Skip to contentSkip to conversation

package/musl: Make scheduler functions Linux-compatible

PATCH New stefan.nickl at gmail.com 4 participants 13 messages May 13, 2019 – Oct 27, 2019 1 message not in the archive

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

---
 package/musl/0002-sched-linux-compat.patch | 74 ++++++++++++++++++++++
 1 file changed, 74 insertions(+)
 create mode 100644 package/musl/0002-sched-linux-compat.patch
(quoted diff) +0 -0

b/package/musl/0002-sched-linux-compat.patch
new file mode 100644
index 0000000000..c1626ea7c3

package/musl/0002-sched-linux-compat.patch new file +38 -0
@@ -0,0 +1,74 @@
+The POSIX functions sched_getscheduler(), sched_setscheduler(),
+sched_getparam(), sched_setparam() are technically not correctly
+implemented by the Linux syscalls of the same name, because what the
+kernel calls a PID and what POSIX calls a PID isn't truly the same,
+resulting in somewhat different semantics as to what these functions
+exactly apply to.
+
+Since the musl developers put a high premium on POSIX compliance, they
+deliberately implement these functions to return -ENOSYS instead of
+relaying them to the respective Linux syscalls as glibc/uClibc do.
+
+This means that virtually all programs written for Linux using these
+functions are currently broken. For example running 'chrt -p 1' fails
+with "Function not implemented" on a musl-libc based system.
+
+While the musl developers are right strictly speaking, it seems
+unfeasible to fix all affected programs in a way that hardly any Linux
+developer will consider to be broken in the first place. So this patch
+simply changes musl to fall in line with the other libcs.
+
+Signed-off-by: Stefan Nickl <[email protected]>
+---
+diff -ru a/src/sched/sched_getparam.c b/src/sched/sched_getparam.c
+--- a/src/sched/sched_getparam.c	2019-04-10 02:39:21.000000000
+0200
++++ b/src/sched/sched_getparam.c	2019-05-07 08:51:00.980363141
+0200
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_getparam(pid_t pid, struct sched_param *param)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_getparam, pid, param);
+ }
+diff -ru a/src/sched/sched_getscheduler.c

b/src/sched/sched_getscheduler.c
+--- a/src/sched/sched_getscheduler.c 2019-04-10 02:39:21.000000000
+0200
++++ b/src/sched/sched_getscheduler.c 2019-05-07 08:50:48.246021455
+0200
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+
+ int sched_getscheduler(pid_t pid)
+ {
+- return __syscall_ret(-ENOSYS);
++ return syscall(SYS_sched_getscheduler, pid);
+ }
+diff -ru a/src/sched/sched_setparam.c b/src/sched/sched_setparam.c
+--- a/src/sched/sched_setparam.c 2019-04-10 02:39:21.000000000
+0200
++++ b/src/sched/sched_setparam.c 2019-05-07 08:51:03.603227550
+0200
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+
+ int sched_setparam(pid_t pid, const struct sched_param *param)
+ {
+- return __syscall_ret(-ENOSYS);
++ return syscall(SYS_sched_setparam, pid, param);
+ }
+diff -ru a/src/sched/sched_setscheduler.c b/src/sched/sched_setscheduler.c
+--- a/src/sched/sched_setscheduler.c 2019-04-10 02:39:21.000000000
+0200
++++ b/src/sched/sched_setscheduler.c 2019-05-07 08:50:53.379756063
+0200
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+
+ int sched_setscheduler(pid_t pid, int sched, const struct sched_param *param)
+ {
+- return __syscall_ret(-ENOSYS);
++ return syscall(SYS_sched_setscheduler, pid, sched, param);
+ }

#1 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

Hello Stefan,

Thanks for your contribution!

