Skip to contentSkip to conversation

io_uring: extend io_uring_sqe flags bits

RFC PATCH Superseded Jens Axboe 3 participants 13 messages Oct 31, 2024 – Nov 1, 2024 io_uring

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Jens Axboe

author maintainer
#0 lore raw

RFC PATCH 1/1 io_uring: extend io_uring_sqe flags bits

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that would have to be done via a uring cmd flag.

---

Was toying with this idea to allow for some more flags, I just don't like grabbing the last flag and punting the problem both to the future and to "somebody elses problem". Here's one way we could do it, without rewriting the entire sqe into a v2. Which does need to happen at some point, but preferably without pressing issues around.

I don't _hate_ it, there's really not a great way to do this. And I do think personality is the least used of all the things, and probably will never get used with uring_cmd. But if it had to work for that, then there are certainly ways to pass in that info. Not that we ever would...

include/linux/io_uring_types.h +11 -2
index 77fd508d043a..8a45bf6a68ca 100644
@@ -433,6 +433,7 @@ struct io_tw_state {
 };
 
 enum {
+	/* 8 bits of sqe->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,
@@ -440,9 +441,13 @@ enum {
 	REQ_F_FORCE_ASYNC_BIT	= IOSQE_ASYNC_BIT,
 	REQ_F_BUFFER_SELECT_BIT	= IOSQE_BUFFER_SELECT_BIT,
 	REQ_F_CQE_SKIP_BIT	= IOSQE_CQE_SKIP_SUCCESS_BIT,
+	REQ_F_FLAGS2_BIT	= IOSQE_FLAGS2_BIT,
 
-	/* first byte is taken by user flags, shift it to not overlap */
-	REQ_F_FAIL_BIT		= 8,
+	/* 16 bits of sqe->flags2 */
+	REQ_F_PERSONALITY_BIT	= IOSQE2_PERSONALITY_BIT + 8,
+
+	/* first byte taken by sqe->flags, next 2 by sqe->flags2 */
+	REQ_F_FAIL_BIT		= 24,
 	REQ_F_INFLIGHT_BIT,
 	REQ_F_CUR_POS_BIT,
 	REQ_F_NOWAIT_BIT,
@@ -492,6 +497,10 @@ enum {
 	REQ_F_BUFFER_SELECT	= IO_REQ_FLAG(REQ_F_BUFFER_SELECT_BIT),
 	/* IOSQE_CQE_SKIP_SUCCESS */
 	REQ_F_CQE_SKIP		= IO_REQ_FLAG(REQ_F_CQE_SKIP_BIT),
+	/* ->flags2 is valid */
+	REQ_F_FLAGS2		= IO_REQ_FLAG(REQ_F_FLAGS2_BIT),
+
+	REQ_F_PERSONALITY	= IO_REQ_FLAG(REQ_F_PERSONALITY_BIT),
 
 	/* fail rest of links */
 	REQ_F_FAIL		= IO_REQ_FLAG(REQ_F_FAIL_BIT),
include/uapi/linux/io_uring.h +30 -7
index ce58c4590de6..c7c3ba69ffdd 100644
@@ -82,8 +82,12 @@ struct io_uring_sqe {
 		/* for grouped buffer selection */
 		__u16	buf_group;
 	} __attribute__((packed));
-	/* personality to use, if used */
-	__u16	personality;
+	union {
+		/* personality to use, if used */
+		__u16	personality;
+		/* 2nd set of flags, can't be used with personality */
+		__u16	flags2;
+	};
 	union {
 		__s32	splice_fd_in;
 		__u32	file_index;
@@ -99,11 +103,17 @@ struct io_uring_sqe {
 			__u64	__pad2[1];
 		};
 		__u64	optval;
-		/*
-		 * If the ring is initialized with IORING_SETUP_SQE128, then
-		 * this field is used for 80 bytes of arbitrary command data
-		 */
-		__u8	cmd[0];
+		struct {
+			/*
+			 * If the ring is initialized with IORING_SETUP_SQE128,
+			 * then this field is used for 80 bytes of arbitrary
+			 * command data
+			 */
+			__u8	cmd[0];
+
+			/* personality to use, if IOSQE2_PERSONALITY set */
+			__u16	personality2;
+		};
 	};
 };
 
@@ -124,6 +134,11 @@ enum io_uring_sqe_flags_bit {
 	IOSQE_ASYNC_BIT,
 	IOSQE_BUFFER_SELECT_BIT,
 	IOSQE_CQE_SKIP_SUCCESS_BIT,
+	IOSQE_FLAGS2_BIT,
+};
+
+enum io_uring_sqe_flags2_bit {
+	IOSQE2_PERSONALITY_BIT,
 };
 
