Re: [PATCH v3 08/10] x86/mtrr: let cache_aps_delayed_init replace mtrr_aps_delayed_init

[Juergen Gross <jgross@xxxxxxxx>]

On 28.09.22 07:30, Juergen Gross wrote:
> On 28.09.22 01:16, Borislav Petkov wrote:
> > On Tue, Sep 27, 2022 at 02:21:17PM +0200, Juergen Gross wrote:
> > > So replacing the bool with "(system_state != SYSTEM_RUNNING)" is fine
> > > with you right now? We can later switch that to the "more elegant"
> > > solution when it shows up.
> > Ok, I think I have something. And it was staring me straight in the
> > face but I didn't see it: the MTRR code needs a hotplug notifier. In
> > that notifier it can do the immediate, i.e., non-delayed init while the
> > delayed init becomes the default, see below.
> >
> > And ignore the pr_info debugging gunk pls.
> >
> > mtrr_ap_init() becomes the notifier callback. It doesn't need to be
> > called in identify_secondary_cpu() anymore as in the init case that
> > function doesn't do anything - delayed=true - and in the hotplug case
> > the notifier runs.
> Are sure the hotplug notifier doesn't get called in the boot and in the
> resume cases? I don't see how those calls are being not done or resulting in
> not doing anything.
In case my suspicion is correct: this can still be solved by adding the
hotplug notifier only in mtrr_aps_init(), and removing it again in
arch_thaw_secondary_cpus_begin().


Juergen

Attachment: OpenPGP_0xB0DE9DD628BF132F.asc
Description: OpenPGP public key

Attachment: OpenPGP_signature
Description: OpenPGP digital signature