[PATCH v2 00/12] add large CQE support for io-uring

[PATCH v2 00/12] add large CQE support for io-uring

[Stefan Roesch]

This adds the large CQE support for io-uring. Large CQE's are 16 bytes longer.
To support the longer CQE's the allocation part is changed and when the CQE is
accessed.

The allocation of the large CQE's is twice as big, so the allocation size is
doubled. The ring size calculation needs to take this into account.

All accesses to the large CQE's need to be shifted by 1 to take the bigger size
of each CQE into account. The existing index manipulation does not need to be
changed and can stay the same.

The setup and the completion processing needs to take the new fields into
account and initialize them. For the completion processing these fields need
to be passed through.

The flush completion processing needs to fill the additional CQE32 fields.

The code for overflows needs to be adapted accordingly: the allocation needs to
take large CQE's into account. This means that the order of the fields in the io
overflow structure needs to be changed and the allocation needs to be enlarged
for big CQE's.
In addition the two new fields need to be copied for large CQE's.

The new fields are added to the tracing statements, so the extra1 and extra2
fields are exposed in tracing. The new fields are also exposed in the /proc
filesystem entry.

For testing purposes the extra1 and extra2 fields are used by the nop operation.


Testing:

The exisiting tests have been run with the following configurations and they all
pass:

- Default config
- Large SQE
- Large CQE
- Large SQE and large CQE.

In addition a new test has been added to liburing to verify that extra1 and extra2
are set as expected for the nop operation.

Note:
To use this patch also the corresponding changes to the client library
liburing are required. A different patch series is sent out for this.


Changes:
  V2: - added support for CQE32 in the /proc filesystem entry output function
      - the definition of the io_uring_cqe_extra field has been changed
        to avoid warning with the /proc changes.


Stefan Roesch (12):
  io_uring: support CQE32 in io_uring_cqe
  io_uring: wire up inline completion path for CQE32
  io_uring: change ring size calculation for CQE32
  io_uring: add CQE32 setup processing
  io_uring: add CQE32 completion processing
  io_uring: modify io_get_cqe for CQE32
  io_uring: flush completions for CQE32
  io_uring: overflow processing for CQE32
  io_uring: add tracing for additional CQE32 fields
  io_uring: support CQE32 in /proc info
  io_uring: enable CQE32
  io_uring: support CQE32 for nop operation

 fs/io_uring.c                   | 224 +++++++++++++++++++++++++++-----
 include/trace/events/io_uring.h |  18 ++-
 include/uapi/linux/io_uring.h   |  12 ++
 3 files changed, 220 insertions(+), 34 deletions(-)


base-commit: fd1cf8f1947eb7b009eb79807ec8af0e920fc57b
-- 
2.30.2

[PATCH v2 01/12] io_uring: support CQE32 in io_uring_cqe

[Stefan Roesch]

This adds the struct io_uring_cqe_extra in the structure io_uring_cqe to
support large CQE's.

Co-developed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 include/uapi/linux/io_uring.h | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index ee677dbd6a6d..c0e9b5e8d20c 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -111,6 +111,7 @@ enum {
 #define IORING_SETUP_R_DISABLED	(1U << 6)	/* start with ring disabled */
 #define IORING_SETUP_SUBMIT_ALL	(1U << 7)	/* continue submit on error */
 #define IORING_SETUP_SQE128	(1U << 8)	/* SQEs are 128b */
+#define IORING_SETUP_CQE32	(1U << 9)	/* CQEs are 32b */
 
 enum {
 	IORING_OP_NOP,
@@ -201,6 +202,11 @@ enum {
 #define IORING_POLL_UPDATE_EVENTS	(1U << 1)
 #define IORING_POLL_UPDATE_USER_DATA	(1U << 2)
 
+struct io_uring_cqe_extra {
+	__u64	extra1;
+	__u64	extra2;
+};
+
 /*
  * IO completion data structure (Completion Queue Entry)
  */
@@ -208,6 +214,12 @@ struct io_uring_cqe {
 	__u64	user_data;	/* sqe->data submission passed back */
 	__s32	res;		/* result code for this event */
 	__u32	flags;
+
+	/*
+	 * If the ring is initialized with IORING_SETUP_CQE32, then this field
+	 * contains 16-bytes of padding, doubling the size of the CQE.
+	 */
+	struct io_uring_cqe_extra	b[];
 };
 
 /*
-- 
2.30.2

[PATCH v2 02/12] io_uring: wire up inline completion path for CQE32

[Stefan Roesch]

Rather than always use the slower locked path, wire up use of the
deferred completion path that normal CQEs can take. This reuses the
hash list node for the storage we need to hold the two 64-bit values
that must be passed back.

Co-developed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 4c32cf987ef3..bf2b02518332 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -964,7 +964,13 @@ struct io_kiocb {
 	atomic_t			poll_refs;
 	struct io_task_work		io_task_work;
 	/* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
-	struct hlist_node		hash_node;
+	union {
+		struct hlist_node	hash_node;
+		struct {
+			u64		extra1;
+			u64		extra2;
+		};
+	};
 	/* internal polling, see IORING_FEAT_FAST_POLL */
 	struct async_poll		*apoll;
 	/* opcode allocated if it needs to store data for async defer */
-- 
2.30.2

[PATCH v2 03/12] io_uring: change ring size calculation for CQE32

[Stefan Roesch]

This changes the function rings_size to take large CQE's into account.

Co-developed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c | 10 +++++++---
 1 file changed, 7 insertions(+), 3 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index bf2b02518332..9712483d3a17 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -9693,8 +9693,8 @@ static void *io_mem_alloc(size_t size)
 	return (void *) __get_free_pages(gfp, get_order(size));
 }
 
-static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
-				size_t *sq_offset)
+static unsigned long rings_size(struct io_ring_ctx *ctx, unsigned int sq_entries,
+				unsigned int cq_entries, size_t *sq_offset)
 {
 	struct io_rings *rings;
 	size_t off, sq_array_size;
@@ -9702,6 +9702,10 @@ static unsigned long rings_size(unsigned sq_entries, unsigned cq_entries,
 	off = struct_size(rings, cqes, cq_entries);
 	if (off == SIZE_MAX)
 		return SIZE_MAX;
+	if (ctx->flags & IORING_SETUP_CQE32) {
+		if (check_shl_overflow(off, 1, &off))
+			return SIZE_MAX;
+	}
 
 #ifdef CONFIG_SMP
 	off = ALIGN(off, SMP_CACHE_BYTES);
@@ -11365,7 +11369,7 @@ static __cold int io_allocate_scq_urings(struct io_ring_ctx *ctx,
 	ctx->sq_entries = p->sq_entries;
 	ctx->cq_entries = p->cq_entries;
 
-	size = rings_size(p->sq_entries, p->cq_entries, &sq_array_offset);
+	size = rings_size(ctx, p->sq_entries, p->cq_entries, &sq_array_offset);
 	if (size == SIZE_MAX)
 		return -EOVERFLOW;
 
-- 
2.30.2

[PATCH v2 04/12] io_uring: add CQE32 setup processing

[Stefan Roesch]

This adds two new function to setup and fill the CQE32 result structure.

Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c | 58 +++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 58 insertions(+)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 9712483d3a17..abbd2efbe255 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2175,12 +2175,70 @@ static inline bool __io_fill_cqe_req_filled(struct io_ring_ctx *ctx,
 					req->cqe.res, req->cqe.flags);
 }
 
+static inline bool __io_fill_cqe32_req_filled(struct io_ring_ctx *ctx,
+					      struct io_kiocb *req)
+{
+	struct io_uring_cqe *cqe;
+	u64 extra1 = req->extra1;
+	u64 extra2 = req->extra2;
+
+	trace_io_uring_complete(req->ctx, req, req->cqe.user_data,
+				req->cqe.res, req->cqe.flags);
+
+	/*
+	 * If we can't get a cq entry, userspace overflowed the
+	 * submission (by quite a lot). Increment the overflow count in
+	 * the ring.
+	 */
+	cqe = io_get_cqe(ctx);
+	if (likely(cqe)) {
+		memcpy(cqe, &req->cqe, sizeof(struct io_uring_cqe));
+		cqe->b[0].extra1 = extra1;
+		cqe->b[0].extra2 = extra2;
+		return true;
+	}
+
+	return io_cqring_event_overflow(ctx, req->cqe.user_data,
+					req->cqe.res, req->cqe.flags, extra1, extra2);
+}
+
 static inline bool __io_fill_cqe_req(struct io_kiocb *req, s32 res, u32 cflags)
 {
 	trace_io_uring_complete(req->ctx, req, req->cqe.user_data, res, cflags);
 	return __io_fill_cqe(req->ctx, req->cqe.user_data, res, cflags);
 }
 
+static void __io_fill_cqe32_req(struct io_kiocb *req, s32 res, u32 cflags,
+				u64 extra1, u64 extra2)
+{
+	struct io_ring_ctx *ctx = req->ctx;
+	struct io_uring_cqe *cqe;
+
+	if (WARN_ON_ONCE(!(ctx->flags & IORING_SETUP_CQE32)))
+		return;
+	if (req->flags & REQ_F_CQE_SKIP)
+		return;
+
+	trace_io_uring_complete(ctx, req, req->user_data, res, cflags);
+
+	/*
+	 * If we can't get a cq entry, userspace overflowed the
+	 * submission (by quite a lot). Increment the overflow count in
+	 * the ring.
+	 */
+	cqe = io_get_cqe(ctx);
+	if (likely(cqe)) {
+		WRITE_ONCE(cqe->user_data, req->cqe.user_data);
+		WRITE_ONCE(cqe->res, res);
+		WRITE_ONCE(cqe->flags, cflags);
+		WRITE_ONCE(cqe->b[0].extra1, extra1);
+		WRITE_ONCE(cqe->b[0].extra2, extra2);
+		return;
+	}
+
+	io_cqring_event_overflow(ctx, req->cqe.user_data, res, cflags);
+}
+
 static noinline bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data,
 				     s32 res, u32 cflags)
 {
-- 
2.30.2

[PATCH v2 05/12] io_uring: add CQE32 completion processing

[Stefan Roesch]

This adds the completion processing for the large CQE's and makes sure
that the extra1 and extra2 fields are passed through.

