login
Search
Great, thanks. [ 3606.561930] ath0: switched to short barker preamble (BSSID=00:0f:24:c2:59:70) [ 3629.398993] ath0: RX deauthentication from 00:0f:24:c2:59:70 (reason=2) [ 3629.398996] ath0: deauthenticated Hmm, reason = 2 means AP disconnected you because authorization was no longer valid. That's not ringing bells for me. 5006s are known to still be a little iffy. That's what I have so I'll try replicating your test case at home. In the meantime can you verify you have the patch below in the compat driver? (Just make sure the "rxs.flag = 0;" part is in base.c.): http://permalink.gmane.org/gmane.linux.kernel.wireless.general/14684 -- Bob Copeland %% www.bobcopeland.com _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
The dmesg log I sent you was run on the 05-07-2008 compat-wireless which did not have the below patch. However, I originally ran into the issue on the 05-21-2008 archive which did have the patch. I am going to put today's compat-wireless drop on the machine and see what the behavior is and if the deauthetication occurs with the same symptoms. I have a AR5212 (Cisco PC Card) that I will try to see if it acts differently as well. Thanks, Phil -----Original Message----- From: Bob Copeland [mailto:me@bobcopeland.com] Sent: Wednesday, May 21, 2008 5:24 PM To: Philip A. Culver; ath5k-devel@lists.ath5k.org Subject: RE: [ath5k-devel] Transfers stall with ath5k on compat-wireless Great, thanks. [ 3606.561930] ath0: switched to short barker preamble (BSSID=00:0f:24:c2:59:70) [ 3629.398993] ath0: RX deauthentication from 00:0f:24:c2:59:70 (reason=2) [ 3629.398996] ath0: deauthenticated Hmm, reason = 2 means AP disconnected you because authorization was no longer valid. That's not ringing bells for me. 5006s are known to still be a little iffy. That's what I have so I'll try replicating your test case at home. In the meantime can you verify you have the patch below in the compat driver? (Just make sure the "rxs.flag = 0;" part is in base.c.): http://permalink.gmane.org/gmane.linux.kernel.wireless.general/14684 -- Bob Copeland %% www.bobcopeland.com _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Hello, I tried this on a AR5212 chipset card with slightly better results but still the behavior was not good. The scp copy would push between 8 and 60MB in about 6 tries across the wireless connection before stalling. The behavior here seems a tad different as well. I believe the driver reassociates with the AP but it still kills the connection and file transfer. On the remote side of the file push, scp would respond "Received disconnect from 192.168.2.253: 2: Corrupted MAC on input." The dmesg trace is as follows: [ 2518.780769] ath0: No ProbeResp from current AP 00:0f:24:c2:59:70 - assume out of range [ 2523.436319] ath0: Initial auth_alg=0 [ 2523.436333] ath0: authenticate with AP 00:0f:24:c2:59:70 [ 2523.439651] ath0: RX authentication from 00:0f:24:c2:59:70 (alg=0 transaction=2 status=0) [ 2523.439659] ath0: authenticated [ 2523.439664] ath0: associate with AP 00:0f:24:c2:59:70 [ 2523.442022] ath0: RX ReassocResp from 00:0f:24:c2:59:70 (capab=0x431 status=0 aid=11) [ 2523.442031] ath0: associated [ 2523.442051] ath0: CTS protection enabled (BSSID=00:0f:24:c2:59:70) [ 2523.442057] ath0: switched to short barker preamble (BSSID=00:0f:24:c2:59:70) I also saw one anomaly. The dmesg filled up with the following over and over again until I killed the scp process. [ 2289.719217] KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1201) [ 2289.720613] recvmsg bug: copied D6E83810 seq D6E83DB8 [ 2289.721032] KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1201) [ 2289.722409] recvmsg bug: copied D6E83810 seq D6E83DB8 [ 2289.722729] KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1201) [ 2289.724123] recvmsg bug: copied D6E83810 seq D6E83DB8 [ 2289.724454] KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1201) [ 2289.725823] recvmsg bug: copied D6E83810 seq D6E83DB8 [ 2289.725990] KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1201) [ 2289.727381] recvmsg bug: copied D6E83810 seq D6E83DB8 I ...