Quoted 24 lines · On Mon, 13 May 2019 22:00:58 +0200 stefan.nickl at gmail.com wrote:
---
 package/musl/0002-sched-linux-compat.patch | 74 ++++++++++++++++++++++
 1 file changed, 74 insertions(+)
 create mode 100644 package/musl/0002-sched-linux-compat.patch
package/musl/0002-sched-linux-compat.patch new file +11 -0
new file mode 100644
index 0000000000..c1626ea7c3
@@ -0,0 +1,74 @@
+The POSIX functions sched_getscheduler(), sched_setscheduler(),
+sched_getparam(), sched_setparam() are technically not correctly
+implemented by the Linux syscalls of the same name, because what the
+kernel calls a PID and what POSIX calls a PID isn't truly the same,
+resulting in somewhat different semantics as to what these functions
+exactly apply to.
+
+Since the musl developers put a high premium on POSIX compliance, they
+deliberately implement these functions to return -ENOSYS instead of
+relaying them to the respective Linux syscalls as glibc/uClibc do.
+

package/musl/0002-sched-linux-compat.patch line 21

Quoted 10 lines
package/musl/0002-sched-linux-compat.patch new file +10 -0
+This means that virtually all programs written for Linux using these
+functions are currently broken. For example running 'chrt -p 1' fails
+with "Function not implemented" on a musl-libc based system.
+
+While the musl developers are right strictly speaking, it seems
+unfeasible to fix all affected programs in a way that hardly any Linux
+developer will consider to be broken in the first place. So this patch
+simply changes musl to fall in line with the other libcs.
+
+Signed-off-by: Stefan Nickl <[email protected]>

On a purely formatting point of view, we will need this patch to be generated by git format-patch.

However, I really wonder if we want to carry this kind of change/fix inside Buildroot. We would have to keep it forever, because it's basically something the musl upstream developers would never accept.

On the other hand, we already have some quirks for musl in the musl-compat-headers, and maybe we want to be more pragmatic than the musl upstream developers are.

Peter, Arnout, Yann, others, any opinion ?

Best regards,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#2 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

Hi Thomas,

thanks for considering.
This has the inner patch also generated by git.

Cheers,
Stefan

---
 ...e-scheduler-functions-Linux-compatib.patch | 92 +++++++++++++++++++
 1 file changed, 92 insertions(+)
 create mode 100644 package/musl/0002-package-musl-Make-scheduler-functions-Linux-compatib.patch
package/musl/0002-package-musl-Make-scheduler-functions-Linux-compatib.patch new file +92 -0
new file mode 100644
index 0000000000..a8dbb9eaa5
@@ -0,0 +1,92 @@
+From 775a6cf46a7918cde335f959e1a63119c74e68d3 Mon Sep 17 00:00:00 2001
+From: Stefan Nickl <[email protected]>
+Date: Mon, 13 May 2019 22:33:21 +0200
+Subject: [PATCH] package/musl: Make scheduler functions Linux-compatible
+
+The POSIX functions sched_getscheduler(), sched_setscheduler(),
+sched_getparam(), sched_setparam() are technically not correctly
+implemented by the Linux syscalls of the same name, because what the
+kernel calls a PID and what POSIX calls a PID isn't truly the same,
+resulting in somewhat different semantics as to what these functions
+exactly apply to.
+
+Since the musl developers put a high premium on POSIX compliance, they
+deliberately implement these functions to return -ENOSYS instead of
+relaying them to the respective Linux syscalls as glibc/uClibc do.
+
+This means that virtually all programs written for Linux using these
+functions are currently broken. For example running 'chrt -p 1' fails
+with "Function not implemented" on a musl-libc based system.
+
+While the musl developers are right strictly speaking, it seems
+unfeasible to fix all affected programs in a way that hardly any Linux
+developer will consider to be broken in the first place. So this patch
+simply changes musl to fall in line with the other libcs.
+
+Signed-off-by: Stefan Nickl <[email protected]>
+---
+ src/sched/sched_getparam.c     | 3 +--
+ src/sched/sched_getscheduler.c | 3 +--
+ src/sched/sched_setparam.c     | 3 +--
+ src/sched/sched_setscheduler.c | 3 +--
+ 4 files changed, 4 insertions(+), 8 deletions(-)
+
+diff --git a/src/sched/sched_getparam.c b/src/sched/sched_getparam.c
+index 76f10e4..65be107 100644
+--- a/src/sched/sched_getparam.c
++++ b/src/sched/sched_getparam.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_getparam(pid_t pid, struct sched_param *param)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_getparam, pid, param);
+ }
+diff --git a/src/sched/sched_getscheduler.c b/src/sched/sched_getscheduler.c
+index 394e508..4c922f6 100644
+--- a/src/sched/sched_getscheduler.c
++++ b/src/sched/sched_getscheduler.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_getscheduler(pid_t pid)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_getscheduler, pid);
+ }
+diff --git a/src/sched/sched_setparam.c b/src/sched/sched_setparam.c
+index 18623ee..f699faf 100644
+--- a/src/sched/sched_setparam.c
++++ b/src/sched/sched_setparam.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_setparam(pid_t pid, const struct sched_param *param)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_setparam, pid, param);
+ }
+diff --git a/src/sched/sched_setscheduler.c b/src/sched/sched_setscheduler.c
+index 4435f21..e678221 100644
+--- a/src/sched/sched_setscheduler.c
++++ b/src/sched/sched_setscheduler.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_setscheduler(pid_t pid, int sched, const struct sched_param *param)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_setscheduler, pid, sched, param);
+ }
+-- 
+2.21.0
+
Signature · 1 line
2.21.0
#3 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