Co-developed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c | 55 +++++++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 47 insertions(+), 8 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index abbd2efbe255..c93a9353c88d 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2247,18 +2247,15 @@ static noinline bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data,
 	return __io_fill_cqe(ctx, user_data, res, cflags);
 }
 
-static void __io_req_complete_post(struct io_kiocb *req, s32 res,
-				   u32 cflags)
+static void __io_req_complete_put(struct io_kiocb *req)
 {
-	struct io_ring_ctx *ctx = req->ctx;
-
-	if (!(req->flags & REQ_F_CQE_SKIP))
-		__io_fill_cqe_req(req, res, cflags);
 	/*
 	 * If we're the last reference to this request, add to our locked
 	 * free_list cache.
 	 */
 	if (req_ref_put_and_test(req)) {
+		struct io_ring_ctx *ctx = req->ctx;
+
 		if (req->flags & IO_REQ_LINK_FLAGS) {
 			if (req->flags & IO_DISARM_MASK)
 				io_disarm_next(req);
@@ -2281,8 +2278,23 @@ static void __io_req_complete_post(struct io_kiocb *req, s32 res,
 	}
 }
 
-static void io_req_complete_post(struct io_kiocb *req, s32 res,
-				 u32 cflags)
+static void __io_req_complete_post(struct io_kiocb *req, s32 res,
+				   u32 cflags)
+{
+	if (!(req->flags & REQ_F_CQE_SKIP))
+		__io_fill_cqe_req(req, res, cflags);
+	__io_req_complete_put(req);
+}
+
+static void __io_req_complete_post32(struct io_kiocb *req, s32 res,
+				   u32 cflags, u64 extra1, u64 extra2)
+{
+	if (!(req->flags & REQ_F_CQE_SKIP))
+		__io_fill_cqe32_req(req, res, cflags, extra1, extra2);
+	__io_req_complete_put(req);
+}
+
+static void io_req_complete_post(struct io_kiocb *req, s32 res, u32 cflags)
 {
 	struct io_ring_ctx *ctx = req->ctx;
 
@@ -2293,6 +2305,18 @@ static void io_req_complete_post(struct io_kiocb *req, s32 res,
 	io_cqring_ev_posted(ctx);
 }
 
+static void io_req_complete_post32(struct io_kiocb *req, s32 res,
+				   u32 cflags, u64 extra1, u64 extra2)
+{
+	struct io_ring_ctx *ctx = req->ctx;
+
+	spin_lock(&ctx->completion_lock);
+	__io_req_complete_post32(req, res, cflags, extra1, extra2);
+	io_commit_cqring(ctx);
+	spin_unlock(&ctx->completion_lock);
+	io_cqring_ev_posted(ctx);
+}
+
 static inline void io_req_complete_state(struct io_kiocb *req, s32 res,
 					 u32 cflags)
 {
@@ -2310,6 +2334,21 @@ static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
 		io_req_complete_post(req, res, cflags);
 }
 
+static inline void __io_req_complete32(struct io_kiocb *req,
+				       unsigned int issue_flags, s32 res,
+				       u32 cflags, u64 extra1, u64 extra2)
+{
+	if (issue_flags & IO_URING_F_COMPLETE_DEFER) {
+		req->cqe.res = res;
+		req->cqe.flags = cflags;
+		req->extra1 = extra1;
+		req->extra2 = extra2;
+		req->flags |= REQ_F_COMPLETE_INLINE;
+	} else {
+		io_req_complete_post32(req, res, cflags, extra1, extra2);
+	}
+}
+
 static inline void io_req_complete(struct io_kiocb *req, s32 res)
 {
 	__io_req_complete(req, 0, res, 0);
-- 
2.30.2

[PATCH v2 06/12] io_uring: modify io_get_cqe for CQE32

[Stefan Roesch]

Modify accesses to the CQE array to take large CQE's into account. The
index needs to be shifted by one for large CQE's.

Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index c93a9353c88d..bd352815b9e7 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1909,8 +1909,12 @@ static noinline struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx)
 {
 	struct io_rings *rings = ctx->rings;
 	unsigned int off = ctx->cached_cq_tail & (ctx->cq_entries - 1);
+	unsigned int shift = 0;
 	unsigned int free, queued, len;
 
+	if (ctx->flags & IORING_SETUP_CQE32)
+		shift = 1;
+
 	/* userspace may cheat modifying the tail, be safe and do min */
 	queued = min(__io_cqring_events(ctx), ctx->cq_entries);
 	free = ctx->cq_entries - queued;
@@ -1922,12 +1926,13 @@ static noinline struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx)
 	ctx->cached_cq_tail++;
 	ctx->cqe_cached = &rings->cqes[off];
 	ctx->cqe_sentinel = ctx->cqe_cached + len;
-	return ctx->cqe_cached++;
+	ctx->cqe_cached++;
+	return &rings->cqes[off << shift];
 }
 
 static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx)
 {
-	if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) {
+	if (likely(ctx->cqe_cached < ctx->cqe_sentinel && !(ctx->flags & IORING_SETUP_CQE32))) {
 		ctx->cached_cq_tail++;
 		return ctx->cqe_cached++;
 	}
-- 
2.30.2

[PATCH v2 07/12] io_uring: flush completions for CQE32

[Stefan Roesch]

This flushes the completions according to their CQE type: the same
processing is done for the default CQE size, but for large CQE's the
extra1 and extra2 fields are filled in.

Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index bd352815b9e7..ff6229b6df16 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2877,8 +2877,12 @@ static void __io_submit_flush_completions(struct io_ring_ctx *ctx)
 			struct io_kiocb *req = container_of(node, struct io_kiocb,
 						    comp_list);
 
-			if (!(req->flags & REQ_F_CQE_SKIP))
-				__io_fill_cqe_req_filled(ctx, req);
+			if (!(req->flags & REQ_F_CQE_SKIP)) {
+				if (!(ctx->flags & IORING_SETUP_CQE32))
+					__io_fill_cqe_req_filled(ctx, req);
+				else
+					__io_fill_cqe32_req_filled(ctx, req);
+			}
 		}
 
 		io_commit_cqring(ctx);
-- 
2.30.2

[PATCH v2 08/12] io_uring: overflow processing for CQE32

[Stefan Roesch]

This adds the overflow processing for large CQE's.

This adds two parameters to the io_cqring_event_overflow function and
uses these fields to initialize the large CQE fields.

Allocate enough space for large CQE's in the overflow structue. If no
large CQE's are used, the size of the allocation is unchanged.

The cqe field can have a different size depending if its a large
CQE or not. To be able to allocate different sizes, the two fields
in the structure are re-ordered.

Co-developed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c | 26 +++++++++++++++++++-------
 1 file changed, 19 insertions(+), 7 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index ff6229b6df16..50efced63ec9 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -220,8 +220,8 @@ struct io_mapped_ubuf {
 struct io_ring_ctx;
 
 struct io_overflow_cqe {
-	struct io_uring_cqe cqe;
 	struct list_head list;
+	struct io_uring_cqe cqe;
 };
 
 struct io_fixed_file {
@@ -2016,13 +2016,17 @@ static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
 	while (!list_empty(&ctx->cq_overflow_list)) {
 		struct io_uring_cqe *cqe = io_get_cqe(ctx);
 		struct io_overflow_cqe *ocqe;
+		size_t cqe_size = sizeof(struct io_uring_cqe);
+
+		if (ctx->flags & IORING_SETUP_CQE32)
+			cqe_size <<= 1;
 
 		if (!cqe && !force)
 			break;
 		ocqe = list_first_entry(&ctx->cq_overflow_list,
 					struct io_overflow_cqe, list);
 		if (cqe)
-			memcpy(cqe, &ocqe->cqe, sizeof(*cqe));
+			memcpy(cqe, &ocqe->cqe, cqe_size);
 		else
 			io_account_cq_overflow(ctx);
 
@@ -2111,11 +2115,15 @@ static __cold void io_uring_drop_tctx_refs(struct task_struct *task)
 }
 
 static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data,
-				     s32 res, u32 cflags)
+				     s32 res, u32 cflags, u64 extra1, u64 extra2)
 {
 	struct io_overflow_cqe *ocqe;
+	size_t ocq_size = sizeof(struct io_overflow_cqe);
 
