What *is* the API for sched_getaffinity? Should sched_getaffinity always succeed when using cpu_set_t?

Quoted 9 lines · Roland McGrath

Quoted 2 lines

I can fix the glibc manual. A 'configured' CPU is one that the OS can bring online.

Where do you get this definition, in the absence of a standard that specifies _SC_NPROCESSORS_CONF? The only definition I've ever known for _SC_NPROCESSORS_CONF is a value that's constant for at least the life of the process (and probably until reboot) that is the upper bound for what _SC_NPROCESSORS_ONLN might ever report. If the implementation for Linux is inconsistent with that definition, then it's just a bug in the implementation.

Let me reiterate my understanding such that you can help me clarify exactly my interpretation of the glibc manual wording regarding the two existing constants.

The reality of the situation is that the linux kernel as an abstraction presents the following:

(a) The number of online cpus.
- Changes dynamically.
- Not constant for the life of the process, but pretty constant.

(b) The number of configured cpus.
- The number of detected cpus that the OS could access.
- Some of them may be offline for various reasons.
- Changes dynamically with hotplug.

(c) The number of possible CPUs the OS or hardware can support.
- The internal software infrastructure is designed to support at most this many cpus.
- Constant for the uptime of the system.
- May be tied in some way to the hardware.

On Linux, glibc currently maps _SC_NPROCESSORS_CONF to (b) via /sys/devices/system/cpu/cpu*, and _SC_NPROCESSORS_ONLN to (a) via /sys/devices/system/cpu/online.

The problem is that sched_getaffinity and sched_setaffinity only cares about (c) since the size of the kernel affinity mask is of size (c).

What Motohiro-san was requesting was that the manual should make it clear that _SC_NPROCESSORS_CONF is distinct from (c) which is an OS limit that the user doesn't know.

We need not expose (c) as a new _SC_* constant since it's not really required, since glibc's sched_getaffinity and sched_setaffinity could hide the fact that (c) exists from userspace (and that's what I suggest should happen).

Does that clarify my statement?

Carlos,

Quoted 44 lines · Carlos O'Donell

Quoted 9 lines · Roland McGrath

Quoted 2 lines

I can fix the glibc manual. A 'configured' CPU is one that the OS can bring online.

Where do you get this definition, in the absence of a standard that specifies _SC_NPROCESSORS_CONF? The only definition I've ever known for _SC_NPROCESSORS_CONF is a value that's constant for at least the life of the process (and probably until reboot) that is the upper bound for what _SC_NPROCESSORS_ONLN might ever report. If the implementation for Linux is inconsistent with that definition, then it's just a bug in the implementation.

Let me reiterate my understanding such that you can help me clarify exactly my interpretation of the glibc manual wording regarding the two existing constants.

The reality of the situation is that the linux kernel as an abstraction presents the following:

(a) The number of online cpus.
- Changes dynamically.
- Not constant for the life of the process, but pretty constant.

(b) The number of configured cpus.
- The number of detected cpus that the OS could access.
- Some of them may be offline for various reasons.
- Changes dynamically with hotplug.

(c) The number of possible CPUs the OS or hardware can support.
- The internal software infrastructure is designed to support at most this many cpus.
- Constant for the uptime of the system.
- May be tied in some way to the hardware.

On Linux, glibc currently maps _SC_NPROCESSORS_CONF to (b) via /sys/devices/system/cpu/cpu*, and _SC_NPROCESSORS_ONLN to (a) via /sys/devices/system/cpu/online.

The problem is that sched_getaffinity and sched_setaffinity only cares about (c) since the size of the kernel affinity mask is of size (c).

What Motohiro-san was requesting was that the manual should make it clear that _SC_NPROCESSORS_CONF is distinct from (c) which is an OS limit that the user doesn't know.

Quoted 6 lines

We need not expose (c) as a new _SC_* constant since it's not really required, since glibc's sched_getaffinity and sched_setaffinity could hide the fact that (c) exists from userspace (and that's what I suggest should happen).

Does that clarify my statement?

It's a long time since the last activity in this discussion, and I see that
https://sourceware.org/bugzilla/show_bug.cgi?id=15630
remains open, I propose to apply the patch below to the sched_setattr/sched_getattr man page. Seem okay?

