[RFC 0/5] support nonblock submission for splice pipe to pipe

[RFC 0/5] support nonblock submission for splice pipe to pipe

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

splice from pipe to pipe is a trivial case, and we can support nonblock
try for it easily. splice depends on iowq at all which is slow. Let's
build a fast submission path for it by supporting nonblock.

Wrote a simple test to test time spent of splicing from pipe to pipe:


Did 50 times test for each, ignore the highest and lowest number,
calculate the average number:

before patchset: 119.85 usec
with patchset: 29.5 usec

----------------
I'm not sure if we should use a io_uring specific flag rather than
SPLICE_F_NONBLOCK since from mutex_lock_nest to mutex_trylock changes
the behavior under debug environment I guess. Or maybe there is another
better option than mutex_trylock?


Hao Xu (5):
  io_uring: move sp->len check up for splice and tee
  pipe: add trylock helpers for pipe lock
  splice: support nonblock for splice from pipe to pipe
  io_uring: support nonblock try for splicing from pipe to pipe
  io_uring: add file_in in io_splice{} to avoid duplicate calculation

 fs/pipe.c                 | 29 +++++++++++++++++++++
 fs/splice.c               | 21 +++++++++++++---
 include/linux/pipe_fs_i.h |  2 ++
 io_uring/splice.c         | 53 +++++++++++++++++++++++++++++----------
 4 files changed, 89 insertions(+), 16 deletions(-)


base-commit: d8271bf021438f468dab3cd84fe5279b5bbcead8
-- 
2.25.1

[PATCH 1/5] io_uring: move sp->len check up for splice and tee

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

The traditional sync splice code return 0 if len is 0 at the beginning
of syscall, similar thing for tee. So move up sp->len zero check so that
it reaches quick ending when len is 0.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 io_uring/splice.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)

diff --git a/io_uring/splice.c b/io_uring/splice.c
index 0e19d6330345..b2cd1044c3ee 100644
--- a/io_uring/splice.c
+++ b/io_uring/splice.c
@@ -53,6 +53,9 @@ int io_tee(struct io_kiocb *req, unsigned int issue_flags)
 	struct file *in;
 	long ret = 0;
 
+	if (unlikely(!sp->len))
+		goto done;
+
 	if (issue_flags & IO_URING_F_NONBLOCK)
 		return -EAGAIN;
 
@@ -65,8 +68,7 @@ int io_tee(struct io_kiocb *req, unsigned int issue_flags)
 		goto done;
 	}
 
-	if (sp->len)
-		ret = do_tee(in, out, sp->len, flags);
+	ret = do_tee(in, out, sp->len, flags);
 
 	if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
 		io_put_file(in);
@@ -95,6 +97,9 @@ int io_splice(struct io_kiocb *req, unsigned int issue_flags)
 	struct file *in;
 	long ret = 0;
 
+	if (unlikely(!sp->len))
+		goto done;
+
 	if (issue_flags & IO_URING_F_NONBLOCK)
 		return -EAGAIN;
 
@@ -110,8 +115,7 @@ int io_splice(struct io_kiocb *req, unsigned int issue_flags)
 	poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
 	poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
 
-	if (sp->len)
-		ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
+	ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
 
 	if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
 		io_put_file(in);
-- 
2.25.1

[PATCH 2/5] pipe: add trylock helpers for pipe lock

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

Add two helpers pipe_trylock() and pipe_double_trylock(), they are used
in nonblock splice(pipe to pipe) scenario in next patches.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 fs/pipe.c                 | 29 +++++++++++++++++++++++++++++
 include/linux/pipe_fs_i.h |  2 ++
 2 files changed, 31 insertions(+)

diff --git a/fs/pipe.c b/fs/pipe.c
index 74ae9fafd25a..033736eb61fb 100644
--- a/fs/pipe.c
+++ b/fs/pipe.c
@@ -98,6 +98,11 @@ void pipe_unlock(struct pipe_inode_info *pipe)
 }
 EXPORT_SYMBOL(pipe_unlock);
 
+int pipe_trylock(struct pipe_inode_info *pipe)
+{
+	return mutex_trylock(&pipe->mutex);
+}
+
 static inline void __pipe_lock(struct pipe_inode_info *pipe)
 {
 	mutex_lock_nested(&pipe->mutex, I_MUTEX_PARENT);
@@ -122,6 +127,30 @@ void pipe_double_lock(struct pipe_inode_info *pipe1,
 	}
 }
 