-	ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT);
+	if (ctx->flags & IORING_SETUP_CQE32)
+		ocq_size += sizeof(struct io_uring_cqe);
+
+	ocqe = kmalloc(ocq_size, GFP_ATOMIC | __GFP_ACCOUNT);
 	if (!ocqe) {
 		/*
 		 * If we're in ring overflow flush mode, or in task cancel mode,
@@ -2134,6 +2142,10 @@ static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data,
 	ocqe->cqe.user_data = user_data;
 	ocqe->cqe.res = res;
 	ocqe->cqe.flags = cflags;
+	if (ctx->flags & IORING_SETUP_CQE32) {
+		ocqe->cqe.b[0].extra1 = extra1;
+		ocqe->cqe.b[0].extra2 = extra2;
+	}
 	list_add_tail(&ocqe->list, &ctx->cq_overflow_list);
 	return true;
 }
@@ -2155,7 +2167,7 @@ static inline bool __io_fill_cqe(struct io_ring_ctx *ctx, u64 user_data,
 		WRITE_ONCE(cqe->flags, cflags);
 		return true;
 	}
-	return io_cqring_event_overflow(ctx, user_data, res, cflags);
+	return io_cqring_event_overflow(ctx, user_data, res, cflags, 0, 0);
 }
 
 static inline bool __io_fill_cqe_req_filled(struct io_ring_ctx *ctx,
@@ -2177,7 +2189,7 @@ static inline bool __io_fill_cqe_req_filled(struct io_ring_ctx *ctx,
 		return true;
 	}
 	return io_cqring_event_overflow(ctx, req->cqe.user_data,
-					req->cqe.res, req->cqe.flags);
+					req->cqe.res, req->cqe.flags, 0, 0);
 }
 
 static inline bool __io_fill_cqe32_req_filled(struct io_ring_ctx *ctx,
@@ -2241,7 +2253,7 @@ static void __io_fill_cqe32_req(struct io_kiocb *req, s32 res, u32 cflags,
 		return;
 	}
 
-	io_cqring_event_overflow(ctx, req->cqe.user_data, res, cflags);
+	io_cqring_event_overflow(ctx, req->cqe.user_data, res, cflags, extra1, extra2);
 }
 
 static noinline bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data,
-- 
2.30.2

[PATCH v2 09/12] io_uring: add tracing for additional CQE32 fields

[Stefan Roesch]

This adds tracing for the extra1 and extra2 fields.

Co-developed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c                   | 11 ++++++-----
 include/trace/events/io_uring.h | 18 ++++++++++++++----
 2 files changed, 20 insertions(+), 9 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 50efced63ec9..366f49969b31 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -2176,7 +2176,7 @@ static inline bool __io_fill_cqe_req_filled(struct io_ring_ctx *ctx,
 	struct io_uring_cqe *cqe;
 
 	trace_io_uring_complete(req->ctx, req, req->cqe.user_data,
-				req->cqe.res, req->cqe.flags);
+				req->cqe.res, req->cqe.flags, 0, 0);
 
 	/*
 	 * If we can't get a cq entry, userspace overflowed the
@@ -2200,7 +2200,7 @@ static inline bool __io_fill_cqe32_req_filled(struct io_ring_ctx *ctx,
 	u64 extra2 = req->extra2;
 
 	trace_io_uring_complete(req->ctx, req, req->cqe.user_data,
-				req->cqe.res, req->cqe.flags);
+				req->cqe.res, req->cqe.flags, extra1, extra2);
 
 	/*
 	 * If we can't get a cq entry, userspace overflowed the
@@ -2221,7 +2221,7 @@ static inline bool __io_fill_cqe32_req_filled(struct io_ring_ctx *ctx,
 
 static inline bool __io_fill_cqe_req(struct io_kiocb *req, s32 res, u32 cflags)
 {
-	trace_io_uring_complete(req->ctx, req, req->cqe.user_data, res, cflags);
+	trace_io_uring_complete(req->ctx, req, req->cqe.user_data, res, cflags, 0, 0);
 	return __io_fill_cqe(req->ctx, req->cqe.user_data, res, cflags);
 }
 
@@ -2236,7 +2236,8 @@ static void __io_fill_cqe32_req(struct io_kiocb *req, s32 res, u32 cflags,
 	if (req->flags & REQ_F_CQE_SKIP)
 		return;
 
-	trace_io_uring_complete(ctx, req, req->user_data, res, cflags);
+	trace_io_uring_complete(ctx, req, req->cqe.user_data, res, cflags,
+				extra1, extra2);
 
 	/*
 	 * If we can't get a cq entry, userspace overflowed the
@@ -2260,7 +2261,7 @@ static noinline bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data,
 				     s32 res, u32 cflags)
 {
 	ctx->cq_extra++;
-	trace_io_uring_complete(ctx, NULL, user_data, res, cflags);
+	trace_io_uring_complete(ctx, NULL, user_data, res, cflags, 0, 0);
 	return __io_fill_cqe(ctx, user_data, res, cflags);
 }
 
diff --git a/include/trace/events/io_uring.h b/include/trace/events/io_uring.h
index 8477414d6d06..2eb4f4e47de4 100644
--- a/include/trace/events/io_uring.h
+++ b/include/trace/events/io_uring.h
@@ -318,13 +318,16 @@ TRACE_EVENT(io_uring_fail_link,
  * @user_data:		user data associated with the request
  * @res:		result of the request
  * @cflags:		completion flags
+ * @extra1:		extra 64-bit data for CQE32
+ * @extra2:		extra 64-bit data for CQE32
  *
  */
 TRACE_EVENT(io_uring_complete,
 
-	TP_PROTO(void *ctx, void *req, u64 user_data, int res, unsigned cflags),
+	TP_PROTO(void *ctx, void *req, u64 user_data, int res, unsigned cflags,
+		 u64 extra1, u64 extra2),
 
-	TP_ARGS(ctx, req, user_data, res, cflags),
+	TP_ARGS(ctx, req, user_data, res, cflags, extra1, extra2),
 
 	TP_STRUCT__entry (
 		__field(  void *,	ctx		)
@@ -332,6 +335,8 @@ TRACE_EVENT(io_uring_complete,
 		__field(  u64,		user_data	)
 		__field(  int,		res		)
 		__field(  unsigned,	cflags		)
+		__field(  u64,		extra1		)
+		__field(  u64,		extra2		)
 	),
 
 	TP_fast_assign(
@@ -340,12 +345,17 @@ TRACE_EVENT(io_uring_complete,
 		__entry->user_data	= user_data;
 		__entry->res		= res;
 		__entry->cflags		= cflags;
+		__entry->extra1		= extra1;
+		__entry->extra2		= extra2;
 	),
 
-	TP_printk("ring %p, req %p, user_data 0x%llx, result %d, cflags 0x%x",
+	TP_printk("ring %p, req %p, user_data 0x%llx, result %d, cflags 0x%x "
+		  "extra1 %llu extra2 %llu ",
 		__entry->ctx, __entry->req,
 		__entry->user_data,
-		__entry->res, __entry->cflags)
+		__entry->res, __entry->cflags,
+		(unsigned long long) __entry->extra1,
+		(unsigned long long) __entry->extra2)
 );
 
 /**
-- 
2.30.2

[PATCH v2 10/12] io_uring: support CQE32 in /proc info

[Stefan Roesch]

This exposes the extra1 and extra2 fields in the /proc output.

Signed-off-by: Stefan Roesch <shr@fb.com>
---
 fs/io_uring.c | 15 +++++++++++++--
 1 file changed, 13 insertions(+), 2 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 366f49969b31..a5fbad91800c 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -11345,10 +11345,14 @@ static __cold void __io_uring_show_fdinfo(struct io_ring_ctx *ctx,
 	unsigned int sq_tail = READ_ONCE(r->sq.tail);
 	unsigned int cq_head = READ_ONCE(r->cq.head);
 	unsigned int cq_tail = READ_ONCE(r->cq.tail);
+	unsigned int cq_shift = 0;
 	unsigned int sq_entries, cq_entries;
 	bool has_lock;
 	unsigned int i;
 
+	if (ctx->flags & IORING_SETUP_CQE32)
+		cq_shift = 1;
+
 	/*
 	 * we may get imprecise sqe and cqe info if uring is actively running
 	 * since we get cached_sq_head and cached_cq_tail without uring_lock
@@ -11381,11 +11385,18 @@ static __cold void __io_uring_show_fdinfo(struct io_ring_ctx *ctx,
 	cq_entries = min(cq_tail - cq_head, ctx->cq_entries);
 	for (i = 0; i < cq_entries; i++) {
 		unsigned int entry = i + cq_head;
-		struct io_uring_cqe *cqe = &r->cqes[entry & cq_mask];
+		struct io_uring_cqe *cqe = &r->cqes[(entry & cq_mask) << cq_shift];
 
-		seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x\n",
+		if (!(ctx->flags & IORING_SETUP_CQE32)) {
+			seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x\n",
 			   entry & cq_mask, cqe->user_data, cqe->res,
 			   cqe->flags);
+		} else {
+			seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x, "
+				"extra1:%llu, extra2:%llu\n",
+				entry & cq_mask, cqe->user_data, cqe->res,
+				cqe->flags, cqe->b[0].extra1, cqe->b[0].extra2);
+		}
 	}
 
 	/*
-- 
2.30.2

[PATCH v2 11/12] io_uring: enable CQE32

[Stefan Roesch]

This enables large CQE's in the uring setup.

