had linked timeout */
-	REQ_F_LINK_TIMEOUT	= BIT(REQ_F_LINK_TIMEOUT_BIT),
+	REQ_F_LINK_TIMEOUT	= IO_REQ_FLAG(REQ_F_LINK_TIMEOUT_BIT),
 	/* needs cleanup */
-	REQ_F_NEED_CLEANUP	= BIT(REQ_F_NEED_CLEANUP_BIT),
+	REQ_F_NEED_CLEANUP	= IO_REQ_FLAG(REQ_F_NEED_CLEANUP_BIT),
 	/* already went through poll handler */
-	REQ_F_POLLED		= BIT(REQ_F_POLLED_BIT),
+	REQ_F_POLLED		= IO_REQ_FLAG(REQ_F_POLLED_BIT),
 	/* buffer already selected */
-	REQ_F_BUFFER_SELECTED	= BIT(REQ_F_BUFFER_SELECTED_BIT),
+	REQ_F_BUFFER_SELECTED	= IO_REQ_FLAG(REQ_F_BUFFER_SELECTED_BIT),
 	/* buffer selected from ring, needs commit */
-	REQ_F_BUFFER_RING	= BIT(REQ_F_BUFFER_RING_BIT),
+	REQ_F_BUFFER_RING	= IO_REQ_FLAG(REQ_F_BUFFER_RING_BIT),
 	/* caller should reissue async */
-	REQ_F_REISSUE		= BIT(REQ_F_REISSUE_BIT),
+	REQ_F_REISSUE		= IO_REQ_FLAG(REQ_F_REISSUE_BIT),
 	/* supports async reads/writes */
-	REQ_F_SUPPORT_NOWAIT	= BIT(REQ_F_SUPPORT_NOWAIT_BIT),
+	REQ_F_SUPPORT_NOWAIT	= IO_REQ_FLAG(REQ_F_SUPPORT_NOWAIT_BIT),
 	/* regular file */
-	REQ_F_ISREG		= BIT(REQ_F_ISREG_BIT),
+	REQ_F_ISREG		= IO_REQ_FLAG(REQ_F_ISREG_BIT),
 	/* has creds assigned */
-	REQ_F_CREDS		= BIT(REQ_F_CREDS_BIT),
+	REQ_F_CREDS		= IO_REQ_FLAG(REQ_F_CREDS_BIT),
 	/* skip refcounting if not set */
-	REQ_F_REFCOUNT		= BIT(REQ_F_REFCOUNT_BIT),
+	REQ_F_REFCOUNT		= IO_REQ_FLAG(REQ_F_REFCOUNT_BIT),
 	/* there is a linked timeout that has to be armed */
-	REQ_F_ARM_LTIMEOUT	= BIT(REQ_F_ARM_LTIMEOUT_BIT),
+	REQ_F_ARM_LTIMEOUT	= IO_REQ_FLAG(REQ_F_ARM_LTIMEOUT_BIT),
 	/* ->async_data allocated */
-	REQ_F_ASYNC_DATA	= BIT(REQ_F_ASYNC_DATA_BIT),
+	REQ_F_ASYNC_DATA	= IO_REQ_FLAG(REQ_F_ASYNC_DATA_BIT),
 	/* don't post CQEs while failing linked requests */
-	REQ_F_SKIP_LINK_CQES	= BIT(REQ_F_SKIP_LINK_CQES_BIT),
+	REQ_F_SKIP_LINK_CQES	= IO_REQ_FLAG(REQ_F_SKIP_LINK_CQES_BIT),
 	/* single poll may be active */
-	REQ_F_SINGLE_POLL	= BIT(REQ_F_SINGLE_POLL_BIT),
+	REQ_F_SINGLE_POLL	= IO_REQ_FLAG(REQ_F_SINGLE_POLL_BIT),
 	/* double poll may active */
-	REQ_F_DOUBLE_POLL	= BIT(REQ_F_DOUBLE_POLL_BIT),
+	REQ_F_DOUBLE_POLL	= IO_REQ_FLAG(REQ_F_DOUBLE_POLL_BIT),
 	/* request has already done partial IO */
-	REQ_F_PARTIAL_IO	= BIT(REQ_F_PARTIAL_IO_BIT),
+	REQ_F_PARTIAL_IO	= IO_REQ_FLAG(REQ_F_PARTIAL_IO_BIT),
 	/* fast poll multishot mode */
-	REQ_F_APOLL_MULTISHOT	= BIT(REQ_F_APOLL_MULTISHOT_BIT),
+	REQ_F_APOLL_MULTISHOT	= IO_REQ_FLAG(REQ_F_APOLL_MULTISHOT_BIT),
 	/* recvmsg special flag, clear EPOLLIN */
-	REQ_F_CLEAR_POLLIN	= BIT(REQ_F_CLEAR_POLLIN_BIT),
+	REQ_F_CLEAR_POLLIN	= IO_REQ_FLAG(REQ_F_CLEAR_POLLIN_BIT),
 	/* hashed into ->cancel_hash_locked, protected by ->uring_lock */
-	REQ_F_HASH_LOCKED	= BIT(REQ_F_HASH_LOCKED_BIT),
+	REQ_F_HASH_LOCKED	= IO_REQ_FLAG(REQ_F_HASH_LOCKED_BIT),
 	/* don't use lazy poll wake for this request */
-	REQ_F_POLL_NO_LAZY	= BIT(REQ_F_POLL_NO_LAZY_BIT),
+	REQ_F_POLL_NO_LAZY	= IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT),
 };
 
 typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
@@ -592,15 +595,14 @@ struct io_kiocb {
 	 * and after selection it points to the buffer ID itself.
 	 */
 	u16				buf_index;
-	unsigned int			flags;
 
-	struct io_cqe			cqe;
+	atomic_t			refs;
+
+	io_req_flags_t			flags;
 
 	struct io_ring_ctx		*ctx;
 	struct task_struct		*task;
 
-	struct io_rsrc_node		*rsrc_node;
-
 	union {
 		/* store used ubuf, so we can prevent reloading */
 		struct io_mapped_ubuf	*imu;
@@ -615,18 +617,23 @@ struct io_kiocb {
 		struct io_buffer_list	*buf_list;
 	};
 
+	/* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
+	struct hlist_node		hash_node;
+
 	union {
 		/* used by request caches, completion batching and iopoll */
 		struct io_wq_work_node	comp_list;
 		/* cache ->apoll->events */
 		__poll_t apoll_events;
 	};
-	atomic_t			refs;
-	atomic_t			poll_refs;
+
+	struct io_rsrc_node		*rsrc_node;
+
+	struct io_cqe			cqe;
+
 	struct io_task_work		io_task_work;
+	atomic_t			poll_refs;
 	unsigned			nr_tw;
-	/* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
-	struct hlist_node		hash_node;
 	/* internal polling, see IORING_FEAT_FAST_POLL */
 	struct async_poll		*apoll;
 	/* opcode allocated if it needs to store data for async defer */
diff --git a/include/trace/events/io_uring.h b/include/trace/events/io_uring.h
index 69454f1f98b0..3d7704a52b73 100644
--- a/include/trace/events/io_uring.h
+++ b/include/trace/events/io_uring.h
@@ -148,7 +148,7 @@ TRACE_EVENT(io_uring_queue_async_work,
 		__field(  void *,			req		)
 		__field(  u64,				user_data	)
 		__field(  u8,				opcode		)
-		__field(  unsigned int,			flags		)
+		__field(  io_req_flags_t,		flags		)
 		__field(  struct io_wq_work *,		work		)
 		__field(  int,				rw		)
 
@@ -167,10 +167,10 @@ TRACE_EVENT(io_uring_queue_async_work,
 		__assign_str(op_str, io_uring_get_opcode(req->opcode));
 	),
 
-	TP_printk("ring %p, request %p, user_data 0x%llx, opcode %s, flags 0x%x, %s queue, work %p",
+	TP_printk("ring %p, request %p, user_data 0x%llx, opcode %s, flags 0x%lx, %s queue, work %p",
 		__entry->ctx, __entry->req, __entry->user_data,
-		__get_str(op_str),
-		__entry->flags, __entry->rw ? "hashed" : "normal", __entry->work)
+		__get_str(op_str), (long) __entry->flags,
+		__entry->rw ? "hashed" : "normal", __entry->work)
 );
 
 /**
@@ -378,7 +378,7 @@ TRACE_EVENT(io_uring_submit_req,
 		__field(  void *,		req		)
 		__field(  unsigned long long,	user_data	)
 		__field(  u8,			opcode		)
-		__field(  u32,			flags		)
+		__field(  io_req_flags_t,	flags		)
 		__field(  bool,			sq_thread	)
 
 		__string( op_str, io_uring_get_opcode(req->opcode) )
@@ -395,10 +395,10 @@ TRACE_EVENT(io_uring_submit_req,
 		__assign_str(op_str, io_uring_get_opcode(req->opcode));
 	),
 
-	TP_printk("ring %p, req %p, user_data 0x%llx, opcode %s, flags 0x%x, "
+	TP_printk("ring %p, req %p, user_data 0x%llx, opcode %s, flags 0x%lx, "
 		  "sq_thread %d", __entry->ctx, __entry->req,
 		  __entry->user_data, __get_str(op_str),
-		  __entry->flags, __entry->sq_thread)
+		  (long) __entry->flags, __entry->sq_thread)
 );
 
