Re: [PATCH resend regression?] AHCI: Fix threaded interrupt setup

[Marc Zyngier <marc.zyngier@arm.com>]

Hi Hans,

On 13/11/14 09:44, Hans de Goede wrote:
> Hi Tejun,
> 
> Since this patch mentioned sunxi it gained my attention, but as it touches
> ahci core code I was expecting you to pick it up, which seems to not have
> happened, so hence this resend.
> 
> FWIW I've tried to reproduce the problem, but it does not reproduce for me,
> that does not mean it is not a real problem though. I'm afraid I'm to
> unfamiliar with the ahci interrupt handling to say anything more sensible
> about this.

I can reproduce it about once every 5 boots or so. You can also notice
that the interrupt rate is much higher with the threaded interrupt code,
which is a sign that something is very fishy (the level interrupt is
never masked...).

> It looks like this is a 3.18 regression, so if this is a real issue, we
> should probably get this patch in for 3.18 .

I think the real fix should be to revert the whole threaded interrupt
thing. Adding the IRQF_ONESHOT breaks some Intel platforms in a rather
bad way in case of a shared interrupt (for gory details, see
http://www.gossamer-threads.com/lists/linux/kernel/2043506).

Tejun did prepare a set of patches to revert this, but I haven't seen it
being pulled in yet. It would be good to get something into 3.18.

Thanks,

	M.
-- 
Jazz is not dead. It just smells funny...