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: <3e3115c0-c3a2-4ec2-8aea-ee1b40057dd6@redhat.com>
Date: Tue, 1 Apr 2025 11:33:59 +0200
From: David Hildenbrand <david@...hat.com>
To: Gregory Price <gourry@...rry.net>, dan.j.williams@...el.com
Cc: nvdimm@...ts.linux.dev, linux-kernel@...r.kernel.org,
 kernel-team@...a.com, vishal.l.verma@...el.com, dave.jiang@...el.com,
 linux-cxl@...r.kernel.org
Subject: Re: [PATCH] DAX: warn when kmem regions are truncated for memory
 block alignment.

On 31.03.25 20:27, Gregory Price wrote:
> On Fri, Mar 21, 2025 at 02:07:31PM -0400, Gregory Price wrote:
>> Device capacity intended for use as system ram should be aligned to the
>> architecture-defined memory block size or that capacity will be silently
>> truncated and capacity stranded.
>>
>> As hotplug dax memory becomes more prevelant, the memory block size
>> alignment becomes more important for platform and device vendors to
>> pay attention to - so this truncation should not be silent.
>>
>> This issue is particularly relevant for CXL Dynamic Capacity devices,
>> whose capacity may arrive in spec-aligned but block-misaligned chunks.
>>
>> Example:
>>   [...] kmem dax0.0: dax region truncated 2684354560 bytes - alignment
>>   [...] kmem dax1.0: dax region truncated 1610612736 bytes - alignment
>>
>> Signed-off-by: Gregory Price <gourry@...rry.net>
> 
> Gentle pokes.  There were a couple questions last week whether we should
> warn here or actually fix something in memory-hotplug.
> 
> Notes from CXL Boot to Bash session discussions:
> 
> 
> We discussed [1] how this auto-sizing can cause 1GB huge page
> allocation failures (assuming you online as ZONE_NORMAL). That means
> ACPI-informed sizing by default would potentially be harmful to existing
> systems and adding yet-another-boot-option just seems nasty.
> 
> I've since dropped acpi-informed block size patch[2].  If there are opinions
> otherwise, I can continue pushing it.

Oh, I thought we would be going forward with that. What's the reason we 
would not want to do that?

> 
> 
> We also discussed[3] variable-sized blocks having some nasty corner cases.
> Not unsolvable, but doesn't help users in the short term.
> 
> 
> There was some brief discussion about whether a hotplug memblock with a
> portion as offline pages would be possible.  This seems hacky?  There
> was another patch set discussing this, but I can't seem to find it.

Yeah, I proposed something like that as well when I started working on 
virtio-mem and did not really understand the whole hot(un)plug model in 
Linux properly. Someone else proposed it again a couple of years ago, 
but it's just wrong and should not be done that way.

One could implement something like virtio-mem, whereby parts of a Linux 
memory block can be added/removed independently ("fake offlined"). But 
the whole idea of virtio-mem is that all memory in the Linux memory 
block range belongs to it. So it doesn't quite apply to DAX where parts 
of a Linux memory block might be from something completely different 
(e.g., boot memory etc).

-- 
Cheers,

David / dhildenb


Powered by blists - more mailing lists

Powered by Openwall GNU/*/Linux Powered by OpenVZ