For review: CPU_SET.3

For review: CPU_SET.3

[Michael Kerrisk]

Hi Bert,

I created this page by removing the CPU_*() material from
sched_setscheduler.2 and adding documentation of the many other macros
that have been added in more recent glibc releases.

I remember looking at a mail thread you pointed at a few days back
where you said the interface of the CPU_*S() macros is somewhat
confusing in its use of bits in some places, and bytes in others.
It's hard to disagree with you -- so I included an example in the
man-pages which hopefully might reduce the confusion.

Of course, since these are macros, my use of types in the prototypes
in the SYNOPSIS is in many cases invention; let me know if something
looks weird.

Also, take a look at
http://sourceware.org/bugzilla/show_bug.cgi?id=7029.  Unless I'm
confused (which is always possible), this is a real bug.  Do you
agree?

Thanks,

Michael

.\" Copyright (C) 2006 Michael Kerrisk
.\" and Copyright (C) 2008 Linux Foundation, written by Michael Kerrisk
.\"     <mtk.manpages-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
.\"
.\" Permission is granted to make and distribute verbatim copies of this
.\" manual provided the copyright notice and this permission notice are
.\" preserved on all copies.
.\"
.\" Permission is granted to copy and distribute modified versions of this
.\" manual under the conditions for verbatim copying, provided that the
.\" entire resulting derived work is distributed under the terms of a
.\" permission notice identical to this one.
.\"
.\" Since the Linux kernel and libraries are constantly changing, this
.\" manual page may be incorrect or out-of-date.  The author(s) assume no
.\" responsibility for errors or omissions, or for damages resulting from
.\" the use of the information contained herein.  The author(s) may not
.\" have taken the same level of care in the production of this manual,
.\" which is licensed free of charge, as they might when working
.\" professionally.
.\"
.\" Formatted or processed versions of this manual, if unaccompanied by
.\" the source, must acknowledge the copyright and authors of this work.
.\"
.TH CPU_SET 2 2008-11-13 "Linux" "Linux Programmer's Manual"
.SH NAME
CPU_SET, CPU_CLR, CPU_ISSET, CPU_ZERO, CPU_COUNT,
CPU_AND, CPU_OR, CPU_XOR, CPU_EQUAL,
CPU_ALLOC, CPU_ALLOC_SIZE, CPU_FREE,
CPU_SET_S, CPU_CLR_S, CPU_ISSET_S, CPU_ZERO_S,
CPU_COUNT_S, CPU_AND_S, CPU_OR_S, CPU_XOR_S, CPU_EQUAL_S \-
macros for manipulating CPU sets
.SH SYNOPSIS
.nf
.B #define _GNU_SOURCE
.B #include <sched.h>
.sp
.BI "void CPU_ZERO(cpu_set_t *" set );
.sp
.BI "void CPU_SET(int " cpu ", cpu_set_t *" set );
.BI "void CPU_CLR(int " cpu ", cpu_set_t *" set );
.BI "int  CPU_ISSET(int " cpu ", cpu_set_t *" set );
.sp
.BI "void CPU_COUNT(cpu_set_t *" set );
.sp
.BI "void CPU_AND(cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.BI "void CPU_OR(cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.BI "void CPU_XOR(cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.sp
.BI "int  CPU_EQUAL(cpu_set_t *" set1 ", cpu_set_t *" set2 );
.sp
.BI "cpu_set_t *CPU_ALLOC(int " num_cpus );
.BR "void CPU_FREE(cpu_set_t *" set );
.BI "size_t CPU_ALLOC_SIZE(int " num_cpus );
.sp
.BI "void CPU_ZERO_S(size_t " setsize ", cpu_set_t *" set );
.sp
.BI "void CPU_SET_S(int " cpu ", size_t " setsize ", cpu_set_t *" set );
.BI "void CPU_CLR_S(int " cpu ", size_t " setsize ", cpu_set_t *" set );
.BI "int  CPU_ISSET_S(int " cpu ", size_t " setsize ", cpu_set_t *" set );
.sp
.BI "void CPU_COUNT_S(size_t " setsize ", cpu_set_t *" set );
.sp
.BI "void CPU_AND_S(size_t " setsize ", cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.BI "void CPU_OR_S(size_t " setsize ", cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.BI "void CPU_XOR_S(size_t " setsize ", cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.sp
.BI "int  CPU_EQUAL_S(size_t " setsize ", cpu_set_t *" set1 \
", cpu_set_t *" set2 );
.fi
.SH DESCRIPTION
The
.I cpu_set_t
data structure represents a set of CPUs.
CPU sets are used by
.BR sched_setaffinity (2)
and similar interfaces.

The following macros are provided to operate on the CPU set
.IR set :
.TP 17
.BR CPU_ZERO ()
Clears
.IR set ,
so that it contains no CPUs.
.TP
.BR CPU_SET ()
Add CPU
.I cpu
to
.IR set .
.TP
.BR CPU_CLR ()
Remove CPU
.I cpu
from
.IR set .
.TP
.BR CPU_ISSET ()
Test to see if CPU
.I cpu
is a member of
.IR set .
.TP
.BR CPU_COUNT ()
Return the number of CPUs in
.IR set .
.PP
Where a
.I cpu
argument is specified, it should not produce side effects,
since the above macros may evaluate the argument more than once.
.PP
The first available CPU on the system corresponds to a
.I cpu
value of 0, the next CPU corresponds to a
.I cpu
value of 1, and so on.
The constant
.B CPU_SETSIZE
(currently 1024) specifies a value one greater than the maximum CPU
number that can be stored in
.IR cpu_set_t .

The following macros perform logical operations on CPU sets:
.TP 17
.BR CPU_AND ()
Store the logical AND of the sets
.I srcset1
and
.I srcset2
in
.I destset
(which may be one of the source sets).
.TP
.BR CPU_OR ()
Store the logical OR of the sets
.I srcset1
and
.I srcset2
in
.I destset
(which may be one of the source sets).
.TP
.BR CPU_XOR ()
Store the logical XOR of the sets
.I srcset1
and
.I srcset2
in
.I destset
(which may be one of the source sets).
.TP
.BR CPU_EQUAL ()
Test whether two CPU set contain exactly the same CPUs.
.SS Dynamically sized CPU sets
Because some applications may require the ability to dynamically
size CPU sets (e.g., to allocate sets larger than that
defined by the standard
.I cpu_set_t
data type), glibc nowadays provides a set of macros to support this.

The following macros are used to allocate and deallocate CPU sets:
.TP 17
.BR CPU_ALLOC ()
Allocate a CPU set large enough to hold CPUs
in the range 0 to
.IR num_cpus-1 .
.TP
.BR CPU_ALLOC_SIZE ()
Return the size in bytes of the CPU set that would be needed to
hold CPUs in the range 0 to
.IR num_cpus-1 .
.\" FIXME . track this bug
.\" Currently (glibc 2.8), CPU_ALLOC_SIZE() returns double the
.\" value that it should.  This also causes CPU_ALLOC() to allocate
.\" twice as much memory as is required.
.\" http://sourceware.org/bugzilla/show_bug.cgi?id=7029
.\" Eventually, write this up in BUGS.
This macro provides the value that can be used for
.I setsize
in the
.BR CPU_*_S ()
macros described below.
.TP
.BR CPU_FREE ()
Free a CPU set previously allocated by
.BR CPU_ALLOC ().
.PP
The macros whose names end with "_S" are the analogs of
the similarly named macros without the suffix.
These macros perform the same tasks as their analogs,
but operate on the dynamically allocated CPU set(s) whose size is
.I setsize
bytes.
.SH "RETURN VALUE"
.BR CPU_ISSET ()
and
.BR CPU_ISSET_S ()
return non-zero if
.I cpu
is in
.IR set ;
otherwise, it returns 0.

.BR CPU_COUNT ()
and
.BR CPU_COUNT_S ()
return the number of CPUs in
.IR set .

.BR CPU_EQUAL ()
and
.BR CPU_EQUAL_S ()
return non-zero if the two CPU sets are equal; otherwise it returns 0.

.BR CPU_ALLOC ()
returns a pointer on success, or NULL on failure.
(Errors are as for
.BR malloc (3).)

.BR CPU_ALLOC_SIZE ()
returns the number of bytes required to store a
CPU set of the specified cardinality.

The other functions do not return a value.
.SH VERSIONS
The
.BR CPU_ZERO (),
.BR CPU_SET (),
.BR CPU_CLR (),
and
.BR CPU_ISSET ()
macros were added in glibc 2.3.3.

.BR CPU_COUNT ()
first appeared in glibc 2.6.

.BR CPU_AND (),
.BR CPU_OR (),
.BR CPU_XOR (),
.BR CPU_EQUAL (),
.BR CPU_ALLOC (),
.BR CPU_ALLOC_SIZE (),
.BR CPU_FREE (),
.BR CPU_ZERO_S (),
.BR CPU_SET_S (),
.BR CPU_CLR_S (),
.BR CPU_ISSET_S (),
.BR CPU_AND_S (),
.BR CPU_OR_S (),
.BR CPU_XOR_S (),
and
.BR CPU_EQUAL_S ()
first appeared in glibc 2.7.
.SH "CONFORMING TO"
These interfaces are Linux-specific.
.SH EXAMPLE
The following program demonstrates the use of some of the macros
used for dynamically allocated CPU sets.

.nf
#define _GNU_SOURCE
#include <sched.h>
#include <stdlib.h>
#include <unistd.h>
#include <stdio.h>
#include <assert.h>

int
main(int argc, char *argv[])
{
    cpu_set_t *cpusetp;
    size_t size;
    int num_cpus, cpu;

    if (argc < 2) {
        fprintf(stderr, "Usage: %s <num\-cpus>\\n", argv[0]);
        exit(EXIT_FAILURE);
    }

    num_cpus = atoi(argv[1]);

    cpusetp = CPU_ALLOC(num_cpus);
    if (cpusetp == NULL) {
        perror("CPU_ALLOC");
        exit(EXIT_FAILURE);
    }

    size = CPU_ALLOC_SIZE(num_cpus);

    CPU_ZERO_S(size, cpusetp);
    for (cpu = 0; cpu < num_cpus; cpu += 2)
        CPU_SET_S(cpu, size, cpusetp);

    printf("CPU_COUNT() of set:    %d\\n", CPU_COUNT_S(size, cpusetp));

    CPU_FREE(cpusetp);
    exit(EXIT_SUCCESS);
}
.fi
.SH "SEE ALSO"
.BR sched_setaffinity (2),
.BR pthread_attr_setaffinity_np (3),
.BR pthread_setaffinity_np (3),
.BR cpuset (7)
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: For review: CPU_SET.3

[Bert Wesarg]

