Re: p54usb broken by commit b19fa1f

[Larry Finger <Larry.Finger@lwfinger.net>]

Chr wrote:
> On Saturday 02 August 2008 05:06:37 Larry Finger wrote:
>> In testing 2.6.27-rc1, I found that p54usb caused my machine to freeze. A
>> few times, I was able to use the KDE system status applet to see that the
>> system was using essentially 100% of the CPU's. One other time, the system
>> responded a little - enough for me to see tha wpa_supplicant was consuming
>> 100% of the my AMD Turion 64 X2 running an x86_64 kernel. From bisection,
>> the problem is due to
>>
>> commit b19fa1fa91845234961c64dbd564671aa7c0fd27
>> Author: David S. Miller <davem@davemloft.net>
>> Date:   Tue Jul 8 23:14:24 2008 -0700
>>
>>      net: Delete NETDEVICES_MULTIQUEUE kconfig option.
>>
>>      Multiple TX queue support is a core networking feature.
>>
>>      Signed-off-by: David S. Miller <davem@davemloft.net>
>>
>> Obviously, something in the driver cannot handle multiqueues, but I
>> have not found anything, and I have not been able to get any logged events
>> that show what is going wrong.
>>
> 
> Hmm that's a bit odd as my p54usb-adapter seems to work usual with
> 2.6.27-rc1?!... That said: it doesn't work nearly as stable as the
> minipci card that I have too. Is there anyone else who can confirm
> that this problem is specific to the p54 driver & 2.6.27-rc1?

The problem is triggered by the following section:

         if (priv->fw_var >= 0x300) {
                 /* Firmware supports QoS, use it! */
                 priv->tx_stats[0].limit = 3;
                 priv->tx_stats[1].limit = 4;
                 priv->tx_stats[2].limit = 3;
                 priv->tx_stats[3].limit = 1;
                 dev->queues = 4;
         }

If fw_var is less than 0x300, then the bug does not trigger. On my system, if I 
change the number of queues to 1, then all is OK.

Larry