[ath9k-devel] ath9k constantly drops network connection

[Mats Johannesson <spamcan1@abrasax.se>]

On Thu, 18 Sep 2008 04:46:18 +0200 Mats Johannesson wrote:
> On Wed, 17 Sep 2008 12:10:03 -0700 Luis R. Rodriguez wrote:
> > On Wed, Sep 17, 2008 at 03:05:59AM -0700, Partha Bagchi wrote:
> [...]
> > > 03:42:27 EDT 2008 i686 i686 i386 GNU/Linux
> > 
> > Please try appling the patch titled:
> > 
> > "[PATCH] ath9k: connectivity is lost after Group rekeying is done"
> > 
> > which was posted on linux-wireless. If that fixes your issue I can
> 
> A question Luis. Looking at the code (and not knowing _anything_ about
> wireless internals):
> 
> -		if (sc->sc_keytype == ATH9K_CIPHER_WEP)
> +		if (tx_info->control.hw_key->alg == ALG_WEP)
>  			txctl->keytype = ATH9K_KEY_TYPE_WEP;
> -		else if (sc->sc_keytype == ATH9K_CIPHER_TKIP)
> +		else if (tx_info->control.hw_key->alg == ALG_TKIP)
>  			txctl->keytype = ATH9K_KEY_TYPE_TKIP;
> -		else if (sc->sc_keytype == ATH9K_CIPHER_AES_CCM)
> +		else if (tx_info->control.hw_key->alg == ALG_CCMP)
>  			txctl->keytype = ATH9K_KEY_TYPE_AES;
> 
> I'm confused regarding the AES cypher. My AP is set through its
> interface as:
> 
> Security Mode: WPA-Personal
> WPA Mode: WPA2 Only
> Cipher Type: AES
> Group Key Update Interval: 2461 seconds (~41 minutes)
> Pre-Shared Key: Loooong long key.
> 
> But the output from "iwlist scan" tells another story:
> 
>                     Encryption key:on
>                     IE: IEEE 802.11i/WPA2 Version 1
>                         Group Cipher : TKIP
>                         Pairwise Ciphers (1) : TKIP
>                         Authentication Suites (1) : PSK
> 
> So someones lying... Does this effect ath9k?
> 
> I'll compile another kernel and try to get rid of TKIP stuff
> completely to test this out. At the moment the environment looks like
> this:
> 
> grep -i tkip my-.config
> grep -i michael my-.config
> # CONFIG_CRYPTO_MICHAEL_MIC is not set
> grep -i ecb my-.config
> CONFIG_CRYPTO_ECB=m
> grep -i aes my-.config
> CONFIG_CRYPTO_AES=m
> CONFIG_CRYPTO_AES_X86_64=m
> 
> loke at sleipner:~$ lsmod|grep -i ecb
> ecb                     1344  2
> crypto_blkcipher        8964  1 ecb
> crypto_algapi           6336  6
> aes_x86_64,aes_generic,arc4,ecb,crypto_blkcipher,cryptomgr
> 
> loke at sleipner:~$ lsmod|grep -i aes
> aes_x86_64              6784  3
> aes_generic            25576  1 aes_x86_64
> crypto_algapi           6336  6
> aes_x86_64,aes_generic,arc4,ecb,crypto_blkcipher,cryptomgr
> 
> Perhaps I'll break ath9k if the kernel lets me compile without an ecb
> module...

Well, it didn't let me deselect ecb since that is selected by MAC80211
(as is arc4).

Anyhow, writing now to share the good news that the patch you refer to:
[PATCH] ath9k: connectivity is lost after Group rekeying is done
(Click on [Download message RAW] at
http://marc.info/?l=linux-wireless&m=122163541519736&w=2 to get a clean
copy) completely fixes the regression between -rc5 and -rc6 I reported
in:

[ath9k-devel] 2.6.27-rc6 (and forwartd to -rc6-git3) dies silently.
https://lists.ath9k.org/pipermail/ath9k-devel/2008-September/000263.html

Patching -rc6-git3 I've now run 4 hours without a connection drop.
Strangely enough it also allowed me to again make the connection
'sticky' over a reboot (Wicd Manager).

This patch is a must have for 2.6.27 no matter what merging policy is
in effect.

Mvh
Mats Johannesson