+int pipe_double_trylock(struct pipe_inode_info *pipe1,
+			 struct pipe_inode_info *pipe2)
+{
+	BUG_ON(pipe1 == pipe2);
+
+	if (pipe1 < pipe2) {
+		if (!pipe_trylock(pipe1))
+			return 0;
+		if (!pipe_trylock(pipe2)) {
+			pipe_unlock(pipe1);
+			return 0;
+		}
+	} else {
+		if (!pipe_trylock(pipe2))
+			return 0;
+		if (!pipe_trylock(pipe1)) {
+			pipe_unlock(pipe2);
+			return 0;
+		}
+	}
+
+	return 1;
+}
+
 static void anon_pipe_buf_release(struct pipe_inode_info *pipe,
 				  struct pipe_buffer *buf)
 {
diff --git a/include/linux/pipe_fs_i.h b/include/linux/pipe_fs_i.h
index cb0fd633a610..78dc2c7c999f 100644
--- a/include/linux/pipe_fs_i.h
+++ b/include/linux/pipe_fs_i.h
@@ -235,8 +235,10 @@ static inline bool pipe_buf_try_steal(struct pipe_inode_info *pipe,
 
 /* Pipe lock and unlock operations */
 void pipe_lock(struct pipe_inode_info *);
+int pipe_trylock(struct pipe_inode_info *);
 void pipe_unlock(struct pipe_inode_info *);
 void pipe_double_lock(struct pipe_inode_info *, struct pipe_inode_info *);
+int pipe_double_trylock(struct pipe_inode_info *, struct pipe_inode_info *);
 
 /* Wait for a pipe to be readable/writable while dropping the pipe lock */
 void pipe_wait_readable(struct pipe_inode_info *);
-- 
2.25.1

[PATCH 3/5] splice: support nonblock for splice from pipe to pipe

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

When SPLICE_F_NONBLOCK is set, splice() still may be blocked by pipe
lock in pipe to pipe scenario. Add trylock logic to make it more
nonblock

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 fs/splice.c | 21 ++++++++++++++++++---
 1 file changed, 18 insertions(+), 3 deletions(-)

diff --git a/fs/splice.c b/fs/splice.c
index 047b79db8eb5..b087e00ed079 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -1372,7 +1372,12 @@ static int ipipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
 		return 0;
 
 	ret = 0;
-	pipe_lock(pipe);
+	if (flags & SPLICE_F_NONBLOCK) {
+		if (!pipe_trylock(pipe))
+			return -EAGAIN;
+	} else {
+		pipe_lock(pipe);
+	}
 
 	while (pipe_empty(pipe->head, pipe->tail)) {
 		if (signal_pending(current)) {
@@ -1408,7 +1413,12 @@ static int opipe_prep(struct pipe_inode_info *pipe, unsigned int flags)
 		return 0;
 
 	ret = 0;
-	pipe_lock(pipe);
+	if (flags & SPLICE_F_NONBLOCK) {
+		if (!pipe_trylock(pipe))
+			return -EAGAIN;
+	} else {
+		pipe_lock(pipe);
+	}
 
 	while (pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
 		if (!pipe->readers) {
@@ -1460,7 +1470,12 @@ static int splice_pipe_to_pipe(struct pipe_inode_info *ipipe,
 	 * grabbing by pipe info address. Otherwise two different processes
 	 * could deadlock (one doing tee from A -> B, the other from B -> A).
 	 */
-	pipe_double_lock(ipipe, opipe);
+	if (flags & SPLICE_F_NONBLOCK) {
+		if (!pipe_double_trylock(ipipe, opipe))
+			return -EAGAIN;
+	} else {
+		pipe_double_lock(ipipe, opipe);
+	}
 
 	i_tail = ipipe->tail;
 	i_mask = ipipe->ring_size - 1;
-- 
2.25.1

[PATCH 4/5] io_uring: support nonblock try for splicing from pipe to pipe

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

splice() in io_uring is running in a slow way since it fully depends on
io workers. While splicing from pipe to pipe is a simpler case compared
with file to pipe and pipe to file. Let's make it support nonblock
first. This way we get a fast path for splicing fom pipe to pipe.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 io_uring/splice.c | 24 +++++++++++++++++++++---
 1 file changed, 21 insertions(+), 3 deletions(-)

diff --git a/io_uring/splice.c b/io_uring/splice.c
index b2cd1044c3ee..650c70e3dde1 100644
--- a/io_uring/splice.c
+++ b/io_uring/splice.c
@@ -88,6 +88,14 @@ int io_splice_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe)
 	return __io_splice_prep(req, sqe);
 }
 
+bool io_splice_support_nowait(struct file *in, struct file *out)
+{
+	if (get_pipe_info(in, true) && get_pipe_info(out, true))
+		return true;
+
+	return false;
+}
+
 int io_splice(struct io_kiocb *req, unsigned int issue_flags)
 {
 	struct io_splice *sp = io_kiocb_to_cmd(req);
@@ -100,9 +108,6 @@ int io_splice(struct io_kiocb *req, unsigned int issue_flags)
 	if (unlikely(!sp->len))
 		goto done;
 
-	if (issue_flags & IO_URING_F_NONBLOCK)
-		return -EAGAIN;
-
 	if (sp->flags & SPLICE_F_FD_IN_FIXED)
 		in = io_file_get_fixed(req, sp->splice_fd_in, issue_flags);
 	else
@@ -112,6 +117,16 @@ int io_splice(struct io_kiocb *req, unsigned int issue_flags)
 		goto done;
 	}
 
+	if (issue_flags & IO_URING_F_NONBLOCK) {
+		if (io_splice_support_nowait(in, out)) {
+			flags |= SPLICE_F_NONBLOCK;
+		} else {
+			if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
+				io_put_file(in);
+			return -EAGAIN;
+		}
+	}
+
 	poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
 	poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
 
@@ -119,6 +134,9 @@ int io_splice(struct io_kiocb *req, unsigned int issue_flags)
 
 	if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
 		io_put_file(in);
+	if (ret == -EAGAIN)
+		return ret;
+
 done:
 	if (ret != sp->len)
 		req_set_fail(req);
-- 
2.25.1

Re: [PATCH 4/5] io_uring: support nonblock try for splicing from pipe to pipe

[kernel test robot]

Hi Hao,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on d8271bf021438f468dab3cd84fe5279b5bbcead8]

url:    https://github.com/intel-lab-lkp/linux/commits/Hao-Xu/support-nonblock-submission-for-splice-pipe-to-pipe/20220607-161605
base:   d8271bf021438f468dab3cd84fe5279b5bbcead8
config: mips-buildonly-randconfig-r001-20220607 (https://download.01.org/0day-ci/archive/20220608/202206080550.7nXmrsiv-lkp@intel.com/config)
compiler: clang version 15.0.0 (https://github.com/llvm/llvm-project b92436efcb7813fc481b30f2593a4907568d917a)
reproduce (this is a W=1 build):
        wget https://raw.githubusercontent.com/intel/lkp-tests/master/sbin/make.cross -O ~/bin/make.cross
        chmod +x ~/bin/make.cross
        # install mips cross compiling tool for clang build
        # apt-get install binutils-mips-linux-gnu
        # https://github.com/intel-lab-lkp/linux/commit/96683840c3f19b77a536a259094d24e0cd93ebc0
        git remote add linux-review https://github.com/intel-lab-lkp/linux
        git fetch --no-tags linux-review Hao-Xu/support-nonblock-submission-for-splice-pipe-to-pipe/20220607-161605
        git checkout 96683840c3f19b77a536a259094d24e0cd93ebc0
        # save the config file
        mkdir build_dir && cp config build_dir/.config
        COMPILER_INSTALL_PATH=$HOME/0day COMPILER=clang make.cross W=1 O=build_dir ARCH=mips SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
Reported-by: kernel test robot <lkp@intel.com>

All warnings (new ones prefixed by >>):

>> io_uring/splice.c:91:6: warning: no previous prototype for function 'io_splice_support_nowait' [-Wmissing-prototypes]
   bool io_splice_support_nowait(struct file *in, struct file *out)
        ^
   io_uring/splice.c:91:1: note: declare 'static' if the function is not intended to be used outside of this translation unit
   bool io_splice_support_nowait(struct file *in, struct file *out)
   ^
   static 
   1 warning generated.


vim +/io_splice_support_nowait +91 io_uring/splice.c

    90	
  > 91	bool io_splice_support_nowait(struct file *in, struct file *out)
    92	{
    93		if (get_pipe_info(in, true) && get_pipe_info(out, true))
    94			return true;
    95	
    96		return false;
    97	}
    98	