Hi Michael,

On Thu, Nov 13, 2008 at 15:06, Michael Kerrisk
<mtk.manpages-gM/Ye1E23mwN+BqQ9rBEUg@public.gmane.org> wrote:
> Hi Bert,
>
> I created this page by removing the CPU_*() material from
> sched_setscheduler.2 and adding documentation of the many other macros
> that have been added in more recent glibc releases.
>
> I remember looking at a mail thread you pointed at a few days back
> where you said the interface of the CPU_*S() macros is somewhat
> confusing in its use of bits in some places, and bytes in others.
> It's hard to disagree with you -- so I included an example in the
> man-pages which hopefully might reduce the confusion.
>
> Of course, since these are macros, my use of types in the prototypes
> in the SYNOPSIS is in many cases invention; let me know if something
> looks weird.
>
> Also, take a look at
> http://sourceware.org/bugzilla/show_bug.cgi?id=7029.  Unless I'm
> confused (which is always possible), this is a real bug.  Do you
> agree?
"It's hard to disagree with you." I can confirm this here on my x86-64
box, a x86-32 exec returns 2048 the x86-64 1024.

>
> Thanks,
>
> Michael
>
> .\" Copyright (C) 2006 Michael Kerrisk
> .\" and Copyright (C) 2008 Linux Foundation, written by Michael Kerrisk
> .\"     <mtk.manpages-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> .\"
> .\" Permission is granted to make and distribute verbatim copies of this
> .\" manual provided the copyright notice and this permission notice are
> .\" preserved on all copies.
> .\"
> .\" Permission is granted to copy and distribute modified versions of this
> .\" manual under the conditions for verbatim copying, provided that the
> .\" entire resulting derived work is distributed under the terms of a
> .\" permission notice identical to this one.
> .\"
> .\" Since the Linux kernel and libraries are constantly changing, this
> .\" manual page may be incorrect or out-of-date.  The author(s) assume no
> .\" responsibility for errors or omissions, or for damages resulting from
> .\" the use of the information contained herein.  The author(s) may not
> .\" have taken the same level of care in the production of this manual,
> .\" which is licensed free of charge, as they might when working
> .\" professionally.
> .\"
> .\" Formatted or processed versions of this manual, if unaccompanied by
> .\" the source, must acknowledge the copyright and authors of this work.
> .\"
> .TH CPU_SET 2 2008-11-13 "Linux" "Linux Programmer's Manual"
> .SH NAME
> CPU_SET, CPU_CLR, CPU_ISSET, CPU_ZERO, CPU_COUNT,
> CPU_AND, CPU_OR, CPU_XOR, CPU_EQUAL,
> CPU_ALLOC, CPU_ALLOC_SIZE, CPU_FREE,
> CPU_SET_S, CPU_CLR_S, CPU_ISSET_S, CPU_ZERO_S,
> CPU_COUNT_S, CPU_AND_S, CPU_OR_S, CPU_XOR_S, CPU_EQUAL_S \-
> macros for manipulating CPU sets
> .SH SYNOPSIS
> .nf
> .B #define _GNU_SOURCE
> .B #include <sched.h>
> .sp
> .BI "void CPU_ZERO(cpu_set_t *" set );
> .sp
> .BI "void CPU_SET(int " cpu ", cpu_set_t *" set );
> .BI "void CPU_CLR(int " cpu ", cpu_set_t *" set );
> .BI "int  CPU_ISSET(int " cpu ", cpu_set_t *" set );
> .sp
> .BI "void CPU_COUNT(cpu_set_t *" set );
> .sp
> .BI "void CPU_AND(cpu_set_t *" destset ,
> .BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
> .BI "void CPU_OR(cpu_set_t *" destset ,
> .BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
> .BI "void CPU_XOR(cpu_set_t *" destset ,
> .BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
> .sp
> .BI "int  CPU_EQUAL(cpu_set_t *" set1 ", cpu_set_t *" set2 );
> .sp
> .BI "cpu_set_t *CPU_ALLOC(int " num_cpus );
> .BR "void CPU_FREE(cpu_set_t *" set );
> .BI "size_t CPU_ALLOC_SIZE(int " num_cpus );
> .sp
> .BI "void CPU_ZERO_S(size_t " setsize ", cpu_set_t *" set );
> .sp
> .BI "void CPU_SET_S(int " cpu ", size_t " setsize ", cpu_set_t *" set );
> .BI "void CPU_CLR_S(int " cpu ", size_t " setsize ", cpu_set_t *" set );
> .BI "int  CPU_ISSET_S(int " cpu ", size_t " setsize ", cpu_set_t *" set );
> .sp
> .BI "void CPU_COUNT_S(size_t " setsize ", cpu_set_t *" set );
> .sp
> .BI "void CPU_AND_S(size_t " setsize ", cpu_set_t *" destset ,
> .BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
> .BI "void CPU_OR_S(size_t " setsize ", cpu_set_t *" destset ,
> .BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
> .BI "void CPU_XOR_S(size_t " setsize ", cpu_set_t *" destset ,
> .BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
> .sp
> .BI "int  CPU_EQUAL_S(size_t " setsize ", cpu_set_t *" set1 \
> ", cpu_set_t *" set2 );
Even I like to see the equivalent C prototypes, I think this should be
noted earlier that these are actually macros.

Also, I would like to see a note that the cpu_set_t type is opaque (as
from the glibc manual) but there is no official copy operator. The
glibc developer suggest using memcpy for this (see the reminder of
this thread I posted). And you should not use '=' because of the
dynamically allocated ones.