Co-developed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index a5fbad91800c..8bdf253b9462 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -11742,7 +11742,7 @@ static long io_uring_setup(u32 entries, struct io_uring_params __user *params)
 			IORING_SETUP_SQ_AFF | IORING_SETUP_CQSIZE |
 			IORING_SETUP_CLAMP | IORING_SETUP_ATTACH_WQ |
 			IORING_SETUP_R_DISABLED | IORING_SETUP_SUBMIT_ALL |
-			IORING_SETUP_SQE128))
+			IORING_SETUP_SQE128 | IORING_SETUP_CQE32))
 		return -EINVAL;
 
 	return  io_uring_create(entries, &p, params);
-- 
2.30.2

[PATCH v2 12/12] io_uring: support CQE32 for nop operation

[Stefan Roesch]

This adds support for filling the extra1 and extra2 fields for large
CQE's.

Co-developed-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Stefan Roesch <shr@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 fs/io_uring.c | 28 ++++++++++++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 8bdf253b9462..3855148b6774 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -744,6 +744,12 @@ struct io_msg {
 	u32 len;
 };
 
+struct io_nop {
+	struct file			*file;
+	u64				extra1;
+	u64				extra2;
+};
+
 struct io_async_connect {
 	struct sockaddr_storage		address;
 };
@@ -937,6 +943,7 @@ struct io_kiocb {
 		struct io_msg		msg;
 		struct io_xattr		xattr;
 		struct io_socket	sock;
+		struct io_nop		nop;
 	};
 
 	u8				opcode;
@@ -4863,6 +4870,19 @@ static int io_splice(struct io_kiocb *req, unsigned int issue_flags)
 	return 0;
 }
 
+static int io_nop_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
+{
+	/*
+	 * If the ring is setup with CQE32, relay back addr/addr
+	 */
+	if (req->ctx->flags & IORING_SETUP_CQE32) {
+		req->nop.extra1 = READ_ONCE(sqe->addr);
+		req->nop.extra2 = READ_ONCE(sqe->addr2);
+	}
+
+	return 0;
+}
+
 /*
  * IORING_OP_NOP just posts a completion event, nothing else.
  */
@@ -4873,7 +4893,11 @@ static int io_nop(struct io_kiocb *req, unsigned int issue_flags)
 	if (unlikely(ctx->flags & IORING_SETUP_IOPOLL))
 		return -EINVAL;
 
-	__io_req_complete(req, issue_flags, 0, 0);
+	if (!(ctx->flags & IORING_SETUP_CQE32))
+		__io_req_complete(req, issue_flags, 0, 0);
+	else
+		__io_req_complete32(req, issue_flags, 0, 0, req->nop.extra1,
+					req->nop.extra2);
 	return 0;
 }
 
@@ -7345,7 +7369,7 @@ static int io_req_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
 	switch (req->opcode) {
 	case IORING_OP_NOP:
-		return 0;
+		return io_nop_prep(req, sqe);
 	case IORING_OP_READV:
 	case IORING_OP_READ_FIXED:
 	case IORING_OP_READ:
-- 
2.30.2

Re: [PATCH v2 00/12] add large CQE support for io-uring

[Jens Axboe]

On Wed, 20 Apr 2022 12:14:39 -0700, Stefan Roesch wrote:
> This adds the large CQE support for io-uring. Large CQE's are 16 bytes longer.
> To support the longer CQE's the allocation part is changed and when the CQE is
> accessed.
> 
> The allocation of the large CQE's is twice as big, so the allocation size is
> doubled. The ring size calculation needs to take this into account.
> 
> [...]

Applied, thanks!

[01/12] io_uring: support CQE32 in io_uring_cqe
        commit: be428af6b204c2b366dd8b838bea87d1d4d9f2bd
[02/12] io_uring: wire up inline completion path for CQE32
        commit: 8fc4fbc38db6538056498c88f606f958fbb24bfd
[03/12] io_uring: change ring size calculation for CQE32
        commit: d09d3b8f2986899ff8f535c91d95c137b03595ec
[04/12] io_uring: add CQE32 setup processing
        commit: a81124f0283879a7c5e77c0def9c725e84e79cb1
[05/12] io_uring: add CQE32 completion processing
        commit: c7050dfe60c484f9084e57c2b1c88b8ab1f8a06d
[06/12] io_uring: modify io_get_cqe for CQE32
        commit: f23855c3511dffa54069c9a0ed513b79bec39938
[07/12] io_uring: flush completions for CQE32
        commit: 8a5be11b11449a412ef89c46a05e9bbeeab6652d
[08/12] io_uring: overflow processing for CQE32
        commit: 2f1bbef557e9b174361ecd2f7c59b683bbca4464
[09/12] io_uring: add tracing for additional CQE32 fields
        commit: b4df41b44f8f358f86533148aa0e56b27bca47d6
[10/12] io_uring: support CQE32 in /proc info
        commit: 9d1b8d722dc06b9ab96db6e2bb967187c6185727
[11/12] io_uring: enable CQE32
        commit: cae6c1bdf9704dee2d3c7803c36ef73ada19e238
[12/12] io_uring: support CQE32 for nop operation
        commit: 460527265a0a6aa5107a7e4e4640f8d4b2088455

Best regards,
-- 
Jens Axboe

Re: [PATCH v2 00/12] add large CQE support for io-uring

[Pavel Begunkov]

On 4/20/22 23:51, Jens Axboe wrote:
> On Wed, 20 Apr 2022 12:14:39 -0700, Stefan Roesch wrote:
>> This adds the large CQE support for io-uring. Large CQE's are 16 bytes longer.
>> To support the longer CQE's the allocation part is changed and when the CQE is
>> accessed.
>>
>> The allocation of the large CQE's is twice as big, so the allocation size is
>> doubled. The ring size calculation needs to take this into account.

I'm missing something here, do we have a user for it apart
from no-op requests?


> Applied, thanks!
> 
> [01/12] io_uring: support CQE32 in io_uring_cqe
>          commit: be428af6b204c2b366dd8b838bea87d1d4d9f2bd
> [02/12] io_uring: wire up inline completion path for CQE32
>          commit: 8fc4fbc38db6538056498c88f606f958fbb24bfd
> [03/12] io_uring: change ring size calculation for CQE32
>          commit: d09d3b8f2986899ff8f535c91d95c137b03595ec
> [04/12] io_uring: add CQE32 setup processing
>          commit: a81124f0283879a7c5e77c0def9c725e84e79cb1
> [05/12] io_uring: add CQE32 completion processing
>          commit: c7050dfe60c484f9084e57c2b1c88b8ab1f8a06d
> [06/12] io_uring: modify io_get_cqe for CQE32
>          commit: f23855c3511dffa54069c9a0ed513b79bec39938
> [07/12] io_uring: flush completions for CQE32
>          commit: 8a5be11b11449a412ef89c46a05e9bbeeab6652d
> [08/12] io_uring: overflow processing for CQE32
>          commit: 2f1bbef557e9b174361ecd2f7c59b683bbca4464
> [09/12] io_uring: add tracing for additional CQE32 fields
>          commit: b4df41b44f8f358f86533148aa0e56b27bca47d6
> [10/12] io_uring: support CQE32 in /proc info
>          commit: 9d1b8d722dc06b9ab96db6e2bb967187c6185727
> [11/12] io_uring: enable CQE32
>          commit: cae6c1bdf9704dee2d3c7803c36ef73ada19e238
> [12/12] io_uring: support CQE32 for nop operation
>          commit: 460527265a0a6aa5107a7e4e4640f8d4b2088455
> 
> Best regards,

-- 
Pavel Begunkov

Re: [PATCH v2 00/12] add large CQE support for io-uring

[Stefan Roesch]

On 4/21/22 11:42 AM, Pavel Begunkov wrote:
> On 4/20/22 23:51, Jens Axboe wrote:
>> On Wed, 20 Apr 2022 12:14:39 -0700, Stefan Roesch wrote:
>>> This adds the large CQE support for io-uring. Large CQE's are 16 bytes longer.
>>> To support the longer CQE's the allocation part is changed and when the CQE is
>>> accessed.
>>>
>>> The allocation of the large CQE's is twice as big, so the allocation size is
>>> doubled. The ring size calculation needs to take this into account.
> 
> I'm missing something here, do we have a user for it apart
> from no-op requests?
> 

Pavel, what started this work is the patch series "io_uring passthru over nvme" from samsung.
(https://lore.kernel.org/io-uring/20220308152105.309618-1-joshi.k@samsung.com/)

They will use the large SQE and CQE support.