Hello Stefan,

Quoted 2 lines · On Tue, 14 May 2019 18:32:32 +0200 stefan.nickl at gmail.com wrote:

thanks for considering.
This has the inner patch also generated by git.

Thanks, but could you make a proper submission of it, with a good commit log, like you did for the first patch ?

With this new submission, the commit log would contain "thanks for considering. This has the inner patch also generated by git", which is not a very good commit log.

Thanks,

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com

A message that is not in the archive (sent off-list or not yet received): [email protected]

#5 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

Hello,

Please keep the mailing list Cc'ed. Thanks!

Quoted 3 lines · On Tue, 14 May 2019 18:51:05 +0200 stefan.nickl at gmail.com wrote:

Should I move the text all out from the patch to the commit log, or can I just duplicate the text inside and outside the patch?
Or should one be more detailed and the other shorter?

The text can be similar, but it has a different target.

The commit log is meant to be useful to Buildroot, so it should explain why in Buildroot we would want this change.

The description of the musl patch itself should theoretically be unrelated to Buildroot: it should ideally be an upstreamable patch, and musl doesn't care about Buildroot specifically.

So the patch description should really describe the change in terms of the upstream project, while the commit log should describe the change in the context of Buildroot.

Does this helps?

Thomas

Signature · 3 lines
Thomas Petazzoni, CTO, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com
#6 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

In reply to Thomas Petazzoni

The POSIX functions sched_getscheduler(), sched_setscheduler(), sched_getparam(), sched_setparam() are technically not correctly implemented by the Linux syscalls of the same name, because what the kernel calls a PID and what POSIX calls a PID isn't truly the same, resulting in somewhat different semantics as to what these functions exactly apply to.
Details: https://sourceware.org/bugzilla/show_bug.cgi?id=14829

Since the musl developers put a high premium on POSIX compliance, they deliberately implement these functions to return -ENOSYS instead of relaying them to the respective Linux syscalls as glibc/uClibc do.

Unfortunally this breaks virtually all Linux programs using these functions under musl. For example running 'chrt -p 1' fails with 'Function not implemented' on a musl-libc based system. In particular, it affects embedded systems using these interfaces for scheduling real-time processes.

As it seems unfeasible to fix all affected programs to manually use syscall wrappers instead of the libc functions, make musl behave the Linux way.