Last but not least, because CPU_ALLOC{,_SIZE} rounds the num_cpus
argument up (i.e. CPU_ALLOC_SIZE(1) == sizeof(unsigned long)) there
are probably more cpus possible in this set than you requested, so
these blind bits should be mentioned as undefined.

Now to the naming confusion, I think the main problem is CPU_SETSIZE
and the setsize arguement to the CPU_*_S macros. The former is in bits
the latter in bytes. Either put a note to CPU_SETSIZE that this is in
bits and you should not pass this as an setsize argument to any
CPU_*_S macro, because of the similar name. Or, rename the setsize
argument to something other, and use the setsize name for the
CPU_ALLOC{,_SIZE}, too.

I think option one should suffice.

> .fi
> .SH DESCRIPTION
> The
> .I cpu_set_t
> data structure represents a set of CPUs.
> CPU sets are used by
> .BR sched_setaffinity (2)
> and similar interfaces.
>
> The following macros are provided to operate on the CPU set
> .IR set :
> .TP 17
> .BR CPU_ZERO ()
> Clears
> .IR set ,
> so that it contains no CPUs.
> .TP
> .BR CPU_SET ()
> Add CPU
> .I cpu
> to
> .IR set .
> .TP
> .BR CPU_CLR ()
> Remove CPU
> .I cpu
> from
> .IR set .
> .TP
> .BR CPU_ISSET ()
> Test to see if CPU
> .I cpu
> is a member of
> .IR set .
> .TP
> .BR CPU_COUNT ()
> Return the number of CPUs in
> .IR set .
> .PP
> Where a
> .I cpu
> argument is specified, it should not produce side effects,
> since the above macros may evaluate the argument more than once.
> .PP
> The first available CPU on the system corresponds to a
> .I cpu
> value of 0, the next CPU corresponds to a
> .I cpu
> value of 1, and so on.
> The constant
> .B CPU_SETSIZE
> (currently 1024) specifies a value one greater than the maximum CPU
> number that can be stored in
> .IR cpu_set_t .
>
> The following macros perform logical operations on CPU sets:
> .TP 17
> .BR CPU_AND ()
> Store the logical AND of the sets
> .I srcset1
> and
> .I srcset2
> in
> .I destset
> (which may be one of the source sets).
> .TP
> .BR CPU_OR ()
> Store the logical OR of the sets
> .I srcset1
> and
> .I srcset2
> in
> .I destset
> (which may be one of the source sets).
> .TP
> .BR CPU_XOR ()
> Store the logical XOR of the sets
> .I srcset1
> and
> .I srcset2
> in
> .I destset
> (which may be one of the source sets).
> .TP
> .BR CPU_EQUAL ()
> Test whether two CPU set contain exactly the same CPUs.
> .SS Dynamically sized CPU sets
> Because some applications may require the ability to dynamically
> size CPU sets (e.g., to allocate sets larger than that
> defined by the standard
> .I cpu_set_t
> data type), glibc nowadays provides a set of macros to support this.
>
> The following macros are used to allocate and deallocate CPU sets:
> .TP 17
> .BR CPU_ALLOC ()
> Allocate a CPU set large enough to hold CPUs
> in the range 0 to
> .IR num_cpus-1 .
> .TP
> .BR CPU_ALLOC_SIZE ()
> Return the size in bytes of the CPU set that would be needed to
> hold CPUs in the range 0 to
> .IR num_cpus-1 .
> .\" FIXME . track this bug
> .\" Currently (glibc 2.8), CPU_ALLOC_SIZE() returns double the
> .\" value that it should.  This also causes CPU_ALLOC() to allocate
> .\" twice as much memory as is required.
> .\" http://sourceware.org/bugzilla/show_bug.cgi?id=7029
> .\" Eventually, write this up in BUGS.
> This macro provides the value that can be used for
the
> .I setsize
argument
> in the
> .BR CPU_*_S ()
> macros described below.
> .TP
> .BR CPU_FREE ()
> Free a CPU set previously allocated by
> .BR CPU_ALLOC ().
> .PP
> The macros whose names end with "_S" are the analogs of
> the similarly named macros without the suffix.
> These macros perform the same tasks as their analogs,
> but operate on the dynamically allocated CPU set(s) whose size is
> .I setsize
> bytes.
> .SH "RETURN VALUE"
> .BR CPU_ISSET ()
> and
> .BR CPU_ISSET_S ()
> return non-zero if
> .I cpu
> is in
> .IR set ;
> otherwise, it returns 0.
>
> .BR CPU_COUNT ()
> and
> .BR CPU_COUNT_S ()
> return the number of CPUs in
> .IR set .
>
> .BR CPU_EQUAL ()
> and
> .BR CPU_EQUAL_S ()
> return non-zero if the two CPU sets are equal; otherwise it returns 0.
>
> .BR CPU_ALLOC ()
> returns a pointer on success, or NULL on failure.
> (Errors are as for
> .BR malloc (3).)
>
> .BR CPU_ALLOC_SIZE ()
> returns the number of bytes required to store a
> CPU set of the specified cardinality.
>
> The other functions do not return a value.
> .SH VERSIONS
> The
> .BR CPU_ZERO (),
> .BR CPU_SET (),
> .BR CPU_CLR (),
> and
> .BR CPU_ISSET ()
> macros were added in glibc 2.3.3.
>
> .BR CPU_COUNT ()
> first appeared in glibc 2.6.
>
> .BR CPU_AND (),
> .BR CPU_OR (),
> .BR CPU_XOR (),
> .BR CPU_EQUAL (),
> .BR CPU_ALLOC (),
> .BR CPU_ALLOC_SIZE (),
> .BR CPU_FREE (),
> .BR CPU_ZERO_S (),
> .BR CPU_SET_S (),
> .BR CPU_CLR_S (),
> .BR CPU_ISSET_S (),
> .BR CPU_AND_S (),
> .BR CPU_OR_S (),
> .BR CPU_XOR_S (),
> and
> .BR CPU_EQUAL_S ()
> first appeared in glibc 2.7.
> .SH "CONFORMING TO"
> These interfaces are Linux-specific.
> .SH EXAMPLE
> The following program demonstrates the use of some of the macros
> used for dynamically allocated CPU sets.
>
> .nf
> #define _GNU_SOURCE
> #include <sched.h>
> #include <stdlib.h>
> #include <unistd.h>
> #include <stdio.h>
> #include <assert.h>
>
> int
> main(int argc, char *argv[])
> {
>    cpu_set_t *cpusetp;
>    size_t size;
>    int num_cpus, cpu;
>
>    if (argc < 2) {
>        fprintf(stderr, "Usage: %s <num\-cpus>\\n", argv[0]);
>        exit(EXIT_FAILURE);
>    }
>
>    num_cpus = atoi(argv[1]);
>
>    cpusetp = CPU_ALLOC(num_cpus);
>    if (cpusetp == NULL) {
>        perror("CPU_ALLOC");
>        exit(EXIT_FAILURE);
>    }
>
>    size = CPU_ALLOC_SIZE(num_cpus);
>
>    CPU_ZERO_S(size, cpusetp);
>    for (cpu = 0; cpu < num_cpus; cpu += 2)
>        CPU_SET_S(cpu, size, cpusetp);
>
>    printf("CPU_COUNT() of set:    %d\\n", CPU_COUNT_S(size, cpusetp));
>
>    CPU_FREE(cpusetp);
>    exit(EXIT_SUCCESS);
> }
> .fi
> .SH "SEE ALSO"
> .BR sched_setaffinity (2),
> .BR pthread_attr_setaffinity_np (3),
> .BR pthread_setaffinity_np (3),
> .BR cpuset (7)
>
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: For review: CPU_SET.3

