[Patch suggestion] Probability flag for 32 bit syscalls

[Patch suggestion] Probability flag for 32 bit syscalls

[Ildar Muslukhov]

Hi everyone,

In the children/random-syscalls.c file (function choose_syscall_table)
there is a FIXME section that right now disables 32 bit version of
syscall if a 64 bit version is present and enabled.

// FIXME: I forgot why this got disabled. Revisit.
//                      if (rand() % 100 < 10)
//

Just want to clear this one out. Should it be a command line
parameter? Something like -P 50 (for 50% probability, lets say of 32
bit). If so, what should be a default behaviour when the parameter is
not present 50/50 or 0/100 as it is right now.

-Ildar

Re: [Patch suggestion] Probability flag for 32 bit syscalls

[Dave Jones]

On Tue, Sep 24, 2013 at 11:36:53AM -0700, Ildar Muslukhov wrote:
 > Hi everyone,
 > 
 > In the children/random-syscalls.c file (function choose_syscall_table)
 > there is a FIXME section that right now disables 32 bit version of
 > syscall if a 64 bit version is present and enabled.
 > 
 > // FIXME: I forgot why this got disabled. Revisit.
 > //                      if (rand() % 100 < 10)
 > //
 > 
 > Just want to clear this one out. Should it be a command line
 > parameter? Something like -P 50 (for 50% probability, lets say of 32
 > bit). If so, what should be a default behaviour when the parameter is
 > not present 50/50 or 0/100 as it is right now.

So that got disabled a long time ago. Back in May of last year in commit
2d4de3c15fa80c42cccb1863fbd7879ef5aeaf38.

I'm not sure that ever got fixed up. It might be worth doing some long
runs to see if it's still a problem.

	Dave

Re: [Patch suggestion] Probability flag for 32 bit syscalls

[Ildar Muslukhov]

Is there any specific steps to reproduce this behaviour? Or just
calling random syscalls will do it?

Ildar

On Tue, Sep 24, 2013 at 11:50 AM, Dave Jones <davej@redhat.com> wrote:
> On Tue, Sep 24, 2013 at 11:36:53AM -0700, Ildar Muslukhov wrote:
>  > Hi everyone,
>  >
>  > In the children/random-syscalls.c file (function choose_syscall_table)
>  > there is a FIXME section that right now disables 32 bit version of
>  > syscall if a 64 bit version is present and enabled.
>  >
>  > // FIXME: I forgot why this got disabled. Revisit.
>  > //                      if (rand() % 100 < 10)
>  > //
>  >
>  > Just want to clear this one out. Should it be a command line
>  > parameter? Something like -P 50 (for 50% probability, lets say of 32
>  > bit). If so, what should be a default behaviour when the parameter is
>  > not present 50/50 or 0/100 as it is right now.
>
> So that got disabled a long time ago. Back in May of last year in commit
> 2d4de3c15fa80c42cccb1863fbd7879ef5aeaf38.
>
> I'm not sure that ever got fixed up. It might be worth doing some long
> runs to see if it's still a problem.
>
>         Dave
>
>
>

Re: [Patch suggestion] Probability flag for 32 bit syscalls

[Dave Jones]

On Tue, Sep 24, 2013 at 12:03:28PM -0700, Ildar Muslukhov wrote:
 > Is there any specific steps to reproduce this behaviour? Or just
 > calling random syscalls will do it?
 
yeah, nothing special.  I've been running it here for the last half hour
without seeing anything weird so far. ( I changed the rand to be < 50 just
to exercise it more).

I'll leave it run for the afternoon, and see if anything falls out.

	Dave

Re: [Patch suggestion] Probability flag for 32 bit syscalls

[Ildar Muslukhov]

Any luck with the test?

On Tue, Sep 24, 2013 at 12:21 PM, Dave Jones <davej@redhat.com> wrote:
> On Tue, Sep 24, 2013 at 12:03:28PM -0700, Ildar Muslukhov wrote:
>  > Is there any specific steps to reproduce this behaviour? Or just
>  > calling random syscalls will do it?
>
> yeah, nothing special.  I've been running it here for the last half hour
> without seeing anything weird so far. ( I changed the rand to be < 50 just
> to exercise it more).
>
> I'll leave it run for the afternoon, and see if anything falls out.
>
>         Dave

Re: [Patch suggestion] Probability flag for 32 bit syscalls

[Dave Jones]

On Wed, Sep 25, 2013 at 01:25:21PM -0700, Ildar Muslukhov wrote:
 > Any luck with the test?

Nothing specifically problematic with 32bit syscalls, but I've got a bunch
of changes that I'll push out over the next day or so as a result of yesterdays
run.  I found a bunch of cases where we end up with all the children stuck
reading fd's, and ignoring sigkill after the sigalarm has gone off. Still tweaking it.

	Dave

Re: [Patch suggestion] Probability flag for 32 bit syscalls

[Ildar Muslukhov]

OK, then I will wait till that push to submit my other patches.

On Wed, Sep 25, 2013 at 1:32 PM, Dave Jones <davej@redhat.com> wrote:
> On Wed, Sep 25, 2013 at 01:25:21PM -0700, Ildar Muslukhov wrote:
>  > Any luck with the test?
>
> Nothing specifically problematic with 32bit syscalls, but I've got a bunch
> of changes that I'll push out over the next day or so as a result of yesterdays
> run.  I found a bunch of cases where we end up with all the children stuck
> reading fd's, and ignoring sigkill after the sigalarm has gone off. Still tweaking it.
>
>         Dave
>