Cheers,

Michael

man2/sched_setaffinity.2 +44 -0

@@ -333,6 +334,57 @@ main(int argc, char *argv[])
     }
 }
 .fi
+.SH BUGS
+The glibc
+.BR sched_setaffinity ()
+and
+.BR sched_getaffinity ()
+wrapper functions do not handle systems with more than 1024 CPUs.
+.\" FIXME . See https://sourceware.org/bugzilla/show_bug.cgi?id=15630
+.\" and https://sourceware.org/ml/libc-alpha/2013-07/msg00288.html
+The
+.I cpu_set_t
+data type used by glibc has a fixed size of 128 bytes,
+meaning that the the maximum CPU number that can be represented is 1023.
+If the system has more than 1024 CPUs, then:
+.IP * 3
+The
+.BR sched_setaffinity ()
+.I mask
+argument is not capable of representing the excess CPUs.
+.IP *
+Calls of the form:
+
+    sched_getaffinity(pid, sizeof(cpu_set_t), &mask);
+
+will fail with error
+.BR EINVAL ,
+the error produced by the underlying system call for the case where the
+.I mask
+size specified in
+.I cpusetsize
+is smaller than the size of the affinity mask used by the kernel.
+.PP
+The workaround for this problem is to fall back to the use of the
+underlying system call (via
+.BR syscall (2)),
+passing
+.I mask
+arguments of a sufficient size.
+Using a value based on the number of online CPUs:
+
+    (sysconf(_SC_NPROCESSORS_CONF) / (sizeof(unsigned long) * 8) + 1)
+                                   * sizeof(unsigned long)
+
+is probably sufficient as the size of the mask,
+although the value returned by the

man2/sched_setaffinity.2 +8 -0

+
+    sched_getaffinity(pid, sizeof(cpu_set_t), &mask);
+
+will fail with the error
+.BR EINVAL ,
+the error produced by the underlying system call for the case where the
+.I mask
+size specified in

+.SS Handling systems with more than 1024 CPUs

man2/sched_setaffinity.2 +7 -0

+The
+.I cpu_set_t
+data type used by glibc has a fixed size of 128 bytes,
+meaning that the maximum CPU number that can be represented is 1023.
+.\" FIXME . See https://sourceware.org/bugzilla/show_bug.cgi?id=15630
+.\" and https://sourceware.org/ml/libc-alpha/2013-07/msg00288.html
+If the system has more than 1024 CPUs, then calls of the form:

+.PP

man2/sched_setaffinity.2 +3 -0

+The underlying system calls (which represent CPU masks as bit masks of type
+.IR "unsigned long\ *" )
+impose no restriction on the size of the mask.

Quoted 13 lines · Michael Kerrisk (man-pages)

+.SS Handling systems with more than 1024 CPUs

man2/sched_setaffinity.2 +12 -0

+The
+.I cpu_set_t
+data type used by glibc has a fixed size of 128 bytes,
+meaning that the maximum CPU number that can be represented is 1023.
+.\" FIXME . See https://sourceware.org/bugzilla/show_bug.cgi?id=15630
+.\" and https://sourceware.org/ml/libc-alpha/2013-07/msg00288.html
+If the system has more than 1024 CPUs, then calls of the form:
+
+    sched_getaffinity(pid, sizeof(cpu_set_t), &mask);
+
+will fail with the error
+.BR EINVAL ,

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 243

Quoted 5 lines

man2/sched_setaffinity.2 +5 -0

+the error produced by the underlying system call for the case where the
+.I mask
+size specified in
+.I cpusetsize
+is smaller than the size of the affinity mask used by the kernel.

I think it is best to leave this as unspecified as possible. Kernel behavior already changed once, and I can imagine it changing again.

Carlos and I tried to get clarification of the future direction of the kernel interface here:

<https://sourceware.org/ml/libc-alpha/2015-06/msg00210.html>

No reply so far, unless I missed something.

Quoted 4 lines

+.PP

man2/sched_setaffinity.2 +3 -0

+The underlying system calls (which represent CPU masks as bit masks of type
+.IR "unsigned long\ *" )
+impose no restriction on the size of the mask.

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 257

Quoted 10 lines

man2/sched_setaffinity.2 +10 -0

+To handle systems with more than 1024 CPUs, one must dynamically allocate the
+.I mask
+argument using
+.BR CPU_ALLOC (3)
+and manipulate the mask using the "_S" macros described in
+.BR CPU_ALLOC (3).
+Using an allocation based on the number of online CPUs:
+
+    cpu_set_t *mask = CPU_ALLOC(CPU_ALLOC_SIZE(
+                                sysconf(_SC_NPROCESSORS_CONF)));

I believe this is incorrect in several ways:

CPU_ALLOC uses the raw CPU counts. CPU_ALLOC_SIZE converts from the raw count to the size in bytes. (This API is misdesigned.)

sysconf(_SC_NPROCESSORS_CONF) is not related to the kernel CPU mask size, so it is not the correct value.