On Thu, May 22, 2008 at 5:36 PM, Philip A. Culver Well, I tried replicating this today. I definitely get stalls when pushing a lot of data, and a fair number of dropped packets. However with the latest code I do not get disconnects from the AP -- even while a scp connection is stalled I can browse the net more or less successfully. I didn't see the 'Corrupted MAC' error but I'm not sure I have ssh logging that to syslog. It would be interesting to trace the packets to see whether the source Strange, we have a packet in the receive queue whose sequence number is higher than the current high-water mark? I don't know the network stack well enough to know why. Maybe something is stomping on that skb. -- Bob Copeland %% www.bobcopeland.com _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Can you verify that rate changes when you force a different rate via iwconfig ? (put a printk when using tx_info stuct) I believe it's a problem that came up during the move of tx_control to skb->cb (API change), don't have time to debug it right now unfortunately ;-( Here is the change... http://git.kernel.org/?p=linux/kernel/git/linville/wireless-testing.git;a=commit;h=fba... -- GPG ID: 0xD21DB2DB As you read this post global entropy rises. Have Fun ;-) Nick _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
I'll give that a shot tonight, been trying to track down an input bug D6E83DB8 - D6E83810 = 5A8.. 0x5a is a funny number (POISON_INUSE), but it's at a weird bit location, so could be coincidence. -- Bob Copeland %% www.bobcopeland.com _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Hi, I tried the scp copy again against the 05-29-2008 compat-wireless drivers. I tracked the results for 5 of the attempts in a row. Some background info: AR5212 PCMCIA card Modified Fedora 2.6.23.15-80 kernel (essentially added some SAS drivers from intel) Compat-wireless 05-29-2008 Connected via WPA-PSK, TKIP at about 40-50% signal strength. Try 1 - Saw recvmsg bug in dmesg. Scp was stalled and ctrl+c to kill it. (dmesg_recvb Try 2 - Saw no probe resp in dmesg. Reassociated with access point. Scp was stalled and ctrl+c to kill Try 3 - Saw no probe resp in dmesg. Scp remoted received disconnect 2: Corrupted MAC on input Try 4 - Saw no probe resp in dmesg. Scp reported closed by remote host. Try 5 - Saw recvmsg bug in dmesg. Scp stalled and ctrl+c to kill. For tries 3-5 a ping was running to that machine as well. 831 packets transmitted, 800 received, 3% packet loss. Rtt min/avg/max/mdev = 2.366/11.292/604.221/42.193 ms For try # 1 & 5. The two numbers subtracted again = 5A8. Also on the system with the wireless card a custom protocol over TCP/IP communicated with a server. Whenever the stall occurred the protocol would timeout and then reconnect. In all cases a standard ssh terminal connection to the system stayed connected. Dmesg traces: Try1: 151770] recvmsg bug: copied CF7782FF seq CF7788A7 [ 1560.151780] KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1201) [ 1560.151785] recvmsg bug: copied CF7782FF seq CF7788A7 [ 1560.151796] KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1201) [ 1560.151801] recvmsg bug: copied CF7782FF seq CF7788A7 [ 1560.151812] KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1201) [ 1560.151817] recvmsg bug: copied CF7782FF seq CF7788A7 This continues on for a 1MB log Try 2-4 all had the same dmesg output: [ 1713.643634] ath0: No ProbeResp from current AP 00:0f:24:c2:59:70 - assume out of range [ 1718.565411] ath0: Initial auth_alg=0 [ 1718.565427] ath0: ...
