On 17/08/2022 19:00, Borislav Petkov wrote: > On Wed, Aug 17, 2022 at 06:56:11PM -0300, Guilherme G. Piccoli wrote: >> But do you agree that currently, in case of a kdump, that information >> *is not collected*, with our without my patch? > > If for some reason that panic notifier does not get run before kdump, > then that's a bug and that driver should not use a panic notifier to log > hw errors in the first place. > Indeed, the notifiers don't run before kdump by default, in a conscious decision of the kdump maintainers. You might be right, in the sense that maybe the edac error handler shouldn't run as a panic notifier. Let's see if Tony / Dinh can chime in on that discussion - we could move it to run in the kexec event as well, so it'd always run before a kdump, but maybe the risk it offers during panic time is not worth. Again - a matter of a trade-off, a good compromise must be agreed by all parties (kdump maintainers are usually extremely afraid of taking risks to not break kdump). Cheers!
