Re: [PATCH 5/6] x86: patch all traced function calls using theint3-based framework

[Steven Rostedt]

On Fri, 18 Oct 2013 10:55:57 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Fri, 18 Oct 2013 16:27:24 +0200
> Petr Mladek <pmladek@xxxxxxx> wrote:
> 
> 
> > +/*
> > + * We do not want to replace ftrace calls one by one because syncing all CPUs
> > + * is quite expensive.
> > + *
> > + * On the other hand, we do not want to modify all calls at once because
> > + * the buffers for text_poke_bp might be quite huge. Also we do not want
> > + * to count the number of records before the allocation.
> > + *
> > + * Let's modify the call in batches defined by this define.
> > + */
> > +#define FTRACE_MAX_RECS_PATCH 8192
> >  
> > -static int finish_update(struct dyn_ftrace *rec, int enable)
> > +static int ftrace_allocate_replace_buffers(unsigned long **addr, void **opcode)
> 
> I absolutely hate this. Current ftrace conversion does not need to
> allocate at all. I want to keep it that way.
> 

I should clarify that I do not hate the patch set. I actually like what
it's trying to achieve (a lot!). But I absolutely hate having to copy
records to do the bulk conversion. That part I'm giving a NAK to.

Add a text_poke_bp_iterate() or something, that removes the need for
allocating extra buffers, then I'll be very happy :-)

-- Steve
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/