Re: irq-disabled vs vmap vs text_poke

[Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>]

* Steven Rostedt (rostedt@goodmis.org) wrote:
> 
> On Tue, 17 Feb 2009, Masami Hiramatsu wrote:
> > 
> > Thanks, I think text_poke() doesn't need high performance currently,
> > because it's not called so frequently, nor from the normal operation.
> > 
> > However, Would dynamic ftrace need performance?
> 
> Are you talking about replacing what dynamic ftrace does now with 
> text_poke? Well, It takes just under a second right now to do all the 
> conversions. Looking at the list on my booted box, it converts 19805 
> locations. It takes care to run stop machine, and it also has code to 
> handle NMIs on the other CPUS while it runs.
> 
> Changing text must be done extremely carefully on a running box. If the 
> code being changed is also executed on another CPU then you will most 
> likely take a GPF on that CPU.
> 
> Also, every arch must do things a bit different, due to the way modules 
> are handled.
> 
> -- Steve
> 

Anyway dynamic ftrace only need to do code patching at early boot,
right? Or do you need also to patch the call sites a bit later too ?
Because there is a text_poke_early for that purpose : modifying code
when it is still writable.

Mathieu


-- 
Mathieu Desnoyers
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68