Phil,
Try applying the following patch. There was a misplaced #endif CONFIG_SMP
in the close() path.
On Wed, Nov 29, 2006 at 05:38:41PM +0100, Philip J. Mucci wrote:
> Sorry, I wasn't clear. Yes, the numbering in /sys starts from 0. Just in
> my text descriptions...There is no PMD watchdog on this system.
>
> However, it certainly appears that this field is not being decremented
> in the code when in UP...because I can produce this reliably. I don't
> believe there is any arch specific code that handles this stuff.
>
> Phil
>
>
> On Wed, 2006-11-29 at 11:34 -0500, William Cohen wrote:
> > Philip J. Mucci wrote:
> > > This happens on a MIPS64 system, running UP.
> > >
> > > I have numbered my PMD's from 1...so I have all the hardware.
> > >
> > > Phil
> >
> > Doesn't the numbering start from 0? That is the case with what is reported
> > by
> > whichpmu. How many registers are there? What information is reported by
> > showreginfo? On the AMD machine I am using pmd0 is used by the watchdog
> > timer
> > and showreginfo isn't reporting pmd0 as available.
> >
> > -Will
> >
> > >
> > > On Wed, 2006-11-29 at 11:17 -0500, William Cohen wrote:
> > >
> > >>Philip J. Mucci wrote:
> > >>
> > >>>Hi Stefane,
> > >>>
> > >>>Well, I never ran these test cases before now....
> > >>>
> > >>>It looks like when I run either rtop or syst, I'm seeing an error
> > >>>because Perfmon2 seems to think that there's other contexts running when
> > >>>in fact there isn't. Something isn't getting cleaned up, because I can
> > >>>increase the number of 'conflicting sessions' simply by running 'self' a
> > >>>number of times. It seems every run increases this value.
> > >>
> > >>I didn't see this problem on the AMD64 machine I am running things on. I
> > >>am
> > >>noticing that PMD1 and PMD2 are being used on the AMD machine Which
> > >>architecture is this on? Is this on an architecture with only 2 counters?
> > >>The
> > >>new kernel could be reserving PMD0.
> > >>
> > >>-Will
> > >
> > >
> >
--
-Stephane
--- perfmon/perfmon_file.c.orig 2006-11-29 08:52:27.000000000 -0800
+++ perfmon/perfmon_file.c 2006-11-29 08:45:11.000000000 -0800
@@ -592,9 +592,9 @@ doit:
#ifdef CONFIG_SMP
free_it:
+#endif
if (can_release)
pfm_release_session(is_system, cpu);
-#endif
if (can_free)
pfm_context_free(ctx);
_______________________________________________
perfmon mailing list
[email protected]
http://www.hpl.hp.com/hosted/linux/mail-archives/perfmon/