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 265

Quoted 7 lines

man2/sched_setaffinity.2 +7 -0

+is probably sufficient, although the value returned by the
+.BR sysconf ()
+call can in theory change during the lifetime of the process.
+Alternatively, one can obtain a value that is guaranteed to be stable for
+the lifetime of the process by proby for the size of the required mask using
+.BR sched_getaffinity ()
+calls with increasing mask sizes until the call does not fail with the error

This is the only possible way right now if you do not want to read sysconf values.

It's also worth noting that the system call and the glibc function have different return values.

Michael,

given the approach is accepted by Carlos and Roland, I have some minor textual suggestions for the patch itself.

On 06/26/2015 10:05 PM, Michael Kerrisk (man-pages) wrote:

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 226

Quoted 7 lines

man2/sched_setaffinity.2 +1 -0

@@ -223,6 +223,47 @@ system call returns the size (in bytes) of the
 .I cpumask_t
 data type that is used internally by the kernel to
 represent the CPU set bit mask.
+.SS Handling systems with more than 1024 CPUs

What if the system has exactly 1024 CPUs ?
Suggestion: systems with 1024 or more CPUs

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 232

Quoted 7 lines

man2/sched_setaffinity.2 +6 -0

+The
+.I cpu_set_t
+data type used by glibc has a fixed size of 128 bytes,
+meaning that the maximum CPU number that can be represented is 1023.
+.\" FIXME . See https://sourceware.org/bugzilla/show_bug.cgi?id=15630
+.\" and https://sourceware.org/ml/libc-alpha/2013-07/msg00288.html

No objection, although I have never really noticed external references in man-pages (esp. web refs). Shouldn't these be generally avoided ? (and yes, I have noticed the FIXME)

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 233

Quoted 1 line

man2/sched_setaffinity.2 +1 -0

+If the system has more than 1024 CPUs, then calls of the form:

1024 or more CPUs.

Quoted 12 lines

man2/sched_setaffinity.2 +12 -0

+
+    sched_getaffinity(pid, sizeof(cpu_set_t), &mask);
+
+will fail with the error
+.BR EINVAL ,
+the error produced by the underlying system call for the case where the
+.I mask
+size specified in
+.I cpusetsize
+is smaller than the size of the affinity mask used by the kernel.
+.PP
+The underlying system calls (which represent CPU masks as bit masks of type

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 251

Quoted 6 lines

man2/sched_setaffinity.2 +6 -0

+.IR "unsigned long\ *" )
+impose no restriction on the size of the mask.
+To handle systems with more than 1024 CPUs, one must dynamically allocate the
+.I mask
+argument using
+.BR CPU_ALLOC (3)

I would rewrite the sentence to avoid "one must".

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 252

Quoted 1 line

man2/sched_setaffinity.2 +1 -0

+and manipulate the mask using the "_S" macros described in

and manipulate the macros ending with "_S" as described in

Quoted 4 lines

man2/sched_setaffinity.2 +4 -0

+.BR CPU_ALLOC (3).
+Using an allocation based on the number of online CPUs:
+
+    cpu_set_t *mask = CPU_ALLOC(CPU_ALLOC_SIZE(

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 262

Quoted 6 lines

man2/sched_setaffinity.2 +6 -0

+                                sysconf(_SC_NPROCESSORS_CONF)));
+
+is probably sufficient, although the value returned by the
+.BR sysconf ()
+call can in theory change during the lifetime of the process.
+Alternatively, one can obtain a value that is guaranteed to be stable for

Like above, I would replace "one can obtain a value" by "a value can be obtained".

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 263

Quoted 1 line

man2/sched_setaffinity.2 +1 -0

+the lifetime of the process by proby for the size of the required mask using

s/proby/probing/.

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 266

Quoted 3 lines

man2/sched_setaffinity.2 +3 -0

+.BR sched_getaffinity ()
+calls with increasing mask sizes until the call does not fail with the error
+.BR EINVAL .

I would replace "until the call does not fail with error ..." by "while the call succeeds".

Hello Florian,

Thanks for your comments, and sorry for the delayed follow-up.