[Michael Kerrisk]

Hi Bert,

Thanks for looking this over.

On Thu, Nov 13, 2008 at 1:45 PM, Bert Wesarg <bert.wesarg-gM/Ye1E23mwN+BqQ9rBEUg@public.gmane.org> wrote:
> Hi Michael,
>
> On Thu, Nov 13, 2008 at 15:06, Michael Kerrisk
> <mtk.manpages-gM/Ye1E23mwN+BqQ9rBEUg@public.gmane.org> wrote:
>> Hi Bert,
>>
>> I created this page by removing the CPU_*() material from
>> sched_setscheduler.2 and adding documentation of the many other macros
>> that have been added in more recent glibc releases.
>>
>> I remember looking at a mail thread you pointed at a few days back
>> where you said the interface of the CPU_*S() macros is somewhat
>> confusing in its use of bits in some places, and bytes in others.
>> It's hard to disagree with you -- so I included an example in the
>> man-pages which hopefully might reduce the confusion.
>>
>> Of course, since these are macros, my use of types in the prototypes
>> in the SYNOPSIS is in many cases invention; let me know if something
>> looks weird.
>>
>> Also, take a look at
>> http://sourceware.org/bugzilla/show_bug.cgi?id=7029.  Unless I'm
>> confused (which is always possible), this is a real bug.  Do you
>> agree?
> "It's hard to disagree with you." I can confirm this here on my x86-64
> box, a x86-32 exec returns 2048 the x86-64 1024.

Yep.  In fact UD has already fixed it today.

[...]

>> .BI "int  CPU_EQUAL_S(size_t " setsize ", cpu_set_t *" set1 \
>> ", cpu_set_t *" set2 );
> Even I like to see the equivalent C prototypes, I think this should be
> noted earlier that these are actually macros.

Well, I do mention that they are macros a number of times.  What more
do you think I should add?

> Also, I would like to see a note that the cpu_set_t type is opaque (as
> from the glibc manual)

I added something on that.

> but there is no official copy operator. The
> glibc developer suggest using memcpy for this (see the reminder of
> this thread I posted).

Yes, good point.  I Added a NOTES section mentioning memcpy(3).

> And you should not use '=' because of the
> dynamically allocated ones.
>
> Last but not least, because CPU_ALLOC{,_SIZE} rounds the num_cpus
> argument up (i.e. CPU_ALLOC_SIZE(1) == sizeof(unsigned long)) there
> are probably more cpus possible in this set than you requested, so
> these blind bits should be mentioned as undefined.

I added something in NOTES about this.

> Now to the naming confusion, I think the main problem is CPU_SETSIZE
> and the setsize arguement to the CPU_*_S macros. The former is in bits
> the latter in bytes. Either put a note to CPU_SETSIZE that this is in
> bits and you should not pass this as an setsize argument to any
> CPU_*_S macro, because of the similar name. Or, rename the setsize
> argument to something other, and use the setsize name for the
> CPU_ALLOC{,_SIZE}, too.
>
> I think option one should suffice.

Again, I added something to NOTES on this.

[...]

>> This macro provides the value that can be used for
> the
>> .I setsize
> argument

Done.