---
 ...e-scheduler-functions-Linux-compatib.patch | 76 +++++++++++++++++++
 1 file changed, 76 insertions(+)
 create mode 100644 package/musl/0002-package-musl-Make-scheduler-functions-Linux-compatib.patch
package/musl/0002-package-musl-Make-scheduler-functions-Linux-compatib.patch new file +76 -0
new file mode 100644
index 0000000000..7c3acf9f02
@@ -0,0 +1,76 @@
+From 407c96fc790d0d11ca9603a2a533216c745b5051 Mon Sep 17 00:00:00 2001
+From: Stefan Nickl <[email protected]>
+Date: Mon, 13 May 2019 22:33:21 +0200
+Subject: [PATCH] Make scheduler functions Linux-compatible
+
+Let sched_getscheduler(), sched_setscheduler(), sched_getparam(),
+sched_setparam() invoke the Linux syscalls of the same name instead
+of returning -ENOSYS.
+
+Signed-off-by: Stefan Nickl <[email protected]>
+---
+ src/sched/sched_getparam.c     | 3 +--
+ src/sched/sched_getscheduler.c | 3 +--
+ src/sched/sched_setparam.c     | 3 +--
+ src/sched/sched_setscheduler.c | 3 +--
+ 4 files changed, 4 insertions(+), 8 deletions(-)
+
+diff --git a/src/sched/sched_getparam.c b/src/sched/sched_getparam.c
+index 76f10e4..65be107 100644
+--- a/src/sched/sched_getparam.c
++++ b/src/sched/sched_getparam.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_getparam(pid_t pid, struct sched_param *param)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_getparam, pid, param);
+ }
+diff --git a/src/sched/sched_getscheduler.c b/src/sched/sched_getscheduler.c
+index 394e508..4c922f6 100644
+--- a/src/sched/sched_getscheduler.c
++++ b/src/sched/sched_getscheduler.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_getscheduler(pid_t pid)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_getscheduler, pid);
+ }
+diff --git a/src/sched/sched_setparam.c b/src/sched/sched_setparam.c
+index 18623ee..f699faf 100644
+--- a/src/sched/sched_setparam.c
++++ b/src/sched/sched_setparam.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_setparam(pid_t pid, const struct sched_param *param)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_setparam, pid, param);
+ }
+diff --git a/src/sched/sched_setscheduler.c b/src/sched/sched_setscheduler.c
+index 4435f21..e678221 100644
+--- a/src/sched/sched_setscheduler.c
++++ b/src/sched/sched_setscheduler.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_setscheduler(pid_t pid, int sched, const struct sched_param *param)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_setscheduler, pid, sched, param);
+ }
+-- 
+2.21.0
+
Signature · 1 line
2.21.0
#7 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

Quoted 18 lines · stefan.nickl at gmail.com

The POSIX functions sched_getscheduler(), sched_setscheduler(), sched_getparam(), sched_setparam() are technically not correctly implemented by the Linux syscalls of the same name, because what the kernel calls a PID and what POSIX calls a PID isn't truly the same, resulting in somewhat different semantics as to what these functions exactly apply to.
Details: https://sourceware.org/bugzilla/show_bug.cgi?id=14829

Since the musl developers put a high premium on POSIX compliance, they deliberately implement these functions to return -ENOSYS instead of relaying them to the respective Linux syscalls as glibc/uClibc do.

Unfortunally this breaks virtually all Linux programs using these functions under musl. For example running 'chrt -p 1' fails with 'Function not implemented' on a musl-libc based system. In particular, it affects embedded systems using these interfaces for scheduling real-time processes.

Quoted 3 lines

As it seems unfeasible to fix all affected programs to manually use syscall wrappers instead of the libc functions, make musl behave the Linux way.

As Thomas wrote, the patch as it is is not upstreamable since musl considers
this a feature.

However, it is possible that Rich can be convinced to actually read the POSIX
spec and decide that linux behaviour is correct (except for the fact that it sched_setscheduler() should return the previous scheduling policy rather than 0, but that can be approximated (in a racy way) by calling sched_getscheduler() first). So it might be worth trying to push the patch upstream.

Quoted 12 lines
---
 ...e-scheduler-functions-Linux-compatib.patch | 76 +++++++++++++++++++
 1 file changed, 76 insertions(+)
 create mode 100644 package/musl/0002-package-musl-Make-scheduler-functions-Linux-compatib.patch
package/musl/0002-package-musl-Make-scheduler-functions-Linux-compatib.patch new file +0 -0
new file mode 100644
index 0000000000..7c3acf9f02

package/musl/0002-package-musl-Make-scheduler-functions-Linux-compatib.patch line 8

Quoted 8 lines
package/musl/0002-package-musl-Make-scheduler-functions-Linux-compatib.patch new file +8 -0
@@ -0,0 +1,76 @@
+From 407c96fc790d0d11ca9603a2a533216c745b5051 Mon Sep 17 00:00:00 2001
+From: Stefan Nickl <[email protected]>
+Date: Mon, 13 May 2019 22:33:21 +0200
+Subject: [PATCH] Make scheduler functions Linux-compatible
+
+Let sched_getscheduler(), sched_setscheduler(), sched_getparam(),
+sched_setparam() invoke the Linux syscalls of the same name instead
+of returning -ENOSYS.

