#74: module unstable on x86_64
------------------------------+---------------------------------------------
Reporter: kelmo | Owner:
Type: defect | Status: new
Priority: critical | Milestone: version 1.0.0 - first stable
release
Component: madwifi: driver | Version: trunk
Resolution: | Keywords:
------------------------------+---------------------------------------------
Comment (by tim):
Ok, I've done some more debugging today, and have learned a little bit.
The line dying for me is indeed the one that calls the macro
STAILQ_REMOVE_HEAD(). However, I've found that the line right before that
always sets lastbuf to NULL. In addition, even if one got past the
STAILQ_REMOVE_HEAD() macro, we'd puke on the dereferences of lastbuf.
Taking a step back, what function are we in? We're in
ath_tx_uapsdqueue(). What's this for? It looks like were enqueue-ing a
frame onto the UAPSD queue. Don't ask me to explain what this is. I am
clueless when it comes to wireless hardware.
Looking at the different cases in this file, I see that we're dying on the
condition when both the current queue, and the overflow queue are full.
So, this case is supposed to drop an old frame, and enqueue the new frame.
However, when we try to pull off the old frame, we get back NULL. How
could this be? I thought we had full queues!! I can think of only two
explanations right now:
1. The test
"an->an_uapsd_overflowqdepth == an->an_node.ni_uapsd_maxsp"
is evaluating to true because someone didn't decrement the overflowqdepth
when they were supposed to, or there's an x86_64 integer conversion issue
with this counter.
2. The comment at the top of the function reads:
"* Needs external locking!"
If this function is getting pre-empted right after the overflow counter
check, then another function could be stealing all of the frames on the
queue. When execution comes back to us, we fail to grab a frame because
they're all gone.
I am leaning toward the latter right now... Please someone who knows
something about this code correct me if I am wrong.
More info in a moment...
--
Ticket URL: <http://madwifi.org/ticket/74>
MadWifi <http://madwifi.org/>
Multiband Atheros Driver for Wireless Fidelity