Thanks Bert.


-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
git://git.kernel.org/pub/scm/docs/man-pages/man-pages.git
man-pages online: http://www.kernel.org/doc/man-pages/online_pages.html
Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: For review: CPU_SET.3

[Michael Kerrisk]

And this is the revised page.

Cheers,

Michael

.\" Copyright (C) 2006 Michael Kerrisk
.\" and Copyright (C) 2008 Linux Foundation, written by Michael Kerrisk
.\"     <mtk.manpages-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
.\"
.\" Permission is granted to make and distribute verbatim copies of this
.\" manual provided the copyright notice and this permission notice are
.\" preserved on all copies.
.\"
.\" Permission is granted to copy and distribute modified versions of this
.\" manual under the conditions for verbatim copying, provided that the
.\" entire resulting derived work is distributed under the terms of a
.\" permission notice identical to this one.
.\"
.\" Since the Linux kernel and libraries are constantly changing, this
.\" manual page may be incorrect or out-of-date.  The author(s) assume no
.\" responsibility for errors or omissions, or for damages resulting from
.\" the use of the information contained herein.  The author(s) may not
.\" have taken the same level of care in the production of this manual,
.\" which is licensed free of charge, as they might when working
.\" professionally.
.\"
.\" Formatted or processed versions of this manual, if unaccompanied by
.\" the source, must acknowledge the copyright and authors of this work.
.\"
.TH CPU_SET 2 2008-11-13 "Linux" "Linux Programmer's Manual"
.SH NAME
CPU_SET, CPU_CLR, CPU_ISSET, CPU_ZERO, CPU_COUNT,
CPU_AND, CPU_OR, CPU_XOR, CPU_EQUAL,
CPU_ALLOC, CPU_ALLOC_SIZE, CPU_FREE,
CPU_SET_S, CPU_CLR_S, CPU_ISSET_S, CPU_ZERO_S,
CPU_COUNT_S, CPU_AND_S, CPU_OR_S, CPU_XOR_S, CPU_EQUAL_S \-
macros for manipulating CPU sets
.SH SYNOPSIS
.nf
.B #define _GNU_SOURCE
.B #include <sched.h>
.sp
.BI "void CPU_ZERO(cpu_set_t *" set );
.sp
.BI "void CPU_SET(int " cpu ", cpu_set_t *" set );
.BI "void CPU_CLR(int " cpu ", cpu_set_t *" set );
.BI "int  CPU_ISSET(int " cpu ", cpu_set_t *" set );
.sp
.BI "void CPU_COUNT(cpu_set_t *" set );
.sp
.BI "void CPU_AND(cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.BI "void CPU_OR(cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.BI "void CPU_XOR(cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.sp
.BI "int  CPU_EQUAL(cpu_set_t *" set1 ", cpu_set_t *" set2 );
.sp
.BI "cpu_set_t *CPU_ALLOC(int " num_cpus );
.BR "void CPU_FREE(cpu_set_t *" set );
.BI "size_t CPU_ALLOC_SIZE(int " num_cpus );
.sp
.BI "void CPU_ZERO_S(size_t " setsize ", cpu_set_t *" set );
.sp
.BI "void CPU_SET_S(int " cpu ", size_t " setsize ", cpu_set_t *" set );
.BI "void CPU_CLR_S(int " cpu ", size_t " setsize ", cpu_set_t *" set );
.BI "int  CPU_ISSET_S(int " cpu ", size_t " setsize ", cpu_set_t *" set );
.sp
.BI "void CPU_COUNT_S(size_t " setsize ", cpu_set_t *" set );
.sp
.BI "void CPU_AND_S(size_t " setsize ", cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.BI "void CPU_OR_S(size_t " setsize ", cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.BI "void CPU_XOR_S(size_t " setsize ", cpu_set_t *" destset ,
.BI "             cpu_set_t *" srcset1 ", cpu_set_t *" srcset2 );
.sp
.BI "int  CPU_EQUAL_S(size_t " setsize ", cpu_set_t *" set1 \
", cpu_set_t *" set2 );
.fi
.SH DESCRIPTION
The
.I cpu_set_t
data structure represents a set of CPUs.
CPU sets are used by
.BR sched_setaffinity (2)
and similar interfaces.

The
.I cpu_set_t
data type is implemented as a bitset.
However, the data structure treated as considered opaque:
all manipulation of CPU sets should be done via the macros
described in this page.

The following macros are provided to operate on the CPU set
.IR set :
.TP 17
.BR CPU_ZERO ()
Clears
.IR set ,
so that it contains no CPUs.
.TP
.BR CPU_SET ()
Add CPU
.I cpu
to
.IR set .
.TP
.BR CPU_CLR ()
Remove CPU
.I cpu
from
.IR set .
.TP
.BR CPU_ISSET ()
Test to see if CPU
.I cpu
is a member of
.IR set .
.TP
.BR CPU_COUNT ()
Return the number of CPUs in
.IR set .
.PP
Where a
.I cpu
argument is specified, it should not produce side effects,
since the above macros may evaluate the argument more than once.
.PP
The first available CPU on the system corresponds to a
.I cpu
value of 0, the next CPU corresponds to a
.I cpu
value of 1, and so on.
The constant
.B CPU_SETSIZE
(currently 1024) specifies a value one greater than the maximum CPU
number that can be stored in
.IR cpu_set_t .

