Re: rt-tests, cyclictest libnuma and isolcpus

Re: rt-tests, cyclictest libnuma and isolcpus

[Clark Williams]

[-- Attachment #1: Type: text/plain, Size: 813 bytes --]

On Mon, 27 Apr 2015 17:32:22 +0200
Henning Schild <henning.schild@siemens.com> wrote:

> Hi,
> 
> i did not quickly find out which list to send this mail to, please cc if
> there is one.
> 
> The current cyclictest from git contains a compatibilty hack for older
> versions of libnuma, specifically the one that debian shipped. That
> hack breaks the -a option on a system that uses isolcpus. Have a look
> at src/cyclictest/rt_numa.h around line 90.
> 
> The HAVE_PARSE_CPUSTRING_ALL-case should be the default, especially now
> that even debian has a more recent version of the numa library.
> 
> Henning

I've CC'd my co-maintainer (John Kacur) and the linux-rt list, which is
where we argue^wdiscuss changes to cyclictest.

Thanks for the info, we'll look at removing the hack.

Clark

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

[PATCH] cyclictest: remove compatibilty hack for old libnuma, repairs "-a" for isolcpus

[Henning Schild]

Recent versions of libnuma >=2.0.9 all have numa_parse_cpustring_all(). Not
using this function will make the parser fail if isolcpus are used.

Signed-off-by: Henning Schild <henning.schild@siemens.com>
---
 src/cyclictest/rt_numa.h | 10 ----------
 1 file changed, 10 deletions(-)

diff --git a/src/cyclictest/rt_numa.h b/src/cyclictest/rt_numa.h
index 06c9420..64516f5 100644
--- a/src/cyclictest/rt_numa.h
+++ b/src/cyclictest/rt_numa.h
@@ -92,17 +92,7 @@ static inline unsigned int rt_numa_bitmask_isbitset( const struct bitmask *mask,
 static inline struct bitmask* rt_numa_parse_cpustring(const char* s,
 	int max_cpus)
 {
-#ifdef HAVE_PARSE_CPUSTRING_ALL		/* Currently not defined anywhere.  No
-					   autotools build. */
 	return numa_parse_cpustring_all(s);
-#else
-	/* We really need numa_parse_cpustring_all(), so we can assign threads
-	 * to cores which are part of an isolcpus set, but early 2.x versions of
-	 * libnuma do not have this function.  A work around should be to run
-	 * your command with e.g. taskset -c 9-15 <command>
-	 */
-	return numa_parse_cpustring((char *)s);
-#endif
 }
 
 static inline void rt_bitmask_free(struct bitmask *mask)
-- 
2.3.6

Re: [PATCH] cyclictest: remove compatibilty hack for old libnuma, repairs "-a" for isolcpus

[John Kacur]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 1842 bytes --]



On Wed, 20 May 2015, Henning Schild wrote:

> Recent versions of libnuma >=2.0.9 all have numa_parse_cpustring_all(). Not
> using this function will make the parser fail if isolcpus are used.
> 
> Signed-off-by: Henning Schild <henning.schild@siemens.com>
> ---
>  src/cyclictest/rt_numa.h | 10 ----------
>  1 file changed, 10 deletions(-)
> 
> diff --git a/src/cyclictest/rt_numa.h b/src/cyclictest/rt_numa.h
> index 06c9420..64516f5 100644
> --- a/src/cyclictest/rt_numa.h
> +++ b/src/cyclictest/rt_numa.h
> @@ -92,17 +92,7 @@ static inline unsigned int rt_numa_bitmask_isbitset( const struct bitmask *mask,
>  static inline struct bitmask* rt_numa_parse_cpustring(const char* s,
>  	int max_cpus)
>  {
> -#ifdef HAVE_PARSE_CPUSTRING_ALL		/* Currently not defined anywhere.  No
> -					   autotools build. */
>  	return numa_parse_cpustring_all(s);
> -#else
> -	/* We really need numa_parse_cpustring_all(), so we can assign threads
> -	 * to cores which are part of an isolcpus set, but early 2.x versions of
> -	 * libnuma do not have this function.  A work around should be to run
> -	 * your command with e.g. taskset -c 9-15 <command>
> -	 */
> -	return numa_parse_cpustring((char *)s);
> -#endif
>  }
>  
>  static inline void rt_bitmask_free(struct bitmask *mask)
> -- 
> 2.3.6
> 

This is a terrible hack, but I'm afraid to remove it yet.
Even systems as recent as Fedora 19 have a libnuma less than that.
pm -qf /usr/lib64/libnuma.so
numactl-devel-2.0.8-4.fc19.x86_64
[jkacur@riemann rt-tests]$ cat /etc/fedora-release 
Fedora release 19 (Schrödinger’s Cat)

This is causing problems for people who want to pin threads to 
CPUs isolated with isolcpus=, so we do need to do something about.

I could perhaps accept an improved hack that tests whether the hack itself 
is needed, until we have a better solution.

