Driver model problems in -test5: usb this time

Driver model problems in -test5: usb this time

[Pavel Machek]

Hi!

Upon attempt to swsusp, I get "Unable to handle kernel paging request
at virtual address fffffff4" in usb_device_suspend, called from
suspend_device, device_suspend, drivers_suspend, do_software_suspend,
....

								Pavel
-- 
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]

Re: Driver model problems in -test5: usb this time

[Patrick Mochel]

> Upon attempt to swsusp, I get "Unable to handle kernel paging request
> at virtual address fffffff4" in usb_device_suspend, called from
> suspend_device, device_suspend, drivers_suspend, do_software_suspend,
> ....

Unfortunately, we don't have a 'taint' flag for when extra patches are 
added to an official release. 

The latter two functions do not exist in -test5. It would helpful if you 
tried to reproduce with a virgin -test5. It would be courteous to state 
what patches you applied on top of the virgin -test5 kernel. 

The entire Oops would also help, captured either by hand or with e.g. a 
digital camera. You may want to cc the usb-devel list, too.

Thanks,


	Pat

Re: Driver model problems in -test5: usb this time

[Pavel Machek]

Hi!

> > Upon attempt to swsusp, I get "Unable to handle kernel paging request
> > at virtual address fffffff4" in usb_device_suspend, called from
> > suspend_device, device_suspend, drivers_suspend, do_software_suspend,
> > ....
> 
> Unfortunately, we don't have a 'taint' flag for when extra patches are 
> added to an official release. 

:-)

> The latter two functions do not exist in -test5. It would helpful if you 
> tried to reproduce with a virgin -test5. It would be courteous to state 
> what patches you applied on top of the virgin -test5 kernel. 

Lot of them, but only "revert to -test3 swsusp" should be important
here.

> The entire Oops would also help, captured either by hand or with e.g. a 
> digital camera. You may want to cc the usb-devel list, too.

Good news is that it only happens when there are devices plugged into
usb. I'll test other machines now.
								Pavel
-- 
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]

Re: Driver model problems in -test5: usb this time

[Patrick Mochel]

> > The latter two functions do not exist in -test5. It would helpful if you 
> > tried to reproduce with a virgin -test5. It would be courteous to state 
> > what patches you applied on top of the virgin -test5 kernel. 
> 
> Lot of them, but only "revert to -test3 swsusp" should be important
> here.

Then all bets are off. I cannot expect to reproduce the problems until you 
narrow down which patch causes the problem or verify that it appears on a 
standard kernel release.


	Pat

Re: Driver model problems in -test5: usb this time

[Pavel Machek]

Hi!

> > > The latter two functions do not exist in -test5. It would helpful if you 
> > > tried to reproduce with a virgin -test5. It would be courteous to state 
> > > what patches you applied on top of the virgin -test5 kernel. 
> > 
> > Lot of them, but only "revert to -test3 swsusp" should be important
> > here.
> 
> Then all bets are off. I cannot expect to reproduce the problems until you 
> narrow down which patch causes the problem or verify that it appears on a 
> standard kernel release.

Meanwhile I reproduced it with echo -n mem > state; that does not
touch "my" codepaths at all. Oops in usb_device_suspend called from
suspend_device, device_suspend, suspend_prepare, enter_state. If I
unplug usb hub with acm modem and zaurus, it will not oops like that.

								Pavel
-- 
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]

[PATCH] Re: Driver model problems in -test5: usb this time

[Pavel Machek]

Hi!

> > > The latter two functions do not exist in -test5. It would helpful if you 
> > > tried to reproduce with a virgin -test5. It would be courteous to state 
> > > what patches you applied on top of the virgin -test5 kernel. 
> > 
> > Lot of them, but only "revert to -test3 swsusp" should be important
> > here.
> 
> Then all bets are off. I cannot expect to reproduce the problems until you 
> narrow down which patch causes the problem or verify that it appears on a 
> standard kernel release.

Here's patch that should fix it. [First part of first hunk defitely
triggered twice during suspend, and made machine survive that.] Please
apply,

									Pavel

--- clean/drivers/usb/core/usb.c	2003-09-09 12:45:35.000000000 +0200
+++ linux/drivers/usb/core/usb.c	2003-09-10 02:16:10.000000000 +0200
@@ -1429,9 +1429,11 @@
 		return 0;
 
 	intf = to_usb_interface(dev);
+	if (!dev->driver)
+		return 0;
 	driver = to_usb_driver(dev->driver);
 
-	if (driver && driver->suspend)
+	if (driver->suspend)
 		return driver->suspend(intf, state);
 	return 0;
 }
@@ -1446,9 +1448,11 @@
 		return 0;
 
 	intf = to_usb_interface(dev);
+	if (!dev->driver)
+		return 0;
 	driver = to_usb_driver(dev->driver);
 
-	if (driver && driver->resume)
+	if (driver->resume)
 		return driver->resume(intf);
 	return 0;
 }



-- 
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]

Re: Driver model problems in -test5: usb this time

[Pavel Machek]

Hi!

> > > The latter two functions do not exist in -test5. It would helpful if you 
> > > tried to reproduce with a virgin -test5. It would be courteous to state 
> > > what patches you applied on top of the virgin -test5 kernel. 
> > 
> > Lot of them, but only "revert to -test3 swsusp" should be important
> > here.
> 
> Then all bets are off. I cannot expect to reproduce the problems until you 
> narrow down which patch causes the problem or verify that it appears on a 
> standard kernel release.

To follow up myself, patch attached to message sent about 5 minutes
ago indeed fixes the oops. [USB still does not survive suspend/resume,
but at least the kernel survives. Probably pm_send_all() stuff needs
to be cleaned up before USB can be fixed.]
								Pavel
-- 
When do you have a heart between your knees?
[Johanka's followup: and *two* hearts?]

Re: [linux-usb-devel] [PATCH] Re: Driver model problems in -test5: usb this time

[Greg KH]

On Wed, Sep 10, 2003 at 02:19:56AM +0200, Pavel Machek wrote:
> Hi!
> 
> > > > The latter two functions do not exist in -test5. It would helpful if you 
> > > > tried to reproduce with a virgin -test5. It would be courteous to state 
> > > > what patches you applied on top of the virgin -test5 kernel. 
> > > 
> > > Lot of them, but only "revert to -test3 swsusp" should be important
> > > here.
> > 
> > Then all bets are off. I cannot expect to reproduce the problems until you 
> > narrow down which patch causes the problem or verify that it appears on a 
> > standard kernel release.
> 
> Here's patch that should fix it. [First part of first hunk defitely
> triggered twice during suspend, and made machine survive that.] Please
> apply,

Doh, thanks for catching this, I'll add this to my tree and send it in
my next batch to Linus.

This was my fault, nothing that Pat added.

thanks,

greg k-h