[2.6.37 -> 2.6.38 regression] "mv64xxx: I2C bus locked, block: 1, time_left: 0" every few seconds

[Roman Mamedov]

On Tue, 31 Jan 2012 02:44:29 +0600
Roman Mamedov <rm at romanrm.ru> wrote:

> On Sat, 7 Jan 2012 16:45:02 -0800
> Guenter Roeck <guenter.roeck at ericsson.com> wrote:
> 
> > >  - Michael Ott reported with Debian kernel 2.6.38-3 (which is closely
> > >    based on stable v2.6.38.2):
> > > 
> > > 	syslog tell me every few seconds:
> > > 	i2c i2c-0: mv64xxx: I2C bus locked, block: 1, time_left: 0
> > > 
> > >  - Manuel Roeder reported the same and found that upstream v2.6.38
> > >    triggers the problem and v2.6.37.6 does not.
> > > 
> > >  - Roman Mamedov finds the bug present in Debian 3.1.6-1 (which
> > >    is closely based on stable v3.1.6) and Debian 2.6.37-2 (based on
> > >    stable v2.6.37.2) but not Debian 2.6.37-1 (based on mainline
> > >    v2.6.37).
> > > 
> > > The regression range described above seems a little inconsistent to
> > > me, so maybe there's something more going on.
> > > 
> > Points to commit eda6bee6c7e67b5bd17bdbced0926f5687f686d5 (i2c-mv64xxx: send repeated START
> > between messages in xfer). Maybe you can back it out and see if it makes a difference.
> 
> Hello,
> 
> I have just confirmed that the problem still occurs in the current
> 3.2.2, and is indeed solved by rolling-back the referred commit.
> 
> Should be noted that beside the "I2C bus locked" messages in dmesg, the described problem
> manifested itself in the inability to read the temperature sensor value or adjust the fan
> speed on the D-Link DNS-323. On all problematic kernels the built-in fan of DNS-323 does not
> rotate and this risks overheating the device and disks installed in it.

Also if anyone decides to further debug this, I should note that once a
'problematic' kernel has booted once on the device, a full power-off seems to
be required, before a 'good' kernel can work properly. Without this, even a
working kernel will manifest the same problem, it looks like something in the
hardware gets locked-up hard and stays that way even across reboots. If
someone's unaware of this, it can completely foil any attempt at git bisecting.

-- 
With respect,
Roman

~~~~~~~~~~~~~~~~~~~~~~~~~~~
"Stallman had a printer,
with code he could not see.
So he began to tinker,
and set the software free."
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20120131/ec3de644/attachment-0001.sig>