Re: [PATCH 3/3] x86/ftrace: Use text_poke()

From: Peter Zijlstra
Date: Wed Oct 02 2019 - 14:21:19 EST


On Wed, Oct 02, 2019 at 06:35:26PM +0200, Daniel Bristot de Oliveira wrote:

> ftrace was already batching the updates, for instance, causing 3 IPIs to enable
> all functions. The text_poke() batching also works. But because of the limited
> buffer [ see the reply to the patch 2/3 ], it is flushing the buffer during the
> operation, causing more IPIs than the previous code. Using the 5.4-rc1 in a VM,
> when enabling the function tracer, I see 250+ intermediate text_poke_finish()
> because of a full buffer...
>
> Would this be the case of trying to use a dynamically allocated buffer?
>
> Thoughts?

Is it a problem? I tried growing the buffer (IIRC I made it 10 times
bigger) and didn't see any performance improvements because of it.