The following macros perform logical operations on CPU sets:
.TP 17
.BR CPU_AND ()
Store the logical AND of the sets
.I srcset1
and
.I srcset2
in
.I destset
(which may be one of the source sets).
.TP
.BR CPU_OR ()
Store the logical OR of the sets
.I srcset1
and
.I srcset2
in
.I destset
(which may be one of the source sets).
.TP
.BR CPU_XOR ()
Store the logical XOR of the sets
.I srcset1
and
.I srcset2
in
.I destset
(which may be one of the source sets).
.TP
.BR CPU_EQUAL ()
Test whether two CPU set contain exactly the same CPUs.
.SS Dynamically sized CPU sets
Because some applications may require the ability to dynamically
size CPU sets (e.g., to allocate sets larger than that
defined by the standard
.I cpu_set_t
data type), glibc nowadays provides a set of macros to support this.

The following macros are used to allocate and deallocate CPU sets:
.TP 17
.BR CPU_ALLOC ()
Allocate a CPU set large enough to hold CPUs
in the range 0 to
.IR num_cpus-1 .
.TP
.BR CPU_ALLOC_SIZE ()
Return the size in bytes of the CPU set that would be needed to
hold CPUs in the range 0 to
.IR num_cpus-1 .
This macro provides the value that can be used for the
.I setsize
argument in the
.BR CPU_*_S ()
macros described below.
.TP
.BR CPU_FREE ()
Free a CPU set previously allocated by
.BR CPU_ALLOC ().
.PP
The macros whose names end with "_S" are the analogs of
the similarly named macros without the suffix.
These macros perform the same tasks as their analogs,
but operate on the dynamically allocated CPU set(s) whose size is
.I setsize
bytes.
.SH "RETURN VALUE"
.BR CPU_ISSET ()
and
.BR CPU_ISSET_S ()
return non-zero if
.I cpu
is in
.IR set ;
otherwise, it returns 0.

.BR CPU_COUNT ()
and
.BR CPU_COUNT_S ()
return the number of CPUs in
.IR set .

.BR CPU_EQUAL ()
and
.BR CPU_EQUAL_S ()
return non-zero if the two CPU sets are equal; otherwise it returns 0.

.BR CPU_ALLOC ()
returns a pointer on success, or NULL on failure.
(Errors are as for
.BR malloc (3).)

.BR CPU_ALLOC_SIZE ()
returns the number of bytes required to store a
CPU set of the specified cardinality.

The other functions do not return a value.
.SH VERSIONS
The
.BR CPU_ZERO (),
.BR CPU_SET (),
.BR CPU_CLR (),
and
.BR CPU_ISSET ()
macros were added in glibc 2.3.3.

.BR CPU_COUNT ()
first appeared in glibc 2.6.

.BR CPU_AND (),
.BR CPU_OR (),
.BR CPU_XOR (),
.BR CPU_EQUAL (),
.BR CPU_ALLOC (),
.BR CPU_ALLOC_SIZE (),
.BR CPU_FREE (),
.BR CPU_ZERO_S (),
.BR CPU_SET_S (),
.BR CPU_CLR_S (),
.BR CPU_ISSET_S (),
.BR CPU_AND_S (),
.BR CPU_OR_S (),
.BR CPU_XOR_S (),
and
.BR CPU_EQUAL_S ()
first appeared in glibc 2.7.
.SH "CONFORMING TO"
These interfaces are Linux-specific.
.SH NOTES
To duplicate a CPU set, use
.BR memcpy (3).

Since CPU sets are bitsets allocated in units of long words,
the actual number of CPUs in a dynamically
allocated CPU set will be rounded up to the next multiple of
.IR "sizeof(unsigned long)" .
An application should consider the contents on these extra bits
to be undefined.

Note that the constant
.B CPU_SETSIZE
indicated the number of CPUs in the
.I cpu_set_t
data type (thus, it is effectively a count of bits in the bitset),
while the
.I setsize
argument of the
.BR CPU_*_S ()
macros is a size in bytes.
.SH EXAMPLE
The following program demonstrates the use of some of the macros
used for dynamically allocated CPU sets.

.nf
#define _GNU_SOURCE
#include <sched.h>
#include <stdlib.h>
#include <unistd.h>
#include <stdio.h>
#include <assert.h>

int
main(int argc, char *argv[])
{
    cpu_set_t *cpusetp;
    size_t size;
    int num_cpus, cpu;

    if (argc < 2) {
        fprintf(stderr, "Usage: %s <num\-cpus>\\n", argv[0]);
        exit(EXIT_FAILURE);
    }

    num_cpus = atoi(argv[1]);

    cpusetp = CPU_ALLOC(num_cpus);
    if (cpusetp == NULL) {
        perror("CPU_ALLOC");
        exit(EXIT_FAILURE);
    }

    size = CPU_ALLOC_SIZE(num_cpus);

    CPU_ZERO_S(size, cpusetp);
    for (cpu = 0; cpu < num_cpus; cpu += 2)
        CPU_SET_S(cpu, size, cpusetp);

    printf("CPU_COUNT() of set:    %d\\n", CPU_COUNT_S(size, cpusetp));

    CPU_FREE(cpusetp);
    exit(EXIT_SUCCESS);
}
.fi
.SH BUGS
On 32-bit platforms with glibc 2.8 and earlier,
.BR CPU_ALLOC ()
allocates twice as much space as is required, and
.BR CPU_ALLOC_SIZE ()
returns a value twice as large as it should.
The only consequences for most programs is wasted space,
and slightly less efficient operation of the macros that
operate on dynamically allocated CPU sets.
These bugs are fixed in glibc 2.9.
.\" http://sourceware.org/bugzilla/show_bug.cgi?id=7029
.SH "SEE ALSO"
.BR sched_setaffinity (2),
.BR pthread_attr_setaffinity_np (3),
.BR pthread_setaffinity_np (3),
.BR cpuset (7)
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: For review: CPU_SET.3

