Re: kprobes performance on i386 vs x86_64

[Jim Keniston <jkenisto at us dot ibm dot com>]

On Tue, 2007-01-30 at 15:10, Satoshi Oshima wrote:
> Martin Hunt wrote:
> > Is there a reason why i386 kprobes are 2-3 times faster than x86_64 on
> > 2.6.19? This is on the same hardware. Is djprobes only working on i386
> > or something?
> 
> True. Kprobes for i386 has booster but for x86_64 doesn't have booster.
> 
> Sorry that's my fault.
> 
> Today, I rebase k*probe-booster x86_64 patch set to latest mm.
> and now under testing.
> 
> It will be posted as soon as it passes the tests.
> 
> Satoshi

It would be good for future k[ret]probe booster patches (and other
performance patches) to include an appropriate patch to the "Probe
Overhead" section of Documentation/kprobes.txt.  You don't have to test
on the same hardware we did; just specify the parameters of the hardware
you used.

The attached tarball contains the (ancient) test modules I used to get
the performance numbers in kprobes.txt.

Jim

Attachment: kprobe_timings.tar.gz
Description: GNU Zip compressed data