[rt2x00-users] [RFC/RFT] rt2x00: Fix PCI irq processing race on SMP systems

[Andreas Hartmann]

Helmut Schaa schrieb:
> Am Sonntag, 24. Juli 2011 schrieb Andreas Hartmann:
>> Helmut Schaa schrieb:
>>> Am Donnerstag, 21. Juli 2011 schrieb Andreas Hartmann:
>>>> Helmut Schaa schrieb:
>>>>> Am Donnerstag, 21. Juli 2011 schrieb Andreas Hartmann:
>>>>>> Sorry, but I have to disappoint you - the warning message came up as
>>>>>> before :-(. I tested two times, in between the unpatched compat-wireless
>>>>>> - always the same result (here one example):
>>>>>
>>>>> Ok, thanks for trying anyway. I'm pretty sure I asked before but do you run
>>>>> 40Mhz or 20Mhz?
>>>>
>>>> 40MHz! I want to have power :-)!
>>>
>>> Oh, Andreas could you please check if reducing the queue limit to 8 as in the
>>> rt2x00 USB RFC I sent before makes any difference?
>>
>> Ok, besides the crash mentioned in another mail, I couldn't see any more
>> crash. Don't know, why this happened.
>>
>> Now to the throughput between the queue limit 8 or 64. I played around
>> very long, changed forth and back between 8 and 64 - but to be honest, I
>> couldn't get any preference for the one or the other.
> 
> Sorry, I wasn't clear on this. I was thinking if using a queue len of 8 would
> cause the queue flush problems to disappear?

Well, besides the described impact at the other mail, I couldn't see any
Warnings on ending hostapd - but they didn't come up with 64, too (as
long as X was off).

I couldn't see a warning too in the case, where X was running and the
queue limit was 8 (where I got the dropping frame errors before). If
this could be representative, I can't say, because the warnings don't
come up on every end of hostapd and queue limit 64 w/ X.


BTW: There is no interrupt sharing between X and the WLAN-card:

X runs with IRQ 18, WLAN-card with 21. There is no other device running
with IRQ 21.


I'm thinking of doing some more tests with reduced queue limit and w/ X,
but to be honest, I fear the hard crashes which could destroy my
filesystems with irreparable damage and data loss at the end. I recently
had to reinstall a machine two times because of hard locks and
afterwards totally broken ext4 filesystem. So, if you really need a
resilient statement, I would have to think about to reduce the potential
damage of the filesystems.


Andreas