 /*
diff --git a/io_uring/filetable.h b/io_uring/filetable.h
index b47adf170c31..b2435c4dca1f 100644
--- a/io_uring/filetable.h
+++ b/io_uring/filetable.h
@@ -17,7 +17,7 @@ int io_fixed_fd_remove(struct io_ring_ctx *ctx, unsigned int offset);
 int io_register_file_alloc_range(struct io_ring_ctx *ctx,
 				 struct io_uring_file_index_range __user *arg);
 
-unsigned int io_file_get_flags(struct file *file);
+io_req_flags_t io_file_get_flags(struct file *file);
 
 static inline void io_file_bitmap_clear(struct io_file_table *table, int bit)
 {
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index cd9a137ad6ce..360a7ee41d3a 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -1768,9 +1768,9 @@ static void io_iopoll_req_issued(struct io_kiocb *req, unsigned int issue_flags)
 	}
 }
 
-unsigned int io_file_get_flags(struct file *file)
+io_req_flags_t io_file_get_flags(struct file *file)
 {
-	unsigned int res = 0;
+	io_req_flags_t res = 0;
 
 	if (S_ISREG(file_inode(file)->i_mode))
 		res |= REQ_F_ISREG;
@@ -2171,7 +2171,8 @@ static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
 	/* req is partially pre-initialised, see io_preinit_req() */
 	req->opcode = opcode = READ_ONCE(sqe->opcode);
 	/* same numerical values with corresponding REQ_F_*, safe to copy */
-	req->flags = sqe_flags = READ_ONCE(sqe->flags);
+	sqe_flags = READ_ONCE(sqe->flags);
+	req->flags = (io_req_flags_t) sqe_flags;
 	req->cqe.user_data = READ_ONCE(sqe->user_data);
 	req->file = NULL;
 	req->rsrc_node = NULL;
@@ -4153,7 +4154,7 @@ static int __init io_uring_init(void)
 	BUILD_BUG_ON(SQE_COMMON_FLAGS >= (1 << 8));
 	BUILD_BUG_ON((SQE_VALID_FLAGS | SQE_COMMON_FLAGS) != SQE_VALID_FLAGS);
 
-	BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(int));
+	BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(u64));
 
 	BUILD_BUG_ON(sizeof(atomic_t) != sizeof(u32));
 
-- 
Jens Axboe

Re: [PATCH 1/6] io_uring: expand main struct io_kiocb flags to 64-bits

[Pavel Begunkov]

On 2/6/24 16:22, Jens Axboe wrote:
> We're out of space here, and none of the flags are easily reclaimable.
> Bump it to 64-bits and re-arrange the struct a bit to avoid gaps.
> 
> Add a specific bitwise type for the request flags, io_request_flags_t.
> This will help catch violations of casting this value to a smaller type
> on 32-bit archs, like unsigned int.
> 
> No functional changes intended in this patch.
> 
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>   include/linux/io_uring_types.h  | 87 ++++++++++++++++++---------------
>   include/trace/events/io_uring.h | 14 +++---
>   io_uring/filetable.h            |  2 +-
>   io_uring/io_uring.c             |  9 ++--
>   4 files changed, 60 insertions(+), 52 deletions(-)
> 
> diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
> index 854ad67a5f70..5ac18b05d4ee 100644
> --- a/include/linux/io_uring_types.h
> +++ b/include/linux/io_uring_types.h
> @@ -428,7 +428,7 @@ struct io_tw_state {
>   	bool locked;
>   };
>   
> -enum {
> +enum io_req_flags {
>   	REQ_F_FIXED_FILE_BIT	= IOSQE_FIXED_FILE_BIT,
>   	REQ_F_IO_DRAIN_BIT	= IOSQE_IO_DRAIN_BIT,
>   	REQ_F_LINK_BIT		= IOSQE_IO_LINK_BIT,
> @@ -468,70 +468,73 @@ enum {
>   	__REQ_F_LAST_BIT,
>   };
>   
> +typedef enum io_req_flags __bitwise io_req_flags_t;
> +#define IO_REQ_FLAG(bitno)	((__force io_req_flags_t) BIT_ULL((bitno)))
> +
>   enum {
>   	/* ctx owns file */
> -	REQ_F_FIXED_FILE	= BIT(REQ_F_FIXED_FILE_BIT),
> +	REQ_F_FIXED_FILE	= IO_REQ_FLAG(REQ_F_FIXED_FILE_BIT),
>   	/* drain existing IO first */
> -	REQ_F_IO_DRAIN		= BIT(REQ_F_IO_DRAIN_BIT),
> +	REQ_F_IO_DRAIN		= IO_REQ_FLAG(REQ_F_IO_DRAIN_BIT),
>   	/* linked sqes */
> -	REQ_F_LINK		= BIT(REQ_F_LINK_BIT),
> +	REQ_F_LINK		= IO_REQ_FLAG(REQ_F_LINK_BIT),
>   	/* doesn't sever on completion < 0 */
> -	REQ_F_HARDLINK		= BIT(REQ_F_HARDLINK_BIT),
> +	REQ_F_HARDLINK		= IO_REQ_FLAG(REQ_F_HARDLINK_BIT),
>   	/* IOSQE_ASYNC */
> -	REQ_F_FORCE_ASYNC	= BIT(REQ_F_FORCE_ASYNC_BIT),
> +	REQ_F_FORCE_ASYNC	= IO_REQ_FLAG(REQ_F_FORCE_ASYNC_BIT),
>   	/* IOSQE_BUFFER_SELECT */
> -	REQ_F_BUFFER_SELECT	= BIT(REQ_F_BUFFER_SELECT_BIT),
> +	REQ_F_BUFFER_SELECT	= IO_REQ_FLAG(REQ_F_BUFFER_SELECT_BIT),
>   	/* IOSQE_CQE_SKIP_SUCCESS */
> -	REQ_F_CQE_SKIP		= BIT(REQ_F_CQE_SKIP_BIT),
> +	REQ_F_CQE_SKIP		= IO_REQ_FLAG(REQ_F_CQE_SKIP_BIT),
>   
>   	/* fail rest of links */
> -	REQ_F_FAIL		= BIT(REQ_F_FAIL_BIT),
> +	REQ_F_FAIL		= IO_REQ_FLAG(REQ_F_FAIL_BIT),
>   	/* on inflight list, should be cancelled and waited on exit reliably */
> -	REQ_F_INFLIGHT		= BIT(REQ_F_INFLIGHT_BIT),
> +	REQ_F_INFLIGHT		= IO_REQ_FLAG(REQ_F_INFLIGHT_BIT),
>   	/* read/write uses file position */
> -	REQ_F_CUR_POS		= BIT(REQ_F_CUR_POS_BIT),
> +	REQ_F_CUR_POS		= IO_REQ_FLAG(REQ_F_CUR_POS_BIT),
>   	/* must not punt to workers */
> -	REQ_F_NOWAIT		= BIT(REQ_F_NOWAIT_BIT),
> +	REQ_F_NOWAIT		= IO_REQ_FLAG(REQ_F_NOWAIT_BIT),
>   	/* has or had linked timeout */
> -	REQ_F_LINK_TIMEOUT	= BIT(REQ_F_LINK_TIMEOUT_BIT),
> +	REQ_F_LINK_TIMEOUT	= IO_REQ_FLAG(REQ_F_LINK_TIMEOUT_BIT),
>   	/* needs cleanup */
> -	REQ_F_NEED_CLEANUP	= BIT(REQ_F_NEED_CLEANUP_BIT),
> +	REQ_F_NEED_CLEANUP	= IO_REQ_FLAG(REQ_F_NEED_CLEANUP_BIT),
>   	/* already went through poll handler */
> -	REQ_F_POLLED		= BIT(REQ_F_POLLED_BIT),
> +	REQ_F_POLLED		= IO_REQ_FLAG(REQ_F_POLLED_BIT),
>   	/* buffer already selected */
> -	REQ_F_BUFFER_SELECTED	= BIT(REQ_F_BUFFER_SELECTED_BIT),
> +	REQ_F_BUFFER_SELECTED	= IO_REQ_FLAG(REQ_F_BUFFER_SELECTED_BIT),
>   	/* buffer selected from ring, needs commit */
> -	REQ_F_BUFFER_RING	= BIT(REQ_F_BUFFER_RING_BIT),
> +	REQ_F_BUFFER_RING	= IO_REQ_FLAG(REQ_F_BUFFER_RING_BIT),
>   	/* caller should reissue async */
> -	REQ_F_REISSUE		= BIT(REQ_F_REISSUE_BIT),
> +	REQ_F_REISSUE		= IO_REQ_FLAG(REQ_F_REISSUE_BIT),
>   	/* supports async reads/writes */
> -	REQ_F_SUPPORT_NOWAIT	= BIT(REQ_F_SUPPORT_NOWAIT_BIT),
> +	REQ_F_SUPPORT_NOWAIT	= IO_REQ_FLAG(REQ_F_SUPPORT_NOWAIT_BIT),
>   	/* regular file */
> -	REQ_F_ISREG		= BIT(REQ_F_ISREG_BIT),
> +	REQ_F_ISREG		= IO_REQ_FLAG(REQ_F_ISREG_BIT),
>   	/* has creds assigned */
> -	REQ_F_CREDS		= BIT(REQ_F_CREDS_BIT),
> +	REQ_F_CREDS		= IO_REQ_FLAG(REQ_F_CREDS_BIT),
>   	/* skip refcounting if not set */
> -	REQ_F_REFCOUNT		= BIT(REQ_F_REFCOUNT_BIT),
> +	REQ_F_REFCOUNT		= IO_REQ_FLAG(REQ_F_REFCOUNT_BIT),
>   	/* there is a linked timeout that has to be armed */
> -	REQ_F_ARM_LTIMEOUT	= BIT(REQ_F_ARM_LTIMEOUT_BIT),
> +	REQ_F_ARM_LTIMEOUT	= IO_REQ_FLAG(REQ_F_ARM_LTIMEOUT_BIT),
>   	/* ->async_data allocated */
> -	REQ_F_ASYNC_DATA	= BIT(REQ_F_ASYNC_DATA_BIT),
> +	REQ_F_ASYNC_DATA	= IO_REQ_FLAG(REQ_F_ASYNC_DATA_BIT),
>   	/* don't post CQEs while failing linked requests */
> -	REQ_F_SKIP_LINK_CQES	= BIT(REQ_F_SKIP_LINK_CQES_BIT),
> +	REQ_F_SKIP_LINK_CQES	= IO_REQ_FLAG(REQ_F_SKIP_LINK_CQES_BIT),
>   	/* single poll may be active */
> -	REQ_F_SINGLE_POLL	= BIT(REQ_F_SINGLE_POLL_BIT),
> +	REQ_F_SINGLE_POLL	= IO_REQ_FLAG(REQ_F_SINGLE_POLL_BIT),
>   	/* double poll may active */
> -	REQ_F_DOUBLE_POLL	= BIT(REQ_F_DOUBLE_POLL_BIT),
> +	REQ_F_DOUBLE_POLL	= IO_REQ_FLAG(REQ_F_DOUBLE_POLL_BIT),
>   	/* request has already done partial IO */
> -	REQ_F_PARTIAL_IO	= BIT(REQ_F_PARTIAL_IO_BIT),
> +	REQ_F_PARTIAL_IO	= IO_REQ_FLAG(REQ_F_PARTIAL_IO_BIT),
>   	/* fast poll multishot mode */
> -	REQ_F_APOLL_MULTISHOT	= BIT(REQ_F_APOLL_MULTISHOT_BIT),
> +	REQ_F_APOLL_MULTISHOT	= IO_REQ_FLAG(REQ_F_APOLL_MULTISHOT_BIT),
>   	/* recvmsg special flag, clear EPOLLIN */
> -	REQ_F_CLEAR_POLLIN	= BIT(REQ_F_CLEAR_POLLIN_BIT),
> +	REQ_F_CLEAR_POLLIN	= IO_REQ_FLAG(REQ_F_CLEAR_POLLIN_BIT),
>   	/* hashed into ->cancel_hash_locked, protected by ->uring_lock */
> -	REQ_F_HASH_LOCKED	= BIT(REQ_F_HASH_LOCKED_BIT),
> +	REQ_F_HASH_LOCKED	= IO_REQ_FLAG(REQ_F_HASH_LOCKED_BIT),
>   	/* don't use lazy poll wake for this request */
> -	REQ_F_POLL_NO_LAZY	= BIT(REQ_F_POLL_NO_LAZY_BIT),
> +	REQ_F_POLL_NO_LAZY	= IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT),
>   };
>   
>   typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
> @@ -592,15 +595,14 @@ struct io_kiocb {
>   	 * and after selection it points to the buffer ID itself.
>   	 */
>   	u16				buf_index;
> -	unsigned int			flags;
>   
> -	struct io_cqe			cqe;