> 
>> Applied, thanks!
>>
>> [01/12] io_uring: support CQE32 in io_uring_cqe
>>          commit: be428af6b204c2b366dd8b838bea87d1d4d9f2bd
>> [02/12] io_uring: wire up inline completion path for CQE32
>>          commit: 8fc4fbc38db6538056498c88f606f958fbb24bfd
>> [03/12] io_uring: change ring size calculation for CQE32
>>          commit: d09d3b8f2986899ff8f535c91d95c137b03595ec
>> [04/12] io_uring: add CQE32 setup processing
>>          commit: a81124f0283879a7c5e77c0def9c725e84e79cb1
>> [05/12] io_uring: add CQE32 completion processing
>>          commit: c7050dfe60c484f9084e57c2b1c88b8ab1f8a06d
>> [06/12] io_uring: modify io_get_cqe for CQE32
>>          commit: f23855c3511dffa54069c9a0ed513b79bec39938
>> [07/12] io_uring: flush completions for CQE32
>>          commit: 8a5be11b11449a412ef89c46a05e9bbeeab6652d
>> [08/12] io_uring: overflow processing for CQE32
>>          commit: 2f1bbef557e9b174361ecd2f7c59b683bbca4464
>> [09/12] io_uring: add tracing for additional CQE32 fields
>>          commit: b4df41b44f8f358f86533148aa0e56b27bca47d6
>> [10/12] io_uring: support CQE32 in /proc info
>>          commit: 9d1b8d722dc06b9ab96db6e2bb967187c6185727
>> [11/12] io_uring: enable CQE32
>>          commit: cae6c1bdf9704dee2d3c7803c36ef73ada19e238
>> [12/12] io_uring: support CQE32 for nop operation
>>          commit: 460527265a0a6aa5107a7e4e4640f8d4b2088455
>>
>> Best regards,
> 

Re: [PATCH v2 00/12] add large CQE support for io-uring

[Jens Axboe]

On 4/21/22 12:49 PM, Stefan Roesch wrote:
> 
> 
> On 4/21/22 11:42 AM, Pavel Begunkov wrote:
>> On 4/20/22 23:51, Jens Axboe wrote:
>>> On Wed, 20 Apr 2022 12:14:39 -0700, Stefan Roesch wrote:
>>>> This adds the large CQE support for io-uring. Large CQE's are 16 bytes longer.
>>>> To support the longer CQE's the allocation part is changed and when the CQE is
>>>> accessed.
>>>>
>>>> The allocation of the large CQE's is twice as big, so the allocation size is
>>>> doubled. The ring size calculation needs to take this into account.
>>
>> I'm missing something here, do we have a user for it apart
>> from no-op requests?
>>
> 
> Pavel, what started this work is the patch series "io_uring passthru over nvme" from samsung.
> (https://lore.kernel.org/io-uring/20220308152105.309618-1-joshi.k@samsung.com/)
> 
> They will use the large SQE and CQE support.

Indeed - and as such, this will just be a base for that. Doesn't make
sense standalone, but with the passthrough support it does.

-- 
Jens Axboe

Re: [PATCH v2 00/12] add large CQE support for io-uring

[Pavel Begunkov]

On 4/21/22 19:49, Stefan Roesch wrote:
> On 4/21/22 11:42 AM, Pavel Begunkov wrote:
>> On 4/20/22 23:51, Jens Axboe wrote:
>>> On Wed, 20 Apr 2022 12:14:39 -0700, Stefan Roesch wrote:
>>>> This adds the large CQE support for io-uring. Large CQE's are 16 bytes longer.
>>>> To support the longer CQE's the allocation part is changed and when the CQE is
>>>> accessed.
>>>>
>>>> The allocation of the large CQE's is twice as big, so the allocation size is
>>>> doubled. The ring size calculation needs to take this into account.
>>
>> I'm missing something here, do we have a user for it apart
>> from no-op requests?
>>
> 
> Pavel, what started this work is the patch series "io_uring passthru over nvme" from samsung.
> (https://lore.kernel.org/io-uring/20220308152105.309618-1-joshi.k@samsung.com/)
> 
> They will use the large SQE and CQE support.

I see, thanks for clarifying. I saw it used in passthrough
patches, but it only got me more confused why it's applied
aforehand separately from the io_uring-cmd and passthrough


>>> Applied, thanks!
>>>
>>> [01/12] io_uring: support CQE32 in io_uring_cqe
>>>           commit: be428af6b204c2b366dd8b838bea87d1d4d9f2bd
>>> [02/12] io_uring: wire up inline completion path for CQE32
>>>           commit: 8fc4fbc38db6538056498c88f606f958fbb24bfd
>>> [03/12] io_uring: change ring size calculation for CQE32
>>>           commit: d09d3b8f2986899ff8f535c91d95c137b03595ec
>>> [04/12] io_uring: add CQE32 setup processing
>>>           commit: a81124f0283879a7c5e77c0def9c725e84e79cb1
>>> [05/12] io_uring: add CQE32 completion processing
>>>           commit: c7050dfe60c484f9084e57c2b1c88b8ab1f8a06d
>>> [06/12] io_uring: modify io_get_cqe for CQE32
>>>           commit: f23855c3511dffa54069c9a0ed513b79bec39938
>>> [07/12] io_uring: flush completions for CQE32
>>>           commit: 8a5be11b11449a412ef89c46a05e9bbeeab6652d
>>> [08/12] io_uring: overflow processing for CQE32
>>>           commit: 2f1bbef557e9b174361ecd2f7c59b683bbca4464
>>> [09/12] io_uring: add tracing for additional CQE32 fields
>>>           commit: b4df41b44f8f358f86533148aa0e56b27bca47d6
>>> [10/12] io_uring: support CQE32 in /proc info
>>>           commit: 9d1b8d722dc06b9ab96db6e2bb967187c6185727
>>> [11/12] io_uring: enable CQE32
>>>           commit: cae6c1bdf9704dee2d3c7803c36ef73ada19e238
>>> [12/12] io_uring: support CQE32 for nop operation
>>>           commit: 460527265a0a6aa5107a7e4e4640f8d4b2088455
>>>
>>> Best regards,
>>

-- 
Pavel Begunkov

Re: [PATCH v2 00/12] add large CQE support for io-uring

[Jens Axboe]

On 4/21/22 12:57 PM, Pavel Begunkov wrote:
> On 4/21/22 19:49, Stefan Roesch wrote:
>> On 4/21/22 11:42 AM, Pavel Begunkov wrote:
>>> On 4/20/22 23:51, Jens Axboe wrote:
>>>> On Wed, 20 Apr 2022 12:14:39 -0700, Stefan Roesch wrote:
>>>>> This adds the large CQE support for io-uring. Large CQE's are 16 bytes longer.
>>>>> To support the longer CQE's the allocation part is changed and when the CQE is
>>>>> accessed.
>>>>>
>>>>> The allocation of the large CQE's is twice as big, so the allocation size is
>>>>> doubled. The ring size calculation needs to take this into account.
>>>
>>> I'm missing something here, do we have a user for it apart
>>> from no-op requests?
>>>
>>
>> Pavel, what started this work is the patch series "io_uring passthru over nvme" from samsung.
>> (https://lore.kernel.org/io-uring/20220308152105.309618-1-joshi.k@samsung.com/)
>>
>> They will use the large SQE and CQE support.
> 
> I see, thanks for clarifying. I saw it used in passthrough
> patches, but it only got me more confused why it's applied
> aforehand separately from the io_uring-cmd and passthrough

It's just applied to a branch so the passthrough folks have something to
base on, io_uring-big-sqe. It's not queued for 5.19 or anything like
that yet.

-- 
Jens Axboe

Re: [PATCH v2 06/12] io_uring: modify io_get_cqe for CQE32

[Kanchan Joshi]

On Thu, Apr 21, 2022 at 3:54 PM Stefan Roesch <shr@fb.com> wrote:
>
> Modify accesses to the CQE array to take large CQE's into account. The
> index needs to be shifted by one for large CQE's.
>
> Signed-off-by: Stefan Roesch <shr@fb.com>
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>  fs/io_uring.c | 9 +++++++--
>  1 file changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index c93a9353c88d..bd352815b9e7 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -1909,8 +1909,12 @@ static noinline struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx)
>  {
>         struct io_rings *rings = ctx->rings;
>         unsigned int off = ctx->cached_cq_tail & (ctx->cq_entries - 1);
> +       unsigned int shift = 0;
>         unsigned int free, queued, len;
>
> +       if (ctx->flags & IORING_SETUP_CQE32)
> +               shift = 1;
> +
>         /* userspace may cheat modifying the tail, be safe and do min */
>         queued = min(__io_cqring_events(ctx), ctx->cq_entries);
>         free = ctx->cq_entries - queued;
> @@ -1922,12 +1926,13 @@ static noinline struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx)
>         ctx->cached_cq_tail++;
>         ctx->cqe_cached = &rings->cqes[off];
>         ctx->cqe_sentinel = ctx->cqe_cached + len;
> -       return ctx->cqe_cached++;
> +       ctx->cqe_cached++;
> +       return &rings->cqes[off << shift];
>  }
>
>  static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx)
>  {
> -       if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) {
> +       if (likely(ctx->cqe_cached < ctx->cqe_sentinel && !(ctx->flags & IORING_SETUP_CQE32))) {
>                 ctx->cached_cq_tail++;
>                 return ctx->cqe_cached++;
>         }

This excludes CQE-caching for 32b CQEs.
How about something like below to have that enabled (adding
io_get_cqe32 for the new ring) -

