[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20170531215855.GA5382@khazad-dum.debian.net>
Date: Wed, 31 May 2017 18:58:55 -0300
From: Henrique de Moraes Holschuh <hmh@....eng.br>
To: Peter Zijlstra <peterz@...radead.org>
Cc: tglx@...utronix.de, x86@...nel.org, linux-kernel@...r.kernel.org,
kevin.b.stanton@...el.com
Subject: Re: [PATCH 2/3] x86/apic: Add TSC_DEADLINE quirk due to errata
On Wed, 31 May 2017, Peter Zijlstra wrote:
> + DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_X, 0x02000014),
...
> + DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_MOBILE, 0xb2),
> + DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_SKYLAKE_DESKTOP, 0xb2),
...
> + DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_MOBILE, 0x52),
> + DEADLINE_MODEL_MATCH_REV ( INTEL_FAM6_KABYLAKE_DESKTOP, 0x52),
Maybe these revisions need to be decreased by one, so that the SGX
behavior of messing with the microcode revision is taken into account?
When the processor microcode is auto-updated from the (signed) FIT, and
SGX's PRMRR feature is supported, the processor will report its
microcode revision as one less. Therefore, a microcode update with
revision 0xb2 auto-loaded from FIT would be reported by RDMSR(0x8b) as
revision 0xb1.
I know about this SGX-related behavior from a coreboot commit from ~two
years ago (link below). As far as I can tell, the Intel 64/IA32 SDM is
*still* missing any mention about this behavior in vol 3A section 9.11
(where it describes microcode updates).
https://review.coreboot.org/cgit/coreboot.git/commit/?id=5042aad4ded1651638ae9b60e34114b65e4f211e
--
Henrique Holschuh
Powered by blists - more mailing lists