Lorenzo Nava wrote:
> All the registers values look correct: rate tables seems ok, general
> purpose registers ok, rx and tx headers are fine too. The only thing
> that I noticed is that SHM reports a TX header which cookie value is
> 0x200C (at 0x08A0). So this means that the dump of the SHM is relative
> to a frame that is different with respect to the one that causes the
> warning.
> 
> Where are 0x206C-0x200A frames?
> 
> Larry, this problem is presenting only when you stress the device, or it
> always causes crashes the card?

In previous crashes, there has always been a number of queued
tx_status responses. Perhaps the missing cookies were there. This time
I stopped the microcode CPU and never restarted it, but now that it
looks as if the dump is OK, I will start it again right after the dump.

So far, I have only seen these crashes when I stress the device using
two transmit streams - one with small packets (a flood ping) and one
with large packets (the tcpperf test). Even then, it has taken as long
as 7 hours to cause the crash. On other occasions, it takes just
minutes. The bug must have only a narrow window to trigger. In
addition, these tests were the first ones I tried.

FWIW, the proprietary firmware ran my test for 12 hours without failing.

Do you suggest any additional checks in the tx_status routine? Could I
compare the received cookie with some part of SHM?

Larry
_______________________________________________
Bcm43xx-dev mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/bcm43xx-dev

Reply via email to