linux-kernel - Re: [PATCH v2 1/2] EDAC/amd64: Check register values from all UMCs

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Wed, 5 Jan 2022 16:12:08 +0000
From:   Yazen Ghannam <yazen.ghannam@....com>
To:     Borislav Petkov <bp@...en8.de>
Cc:     linux-edac@...r.kernel.org, linux-kernel@...r.kernel.org,
        mchehab@...nel.org, tony.luck@...el.com, james.morse@....com,
        rric@...nel.org, Smita.KoralahalliChannabasappa@....com,
        william.roche@...cle.com
Subject: Re: [PATCH v2 1/2] EDAC/amd64: Check register values from all UMCs

On Thu, Dec 30, 2021 at 12:36:44PM +0100, Borislav Petkov wrote:
> On Thu, Dec 16, 2021 at 04:08:18PM +0000, Yazen Ghannam wrote:
> > No, that's a good question. And actually the assumption is incorrect. It is
> > allowed to have different DIMM types in a system though all DIMMs on a single
> > UMC must match.
> 
> Oh fun, really?!
> 
> So a single system can have DDR4 *and* DDR5 on the same board?!
>

Well, I don't know about that specifically. There are some restrictions, but
you could have UDIMMs and RDIMMs of the same generation, at least.
 
> So then that
> 
> 	pvt->dram_type
> 
> is insufficient to store the DIMM type for a pvt. If you have multiple
> UMCs on a pvt and all have different type DIMMs, then you need the
> relevant DIMM type when you dump it in sysfs...
> 
> Which then means, you need ->dram_type to be per UMC...
> 
> And also, I'm assuming the hw already enforces that DIMMs on a single
> UMC must match - it simply won't boot if they don't so you don't have to
> enforce that, at least.
> 
> > Do you recommend a follow up patch or should this one be reworked?
> 
> This one is insufficient, I'm afraid.
> 
> One way to address this is, you could use pvt->umc at the places where
> dram_type is used and assign directly to the dimm->mtype thing. But then
> you'd need a way to map each struct dimm_info *dimm to the UMC so that
> you can determine the correct DIMM type.
> 

I did send a patch that did something like this.
https://lkml.kernel.org/r/20211228200615.412999-2-yazen.ghannam@amd.com

Though this got a build warning report, so I need to follow up on that.

> Which would make pvt->dram_type redundant and can be removed.
>

I kept this so as to not break legacy systems. But I'll look at it again. I
think you may be right.

Thanks,
Yazen