[Bert Wesarg]

On Fri, Nov 14, 2008 at 04:08, Michael Kerrisk
<mtk.manpages-gM/Ye1E23mwN+BqQ9rBEUg@public.gmane.org> wrote:
> Hi Bert,
>
> Thanks for looking this over.
>
> On Thu, Nov 13, 2008 at 1:45 PM, Bert Wesarg <bert.wesarg-gM/Ye1E23mwN+BqQ9rBEUg@public.gmane.org> wrote:
>> Hi Michael,
>>
>> On Thu, Nov 13, 2008 at 15:06, Michael Kerrisk
>> <mtk.manpages-gM/Ye1E23mwN+BqQ9rBEUg@public.gmane.org> wrote:
>>> Hi Bert,
>>>
>>> I created this page by removing the CPU_*() material from
>>> sched_setscheduler.2 and adding documentation of the many other macros
>>> that have been added in more recent glibc releases.
>>>
>>> I remember looking at a mail thread you pointed at a few days back
>>> where you said the interface of the CPU_*S() macros is somewhat
>>> confusing in its use of bits in some places, and bytes in others.
>>> It's hard to disagree with you -- so I included an example in the
>>> man-pages which hopefully might reduce the confusion.
>>>
>>> Of course, since these are macros, my use of types in the prototypes
>>> in the SYNOPSIS is in many cases invention; let me know if something
>>> looks weird.
>>>
>>> Also, take a look at
>>> http://sourceware.org/bugzilla/show_bug.cgi?id=7029.  Unless I'm
>>> confused (which is always possible), this is a real bug.  Do you
>>> agree?
>> "It's hard to disagree with you." I can confirm this here on my x86-64
>> box, a x86-32 exec returns 2048 the x86-64 1024.
>
> Yep.  In fact UD has already fixed it today.
>
> [...]
>
>>> .BI "int  CPU_EQUAL_S(size_t " setsize ", cpu_set_t *" set1 \
>>> ", cpu_set_t *" set2 );
>> Even I like to see the equivalent C prototypes, I think this should be
>> noted earlier that these are actually macros.
>
> Well, I do mention that they are macros a number of times.  What more
> do you think I should add?
Hmm, I thought something like "hey these are only theoretical C prototypes"

>
>> Also, I would like to see a note that the cpu_set_t type is opaque (as
>> from the glibc manual)
>
> I added something on that.
>
>> but there is no official copy operator. The
>> glibc developer suggest using memcpy for this (see the reminder of
>> this thread I posted).
>
> Yes, good point.  I Added a NOTES section mentioning memcpy(3).
>
>> And you should not use '=' because of the
>> dynamically allocated ones.
>>
>> Last but not least, because CPU_ALLOC{,_SIZE} rounds the num_cpus
>> argument up (i.e. CPU_ALLOC_SIZE(1) == sizeof(unsigned long)) there
>> are probably more cpus possible in this set than you requested, so
>> these blind bits should be mentioned as undefined.
>
> I added something in NOTES about this.
>
>> Now to the naming confusion, I think the main problem is CPU_SETSIZE
>> and the setsize arguement to the CPU_*_S macros. The former is in bits
>> the latter in bytes. Either put a note to CPU_SETSIZE that this is in
>> bits and you should not pass this as an setsize argument to any
>> CPU_*_S macro, because of the similar name. Or, rename the setsize
>> argument to something other, and use the setsize name for the
>> CPU_ALLOC{,_SIZE}, too.
>>
>> I think option one should suffice.
>
> Again, I added something to NOTES on this.
>
> [...]
>
>>> This macro provides the value that can be used for
>> the
>>> .I setsize
>> argument
>
> Done.
>
> Thanks Bert.
New version looks good.

Thank You.

Bert
>
>
> --
> Michael Kerrisk
> Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
> git://git.kernel.org/pub/scm/docs/man-pages/man-pages.git
> man-pages online: http://www.kernel.org/doc/man-pages/online_pages.html
> Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html
>
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: For review: CPU_SET.3

[Michael Kerrisk]

Hi Bert,

[...]

>>> Even I like to see the equivalent C prototypes, I think this should be
>>> noted earlier that these are actually macros.
>>
>> Well, I do mention that they are macros a number of times.  What more
>> do you think I should add?
> Hmm, I thought something like "hey these are only theoretical C prototypes"

How about this text, which I added to notes:

==
The data types for arguments and return values shown
in the SYNOPSIS are hints what about is expected in each case.
However, since these these interfaces are implemented as macros,
the compiler won't necessarily catch all type errors
if you violate the suggestions.
==

> New version looks good.

Thanks for looking it over.

Cheers,

Michael


-- 
Michael Kerrisk
Linux man-pages maintainer; http://www.kernel.org/doc/man-pages/
git://git.kernel.org/pub/scm/docs/man-pages/man-pages.git
man-pages online: http://www.kernel.org/doc/man-pages/online_pages.html
Found a bug? http://www.kernel.org/doc/man-pages/reporting_bugs.html
--
To unsubscribe from this list: send the line "unsubscribe linux-man" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html