With the current layout the min number of lines we touch per
request is 2 (including the op specific 64B), that's includes
setting up cqe at init and using it for completing. Moving cqe
down makes it 3.

> +	atomic_t			refs;

We're pulling it refs, which is not touched at all in the hot
path. Even if there's a hole I'd argue it's better to leave it
at the end.

> +
> +	io_req_flags_t			flags;
>   
>   	struct io_ring_ctx		*ctx;
>   	struct task_struct		*task;
>   
> -	struct io_rsrc_node		*rsrc_node;

It's used in hot paths, registered buffers/files, would be
unfortunate to move it to the next line.

> -
>   	union {
>   		/* store used ubuf, so we can prevent reloading */
>   		struct io_mapped_ubuf	*imu;
> @@ -615,18 +617,23 @@ struct io_kiocb {
>   		struct io_buffer_list	*buf_list;
>   	};
>   
> +	/* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
> +	struct hlist_node		hash_node;
> +

And we're pulling hash_node into the hottest line, which is
used only when we arm a poll and remove poll. So, it's mostly
for networking, sends wouldn't use it much, and multishots
wouldn't normally touch it.

As for ideas how to find space:
1) iopoll_completed completed can be converted to flags2

2) REQ_F_{SINGLE,DOUBLE}_POLL is a weird duplication. Can
probably be combined into one flag, or removed at all.
Again, sends are usually not so poll heavy and the hot
path for recv is multishot.

3) we can probably move req->task down and replace it with

get_task() {
     if (req->ctx->flags & DEFER_TASKRUN)
         task = ctx->submitter_task;
     else
         task = req->task;
}

The most common user of it -- task_work_add -- already
checks the flag and has separate paths, and
init/completion paths can be optimised.


>   	union {
>   		/* used by request caches, completion batching and iopoll */
>   		struct io_wq_work_node	comp_list;
>   		/* cache ->apoll->events */
>   		__poll_t apoll_events;
>   	};
> -	atomic_t			refs;
> -	atomic_t			poll_refs;
> +
> +	struct io_rsrc_node		*rsrc_node;
> +
> +	struct io_cqe			cqe;
> +
>   	struct io_task_work		io_task_work;
> +	atomic_t			poll_refs;
>   	unsigned			nr_tw;
> -	/* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
> -	struct hlist_node		hash_node;
>   	/* internal polling, see IORING_FEAT_FAST_POLL */
>   	struct async_poll		*apoll;
>   	/* opcode allocated if it needs to store data for async defer */
> diff --git a/include/trace/events/io_uring.h b/include/trace/events/io_uring.h
> index 69454f1f98b0..3d7704a52b73 100644
> --- a/include/trace/events/io_uring.h
> +++ b/include/trace/events/io_uring.h
> @@ -148,7 +148,7 @@ TRACE_EVENT(io_uring_queue_async_work,
>   		__field(  void *,			req		)
>   		__field(  u64,				user_data	)
>   		__field(  u8,				opcode		)
> -		__field(  unsigned int,			flags		)
> +		__field(  io_req_flags_t,		flags		)
>   		__field(  struct io_wq_work *,		work		)
>   		__field(  int,				rw		)
>   
> @@ -167,10 +167,10 @@ TRACE_EVENT(io_uring_queue_async_work,
>   		__assign_str(op_str, io_uring_get_opcode(req->opcode));
>   	),
>   
> -	TP_printk("ring %p, request %p, user_data 0x%llx, opcode %s, flags 0x%x, %s queue, work %p",
> +	TP_printk("ring %p, request %p, user_data 0x%llx, opcode %s, flags 0x%lx, %s queue, work %p",
>   		__entry->ctx, __entry->req, __entry->user_data,
> -		__get_str(op_str),
> -		__entry->flags, __entry->rw ? "hashed" : "normal", __entry->work)
> +		__get_str(op_str), (long) __entry->flags,
> +		__entry->rw ? "hashed" : "normal", __entry->work)
>   );
>   
>   /**
> @@ -378,7 +378,7 @@ TRACE_EVENT(io_uring_submit_req,
>   		__field(  void *,		req		)
>   		__field(  unsigned long long,	user_data	)
>   		__field(  u8,			opcode		)
> -		__field(  u32,			flags		)
> +		__field(  io_req_flags_t,	flags		)
>   		__field(  bool,			sq_thread	)
>   
>   		__string( op_str, io_uring_get_opcode(req->opcode) )
> @@ -395,10 +395,10 @@ TRACE_EVENT(io_uring_submit_req,
>   		__assign_str(op_str, io_uring_get_opcode(req->opcode));
>   	),
>   
> -	TP_printk("ring %p, req %p, user_data 0x%llx, opcode %s, flags 0x%x, "
> +	TP_printk("ring %p, req %p, user_data 0x%llx, opcode %s, flags 0x%lx, "
>   		  "sq_thread %d", __entry->ctx, __entry->req,
>   		  __entry->user_data, __get_str(op_str),
> -		  __entry->flags, __entry->sq_thread)
> +		  (long) __entry->flags, __entry->sq_thread)
>   );
>   
>   /*
> diff --git a/io_uring/filetable.h b/io_uring/filetable.h
> index b47adf170c31..b2435c4dca1f 100644
> --- a/io_uring/filetable.h
> +++ b/io_uring/filetable.h
> @@ -17,7 +17,7 @@ int io_fixed_fd_remove(struct io_ring_ctx *ctx, unsigned int offset);
>   int io_register_file_alloc_range(struct io_ring_ctx *ctx,
>   				 struct io_uring_file_index_range __user *arg);
>   
> -unsigned int io_file_get_flags(struct file *file);
> +io_req_flags_t io_file_get_flags(struct file *file);
>   
>   static inline void io_file_bitmap_clear(struct io_file_table *table, int bit)
>   {
> diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
> index cd9a137ad6ce..360a7ee41d3a 100644
> --- a/io_uring/io_uring.c
> +++ b/io_uring/io_uring.c
> @@ -1768,9 +1768,9 @@ static void io_iopoll_req_issued(struct io_kiocb *req, unsigned int issue_flags)
>   	}
>   }
>   
> -unsigned int io_file_get_flags(struct file *file)
> +io_req_flags_t io_file_get_flags(struct file *file)
>   {
> -	unsigned int res = 0;
> +	io_req_flags_t res = 0;
>   
>   	if (S_ISREG(file_inode(file)->i_mode))
>   		res |= REQ_F_ISREG;
> @@ -2171,7 +2171,8 @@ static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
>   	/* req is partially pre-initialised, see io_preinit_req() */
>   	req->opcode = opcode = READ_ONCE(sqe->opcode);
>   	/* same numerical values with corresponding REQ_F_*, safe to copy */
> -	req->flags = sqe_flags = READ_ONCE(sqe->flags);
> +	sqe_flags = READ_ONCE(sqe->flags);
> +	req->flags = (io_req_flags_t) sqe_flags;
>   	req->cqe.user_data = READ_ONCE(sqe->user_data);
>   	req->file = NULL;
>   	req->rsrc_node = NULL;
> @@ -4153,7 +4154,7 @@ static int __init io_uring_init(void)
>   	BUILD_BUG_ON(SQE_COMMON_FLAGS >= (1 << 8));
>   	BUILD_BUG_ON((SQE_VALID_FLAGS | SQE_COMMON_FLAGS) != SQE_VALID_FLAGS);
>   
> -	BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(int));
> +	BUILD_BUG_ON(__REQ_F_LAST_BIT > 8 * sizeof(u64));
>   
>   	BUILD_BUG_ON(sizeof(atomic_t) != sizeof(u32));
>   

