Re: [PATCH] Linux Kernel Markers 0.13 for 2.6.17

[Mathieu Desnoyers <compudj@krystal.dyndns.org>]

* Mathieu Desnoyers (mathieu.desnoyers@polymtl.ca) wrote:
> Yes, preempt_disable() has a barrier(), on gcc :
> __asm__ __volatile__("": : :"memory").
> 
> 
> > Either way, this doesn't prevent some otherwise unrelated 
> > non-memory-using code from being scheduled in there, which would not be 
> > executed.  The gcc manual really strongly discourages jumping between 
> > inline asms, because they have basically unpredictable results.
> > 
> 
> Ok, I will do the call in assembly then.
> 

Before I rush on a solution too fast... I have a question for you :

To protect code from being preempted, the macros preempt_disable and
preempt_enable must normally be used. Logically, this macro must make sure gcc
doesn't interleave preemptible code and non-preemptible code.

Starting with this hypothesis, what makes gcc aware of this ? If we check
preempt_disable (the disable call is almost symmetric) :

linux/preempt.h:
define add_preempt_count(val) do { preempt_count() += (val); } while (0)

#define inc_preempt_count() add_preempt_count(1)

#define preempt_disable() \
do { \
        inc_preempt_count(); \
        barrier(); \
} while (0)

So the magic must be in the barrier() macro :

linux/compiler-gcc.h:
/* Optimization barrier */
/* The "volatile" is due to gcc bugs */
#define barrier() __asm__ __volatile__("": : :"memory")

Which makes me think that if I put barriers around my asm, call, asm trio, no
other code will be interleaved. Is it right ?

Mathieu


OpenPGP public key:              http://krystal.dyndns.org:8080/key/compudj.gpg
Key fingerprint:     8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68