linux-mips
[Top] [All Lists]

Re: [RFC] count TLB refill for Netlogic XLR chip

To: "Jayachandran C." <jayachandranc@netlogicmicro.com>
Subject: Re: [RFC] count TLB refill for Netlogic XLR chip
From: Hillf Danton <dhillf@gmail.com>
Date: Sun, 2 Oct 2011 15:13:29 +0800
Cc: linux-mips@linux-mips.org, Ralf Baechle <ralf@linux-mips.org>
Dkim-signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=gamma; h=mime-version:in-reply-to:references:date:message-id:subject:from:to :cc:content-type; bh=InfL9PfFLh7NXF75ElkvYYLBfxg1FvvPycS5vaOkt8M=; b=pAwkfxtruTkfyXVdiDIEHCnsRmzaFUU9VsydV/wH2yfn547+DnLgTEU92R/XiqEZWO HZEjjmQgIteW41NeTplA+oG/wiszcgF0N0G4n9cr1t+YzNVQZhCBr4ndZSceyDH3EzXB ArPZX8vMLo1CGViLHEotOkQQFL3JmPzNjCx98=
In-reply-to: <20111001073936.GA15674@jayachandranc.netlogicmicro.com>
References: <CAJd=RBDQ9eyfgWkgsdUrojteqbnribZyk0QATr3CgPXLbBDkPQ@mail.gmail.com> <20111001073936.GA15674@jayachandranc.netlogicmicro.com>
Sender: linux-mips-bounce@linux-mips.org
On Sat, Oct 1, 2011 at 3:39 PM, Jayachandran C.
<jayachandranc@netlogicmicro.com> wrote:
> On Sat, Oct 01, 2011 at 01:19:53PM +0800, Hillf Danton wrote:
>> TLB miss is one of the concerned factors when tuning the performance of user
>> applications, and there are on netlogic XLR chip eight 64-bit registers,
>> c0 register 22 select 0-7, which could be used as temporary storage.
>>
>> One of them is used for counting TLB refill, and any comment is appreciated.
>>
>
> Few comments:
>
> Adding this unconditionally will add overhead to the TLB refill handler, this
> should be probably controlled by a boot-time variable, or with some
> perf/profile framework.
>

Boot-time variable will be added.

> If you have access to our SDK, it uses register 22,2 for counting TLB misses
> it would be good to retain that compability, otherwise it will mess up when
> we merge this back to our internal code.
>

It is clear for all users if the usage of reg 22 is defined in , say,
mips-extns.h?

> The reporting part currently will only print the misses on a crash (if I am
> not mistaken), again this has to be tied in with some standard framework.
>

How about one line added in /proc/interrupts?

Thanks
Hillf

<Prev in Thread] Current Thread [Next in Thread>