Corinna Vinschen wrote:
On Nov 27 10:14, Christian Franke wrote:
Corinna Vinschen wrote:
On Nov 25 21:20, Christian Franke wrote:
Corinna Vinschen wrote:
- Isn't returning SCHED_FIFO sched_getscheduler() just as wrong?
Definitly. SCHED_FIFO is a non-preemptive(!) real-time policy. Windows does
not offer anything like that to userland (AFAIK).
https://man7.org/linux/man-pages/man7/sched.7.html
I wonder whether there was a use case for this emulation when this module
was introduced in 2001.
Just guessing here, but using one of the RT schedulers was the only way
to enable changing the priority from user space and using SCHED_FIFO was
maybe in error.
Shouldn't that be SCHED_OTHER, and sched_setscheduler() should check
for that instead? Cygwin in a real-time scenario sounds a bit
far-fetched...
Agree.
Note that SCHED_OTHER requires sched_priority == 0, so most of the
sched_get/set*() priority related code would no longer make sense then.
This is the other problem. Changing this to SCHED_OTHER sounds like
dropping potentially used functionality. Maybe we should just switch to
SCHED_RR?
Yes, it at least would be closer to what windows does. It is still
non-preemptive from the point of view of lower priorities. I don't know what
the Win32 *_PRIORITY_CLASSes actually do.
Who does? :}
As far as I understand the related documentation, a more sophisticated
emulation (aka fake) of SCHED_* would be:
- Allow to switch between SCHED_OTHER (default) and SCHED_RR with
sched_setscheduler().
- If SCHED_OTHER is selected, change PRIORITY_CLASS with setpriority() and
ignore (or fail on?) attempts to change sched_priority with
sched_setparam().
- If SCHED_RR is selected, ignore setpriority() and change PRIORITY_CLASS
with sched_setparam().
Possibly not worth the effort...
Why not? It should probably also allow SCHED_FIFO for backward compat,
since, in a way, it doesn't really matter anyway. This would be nice
for 3.6.
OK, I'll try to provide a patch.
And I think your patch here should go in as is, just with the release
message in release/3.5.5 so we can cherry-pick it to the 3.5 branch.
Attached. Message moved to 3.5.5 and "Fixes:" changed as suggested.
From 86266b67334d43ac52a9b7ac1ee879a8d34f0c62 Mon Sep 17 00:00:00 2001
From: Christian Franke <christian.fra...@t-online.de>
Date: Wed, 27 Nov 2024 16:39:37 +0100
Subject: [PATCH] Cygwin: sched_setscheduler: allow changes of the priority
Behave like sched_setparam() if the requested policy is identical
to the fixed value (SCHED_FIFO) returned by sched_getscheduler().
Fixes: 9a08b2c02eea ("* sched.cc: New file. Implement sched*.")
Signed-off-by: Christian Franke <christian.fra...@t-online.de>
---
winsup/cygwin/release/3.5.5 | 3 +++
winsup/cygwin/sched.cc | 5 ++++-
2 files changed, 7 insertions(+), 1 deletion(-)
diff --git a/winsup/cygwin/release/3.5.5 b/winsup/cygwin/release/3.5.5
index d91f2b92c..d41d168c6 100644
--- a/winsup/cygwin/release/3.5.5
+++ b/winsup/cygwin/release/3.5.5
@@ -45,3 +45,6 @@ Fixes:
- Fix segfault in sigtimedwait() when using timeout.
Addresses: https://cygwin.com/pipermail/cygwin/2024-November/256762.html
+
+- sched_setscheduler(2) allows to change the priority if the policy is
+ equal to the value returned by sched_getscheduler(2).
diff --git a/winsup/cygwin/sched.cc b/winsup/cygwin/sched.cc
index 71a1e868f..333786f44 100644
--- a/winsup/cygwin/sched.cc
+++ b/winsup/cygwin/sched.cc
@@ -399,8 +399,11 @@ int
sched_setscheduler (pid_t pid, int policy,
const struct sched_param *param)
{
+ if (policy == SCHED_FIFO) /* returned by sched_getscheduler. */
+ return sched_setparam (pid, param);
+
/* on win32, you can't change the scheduler. Doh! */
- set_errno (ENOSYS);
+ set_errno (EINVAL);
return -1;
}
--
2.45.1
