linux-kernel - Re: [PATCH] regmap: regmap-irq: Make irq-type callbak optional

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Date:   Fri, 4 Jan 2019 14:14:34 +0200
From:   Matti Vaittinen <matti.vaittinen@...rohmeurope.com>
To:     Charles Keepax <ckeepax@...nsource.cirrus.com>
Cc:     mazziesaccount@...il.com, mikko.mutanen@...rohmeurope.com,
        heikki.haikola@...rohmeurope.com, broonie@...nel.org,
        gregkh@...uxfoundation.org, rafael@...nel.org,
        linux-kernel@...r.kernel.org, geert@...ux-m68k.org
Subject: Re: [PATCH] regmap: regmap-irq: Make irq-type callbak optional

Big thanks for testing this so early Charles =) It's nice to get this
report before I broke things more widely!

On Fri, Jan 04, 2019 at 11:32:07AM +0000, Charles Keepax wrote:
> On Fri, Jan 04, 2019 at 11:14:43AM +0000, Charles Keepax wrote:
> > On Fri, Jan 04, 2019 at 12:31:15PM +0200, Matti Vaittinen wrote:
> > > +	if (num_type_reg)
> > > +		d->irq_chip.irq_set_type = regmap_irq_set_type;
> > > +
> > 
> > Afraid this also causes regressions at my end, still having a bit
> > of a look but it looks like some how this prevents properties of the
> > IRQ getting passed along which causes my system to not probe
> > properly with:
> > 
> > genirq: Flags mismatch irq 58. 00002088 (cs35l35) vs. 00002088 (cs35l35)
> > cs35l35 0-0041: Failed to request IRQ: -16
> > 
> 
> My case is a shared IRQ with 2 amps (cs35l35) connected to a CODEC
> (wm8280).
> 
> So looks like the issue is if you don't have a set_type callback
> then the IRQ ends up as IRQF_TRIGGER_NONE, which causes the
> second IRQ to fail the middle check here in __setup_irq:
> 
> 		if (!((old->flags & new->flags) & IRQF_SHARED) ||
> 		    (oldtype != (new->flags & IRQF_TRIGGER_MASK)) ||
> 		    ((old->flags ^ new->flags) & IRQF_ONESHOT)) {

Right. Thanks for pinpointing the issue. This thing is a bit fishy.
There should not be a need for a 'fake type setting callback' when we
have shared IRQ with two or more devices who actually agree on IRQ type
- assuming the HW default irq type is correct.

For me the correct thing would that the HW default type should be
stored in desc and returned by irqd_get_trigger_type instead of
IRQF_TRIGGER_NONE. But I have no idea how that would be nicely
implemented w/o going trough all the irqchips.

The other option would be being more permissive when IRQF_TRIGGER_NONE
is set as oldtype. But I am not feeling like an expert on this area.

> Kinda inclined to just leave the fix as currently submitted and
> just drop this patch? But I can do more testing etc. if we want
> to push further down this road.

Huge thanks for the offer Charles - I do really appreciate the testing.
Touching the IRQ core sounds quite scary to me. Changing it feels risky
and would probably involve bunch of other people too. 10-years ago I
would have taken the challenge and tried to get this "correct" - but
nowadays I have learned to accept some small shortcuts :p

I will gladly follow things and participate the discussions/development
if someone wants to see how this should be done - but I don't think I
have the energy and time to drive this change further... So Mark, please
just drop this patch and keep the original fix unless you want to drive
this further yourself =)

Br,
	Matti Vaittinen

> 
> Thanks,
> Charles

-- 
Matti Vaittinen
ROHM Semiconductors

~~~ "I don't think so," said Rene Descartes.  Just then, he vanished ~~~