linux-kernel - Re: [PATCH] MIPS: SMP: Move the AP sync point before the non-parallel aware functions

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: <87frgvokga.fsf@BLaptop.bootlin.com>
Date: Fri, 23 May 2025 09:27:49 +0200
From: Gregory CLEMENT <gregory.clement@...tlin.com>
To: Huacai Chen <chenhuacai@...nel.org>
Cc: Thomas Bogendoerfer <tsbogend@...ha.franken.de>, Jiaxun Yang
 <jiaxun.yang@...goat.com>, Vladimir Kondratiev
 <vladimir.kondratiev@...ileye.com>, Théo Lebrun
 <theo.lebrun@...tlin.com>,
 Tawfik Bayouk <tawfik.bayouk@...ileye.com>, Thomas Petazzoni
 <thomas.petazzoni@...tlin.com>, linux-mips@...r.kernel.org,
 linux-kernel@...r.kernel.org
Subject: Re: [PATCH] MIPS: SMP: Move the AP sync point before the
 non-parallel aware functions

Hello Huacai,

> Hi, Gregory,
>
> On Mon, May 5, 2025 at 8:58 PM Gregory CLEMENT
> <gregory.clement@...tlin.com> wrote:
>>
>> When CONFIG_HOTPLUG_PARALLEL is enabled, the code executing before
>> cpuhp_ap_sync_alive() is executed in parallel, while after it is
>> serialized. The functions set_cpu_sibling_map() and set_cpu_core_map()
>> were not designed to be executed in parallel, so by moving the
>> cpuhp_ap_sync_alive() before cpuhp_ap_sync_alive(), we then ensure
>> they will be called serialized.
> set_cpu_sibling_map() and set_cpu_core_map() are the most obvious
> functions that need serialization, but you'd better check other places
> to make sure there are no similar functions executed in parallel.

I conducted an exhaustive review of all called functions, which allowed
me to identify an opportunity for improving boot time:
[1]. Additionally, I noted that CPU delay calibration is the only
remaining part where concurrent access could occur. This was indeed the
case prior to my previous patch submission[1]. To ensure safety, I am
about sending a new fix addressing this issue.

Gregory

[1]: https://lore.kernel.org/linux-mips/20250520-smp_calib-v1-1-cd04f0a78648@bootlin.com/
>
> Huacai
>
>>
>> The measurement done on EyeQ5 did not show any relevant boot time
>> increase after applying this patch.
>>
>> Reported-by: Huacai Chen <chenhuacai@...nel.org>
>> Signed-off-by: Gregory CLEMENT <gregory.clement@...tlin.com>
>> ---
>> Hello,
>>
>> As discussed last week [1], this is the patch that fixes the potential
>> issue with the functions set_cpu_sibling_map() and set_cpu_core_map().
>>
>> Gregory
>>
>> [1]: https://lore.kernel.org/all/aBVBHFGH2kICjnT3@alpha.franken.de/
>> ---
>>  arch/mips/kernel/smp.c | 6 +++---
>>  1 file changed, 3 insertions(+), 3 deletions(-)
>>
>> diff --git a/arch/mips/kernel/smp.c b/arch/mips/kernel/smp.c
>> index 1726744f2b2ec10a44420a7b9b9cd04f06c4d2f6..7901b59d8f60eddefc020cf2a137716af963f09e 100644
>> --- a/arch/mips/kernel/smp.c
>> +++ b/arch/mips/kernel/smp.c
>> @@ -374,13 +374,13 @@ asmlinkage void start_secondary(void)
>>         calibrate_delay();
>>         cpu_data[cpu].udelay_val = loops_per_jiffy;
>>
>> +#ifdef CONFIG_HOTPLUG_PARALLEL
>> +       cpuhp_ap_sync_alive();
>> +#endif
>>         set_cpu_sibling_map(cpu);
>>         set_cpu_core_map(cpu);
>>
>>         cpumask_set_cpu(cpu, &cpu_coherent_mask);
>> -#ifdef CONFIG_HOTPLUG_PARALLEL
>> -       cpuhp_ap_sync_alive();
>> -#endif
>>         notify_cpu_starting(cpu);
>>
>>  #ifndef CONFIG_HOTPLUG_PARALLEL
>>
>> ---
>> base-commit: 3b3704261e851e25983860e4c352f1f73786f4ab
>> change-id: 20250505-hotplug-paralell-fix-25224cd229c6
>>
>> Best regards,
>> --
>> Grégory CLEMENT, Bootlin
>> Embedded Linux and Kernel engineering
>> https://bootlin.com
>>

-- 
Grégory CLEMENT, Bootlin
Embedded Linux and Kernel engineering
https://bootlin.com