linux-kernel - Re: [PATCH] mm: Make SPLIT_PTE_PTLOCKS depend on the existence of NR

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: <a5095b37-ab61-4789-973a-aee3a748fe12@roeck-us.net>
Date: Tue, 24 Sep 2024 07:16:22 -0700
From: Guenter Roeck <linux@...ck-us.net>
To: Geert Uytterhoeven <geert@...ux-m68k.org>
Cc: David Hildenbrand <david@...hat.com>,
 Andrew Morton <akpm@...ux-foundation.org>, linux-mm@...ck.org,
 linux-kernel@...r.kernel.org, Masahiro Yamada <masahiroy@...nel.org>,
 linux-kbuild <linux-kbuild@...r.kernel.org>
Subject: Re: [PATCH] mm: Make SPLIT_PTE_PTLOCKS depend on the existence of
 NR_CPUS

On 9/24/24 00:45, Geert Uytterhoeven wrote:
> Hi Günter,
> 
> CC kbuild
> 
> I have two comments...
> 
> On Tue, Sep 24, 2024 at 1:52 AM Guenter Roeck <linux@...ck-us.net> wrote:
>> On 9/23/24 15:08, Guenter Roeck wrote:
>>> On 9/23/24 08:23, David Hildenbrand wrote:
>>>> On 23.09.24 16:25, Guenter Roeck wrote:
>>>>> SPLIT_PTE_PTLOCKS already depends on "NR_CPUS >= 4", but that evaluates
>>>>> to true if there is no NR_CPUS configuration option (such as for m68k).
>>>>> This results in CONFIG_SPLIT_PTE_PTLOCKS=y for mac_defconfig.
>>>>> This in turn causes the m68k "q800" machine to crash in qemu.
> 
> Should this be fixed in Kconfig (too)?
> 

I don't know. I thought that was intentional, though I don't understand
the logic. I didn't find a documentation that would explain how boolean
dependencies of integer objects are supposed to be handled. My expectation
that it would be similar to C was obviously wrong.

>>>> Oh, that's why my compile tests still worked ... I even removed the additional NR_CPUS check, assuming it's not required ...
>>>>
>>>> Thanks for debugging and fixing!
>>>>
>>>> Acked-by: David Hildenbrand <david@...hat.com>
>>>>
>>>
>>> Apparently it wasn't that simple :-(. 0-day reports a build failure
>>> with s390 builds.
>>>
>>> arch/s390/mm/gmap.c:357:16: error: implicit declaration of function 'pmd_pgtable_page'.
>>>
>>> Turns out that
>>>       depends on NR_CPUS && NR_CPUS >= 4
>>>
>>> doesn't work and disables SPLIT_PTE_PTLOCKS even if NR_CPUS _is_ defined.
>>> I have no idea how to declare the dependency correctly.
>>> Sorry, I did not expect that.
>>
>> The only solution I found was to define NR_CPUS for m68k. That seems to be
>> the only architecture not defining it, so hopefully that is an acceptable
>> solution. I'll send v2 of the patch shortly.
> 
> My first thought was to agree, as m68k is indeed the only architecture
> that does not define NR_CPUS. Upon closer look, most architectures
> have NR_CPUS depend on SMP, hence I assume the issue could happen for
> those too (although I didn't manage to create such a config on anything
> but m68k)?  So the simple solution would be to add a dependency on
> SMP to SPLIT_PTE_PTLOCKS.
> 
Makes sense. I'll send v3.

Guenter