 /*
@@ -143,6 +158,14 @@ enum io_uring_sqe_flags_bit {
 #define IOSQE_BUFFER_SELECT	(1U << IOSQE_BUFFER_SELECT_BIT)
 /* don't post CQE if request succeeded */
 #define IOSQE_CQE_SKIP_SUCCESS	(1U << IOSQE_CQE_SKIP_SUCCESS_BIT)
+/* ->flags2 is valid */
+#define IOSQE_FLAGS2		(1U << IOSQE_FLAGS2_BIT)
+
+/*
+ * sqe->flags2
+ */
+ /* if set, sqe->personality2 contains personality */
+#define IOSQE2_PERSONALITY	(1U << IOSQE2_PERSONALITY_BIT)
 
 /*
  * io_uring_setup() flags
io_uring/io_uring.c +10 -3
index 1149fba20503..c2bbadd5640d 100644
@@ -109,7 +109,8 @@
 			  IOSQE_IO_HARDLINK | IOSQE_ASYNC)
 
 #define SQE_VALID_FLAGS	(SQE_COMMON_FLAGS | IOSQE_BUFFER_SELECT | \
-			IOSQE_IO_DRAIN | IOSQE_CQE_SKIP_SUCCESS)
+			IOSQE_IO_DRAIN | IOSQE_CQE_SKIP_SUCCESS | \
+			IOSQE_FLAGS2 | IOSQE2_PERSONALITY)
 
 #define IO_REQ_CLEAN_FLAGS (REQ_F_BUFFER_SELECTED | REQ_F_NEED_CLEANUP | \
 				REQ_F_POLLED | REQ_F_INFLIGHT | REQ_F_CREDS | \
@@ -2032,6 +2033,8 @@ static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
 	req->opcode = opcode = READ_ONCE(sqe->opcode);
 	/* same numerical values with corresponding REQ_F_*, safe to copy */
 	sqe_flags = READ_ONCE(sqe->flags);
+	if (sqe_flags & REQ_F_FLAGS2)
+		sqe_flags |= (__u32) READ_ONCE(sqe->flags2) << 8;
 	req->flags = (__force io_req_flags_t) sqe_flags;
 	req->cqe.user_data = READ_ONCE(sqe->user_data);
 	req->file = NULL;
@@ -2095,8 +2098,12 @@ static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req,
 		}
 	}
 
-	personality = READ_ONCE(sqe->personality);
-	if (personality) {
+	personality = 0;
+	if (req->flags & REQ_F_PERSONALITY)
+		personality = READ_ONCE(sqe->personality2);
+	else if (!(req->flags & REQ_F_FLAGS2))
+		personality = READ_ONCE(sqe->personality);
+	if (unlikely(personality)) {
 		int ret;
 
 		req->creds = xa_load(&ctx->personalities, personality);
io_uring/uring_cmd.c +1 -1
index 535909a38e76..ee04e0c48672 100644
@@ -200,7 +200,7 @@ int io_uring_cmd_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 {
 	struct io_uring_cmd *ioucmd = io_kiocb_to_cmd(req, struct io_uring_cmd);
 
-	if (sqe->__pad1)
+	if (sqe->__pad1 || req->flags & REQ_F_PERSONALITY)
 		return -EINVAL;
 
 	ioucmd->flags = READ_ONCE(sqe->uring_cmd_flags);
Signature · 1 line
Jens Axboe
#1 lore raw
Quoted 11 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that

The space is the 1st short for uring_cmd, instead of SQE128 only.

Also it is overlapped with ->optval and ->addr3, so just wondering why not use ->__pad2?

Another ways is to use __pad2 for sqe2_flags for non-uring_cmd, and for uring_cmd, use its top 16 as sqe2_flags, this way does work, but it is just a bit ugly to use.

Thanks,
Ming

#2 lore raw
Quoted 13 lines · Ming Lei
Quoted 11 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that

Quoted 8 lines

The space is the 1st short for uring_cmd, instead of SQE128 only.

Also it is overlapped with ->optval and ->addr3, so just wondering why not use ->__pad2?

Another ways is to use __pad2 for sqe2_flags for non-uring_cmd, and for uring_cmd, use its top 16 as sqe2_flags, this way does work, but it is just a bit ugly to use.

Also IOSQE2_PERSONALITY doesn't have to be per-SQE, and it can be one feature of IORING_FEAT_IOSQE2_PERSONALITY, that is why I thought it is fine to take the 7th bit as SQE_GROUP now.

