Re: [Intel-gfx] [PATCH] drm/i915: move i915_trace_irq_get() out of the tracing macro

[Daniel Vetter <daniel@ffwll.ch>]

On Fri, Jun 21, 2013 at 02:51:07PM +0200, Sebastian Andrzej Siewior wrote:
> On 06/21/2013 01:08 PM, Chris Wilson wrote:
> > On Fri, Jun 21, 2013 at 12:15:53PM +0200, Sebastian Andrzej Siewior wrote:
> >> There is a report on RT about "BUG: scheduling while atomic" because the
> >> sleeping lock is taken in tracing context. This patch simply moves
> >> locking operation out of the tracing macro.
> > 
> > No. This enables the IRQ, as well as making a number of
> > very expensively serialised read, unconditionally.
> 
> Ach in case CONFIG_TRACING is enabled but the tracepoint itself is
> disabled. In that case it is probably best to drop this tracepoint from
> -RT.

Iirc when we've last discussed this with Thomas Gleixner he mentioned that
there's a setup/teardown hook when enabling/disabling a tracepoint. We
could use that to enable/disable interrupts. Needs a notch of logic rework
though since the auto-irq disable code needs adjusting ... And on a 2nd
look that trace_irq_seqno logic looks a bit broken anyway.

It's somewhere on my todo list, but patches very much welcome.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch