[PATCH 0/3] Discourage sched_yield(2)

[PATCH 0/3] Discourage sched_yield(2)

[Alejandro Colomar]

Hi Andrew,

Here's a patch set for discouraging sched_yield(2).  See the formatted
page at the bottom.  I also updated some POSIX historic detail.

Cheers,
Alex


Alejandro Colomar (3):
  sched_yield.2: HISTORY: POSIX.1-2008 makes this non-optional
  sched_yield.2: NOTES: Remove misleading sentence
  sched_yield.2: Rename NOTES to CAVEATS, and reorder contents

 man2/sched_yield.2 | 41 +++++++++++++++++++----------------------
 1 file changed, 19 insertions(+), 22 deletions(-)

---
$ MANWIDTH=72 man ./man2/sched_yield.2 | cat
sched_yield(2)            System Calls Manual           sched_yield(2)

NAME
       sched_yield - yield the processor

LIBRARY
       Standard C library (libc, -lc)

SYNOPSIS
       #include <sched.h>

       int sched_yield(void);

DESCRIPTION
       sched_yield()  causes the calling thread to relinquish the CPU.
       The thread is moved to the end of the queue for its static pri‐
       ority and a new thread gets to run.

RETURN VALUE
       On success, sched_yield() returns 0.  On error, -1 is returned,
       and errno is set to indicate the error.

ERRORS
       In the Linux implementation, sched_yield() always succeeds.

STANDARDS
       POSIX.1‐2008.

HISTORY
       POSIX.1‐2001 (but optional).  POSIX.1‐2008.

       Before POSIX.1‐2008, systems on which sched_yield()  is  avail‐
       able defined _POSIX_PRIORITY_SCHEDULING in <unistd.h>.

CAVEATS
       sched_yield()  is  intended  for  use with real‐time scheduling
       policies (i.e., SCHED_FIFO or SCHED_RR).  Use of  sched_yield()
       with  nondeterministic  scheduling policies such as SCHED_OTHER
       is unspecified and very likely means your application design is
       broken.

       If the calling thread is the only thread in the highest  prior‐
       ity  list at that time, it will continue to run after a call to
       sched_yield().

       Avoid calling sched_yield()  unnecessarily  or  inappropriately
       (e.g.,  when  resources needed by other schedulable threads are
       still held by the caller), since doing so will result in unnec‐
       essary context switches, which will degrade system performance.

SEE ALSO
       sched(7)

Linux man‐pages (unreleased)    (date)                  sched_yield(2)
-- 
2.40.1

[PATCH 1/3] sched_yield.2: HISTORY: POSIX.1-2008 makes this non-optional

[Alejandro Colomar]

Signed-off-by: Alejandro Colomar <alx@kernel.org>
---
 man2/sched_yield.2 | 18 ++++++++++--------
 1 file changed, 10 insertions(+), 8 deletions(-)

diff --git a/man2/sched_yield.2 b/man2/sched_yield.2
index f1024762a..bab0f9569 100644
--- a/man2/sched_yield.2
+++ b/man2/sched_yield.2
@@ -38,20 +38,22 @@ .SH ERRORS
 .SH STANDARDS
 POSIX.1-2008.
 .SH HISTORY
-POSIX.1-2001.
+POSIX.1-2001 (but optional).
+POSIX.1-2008.
+.PP
+Before POSIX.1-2008,
+systems on which
+.BR sched_yield ()
+is available defined
+.B _POSIX_PRIORITY_SCHEDULING
+in
+.IR <unistd.h> .
 .SH NOTES
 If the calling thread is the only thread in the highest
 priority list at that time,
 it will continue to run after a call to
 .BR sched_yield ().
 .PP
-POSIX systems on which
-.BR sched_yield ()
-is available define
-.B _POSIX_PRIORITY_SCHEDULING
-in
-.IR <unistd.h> .
-.PP
 Strategic calls to
 .BR sched_yield ()
 can improve performance by giving other threads or processes
-- 
2.40.1

[PATCH 2/3] sched_yield.2: NOTES: Remove misleading sentence

[Alejandro Colomar]

sched_yield(2) is not the right thing for heavily contended resources.
The right thing to do is to call functions that wake the waiting threads.

Link: <https://www.realworldtech.com/forum/?threadid=189711&curpostid=189752>
Cc: Andrew Clayton <a.clayton@nginx.com>
Signed-off-by: Alejandro Colomar <alx@nginx.com>
---
 man2/sched_yield.2 | 5 -----
 1 file changed, 5 deletions(-)

diff --git a/man2/sched_yield.2 b/man2/sched_yield.2
index bab0f9569..5e5b45a48 100644
--- a/man2/sched_yield.2
+++ b/man2/sched_yield.2
@@ -54,11 +54,6 @@ .SH NOTES
 it will continue to run after a call to
 .BR sched_yield ().
 .PP
-Strategic calls to
-.BR sched_yield ()
-can improve performance by giving other threads or processes
-a chance to run when (heavily) contended resources (e.g., mutexes)
-have been released by the caller.
 Avoid calling
 .BR sched_yield ()
 unnecessarily or inappropriately
-- 
2.40.1

[PATCH 3/3] sched_yield.2: Rename NOTES to CAVEATS, and reorder contents

[Alejandro Colomar]

Put the last paragraph at the top of the CAVEATS section, since it's
probably the most important for readers.  This system call is likely not
the right one for most programs; let's discourage its use.

Link: <https://www.realworldtech.com/forum/?threadid=189711&curpostid=189752>
Cc: Andrew Clayton <a.clayton@nginx.com>
Signed-off-by: Alejandro Colomar <alx@nginx.com>
---
 man2/sched_yield.2 | 24 ++++++++++++------------
 1 file changed, 12 insertions(+), 12 deletions(-)

diff --git a/man2/sched_yield.2 b/man2/sched_yield.2
index 5e5b45a48..4eb858018 100644
--- a/man2/sched_yield.2
+++ b/man2/sched_yield.2
@@ -48,7 +48,18 @@ .SH HISTORY
 .B _POSIX_PRIORITY_SCHEDULING
 in
 .IR <unistd.h> .
-.SH NOTES
+.SH CAVEATS
+.BR sched_yield ()
+is intended for use with real-time scheduling policies (i.e.,
+.B SCHED_FIFO
+or
+.BR SCHED_RR ).
+Use of
+.BR sched_yield ()
+with nondeterministic scheduling policies such as
+.B SCHED_OTHER
+is unspecified and very likely means your application design is broken.
+.PP
 If the calling thread is the only thread in the highest
 priority list at that time,
 it will continue to run after a call to
@@ -61,16 +72,5 @@ .SH NOTES
 schedulable threads are still held by the caller),
 since doing so will result in unnecessary context switches,
 which will degrade system performance.
-.PP
-.BR sched_yield ()
-is intended for use with real-time scheduling policies (i.e.,
-.B SCHED_FIFO
-or
-.BR SCHED_RR ).
-Use of
-.BR sched_yield ()
-with nondeterministic scheduling policies such as
-.B SCHED_OTHER
-is unspecified and very likely means your application design is broken.
 .SH SEE ALSO
 .BR sched (7)
-- 
2.40.1