[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <87zhajekt6.fsf@nanos.tec.linutronix.de>
Date: Thu, 07 May 2020 19:06:45 +0200
From: Thomas Gleixner <tglx@...utronix.de>
To: Alexandre Chartre <alexandre.chartre@...cle.com>,
LKML <linux-kernel@...r.kernel.org>
Subject: Re: [patch V4 part 2 05/18] x86/entry: Move irq tracing on syscall entry to C-code
Thomas Gleixner <tglx@...utronix.de> writes:
> Thomas Gleixner <tglx@...utronix.de> writes:
>> Alexandre Chartre <alexandre.chartre@...cle.com> writes:
>>> On 5/5/20 3:41 PM, Thomas Gleixner wrote:
>>>> - /*
>>>> - * User mode is traced as though IRQs are on, and the interrupt
>>>> - * gate turned them off.
>>>> - */
>>>> - TRACE_IRQS_OFF
>>>> -
>>>> movq %rsp, %rdi
>>>> call do_int80_syscall_32
>>>> .Lsyscall_32_done:
>>>>
>>>
>>> enter_from_user_mode() is also called with the CALL_enter_from_user_mode macro,
>>> which is used in interrupt_entry() and identry. Don't you need to also remove
>>> the TRACE_IRQS_OFF there now?
>>
>> Hrm. right. OTOH, it's just redundant and should be no harm, but let me have a
>> look at that again.
>
> Grr, no. It'll trigger the warnon when context tracking is enabled. /me
> scratches head and goes to fix.
Scratch that. After unfrying my brain by walking the dogs for an hour,
it's really just redundant calls into lockdep and tracing and both are
happy about it.
I could do a temporary function for that or just mention it in the
changelog.
Thanks,
tglx
Powered by blists - more mailing lists