Just to be clear, I should see changes to the rate in say, ath5k_txbuf_setup? When I do 'iwconfig rate X' while connected, the rate changes according to iwconfig, but the printk-ed rate indeed I still haven't seen this bug here. Philip, which kernel are you using? Can you post your .config? -- Bob Copeland %% www.bobcopeland.com _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Hi Bob, I am using a modified Fedora 2.6.23.15-80 kernel. It has added patches for an Intel SAS drive chipset. The .config file is attached. -- Phil -----Original Message----- From: bcopeland@gmail.com [mailto:bcopeland@gmail.com] On Behalf Of Bob Copeland Sent: Sunday, June 01, 2008 8:32 AM To: Nick Kossifidis Cc: ath5k-devel@lists.ath5k.org; mcgrof@gmail.com; pculver@bostonatlantic.net Subject: Re: [ath5k-devel] Transfers stall with ath5k on compat-wireless drivers Just to be clear, I should see changes to the rate in say, ath5k_txbuf_setup? When I do 'iwconfig rate X' while connected, the rate changes according to iwconfig, but the printk-ed rate indeed I still haven't seen this bug here. Philip, which kernel are you using? Can you post your .config? -- Bob Copeland %% www.bobcopeland.com
Latest wireless-testing, particularly fixes this particular rate control issue. Particularly: http://git.kernel.org/?p=linux/kernel/git/linville/wireless-testing.git;a=commit;h=463... Ok I'll try to get this set up to see if there's something weird about 2.6.23 and compat-wireless -- Bob Copeland %% www.bobcopeland.com _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
I 'particularly' shouldn't write emails before 8am :o -- Bob Copeland %% www.bobcopeland.com _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Don't think so. I'm seeing the bug here with vanilla 2.6.25.4 + compat-wireless-2008-06-04 and 2.6.26-rc4. kernel: ath5k phy0: Atheros AR5213A chip found (MAC: 0x59, PHY: 0x43) kernel: ath5k phy0: RF5112A multiband radio found (0x36) pairwise_cipher=CCMP group_cipher=TKIP key_mgmt=WPA-PSK Kernel log (2.6.25.4 + compat-wireless): Jun 3 13:28:13 kernel: wlan0: No ProbeResp from current AP 00:0f:66:c7:85:48 - assume out of range Jun 3 13:28:13 kernel: ADDRCONF(NETDEV_CHANGE): wlan0: link becomes ready Jun 3 13:28:18 kernel: wlan0: Initial auth_alg=0 Jun 3 13:28:18 kernel: wlan0: authenticate with AP 00:0f:66:c7:85:48 Jun 3 13:28:18 kernel: wlan0: RX authentication from 00:0f:66:c7:85:48 (alg=0 transaction=2 status=0) Jun 3 13:28:18 kernel: wlan0: authenticated Jun 3 13:28:18 kernel: wlan0: associate with AP 00:0f:66:c7:85:48 Jun 3 13:28:18 kernel: wlan0: RX ReassocResp from 00:0f:66:c7:85:48 (capab=0x411 status=0 aid=3) Jun 3 13:28:18 kernel: wlan0: associated Jun 3 13:28:18 kernel: ADDRCONF(NETDEV_CHANGE): wlan0: link becomes ready Jun 3 13:28:28 kernel: wlan0: no IPv6 routers present Jun 3 13:28:46 kernel: KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1354) Jun 3 13:28:46 kernel: KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1354) Jun 3 13:28:46 kernel: recvmsg bug: copied E3C09CF2 seq E3C0A29A Jun 3 13:28:46 kernel: KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1354) Jun 3 13:28:46 kernel: recvmsg bug: copied E3C09CF2 seq E3C0A29A Jun 3 13:28:46 kernel: KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1354) Jun 3 13:28:46 kernel: recvmsg bug: copied E3C09CF2 seq E3C0A29A Jun 3 13:28:46 kernel: KERNEL: assertion (flags & MSG_PEEK) failed at net/ipv4/tcp.c (1354) Jun 3 13:28:46 kernel: recvmsg bug: copied E3C09CF2 seq E3C0A29A Kernel log (2.6.26-rc4): Jun 4 16:30:40 kernel: <7><7>wlan0: No ProbeResp from current AP 00:0f:66:c7:85:48 - assume out of range Jun 4 16:30:44 kernel: wlan0: ...