-- 
Pavel Begunkov

Re: [PATCH 1/6] io_uring: expand main struct io_kiocb flags to 64-bits

[Jens Axboe]

On 2/6/24 5:43 PM, Pavel Begunkov wrote:
> On 2/6/24 16:22, Jens Axboe wrote:
>> We're out of space here, and none of the flags are easily reclaimable.
>> Bump it to 64-bits and re-arrange the struct a bit to avoid gaps.
>>
>> Add a specific bitwise type for the request flags, io_request_flags_t.
>> This will help catch violations of casting this value to a smaller type
>> on 32-bit archs, like unsigned int.
>>
>> No functional changes intended in this patch.
>>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>> ---
>>   include/linux/io_uring_types.h  | 87 ++++++++++++++++++---------------
>>   include/trace/events/io_uring.h | 14 +++---
>>   io_uring/filetable.h            |  2 +-
>>   io_uring/io_uring.c             |  9 ++--
>>   4 files changed, 60 insertions(+), 52 deletions(-)
>>
>> diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
>> index 854ad67a5f70..5ac18b05d4ee 100644
>> --- a/include/linux/io_uring_types.h
>> +++ b/include/linux/io_uring_types.h
>> @@ -428,7 +428,7 @@ struct io_tw_state {
>>       bool locked;
>>   };
>>   -enum {
>> +enum io_req_flags {
>>       REQ_F_FIXED_FILE_BIT    = IOSQE_FIXED_FILE_BIT,
>>       REQ_F_IO_DRAIN_BIT    = IOSQE_IO_DRAIN_BIT,
>>       REQ_F_LINK_BIT        = IOSQE_IO_LINK_BIT,
>> @@ -468,70 +468,73 @@ enum {
>>       __REQ_F_LAST_BIT,
>>   };
>>   +typedef enum io_req_flags __bitwise io_req_flags_t;
>> +#define IO_REQ_FLAG(bitno)    ((__force io_req_flags_t) BIT_ULL((bitno)))
>> +
>>   enum {
>>       /* ctx owns file */
>> -    REQ_F_FIXED_FILE    = BIT(REQ_F_FIXED_FILE_BIT),
>> +    REQ_F_FIXED_FILE    = IO_REQ_FLAG(REQ_F_FIXED_FILE_BIT),
>>       /* drain existing IO first */
>> -    REQ_F_IO_DRAIN        = BIT(REQ_F_IO_DRAIN_BIT),
>> +    REQ_F_IO_DRAIN        = IO_REQ_FLAG(REQ_F_IO_DRAIN_BIT),
>>       /* linked sqes */
>> -    REQ_F_LINK        = BIT(REQ_F_LINK_BIT),
>> +    REQ_F_LINK        = IO_REQ_FLAG(REQ_F_LINK_BIT),
>>       /* doesn't sever on completion < 0 */
>> -    REQ_F_HARDLINK        = BIT(REQ_F_HARDLINK_BIT),
>> +    REQ_F_HARDLINK        = IO_REQ_FLAG(REQ_F_HARDLINK_BIT),
>>       /* IOSQE_ASYNC */
>> -    REQ_F_FORCE_ASYNC    = BIT(REQ_F_FORCE_ASYNC_BIT),
>> +    REQ_F_FORCE_ASYNC    = IO_REQ_FLAG(REQ_F_FORCE_ASYNC_BIT),
>>       /* IOSQE_BUFFER_SELECT */
>> -    REQ_F_BUFFER_SELECT    = BIT(REQ_F_BUFFER_SELECT_BIT),
>> +    REQ_F_BUFFER_SELECT    = IO_REQ_FLAG(REQ_F_BUFFER_SELECT_BIT),
>>       /* IOSQE_CQE_SKIP_SUCCESS */
>> -    REQ_F_CQE_SKIP        = BIT(REQ_F_CQE_SKIP_BIT),
>> +    REQ_F_CQE_SKIP        = IO_REQ_FLAG(REQ_F_CQE_SKIP_BIT),
>>         /* fail rest of links */
>> -    REQ_F_FAIL        = BIT(REQ_F_FAIL_BIT),
>> +    REQ_F_FAIL        = IO_REQ_FLAG(REQ_F_FAIL_BIT),
>>       /* on inflight list, should be cancelled and waited on exit reliably */
>> -    REQ_F_INFLIGHT        = BIT(REQ_F_INFLIGHT_BIT),
>> +    REQ_F_INFLIGHT        = IO_REQ_FLAG(REQ_F_INFLIGHT_BIT),
>>       /* read/write uses file position */
>> -    REQ_F_CUR_POS        = BIT(REQ_F_CUR_POS_BIT),
>> +    REQ_F_CUR_POS        = IO_REQ_FLAG(REQ_F_CUR_POS_BIT),
>>       /* must not punt to workers */
>> -    REQ_F_NOWAIT        = BIT(REQ_F_NOWAIT_BIT),
>> +    REQ_F_NOWAIT        = IO_REQ_FLAG(REQ_F_NOWAIT_BIT),
>>       /* has or had linked timeout */
>> -    REQ_F_LINK_TIMEOUT    = BIT(REQ_F_LINK_TIMEOUT_BIT),
>> +    REQ_F_LINK_TIMEOUT    = IO_REQ_FLAG(REQ_F_LINK_TIMEOUT_BIT),
>>       /* needs cleanup */
>> -    REQ_F_NEED_CLEANUP    = BIT(REQ_F_NEED_CLEANUP_BIT),
>> +    REQ_F_NEED_CLEANUP    = IO_REQ_FLAG(REQ_F_NEED_CLEANUP_BIT),
>>       /* already went through poll handler */
>> -    REQ_F_POLLED        = BIT(REQ_F_POLLED_BIT),
>> +    REQ_F_POLLED        = IO_REQ_FLAG(REQ_F_POLLED_BIT),
>>       /* buffer already selected */
>> -    REQ_F_BUFFER_SELECTED    = BIT(REQ_F_BUFFER_SELECTED_BIT),
>> +    REQ_F_BUFFER_SELECTED    = IO_REQ_FLAG(REQ_F_BUFFER_SELECTED_BIT),
>>       /* buffer selected from ring, needs commit */
>> -    REQ_F_BUFFER_RING    = BIT(REQ_F_BUFFER_RING_BIT),
>> +    REQ_F_BUFFER_RING    = IO_REQ_FLAG(REQ_F_BUFFER_RING_BIT),
>>       /* caller should reissue async */
>> -    REQ_F_REISSUE        = BIT(REQ_F_REISSUE_BIT),
>> +    REQ_F_REISSUE        = IO_REQ_FLAG(REQ_F_REISSUE_BIT),
>>       /* supports async reads/writes */
>> -    REQ_F_SUPPORT_NOWAIT    = BIT(REQ_F_SUPPORT_NOWAIT_BIT),
>> +    REQ_F_SUPPORT_NOWAIT    = IO_REQ_FLAG(REQ_F_SUPPORT_NOWAIT_BIT),
>>       /* regular file */
>> -    REQ_F_ISREG        = BIT(REQ_F_ISREG_BIT),
>> +    REQ_F_ISREG        = IO_REQ_FLAG(REQ_F_ISREG_BIT),
>>       /* has creds assigned */
>> -    REQ_F_CREDS        = BIT(REQ_F_CREDS_BIT),
>> +    REQ_F_CREDS        = IO_REQ_FLAG(REQ_F_CREDS_BIT),
>>       /* skip refcounting if not set */
>> -    REQ_F_REFCOUNT        = BIT(REQ_F_REFCOUNT_BIT),
>> +    REQ_F_REFCOUNT        = IO_REQ_FLAG(REQ_F_REFCOUNT_BIT),
>>       /* there is a linked timeout that has to be armed */
>> -    REQ_F_ARM_LTIMEOUT    = BIT(REQ_F_ARM_LTIMEOUT_BIT),
>> +    REQ_F_ARM_LTIMEOUT    = IO_REQ_FLAG(REQ_F_ARM_LTIMEOUT_BIT),
>>       /* ->async_data allocated */
>> -    REQ_F_ASYNC_DATA    = BIT(REQ_F_ASYNC_DATA_BIT),
>> +    REQ_F_ASYNC_DATA    = IO_REQ_FLAG(REQ_F_ASYNC_DATA_BIT),
>>       /* don't post CQEs while failing linked requests */
>> -    REQ_F_SKIP_LINK_CQES    = BIT(REQ_F_SKIP_LINK_CQES_BIT),
>> +    REQ_F_SKIP_LINK_CQES    = IO_REQ_FLAG(REQ_F_SKIP_LINK_CQES_BIT),
>>       /* single poll may be active */
>> -    REQ_F_SINGLE_POLL    = BIT(REQ_F_SINGLE_POLL_BIT),
>> +    REQ_F_SINGLE_POLL    = IO_REQ_FLAG(REQ_F_SINGLE_POLL_BIT),
>>       /* double poll may active */
>> -    REQ_F_DOUBLE_POLL    = BIT(REQ_F_DOUBLE_POLL_BIT),
>> +    REQ_F_DOUBLE_POLL    = IO_REQ_FLAG(REQ_F_DOUBLE_POLL_BIT),
>>       /* request has already done partial IO */
>> -    REQ_F_PARTIAL_IO    = BIT(REQ_F_PARTIAL_IO_BIT),
>> +    REQ_F_PARTIAL_IO    = IO_REQ_FLAG(REQ_F_PARTIAL_IO_BIT),
>>       /* fast poll multishot mode */
>> -    REQ_F_APOLL_MULTISHOT    = BIT(REQ_F_APOLL_MULTISHOT_BIT),
>> +    REQ_F_APOLL_MULTISHOT    = IO_REQ_FLAG(REQ_F_APOLL_MULTISHOT_BIT),
>>       /* recvmsg special flag, clear EPOLLIN */
>> -    REQ_F_CLEAR_POLLIN    = BIT(REQ_F_CLEAR_POLLIN_BIT),
>> +    REQ_F_CLEAR_POLLIN    = IO_REQ_FLAG(REQ_F_CLEAR_POLLIN_BIT),
>>       /* hashed into ->cancel_hash_locked, protected by ->uring_lock */
>> -    REQ_F_HASH_LOCKED    = BIT(REQ_F_HASH_LOCKED_BIT),
>> +    REQ_F_HASH_LOCKED    = IO_REQ_FLAG(REQ_F_HASH_LOCKED_BIT),
>>       /* don't use lazy poll wake for this request */
>> -    REQ_F_POLL_NO_LAZY    = BIT(REQ_F_POLL_NO_LAZY_BIT),
>> +    REQ_F_POLL_NO_LAZY    = IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT),
>>   };
>>     typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
>> @@ -592,15 +595,14 @@ struct io_kiocb {
>>        * and after selection it points to the buffer ID itself.
>>        */
>>       u16                buf_index;
>> -    unsigned int            flags;
>>   -    struct io_cqe            cqe;
> 
> With the current layout the min number of lines we touch per
> request is 2 (including the op specific 64B), that's includes
> setting up cqe at init and using it for completing. Moving cqe
> down makes it 3.
> 
>> +    atomic_t            refs;
> 
> We're pulling it refs, which is not touched at all in the hot
> path. Even if there's a hole I'd argue it's better to leave it
> at the end.
> 
>> +
>> +    io_req_flags_t            flags;
>>         struct io_ring_ctx        *ctx;
>>       struct task_struct        *task;
>>   -    struct io_rsrc_node        *rsrc_node;
> 
> It's used in hot paths, registered buffers/files, would be
> unfortunate to move it to the next line.