... but then you'd need to refer to
https://pubs.opengroup.org/onlinepubs/7908799/xsh/sched_setparam.html
and explain that in Linux, a pid_t can only refer to a process (because a thread's PID can only be obtained with gettid() or clone(), neither of which is available in musl), and that POSIX specifies that in a process with multiple threads, sched_setscheduler should affect only the main thread, which is exactly what Linux does.

Either way (whether you upstream it or not), this patch looks good to me.

Regards,
Arnout

Quoted 69 lines
package/musl/0002-package-musl-Make-scheduler-functions-Linux-compatib.patch new file +68 -0
+
+Signed-off-by: Stefan Nickl <[email protected]>
+---
+ src/sched/sched_getparam.c     | 3 +--
+ src/sched/sched_getscheduler.c | 3 +--
+ src/sched/sched_setparam.c     | 3 +--
+ src/sched/sched_setscheduler.c | 3 +--
+ 4 files changed, 4 insertions(+), 8 deletions(-)
+
+diff --git a/src/sched/sched_getparam.c b/src/sched/sched_getparam.c
+index 76f10e4..65be107 100644
+--- a/src/sched/sched_getparam.c
++++ b/src/sched/sched_getparam.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_getparam(pid_t pid, struct sched_param *param)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_getparam, pid, param);
+ }
+diff --git a/src/sched/sched_getscheduler.c b/src/sched/sched_getscheduler.c
+index 394e508..4c922f6 100644
+--- a/src/sched/sched_getscheduler.c
++++ b/src/sched/sched_getscheduler.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_getscheduler(pid_t pid)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_getscheduler, pid);
+ }
+diff --git a/src/sched/sched_setparam.c b/src/sched/sched_setparam.c
+index 18623ee..f699faf 100644
+--- a/src/sched/sched_setparam.c
++++ b/src/sched/sched_setparam.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_setparam(pid_t pid, const struct sched_param *param)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_setparam, pid, param);
+ }
+diff --git a/src/sched/sched_setscheduler.c b/src/sched/sched_setscheduler.c
+index 4435f21..e678221 100644
+--- a/src/sched/sched_setscheduler.c
++++ b/src/sched/sched_setscheduler.c
+@@ -1,8 +1,7 @@
+ #include <sched.h>
+-#include <errno.h>
+ #include "syscall.h"
+ 
+ int sched_setscheduler(pid_t pid, int sched, const struct sched_param *param)
+ {
+-	return __syscall_ret(-ENOSYS);
++	return syscall(SYS_sched_setscheduler, pid, sched, param);
+ }
+-- 
+2.21.0
+
#8 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

Quoted 4 lines · Arnout Vandecappelle

As Thomas wrote, the patch as it is is not upstreamable since musl
considers
this a feature.

Quoted 9 lines

However, it is possible that Rich can be convinced to actually read
the POSIX
spec and decide that linux behaviour is correct (except for the fact that it
sched_setscheduler() should return the previous scheduling policy rather than 0,
but that can be approximated (in a racy way) by calling sched_getscheduler()
first). So it might be worth trying to push the patch upstream.

Hi Arnout,

I think upstreaming is out of the question since Rich has actually *undone* it before
https://git.musl-libc.org/cgit/musl/commit/?id=1e21e78bf7a5

Also asked on #musl a couple days ago whether they might be willing to at least #ifdef it, and only got suggestions for application side fixes. So, no dice it seems.

Cheers,
Stefan

#9 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

Quoted 17 lines · stefan.nickl at gmail.com
Quoted 4 lines · Arnout Vandecappelle

As Thomas wrote, the patch as it is is not upstreamable since musl
considers
this a feature.

Quoted 9 lines

However, it is possible that Rich can be convinced to actually read
the POSIX
spec and decide that linux behaviour is correct (except for the fact that it
sched_setscheduler() should return the previous scheduling policy rather than 0,
but that can be approximated (in a racy way) by calling sched_getscheduler()
first). So it might be worth trying to push the patch upstream.

Hi Arnout,

Quoted 7 lines

I think upstreaming is out of the question since Rich has actually *undone* it before
https://git.musl-libc.org/cgit/musl/commit/?id=1e21e78bf7a5

Also asked on #musl a couple days ago whether they might be willing to at least #ifdef it, and only got suggestions for application side fixes. So, no dice it seems.

Yes, but has the argument "Linux does in fact implement POSIX" been used already?

