Re: 3.12.x looses serial mouse over hibernate + resume

From: Shuah Khan
Date: Mon Dec 02 2013 - 13:45:16 EST


On 12/02/2013 11:35 AM, Manuel Krause wrote:
On 2013-12-02 17:45, Dmitry Torokhov wrote:
On Mon, Dec 02, 2013 at 08:38:16AM -0800, Dmitry Torokhov wrote:
On Monday, December 02, 2013 05:08:28 PM Manuel Krause wrote:
On 2013-12-01 16:43, Peter Hurley wrote:
[ +cc Dmitry Torokhov, Greg Kroah-Hartman, linux-input,
linux-serial ]

On 11/26/2013 05:19 PM, Manuel Krause wrote:
Since kernel 3.12.0 I have a problem with hibernate+resume
not reactivating my serial mouse (trackball) with my HP notebook.
Kernels 3.11.0 til 9 don't show this behaviour.

Machine: HP Notebook with Core2Duo CPU (Penryn)
Distro: openSUSE 12.3, 64bit, continuously updated
Desktop: KDE 4.11.3
MESA & drm & Xorg: most recent ones from:

http://download.opensuse.org/repositories/home:/pontostroy:/X11/openSUSE_

12.3/x86_64/

Current kernel: 3.12.1 vanilla from openSUSE repos, with

-ck1 and BFQ patches

The Logitech Trackman Marble FX is a PS/2 device and connected
via an original Logitech
PS/2-COM-port adapter and manually configured via my xorg.conf.

At first, I blamed the -ck1 patches from Con Kolivas for this
behaviour that I use in
addition to the BFQ patches, what has showed up as not right:
This happens with the
normal vanilla kernel
schedulers for CPU and disk I/O, too.

By coincidence I found a weird(!) way to reactivate the serial
mouse:
(1) call Hibernate (suspend-to-disk) from KDE desktop as normal
(2) resume --> the PS/2 touchpad is working, the serial
trackball NOT
(3) call suspend-to-RAM (Sleep) from KDE, serial trackball
still dead
(4) execute `setserial -a /dev/ttyS0` in a konsole window or a
tty* console
(5) ==> serial trackball is back with all configuration from
xorg.conf

It's fully reproducible over multiple hibernations. This also
happens when calling
`pm-hibernate` (to-disk) and `pm-suspend` (to-RAM) and the
setserial from a root shell
in KDE or any tty*.

Please, _always_CC_me_ -- as I'm not on the kernel mailing list.

Manuel,

Please attach complete dmesgs (zipped, if necessary) of a
suspend/resume cycle
on a vanilla 3.12.x (where resume fails) _and_ a vanilla 3.11.x
(where resume succeeds).

For the test configurations, please do not apply patches.

Regards,
Peter Hurley

Thank you very much for your reply!
Attached you'll find a zip file with the two edited dmesg logs of
plain vanilla kernel runs.

I have to add, that the resumes _do_ succeed in both cases, only
the serial mouse doesn't get activated after hibernate in 3.12.x
automatically. Just scan for and compare the lines indicating
"serial 00:08: disabled" or "serial 00:08: activated". In 3.12.x
the activation doesn't happen after hibernate, but after
suspend-to-ram (sleep). That only after STR and not before a
setserial gets my mouse back... a miracle. ;-)

I do not see

[ 206.577370] serial 00:08: activated

in the restore from hibernation log of 3.12 which shoudl come from PNP
layer.

[dtor@dtor-d630 work]$ git log --oneline v3.11..v3.12 -- drivers/pnp/
729377d pnp: change pnp bus pm_ops to invoke pnp driver dev_pm_ops if
specified
ce63e18 Merge branch 'pnp'
8ad928d ACPI / PM: Use ACPI_STATE_D3_COLD instead of ACPI_STATE_D3
everywhere
eaf140b PNP: convert PNP driver bus legacy pm_ops to dev_pm_ops

I'd start looking into these commits.


Does the following patch fixes the issue?


PNP: fix restoring devices after hibernation

From: Dmitry Torokhov <dmitry.torokhov@xxxxxxxxx>

On returning from hibernation 'restore; callback is called, not 'resume'.
This fixes breakage introduced by commit
eaf140b60ec961252083ab8adaf67aef29a362dd

Signed-off-by: Dmitry Torokhov <dmitry.torokhov@xxxxxxxxx>
---
drivers/pnp/driver.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/pnp/driver.c b/drivers/pnp/driver.c
index a39ee38..185a24a 100644
--- a/drivers/pnp/driver.c
+++ b/drivers/pnp/driver.c
@@ -235,8 +235,9 @@ static int pnp_bus_resume(struct device *dev)

static const struct dev_pm_ops pnp_bus_dev_pm_ops = {
.suspend = pnp_bus_suspend,
- .freeze = pnp_bus_freeze,
.resume = pnp_bus_resume,
+ .freeze = pnp_bus_freeze,
+ .restore = pnp_bus_resume,
};

struct bus_type pnp_bus_type = {


YES! This patch fixes the issue!!! (Even if compiled with a patched
kernel.) ;-)

Many thanks for your work!

Manuel Krause


I am glad the problem is fixed. But I am puzzled. pnp_bus_resume() didn't handle restore prior to this change. state doesn't get passed in to legacy resume routines. I had to add freeze to handle PMSG_FREEZE case, sounds like restore is needed as well, however I don't see where how restore is handled prior to this change.

-- Shuah

--
Shuah Khan
Senior Linux Kernel Developer - Open Source Group
Samsung Research America(Silicon Valley)
shuah.kh@xxxxxxxxxxx | (970) 672-0658
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/