+static noinline struct io_uring_cqe *__io_get_cqe32(struct io_ring_ctx *ctx)
+{
+       struct io_rings *rings = ctx->rings;
+       unsigned int off = ctx->cached_cq_tail & (ctx->cq_entries - 1);
+       unsigned int free, queued, len;
+
+       /* userspace may cheat modifying the tail, be safe and do min */
+       queued = min(__io_cqring_events(ctx), ctx->cq_entries);
+       free = ctx->cq_entries - queued;
+       /* we need a contiguous range, limit based on the current
array offset */
+       len = min(free, ctx->cq_entries - off);
+       if (!len)
+               return NULL;
+
+       ctx->cached_cq_tail++;
+       /* double increment for 32 CQEs */
+       ctx->cqe_cached = &rings->cqes[off << 1];
+       ctx->cqe_sentinel = ctx->cqe_cached + (len << 1);
+       return ctx->cqe_cached;
+}
+
+static inline struct io_uring_cqe *io_get_cqe32(struct io_ring_ctx *ctx)
+{
+       struct io_uring_cqe *cqe32;
+       if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) {
+               ctx->cached_cq_tail++;
+               cqe32 = ctx->cqe_cached;
+       } else
+               cqe32 = __io_get_cqe32(ctx);
+       /* double increment for 32b CQE*/
+       ctx->cqe_cached += 2;
+       return cqe32;
+}

Re: [PATCH v2 05/12] io_uring: add CQE32 completion processing

[Kanchan Joshi]

On Thu, Apr 21, 2022 at 10:44 AM Stefan Roesch <shr@fb.com> wrote:
>
> This adds the completion processing for the large CQE's and makes sure
> that the extra1 and extra2 fields are passed through.
>
> Co-developed-by: Jens Axboe <axboe@kernel.dk>
> Signed-off-by: Stefan Roesch <shr@fb.com>
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>  fs/io_uring.c | 55 +++++++++++++++++++++++++++++++++++++++++++--------
>  1 file changed, 47 insertions(+), 8 deletions(-)
>
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index abbd2efbe255..c93a9353c88d 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -2247,18 +2247,15 @@ static noinline bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data,
>         return __io_fill_cqe(ctx, user_data, res, cflags);
>  }
>
> -static void __io_req_complete_post(struct io_kiocb *req, s32 res,
> -                                  u32 cflags)
> +static void __io_req_complete_put(struct io_kiocb *req)
>  {
> -       struct io_ring_ctx *ctx = req->ctx;
> -
> -       if (!(req->flags & REQ_F_CQE_SKIP))
> -               __io_fill_cqe_req(req, res, cflags);
>         /*
>          * If we're the last reference to this request, add to our locked
>          * free_list cache.
>          */
>         if (req_ref_put_and_test(req)) {
> +               struct io_ring_ctx *ctx = req->ctx;
> +
>                 if (req->flags & IO_REQ_LINK_FLAGS) {
>                         if (req->flags & IO_DISARM_MASK)
>                                 io_disarm_next(req);
> @@ -2281,8 +2278,23 @@ static void __io_req_complete_post(struct io_kiocb *req, s32 res,
>         }
>  }
>
> -static void io_req_complete_post(struct io_kiocb *req, s32 res,
> -                                u32 cflags)
> +static void __io_req_complete_post(struct io_kiocb *req, s32 res,
> +                                  u32 cflags)
> +{
> +       if (!(req->flags & REQ_F_CQE_SKIP))
> +               __io_fill_cqe_req(req, res, cflags);
> +       __io_req_complete_put(req);
> +}
> +
> +static void __io_req_complete_post32(struct io_kiocb *req, s32 res,
> +                                  u32 cflags, u64 extra1, u64 extra2)
> +{
> +       if (!(req->flags & REQ_F_CQE_SKIP))
> +               __io_fill_cqe32_req(req, res, cflags, extra1, extra2);
> +       __io_req_complete_put(req);
> +}
> +
> +static void io_req_complete_post(struct io_kiocb *req, s32 res, u32 cflags)
>  {
>         struct io_ring_ctx *ctx = req->ctx;
>
> @@ -2293,6 +2305,18 @@ static void io_req_complete_post(struct io_kiocb *req, s32 res,
>         io_cqring_ev_posted(ctx);
>  }
>
> +static void io_req_complete_post32(struct io_kiocb *req, s32 res,
> +                                  u32 cflags, u64 extra1, u64 extra2)
> +{
> +       struct io_ring_ctx *ctx = req->ctx;
> +
> +       spin_lock(&ctx->completion_lock);
> +       __io_req_complete_post32(req, res, cflags, extra1, extra2);
> +       io_commit_cqring(ctx);
> +       spin_unlock(&ctx->completion_lock);
> +       io_cqring_ev_posted(ctx);
> +}
> +
>  static inline void io_req_complete_state(struct io_kiocb *req, s32 res,
>                                          u32 cflags)
>  {
> @@ -2310,6 +2334,21 @@ static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
>                 io_req_complete_post(req, res, cflags);
>  }
>
> +static inline void __io_req_complete32(struct io_kiocb *req,
> +                                      unsigned int issue_flags, s32 res,
> +                                      u32 cflags, u64 extra1, u64 extra2)
> +{
> +       if (issue_flags & IO_URING_F_COMPLETE_DEFER) {
> +               req->cqe.res = res;
> +               req->cqe.flags = cflags;
> +               req->extra1 = extra1;
> +               req->extra2 = extra2;
> +               req->flags |= REQ_F_COMPLETE_INLINE;

nit: we can use the existing helper (io_req_complete_state) to
populate these fields rather than open-coding.

Re: [PATCH v2 08/12] io_uring: overflow processing for CQE32

[Kanchan Joshi]

On Thu, Apr 21, 2022 at 1:37 PM Stefan Roesch <shr@fb.com> wrote:
>
> This adds the overflow processing for large CQE's.
>
> This adds two parameters to the io_cqring_event_overflow function and
> uses these fields to initialize the large CQE fields.
>
> Allocate enough space for large CQE's in the overflow structue. If no
> large CQE's are used, the size of the allocation is unchanged.
>
> The cqe field can have a different size depending if its a large
> CQE or not. To be able to allocate different sizes, the two fields
> in the structure are re-ordered.
>
> Co-developed-by: Jens Axboe <axboe@kernel.dk>
> Signed-off-by: Stefan Roesch <shr@fb.com>
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>  fs/io_uring.c | 26 +++++++++++++++++++-------
>  1 file changed, 19 insertions(+), 7 deletions(-)
>
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index ff6229b6df16..50efced63ec9 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -220,8 +220,8 @@ struct io_mapped_ubuf {
>  struct io_ring_ctx;
>
>  struct io_overflow_cqe {
> -       struct io_uring_cqe cqe;
>         struct list_head list;
> +       struct io_uring_cqe cqe;
>  };
>
>  struct io_fixed_file {
> @@ -2016,13 +2016,17 @@ static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
>         while (!list_empty(&ctx->cq_overflow_list)) {
>                 struct io_uring_cqe *cqe = io_get_cqe(ctx);
>                 struct io_overflow_cqe *ocqe;
> +               size_t cqe_size = sizeof(struct io_uring_cqe);
> +
> +               if (ctx->flags & IORING_SETUP_CQE32)
> +                       cqe_size <<= 1;
>
>                 if (!cqe && !force)
>                         break;
>                 ocqe = list_first_entry(&ctx->cq_overflow_list,
>                                         struct io_overflow_cqe, list);
>                 if (cqe)
> -                       memcpy(cqe, &ocqe->cqe, sizeof(*cqe));
> +                       memcpy(cqe, &ocqe->cqe, cqe_size);
>                 else
>                         io_account_cq_overflow(ctx);
>
> @@ -2111,11 +2115,15 @@ static __cold void io_uring_drop_tctx_refs(struct task_struct *task)
>  }
>
>  static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data,
> -                                    s32 res, u32 cflags)
> +                                    s32 res, u32 cflags, u64 extra1, u64 extra2)
>  {
>         struct io_overflow_cqe *ocqe;
> +       size_t ocq_size = sizeof(struct io_overflow_cqe);
>
> -       ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT);
> +       if (ctx->flags & IORING_SETUP_CQE32)

This can go inside in a bool variable, as this check is repeated in
this function.