Thanks,
Ming

author maintainer
#3 lore raw
Quoted 19 lines · Ming Lei
Quoted 18 lines · Ming Lei
Quoted 11 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that

The space is the 1st short for uring_cmd, instead of SQE128 only.

Also it is overlapped with ->optval and ->addr3, so just wondering why not use ->__pad2?

Quoted 7 lines
Quoted 3 lines

Another ways is to use __pad2 for sqe2_flags for non-uring_cmd, and for uring_cmd, use its top 16 as sqe2_flags, this way does work, but it is just a bit ugly to use.

Also IOSQE2_PERSONALITY doesn't have to be per-SQE, and it can be one feature of IORING_FEAT_IOSQE2_PERSONALITY, that is why I thought it is fine to take the 7th bit as SQE_GROUP now.

Not sure I follow your thinking there, can you expand?

Signature · 1 line
Jens Axboe
#4 lore raw
Quoted 24 lines · Jens Axboe
Quoted 23 lines · Ming Lei
Quoted 21 lines · Ming Lei
Quoted 11 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that

The space is the 1st short for uring_cmd, instead of SQE128 only.

Also it is overlapped with ->optval and ->addr3, so just wondering why not use ->__pad2?

Another ways is to use __pad2 for sqe2_flags for non-uring_cmd, and for uring_cmd, use its top 16 as sqe2_flags, this way does work, but it is just a bit ugly to use.

Quoted 5 lines
Quoted 3 lines

Also IOSQE2_PERSONALITY doesn't have to be per-SQE, and it can be one feature of IORING_FEAT_IOSQE2_PERSONALITY, that is why I thought it is fine to take the 7th bit as SQE_GROUP now.

Not sure I follow your thinking there, can you expand?

It could be one io_uring setup flag, such as IORING_SETUP_IOSQE2_PERSONALITY.

If this flag is set, take __pad2 as sqe2_flags, otherwise use current way, so it doesn't have to take bit7 of sqe_flags for this purpose.

Also in future, if uring_cmd needs personality, it still may reuse top 16bit of uring_cmd_flags for that.

Thanks,
Ming

author maintainer
#5 lore raw
Quoted 29 lines · Ming Lei
Quoted 28 lines · Jens Axboe
Quoted 26 lines · Ming Lei
Quoted 13 lines · Ming Lei
Quoted 11 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that

Quoted 8 lines

The space is the 1st short for uring_cmd, instead of SQE128 only.

Also it is overlapped with ->optval and ->addr3, so just wondering why not use ->__pad2?

Another ways is to use __pad2 for sqe2_flags for non-uring_cmd, and for uring_cmd, use its top 16 as sqe2_flags, this way does work, but it is just a bit ugly to use.

Also IOSQE2_PERSONALITY doesn't have to be per-SQE, and it can be one feature of IORING_FEAT_IOSQE2_PERSONALITY, that is why I thought it is fine to take the 7th bit as SQE_GROUP now.

Quoted 7 lines
Quoted 1 line

Not sure I follow your thinking there, can you expand?

It could be one io_uring setup flag, such as
IORING_SETUP_IOSQE2_PERSONALITY.

If this flag is set, take __pad2 as sqe2_flags, otherwise use current way, so it doesn't have to take bit7 of sqe_flags for this purpose.

Would probably have to be a IORING_SETUP_IOSQE2_FLAGS or something in general. And while that could work, not a huge fan of that. I think we should retain that for when a v2 of the sqe is done, to coordinate which version to use.

Quoted 2 lines

Also in future, if uring_cmd needs personality, it still may reuse top 16bit of uring_cmd_flags for that.

Right, that's what I referred to in terms of uring_cmd just having its own way to set personality.

Signature · 1 line
Jens Axboe
#6 lore raw
Quoted 35 lines · Jens Axboe
Quoted 34 lines · Ming Lei
Quoted 24 lines · Jens Axboe
Quoted 23 lines · Ming Lei
Quoted 21 lines · Ming Lei
Quoted 11 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that

The space is the 1st short for uring_cmd, instead of SQE128 only.

Also it is overlapped with ->optval and ->addr3, so just wondering why not use ->__pad2?

Another ways is to use __pad2 for sqe2_flags for non-uring_cmd, and for uring_cmd, use its top 16 as sqe2_flags, this way does work, but it is just a bit ugly to use.

