Stefano Brivio wrote:
[D'oh, I missed this report until now.]
On Thu, 24 Jan 2008 21:35:13 -0700
Larry Finger <larry.finger@xxxxxxxxxxxx> wrote:
Johannes Berg wrote:
On Thu, 2008-01-24 at 00:16 -0700, Larry Finger wrote:
I have been having "random" kernel panics where the "Caps Lock" LED is flashing at ~1 Hz. These
crashes only occur for the wireless-2.6 tree and have been happening for roughly 3 weeks. After
running a memory test to ensure that these panics were not caused by a hardware problem, I enabled
netconsole logging and caught the following crash report for my x86_64 system:
Code: f6 44 02 08 10 74 12 45 85 ed 78 05 44 39 e9 7f 08 89 8f 24
RIP [<ffffffff88202940>] :mac80211:rate_control_pid_tx_status+0x426/0x45a
Damn, I've seen that too but blamed it on my own patching. Stefano, any
idea? IIRC some sta struct was NULL in pid_tx_status.
The problem is not a NULL in one of the structs, but a runaway loop. The error occurs in the
following loop in rate_control_pid_adjust_rate():
while (newidx != sta->txrate) {
if (rate_supported(sta, mode, newidx) &&
(maxrate < 0 || newidx <= maxrate)) {
sta->txrate = newidx;
break;
}
newidx += back;
}
Is this commit in the tree you are testing?
commit 5bfcaca1279835867e2aa3406cfaf2fd7d92ff7c
Author: Stefano Brivio <stefano.brivio@xxxxxxxxx>
Date: Sun Dec 23 04:41:19 2007 +0100
rc80211-pid: simplify and fix shift_adjust
The panic triggers in rate_supported(), which is compiled in-line, with newidx having a value of 576
at the time of the panic!! I'm not sure of the fix, but I think newindex should always be <=
mode->num_rates. The following patch should cure the crash, but may not be the best fix.
Sure, but rate_control_pid_shift_adjust() ensures that the newindex we
start with is within the ranges, so that I can't actually explain how
run-away can ever happen (because as soon as we hit the lower or the higher
limit, that should be a supported rate!) However, a bug prevented that from
working correctly, but should be fixed by the commit I mentioned above.
Index: wireless-2.6/net/mac80211/rc80211_pid_algo.c
===================================================================
--- wireless-2.6.orig/net/mac80211/rc80211_pid_algo.c
+++ wireless-2.6/net/mac80211/rc80211_pid_algo.c
@@ -123,6 +123,8 @@ static void rate_control_pid_adjust_rate
}
newidx += back;
+ if (newidx < 0 || newidx >= mode->num_rates)
+ return;
}
#ifdef CONFIG_MAC80211_DEBUGFS
This patch has been compile tested at the moment, but it will get further testing after this E-mail
is sent.
ACK, this can be useful as an additional sanity check, even if that
shouldn't be needed. Please ensure you have that commit in your tree -- I'll
investigate further, in case. Thank you for the report, I never hit that!
Yes, the commit you mentioned is in my tree. I'm now running with some additional diagnostics. If my
new check is triggered. I'll let you know what I find.
Larry
-
To unsubscribe from this list: send the line "unsubscribe linux-wireless" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