-- 
0-DAY CI Kernel Test Service
https://01.org/lkp

Re: [PATCH 4/5] io_uring: support nonblock try for splicing from pipe to pipe

[kernel test robot]

Hi Hao,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on d8271bf021438f468dab3cd84fe5279b5bbcead8]

url:    https://github.com/intel-lab-lkp/linux/commits/Hao-Xu/support-nonblock-submission-for-splice-pipe-to-pipe/20220607-161605
base:   d8271bf021438f468dab3cd84fe5279b5bbcead8
config: x86_64-defconfig (https://download.01.org/0day-ci/archive/20220608/202206082229.xoMAgqIw-lkp@intel.com/config)
compiler: gcc-11 (Debian 11.3.0-1) 11.3.0
reproduce (this is a W=1 build):
        # https://github.com/intel-lab-lkp/linux/commit/96683840c3f19b77a536a259094d24e0cd93ebc0
        git remote add linux-review https://github.com/intel-lab-lkp/linux
        git fetch --no-tags linux-review Hao-Xu/support-nonblock-submission-for-splice-pipe-to-pipe/20220607-161605
        git checkout 96683840c3f19b77a536a259094d24e0cd93ebc0
        # save the config file
        mkdir build_dir && cp config build_dir/.config
        make W=1 O=build_dir ARCH=x86_64 SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
Reported-by: kernel test robot <lkp@intel.com>

All warnings (new ones prefixed by >>):

>> io_uring/splice.c:91:6: warning: no previous prototype for 'io_splice_support_nowait' [-Wmissing-prototypes]
      91 | bool io_splice_support_nowait(struct file *in, struct file *out)
         |      ^~~~~~~~~~~~~~~~~~~~~~~~


vim +/io_splice_support_nowait +91 io_uring/splice.c

    90	
  > 91	bool io_splice_support_nowait(struct file *in, struct file *out)
    92	{
    93		if (get_pipe_info(in, true) && get_pipe_info(out, true))
    94			return true;
    95	
    96		return false;
    97	}
    98	

-- 
0-DAY CI Kernel Test Service
https://01.org/lkp

Re: [PATCH 4/5] io_uring: support nonblock try for splicing from pipe to pipe

[kernel test robot]

Hi Hao,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on d8271bf021438f468dab3cd84fe5279b5bbcead8]

url:    https://github.com/intel-lab-lkp/linux/commits/Hao-Xu/support-nonblock-submission-for-splice-pipe-to-pipe/20220607-161605
base:   d8271bf021438f468dab3cd84fe5279b5bbcead8
config: i386-randconfig-s001 (https://download.01.org/0day-ci/archive/20220609/202206090959.CEWN5SEB-lkp@intel.com/config)
compiler: gcc-11 (Debian 11.3.0-1) 11.3.0
reproduce:
        # apt-get install sparse
        # sparse version: v0.6.4-26-gb3cf30ba-dirty
        # https://github.com/intel-lab-lkp/linux/commit/96683840c3f19b77a536a259094d24e0cd93ebc0
        git remote add linux-review https://github.com/intel-lab-lkp/linux
        git fetch --no-tags linux-review Hao-Xu/support-nonblock-submission-for-splice-pipe-to-pipe/20220607-161605
        git checkout 96683840c3f19b77a536a259094d24e0cd93ebc0
        # save the config file
        mkdir build_dir && cp config build_dir/.config
        make W=1 C=1 CF='-fdiagnostic-prefix -D__CHECK_ENDIAN__' O=build_dir ARCH=i386 SHELL=/bin/bash

If you fix the issue, kindly add following tag where applicable
Reported-by: kernel test robot <lkp@intel.com>


sparse warnings: (new ones prefixed by >>)
>> io_uring/splice.c:91:6: sparse: sparse: symbol 'io_splice_support_nowait' was not declared. Should it be static?

-- 
0-DAY CI Kernel Test Service
https://01.org/lkp

[PATCH 5/5] io_uring: add file_in in io_splice{} to avoid duplicate calculation

[Hao Xu]

From: Hao Xu <howeyxu@tencent.com>

Add a member file_in in io_splice{} to avoid duplicate calculation of
input file for splice. This is for the case where we do splice from
pipe to pipe and get -EAGAIN for the inline submission.