Quoted 5 lines
Quoted 3 lines

Also IOSQE2_PERSONALITY doesn't have to be per-SQE, and it can be one feature of IORING_FEAT_IOSQE2_PERSONALITY, that is why I thought it is fine to take the 7th bit as SQE_GROUP now.

Not sure I follow your thinking there, can you expand?

It could be one io_uring setup flag, such as
IORING_SETUP_IOSQE2_PERSONALITY.

Quoted 7 lines
Quoted 2 lines

If this flag is set, take __pad2 as sqe2_flags, otherwise use current way, so it doesn't have to take bit7 of sqe_flags for this purpose.

Would probably have to be a IORING_SETUP_IOSQE2_FLAGS or something in general. And while that could work, not a huge fan of that. I think we should retain that for when a v2 of the sqe is done, to coordinate which version to use.

Fair enough.

Now there are 16bits for new features, which may put v2 off long enough.

Quoted 6 lines
Quoted 3 lines

Also in future, if uring_cmd needs personality, it still may reuse top 16bit of uring_cmd_flags for that.

Right, that's what I referred to in terms of uring_cmd just having its own way to set personality.

Then this approach is safe to go, imo.

Thanks,
Ming

author maintainer
#7 lore raw
Quoted 39 lines · Ming Lei
Quoted 38 lines · Jens Axboe
Quoted 36 lines · Ming Lei
Quoted 24 lines · Jens Axboe
Quoted 23 lines · Ming Lei
Quoted 21 lines · Ming Lei
Quoted 11 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that

The space is the 1st short for uring_cmd, instead of SQE128 only.

Also it is overlapped with ->optval and ->addr3, so just wondering why not use ->__pad2?

Another ways is to use __pad2 for sqe2_flags for non-uring_cmd, and for uring_cmd, use its top 16 as sqe2_flags, this way does work, but it is just a bit ugly to use.

Quoted 5 lines
Quoted 3 lines

Also IOSQE2_PERSONALITY doesn't have to be per-SQE, and it can be one feature of IORING_FEAT_IOSQE2_PERSONALITY, that is why I thought it is fine to take the 7th bit as SQE_GROUP now.

Not sure I follow your thinking there, can you expand?

It could be one io_uring setup flag, such as
IORING_SETUP_IOSQE2_PERSONALITY.

If this flag is set, take __pad2 as sqe2_flags, otherwise use current way, so it doesn't have to take bit7 of sqe_flags for this purpose.

Quoted 8 lines
Quoted 4 lines

Would probably have to be a IORING_SETUP_IOSQE2_FLAGS or something in general. And while that could work, not a huge fan of that. I think we should retain that for when a v2 of the sqe is done, to coordinate which version to use.

Fair enough.

Now there are 16bits for new features, which may put v2 off long enough.

Exactly, hopefully that'll push the need out quite a bit, so we have time to do something nice for v2.

Quoted 7 lines
Quoted 5 lines
Quoted 2 lines

Also in future, if uring_cmd needs personality, it still may reuse top 16bit of uring_cmd_flags for that.

Right, that's what I referred to in terms of uring_cmd just having its own way to set personality.

Then this approach is safe to go, imo.

Thanks I think so too, and it'll unblock the sqe grouping. So at least that paves the way for the first part of your patchset. I'll post a v2 of it shortly.

Signature · 1 line
Jens Axboe
#8 lore raw
In reply to Jens Axboe
Quoted 35 lines · Jens Axboe
Quoted 34 lines · Ming Lei
Quoted 24 lines · Jens Axboe
Quoted 23 lines · Ming Lei
Quoted 21 lines · Ming Lei
Quoted 11 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that

The space is the 1st short for uring_cmd, instead of SQE128 only.

Also it is overlapped with ->optval and ->addr3, so just wondering why not use ->__pad2?

Another ways is to use __pad2 for sqe2_flags for non-uring_cmd, and for uring_cmd, use its top 16 as sqe2_flags, this way does work, but it is just a bit ugly to use.

Quoted 5 lines
Quoted 3 lines

Also IOSQE2_PERSONALITY doesn't have to be per-SQE, and it can be one feature of IORING_FEAT_IOSQE2_PERSONALITY, that is why I thought it is fine to take the 7th bit as SQE_GROUP now.

Not sure I follow your thinking there, can you expand?

It could be one io_uring setup flag, such as
IORING_SETUP_IOSQE2_PERSONALITY.

Quoted 7 lines
Quoted 2 lines

If this flag is set, take __pad2 as sqe2_flags, otherwise use current way, so it doesn't have to take bit7 of sqe_flags for this purpose.