Good to know. On the poison theory, do you also have CONFIG_SLUB_DEBUG=y (or SLAB)? Does wireless-testing work any better? -- Bob Copeland %% www.bobcopeland.com _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Not really. With 2.6.25.4 and wireless-testing I also just got Jun 4 17:45:52 abraxas kernel: wlan0: No ProbeResp from current AP 00:0f:66:c7:85:48 - assume out of range Jun 4 17:45:52 abraxas kernel: ------------[ cut here ]------------ Jun 4 17:45:52 abraxas kernel: WARNING: at kernel/softirq.c:141 local_bh_enable+0x55/0x70() Jun 4 17:45:52 abraxas kernel: Modules linked in: radeon drm nfs lockd sunrpc arc4 ecb ath5k mac80211 cfg80211 binfmt_misc rfcomm l2cap bluetooth ppdev lp sbs container sbshc acpi_cpufreq cpufreq_stats cpufreq_powersave cpufreq_conservative ipv6 af_packet fuse joydev irtty_sir sir_dev pcmcia firmware_class nsc_ircc irda crc_ccitt parport_pc parport rtc snd_intel8x0m snd_pcm_oss snd_mixer_oss psmouse snd_intel8x0 snd_ac97_codec pcspkr ac97_bus snd_pcm snd_timer snd serio_raw i2c_i801 soundcore snd_page_alloc battery ac yenta_socket rsrc_nonstatic pcmcia_core bay video output dock button thinkpad_acpi hwmon intel_agp agpgart nvram evdev shpchp pci_hotplug sha256_generic aes_i586 aes_generic cbc dm_crypt crypto_blkcipher dm_mirror dm_snapshot dm_mod raid10 raid456 async_xor async_memcpy async_tx xor raid1 raid0 multipath linear md_mod floppy ide_cd_mod cdrom e1000 ehci_hcd uhci_hcd usbcore thermal processor fan unix Jun 4 17:45:52 abraxas kernel: Pid: 5698, comm: ath5k_pci Not tainted 2.6.25.4 #1 Jun 4 17:45:52 abraxas kernel: [<c011e044>] warn_on_slowpath+0x54/0x70 Jun 4 17:45:52 abraxas kernel: [<c01179ab>] ? default_wake_function+0xb/0x10 Jun 4 17:45:52 abraxas kernel: [<c012f2b6>] ? autoremove_wake_function+0x16/0x40 Jun 4 17:45:52 abraxas kernel: [<c01163be>] ? __wake_up_common+0x3e/0x70 Jun 4 17:45:52 abraxas kernel: [<c012c1ba>] ? insert_work+0x4a/0x70 Jun 4 17:45:52 abraxas kernel: [<c012c391>] ? queue_work+0x31/0x40 Jun 4 17:45:52 abraxas kernel: [<c012c3af>] ? schedule_work+0xf/0x20 Jun 4 17:45:52 abraxas kernel: [<f9cb3748>] ? add_todo+0x38/0x60 [mac80211] Jun 4 17:45:52 abraxas kernel: [<f9cb38ed>] ? __ieee80211_key_free+0x2d/0x40 ...