Re: [PATCH v2 00/12] add large CQE support for io-uring

[Kanchan Joshi]

[-- Attachment #1: Type: text/plain, Size: 1641 bytes --]

On Thu, Apr 21, 2022 at 12:59:42PM -0600, Jens Axboe wrote:
>On 4/21/22 12:57 PM, Pavel Begunkov wrote:
>> On 4/21/22 19:49, Stefan Roesch wrote:
>>> On 4/21/22 11:42 AM, Pavel Begunkov wrote:
>>>> On 4/20/22 23:51, Jens Axboe wrote:
>>>>> On Wed, 20 Apr 2022 12:14:39 -0700, Stefan Roesch wrote:
>>>>>> This adds the large CQE support for io-uring. Large CQE's are 16 bytes longer.
>>>>>> To support the longer CQE's the allocation part is changed and when the CQE is
>>>>>> accessed.
>>>>>>
>>>>>> The allocation of the large CQE's is twice as big, so the allocation size is
>>>>>> doubled. The ring size calculation needs to take this into account.
>>>>
>>>> I'm missing something here, do we have a user for it apart
>>>> from no-op requests?
>>>>
>>>
>>> Pavel, what started this work is the patch series "io_uring passthru over nvme" from samsung.
>>> (https://lore.kernel.org/io-uring/20220308152105.309618-1-joshi.k@samsung.com/)
>>>
>>> They will use the large SQE and CQE support.
>>
>> I see, thanks for clarifying. I saw it used in passthrough
>> patches, but it only got me more confused why it's applied
>> aforehand separately from the io_uring-cmd and passthrough
>
>It's just applied to a branch so the passthrough folks have something to
>base on, io_uring-big-sqe. It's not queued for 5.19 or anything like
>that yet.
>
Thanks for putting this up.
I am bit confused whether these (big-cqe) and big-sqe patches should
continue be sent (to nvme list too) as part of next
uring-cmd/passthrough series?

And does it make sense to squash somes patches of this series; at
high-level there is 32b-CQE support, and no-op support.

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: [PATCH v2 00/12] add large CQE support for io-uring

[Kanchan Joshi]

[-- Attachment #1: Type: text/plain, Size: 1821 bytes --]

On Fri, Apr 22, 2022 at 08:39:18AM +0530, Kanchan Joshi wrote:
>On Thu, Apr 21, 2022 at 12:59:42PM -0600, Jens Axboe wrote:
>>On 4/21/22 12:57 PM, Pavel Begunkov wrote:
>>>On 4/21/22 19:49, Stefan Roesch wrote:
>>>>On 4/21/22 11:42 AM, Pavel Begunkov wrote:
>>>>>On 4/20/22 23:51, Jens Axboe wrote:
>>>>>>On Wed, 20 Apr 2022 12:14:39 -0700, Stefan Roesch wrote:
>>>>>>>This adds the large CQE support for io-uring. Large CQE's are 16 bytes longer.
>>>>>>>To support the longer CQE's the allocation part is changed and when the CQE is
>>>>>>>accessed.
>>>>>>>
>>>>>>>The allocation of the large CQE's is twice as big, so the allocation size is
>>>>>>>doubled. The ring size calculation needs to take this into account.
>>>>>
>>>>>I'm missing something here, do we have a user for it apart
>>>>>from no-op requests?
>>>>>
>>>>
>>>>Pavel, what started this work is the patch series "io_uring passthru over nvme" from samsung.
>>>>(https://lore.kernel.org/io-uring/20220308152105.309618-1-joshi.k@samsung.com/)
>>>>
>>>>They will use the large SQE and CQE support.
>>>
>>>I see, thanks for clarifying. I saw it used in passthrough
>>>patches, but it only got me more confused why it's applied
>>>aforehand separately from the io_uring-cmd and passthrough
>>
>>It's just applied to a branch so the passthrough folks have something to
>>base on, io_uring-big-sqe. It's not queued for 5.19 or anything like
>>that yet.
>>
>Thanks for putting this up.
>I am bit confused whether these (big-cqe) and big-sqe patches should
>continue be sent (to nvme list too) as part of next
>uring-cmd/passthrough series?
>
>And does it make sense to squash somes patches of this series; at
>high-level there is 32b-CQE support, and no-op support.

Maybe as part of v3, as there seems some scope for that (I made comments
at respective places).

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]

Re: [PATCH v2 00/12] add large CQE support for io-uring

[Stefan Roesch]

On 4/21/22 8:09 PM, Kanchan Joshi wrote:
> On Thu, Apr 21, 2022 at 12:59:42PM -0600, Jens Axboe wrote:
>> On 4/21/22 12:57 PM, Pavel Begunkov wrote:
>>> On 4/21/22 19:49, Stefan Roesch wrote:
>>>> On 4/21/22 11:42 AM, Pavel Begunkov wrote:
>>>>> On 4/20/22 23:51, Jens Axboe wrote:
>>>>>> On Wed, 20 Apr 2022 12:14:39 -0700, Stefan Roesch wrote:
>>>>>>> This adds the large CQE support for io-uring. Large CQE's are 16 bytes longer.
>>>>>>> To support the longer CQE's the allocation part is changed and when the CQE is
>>>>>>> accessed.
>>>>>>>
>>>>>>> The allocation of the large CQE's is twice as big, so the allocation size is
>>>>>>> doubled. The ring size calculation needs to take this into account.
>>>>>
>>>>> I'm missing something here, do we have a user for it apart
>>>>> from no-op requests?
>>>>>
>>>>
>>>> Pavel, what started this work is the patch series "io_uring passthru over nvme" from samsung.
>>>> (https://lore.kernel.org/io-uring/20220308152105.309618-1-joshi.k@samsung.com/)
>>>>
>>>> They will use the large SQE and CQE support.
>>>
>>> I see, thanks for clarifying. I saw it used in passthrough
>>> patches, but it only got me more confused why it's applied
>>> aforehand separately from the io_uring-cmd and passthrough
>>
>> It's just applied to a branch so the passthrough folks have something to
>> base on, io_uring-big-sqe. It's not queued for 5.19 or anything like
>> that yet.
>>
> Thanks for putting this up.
> I am bit confused whether these (big-cqe) and big-sqe patches should
> continue be sent (to nvme list too) as part of next
> uring-cmd/passthrough series?
> 

I'll sent version 3 also to the nvme list.

> And does it make sense to squash somes patches of this series; at
> high-level there is 32b-CQE support, and no-op support.
> 

Re: [PATCH v2 08/12] io_uring: overflow processing for CQE32

[Stefan Roesch]

On 4/21/22 7:15 PM, Kanchan Joshi wrote:
> On Thu, Apr 21, 2022 at 1:37 PM Stefan Roesch <shr@fb.com> wrote:
>>
>> This adds the overflow processing for large CQE's.
>>
>> This adds two parameters to the io_cqring_event_overflow function and
>> uses these fields to initialize the large CQE fields.
>>
>> Allocate enough space for large CQE's in the overflow structue. If no
>> large CQE's are used, the size of the allocation is unchanged.
>>
>> The cqe field can have a different size depending if its a large
>> CQE or not. To be able to allocate different sizes, the two fields
>> in the structure are re-ordered.
>>
>> Co-developed-by: Jens Axboe <axboe@kernel.dk>
>> Signed-off-by: Stefan Roesch <shr@fb.com>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>> ---
>>  fs/io_uring.c | 26 +++++++++++++++++++-------
>>  1 file changed, 19 insertions(+), 7 deletions(-)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index ff6229b6df16..50efced63ec9 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -220,8 +220,8 @@ struct io_mapped_ubuf {
>>  struct io_ring_ctx;
>>
>>  struct io_overflow_cqe {
>> -       struct io_uring_cqe cqe;
>>         struct list_head list;
>> +       struct io_uring_cqe cqe;
>>  };
>>
>>  struct io_fixed_file {
>> @@ -2016,13 +2016,17 @@ static bool __io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force)
>>         while (!list_empty(&ctx->cq_overflow_list)) {
>>                 struct io_uring_cqe *cqe = io_get_cqe(ctx);
>>                 struct io_overflow_cqe *ocqe;
>> +               size_t cqe_size = sizeof(struct io_uring_cqe);
>> +
>> +               if (ctx->flags & IORING_SETUP_CQE32)
>> +                       cqe_size <<= 1;
>>
>>                 if (!cqe && !force)
>>                         break;
>>                 ocqe = list_first_entry(&ctx->cq_overflow_list,
>>                                         struct io_overflow_cqe, list);
>>                 if (cqe)
>> -                       memcpy(cqe, &ocqe->cqe, sizeof(*cqe));
>> +                       memcpy(cqe, &ocqe->cqe, cqe_size);
>>                 else
>>                         io_account_cq_overflow(ctx);
>>
>> @@ -2111,11 +2115,15 @@ static __cold void io_uring_drop_tctx_refs(struct task_struct *task)
>>  }
>>
>>  static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data,
>> -                                    s32 res, u32 cflags)
>> +                                    s32 res, u32 cflags, u64 extra1, u64 extra2)
>>  {
>>         struct io_overflow_cqe *ocqe;
>> +       size_t ocq_size = sizeof(struct io_overflow_cqe);
>>
>> -       ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT);
>> +       if (ctx->flags & IORING_SETUP_CQE32)
> 
> This can go inside in a bool variable, as this check is repeated in
> this function.

V3 will have this change.