Signed-off-by: Hao Xu <howeyxu@tencent.com>
---
 io_uring/splice.c | 33 +++++++++++++++++++--------------
 1 file changed, 19 insertions(+), 14 deletions(-)

diff --git a/io_uring/splice.c b/io_uring/splice.c
index 650c70e3dde1..c97f2971fe7e 100644
--- a/io_uring/splice.c
+++ b/io_uring/splice.c
@@ -21,6 +21,7 @@ struct io_splice {
 	loff_t				off_in;
 	u64				len;
 	int				splice_fd_in;
+	struct file			*file_in;
 	unsigned int			flags;
 };
 
@@ -35,6 +36,7 @@ static int __io_splice_prep(struct io_kiocb *req,
 	if (unlikely(sp->flags & ~valid_flags))
 		return -EINVAL;
 	sp->splice_fd_in = READ_ONCE(sqe->splice_fd_in);
+	sp->file_in = NULL;
 	return 0;
 }
 
@@ -108,34 +110,37 @@ int io_splice(struct io_kiocb *req, unsigned int issue_flags)
 	if (unlikely(!sp->len))
 		goto done;
 
-	if (sp->flags & SPLICE_F_FD_IN_FIXED)
-		in = io_file_get_fixed(req, sp->splice_fd_in, issue_flags);
-	else
-		in = io_file_get_normal(req, sp->splice_fd_in);
-	if (!in) {
-		ret = -EBADF;
-		goto done;
+	if (!sp->file_in) {
+		if (sp->flags & SPLICE_F_FD_IN_FIXED)
+			in = io_file_get_fixed(req, sp->splice_fd_in, issue_flags);
+		else
+			in = io_file_get_normal(req, sp->splice_fd_in);
+
+		if (!in) {
+			ret = -EBADF;
+			goto done;
+		}
+		sp->file_in = in;
+	} else {
+		in = sp->file_in;
 	}
 
 	if (issue_flags & IO_URING_F_NONBLOCK) {
-		if (io_splice_support_nowait(in, out)) {
+		if (io_splice_support_nowait(in, out))
 			flags |= SPLICE_F_NONBLOCK;
-		} else {
-			if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
-				io_put_file(in);
+		else
 			return -EAGAIN;
-		}
 	}
 
 	poff_in = (sp->off_in == -1) ? NULL : &sp->off_in;
 	poff_out = (sp->off_out == -1) ? NULL : &sp->off_out;
 
 	ret = do_splice(in, poff_in, out, poff_out, sp->len, flags);
+	if (ret == -EAGAIN)
+		return ret;
 
 	if (!(sp->flags & SPLICE_F_FD_IN_FIXED))
 		io_put_file(in);
-	if (ret == -EAGAIN)
-		return ret;
 
 done:
 	if (ret != sp->len)
-- 
2.25.1

Re: [RFC 0/5] support nonblock submission for splice pipe to pipe

[Hao Xu]

Hi,
On 6/7/22 16:06, Hao Xu wrote:
> From: Hao Xu <howeyxu@tencent.com>
> 
> splice from pipe to pipe is a trivial case, and we can support nonblock
> try for it easily. splice depends on iowq at all which is slow. Let's
                         ^ should be 'splice in io_uring'
> build a fast submission path for it by supporting nonblock.
> 
> Wrote a simple test to test time spent of splicing from pipe to pipe:
> 

Sorry, forgot to attach the test:
https://github.com/HowHsu/liburing/commit/3ff901b4fb80caf66a520742c6542e35111493a0

Regards,
Hao

Re: [RFC 0/5] support nonblock submission for splice pipe to pipe

[Pavel Begunkov]

On 6/7/22 09:06, Hao Xu wrote:
> From: Hao Xu <howeyxu@tencent.com>
> 
> splice from pipe to pipe is a trivial case, and we can support nonblock
> try for it easily. splice depends on iowq at all which is slow. Let's
> build a fast submission path for it by supporting nonblock.

