On Mon, 13 Aug 2012 17:32:34 +0300 Felipe Balbi <balbi@ti.com> wrote:

> Hi,
> 
> On Mon, Aug 13, 2012 at 12:34:53PM +1000, NeilBrown wrote:
> > On Thu, 9 Aug 2012 14:15:51 +0300 Felipe Balbi <balbi@ti.com> wrote:
> > 
> > 
> > > hehe, that's nasty. Please send a patch converting to a try count and a
> > > udelay_range(), or something.
> > > 
> > 
> > how's this?
> > 
> > Thanks,
> > NeilBrown
> > 
> > 
> > From: NeilBrown <neilb@suse.de>
> > Date: Mon, 13 Aug 2012 12:32:58 +1000
> > Subject: [PATCH] omap2430: don't loop indefinitely in interrupt.
> > 
> > When called during resume_irqs, omap2430_musb_set_vbus() is run with
> > interrupts disabled,  In that case 'jiffies' never changes so the loop
> > can loop forever.
> > 
> > So impose a maximum loop count and add an 'mdelay' to ensure we wait
> > a reasonable amount of time for bit to be cleared.
> > 
> > This fixes a hang on resume.
> > 
> > Signed-of-by: NeilBrown <neilb@suse.de>
> > 
> > diff --git a/drivers/usb/musb/omap2430.c b/drivers/usb/musb/omap2430.c
> > index c7785e8..8a93381 100644
> > --- a/drivers/usb/musb/omap2430.c
> > +++ b/drivers/usb/musb/omap2430.c
> > @@ -34,6 +34,7 @@
> >  #include <linux/dma-mapping.h>
> >  #include <linux/pm_runtime.h>
> >  #include <linux/err.h>
> > +#include <linux/delay.h>
> >  
> >  #include "musb_core.h"
> >  #include "omap2430.h"
> > @@ -145,6 +146,7 @@ static void omap2430_musb_set_vbus(struct musb *musb, int is_on)
> >  
> >  	if (is_on) {
> >  		if (musb->xceiv->state == OTG_STATE_A_IDLE) {
> > +			int loops = 100;
> >  			/* start the session */
> >  			devctl |= MUSB_DEVCTL_SESSION;
> >  			musb_writeb(musb->mregs, MUSB_DEVCTL, devctl);
> > @@ -154,9 +156,11 @@ static void omap2430_musb_set_vbus(struct musb *musb, int is_on)
> >  			 */
> >  			while (musb_readb(musb->mregs, MUSB_DEVCTL) & 0x80) {
> >  
> > +				mdelay(5);
> 
> I would prefer udelay_range() as it will let scheduler group timers.
> Something like:
> 
> udelay_range(3000, 5000);
> 
> should do, I gues...
> 

Except that there is no udelay_range :-(
There is a usleep_range, but that can only be used from non-atomic context
and in the problem case interrupts are disabled and a spinlock is held so we
very definitely are not in non-atomic context.  If we need a delay at all, it
has to be udelay or mdelay.
If we could do this in a work function rather than directly from the
interrupt handler that would be best but I have no idea what dependencies
there are..  Would it be safe for musb_stage0_irq() to ask a workqueue to run
musb_platform_set_vbus rather than doing it directly?

NeilBrown