Regards,
Arnout

#10 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

Quoted 26 lines · Arnout Vandecappelle
Quoted 24 lines · stefan.nickl at gmail.com
Quoted 5 lines · Arnout Vandecappelle

As Thomas wrote, the patch as it is is not upstreamable since
musl
considers
this a feature.

Quoted 11 lines

However, it is possible that Rich can be convinced to actually
read
the POSIX
spec and decide that linux behaviour is correct (except for the fact
that it
sched_setscheduler() should return the previous scheduling policy rather than 0,
but that can be approximated (in a racy way) by calling sched_getscheduler()
first). So it might be worth trying to push the patch upstream.

Hi Arnout,

I think upstreaming is out of the question since Rich has actually *undone* it before
https://git.musl-libc.org/cgit/musl/commit/?id=1e21e78bf7a5

Quoted 7 lines
Quoted 4 lines

Also asked on #musl a couple days ago whether they might be willing to
at least #ifdef it, and only got suggestions for application side fixes. So, no dice it seems.

Yes, but has the argument "Linux does in fact implement POSIX" been
used already?

I haven't used that argument and I'm not aware of anyone that has.

This comes pretty close though:
https://github.com/MusicPlayerDaemon/MPD/issues/218

And just recently Rich managed to convince even the glibc developers of the opposite, so I certainly don't feel up to arguing that position ;-)
https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c70824b9a4645c0ecd049da8cfdb2c28ae7ada23

Cheers,
Stefan

#11 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

Stefan, All,

On 2019-05-14 22:03 +0200, stefan.nickl at gmail.com spake thusly:

Quoted 35 lines · Arnout Vandecappelle
Quoted 26 lines · stefan.nickl at gmail.com
Quoted 24 lines · Arnout Vandecappelle
Quoted 5 lines

As Thomas wrote, the patch as it is is not upstreamable since
musl
considers
this a feature.

Quoted 11 lines

However, it is possible that Rich can be convinced to actually
read
the POSIX
spec and decide that linux behaviour is correct (except for the fact
that it
sched_setscheduler() should return the previous scheduling policy rather than 0,
but that can be approximated (in a racy way) by calling sched_getscheduler()
first). So it might be worth trying to push the patch upstream.

Hi Arnout,

I think upstreaming is out of the question since Rich has actually *undone* it before
https://git.musl-libc.org/cgit/musl/commit/?id=1e21e78bf7a5

Quoted 7 lines
Quoted 4 lines

Also asked on #musl a couple days ago whether they might be willing to
at least #ifdef it, and only got suggestions for application side fixes. So, no dice it seems.

Yes, but has the argument "Linux does in fact implement POSIX" been
used already?

Quoted 8 lines

I haven't used that argument and I'm not aware of anyone that has.

This comes pretty close though:
https://github.com/MusicPlayerDaemon/MPD/issues/218

And just recently Rich managed to convince even the glibc developers of the opposite, so I certainly don't feel up to arguing that position ;-)
https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c70824b9a4645c0ecd049da8cfdb2c28ae7ada23

So, if glibc is going the musl way, because Linux does not allow to properly implement the sched_* stuff, when why do we even attempt to implement them in musl at all?

Yes, I read the argument that "applications are broken without that", but they are already brioken anyway because the underlyign syscalls are unsound (from what I read in the various threads and the bugzilla).

As Thomas said, if we add this patch, we'll have to carry it ad libitum, because there are absolutely zero chance this is upstreamed one day.

Besides, it also does not address the case for pre-built toolchains, which leaves users out in the cold with broken applications anway.

So, as painful as it might be, I'd say no to this patch.

Regards,
Yann E. MORIN.

Signature · 6 lines
.-----------------.--------------------.------------------.--------------------.
|  Yann E. MORIN  | Real-Time Embedded | /"\ ASCII RIBBON | Erics' conspiracy: |
| +33 662 376 056 | Software  Designer | \ / CAMPAIGN     |  ___               |
| +33 561 099 427 `------------.-------:  X  AGAINST      |  \e/  There is no  |
| http://ymorin.is-a-geek.org/ | _/*\_ | / \ HTML MAIL    |   v   conspiracy.  |
'------------------------------^-------^------------------^--------------------'
#12 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

Quoted 57 lines · Yann E. MORIN

Stefan, All,

On 2019-05-14 22:03 +0200, stefan.nickl at gmail.com spake thusly:

Quoted 45 lines · Arnout Vandecappelle
Quoted 38 lines · stefan.nickl at gmail.com
Quoted 30 lines · Arnout Vandecappelle
Quoted 6 lines

As Thomas wrote, the patch as it is is not upstreamable
since
musl
considers
this a feature.