Well, I looked through the code and I didn't see anything obvious. My guess was that an skb was being freed while something still held onto it. Then the skb came back out of the slab allocator for a new packet and process A overwrote part of the sequence number. It's telling that the sequence is right in the middle of skb->cb where mac80211 holds the tx info. Still, I couldn't find any problems with the current code. At this point, all I can suggest is adding printks in the disassociate path and at the assert line that is triggering, which would rule that out (e.g. the address of the skb should be the same if the above is at all on track). Also this is probably unrelated but worth a shot: Johannes Berg is working on that one, there's a patch here if it's not already fixed: http://marc.info/?l=linux-wireless&m=121378343122178&w=2 -- Bob Copeland %% www.bobcopeland.com _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Hi, I can't actually compile any compat-wireless snapshots past 05-21-2008. Starting on the 22nd it fails with implicit declaration of function 'skb_get_queue_mapping'. I don't see a definition of this function anywhere in the linux headers or the compat-wireless headers. Any ideas? I will try the latest drivers once I can compile :) Thanks, Phil -----Original Message----- From: bcopeland@gmail.com [mailto:bcopeland@gmail.com] On Behalf Of Bob Copeland Sent: Sunday, May 25, 2008 5:13 PM To: Philip A. Culver Cc: ath5k-devel@lists.ath5k.org Subject: Re: [ath5k-devel] Transfers stall with ath5k on compat-wireless drivers On Thu, May 22, 2008 at 5:36 PM, Philip A. Culver Well, I tried replicating this today. I definitely get stalls when pushing a lot of data, and a fair number of dropped packets. However with the latest code I do not get disconnects from the AP -- even while a scp connection is stalled I can browse the net more or less successfully. I didn't see the 'Corrupted MAC' error but I'm not sure I have ssh logging that to syslog. It would be interesting to trace the packets to see whether the source Strange, we have a packet in the receive queue whose sequence number is higher than the current high-water mark? I don't know the network stack well enough to know why. Maybe something is stomping on that skb. -- Bob Copeland %% www.bobcopeland.com _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
On Thu, May 29, 2008 at 1:37 AM, Philip A. Culver Give me a few minutes. Luis _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
OK try now. Luis _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
I see no difference in the 05-29-2008 snapshot vs the 05-28-2008 one from linuxwireless.org. I will check again tomorrow in case the changes missed the snapshot generation time. Is this a 2.6.23 and earlier issue? My research seems to indicate the missing declaration was added to 2.6.24 in skbuff.h. Thanks, --Phil -----Original Message----- From: Luis R. Rodriguez [mailto:mcgrof@gmail.com] Sent: Thursday, May 29, 2008 8:17 AM To: Philip A. Culver Cc: Bob Copeland; ath5k-devel@lists.ath5k.org Subject: Re: [ath5k-devel] Transfers stall with ath5k on compat-wireless drivers On Thu, May 29, 2008 at 12:51 PM, Luis R. Rodriguez <mcgrof@gmail.com> OK try now. Luis _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
On Thu, May 29, 2008 at 9:45 PM, Philip A. Culver It should be: wireless@xc:~/compat-wireless-2.6$ cat compat-release 3.0-16-ge67fc86 wireless@xc:~/compat-wireless-2.6$ cat git-describe v2.6.26-rc3-8077-g71f33ec Luis _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
Hi, Sorry about that, a brain freeze on my part. I see the changes. I ran the wrong diff. Everything compiles fine. Thanks for fixing the issues! Thanks, Phil -----Original Message----- From: Luis R. Rodriguez [mailto:mcgrof@gmail.com] Sent: Thursday, May 29, 2008 12:29 PM To: Philip A. Culver Cc: ath5k-devel@lists.ath5k.org Subject: Re: [ath5k-devel] Transfers stall with ath5k on compat-wireless drivers On Thu, May 29, 2008 at 9:45 PM, Philip A. Culver It should be: wireless@xc:~/compat-wireless-2.6$ cat compat-release 3.0-16-ge67fc86 wireless@xc:~/compat-wireless-2.6$ cat git-describe v2.6.26-rc3-8077-g71f33ec Luis _______________________________________________ ath5k-devel mailing list ath5k-devel@lists.ath5k.org https://lists.ath5k.org/mailman/listinfo/ath5k-devel
