linux-kernel - Re: [RFC PATCH 1/6] PCI/RCEC: Introduce pcie_walk_rcec

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

Message-ID: <6626ecb93f752_690a29483@dwillia2-xfh.jf.intel.com.notmuch>
Date: Mon, 22 Apr 2024 16:03:21 -0700
From: Dan Williams <dan.j.williams@...el.com>
To: Terry Bowman <Terry.Bowman@....com>, Dan Williams
	<dan.j.williams@...el.com>, Li Ming <ming4.li@...el.com>, <rrichter@....com>
CC: <linux-cxl@...r.kernel.org>, <linux-kernel@...r.kernel.org>
Subject: Re: [RFC PATCH 1/6] PCI/RCEC: Introduce pcie_walk_rcec_all()

Terry Bowman wrote:
[..]
> > Hi Terry,
> > 
> > This patchset is responding to the implications of the implementation
> > note in 9.18.1.5 RCEC Downstream Port Association Structure (RDPAS).
> > That says that CXL.io and CXL.cachemem errors in Root Ports may indeed
> > be signaled to an RCEC. Do you expect that implementation note to cause
> > any issues on platforms that do not follow that CXL spec behavior?
> > 
> > My expectation is that it may just cause extra polling for errors, but
> > not cause any harm.
> 
> AMD platforms in RCH/RCD mode consume protocol errors in the RCEC's AER driver. AMD 
> platforms in VH mode consume protocol errors (including root port errors) in the 
> root port's AER driver. The exception is the VH mode host with CXL1.1 endpoint and 
> RCH downstream errors. CXL1.1 endpoint and RCH downstream errors in a VH host are 
> consumed in the RCEC.

I agree that's the most compatible path for existing software.

> I don't believe these patchset changes would affect this behavior. But, I will need 
> to test to confirm.

As I wrote to Li Ming, I think any potential conflict can further be
limited by the fact that this extra scanning is limited to CXL.cachemem,
not typical PCI AER flows.