Re: e1000e: 2.6.27-rc1 corrupts EEPROM/NVM

[Andrew Morton <akpm@linux-foundation.org>]

On Tue, 19 Aug 2008 10:50:15 +0100 David Vrabel <david.vrabel@csr.com> wrote:

> Andrew Morton wrote:
> > On Fri, 08 Aug 2008 11:47:51 +0100 David Vrabel <david.vrabel@csr.com> wrote:
> > 
> >> 2.6.27-rc1 kernels appear to corrupt the EEPROM/NVM of the 8086:294c
> >> ethernet device (ICH9 I think).  After hacking the driver to ignore the
> >> bad CRC and MAC address I could view the EEPROM with ethtool and it was
> >> all 0xff.
> >>
> >> Testing EEPROM reads and writes with ethtool appeared to work but after
> >> a reboot of the system the ethernet device (see lspci -v output) was no
> >> longer present.  Could invalid EEPROM contents with a valid CRC (as
> >> would be the case after my read/write tests) result in the device
> >> failing to enumerate on the PCIe bus?
> 
> FYI, Jesse Brandeburg replied off-list and said that a corrupt EEPROM
> with a valid CRC can result in the device failing to enumerate.

I seem to recall one David S. Miller expressing strong opposition to
the e100[0] policy of escalating a dud EEPROM into a dead device.

> > Is this still happening in Linus's current tree?
> 
> I don't know, and I'm reluctant to try until I have a reliable way to
> restore an erased EEPROM.
> 
> > Was 2.6.26 OK?
> 
> 2.6.26 was fine.
> 

So we don't know why it corrupted the EEPROM?