Re: BUG while suspending to swap on LVM

[Laurent Riffard <laurent.riffard@free.fr>]

Le 14.08.2006 22:50, Rafael J. Wysocki a écrit :
> On Monday 14 August 2006 22:03, Laurent Riffard wrote:
>> Le 14.08.2006 21:16, Rafael J. Wysocki a écrit :
>>> On Monday 14 August 2006 19:48, Laurent Riffard wrote:
>>>> Le 14.08.2006 19:30, Rafael J. Wysocki a écrit :
>>>>> Hi,
>>>>>
>>>>> Thanks for the report.
>>>>>
>>>>> On Monday 14 August 2006 18:49, Laurent Riffard wrote:
>>>>>> Hello, 
>>>>>>
>>>>>> I hit a BUG while I was trying to suspend to a swap device on LVM. 
>>>>>> The call trace looks like this (see attached picture for detailed 
>>>>>> stack trace):
>>>>>>
>>>>>> EIP is at _raw_spin_lock
>>>>>> Call Trace:
>>>>>>   die
>>>>>>   do_page_fault
>>>>>>   error_code
>>>>>>   _spin_lock_irq
>>>>>>   generic_unplug_device
>>>>>>   wait_on_bio_chain
>>>>>>   swap_write_image
>>>>> This should be swap_write_page().
>>>> right.
>>>>>>   swsusp_write
>>>>>>   pm_suspend_disk
>>>>>>   enter_state
>>>>>>   state_store
>>>>>>   susbsys_attr_store
>>>>>>   sysfs_write_file
>>>>>>   vfs_write
>>>>>>   sys_write
>>>>>>   sysenter_past_esp
>>>>>>
>>>>>> It was Linux 2.6.18-rc4-mm1 booted with:
>>>>>> root=/dev/vglinux1/lvroot video=vesafb:mtrr splash=silent resume=/dev/mapper/vglinux1-lvswap init 1
>>>>> Does the unpatched 2.6.18-rc4 work in this configuration?
>>>> Will try.
>>> Well, I'm afraid it won't work.
>> guess what? 2.6.18-rc4 worked!
> 
> Strange ...
> 
>>> Can you please try the appended patch?
>> Ok, I'll try on 2.6.18-rc4-mm1.
>>  
>>> Rafael
>>>
>>>
>>> ---
>>>  mm/swapfile.c |    3 ++-
>>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>>
>>> Index: linux-2.6.18-rc4-mm1/mm/swapfile.c
>>> ===================================================================
>>> --- linux-2.6.18-rc4-mm1.orig/mm/swapfile.c	2006-08-13 14:54:43.000000000 +0200
>>> +++ linux-2.6.18-rc4-mm1/mm/swapfile.c	2006-08-14 21:09:09.000000000 +0200
>>> @@ -442,11 +442,12 @@ int swap_type_of(dev_t device)
>>>  
>>>  		if (!(swap_info[i].flags & SWP_WRITEOK))
>>>  			continue;
>>> +
>>>  		if (!device) {
>>>  			spin_unlock(&swap_lock);
>>>  			return i;
>>>  		}
>>> -		inode = swap_info->swap_file->f_dentry->d_inode;
>>> +		inode = swap_info[i].swap_file->f_dentry->d_inode;
>>>  		if (S_ISBLK(inode->i_mode) &&
>>>  		    device == MKDEV(imajor(inode), iminor(inode))) {
>>>  			spin_unlock(&swap_lock);
>> Does it mean it was mistaken on swap device ?
> 
> Yes.  If your resume device is not the first swap, it won't work without
> the patch.

Ok, that's why I had that strange behaviour:
- 2 swap devices online, suspend to second one: suspend works.
- on resume, it can't find suspend signature on the 2nd swap device, so
  it continue with normal boot, fsck all FS and say something like "swapon: 
  1st swap device busy". I had to mkswap it.

I was thinking that my initrd script was broken or I was mistyping the resume 
device name in GRUB. Was funny...
 
>> I did not pay attention on which swap device was online or not.
> 
> Please do when you test it. ;-)

I tested with only 1 swap device online: /dev/mapper/vglinux1-lvswap: 
same BUG happened with 2.6.18-rc4-mm1 patched. 

-- 
laurent