[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <a25f878e-83d9-440a-9741-4cf86db4a716@intel.com>
Date: Fri, 13 Sep 2024 08:47:52 -0700
From: Dave Hansen <dave.hansen@...el.com>
To: Borislav Petkov <bp@...en8.de>, Tony W Wang-oc <TonyWWang-oc@...oxin.com>
Cc: tglx@...utronix.de, mingo@...hat.com, dave.hansen@...ux.intel.com,
x86@...nel.org, hpa@...or.com, tony.luck@...el.com,
linux-kernel@...r.kernel.org, linux-edac@...r.kernel.org,
CobeChen@...oxin.com, TimGuo@...oxin.com, LeoLiu-oc@...oxin.com,
Lyle Li <LyleLi@...oxin.com>
Subject: Re: [PATCH v1 1/3] x86/mce: Add centaur vendor to support Zhaoxin MCA
On 9/13/24 07:27, Borislav Petkov wrote:
>> + if (c->x86_vendor == X86_VENDOR_CENTAUR) {
>> + /*
>> + * All newer Centaur CPUs support MCE broadcasting. Enable
>> + * synchronization with a one second timeout.
>> + */
>> + if ((c->x86 == 6 && c->x86_model == 0xf && c->x86_stepping >= 0xe) ||
>> + c->x86 > 6) {
>> + if (cfg->monarch_timeout < 0)
>> + cfg->monarch_timeout = USEC_PER_SEC;
>> + }
>> + }
> So if centaur == zhaoxin, why aren't you moving this hunk to
> mce_zhaoxin_feature_init() instead?
The centaur and zhaoxin logic is also _really_ close here:
> if (c->x86 > 6 || (c->x86_model == 0x19 || c->x86_model == 0x1f)) {
> if (cfg->monarch_timeout < 0)
> cfg->monarch_timeout = USEC_PER_SEC;
> }
vs
> if ((c->x86 == 6 && c->x86_model == 0xf && c->x86_stepping >= 0xe) ||
> c->x86 > 6) {
> if (cfg->monarch_timeout < 0)
> cfg->monarch_timeout = USEC_PER_SEC;
> }
I'd just randomly guess that the zhaoxin version is buggy because it
doesn't do a c->x86 check before the "(c->x86_model == 0x19 ||
c->x86_model == 0x1f)".
So instead of copying and pasting the same block over and over, can we
consolidate it a bit?
foo()
{
/* Older CPUs do not do MCE broadcast: */
if (c->x86 < 6)
return;
/* All newer ones do: */
if (c->x86 > 6)
goto mce_broadcast;
/* Family 6 is mixed: */
if (c->x86_vendor == X86_VENDOR_CENTAUR) {
if (c->x86_model == 0xf &&
c->x86_stepping >= 0xe)
goto mce_broadcast;
} else if (c->x86_vendor == X86_VENDOR_ZHAOXIN) {
if (c->x86_model == 0x19 ||
c->x86_model == 0x1f))
goto mce_broadcast;
}
return;
mce_broadcast:
if (cfg->monarch_timeout < 0)
cfg->monarch_timeout = USEC_PER_SEC;
}
Heck, the Intel code can even go in there I think. Wouldn't that tell
the story a bit better?
Powered by blists - more mailing lists