[Qemu-devel] [6343] Fix false positive for AIO on OpenBSD

[Qemu-devel] [6343] Fix false positive for AIO on OpenBSD

[Blue Swirl]

Revision: 6343
          http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=6343
Author:   blueswir1
Date:     2009-01-16 18:13:32 +0000 (Fri, 16 Jan 2009)

Log Message:
-----------
Fix false positive for AIO on OpenBSD

Modified Paths:
--------------
    trunk/configure

Modified: trunk/configure
===================================================================
--- trunk/configure	2009-01-16 14:27:37 UTC (rev 6342)
+++ trunk/configure	2009-01-16 18:13:32 UTC (rev 6343)
@@ -1005,7 +1005,9 @@
   aio=no
   cat > $TMPC << EOF
 #include <pthread.h>
-int main(void) { pthread_mutex_t lock; return 0; }
+#include <signal.h>
+int main(void) { struct sigevent s; pthread_mutex_t lock;
+    return sigqueue(0, 0, s.sigev_value); }
 EOF
   if $cc $ARCH_CFLAGS -o $TMPE $AIOLIBS $TMPC 2> /dev/null ; then
     aio=yes

Re: [Qemu-devel] [6343] Fix false positive for AIO on OpenBSD

[Anthony Liguori]

Blue Swirl wrote:
> Revision: 6343
>           http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=6343
> Author:   blueswir1
> Date:     2009-01-16 18:13:32 +0000 (Fri, 16 Jan 2009)
>
> Log Message:
> -----------
> Fix false positive for AIO on OpenBSD
>   

If OpenBSD doesn't have sigqueue, we could just replace it with kill.  I 
don't think we need to disable AIO.

Regards,

Anthony Liguori

> Modified Paths:
> --------------
>     trunk/configure
>
> Modified: trunk/configure
> ===================================================================
> --- trunk/configure	2009-01-16 14:27:37 UTC (rev 6342)
> +++ trunk/configure	2009-01-16 18:13:32 UTC (rev 6343)
> @@ -1005,7 +1005,9 @@
>    aio=no
>    cat > $TMPC << EOF
>  #include <pthread.h>
> -int main(void) { pthread_mutex_t lock; return 0; }
> +#include <signal.h>
> +int main(void) { struct sigevent s; pthread_mutex_t lock;
> +    return sigqueue(0, 0, s.sigev_value); }
>  EOF
>    if $cc $ARCH_CFLAGS -o $TMPE $AIOLIBS $TMPC 2> /dev/null ; then
>      aio=yes
>
>
>
>
>   

Re: [Qemu-devel] [6343] Fix false positive for AIO on OpenBSD

[Blue Swirl]

On 1/16/09, Anthony Liguori <anthony@codemonkey.ws> wrote:
> Blue Swirl wrote:
>
> > Revision: 6343
> >
> http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=6343
> > Author:   blueswir1
> > Date:     2009-01-16 18:13:32 +0000 (Fri, 16 Jan 2009)
> >
> > Log Message:
> > -----------
> > Fix false positive for AIO on OpenBSD
> >
> >
>
>  If OpenBSD doesn't have sigqueue, we could just replace it with kill.  I
> don't think we need to disable AIO.

struct sigevent is also missing, is that a problem?

Re: [Qemu-devel] [6343] Fix false positive for AIO on OpenBSD

[Anthony Liguori]

Blue Swirl wrote:
> On 1/16/09, Anthony Liguori <anthony@codemonkey.ws> wrote:
>   
>> Blue Swirl wrote:
>>
>>     
>>> Revision: 6343
>>>
>>>       
>> http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=6343
>>     
>>> Author:   blueswir1
>>> Date:     2009-01-16 18:13:32 +0000 (Fri, 16 Jan 2009)
>>>
>>> Log Message:
>>> -----------
>>> Fix false positive for AIO on OpenBSD
>>>
>>>
>>>       
>>  If OpenBSD doesn't have sigqueue, we could just replace it with kill.  I
>> don't think we need to disable AIO.
>>     
>
> struct sigevent is also missing, is that a problem?
>   

It's just used for assigning the signal value.  We could have a #ifdef 
HAVE_SIGEVENT and define it if it doesn't exist--or just change the 
aioinit structure to use an int instead of the struct sigevent 
structure.  It's sigevent for glibc compatibility but that's not really 
important at this stage.

Regards,

Anthony Liguori

Re: [Qemu-devel] [6343] Fix false positive for AIO on OpenBSD

[Blue Swirl]

On 1/16/09, Anthony Liguori <anthony@codemonkey.ws> wrote:
> Blue Swirl wrote:
>
> > On 1/16/09, Anthony Liguori <anthony@codemonkey.ws> wrote:
> >
> >
> > > Blue Swirl wrote:
> > >
> > >
> > >
> > > > Revision: 6343
> > > >
> > > >
> > > >
> > >
> http://svn.sv.gnu.org/viewvc/?view=rev&root=qemu&revision=6343
> > >
> > >
> > > > Author:   blueswir1
> > > > Date:     2009-01-16 18:13:32 +0000 (Fri, 16 Jan 2009)
> > > >
> > > > Log Message:
> > > > -----------
> > > > Fix false positive for AIO on OpenBSD
> > > >
> > > >
> > > >
> > > >
> > >  If OpenBSD doesn't have sigqueue, we could just replace it with kill.
> I
> > > don't think we need to disable AIO.
> > >
> > >
> >
> > struct sigevent is also missing, is that a problem?
> >
> >
>
>  It's just used for assigning the signal value.  We could have a #ifdef
> HAVE_SIGEVENT and define it if it doesn't exist--or just change the aioinit
> structure to use an int instead of the struct sigevent structure.  It's
> sigevent for glibc compatibility but that's not really important at this
> stage.

Ok. Should kill be used always or just when sigqueue is not available?

Re: [Qemu-devel] [6343] Fix false positive for AIO on OpenBSD

[Anthony Liguori]

Blue Swirl wrote:
> On 1/16/09, Anthony Liguori <anthony@codemonkey.ws> wrote:
>   
>>
>>  It's just used for assigning the signal value.  We could have a #ifdef
>> HAVE_SIGEVENT and define it if it doesn't exist--or just change the aioinit
>> structure to use an int instead of the struct sigevent structure.  It's
>> sigevent for glibc compatibility but that's not really important at this
>> stage.
>>     
>
> Ok. Should kill be used always or just when sigqueue is not available?
>   

Since we use it for process instead of thread signaling, I think it's 
safe to always use kill().  I'm not completely sure though, having it 
conditional on sigqueue would certainly be the conservative thing to do.

Regards,

Anthony Liguori

Re: [Qemu-devel] [6343] Fix false positive for AIO on OpenBSD

[Blue Swirl]

[-- Attachment #1: Type: text/plain, Size: 877 bytes --]

On 1/16/09, Anthony Liguori <anthony@codemonkey.ws> wrote:
> Blue Swirl wrote:
>
> > On 1/16/09, Anthony Liguori <anthony@codemonkey.ws> wrote:
> >
> >
> > >
> > >  It's just used for assigning the signal value.  We could have a #ifdef
> > > HAVE_SIGEVENT and define it if it doesn't exist--or just change the
> aioinit
> > > structure to use an int instead of the struct sigevent structure.  It's
> > > sigevent for glibc compatibility but that's not really important at this
> > > stage.
> > >
> > >
> >
> > Ok. Should kill be used always or just when sigqueue is not available?
> >
> >
>
>  Since we use it for process instead of thread signaling, I think it's safe
> to always use kill().  I'm not completely sure though, having it conditional
> on sigqueue would certainly be the conservative thing to do.

This version uses just kill. Seems to work on Linux and OpenBSD.

[-- Attachment #2: obsd_aio.diff --]
[-- Type: plain/text, Size: 2054 bytes --]

Re: [Qemu-devel] [6343] Fix false positive for AIO on OpenBSD

[Anthony Liguori]

Blue Swirl wrote:
> On 1/16/09, Anthony Liguori <anthony@codemonkey.ws> wrote:
>   
>>
>
> This version uses just kill. Seems to work on Linux and OpenBSD.
>   
Looks good to me.

Regards,

Anthony Liguori