On Thu, Dec 12, 2013 at 11:29 PM, Andreas Naumann <dev@xxxxxxxx> wrote: > Hi Grazvydas, > >> Von: Grazvydas Ignotas [mailto:notasas@xxxxxxxxx] >> Gesendet: Donnerstag, 12. Dezember 2013 01:21 >> An: linux-usb@xxxxxxxxxxxxxxx >> Cc: Felipe Balbi; linux-omap@xxxxxxxxxxxxxxx; Naumann Andreas; Grazvydas >> Ignotas; stable@xxxxxxxxxxxxxxx >> Betreff: [PATCH] usb: musb: omap2430: fix occasional musb breakage on boot >> >> >> This is a hard to reproduce problem which leads to non-functional >> USB-OTG port in 0.1%-1% of all boots. Tracked it down to commit >> e25bec160158abe86c "omap2+: save and restore OTG_INTERFSEL", >> which introduces save/restore of OTG_INTERFSEL over suspend. >> >> Since the resume function is also called early in driver init, it uses a >> non-initialized value (which is 0 and a non-supported setting in DM37xx >> for INTERFSEL). Shortly after the correct value is set. Apparently this >> works most time, but not always. >> >> Fix it by not writing the value on runtime resume if it is 0 >> (0 should never be saved in the context as it's invalid value, >> so we use it as an indicator that context hasn't been saved yet). >> >> This issue was originally found by Andreas Naumann: >> http://marc.info/?l=linux-usb&m=138562574719654&w=2 >> >> Reported-and-bisected-by: Andreas Naumann <anaumann@xxxxxxxxxxxxxx> >> Signed-off-by: Grazvydas Ignotas <notasas@xxxxxxxxx> >> Cc: <stable@xxxxxxxxxxxxxxx> >> --- >> This is a regression from 3.2, so should go to -rc and stable, IMO. >> It's really annoying issue if you want to have a stable OTG behavior, >> I've burned quite a lot of time on it myself over a year ago and gave up >> eventually. Good thing Andreas finally found it, many thanks to him! >> >> drivers/usb/musb/omap2430.c | 3 ++- >> 1 file changed, 2 insertions(+), 1 deletion(-) >> >> diff --git a/drivers/usb/musb/omap2430.c b/drivers/usb/musb/omap2430.c >> index 2a408cd..737b3da 100644 >> --- a/drivers/usb/musb/omap2430.c >> +++ b/drivers/usb/musb/omap2430.c >> @@ -672,7 +672,8 @@ static int omap2430_runtime_resume(struct device *dev) >> >> if (musb) { >> omap2430_low_level_init(musb); >> - musb_writel(musb->mregs, OTG_INTERFSEL, >> + if (musb->context.otg_interfsel != 0) >> + musb_writel(musb->mregs, OTG_INTERFSEL, >> musb->context.otg_interfsel); >> phy_power_on(musb->phy); >> } >> > > Oh, easy way out. I like it but I've also been thinking about your comment > on my original post, which was that initializing otg_interfsel to the PHYSEL > bits only might be dangerous because we cant be sure that there are other > bits in the register. > However, isnt assuming that 0 is invalid on all OMAPs just as dangerous? Well I was trying to do a minimal fix so that it could be suitable for merging to stable kernels. But yes you're right, I've just checked OMAP4 TRM and 0 is actually valid value there.. > After thinking about my patch again, I would propose to change otg_interfsel > into otg_physel and read-modify-write only those bits in resume() as you > suggested in your first answer. That way I could discard the problematic > first read in probe() while leaving other bits untouched. If you agree I > post a patch for this tomorrow. Hmm I don't know about that, this would be inconsistent with what all other OMAP drivers do. Maybe we should do what musb_core.c does just to be consistent and add a similar comment. Only the static variable could be avoided in favor of struct omap2430_glue member. Gražvydas -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
