[PATCH] Fix USB suspend/resume crasher

[Rafael J. Wysocki]

On Thursday, 24 of November 2005 22:01, Benjamin Herrenschmidt wrote:
> 
> > 
> > Well, it's there (actually the problem occurs in vanilla 2.6.15-rc2-mm1 that
> > contains the patch).  Do you mean it should go before the
> > 
> > if (readl(&ehci->regs->configured_flag) != FLAG_CF)
> > 		goto restart;
> > 
> > thing?
> 
> Yes.
> 
> > > It may be worth following it with a memory barrier actually... just in case
> > > (due to the absence of locks in that area).
> > 
> > wmb()?
> 
> Yup.
> 
> I wrote that patch against a tree that had different things in that
> function, Greg merged it by hand but he got that little bit wrong
> unfortunately. I'll send a new patch later today.

Thanks.

FWIW, does the appended change look reasonable to you?  (It apparently
helps. ;-))

Rafael


Index: linux-2.6.15-rc2-mm1/drivers/usb/host/ehci-pci.c
===================================================================
--- linux-2.6.15-rc2-mm1.orig/drivers/usb/host/ehci-pci.c	2005-11-24 21:42:34.000000000 +0100
+++ linux-2.6.15-rc2-mm1/drivers/usb/host/ehci-pci.c	2005-11-24 21:50:38.000000000 +0100
@@ -281,12 +281,13 @@
 	if (time_before(jiffies, ehci->next_statechange))
 		msleep(100);
 
+	set_bit(HCD_FLAG_HW_ACCESSIBLE, &hcd->flags);
+	wmb();
+
 	/* If CF is clear, we lost PCI Vaux power and need to restart.  */
 	if (readl(&ehci->regs->configured_flag) != FLAG_CF)
 		goto restart;
 
-	set_bit(HCD_FLAG_HW_ACCESSIBLE, &hcd->flags);
-
 	/* If any port is suspended (or owned by the companion),
 	 * we know we can/must resume the HC (and mustn't reset it).
 	 * We just defer that to the root hub code.