Message-ID: <20080425214704.GD25950@Krystal>
Date:	Fri, 25 Apr 2008 17:47:04 -0400
From:	Mathieu Desnoyers <mathieu.desnoyers@...ymtl.ca>
To:	"H. Peter Anvin" <hpa@...or.com>
Cc:	Linus Torvalds <torvalds@...ux-foundation.org>,
	Andi Kleen <andi@...stfloor.org>, Ingo Molnar <mingo@...e.hu>,
	Jiri Slaby <jirislaby@...il.com>,
	David Miller <davem@...emloft.net>, zdenek.kabelac@...il.com,
	rjw@...k.pl, paulmck@...ux.vnet.ibm.com, akpm@...ux-foundation.org,
	linux-ext4@...r.kernel.org, herbert@...dor.apana.org.au,
	penberg@...helsinki.fi, clameter@....com,
	linux-kernel@...r.kernel.org, pageexec@...email.hu,
	Jeremy Fitzhardinge <jeremy@...p.org>
Subject: Re: [PATCH 1/1] x86: fix text_poke

* H. Peter Anvin (hpa@...or.com) wrote:
> Mathieu Desnoyers wrote:
>> The point is to provide a way to dynamically enable code at runtime
>> without noticeable performance impact on the system. It's principally
>> useful to control the markers in the kernel, which can be placed in very
>> frequently executed code paths. The original markers add a memory read,
>> test and conditional branch at each marker site. By using the immediate
>> values patchset, it goes down to a load immediate value, test and branch.
>> However, Ingo was still unhappy with the conditional branch, so I cooked
>> this jump patching optimization on top of the immediate values. It
>> looks for an expected pattern which limits the liveliness of the %al and
>> ZF registers to the 3 instructions and, if it finds it, patches a jump
>> located just before the mov instruction to skip the whole pattern and
>> behave exactly like the conditional branch.
>> So basically we get code dynamically actvated by patching a single jump.
>
> Note that all these optimizations only make sense if the case where we 
> *take* the "marker" is frequent, *and* the marker itself is not too 
> expensive.
>

Yes, this is the case. Using breakpoints for markers quickly becomes
noticeable for thing such as scheduler instrumentation, page fault
handler instrumentation, etc. And yes, I have developed kernel tracer,
LTTng, which takes care of writing the data to trace buffers
efficiently. The last time I took performance measurements, it was
performing locking and writing to the memory buffer in about 270ns on a
3GHz Pentium 4. It might be a tiny bit slower now that it parses the
markers format strings dynamically, but nothing very significant.

But there is another point that markers do which the breakpoint won't
give you : they extract local variables from functions and they identify
them with field names which separates the instrumentation from the
actual kernel implementation details. In order to do that, I rely on gcc
building a stack frame for a function call, which I don't want to build
unnecessarity when the marker is disabled. This is why I use a jump to
skip passing the arguments on the stack and the function call.

Mathieu

> If that is not the case, just put in a noop that is dynamically patched to 
> an INT3 or ICEBP instruction (one byte) or an INT instruction (two bytes), 
> take the exception, look up the source address and revector to the marker 
> code.
>
> 	-hpa

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists