Skip to contentSkip to conversation

Linux 3.1.5/6 regression: fails to resume from suspend (bisected)

PATCH Philip Langdale 2 participants 2 messages Dec 27, 2011 – Dec 29, 2011 High-resolution timers, timer wheel, clockevents

lore
Share & tools
lore
#0 lore raw

Hi,

After upgrading to 3.1.5, and still in 3.1.6, I found myself unable to resume from suspend. I did a bisect and identified the following change as the cause:

commit aeed6baa702a285cf03b7dc4182ffc1a7f4e4ed6

Author: Thomas Gleixner <[email protected]>
Date:   Fri Dec 2 16:02:45 2011 +0100

clockevents: Set noop handler in clockevents_exchange_device()

commit de28f25e8244c7353abed8de0c7792f5f883588c upstream.

If a device is shutdown, then there might be a pending interrupt, which will be processed after we reenable interrupts, which causes the original handler to be run. If the old handler is the (broadcast) periodic handler the shutdown state might hang the kernel completely.

kernel/time/clockevents.c +0 -0
index e4c699d..13dfaab 100644
@@ -286,6 +286,7 @@ void clockevents_exchange_device(struct

clock_event_device *old,

         * released list and do a notify add later.
         */
        if (old) {
+               old->event_handler = clockevents_handle_noop;
                clockevents_set_mode(old, CLOCK_EVT_MODE_UNUSED);
                list_del(&old->list);
                list_add(&old->list, &clockevents_released);

If I undo this change in my 3.1.6 tree, I am then able to resume as before.

This was also reported upstream at fedora but was not fully diagnosed:

https://bugzilla.redhat.com/show_bug.cgi?id=767248

I wouldn't be surprised if it's related to the nvidia binary drivers in some way (I use them and so does the fedora bug reporter), but it's not practical to avoid them.

--phil

#1 lore raw

2011/12/27 Philip Langdale <[email protected]>:

Quoted 5 lines

Hi,

After upgrading to 3.1.5, and still in 3.1.6, I found myself unable to resume from suspend. I did a bisect and identified the following change as the cause:

Hey,

I can confirm this breaks things.
My router box (a Soekris net5011-70) is unable to boot with this patch applied (it hangs when control is passed onto init, with the text "INIT:").
Reverting this patch makes it boot fine.

/Bjarke

Quoted 48 lines

commit aeed6baa702a285cf03b7dc4182ffc1a7f4e4ed6
Author: Thomas Gleixner <[email protected]>
Date:   Fri Dec 2 16:02:45 2011 +0100

   clockevents: Set noop handler in clockevents_exchange_device()

   commit de28f25e8244c7353abed8de0c7792f5f883588c upstream.

   If a device is shutdown, then there might be a pending interrupt, which will be processed after we reenable interrupts, which causes the original handler to be run. If the old handler is the (broadcast) periodic handler the shutdown state might hang the kernel completely.

kernel/time/clockevents.c +1 -0
index e4c699d..13dfaab 100644
@@ -286,6 +286,7 @@ void clockevents_exchange_device(struct
 clock_event_device *old,
         * released list and do a notify add later.
         */
        if (old) {
+               old->event_handler = clockevents_handle_noop;
                clockevents_set_mode(old, CLOCK_EVT_MODE_UNUSED);
                list_del(&old->list);

               list_add(&old->list, &clockevents_released);

If I undo this change in my 3.1.6 tree, I am then able to resume as before.

This was also reported upstream at fedora but was not fully diagnosed:

https://bugzilla.redhat.com/show_bug.cgi?id=767248

I wouldn't be surprised if it's related to the nvidia binary drivers in some way (I use them and so does the fedora bug reporter), but it's not practical to avoid them.

--phil
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/