On Thu, 10 Oct 2019 09:29:45 +0200
Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote:
> On Wed, 9 Oct 2019 at 18:43, Tony Lindgren <tony@xxxxxxxxxxx> wrote:
> >
> > Commit 572cf7d7b07d ("ARM: dts: Improve omap l4per idling with wlcore edge
> > sensitive interrupt") changed wlcore interrupts to use edge interrupt based
> > on what's specified in the wl1835mod.pdf data sheet.
> >
> > However, there are still cases where we can have lost interrupts as
> > described in omap_gpio_unidle(). And using a level interrupt instead of edge
> > interrupt helps as we avoid the check for untriggered GPIO interrupts in
> > omap_gpio_unidle().
> >
> > And with commit e6818d29ea15 ("gpio: gpio-omap: configure edge detection
> > for level IRQs for idle wakeup") GPIOs idle just fine with level interrupts.
> >
> > Let's change omap4 and 5 wlcore users back to using level interrupt
> > instead of edge interrupt. Let's not change the others as I've only seen
> > this on omap4 and 5, probably because the other SoCs don't have l4per idle
> > independent of the CPUs.
>
> I assume this relates to the implementation for support of SDIO IRQs
> (and wakeups) in the omap_hsmmc driver?
>
> In any case, just wanted to share some experience in the field, feel
> free to do whatever you want with the below information. :-)
>
> So, while I was working for ST-Ericsson on ux500, we had a very
> similar approach to re-route the SDIO bus DAT1 line to a GPIO IRQ as a
> remote/system wakeup (vendor hack in the mmci driver). In other words,
> while runtime suspending the mmc host controller, we configured a GPIO
> IRQ, via an always on logic, to capture the IRQ instead. The point is,
> I believe we may have ended up looking at similar problems as you have
> been facing on OMAP.
>
> In hindsight, I realized that we actually violated the SDIO spec by
> using this approach. More precisely, during runtime suspend we do
> clock gating and then re-routes the IRQ. However, clock gating isn't
> allowed before the SDIO bus width have been changed back from 4-bit
> into 1-bit. This last piece of action, would be an interesting change
> to see if it could affect the behaviour, but unfortunately I have
> never been able to check this.
>
> The tricky part, is that we can't issue a command to change the bus to
> 1-bit in omap_hsmmc ->runtime_suspend() callback (this needs to be
> managed by the core in some way). However, we can make a simple test,
> by simply always limit the bus width to 1-bit, as that should mean we
> should conform to the SDIO spec.
>
somehow matches that with my experiences with libertas + omap3.
SDIO irq seems to work only with runtime force-enabled in omap_hsmmc
or using 1bit mode.
And yes, I tried switching mode to 1bit in runtime_suspend() but as
you said, that cannot easily done.
Regards,
Andreas
Attachment:
pgpzLLJ2IPF3a.pgp
Description: OpenPGP digital signature
