Hi, On Fri, 7 Sep 2018 09:19:06 -0700 Tony Lindgren <tony@xxxxxxxxxxx> wrote: > * Andreas Kemnade <andreas@xxxxxxxxxxxx> [180904 20:57]: > > Ok, what fixes the problem: > > use above mentioned devicetree, reverting 5b83b2234be6 and > > applying this patch: > > Hmm well the Linux generic wakeirqs are known to work, so > if there's a bug lurking around somewhere we should fix it > naturally :) > > But I'm wondering if you have some SDIO pins floating type > issue when your device enters deeper idle states? And maybe > your patches actually end up blocking the deeper idle states > hiding the issue? > I guess you have read my mails a bit too fast and overread many details, so I try to sumarize Summary: sdio irq + blocked runtime suspend = works sdio irq + pinctl wakeirq + 1 bit = works sdio irq + pinctl wakeirq + 4 bit = does not work sdio irq + wakeirq via remapped dat1 as gpio + 4 bit= works but too many interrupts if wlan is down (*) sdio irq + wakeirq via remapped dat1 as gpio + 4 bit -(=reverted) generic_wakeirq_patch = does not work sdio irq + wakeirq via remapped dat1 as gpio + 1 bit -(=reverted) generic_wakeirq_patch + reordering some things = works Note: works means: sane number of wake interrupts seen in a low-traffic wlan, network traffic works reliable and performant. And there are some interesting sections in the sdio simplificated spec regarding 4 bit mode and irqs missing... > If that's the case, you could define separate pinctrl > idle and default states. You can grep for something like > pinctrl-names = "default", "idle" for some examples. > > I fairly regularly test the mainline kernel with omap3 > hitting off mode during idle with wlan connected and > things are working for me with no issues and the device > wakes to a ping over wlan just fine. What sdio device are you testing 1bit or 4bit? with or without level shifter in between? At least I now have a working not-so-clean solution, so I have a base to experiment and analyze to get a better solution. (*) here probably comes into account that the wlan chip is powered of and as a side effect dat1 gets low (through a level shifter). As the generic wakeup irq is not turned off (contrary to the older solution) when sdio irq is disabled, it will always runtime resume quickly. Regards, Andreas
Attachment:
pgpEtCis4IrDX.pgp
Description: OpenPGP digital signature
