Re: [PATCH 1/2] IB/hfi1: Use preempt_{dis,en}able_nort()

[Arnaldo Carvalho de Melo]

Em Thu, Oct 05, 2017 at 09:17:44AM -0500, Julia Cartwright escreveu:
> On Tue, Oct 03, 2017 at 12:49:19PM -0300, Arnaldo Carvalho de Melo wrote:
> > +++ b/drivers/infiniband/hw/hfi1/pio.c
> > @@ -1421,7 +1421,7 @@ struct pio_buf *sc_buffer_alloc(struct send_context *sc, u32 dw_len,

> >  	/* there is enough room */

> > -	preempt_disable();
> > +	preempt_disable_nort();
> >  	this_cpu_inc(*sc->buffers_allocated);

> Have you tried this on RT w/ CONFIG_DEBUG_PREEMPT?

No

> I believe that the this_cpu_* operations perform a preemption check, which we'd trip.

Humm, looking at include/linux/percpu-defs.h on v4.11.12-rt14 I see
(trimmed to what we're discussing here):

#ifdef CONFIG_DEBUG_PREEMPT
extern void __this_cpu_preempt_check(const char *op);
#else
static inline void __this_cpu_preempt_check(const char *op) { }
#endif

#define __this_cpu_add(pcp, val)					\
({									\
	__this_cpu_preempt_check("add");				\
	raw_cpu_add(pcp, val);						\
})
#define __this_cpu_inc(pcp)		__this_cpu_add(pcp, 1)

/*
 * Operations with implied preemption/interrupt protection.  These
 * operations can be used without worrying about preemption or interrupt.
 */
#define this_cpu_add(pcp, val)          __pcpu_size_call(this_cpu_add_, pcp, val)
#define this_cpu_inc(pcp)               this_cpu_add(pcp, 1)
 
> You may also have to change these to the non-preempt checked variants.

So __this_cpu_inc() checks preemption but this_cpu_inc() doesn't and
thus we're ok here? Or am I getting lost in this maze of defines? :-)

- Arnaldo