question on FSL_EMB perf

question on FSL_EMB perf

[Peter Zijlstra]

Hi Mike, Ben,

I just noticed:

arch/powerpc/Kconfig:   select HAVE_PERF_EVENTS_NMI if PPC64

But can't ppc32 have FSL_EMB perf?

Re: question on FSL_EMB perf

[Scott Wood]

On Wed, 2015-06-10 at 13:41 +0200, Peter Zijlstra wrote:
> Hi Mike, Ben,
> 
> I just noticed:
> 
> arch/powerpc/Kconfig:   select HAVE_PERF_EVENTS_NMI if PPC64
> 
> But can't ppc32 have FSL_EMB perf?

Yes, but it doesn't use NMIs.  ppc64 has lazy interrupt disabling 
which functions as a pseudo-NMI.

-Scott

Re: question on FSL_EMB perf

[Peter Zijlstra]

On Wed, 2015-06-10 at 14:17 -0500, Scott Wood wrote:
> On Wed, 2015-06-10 at 13:41 +0200, Peter Zijlstra wrote:
> > Hi Mike, Ben,
> >=20
> > I just noticed:
> >=20
> > arch/powerpc/Kconfig:   select HAVE_PERF_EVENTS_NMI if PPC64
> >=20
> > But can't ppc32 have FSL_EMB perf?
>=20
> Yes, but it doesn't use NMIs.  ppc64 has lazy interrupt disabling=20
> which functions as a pseudo-NMI.

I know. But you can get the same nesting nonsense as with actual real
NMIs.

And seeing how you select HAVE_PERF_EVENT_NMI for PPC64, I figure you
ought to select it too for whatever fsl-emb is.

# git grep nmi_enter arch/powerpc/
arch/powerpc/perf/core-book3s.c:                nmi_enter();
arch/powerpc/perf/core-fsl-emb.c:               nmi_enter();

Re: question on FSL_EMB perf

[Scott Wood]

On Wed, 2015-06-10 at 21:27 +0200, Peter Zijlstra wrote:
> On Wed, 2015-06-10 at 14:17 -0500, Scott Wood wrote:
> > On Wed, 2015-06-10 at 13:41 +0200, Peter Zijlstra wrote:
> > > Hi Mike, Ben,
> > > 
> > > I just noticed:
> > > 
> > > arch/powerpc/Kconfig:   select HAVE_PERF_EVENTS_NMI if PPC64
> > > 
> > > But can't ppc32 have FSL_EMB perf?
> > 
> > Yes, but it doesn't use NMIs.  ppc64 has lazy interrupt disabling 
> > which functions as a pseudo-NMI.
> 
> I know. But you can get the same nesting nonsense as with actual real
> NMIs.

What sort of "nesting nonsense" would we get on ppc32?  I wasn't 
trying to say that the pseudo-NMIs didn't count -- I was pointing out 
that we don't even have that on ppc32.

> And seeing how you select HAVE_PERF_EVENT_NMI for PPC64, I figure you
> ought to select it too for whatever fsl-emb is.

Why?

> # git grep nmi_enter arch/powerpc/
> arch/powerpc/perf/core-book3s.c:                nmi_enter();
> arch/powerpc/perf/core-fsl-emb.c:               nmi_enter();

nmi_enter() only gets called when perf_intr_is_nmi() returns non-zero, 
which only happens on ppc64.

-Scott

Re: question on FSL_EMB perf

[Peter Zijlstra]

On Wed, 2015-06-10 at 15:13 -0500, Scott Wood wrote:

> What sort of "nesting nonsense" would we get on ppc32?  I wasn't=20
> trying to say that the pseudo-NMIs didn't count -- I was pointing out=20
> that we don't even have that on ppc32.

Oh, I was not aware ppc32 did not have that.

> > # git grep nmi_enter arch/powerpc/
> > arch/powerpc/perf/core-book3s.c:                nmi_enter();
> > arch/powerpc/perf/core-fsl-emb.c:               nmi_enter();
>=20
> nmi_enter() only gets called when perf_intr_is_nmi() returns non-zero,=
=20
> which only happens on ppc64.

ok, so fsl-emb is ppc32 and ppc64, not confusing at all this stuff.

Re: question on FSL_EMB perf

[Scott Wood]

On Wed, 2015-06-10 at 23:14 +0200, Peter Zijlstra wrote:
> On Wed, 2015-06-10 at 15:13 -0500, Scott Wood wrote:
> 
> > What sort of "nesting nonsense" would we get on ppc32?  I wasn't 
> > trying to say that the pseudo-NMIs didn't count -- I was pointing 
> > out 
> > that we don't even have that on ppc32.
> 
> Oh, I was not aware ppc32 did not have that.
> 
> > > # git grep nmi_enter arch/powerpc/
> > > arch/powerpc/perf/core-book3s.c:                nmi_enter();
> > > arch/powerpc/perf/core-fsl-emb.c:               nmi_enter();
> > 
> > nmi_enter() only gets called when perf_intr_is_nmi() returns non-
> > zero, 
> > which only happens on ppc64.
> 
> ok, so fsl-emb is ppc32 and ppc64, not confusing at all this stuff.

Both drivers can be used on either 32 or 64 bits.

-Scott

Re: question on FSL_EMB perf

[Michael Ellerman]

On Wed, 2015-06-10 at 13:41 +0200, Peter Zijlstra wrote:
> Hi Mike, Ben,
> 
> I just noticed:
> 
> arch/powerpc/Kconfig:   select HAVE_PERF_EVENTS_NMI if PPC64

We actually only added that recently, in c54b2bf1b5e9 "powerpc: Add ppc64 hard
lockup detector support".

AFAICS it doesn't actually have anything to do with the NMI behaviour of the
perf interrupt. It controls whether the hardlockup detector is enabled and/or
the oprofile NMI timer.

cheers