Bug with (maybe not *in*) sysfs

Bug with (maybe not *in*) sysfs

[Martin J. Bligh]

Customer is getting a panic from x440 w/ sysfs. May well not be
sysfs's fault ... but it's difficult to tell without knowing
what's meant to happen here ;-) debug is turned on.

Looks like kobject_register is just getting an error from kobject_add,
but as we don't see this:

        pr_debug("kobject %s: registering. parent: %s, set: %s\n",
                 kobj->name, parent ? parent->name : "<NULL>",
                 kobj->kset ? kobj->kset->kobj.name : "<NULL>" );

presumably we're hitting:

        if (!(kobj = kobject_get(kobj)))
                return -ENOENT;

But withh no real context on all the kobject stuff it's hard to tell what
that really means ;-)

M.


  Type:   Processor                          ANSI SCSI revision: 02

DEV: registering device: ID = '1:0:9:0', name = ZIBM     GNHv1 S2        0

kobject 1:0:9:0: registering. parent: 01:03.1, set: devices

bus scsi: add device 1:0:9:0

sysfs_create_link: depth = 3, size = 38

sysfs_create_link: path = '../../../devices/pci1/01:03.1/1:0:9:0'

(scsi1:A:12): 160.000MB/s transfers (80.000MHz DT, offset 63, 16bit)

  Vendor: IBM-ESXS  Model: ST336607LC    F   Rev: B258

  Type:   Direct-Access                      ANSI SCSI revision: 03

DEV: registering device: ID = '1:0:12:0', name = SIBM-ESXSST336607LC    F
3JA0JR
4X000073235STN

kobject 1:0:12:0: registering. parent: 01:03.1, set: devices

bus scsi: add device 1:0:12:0

sysfs_create_link: depth = 3, size = 39

sysfs_create_link: path = '../../../devices/pci1/01:03.1/1:0:12:0'

scsi1:A:12:0: Tagged Queuing enabled.  Depth 32

(scsi1:A:13): 160.000MB/s transfers (80.000MHz DT, offset 63, 16bit)

  Vendor: IBM-ESXS  Model: ST336607LC    F   Rev: B258

  Type:   Direct-Access                      ANSI SCSI revision: 03

DEV: registering device: ID = '1:0:13:0', name = SIBM-ESXSST336607LC    F
3JA0J7
EV000073235W9F

kobject 1:0:13:0: registering. parent: 01:03.1, set: devices

bus scsi: add device 1:0:13:0
(scsi1:A:13): 160.000MB/s transfers (80.000MHz DT, offset 63, 16bit)

  Vendor: IBM-ESXS  Model: ST336607LC    F   Rev: B258

  Type:   Direct-Access                      ANSI SCSI revision: 03

DEV: registering device: ID = '1:0:13:0', name = SIBM-ESXSST336607LC    F
3JA0J7
EV000073235W9F

kobject 1:0:13:0: registering. parent: 01:03.1, set: devices

bus scsi: add device 1:0:13:0

sysfs_create_link: depth = 3, size = 39

sysfs_create_link: path = '../../../devices/pci1/01:03.1/1:0:13:0'

scsi1:A:13:0: Tagged Queuing enabled.  Depth 32

bus pci: add driver ips

kobject ips: registering. parent: <NULL>, set: drivers

bus pci: add driver ips

kobject ips: registering. parent: <NULL>, set: drivers

Badness in kobject_register at lib/kobject.c:152

Call Trace:

 [<c019a47b>] kobject_register+0x3b/0x50

 [<c01bf3f5>] bus_add_driver+0x65/0xe0

 [<c019cec6>] pci_register_driver+0x46/0x60

 [<c0209b8a>] ips_detect+0x8a/0xc0

 [<c01e82c6>] __scsi_add_host+0x56/0x70

 [<c01e8679>] scsi_register_host+0x39/0x80

 [<c01050ae>] init+0x4e/0x1c0

 [<c0105060>] init+0x0/0x1c0

Re: Bug with (maybe not *in*) sysfs

[Patrick Mochel]

Thanks for the debug output and backtrace - that's really helpful. :)

> bus pci: add driver ips
> 
> kobject ips: registering. parent: <NULL>, set: drivers
> 
> bus pci: add driver ips
> 
> kobject ips: registering. parent: <NULL>, set: drivers
> 
> Badness in kobject_register at lib/kobject.c:152
> 
> Call Trace:

[...]

This is typically caused by the same object being added twice at the same 
level in the hierarchy, which appears to be happening. Is the ips driver 
calling pci_register_driver() twice? 

	-pat

Re: Bug with (maybe not *in*) sysfs

[Mike Anderson]

Patrick Mochel [mochel@osdl.org] wrote:
> 
> This is typically caused by the same object being added twice at the same 
> level in the hierarchy, which appears to be happening. Is the ips driver 
> calling pci_register_driver() twice? 
> 
> 	-pat

It was possible in the past that pci_module_init could be called more
than once with non-unique pci_driver names. It is fixed in the current
trees, but I do not have the date when it was pushed. Here is some
context mail links.

http://marc.theaimsgroup.com/?l=linux-scsi&m=104275858704733&w=2

http://marc.theaimsgroup.com/?l=linux-scsi&m=104455557710731&w=2

-andmike
--
Michael Anderson
andmike@us.ibm.com

Re: Bug with (maybe not *in*) sysfs

[Martin J. Bligh]

>> This is typically caused by the same object being added twice at the
>> same  level in the hierarchy, which appears to be happening. Is the ips
>> driver  calling pci_register_driver() twice? 
>> 
>> 	-pat
> 
> It was possible in the past that pci_module_init could be called more
> than once with non-unique pci_driver names. It is fixed in the current
> trees, but I do not have the date when it was pushed. Here is some
> context mail links.
> 
> http://marc.theaimsgroup.com/?l=linux-scsi&m=104275858704733&w=2
> 
> http://marc.theaimsgroup.com/?l=linux-scsi&m=104455557710731&w=2

OK, so two similar, but not identical cards using the same driver?
First patch mentioned looks very simple, but won't apply to 2.5.62
There's now just:

   static char ips_hot_plug_name[] = "ips";

   static int __devinit  ips_insert_device(struct pci_dev *pci_dev, const
struct pci_device_id *ent);
   static void __devexit ips_remove_device(struct pci_dev *pci_dev);

   struct pci_driver ips_pci_driver = {
       .name            = ips_hot_plug_name,
       .id_table        = ips_pci_table,
       .probe           = ips_insert_device,
       .remove          = __devexit_p(ips_remove_device),
   };

Second patch looks a little intimidating ... more like a full blown update
of the driver ... I'd rather not go there unless I really have to.

M.

Re: Bug with (maybe not *in*) sysfs

[Mike Anderson]

Martin J. Bligh [mbligh@aracnet.com] wrote:
> OK, so two similar, but not identical cards using the same driver?
> First patch mentioned looks very simple, but won't apply to 2.5.62

Are you seeing this problem on 2.5.62? The registration problem I
described should be gone in 2.5.62.

-andmike
--
Michael Anderson
andmike@us.ibm.com

Re: Bug with (maybe not *in*) sysfs

[Martin J. Bligh]

>> OK, so two similar, but not identical cards using the same driver?
>> First patch mentioned looks very simple, but won't apply to 2.5.62
> 
> Are you seeing this problem on 2.5.62? The registration problem I
> described should be gone in 2.5.62.

Yup. But I can't see it now we avoided a VM bug ... maybe it was indirect
fallout from that ... sorry. If it's still happening, and I just missed it,
will let you know. Thanks for the help,

M.