fwiw,

https://www.spinics.net/lists/kernel/msg3652757.html



> Wrote a simple test to test time spent of splicing from pipe to pipe:
> 
> 
> Did 50 times test for each, ignore the highest and lowest number,
> calculate the average number:
> 
> before patchset: 119.85 usec
> with patchset: 29.5 usec
> 
> ----------------
> I'm not sure if we should use a io_uring specific flag rather than
> SPLICE_F_NONBLOCK since from mutex_lock_nest to mutex_trylock changes
> the behavior under debug environment I guess. Or maybe there is another
> better option than mutex_trylock?
> 
> 
> Hao Xu (5):
>    io_uring: move sp->len check up for splice and tee
>    pipe: add trylock helpers for pipe lock
>    splice: support nonblock for splice from pipe to pipe
>    io_uring: support nonblock try for splicing from pipe to pipe
>    io_uring: add file_in in io_splice{} to avoid duplicate calculation
> 
>   fs/pipe.c                 | 29 +++++++++++++++++++++
>   fs/splice.c               | 21 +++++++++++++---
>   include/linux/pipe_fs_i.h |  2 ++
>   io_uring/splice.c         | 53 +++++++++++++++++++++++++++++----------
>   4 files changed, 89 insertions(+), 16 deletions(-)
> 
> 
> base-commit: d8271bf021438f468dab3cd84fe5279b5bbcead8

-- 
Pavel Begunkov

Re: [RFC 0/5] support nonblock submission for splice pipe to pipe

[Hao Xu]

On 6/7/22 17:27, Pavel Begunkov wrote:
> On 6/7/22 09:06, Hao Xu wrote:
>> From: Hao Xu <howeyxu@tencent.com>
>>
>> splice from pipe to pipe is a trivial case, and we can support nonblock
>> try for it easily. splice depends on iowq at all which is slow. Let's
>> build a fast submission path for it by supporting nonblock.
> 
> fwiw,
> 
> https://www.spinics.net/lists/kernel/msg3652757.html
> 

Thanks, Pavel. Seems it has been discussed for a long time but the
result remains unclear...For me, I think this patch is necessary for 
getting a good SPLICE_F_NONBLOCK user experience.

Re: [RFC 0/5] support nonblock submission for splice pipe to pipe

[Jens Axboe]

On 6/7/22 10:19 PM, Hao Xu wrote:
> On 6/7/22 17:27, Pavel Begunkov wrote:
>> On 6/7/22 09:06, Hao Xu wrote:
>>> From: Hao Xu <howeyxu@tencent.com>
>>>
>>> splice from pipe to pipe is a trivial case, and we can support nonblock
>>> try for it easily. splice depends on iowq at all which is slow. Let's
>>> build a fast submission path for it by supporting nonblock.
>>
>> fwiw,
>>
>> https://www.spinics.net/lists/kernel/msg3652757.html
>>
> 
> Thanks, Pavel. Seems it has been discussed for a long time but the
> result remains unclear...For me, I think this patch is necessary for
> getting a good SPLICE_F_NONBLOCK user experience.

I'd just take it up again, this is something we do need to get done to
avoid io-wq offload. Which is important...

-- 
Jens Axboe

Re: [RFC 0/5] support nonblock submission for splice pipe to pipe

[Pavel Begunkov]

On 6/8/22 05:19, Hao Xu wrote:
> On 6/7/22 17:27, Pavel Begunkov wrote:
>> On 6/7/22 09:06, Hao Xu wrote:
>>> From: Hao Xu <howeyxu@tencent.com>
>>>
>>> splice from pipe to pipe is a trivial case, and we can support nonblock
>>> try for it easily. splice depends on iowq at all which is slow. Let's
>>> build a fast submission path for it by supporting nonblock.
>>
>> fwiw,
>>
>> https://www.spinics.net/lists/kernel/msg3652757.html
>>
> 
> Thanks, Pavel. Seems it has been discussed for a long time but the
> result remains unclear...For me, I think this patch is necessary for getting a good SPLICE_F_NONBLOCK user experience.

I quite agree here, something like this is much needed.
The performance of io_uring pipe I/O is miserably slow

-- 
Pavel Begunkov

Re: [RFC 0/5] support nonblock submission for splice pipe to pipe

[Hao Xu]

On 6/7/22 16:06, Hao Xu wrote:
> From: Hao Xu <howeyxu@tencent.com>
> 
> splice from pipe to pipe is a trivial case, and we can support nonblock
> try for it easily. splice depends on iowq at all which is slow. Let's
> build a fast submission path for it by supporting nonblock.
> 
> Wrote a simple test to test time spent of splicing from pipe to pipe:
> 
> 
> Did 50 times test for each, ignore the highest and lowest number,
> calculate the average number:
> 
> before patchset: 119.85 usec
> with patchset: 29.5 usec
> 

Hi Al,
Any comments on this one?

Regards,
Hao