Kernel splats in 3.3.7+, mlme.c

Kernel splats in 3.3.7+, mlme.c

[Ben Greear]

We ran a torture test where we try to associate 130 stations against an AP
that can only do 128.  The end result is constant resetting of many of the
virtual station interfaces.

It ran overnight w/out crashing, but spewed quite a few warnings.

I will be happy to test patches or attempt fixes if anyone has suggestions
for what the cause might be....


------------[ cut here ]------------
ADDRCONF(NETDEV_UP): sta128: link is not ready
WARNING: at /home/greearb/git/linux-3.3.dev.y/net/wireless/mlme.c:310 __cfg80211_auth_remove+0x9b/0xaa [cfg80211]()
Hardware name: To be filled by O.E.M.
Modules linked in: aesni_intel cryptd aes_x86_64 aes_generic 8021q garp stp llc fuse macvlan wanlink(PO) pktgen coretemp hwmon sunrpc ipv6 uinput arc4 ath9k 
snd_hda_codec_realtek mac80211 snd_hda_intel snd_hda_codec snd_hwdep snd_seq ath9k_common ath9k_hw snd_seq_device snd_pcm ath snd_timer e1000e cfg80211 snd 
microcode ppdev iTCO_wdt soundcore pcspkr mei(C) serio_raw i2c_i801 snd_page_alloc iTCO_vendor_support parport_pc parport i915 drm_kms_helper drm i2c_algo_bit 
i2c_core video [last unloaded: nf_nat]
Pid: 9598, comm: kworker/u:1 Tainted: P         C O 3.3.6+ #43
Call Trace:
  [<ffffffff81036ef9>] warn_slowpath_common+0x80/0x98
  [<ffffffff81036f26>] warn_slowpath_null+0x15/0x17
  [<ffffffffa01509da>] __cfg80211_auth_remove+0x9b/0xaa [cfg80211]
  [<ffffffffa0151259>] cfg80211_send_auth_timeout+0x8e/0xa5 [cfg80211]
  [<ffffffffa023fb67>] ieee80211_probe_auth_done+0x2f/0xc5 [mac80211]
  [<ffffffff810e0eec>] ? kfree+0x85/0x8a
  [<ffffffffa0242410>] ieee80211_work_work+0xf6f/0xfd9 [mac80211]
  [<ffffffff81448cf6>] ? mutex_unlock+0x9/0xb
  [<ffffffff81057b87>] ? get_parent_ip+0x11/0x41
  [<ffffffff81057b87>] ? get_parent_ip+0x11/0x41
  [<ffffffffa02414a1>] ? free_work+0x14/0x14 [mac80211]
  [<ffffffff810496b5>] process_one_work+0x1a6/0x278
  [<ffffffff8104b708>] worker_thread+0x136/0x255
  [<ffffffff8104b5d2>] ? manage_workers+0x190/0x190
  [<ffffffff8104ed09>] kthread+0x84/0x8c
  [<ffffffff81450924>] kernel_thread_helper+0x4/0x10
  [<ffffffff8104ec85>] ? __init_kthread_worker+0x37/0x37
  [<ffffffff81450920>] ? gs_change+0x13/0x13
---[ end trace d1b4803845952d43 ]---
ADDRCONF(NETDEV_UP): sta302: link is not ready


static void __cfg80211_auth_remove(struct wireless_dev *wdev, const u8 *addr)
{
	int i;
	bool done = true;

	ASSERT_WDEV_LOCK(wdev);

	for (i = 0; addr && i < MAX_AUTH_BSSES; i++) {
		if (wdev->authtry_bsses[i] &&
		    memcmp(wdev->authtry_bsses[i]->pub.bssid,
			   addr, ETH_ALEN) == 0) {
			cfg80211_unhold_bss(wdev->authtry_bsses[i]);
			cfg80211_put_bss(&wdev->authtry_bsses[i]->pub);
			wdev->authtry_bsses[i] = NULL;
			break;
		}
		else if (wdev->authtry_bsses[i])
			done = false;
	}

	WARN_ON(!done);
}


-- 
Ben Greear <greearb@candelatech.com>
Candela Technologies Inc  http://www.candelatech.com

Re: Kernel splats in 3.3.7+, mlme.c

[Johannes Berg]

On Wed, 2012-05-23 at 09:17 -0700, Ben Greear wrote:
> We ran a torture test where we try to associate 130 stations against an AP
> that can only do 128.  The end result is constant resetting of many of the
> virtual station interfaces.
> 
> It ran overnight w/out crashing, but spewed quite a few warnings.
> 
> I will be happy to test patches or attempt fixes if anyone has suggestions
> for what the cause might be....
> 
> 
> ------------[ cut here ]------------
> ADDRCONF(NETDEV_UP): sta128: link is not ready
> WARNING: at /home/greearb/git/linux-3.3.dev.y/net/wireless/mlme.c:310 __cfg80211_auth_remove+0x9b/0xaa [cfg80211]()

This was kinda a known race condition but nobody could ever figure out
what exactly is going on ... the code is gone in 3.4 though :)

johannes