Yep I did feel a bit bad about that one... Let me take another stab at
it.

>> -
>>       union {
>>           /* store used ubuf, so we can prevent reloading */
>>           struct io_mapped_ubuf    *imu;
>> @@ -615,18 +617,23 @@ struct io_kiocb {
>>           struct io_buffer_list    *buf_list;
>>       };
>>   +    /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
>> +    struct hlist_node        hash_node;
>> +
> 
> And we're pulling hash_node into the hottest line, which is
> used only when we arm a poll and remove poll. So, it's mostly
> for networking, sends wouldn't use it much, and multishots
> wouldn't normally touch it.
> 
> As for ideas how to find space:
> 1) iopoll_completed completed can be converted to flags2

That's a good idea, but won't immediately find any space as it'd just
leave a hole anyway. But would be good to note in there perhaps, you
never know when it needs re-arranging again.

> 2) REQ_F_{SINGLE,DOUBLE}_POLL is a weird duplication. Can
> probably be combined into one flag, or removed at all.
> Again, sends are usually not so poll heavy and the hot
> path for recv is multishot.

Normal receive is also a hot path, even if multishot should be preferred
in general. Ditto on non-sockets but still pollable files, doing eg read
for example.

> 3) we can probably move req->task down and replace it with
> 
> get_task() {
>     if (req->ctx->flags & DEFER_TASKRUN)
>         task = ctx->submitter_task;
>     else
>         task = req->task;
> }

Assuming ctx flags is hot, which is would generally be, that's not a bad
idea at all.

I'll do another loop over this one.

-- 
Jens Axboe

Re: [PATCH 1/6] io_uring: expand main struct io_kiocb flags to 64-bits

[Pavel Begunkov]

On 2/7/24 02:18, Jens Axboe wrote:
> On 2/6/24 5:43 PM, Pavel Begunkov wrote:
>> On 2/6/24 16:22, Jens Axboe wrote:
>>> We're out of space here, and none of the flags are easily reclaimable.
>>> Bump it to 64-bits and re-arrange the struct a bit to avoid gaps.
>>>
>>> Add a specific bitwise type for the request flags, io_request_flags_t.
>>> This will help catch violations of casting this value to a smaller type
>>> on 32-bit archs, like unsigned int.
>>>
>>> No functional changes intended in this patch.
>>>
>>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>> ---
...
>>>      typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
>>> @@ -592,15 +595,14 @@ struct io_kiocb {
>>>         * and after selection it points to the buffer ID itself.
>>>         */
>>>        u16                buf_index;
>>> -    unsigned int            flags;
>>>    -    struct io_cqe            cqe;
>>
>> With the current layout the min number of lines we touch per
>> request is 2 (including the op specific 64B), that's includes
>> setting up cqe at init and using it for completing. Moving cqe
>> down makes it 3.
>>
>>> +    atomic_t            refs;
>>
>> We're pulling it refs, which is not touched at all in the hot
>> path. Even if there's a hole I'd argue it's better to leave it
>> at the end.
>>
>>> +
>>> +    io_req_flags_t            flags;
>>>          struct io_ring_ctx        *ctx;
>>>        struct task_struct        *task;
>>>    -    struct io_rsrc_node        *rsrc_node;
>>
>> It's used in hot paths, registered buffers/files, would be
>> unfortunate to move it to the next line.
> 
> Yep I did feel a bit bad about that one... Let me take another stab at
> it.
> 
>>> -
>>>        union {
>>>            /* store used ubuf, so we can prevent reloading */
>>>            struct io_mapped_ubuf    *imu;
>>> @@ -615,18 +617,23 @@ struct io_kiocb {
>>>            struct io_buffer_list    *buf_list;
>>>        };
>>>    +    /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
>>> +    struct hlist_node        hash_node;
>>> +
>>
>> And we're pulling hash_node into the hottest line, which is
>> used only when we arm a poll and remove poll. So, it's mostly
>> for networking, sends wouldn't use it much, and multishots
>> wouldn't normally touch it.
>>
>> As for ideas how to find space:
>> 1) iopoll_completed completed can be converted to flags2
> 
> That's a good idea, but won't immediately find any space as it'd just
> leave a hole anyway. But would be good to note in there perhaps, you
> never know when it needs re-arranging again.

struct io_kiocb {
	unsigned flags;
	...
	u8 	 flags2;
};

I rather proposed to have this, which is definitely borderline
ugly but certainly an option.


>> 2) REQ_F_{SINGLE,DOUBLE}_POLL is a weird duplication. Can
>> probably be combined into one flag, or removed at all.
>> Again, sends are usually not so poll heavy and the hot
>> path for recv is multishot.
> 
> Normal receive is also a hot path, even if multishot should be preferred

The degree of hotness is arguable. It's poll, which takes a
spinlock (and disables irqs), does an indirect call, goes into
he socket internals there touching pretty contended parts
like sock_wq. The relative overhead of looking at f_ops should
be nothing.