Re: [PATCH v2 05/12] io_uring: add CQE32 completion processing

[Stefan Roesch]

On 4/21/22 6:34 PM, Kanchan Joshi wrote:
> On Thu, Apr 21, 2022 at 10:44 AM Stefan Roesch <shr@fb.com> wrote:
>>
>> This adds the completion processing for the large CQE's and makes sure
>> that the extra1 and extra2 fields are passed through.
>>
>> Co-developed-by: Jens Axboe <axboe@kernel.dk>
>> Signed-off-by: Stefan Roesch <shr@fb.com>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>> ---
>>  fs/io_uring.c | 55 +++++++++++++++++++++++++++++++++++++++++++--------
>>  1 file changed, 47 insertions(+), 8 deletions(-)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index abbd2efbe255..c93a9353c88d 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -2247,18 +2247,15 @@ static noinline bool io_fill_cqe_aux(struct io_ring_ctx *ctx, u64 user_data,
>>         return __io_fill_cqe(ctx, user_data, res, cflags);
>>  }
>>
>> -static void __io_req_complete_post(struct io_kiocb *req, s32 res,
>> -                                  u32 cflags)
>> +static void __io_req_complete_put(struct io_kiocb *req)
>>  {
>> -       struct io_ring_ctx *ctx = req->ctx;
>> -
>> -       if (!(req->flags & REQ_F_CQE_SKIP))
>> -               __io_fill_cqe_req(req, res, cflags);
>>         /*
>>          * If we're the last reference to this request, add to our locked
>>          * free_list cache.
>>          */
>>         if (req_ref_put_and_test(req)) {
>> +               struct io_ring_ctx *ctx = req->ctx;
>> +
>>                 if (req->flags & IO_REQ_LINK_FLAGS) {
>>                         if (req->flags & IO_DISARM_MASK)
>>                                 io_disarm_next(req);
>> @@ -2281,8 +2278,23 @@ static void __io_req_complete_post(struct io_kiocb *req, s32 res,
>>         }
>>  }
>>
>> -static void io_req_complete_post(struct io_kiocb *req, s32 res,
>> -                                u32 cflags)
>> +static void __io_req_complete_post(struct io_kiocb *req, s32 res,
>> +                                  u32 cflags)
>> +{
>> +       if (!(req->flags & REQ_F_CQE_SKIP))
>> +               __io_fill_cqe_req(req, res, cflags);
>> +       __io_req_complete_put(req);
>> +}
>> +
>> +static void __io_req_complete_post32(struct io_kiocb *req, s32 res,
>> +                                  u32 cflags, u64 extra1, u64 extra2)
>> +{
>> +       if (!(req->flags & REQ_F_CQE_SKIP))
>> +               __io_fill_cqe32_req(req, res, cflags, extra1, extra2);
>> +       __io_req_complete_put(req);
>> +}
>> +
>> +static void io_req_complete_post(struct io_kiocb *req, s32 res, u32 cflags)
>>  {
>>         struct io_ring_ctx *ctx = req->ctx;
>>
>> @@ -2293,6 +2305,18 @@ static void io_req_complete_post(struct io_kiocb *req, s32 res,
>>         io_cqring_ev_posted(ctx);
>>  }
>>
>> +static void io_req_complete_post32(struct io_kiocb *req, s32 res,
>> +                                  u32 cflags, u64 extra1, u64 extra2)
>> +{
>> +       struct io_ring_ctx *ctx = req->ctx;
>> +
>> +       spin_lock(&ctx->completion_lock);
>> +       __io_req_complete_post32(req, res, cflags, extra1, extra2);
>> +       io_commit_cqring(ctx);
>> +       spin_unlock(&ctx->completion_lock);
>> +       io_cqring_ev_posted(ctx);
>> +}
>> +
>>  static inline void io_req_complete_state(struct io_kiocb *req, s32 res,
>>                                          u32 cflags)
>>  {
>> @@ -2310,6 +2334,21 @@ static inline void __io_req_complete(struct io_kiocb *req, unsigned issue_flags,
>>                 io_req_complete_post(req, res, cflags);
>>  }
>>
>> +static inline void __io_req_complete32(struct io_kiocb *req,
>> +                                      unsigned int issue_flags, s32 res,
>> +                                      u32 cflags, u64 extra1, u64 extra2)
>> +{
>> +       if (issue_flags & IO_URING_F_COMPLETE_DEFER) {
>> +               req->cqe.res = res;
>> +               req->cqe.flags = cflags;
>> +               req->extra1 = extra1;
>> +               req->extra2 = extra2;
>> +               req->flags |= REQ_F_COMPLETE_INLINE;
> 
> nit: we can use the existing helper (io_req_complete_state) to
> populate these fields rather than open-coding.

V3 will have that change.

Re: [PATCH v2 06/12] io_uring: modify io_get_cqe for CQE32

[Stefan Roesch]

On 4/21/22 6:25 PM, Kanchan Joshi wrote:
> On Thu, Apr 21, 2022 at 3:54 PM Stefan Roesch <shr@fb.com> wrote:
>>
>> Modify accesses to the CQE array to take large CQE's into account. The
>> index needs to be shifted by one for large CQE's.
>>
>> Signed-off-by: Stefan Roesch <shr@fb.com>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>> ---
>>  fs/io_uring.c | 9 +++++++--
>>  1 file changed, 7 insertions(+), 2 deletions(-)
>>
>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>> index c93a9353c88d..bd352815b9e7 100644
>> --- a/fs/io_uring.c
>> +++ b/fs/io_uring.c
>> @@ -1909,8 +1909,12 @@ static noinline struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx)
>>  {
>>         struct io_rings *rings = ctx->rings;
>>         unsigned int off = ctx->cached_cq_tail & (ctx->cq_entries - 1);
>> +       unsigned int shift = 0;
>>         unsigned int free, queued, len;
>>
>> +       if (ctx->flags & IORING_SETUP_CQE32)
>> +               shift = 1;
>> +
>>         /* userspace may cheat modifying the tail, be safe and do min */
>>         queued = min(__io_cqring_events(ctx), ctx->cq_entries);
>>         free = ctx->cq_entries - queued;
>> @@ -1922,12 +1926,13 @@ static noinline struct io_uring_cqe *__io_get_cqe(struct io_ring_ctx *ctx)
>>         ctx->cached_cq_tail++;
>>         ctx->cqe_cached = &rings->cqes[off];
>>         ctx->cqe_sentinel = ctx->cqe_cached + len;
>> -       return ctx->cqe_cached++;
>> +       ctx->cqe_cached++;
>> +       return &rings->cqes[off << shift];
>>  }
>>
>>  static inline struct io_uring_cqe *io_get_cqe(struct io_ring_ctx *ctx)
>>  {
>> -       if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) {
>> +       if (likely(ctx->cqe_cached < ctx->cqe_sentinel && !(ctx->flags & IORING_SETUP_CQE32))) {
>>                 ctx->cached_cq_tail++;
>>                 return ctx->cqe_cached++;
>>         }
> 
> This excludes CQE-caching for 32b CQEs.
> How about something like below to have that enabled (adding
> io_get_cqe32 for the new ring) -
> 

What you describe below I tried to avoid: keep the current indexes and pointers
as they are and only when we access an element calculate the correct offset into the
cqe array.

I'll add caching support for V3 in a slightly different way.

> +static noinline struct io_uring_cqe *__io_get_cqe32(struct io_ring_ctx *ctx)
> +{
> +       struct io_rings *rings = ctx->rings;
> +       unsigned int off = ctx->cached_cq_tail & (ctx->cq_entries - 1);
> +       unsigned int free, queued, len;
> +
> +       /* userspace may cheat modifying the tail, be safe and do min */
> +       queued = min(__io_cqring_events(ctx), ctx->cq_entries);
> +       free = ctx->cq_entries - queued;
> +       /* we need a contiguous range, limit based on the current
> array offset */
> +       len = min(free, ctx->cq_entries - off);
> +       if (!len)
> +               return NULL;
> +
> +       ctx->cached_cq_tail++;
> +       /* double increment for 32 CQEs */
> +       ctx->cqe_cached = &rings->cqes[off << 1];
> +       ctx->cqe_sentinel = ctx->cqe_cached + (len << 1);
> +       return ctx->cqe_cached;
> +}
> +
> +static inline struct io_uring_cqe *io_get_cqe32(struct io_ring_ctx *ctx)
> +{
> +       struct io_uring_cqe *cqe32;
> +       if (likely(ctx->cqe_cached < ctx->cqe_sentinel)) {
> +               ctx->cached_cq_tail++;
> +               cqe32 = ctx->cqe_cached;
> +       } else
> +               cqe32 = __io_get_cqe32(ctx);
> +       /* double increment for 32b CQE*/
> +       ctx->cqe_cached += 2;
> +       return cqe32;
> +}

Re: [PATCH v2 08/12] io_uring: overflow processing for CQE32

[Kanchan Joshi]

[-- Attachment #1: Type: text/plain, Size: 835 bytes --]

On Fri, Apr 22, 2022 at 02:27:01PM -0700, Stefan Roesch wrote:
>
>
>On 4/21/22 7:15 PM, Kanchan Joshi wrote:
>> On Thu, Apr 21, 2022 at 1:37 PM Stefan Roesch <shr@fb.com> wrote:
<snip>
>>>  static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data,
>>> -                                    s32 res, u32 cflags)
>>> +                                    s32 res, u32 cflags, u64 extra1, u64 extra2)
>>>  {
>>>         struct io_overflow_cqe *ocqe;
>>> +       size_t ocq_size = sizeof(struct io_overflow_cqe);
>>>
>>> -       ocqe = kmalloc(sizeof(*ocqe), GFP_ATOMIC | __GFP_ACCOUNT);
>>> +       if (ctx->flags & IORING_SETUP_CQE32)
>>
>> This can go inside in a bool variable, as this check is repeated in
>> this function.
>
>V3 will have this change.
While you are at it, good to have this changed in patch 10 too.

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]