Re: [PATCH v4 5/7] PCI/AER: Introduce ratelimit for error logs

[Bjorn Helgaas <helgaas@kernel.org>]

On Fri, Mar 21, 2025 at 02:47:36PM -0700, Sathyanarayanan Kuppuswamy wrote:
> On 3/21/25 12:24 PM, Jon Pan-Doh wrote:
> > On Thu, Mar 20, 2025 at 6:00 PM Sathyanarayanan Kuppuswamy
> > <sathyanarayanan.kuppuswamy@linux.intel.com> wrote:
> > > Should we exclude fatal errors from the rate limit? Fatal error
> > > logs would be really useful for debug analysis, and they not
> > > happen very frequently.
>
> > The logs today only make the distinction between correctable vs.
> > uncorrectable so I thought it made sense to be consistent.
> 
> You're right. From a logging perspective, the current driver only
> differentiates between correctable and uncorrectable errors.
> However, the goal of your patch series is to reduce the spam of
> frequent errors.  While we are rate-limiting these frequent logs, we
> must ensure that we don't miss important logs. I believe we did not
> rate-limit DPC logs for this very reason.
> 
> > Maybe this is something that could be deferred? The only fixed
> 
> I am fine with deferring. IIUC, if needed, through sysfs user can
> skip rate-limit for uncorrectable errors, right?
> 
> But, is the required change to do this complex? Won't skipping the
> rate limit check for fatal errors solve the problem?
> 
> Bjorn, any comments? Do you think Fatal errors should be
> rate-limited?

I'm inclined to not ratelimit fatal errors unless we've seen issues
with a flood of them.

Bjorn