Quoted 15 lines

However, it is possible that Rich can be convinced to
actually
read
the POSIX
spec and decide that linux behaviour is correct (except for the
fact
that it
sched_setscheduler() should return the previous scheduling policy
rather than 0,
but that can be approximated (in a racy way) by calling sched_getscheduler()
first). So it might be worth trying to push the patch upstream.

Hi Arnout,

I think upstreaming is out of the question since Rich has actually
*undone* it before
https://git.musl-libc.org/cgit/musl/commit/?id=1e21e78bf7a5

Quoted 6 lines

Also asked on #musl a couple days ago whether they might be willing
to
at least #ifdef it, and only got suggestions for application side
fixes. So, no dice it seems.

Quoted 3 lines

Yes, but has the argument "Linux does in fact implement POSIX"
been
used already?

I haven't used that argument and I'm not aware of anyone that has.

Quoted 8 lines

This comes pretty close though:
https://github.com/MusicPlayerDaemon/MPD/issues/218

And just recently Rich managed to convince even the glibc developers of
the opposite, so I certainly don't feel up to arguing that position ;-)
https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=c70824b9a4645c0ecd049da8cfdb2c28ae7ada23

Quoted 3 lines

So, if glibc is going the musl way, because Linux does not allow to properly implement the sched_* stuff, when why do we even attempt to implement them in musl at all?

Hi Yann,

that's not what I'm seeing.

The glibc folks are not going to change the library call <-> Linux syscall correspondence.
They merely make note of the fact that they've screwed up in the past, and that now it is too late to change.
All that happens is that the non-conformance gets documented.

What makes this extra confusing is that there are source files in the glibc tree with -ENOSYS-returning implementations of these functions, but it seems the functions that really get used for Linux are generated from a syscalls.list file to use the syscall.
Maybe someone more familiar with glibc can confirm this.
In the uclibc-ng case, it seems rather straightforward, see e.g. libc/sysdeps/linux/common/sched_setscheduler.c

Cheers,
Stefan

#13 lore raw

PATCH 1/1 package/musl: Make scheduler functions Linux-compatible

In reply to stefan.nickl at gmail.com
Quoted 24 lines · stefan.nickl at gmail.com
---
 package/musl/0002-sched-linux-compat.patch | 74 ++++++++++++++++++++++
 1 file changed, 74 insertions(+)
 create mode 100644 package/musl/0002-sched-linux-compat.patch
package/musl/0002-sched-linux-compat.patch new file +11 -0
new file mode 100644
index 0000000000..c1626ea7c3
@@ -0,0 +1,74 @@
+The POSIX functions sched_getscheduler(), sched_setscheduler(),
+sched_getparam(), sched_setparam() are technically not correctly
+implemented by the Linux syscalls of the same name, because what the
+kernel calls a PID and what POSIX calls a PID isn't truly the same,
+resulting in somewhat different semantics as to what these functions
+exactly apply to.
+
+Since the musl developers put a high premium on POSIX compliance, they
+deliberately implement these functions to return -ENOSYS instead of
+relaying them to the respective Linux syscalls as glibc/uClibc do.
+

package/musl/0002-sched-linux-compat.patch line 21

Quoted 10 lines
package/musl/0002-sched-linux-compat.patch new file +10 -0
+This means that virtually all programs written for Linux using these
+functions are currently broken. For example running 'chrt -p 1' fails
+with "Function not implemented" on a musl-libc based system.
+
+While the musl developers are right strictly speaking, it seems
+unfeasible to fix all affected programs in a way that hardly any Linux
+developer will consider to be broken in the first place. So this patch
+simply changes musl to fall in line with the other libcs.
+
+Signed-off-by: Stefan Nickl <[email protected]>

This is really the only way that we can make musl usable in practice. Ideally
it should be merged into musl-compat-headers or something like that, but that's a lot more complicated.

Therefore, I've applied to master, thanks.

Regards,
Arnout