On Mon, 2 Nov 2015, Michael Reutman wrote:
> On Wed, Oct 28, 2015 at 10:48 AM, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
> > The patch below is meant to apply on top of the previous patch. It's a
> > quick-and-dirty workaround which ought to prevent the problem from
> > occurring. Let's see if it does.
>
> Unfortunately, I have some bad news. It appears that the last patch
> just pushes the issue further on down in time. Ran the test fresh
> today and usb got "stuck" in around 15 to 20 minutes. This has
> definitely improved things from before where it took less than a
> minute to get everything out of sorts, so maybe we are on the right
> track?
Here's a revised quick-and-dirty workaround. It replaces the earlier
one. (And like the earlier one, it is meant to apply on top of
the usbfs-snoop patch, but I don't think that makes any difference
now.)
The idea is to wait until the QH has remain unchanged for at least 2 ms
before deciding that the host controller has stopped using it. If this
doesn't fix the problem then I will be completely out of ideas and will
have to ask for help from someone at AMD.
When testing this, don't bother with the usbfs-snoop stuff. With luck
it won't matter, because the test program won't hang.
Alan Stern
Index: usb-4.3/drivers/usb/host/ehci-q.c
===================================================================
--- usb-4.3.orig/drivers/usb/host/ehci-q.c
+++ usb-4.3/drivers/usb/host/ehci-q.c
@@ -1341,8 +1341,22 @@ static void end_unlink_async(struct ehci
}
/* Otherwise process only the first waiting QH (NVIDIA bug?) */
- else
+ else {
+ __hc32 qh_current, qh_token, old_current, old_token;
+
list_move_tail(&qh->unlink_node, &ehci->async_idle);
+ old_current = qh->hw->hw_current;
+ old_token = qh->hw->hw_token;
+ for (;;) {
+ udelay(2000);
+ qh_current = READ_ONCE(qh->hw->hw_current);
+ qh_token = READ_ONCE(qh->hw->hw_token);
+ if (qh_current == old_current && qh_token == old_token)
+ break;
+ old_current = qh_current;
+ old_token = qh_token;
+ }
+ }
/* Start a new IAA cycle if any QHs are waiting for it */
if (!list_empty(&ehci->async_unlink))
--
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