Would probably have to be a IORING_SETUP_IOSQE2_FLAGS or something in general. And while that could work, not a huge fan of that. I think we should retain that for when a v2 of the sqe is done, to coordinate which version to use.

A setup flag over an sqe flag for marking IMHO would be a _much_ better approach. It doesn't take an SQE bit for nothing, you can parse and process it in the slow setup path, enable static keys for the hot path and so on.

Signature · 1 line
Pavel Begunkov
author maintainer
#9 lore raw
Quoted 44 lines · Pavel Begunkov
Quoted 35 lines · Jens Axboe
Quoted 34 lines · Ming Lei
Quoted 24 lines · Jens Axboe
Quoted 23 lines · Ming Lei
Quoted 21 lines · Ming Lei
Quoted 11 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that

The space is the 1st short for uring_cmd, instead of SQE128 only.

Also it is overlapped with ->optval and ->addr3, so just wondering why not use ->__pad2?

Another ways is to use __pad2 for sqe2_flags for non-uring_cmd, and for uring_cmd, use its top 16 as sqe2_flags, this way does work, but it is just a bit ugly to use.

Quoted 5 lines
Quoted 3 lines

Also IOSQE2_PERSONALITY doesn't have to be per-SQE, and it can be one feature of IORING_FEAT_IOSQE2_PERSONALITY, that is why I thought it is fine to take the 7th bit as SQE_GROUP now.

Not sure I follow your thinking there, can you expand?

It could be one io_uring setup flag, such as
IORING_SETUP_IOSQE2_PERSONALITY.

Quoted 7 lines
Quoted 2 lines

If this flag is set, take __pad2 as sqe2_flags, otherwise use current way, so it doesn't have to take bit7 of sqe_flags for this purpose.

Would probably have to be a IORING_SETUP_IOSQE2_FLAGS or something in general. And while that could work, not a huge fan of that. I think we should retain that for when a v2 of the sqe is done, to coordinate which version to use.

Quoted 4 lines

A setup flag over an sqe flag for marking IMHO would be a _much_ better approach. It doesn't take an SQE bit for nothing, you can parse and process it in the slow setup path, enable static keys for the hot path and so on.

Alright, if both of you like that, then let me rework the flags patch to do that instead. If we go that route, then we can just defer doing the actual setup flag until we need a new bit beyond filling the last sqe->flags bit with GROUP.

Signature · 1 line
Jens Axboe
author maintainer
#10 lore raw
In reply to Ming Lei
Quoted 13 lines · Ming Lei
Quoted 11 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that

Quoted 8 lines

The space is the 1st short for uring_cmd, instead of SQE128 only.

Also it is overlapped with ->optval and ->addr3, so just wondering why not use ->__pad2?

Another ways is to use __pad2 for sqe2_flags for non-uring_cmd, and for uring_cmd, use its top 16 as sqe2_flags, this way does work, but it is just a bit ugly to use.

Agree, __pad2 is the better place, we don't want it overlapping with add3/optval.

Signature · 1 line
Jens Axboe
#11 lore raw
In reply to Jens Axboe
Quoted 7 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

Quoted 5 lines

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that would have to be done via a uring cmd flag.

Interesting, I was just experimenting using the personality bits for similar purposes I mentioned but in a different way, and I even thought if anything it could be used to extend sqe flags though I'm not a huge fan of that.

Signature · 1 line
Pavel Begunkov
author maintainer
#12 lore raw
Quoted 14 lines · Pavel Begunkov
Quoted 7 lines · Jens Axboe

In hindsight everything is clearer, but it probably should've been known that 8 bits of ->flags would run out sooner than later. Rather than gobble up the last bit for a random use case, add a bit that controls whether or not ->personality is used as a flags2 argument. If that is the case, then there's a new IOSQE2_PERSONALITY flag that tells io_uring which personality field to read.

Quoted 5 lines

While this isn't the prettiest, it does allow extending with 15 extra flags, and retains being able to use personality with any kind of command. The exception is uring cmd, where personality2 will overlap with the space set aside for SQE128. If they really need that, then that would have to be done via a uring cmd flag.

Quoted 4 lines

Interesting, I was just experimenting using the personality bits for similar purposes I mentioned but in a different way, and I even thought if anything it could be used to extend sqe flags though I'm not a huge fan of that.

We're going to need more SQE flags at some point. At least with the potential to extend it with a setup flag in the future, we can grab the last one and have that other option down the line.

I don't mind grabbing personality. Obviously it'd be better to get free space somewhere, but there's no free real estate...

Signature · 1 line
Jens Axboe