linux-kernel - Re: [RFC][PATCH v2 22/29] mm/numa: Register information into Kmemdump

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]

Message-ID: <cab1db60-e7bc-4f31-b781-c52ad1b24da6@redhat.com>
Date: Mon, 1 Sep 2025 14:17:18 +0200
From: David Hildenbrand <david@...hat.com>
To: Eugen Hristev <eugen.hristev@...aro.org>, Michal Hocko <mhocko@...e.com>
Cc: linux-kernel@...r.kernel.org, linux-arm-msm@...r.kernel.org,
 linux-arch@...r.kernel.org, linux-mm@...ck.org, tglx@...utronix.de,
 andersson@...nel.org, pmladek@...e.com,
 linux-arm-kernel@...ts.infradead.org, linux-hardening@...r.kernel.org,
 corbet@....net, mojha@....qualcomm.com, rostedt@...dmis.org,
 jonechou@...gle.com, tudor.ambarus@...aro.org,
 Christoph Hellwig <hch@...radead.org>,
 Sergey Senozhatsky <senozhatsky@...omium.org>
Subject: Re: [RFC][PATCH v2 22/29] mm/numa: Register information into Kmemdump

On 01.09.25 14:02, Eugen Hristev wrote:
> 
> 
> On 9/1/25 13:01, David Hildenbrand wrote:
>>>>> What do you think ?
>>>>
>>>> Looks a bit over-engineered, and will require us to import a header
>>>> (likely kmemdump.h) in these files, which I don't really enjoy.
>>>>
>>>> I would start simple, without any such macro-magic. It's a very simple
>>>> function after all, and likely you won't end up having many of these?
>>>>
>>>
>>> Thanks David, I will do it as you suggested and see what comes out of it.
>>>
>>> I have one side question you might know much better to answer:
>>> As we have a start and a size for each region, this start is a virtual
>>> address. The firmware/coprocessor that reads the memory and dumps it,
>>> requires physical addresses.
>>
>> Right. I was asking myself the same question while reviewing: should we
>> directly export physical ranges here instead of virtual ones. I guess
>> virtual ones is ok.
> 
> In patch 22/29, some areas are registered using
> memblock_phys_alloc_try_nid() which allocates physical.
> In this case , phys_to_virt() didn't work for me, it was returning a
> wrong address. I used __va() and this worked. So there is a difference
> between them.

memblock_alloc_internal() calls memblock_alloc_range_nid() to then 
perform a phys_to_virt().

memblock_phys_alloc_try_nid() calls memblock_alloc_range_nid() without 
the phys_to_virt().

So it's rather surprising the a phys_to_virt() would not work in that case.

Maybe for these cases where you export the area through a new helper, 
you can just export the physical addr + length instead.

Then, it's also clear that this area is actually physically contiguous.

> 
>>
>> What do you suggest to use to retrieve that
>>> address ? virt_to_phys might be problematic, __pa or __pa_symbol? or
>>> better lm_alias ?
>>
>> All areas should either come from memblock or be global variables, right?
> 
> I would like to be able to register from anywhere. For example someone
> debugging their driver, to just register kmalloc'ed struct.
> Other use case is to register dma coherent CMA areas.

Then probably better to export physical addresses (that you need either 
way) directly from the helpers you have to add.

> 
>>
>> IIRC, virt_to_phys() should work for these. Did you run into any
>> problems with them or why do you think virt_to_phys could be problematic?
>>
> 
> I am pondering about whether it would work in all cases, considering
> it's source code comments that it shall not be used because it does not
> work for any address.

Yeah, it does for example not work for kernel stacks IIRC.

-- 
Cheers

David / dhildenb