But the thought was more about combining them,
REQ_F_POLL_ACTIVE, and clear only if it's not double poll.

> in general. Ditto on non-sockets but still pollable files, doing eg read
> for example.
> 
>> 3) we can probably move req->task down and replace it with
>>
>> get_task() {
>>      if (req->ctx->flags & DEFER_TASKRUN)
>>          task = ctx->submitter_task;
>>      else
>>          task = req->task;
>> }
> 
> Assuming ctx flags is hot, which is would generally be, that's not a bad
> idea at all.

As mentioned, task_work_add would be the main user, and
there is already a different branch for DEFER_TASKRUN,
to it implicitly knows that ctx->submitter_task is correct.

> I'll do another loop over this one.

-- 
Pavel Begunkov

[PATCH 2/6] io_uring: add io_file_can_poll() helper

[Jens Axboe]

This adds a flag to avoid dipping dereferencing file and then f_op
to figure out if the file has a poll handler defined or not. We
generally call this at least twice for networked workloads.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/linux/io_uring_types.h |  3 +++
 io_uring/io_uring.c            |  2 +-
 io_uring/io_uring.h            | 12 ++++++++++++
 io_uring/kbuf.c                |  2 +-
 io_uring/poll.c                |  2 +-
 io_uring/rw.c                  |  6 +++---
 6 files changed, 21 insertions(+), 6 deletions(-)

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 5ac18b05d4ee..7f06cee02b58 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -463,6 +463,7 @@ enum io_req_flags {
 	REQ_F_SUPPORT_NOWAIT_BIT,
 	REQ_F_ISREG_BIT,
 	REQ_F_POLL_NO_LAZY_BIT,
+	REQ_F_CAN_POLL_BIT,
 
 	/* not a real bit, just to check we're not overflowing the space */
 	__REQ_F_LAST_BIT,
@@ -535,6 +536,8 @@ enum {
 	REQ_F_HASH_LOCKED	= IO_REQ_FLAG(REQ_F_HASH_LOCKED_BIT),
 	/* don't use lazy poll wake for this request */
 	REQ_F_POLL_NO_LAZY	= IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT),
+	/* file is pollable */
+	REQ_F_CAN_POLL		= IO_REQ_FLAG(REQ_F_CAN_POLL_BIT),
 };
 
 typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 360a7ee41d3a..d0e06784926f 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -1969,7 +1969,7 @@ void io_wq_submit_work(struct io_wq_work *work)
 	if (req->flags & REQ_F_FORCE_ASYNC) {
 		bool opcode_poll = def->pollin || def->pollout;
 
-		if (opcode_poll && file_can_poll(req->file)) {
+		if (opcode_poll && io_file_can_poll(req)) {
 			needs_poll = true;
 			issue_flags |= IO_URING_F_NONBLOCK;
 		}
diff --git a/io_uring/io_uring.h b/io_uring/io_uring.h
index d5495710c178..2952551fe345 100644
--- a/io_uring/io_uring.h
+++ b/io_uring/io_uring.h
@@ -5,6 +5,7 @@
 #include <linux/lockdep.h>
 #include <linux/resume_user_mode.h>
 #include <linux/kasan.h>
+#include <linux/poll.h>
 #include <linux/io_uring_types.h>
 #include <uapi/linux/eventpoll.h>
 #include "io-wq.h"
@@ -398,4 +399,15 @@ static inline size_t uring_sqe_size(struct io_ring_ctx *ctx)
 		return 2 * sizeof(struct io_uring_sqe);
 	return sizeof(struct io_uring_sqe);
 }
+
+static inline bool io_file_can_poll(struct io_kiocb *req)
+{
+	if (req->flags & REQ_F_CAN_POLL)
+		return true;
+	if (file_can_poll(req->file)) {
+		req->flags |= REQ_F_CAN_POLL;
+		return true;
+	}
+	return false;
+}
 #endif
diff --git a/io_uring/kbuf.c b/io_uring/kbuf.c
index 18df5a9d2f5e..71880615bb78 100644
--- a/io_uring/kbuf.c
+++ b/io_uring/kbuf.c
@@ -180,7 +180,7 @@ static void __user *io_ring_buffer_select(struct io_kiocb *req, size_t *len,
 	req->buf_list = bl;
 	req->buf_index = buf->bid;
 
-	if (issue_flags & IO_URING_F_UNLOCKED || !file_can_poll(req->file)) {
+	if (issue_flags & IO_URING_F_UNLOCKED || !io_file_can_poll(req)) {
 		/*
 		 * If we came in unlocked, we have no choice but to consume the
 		 * buffer here, otherwise nothing ensures that the buffer won't
diff --git a/io_uring/poll.c b/io_uring/poll.c
index 7513afc7b702..4afec733fef6 100644
--- a/io_uring/poll.c
+++ b/io_uring/poll.c
@@ -727,7 +727,7 @@ int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
 
 	if (!def->pollin && !def->pollout)
 		return IO_APOLL_ABORTED;
-	if (!file_can_poll(req->file))
+	if (!io_file_can_poll(req))
 		return IO_APOLL_ABORTED;
 	if (!(req->flags & REQ_F_APOLL_MULTISHOT))
 		mask |= EPOLLONESHOT;
diff --git a/io_uring/rw.c b/io_uring/rw.c
index d5e79d9bdc71..0fb7a045163a 100644
--- a/io_uring/rw.c
+++ b/io_uring/rw.c
@@ -682,7 +682,7 @@ static bool io_rw_should_retry(struct io_kiocb *req)
 	 * just use poll if we can, and don't attempt if the fs doesn't
 	 * support callback based unlocks
 	 */
-	if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
+	if (io_file_can_poll(req) || !(req->file->f_mode & FMODE_BUF_RASYNC))
 		return false;
 
 	wait->wait.func = io_async_buf_func;
@@ -831,7 +831,7 @@ static int __io_read(struct io_kiocb *req, unsigned int issue_flags)
 		 * If we can poll, just do that. For a vectored read, we'll
 		 * need to copy state first.
 		 */
-		if (file_can_poll(req->file) && !io_issue_defs[req->opcode].vectored)
+		if (io_file_can_poll(req) && !io_issue_defs[req->opcode].vectored)
 			return -EAGAIN;
 		/* IOPOLL retry should happen for io-wq threads */
 		if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
@@ -930,7 +930,7 @@ int io_read_mshot(struct io_kiocb *req, unsigned int issue_flags)
 	/*
 	 * Multishot MUST be used on a pollable file
 	 */
-	if (!file_can_poll(req->file))
+	if (!io_file_can_poll(req))
 		return -EBADFD;
 
 	ret = __io_read(req, issue_flags);
-- 
2.43.0

Re: [PATCH 2/6] io_uring: add io_file_can_poll() helper

[Pavel Begunkov]

On 2/6/24 16:22, Jens Axboe wrote:
> This adds a flag to avoid dipping dereferencing file and then f_op
> to figure out if the file has a poll handler defined or not. We
> generally call this at least twice for networked workloads.

Sends are not using poll every time. For recv, we touch it
in io_arm_poll_handler(), which is done only once, and so
ammortised to 0 for multishots.

Looking at the patch, the second time we might care about is
in io_ring_buffer_select(), but I'd argue that it shouldn't
be there in the first place. It's fragile, and I don't see
why selected buffers would care specifically about polling
but not asking more generally "can it go true async"? For
reads you might want to also test FMODE_BUF_RASYNC.

Also note that when called from recv we already know that
it's pollable, it might be much easier to pass it in as an
argument.


> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>   include/linux/io_uring_types.h |  3 +++
>   io_uring/io_uring.c            |  2 +-
>   io_uring/io_uring.h            | 12 ++++++++++++
>   io_uring/kbuf.c                |  2 +-
>   io_uring/poll.c                |  2 +-
>   io_uring/rw.c                  |  6 +++---
>   6 files changed, 21 insertions(+), 6 deletions(-)
> 
> diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
> index 5ac18b05d4ee..7f06cee02b58 100644
> --- a/include/linux/io_uring_types.h
> +++ b/include/linux/io_uring_types.h
> @@ -463,6 +463,7 @@ enum io_req_flags {
>   	REQ_F_SUPPORT_NOWAIT_BIT,
>   	REQ_F_ISREG_BIT,
>   	REQ_F_POLL_NO_LAZY_BIT,
> +	REQ_F_CAN_POLL_BIT,
>   
>   	/* not a real bit, just to check we're not overflowing the space */
>   	__REQ_F_LAST_BIT,
> @@ -535,6 +536,8 @@ enum {
>   	REQ_F_HASH_LOCKED	= IO_REQ_FLAG(REQ_F_HASH_LOCKED_BIT),
>   	/* don't use lazy poll wake for this request */
>   	REQ_F_POLL_NO_LAZY	= IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT),
> +	/* file is pollable */
> +	REQ_F_CAN_POLL		= IO_REQ_FLAG(REQ_F_CAN_POLL_BIT),
>   };
>   
>   typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
> diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
> index 360a7ee41d3a..d0e06784926f 100644
> --- a/io_uring/io_uring.c
> +++ b/io_uring/io_uring.c
> @@ -1969,7 +1969,7 @@ void io_wq_submit_work(struct io_wq_work *work)
>   	if (req->flags & REQ_F_FORCE_ASYNC) {
>   		bool opcode_poll = def->pollin || def->pollout;
>   
> -		if (opcode_poll && file_can_poll(req->file)) {
> +		if (opcode_poll && io_file_can_poll(req)) {
>   			needs_poll = true;
>   			issue_flags |= IO_URING_F_NONBLOCK;
>   		}
> diff --git a/io_uring/io_uring.h b/io_uring/io_uring.h
> index d5495710c178..2952551fe345 100644
> --- a/io_uring/io_uring.h
> +++ b/io_uring/io_uring.h
> @@ -5,6 +5,7 @@
>   #include <linux/lockdep.h>
>   #include <linux/resume_user_mode.h>
>   #include <linux/kasan.h>
> +#include <linux/poll.h>
>   #include <linux/io_uring_types.h>
>   #include <uapi/linux/eventpoll.h>
>   #include "io-wq.h"
> @@ -398,4 +399,15 @@ static inline size_t uring_sqe_size(struct io_ring_ctx *ctx)
>   		return 2 * sizeof(struct io_uring_sqe);
>   	return sizeof(struct io_uring_sqe);
>   }
> +
> +static inline bool io_file_can_poll(struct io_kiocb *req)
> +{
> +	if (req->flags & REQ_F_CAN_POLL)
> +		return true;
> +	if (file_can_poll(req->file)) {
> +		req->flags |= REQ_F_CAN_POLL;
> +		return true;
> +	}
> +	return false;
> +}
>   #endif
> diff --git a/io_uring/kbuf.c b/io_uring/kbuf.c
> index 18df5a9d2f5e..71880615bb78 100644
> --- a/io_uring/kbuf.c
> +++ b/io_uring/kbuf.c
> @@ -180,7 +180,7 @@ static void __user *io_ring_buffer_select(struct io_kiocb *req, size_t *len,
>   	req->buf_list = bl;
>   	req->buf_index = buf->bid;
>   
> -	if (issue_flags & IO_URING_F_UNLOCKED || !file_can_poll(req->file)) {
> +	if (issue_flags & IO_URING_F_UNLOCKED || !io_file_can_poll(req)) {
>   		/*
>   		 * If we came in unlocked, we have no choice but to consume the
>   		 * buffer here, otherwise nothing ensures that the buffer won't
> diff --git a/io_uring/poll.c b/io_uring/poll.c
> index 7513afc7b702..4afec733fef6 100644
> --- a/io_uring/poll.c
> +++ b/io_uring/poll.c
> @@ -727,7 +727,7 @@ int io_arm_poll_handler(struct io_kiocb *req, unsigned issue_flags)
>   
>   	if (!def->pollin && !def->pollout)
>   		return IO_APOLL_ABORTED;
> -	if (!file_can_poll(req->file))
> +	if (!io_file_can_poll(req))
>   		return IO_APOLL_ABORTED;
>   	if (!(req->flags & REQ_F_APOLL_MULTISHOT))
>   		mask |= EPOLLONESHOT;
> diff --git a/io_uring/rw.c b/io_uring/rw.c
> index d5e79d9bdc71..0fb7a045163a 100644
> --- a/io_uring/rw.c
> +++ b/io_uring/rw.c
> @@ -682,7 +682,7 @@ static bool io_rw_should_retry(struct io_kiocb *req)
>   	 * just use poll if we can, and don't attempt if the fs doesn't
>   	 * support callback based unlocks
>   	 */
> -	if (file_can_poll(req->file) || !(req->file->f_mode & FMODE_BUF_RASYNC))
> +	if (io_file_can_poll(req) || !(req->file->f_mode & FMODE_BUF_RASYNC))
>   		return false;
>   
>   	wait->wait.func = io_async_buf_func;
> @@ -831,7 +831,7 @@ static int __io_read(struct io_kiocb *req, unsigned int issue_flags)
>   		 * If we can poll, just do that. For a vectored read, we'll
>   		 * need to copy state first.
>   		 */
> -		if (file_can_poll(req->file) && !io_issue_defs[req->opcode].vectored)
> +		if (io_file_can_poll(req) && !io_issue_defs[req->opcode].vectored)
>   			return -EAGAIN;
>   		/* IOPOLL retry should happen for io-wq threads */
>   		if (!force_nonblock && !(req->ctx->flags & IORING_SETUP_IOPOLL))
> @@ -930,7 +930,7 @@ int io_read_mshot(struct io_kiocb *req, unsigned int issue_flags)
>   	/*
>   	 * Multishot MUST be used on a pollable file
>   	 */
> -	if (!file_can_poll(req->file))
> +	if (!io_file_can_poll(req))
>   		return -EBADFD;
>   
>   	ret = __io_read(req, issue_flags);

-- 
Pavel Begunkov

Re: [PATCH 2/6] io_uring: add io_file_can_poll() helper

[Jens Axboe]

On 2/6/24 5:57 PM, Pavel Begunkov wrote:
> On 2/6/24 16:22, Jens Axboe wrote:
>> This adds a flag to avoid dipping dereferencing file and then f_op
>> to figure out if the file has a poll handler defined or not. We
>> generally call this at least twice for networked workloads.
> 
> Sends are not using poll every time. For recv, we touch it
> in io_arm_poll_handler(), which is done only once, and so
> ammortised to 0 for multishots.

Correct

> Looking at the patch, the second time we might care about is
> in io_ring_buffer_select(), but I'd argue that it shouldn't
> be there in the first place. It's fragile, and I don't see
> why selected buffers would care specifically about polling
> but not asking more generally "can it go true async"? For
> reads you might want to also test FMODE_BUF_RASYNC.

That is indeed the second case that is hit, and I don't think we can
easily get around that which is the reason for the hint.

> Also note that when called from recv we already know that
> it's pollable, it might be much easier to pass it in as an
> argument.

I did think about that, but I don't see a clean way to do it. We could
potentially do it as an issue flag, but that seems kind of ugly to me.
Open to suggestions!

-- 
Jens Axboe

Re: [PATCH 2/6] io_uring: add io_file_can_poll() helper

[Pavel Begunkov]

On 2/7/24 02:15, Jens Axboe wrote:
> On 2/6/24 5:57 PM, Pavel Begunkov wrote:
>> On 2/6/24 16:22, Jens Axboe wrote:
>>> This adds a flag to avoid dipping dereferencing file and then f_op
>>> to figure out if the file has a poll handler defined or not. We
>>> generally call this at least twice for networked workloads.
>>
>> Sends are not using poll every time. For recv, we touch it
>> in io_arm_poll_handler(), which is done only once, and so
>> ammortised to 0 for multishots.
> 
> Correct
> 
>> Looking at the patch, the second time we might care about is
>> in io_ring_buffer_select(), but I'd argue that it shouldn't
>> be there in the first place. It's fragile, and I don't see
>> why selected buffers would care specifically about polling
>> but not asking more generally "can it go true async"? For
>> reads you might want to also test FMODE_BUF_RASYNC.
> 
> That is indeed the second case that is hit, and I don't think we can
> easily get around that which is the reason for the hint.
> 
>> Also note that when called from recv we already know that
>> it's pollable, it might be much easier to pass it in as an
>> argument.
> 
> I did think about that, but I don't see a clean way to do it. We could
> potentially do it as an issue flag, but that seems kind of ugly to me.
> Open to suggestions!

I'd argue passing it as an argument is much much cleaner
and more robust design wise, those leaked abstractions are
always fragile and unreliable. And now there is an argument
that it's even faster because for recv you can just pass "true".
IOW, I'd prefer here potentially a slightly uglier but safer
code.

Surely it'd have been be great to move this "eject buffer"
thing out of the selection func and let the caller decide,
but I haven't stared at the code for long enough to say
anything concrete.