Quoted 15 lines · Florian Weimer

Quoted 13 lines · Michael Kerrisk (man-pages)

+.SS Handling systems with more than 1024 CPUs

man2/sched_setaffinity.2 +12 -0

+The
+.I cpu_set_t
+data type used by glibc has a fixed size of 128 bytes,
+meaning that the maximum CPU number that can be represented is 1023.
+.\" FIXME . See https://sourceware.org/bugzilla/show_bug.cgi?id=15630
+.\" and https://sourceware.org/ml/libc-alpha/2013-07/msg00288.html
+If the system has more than 1024 CPUs, then calls of the form:
+
+    sched_getaffinity(pid, sizeof(cpu_set_t), &mask);
+
+will fail with the error
+.BR EINVAL ,

Quoted 8 lines

man2/sched_setaffinity.2 system call returns the size (in bytes) of the · line 243

Quoted 5 lines

man2/sched_setaffinity.2 +5 -0

+the error produced by the underlying system call for the case where the
+.I mask
+size specified in
+.I cpusetsize
+is smaller than the size of the affinity mask used by the kernel.

I think it is best to leave this as unspecified as possible. Kernel behavior already changed once, and I can imagine it changing again.

Hmmm. Something needs to be said about what the kernel is doing though. Otherwise, it's hard to make sense of this subsection. Did you have a suggested rewording that removes the piece you find problematic?

Handling systems with more than 1024 CPUs
The underlying system calls (which represent CPU masks as bit masks of type unsigned long *) impose no restriction on the size of the CPU mask. However, the cpu_set_t data type used by glibc has a fixed size of 128 bytes, meaning that the maximum CPU number that can be represented is 1023. If the system has more than 1024 CPUs, then calls of the form:

sched_getaffinity(pid, sizeof(cpu_set_t), &mask);

will fail with the error EINVAL, the error produced by the underlying system call for the case where the mask size speci‐ fied in cpusetsize is smaller than the size of the affinity mask used by the kernel.

When working on systems with more than 1024 CPUs, one must
dynamically allocate the mask argument. Currently, the only
way to do this is by probing for the size of the required mask using sched_getaffinity() calls with increasing mask sizes (until the call does not fail with the error EINVAL).

Better?

Quoted 29 lines · Michael Kerrisk (man-pages)

Hello Florian,

Thanks for your comments, and sorry for the delayed follow-up.

Quoted 15 lines · Florian Weimer

Quoted 13 lines · Michael Kerrisk (man-pages)

+.SS Handling systems with more than 1024 CPUs

(quoted diff) +12 -0

+The
+.I cpu_set_t
+data type used by glibc has a fixed size of 128 bytes,
+meaning that the maximum CPU number that can be represented is 1023.
+.\" FIXME . See https://sourceware.org/bugzilla/show_bug.cgi?id=15630
+.\" and https://sourceware.org/ml/libc-alpha/2013-07/msg00288.html
+If the system has more than 1024 CPUs, then calls of the form:
+
+    sched_getaffinity(pid, sizeof(cpu_set_t), &mask);
+
+will fail with the error
+.BR EINVAL ,

Quoted 8 lines

Review

Quoted 5 lines

(quoted diff) +5 -0

+the error produced by the underlying system call for the case where the
+.I mask
+size specified in
+.I cpusetsize
+is smaller than the size of the affinity mask used by the kernel.

I think it is best to leave this as unspecified as possible. Kernel behavior already changed once, and I can imagine it changing again.

Quoted 3 lines

Hmmm. Something needs to be said about what the kernel is doing though. Otherwise, it's hard to make sense of this subsection. Did you have a suggested rewording that removes the piece you find problematic?

What about this?

“If the kernel affinity mask is larger than 1024 then
…
is smaller than the size of the affinity mask used by the kernel. Depending on the system CPU topology, the kernel affinity mask can be substantially larger than the number of active CPUs in the system. ”

I.e., make clear that the size of the mask can be quite different from the CPU count.

Quoted 21 lines

Handling systems with more than 1024 CPUs
The underlying system calls (which represent CPU masks as bit masks of type unsigned long *) impose no restriction on the size of the CPU mask. However, the cpu_set_t data type used by glibc has a fixed size of 128 bytes, meaning that the maximum CPU number that can be represented is 1023. If the system has more than 1024 CPUs, then calls of the form:

sched_getaffinity(pid, sizeof(cpu_set_t), &mask);

will fail with the error EINVAL, the error produced by the underlying system call for the case where the mask size speci‐ fied in cpusetsize is smaller than the size of the affinity mask used by the kernel.

When working on systems with more than 1024 CPUs, one must
dynamically allocate the mask argument. Currently, the only
way to do this is by probing for the size of the required mask using sched_getaffinity() calls with increasing mask sizes (until the call does not fail with the error EINVAL).

Quoted 1 line

Better?

“more than 1024 CPUs” should be “large [kernel CPU] affinity masks” throughout.