Transfer zeroes via ssh test (was: Re: PIO mode)

[Michael Büsch]

On Thu, 2010-10-21 at 09:27 -0500, Larry Finger wrote: 
> On 10/21/2010 03:53 AM, Paul Fertser wrote:
> > On Wed, Oct 20, 2010 at 01:31:54PM -0500, Larry Finger wrote:
> >> On 10/20/2010 12:54 PM, Paul Fertser wrote:
> >>> It looks like we're seeing two different issues.
> >>
> >> I agree.
> > 
> > With this test (transfer via ssh to cold-booted b43) i also get some
> > interesting results:
> > 
> > [ 2140.029809] b43-phy0 debug: Using hardware based encryption for keyidx: 0, mac: 00:02:84:7d:3c:02
> > [ 2148.864071] wlan0: no IPv6 routers present
> > [ 2797.922932] CE: hpet increased min_delta_ns to 11250 nsec
> > [ 3744.258158] ACPI: EC: GPE storm detected, transactions will use polling mode
> > [ 4417.907722] b43-phy0 debug: TX-status contains invalid cookie: 0x1349
> > [ 4417.909318] b43-phy0 debug: Out of order TX status report on DMA ring 1. Expected 74, but got 76
> > [ 4417.910899] b43-phy0 debug: Out of order TX status report on DMA ring 1. Expected 74, but got 78
> > [ 4417.912325] b43-phy0 debug: Out of order TX status report on DMA ring 1. Expected 74, but got 80
> > ...
> > [ 4418.146891] b43-phy0 debug: Out of order TX status report on DMA ring 1. Expected 74, but got 70
> > [ 4418.148529] b43-phy0 debug: Out of order TX status report on DMA ring 1. Expected 74, but got 72
> > [ 4565.170928] wlan0: disassociated from 00:02:84:7d:3c:02 (Reason: 1)
> > 
> > Then repeated associations and deassociations until i unload the driver.
> > Reloading brings functionality back.
> > 
> > I need to do more tests to say if i consistently see the same time interval
> > (between starting and failing) in this test.
> 
> When I had failures during stress testing of the firmware from OpenFWWF, Michael
> helped diagnose the problem and added the diagnostics indicating the "out of
> order" TX status. Over the past months, I have seen them on occasion with
> proprietary firmware, but I have never been able to reproduce these failures. As
> you note, the driver cannot recover.
> 
> It is looking more likely that the bug that causes them is in b43, not in the
> firmware, and that the open-source firmware just triggers the bug more
> "efficiently".
> 
> Michael: Any more thoughts on this issue?

Hm, I hardly remember anything from the debugging.

However, one should check whether the firmware does send those
out-of-order reports actively (because it receives some invalid external
condition from the hardware or something like that), or if those reports
are sent by the hardware without firmware help.
The first case (which is the likely one, IMO), would probably make a
firmware-side workaround possible.

I'm not sure how to test this properly. Maybe some array in SHM that
dumps cookies along with a monotonically increasing number on every
active tx-status-send attempt. This way one could possibly check whether
the out-of-order report was sent by the firmware, or due to hardware
burp.

-- 
Greetings Michael.