Re: port-amd64/52596 (Another netbsd-8 panic)

[Dominik Bialy <dmb%yenn.ulegend.net@localhost>]

The following reply was made to PR port-amd64/52596; it has been noted by GNATS.

From: Dominik Bialy <dmb%yenn.ulegend.net@localhost>
To: coypu%sdf.org@localhost
Cc: Dominik Bialy <dmb%yenn.ulegend.net@localhost>, gnats-bugs%NetBSD.org@localhost
Subject: Re: port-amd64/52596 (Another netbsd-8 panic)
Date: Thu, 12 Oct 2017 08:14:19 +0200

 On Thu, Oct 12, 2017 at 12:18:29AM +0000, coypu%sdf.org@localhost wrote:
 > Complete shot in the dark, this is an errata workaround for similar
 > AMD machines:
 > 
 > The SWAPGS instruction fails to cause an input dependency on the GS
 > segment register. If the GS segment register has been recently
 > changed via a MOV or POP segment register instruction, SWAPGS may
 > incorrectly save the old value of GS base into the KernelGSbase MSR.
 > 
 > Index: arch/amd64/amd64/locore.S
 > ===================================================================
 > RCS file: /cvsroot/src/sys/arch/amd64/amd64/locore.S,v
 > retrieving revision 1.131
 > diff -u -r1.131 locore.S
 > --- arch/amd64/amd64/locore.S	30 Sep 2017 11:43:57 -0000	1.131
 > +++ arch/amd64/amd64/locore.S	12 Oct 2017 00:12:08 -0000
 > @@ -1165,6 +1165,7 @@
 >  	CLI(cx)
 >  	SWAPGS
 >  	movw	%ax,%gs
 > +	mfence
 >  	SWAPGS
 >  	STI(cx)
 >  
 > @@ -1199,6 +1200,7 @@
 >  	CLI(ax)
 >  	SWAPGS
 >  	movw	TF_GS(%rbx),%gs
 > +	mfence
 >  	SWAPGS
 >  	STI(ax)
 >  #else
 > 
 I remembered today, that while doing the upgrade to -8,
 I also enabled some AMD PowerNow settings in the BIOS.
 Yesterday I toggled them off, enabled ACPI 3.0, and
 it booted OK, and now has ~20 hours uptime.
 
 I have a theory that these settings were causing some
 CPU instability.
 
 Now there's no warning in the dmesg about errors in
 setting CPU frequency.
 
 I'm planning to see if it can keep 2 weeks of uptime,
 because the farthest it went so far was 13 days, and let
 you know. OK?
 
 Dominik