-- 
Pavel Begunkov

[PATCH 3/6] io_uring/cancel: don't default to setting req->work.cancel_seq

[Jens Axboe]

Just leave it unset by default, avoiding dipping into the last
cacheline (which is otherwise untouched) for the fast path of using
poll to drive networked traffic. Add a flag that tells us if the
sequence is valid or not, and then we can defer actually assigning
the flag and sequence until someone runs cancelations.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/linux/io_uring_types.h |  3 +++
 io_uring/cancel.c              |  3 +--
 io_uring/cancel.h              | 10 ++++++++++
 io_uring/io_uring.c            |  1 -
 io_uring/poll.c                |  6 +-----
 5 files changed, 15 insertions(+), 8 deletions(-)

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 7f06cee02b58..69a043ff8460 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -464,6 +464,7 @@ enum io_req_flags {
 	REQ_F_ISREG_BIT,
 	REQ_F_POLL_NO_LAZY_BIT,
 	REQ_F_CAN_POLL_BIT,
+	REQ_F_CANCEL_SEQ_BIT,
 
 	/* not a real bit, just to check we're not overflowing the space */
 	__REQ_F_LAST_BIT,
@@ -538,6 +539,8 @@ enum {
 	REQ_F_POLL_NO_LAZY	= IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT),
 	/* file is pollable */
 	REQ_F_CAN_POLL		= IO_REQ_FLAG(REQ_F_CAN_POLL_BIT),
+	/* cancel sequence is set and valid */
+	REQ_F_CANCEL_SEQ	= IO_REQ_FLAG(REQ_F_CANCEL_SEQ_BIT),
 };
 
 typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
diff --git a/io_uring/cancel.c b/io_uring/cancel.c
index 8a8b07dfc444..acfcdd7f059a 100644
--- a/io_uring/cancel.c
+++ b/io_uring/cancel.c
@@ -58,9 +58,8 @@ bool io_cancel_req_match(struct io_kiocb *req, struct io_cancel_data *cd)
 		return false;
 	if (cd->flags & IORING_ASYNC_CANCEL_ALL) {
 check_seq:
-		if (cd->seq == req->work.cancel_seq)
+		if (io_cancel_match_sequence(req, cd->seq))
 			return false;
-		req->work.cancel_seq = cd->seq;
 	}
 
 	return true;
diff --git a/io_uring/cancel.h b/io_uring/cancel.h
index c0a8e7c520b6..76b32e65c03c 100644
--- a/io_uring/cancel.h
+++ b/io_uring/cancel.h
@@ -25,4 +25,14 @@ void init_hash_table(struct io_hash_table *table, unsigned size);
 int io_sync_cancel(struct io_ring_ctx *ctx, void __user *arg);
 bool io_cancel_req_match(struct io_kiocb *req, struct io_cancel_data *cd);
 
+static inline bool io_cancel_match_sequence(struct io_kiocb *req, int sequence)
+{
+	if ((req->flags & REQ_F_CANCEL_SEQ) && sequence == req->work.cancel_seq)
+		return true;
+
+	req->flags |= REQ_F_CANCEL_SEQ;
+	req->work.cancel_seq = sequence;
+	return false;
+}
+
 #endif
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index d0e06784926f..9b499864f10d 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -463,7 +463,6 @@ static void io_prep_async_work(struct io_kiocb *req)
 
 	req->work.list.next = NULL;
 	req->work.flags = 0;
-	req->work.cancel_seq = atomic_read(&ctx->cancel_seq);
 	if (req->flags & REQ_F_FORCE_ASYNC)
 		req->work.flags |= IO_WQ_WORK_CONCURRENT;
 
diff --git a/io_uring/poll.c b/io_uring/poll.c
index 4afec733fef6..3f3380dc5f68 100644
--- a/io_uring/poll.c
+++ b/io_uring/poll.c
@@ -588,10 +588,7 @@ static int __io_arm_poll_handler(struct io_kiocb *req,
 				 struct io_poll_table *ipt, __poll_t mask,
 				 unsigned issue_flags)
 {
-	struct io_ring_ctx *ctx = req->ctx;
-
 	INIT_HLIST_NODE(&req->hash_node);
-	req->work.cancel_seq = atomic_read(&ctx->cancel_seq);
 	io_init_poll_iocb(poll, mask);
 	poll->file = req->file;
 	req->apoll_events = poll->events;
@@ -818,9 +815,8 @@ static struct io_kiocb *io_poll_find(struct io_ring_ctx *ctx, bool poll_only,
 		if (poll_only && req->opcode != IORING_OP_POLL_ADD)
 			continue;
 		if (cd->flags & IORING_ASYNC_CANCEL_ALL) {
-			if (cd->seq == req->work.cancel_seq)
+			if (io_cancel_match_sequence(req, cd->seq))
 				continue;
-			req->work.cancel_seq = cd->seq;
 		}
 		*out_bucket = hb;
 		return req;
-- 
2.43.0

[PATCH 4/6] io_uring: move io_kiocb->nr_tw into comp_list union

[Jens Axboe]

comp_list is only used for completion purposes, which it why it
currently shares space with apoll_events (which is only used for poll
triggering). nr_rw is also not used with comp_list, the former is just
used for local task_list wakeup optimizations.

This doesn't save any space in io_kiocb, rather it now leaves a 32-bit
hole that can be used for something else, when the need arises.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/linux/io_uring_types.h | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/include/linux/io_uring_types.h b/include/linux/io_uring_types.h
index 69a043ff8460..8c0742f5b57e 100644
--- a/include/linux/io_uring_types.h
+++ b/include/linux/io_uring_types.h
@@ -602,8 +602,6 @@ struct io_kiocb {
 	 */
 	u16				buf_index;
 
-	atomic_t			refs;
-
 	io_req_flags_t			flags;
 
 	struct io_ring_ctx		*ctx;
@@ -629,8 +627,11 @@ struct io_kiocb {
 	union {
 		/* used by request caches, completion batching and iopoll */
 		struct io_wq_work_node	comp_list;
-		/* cache ->apoll->events */
-		__poll_t apoll_events;
+		struct {
+			/* cache ->apoll->events */
+			__poll_t apoll_events;
+			unsigned nr_tw;
+		};
 	};
 
 	struct io_rsrc_node		*rsrc_node;
@@ -639,7 +640,7 @@ struct io_kiocb {
 
 	struct io_task_work		io_task_work;
 	atomic_t			poll_refs;
-	unsigned			nr_tw;
+	atomic_t			refs;
 	/* internal polling, see IORING_FEAT_FAST_POLL */
 	struct async_poll		*apoll;
 	/* opcode allocated if it needs to store data for async defer */
-- 
2.43.0

[PATCH 5/6] io_uring: mark the need to lock/unlock the ring as unlikely

[Jens Axboe]

Any of the fast paths will already have this locked, this helper only
exists to deal with io-wq invoking request issue where we do not have
the ctx->uring_lock held already. This means that any common or fast
path will already have this locked, mark it as such.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/io_uring.h | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/io_uring/io_uring.h b/io_uring/io_uring.h
index 2952551fe345..46795ee462df 100644
--- a/io_uring/io_uring.h
+++ b/io_uring/io_uring.h
@@ -208,7 +208,7 @@ static inline void io_ring_submit_unlock(struct io_ring_ctx *ctx,
 					 unsigned issue_flags)
 {
 	lockdep_assert_held(&ctx->uring_lock);
-	if (issue_flags & IO_URING_F_UNLOCKED)
+	if (unlikely(issue_flags & IO_URING_F_UNLOCKED))
 		mutex_unlock(&ctx->uring_lock);
 }
 
@@ -221,7 +221,7 @@ static inline void io_ring_submit_lock(struct io_ring_ctx *ctx,
 	 * The only exception is when we've detached the request and issue it
 	 * from an async worker thread, grab the lock for that case.
 	 */
-	if (issue_flags & IO_URING_F_UNLOCKED)
+	if (unlikely(issue_flags & IO_URING_F_UNLOCKED))
 		mutex_lock(&ctx->uring_lock);
 	lockdep_assert_held(&ctx->uring_lock);
 }
-- 
2.43.0

[PATCH 6/6] io_uring/rw: remove dead file == NULL check

[Jens Axboe]

Any read/write opcode has needs_file == true, which means that we
would've failed the request long before reaching the issue stage if we
didn't successfully assign a file. This check has been dead forever,
and is really a leftover from generic code.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/rw.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/io_uring/rw.c b/io_uring/rw.c
index 0fb7a045163a..8ba93fffc23a 100644
--- a/io_uring/rw.c
+++ b/io_uring/rw.c
@@ -721,7 +721,7 @@ static int io_rw_init_file(struct io_kiocb *req, fmode_t mode)
 	struct file *file = req->file;
 	int ret;
 
-	if (unlikely(!file || !(file->f_mode & mode)))
+	if (unlikely(!(file->f_mode & mode)))
 		return -EBADF;
 
 	if (!(req->flags & REQ_F_FIXED_FILE))
-- 
2.43.0

[PATCHSET v2 0/6] Misc cleanups / optimizations

[Jens Axboe]

Hi,

Nothing major in here:

- Expand io_kiocb flags to 64-bits, so we can use two more bits for
  caching cancelation sequence and pollable state.
- Misc cleanups

Changes since v1:
- Drop nr_tw union with comp_list, that breaks iopoll with DEFER_TASKRUN
  usage.
- Rearrange io_kiocb again in patch 1, now just moving nr_tw up to fill
  the new hole, and shifting rsrc_node down to keep io_comp_list in the
  2nd cacheline.
- Add cleanup patch for io_req_complete_post()

-- 
Jens Axboe

[PATCH 6/6] io_uring/rw: remove dead file == NULL check

[Jens Axboe]

Any read/write opcode has needs_file == true, which means that we
would've failed the request long before reaching the issue stage if we
didn't successfully assign a file. This check has been dead forever,
and is really a leftover from generic code.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 io_uring/rw.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/io_uring/rw.c b/io_uring/rw.c
index 0fb7a045163a..8ba93fffc23a 100644
--- a/io_uring/rw.c
+++ b/io_uring/rw.c
@@ -721,7 +721,7 @@ static int io_rw_init_file(struct io_kiocb *req, fmode_t mode)
 	struct file *file = req->file;
 	int ret;
 
-	if (unlikely(!file || !(file->f_mode & mode)))
+	if (unlikely(!(file->f_mode & mode)))
 		return -EBADF;
 
 	if (!(req->flags & REQ_F_FIXED_FILE))
-- 
2.43.0