From: Julien Grall <julien.grall@arm.com>
To: Dave Martin <Dave.Martin@arm.com>
Cc: linux-arm-kernel@lists.infradead.org, bigeasy@linutronix.de,
linux-rt-users@vger.kernel.org, catalin.marinas@arm.com,
will.deacon@arm.com, ard.biesheuvel@linaro.org,
linux-kernel@vger.kernel.org
Subject: Re: [RFC PATCH] arm64/fpsimd: Don't disable softirq when touching FPSIMD/SVE state
Date: Thu, 11 Apr 2019 16:58:41 +0100 [thread overview]
Message-ID: <899713e0-2d32-04a0-b2f2-3493f7299033@arm.com> (raw)
In-Reply-To: <20190405150722.GE3567@e103592.cambridge.arm.com>
Hi Dave,
On 4/5/19 4:07 PM, Dave Martin wrote:
> On Fri, Apr 05, 2019 at 10:02:45AM +0100, Julien Grall wrote:
>>>> +#ifdef CONFIG_KERNEL_MODE_NEON
>>>> +
>>>> /*
>>>> * may_use_simd - whether it is allowable at this time to issue SIMD
>>>> * instructions or access the SIMD register file
>>>> diff --git a/arch/arm64/kernel/fpsimd.c b/arch/arm64/kernel/fpsimd.c
>>>> index 5ebe73b69961..b7e5dac26190 100644
>>>> --- a/arch/arm64/kernel/fpsimd.c
>>>> +++ b/arch/arm64/kernel/fpsimd.c
>>>> @@ -90,7 +90,8 @@
>>>> * To prevent this from racing with the manipulation of the task's FPSIMD state
>>>> * from task context and thereby corrupting the state, it is necessary to
>>>> * protect any manipulation of a task's fpsimd_state or TIF_FOREIGN_FPSTATE
>>>> - * flag with local_bh_disable() unless softirqs are already masked.
>>>> + * flag with kernel_neon_{disable, enable}. This will still allow softirqs to
>>>> + * run but prevent them to use FPSIMD.
>>>> *
>>>> * For a certain task, the sequence may look something like this:
>>>> * - the task gets scheduled in; if both the task's fpsimd_cpu field
>>>> @@ -142,6 +143,9 @@ extern void __percpu *efi_sve_state;
>>>> #endif /* ! CONFIG_ARM64_SVE */
>>>> +static void kernel_neon_disable(void);
>>>> +static void kernel_neon_enable(void);
>>>
>>> I find these names a bit confusing: _disable() actualy enables FPSIMD/SVE
>>> context access for the current context (i.e., makes it safe).
>>>
>>> Since these also disable/enable preemption, perhaps we can align them
>>> with the existing get_cpu()/put_cpu(), something like:
>>>
>>> void get_cpu_fpsimd_context();
>>> vold put_cpu_fpsimd_context();
>>>
>>> If you consider it's worth adding the checking helper I alluded to
>>> above, it could then be called something like:
>>>
>>> bool have_cpu_fpsimd_context();
>>
>> I am not sure where you suggested a checking helper above. Do you refer to
>> exposing kernel_neon_busy even for !CONFIG_KERNEL_MODE_NEON?
>
> Hmmm, looks like I got my reply out of order here.
>
> I meant the helper (if any) to check
> !preemptible() && !__this_cpu_read(kernel_neon_busy).
I guess you are using && instead of || because some of the caller may
not call get_cpu_fpsimd_context() before but still disable preemption,
right?
Wouldn't it be better to have all the user calling
get_cpu_fpsimd_context() and put_cpu_fpsimd_context()?
This has the advantage to uniformize how the way FPSIMD is protected and
also...
>
> Looks like you inferred what I meant later on anyway.
>
>>
>>>
>>>> +
>>>> /*
>>>> * Call __sve_free() directly only if you know task can't be scheduled
>>>> * or preempted.
>>>> @@ -213,11 +217,11 @@ static void sve_free(struct task_struct *task)
>>>> * thread_struct is known to be up to date, when preparing to enter
>>>> * userspace.
>>>> *
>>>> - * Softirqs (and preemption) must be disabled.
>>>> + * Preemption must be disabled.
>>>
>>> [*] That's not enough: we need to be in kernel_neon_disable()...
>>> _enable() (i.e., kernel_neon_busy needs to be true to prevent softirqs
>>> messing with the FPSIMD state).
>>
>> How about not mentioning preemption at all and just say:
>>
>> "The fpsimd context should be acquired before hand".
>>
>> This would help if we ever decide to protect critical section differently.
>
> Yes, or even better, name the function used to do this (i.e.,
> kernel_neon_disable() or get_cpu_fpsimd_context() or whatever it's
> called).
... would make the comments simpler because we would have only one
possible case to care.
Cheers,
--
Julien Grall
next prev parent reply other threads:[~2019-04-11 15:58 UTC|newest]
Thread overview: 29+ messages / expand[flat|nested] mbox.gz Atom feed top
2019-02-08 16:55 [RFC PATCH] arm64/fpsimd: Don't disable softirq when touching FPSIMD/SVE state Julien Grall
2019-02-12 17:13 ` Julia Cartwright
2019-02-18 14:07 ` Julien Grall
2019-02-13 14:30 ` Sebastian Andrzej Siewior
2019-02-13 15:36 ` Dave Martin
2019-02-13 15:40 ` Ard Biesheuvel
2019-02-13 15:42 ` Dave Martin
2019-02-13 16:52 ` Sebastian Andrzej Siewior
2019-02-14 10:34 ` Dave Martin
2019-02-18 15:07 ` Julien Grall
2019-03-04 17:25 ` Sebastian Andrzej Siewior
2019-03-14 18:07 ` Julien Grall
2019-03-15 10:06 ` Dave Martin
2019-03-15 10:22 ` Julien Grall
2019-02-13 16:50 ` Sebastian Andrzej Siewior
2019-02-18 14:33 ` Julien Grall
2019-02-18 16:32 ` Julien Grall
2019-04-04 10:52 ` Dave Martin
2019-04-05 9:02 ` Julien Grall
2019-04-05 14:39 ` Sebastian Andrzej Siewior
2019-04-05 15:17 ` Julien Grall
2019-04-05 15:42 ` Sebastian Andrzej Siewior
2019-04-11 15:12 ` Julien Grall
2019-04-05 15:07 ` Dave Martin
2019-04-11 15:58 ` Julien Grall [this message]
2019-04-11 16:34 ` Dave Martin
2019-04-11 16:50 ` Julien Grall
2019-04-11 14:10 ` Julien Grall
2019-02-08 17:03 [PATCH 0/3] Remove reference of TIF_USEDFPU on arch not using it Julien Grall
2019-02-08 17:03 ` [RFC PATCH] arm64/fpsimd: Don't disable softirq when touching FPSIMD/SVE state Julien Grall
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=899713e0-2d32-04a0-b2f2-3493f7299033@arm.com \
--to=julien.grall@arm.com \
--cc=Dave.Martin@arm.com \
--cc=ard.biesheuvel@linaro.org \
--cc=bigeasy@linutronix.de \
--cc=catalin.marinas@arm.com \
--cc=linux-arm-kernel@lists.infradead.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-rt-users@vger.kernel.org \
--cc=will.deacon@arm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).