From mboxrd@z Thu Jan 1 00:00:00 1970 From: Ben Greear Date: Mon, 04 Oct 2010 10:48:36 -0700 Subject: [ath9k-devel] Ath9k in funky state after adding 130 STA interfaces. In-Reply-To: <20101004172827.GM2105@tux> References: <4CA81265.8030403@candelatech.com> <20101004170732.GH2105@tux> <4CAA0C54.5050106@candelatech.com> <20101004172422.24699.qmail@stuge.se> <20101004172827.GM2105@tux> Message-ID: <4CAA1374.1040707@candelatech.com> List-Id: MIME-Version: 1.0 Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: 7bit To: ath9k-devel@lists.ath9k.org On 10/04/2010 10:28 AM, Luis R. Rodriguez wrote: > On Mon, Oct 04, 2010 at 10:24:21AM -0700, Peter Stuge wrote: >> Ben Greear wrote: >>> I was hoping the 'deadbeef' registers indicated a particular error >>> in the NIC that I might could use for further debugging. >> >> I'd guess it's a general error rather than something specific. I get >> it on most boots and the driver needs some jerking around before the >> card will start working. Sometimes it doesn't matter what I do and >> power cycle is all that seems to help. I haven't spent much time on >> thorough research yet though, in part because there's so little >> response here. :\ > > yeah the deadbeef stuff typically is a misprogramming on the > driver to not wake the chip up for a read/write operation. > We have a lot of fixes for this over the last few kernels > and if you using 2.6.35 chances are a few fixes may not > have trickled down yet. For development purposes you are > best to use the latest always. I'm on wireless-testing, totally up-to-date as I can get. It seems trivial to kill the system using lots of STA devices on ath9k, so if anyone needs help reproducing this, I'm more than happy to help. In general, I'd guess that the problems I hit very quickly could also be hit by other users if they got unlucky, so it may be worth some time to debug this in detail. Basically, create 130 (don't think exact number matters) STA devices, configure a wpa_supplicant for each, and stand back and watch things go south. I do have a few extra patches..like to restrict scanning to the single associated channel, but I doubt that has much impact on the bug (and it works fine in ath5k). I have lockdep turned on, and pre-empt enabled in the kernel configs. If someone prefers, I can make my entire tree available through git and post .config files, etc. Thanks, Ben -- Ben Greear Candela Technologies Inc http://www.candelatech.com