Re: perfmon trouble
From: Stephane Eranian
Date: Mon Jun 11 2018 - 14:51:55 EST
On Mon, Jun 11, 2018 at 10:18 AM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> On Mon, Jun 11, 2018 at 10:04:00AM -0700, Linus Torvalds wrote:
> > On Mon, Jun 11, 2018 at 9:49 AM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> > >
> > > The problem is that even oprofile on ia64 depends on perfmon.
> > Hmm? You can definitely enable ia64 support for oprofile even without perfmon.
> Oh, I think my memory is playing tricks on me. This is my confusion, I think:
> oprofile-$(CONFIG_PERFMON) += perfmon.o
> so perfmon events are exposed through oprofile, but you can disable
> perfmon without disabling oprofile.
> > Because I'd be inclined to just remove CONFIG_PERFMON support, and see
> > if anybody even notices..
> > I'm not expecting a lot of people to do a lot of oprofile on ia64
> > anyway. It's a bit late to start optimizing things now.
> > Do people use perfmon still? Maybe. Maybe not. Perhaps we could just
> > mark it as broken in the Kconfig file for now, and see if somebody
> > says something?
Thanks Al for the careful analysis. If I understand correctly, the
problem seems to be that on
execve the mapping of the sampling buffer is not destroyed and then on
close, the kernel
unmaps whatever the new binary had in that address range. The fix
would be to ensure
the mmap is destroyed on execve.
My problem is that I do not have IA64 hw anymore, so whatever the fix,
I will not be able
to test this. In the meantime, I agree with Matthew, simply disable
> That gets my vote.
> Tony? Fenghua?
> diff --git a/arch/ia64/Kconfig b/arch/ia64/Kconfig
> index 792437d526c6..ff861420b8f5 100644
> --- a/arch/ia64/Kconfig
> +++ b/arch/ia64/Kconfig
> @@ -455,6 +455,7 @@ config IA64_MCA_RECOVERY
> config PERFMON
> bool "Performance monitor support"
> + depends on BROKEN
> Selects whether support for the IA-64 performance monitor hardware
> is included in the kernel. This makes some kernel data-structures a