Thanks

John

Re: [PATCH] cyclictest: remove compatibilty hack for old libnuma, repairs "-a" for isolcpus

[Thomas Gleixner]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 742 bytes --]

On Wed, 20 May 2015, John Kacur wrote:
> This is a terrible hack, but I'm afraid to remove it yet.
> Even systems as recent as Fedora 19 have a libnuma less than that.
> pm -qf /usr/lib64/libnuma.so
> numactl-devel-2.0.8-4.fc19.x86_64
> [jkacur@riemann rt-tests]$ cat /etc/fedora-release 
> Fedora release 19 (Schrödinger’s Cat)
> 
> This is causing problems for people who want to pin threads to 
> CPUs isolated with isolcpus=, so we do need to do something about.
> 
> I could perhaps accept an improved hack that tests whether the hack itself 
> is needed, until we have a better solution.

I'm not a great fan of it, but I think that rt-tests really needs
proper automake with proper tests for various bits and pieces.

Thanks,

	tglx

Re: [PATCH] cyclictest: remove compatibilty hack for old libnuma, repairs "-a" for isolcpus

[Henning Schild]

On Wed, 20 May 2015 16:38:25 +0200
John Kacur <jkacur@redhat.com> wrote:

> This is a terrible hack, but I'm afraid to remove it yet.
> Even systems as recent as Fedora 19 have a libnuma less than that.
> pm -qf /usr/lib64/libnuma.so
> numactl-devel-2.0.8-4.fc19.x86_64
> [jkacur@riemann rt-tests]$ cat /etc/fedora-release 
> Fedora release 19 (Schrödinger’s Cat)

According to https://fedoraproject.org/wiki/Releases that is an old
unsupported release, so i would not call it recent. No funny state
inbetween, that cat is dead. As far as i can see Fedora 20+ should have
2.0.9+, which is fine.
If you ask me, a compatibilty hack that brakes all modern systems is not
acceptable.

I do not know how many bits and pieces could be tuned, but automake for
that one issue seems overkill.

Henning
--
To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: [PATCH] cyclictest: remove compatibilty hack for old libnuma, repairs "-a" for isolcpus

[John Kacur]

[-- Attachment #1: Type: TEXT/PLAIN, Size: 1091 bytes --]



On Thu, 21 May 2015, John Kacur wrote:

> Sigh, okay, I'll look into that then.
> 
> On 20 May 2015 at 21:54, Thomas Gleixner <tglx@linutronix.de> wrote:
>       On Wed, 20 May 2015, John Kacur wrote:
>       > This is a terrible hack, but I'm afraid to remove it yet.
>       > Even systems as recent as Fedora 19 have a libnuma less than that.
>       > pm -qf /usr/lib64/libnuma.so
>       > numactl-devel-2.0.8-4.fc19.x86_64
>       > [jkacur@riemann rt-tests]$ cat /etc/fedora-release
>       > Fedora release 19 (Schrödinger’s Cat)
>       >
>       > This is causing problems for people who want to pin threads to
>       > CPUs isolated with isolcpus=, so we do need to do something about.
>       >
>       > I could perhaps accept an improved hack that tests whether the hack itself
>       > is needed, until we have a better solution.
> 
>       I'm not a great fan of it, but I think that rt-tests really needs
>       proper automake with proper tests for various bits and pieces.
> 
>       Thanks,
> 
>               tglx

Sigh, okay, I'll look into that then.

John

Re: [PATCH] cyclictest: remove compatibilty hack for old libnuma, repairs "-a" for isolcpus

[Staffan Tjernstrom]

On Thu, 2015-05-21 at 13:11 +0200, Henning Schild wrote:
> On Wed, 20 May 2015 16:38:25 +0200
> John Kacur <jkacur@redhat.com> wrote:
> 
> > This is a terrible hack, but I'm afraid to remove it yet.
> > Even systems as recent as Fedora 19 have a libnuma less than that.
> > pm -qf /usr/lib64/libnuma.so
> > numactl-devel-2.0.8-4.fc19.x86_64
> > [jkacur@riemann rt-tests]$ cat /etc/fedora-release 
> > Fedora release 19 (Schrödinger’s Cat)
> 
> According to https://fedoraproject.org/wiki/Releases that is an old
> unsupported release, so i would not call it recent. No funny state
> inbetween, that cat is dead. As far as i can see Fedora 20+ should have
> 2.0.9+, which is fine.
> If you ask me, a compatibilty hack that brakes all modern systems is not
> acceptable.
> 
> I do not know how many bits and pieces could be tuned, but automake for
> that one issue seems overkill.
> 
> Henning
> --
> To unsubscribe from this list: send the line "unsubscribe linux-rt-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html

Then again, CentOS 5 (and other variants of that ilk) is under official
support until early 2017[1], and it uses 0.9.8[2]
-- 
Kind Regards / med vänlig hälsning,

Staffan Tjernström

[1]http://wiki.centos.org/FAQ/General#head-fe8a0be91ee3e7dea812e8694491e1dde5b75e6d 
[2]http://vault.centos.org/centos/5/os/SRPMS/numactl-0.9.8-12.el5_6.src.rpm

Re: [PATCH] cyclictest: remove compatibilty hack for old libnuma, repairs "-a" for isolcpus

[Henning Schild]

On Thu, 21 May 2015 12:23:25 +0000
Staffan Tjernstrom <stjernstrom@eagleseven.com> wrote:

> Then again, CentOS 5 (and other variants of that ilk) is under
> official support until early 2017[1], and it uses 0.9.8[2]

The problem only appears >=2.0 && <2.0.9.

Henning