Message-ID: <59bc81b5-820e-40ff-9159-c03e429af9a6@gmx.de>
Date: Sat, 30 Dec 2023 08:33:24 +0100
From: Helge Deller <deller@....de>
To: Luis Chamberlain <mcgrof@...nel.org>
Cc: deller@...nel.org, linux-kernel@...r.kernel.org,
 Masahiro Yamada <masahiroy@...nel.org>, Arnd Bergmann <arnd@...db.de>,
 linux-modules@...r.kernel.org, linux-arch@...r.kernel.org
Subject: Re: [PATCH 2/4] modules: Ensure 64-bit alignment on __ksymtab_*
 sections

Hi Luis,

On 12/22/23 21:10, Luis Chamberlain wrote:
> On Fri, Dec 22, 2023 at 01:13:26PM +0100, Helge Deller wrote:
>> On 12/22/23 06:59, Luis Chamberlain wrote:
>>> On Wed, Nov 22, 2023 at 11:18:12PM +0100, deller@...nel.org wrote:
>>>> On 64-bit architectures without CONFIG_HAVE_ARCH_PREL32_RELOCATIONS
>>>> (e.g. ppc64, ppc64le, parisc, s390x,...) the __KSYM_REF() macro stores
>>>> 64-bit pointers into the __ksymtab* sections.
>>>> Make sure that those sections will be correctly aligned at module link time,
>>>> otherwise unaligned memory accesses may happen at runtime.
>>> ...
...
>> So, honestly I don't see a real reason why it shouldn't be applied...
>
> Like I said, you Cc'd stable as a fix,

I added "Cc: stable@...r.kernel.org" on the patch itself, so *if* the patch
would have been applied by you, it would later end up in stable kernel series too.
But I did not CC'ed the stable mailing list directly, so my patch was never
sent to that mailing list.

> as a maintainer it is my job to
> verify how critical this is and ask for more details about how you found
> it and evaluate the real impact. Even if it was not a stable fix I tend
> to ask this for patches, even if they are trivial.
> ...
> OK, can you extend the patch below with something like:
>
> perf stat --repeat 100 --pre 'modprobe -r b a b c' -- ./tools/testing/selftests/module/find_symbol.sh
>
> And test before and after?
>
> I ran a simple test as-is and the data I get is within noise, and so
> I think we need the --repeat 100 thing.

Your selftest code is based on perf.
AFAICS we don't have perf on parisc/hppa, so I can't test your selftest code
on that architecture.
I assume you tested on x86, where the CPU will transparently take care of
unaligned accesses. This is probably why the results are within
the noise.
But on some platforms the CPU raises an exception on unaligned accesses
and jumps into special exception handler assembler code inside the kernel.
This is much more expensive than on x86, which is why we track on parisc
in /proc/cpuinfo counters on how often this exception handler is called:
IRQ:       CPU0       CPU1
   3:       1332          0         SuperIO  ttyS0
   7:    1270013          0         SuperIO  pata_ns87415
  64:  320023012  320021431             CPU  timer
  65:   17080507   20624423             CPU  IPI
UAH:   10948640      58104   Unaligned access handler traps

This "UAH" field could theoretically be used to extend your selftest.
But is it really worth it? The outcome is very much architecture and CPU
specific, maybe it's just within the noise as you measured.

IMHO we should always try to natively align structures, and if we see
we got it wrong in kernel code, we should fix it.
My patches just fix those memory sections where we use inline
assembly (instead of C) and thus missed to provide the correct alignments.

Helge

> -----------------------------------------------------------------------------------
> before:
> sudo ./tools/testing/selftests/module/find_symbol.sh
>
>   Performance counter stats for '/sbin/modprobe test_kallsyms_b':
>
>          81,956,206 ns   duration_time
>          81,883,000 ns   system_time
>                 210      page-faults
>
>         0.081956206 seconds time elapsed
>
>         0.000000000 seconds user
>         0.081883000 seconds sys
>
>
>
>   Performance counter stats for '/sbin/modprobe test_kallsyms_b':
>
>          85,960,863 ns   duration_time
>          84,679,000 ns   system_time
>                 212      page-faults
>
>         0.085960863 seconds time elapsed
>
>         0.000000000 seconds user
>         0.084679000 seconds sys
>
>
>
>   Performance counter stats for '/sbin/modprobe test_kallsyms_b':
>
>          86,484,868 ns   duration_time
>          86,541,000 ns   system_time
>                 213      page-faults
>
>         0.086484868 seconds time elapsed
>
>         0.000000000 seconds user
>         0.086541000 seconds sys
>
> -----------------------------------------------------------------------------------
> After your modules alignement fix:
> sudo ./tools/testing/selftests/module/find_symbol.sh
>   Performance counter stats for '/sbin/modprobe test_kallsyms_b':
>
>          83,579,980 ns   duration_time
>          83,530,000 ns   system_time
>                 212      page-faults
>
>         0.083579980 seconds time elapsed
>
>         0.000000000 seconds user
>         0.083530000 seconds sys
>
>
>
>   Performance counter stats for '/sbin/modprobe test_kallsyms_b':
>
>          70,721,786 ns   duration_time
>          69,289,000 ns   system_time
>                 211      page-faults
>
>         0.070721786 seconds time elapsed
>
>         0.000000000 seconds user
>         0.069289000 seconds sys
>
>
>
>   Performance counter stats for '/sbin/modprobe test_kallsyms_b':
>
>          76,513,219 ns   duration_time
>          76,381,000 ns   system_time
>                 214      page-faults
>
>         0.076513219 seconds time elapsed
>
>         0.000000000 seconds user
>         0.076381000 seconds sys
>
> After your modules alignement fix:
> sudo ./tools/testing/selftests/module/find_symbol.sh
>   Performance counter stats for '/sbin/modprobe test_kallsyms_b':
>
>          83,579,980 ns   duration_time
>          83,530,000 ns   system_time
>                 212      page-faults
>
>         0.083579980 seconds time elapsed
>
>         0.000000000 seconds user
>         0.083530000 seconds sys
>
>
>
>   Performance counter stats for '/sbin/modprobe test_kallsyms_b':
>
>          70,721,786 ns   duration_time
>          69,289,000 ns   system_time
>                 211      page-faults
>
>         0.070721786 seconds time elapsed
>
>         0.000000000 seconds user
>         0.069289000 seconds sys
>
>
>
>   Performance counter stats for '/sbin/modprobe test_kallsyms_b':
>
>          76,513,219 ns   duration_time
>          76,381,000 ns   system_time
>                 214      page-faults
>
>         0.076513219 seconds time elapsed
>
>         0.000000000 seconds user
>         0.076381000 seconds sys
> -----------------------------------------------------------------------------------
>
> [perf-based selftest patch from Luis stripped]

Powered by blists - more mailing lists