Re: network driver disabled interrupts in PREEMPT_RT

[Ingo Molnar <mingo@elte.hu>]

* Christoph Hellwig <hch@infradead.org> wrote:

> On Mon, Jun 13, 2005 at 08:56:42PM +0200, Ingo Molnar wrote:
> > 
> > * Kristian Benoit <kbenoit@opersys.com> wrote:
> > 
> > > Hi,
> > > I got lots of these messages when accessing the net running
> > > 2.6.12-rc6-RT-V0.7.48-25 :
> > > 
> > > "network driver disabled interrupts: tg3_start_xmit+0x0/0x629 [tg3]"
> > > 
> > > it seem to come from net/sched/sch_generic.c.
> > 
> > does the patch below fix it?
> 
> Wouldn't it be much more useful to add spin_trylock_irq?

you are right, a spin_trylock_irqsave() call would be much cleaner.

i think i should explain why you are seeing such hacks in the PREEMPT_RT 
patch: generally we are trying to achieve near-100% driver API 
compatibility (that of course is also one of the requirements for any of 
this to hit mainline), so i'm marking any deviations with _nort or _rt 
postfixes (depending on where the change in semantics is needed). I 
consider them temporary hacks, so i'm usually trying to keep the 
original form of the code, so that i can go back later and fix it.

I had these hacks in tg3.c for some time, and i thought i could drop 
them now that local_irq_*() uses the soft IRQ-flag - but i was wrong.  
One example of the _rt/_nort marking process is e.g. the earlier RCU API 
related grossness, which went away with Paul's PREEMPT_RCU aproach and 
now PREEMPT_RT is fully compatible with the RCU API. Fortunately, 
assymetric local_irq_* + spin_lock_* uses (which are perfectly legal!)  
are relatively rare. Extending the spinlock APIs and converting all 
upstream code would be a good approach to solve this problem, as it's 
also a cleanup. (It would probably also make static lock analysis 
easier.)

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