RE: Simultaneous access to same CPU/counter across processes crashes hwpmc (TSC and other classes)

Public

> -----Original Message-----
> From: Adrian Chadd <adrian@freebsd.org>
> Sent: Saturday, August 22, 2026 4:04 PM
> To: Nascimento, Anderson Eduardo
> <AndersonEduardo.Nascimento@amd.com>
> Cc: freebsd-hackers@freebsd.org
> Subject: Re: Simultaneous access to same CPU/counter across processes
> crashes hwpmc (TSC and other classes)
>
> [You don't often get email from adrian@freebsd.org. Learn why this is
> important at https://aka.ms/LearnAboutSenderIdentification ]
>
> Caution: This message originated from an External Source. Use proper caution
> when opening attachments, clicking links, or responding.
>
>
> On Wed, 19 Aug 2026 at 07:18, Nascimento, Anderson Eduardo
> <AndersonEduardo.Nascimento@amd.com> wrote:
> >
> > Public
> >
> > Hello all,
> >
> > I am working on developing a new class of PMCs for the hwpmc subsystem
> and I ran into a kernel crash while testing my implementation. I believe I have
> identified the reason the crash happens, and I could try to fix it on the affected
> classes, but so far it's still not clear to me what model the hwpmc subsystem
> intends to enforce. Should simultaneous access to the same CPU and counter
> by different processes be allowed?
>
> The initial design goals and list from the original author (jkoshy@) is still
> available on the wiki, eg see https://wiki.freebsd.org/PmcTools/Features .
>
> So yes, one of the design goals of hwpmc versus what was available in Linux at
> the time (oprofile) was that counters are virtualised both for context switching
> between processes /and/ multiple processes accessing the same counters on a
> given CPU (and as part of this, handling context switching appropriately.)
>
> It's what made it much nicer to use back when it was introduced versus
> oprofile. (of course, oprofiles tooling 'just worked' versus pmc..)
>
>

Hello Adrian,

This adds important context. I had checked other pages related to PMC on the Wiki, but I missed that one. Thanks for sharing this.

>
> -adrian