On Sat, Mar 19, 2011 at 7:21 AM, Richard Schütz <r.schtz@t-online.de> wrote:
On Sat, Mar 19, 2011 at 12:57 AM, Jeff Cook<jeff@deserettechnology.com> wrote:
Having issues here with ath9k, much slower than it was with 2.6.37. Found this bug re: Ubuntu on Launchpad, haven't checked the kernel tracker: https://bugs.launchpad.net/ubuntu/+source/linux/+bug/735171
Looks like I'm also getting bit by this on another machine that uses ath5k. Watching wireshark on two machines, almost all packets sent to the ath5k card never make it with 2.6.38. I also tested with a compat-wireless tarball from 2011-03-18 with the same results. Reverting to 2.6.37 makes the issue go away. Definitely seems like an unsafe upgrade at least for Atheros users.
I can confirm that. When running 2.6.38 my downstream with ath9k is about 13 times slower compared to 2.6.37.4.
-- Regards, Richard Schütz
What kind of network are you using? A person in IRC suggested that these issues might only exist on certain (relatively rare) networks, like 802.11n or ad-hoc. I bisected and wound up with 8aec7af99b1e45 as the culprit, though that doesn't make much sense as that change was merged into 2.6.37. Here is my git bisect log: git bisect start '--' 'drivers/net/wireless/' # bad: [7d2c16befae67b901e6750b845661c1fdffd19f1] ath9k: fix aggregation related interoperability issues git bisect bad 7d2c16befae67b901e6750b845661c1fdffd19f1 # good: [3c0eee3fe6a3a1c745379547c7e7c904aa64f6d5] Linux 2.6.37 git bisect good 3c0eee3fe6a3a1c745379547c7e7c904aa64f6d5 # bad: [bfe3850b0cfca6ba64395e2705d9a51cd044f374] rndis_wlan: scanning, workaround device returning incorrect bssid-list item count. git bisect bad bfe3850b0cfca6ba64395e2705d9a51cd044f374 # good: [f7ec8fb4d6f8f3ecb8b11e9e46ece95aa66139cc] ath9k_hw: Fix eeprom offset for AR9287 devices (PCI/USB) git bisect good f7ec8fb4d6f8f3ecb8b11e9e46ece95aa66139cc # bad: [692d2c0fb36c02ad07d54641c26f48e644b27fbd] b43: rename config option for N-PHY, drop BROKEN git bisect bad 692d2c0fb36c02ad07d54641c26f48e644b27fbd # bad: [8efa5d7d6ad307ae2d220def37ca89594062c40d] ath5k: Check if pci pdev struct is initialized in common functions. git bisect bad 8efa5d7d6ad307ae2d220def37ca89594062c40d # good: [61cde037234c4b8e6497a23f5f236c64cbf9d41d] ath5k: Extend rate_duration git bisect good 61cde037234c4b8e6497a23f5f236c64cbf9d41d # bad: [61790c5f3c5f158821821a00797d94504531839f] iwlagn: fix microcode error on 4965 git bisect bad 61790c5f3c5f158821821a00797d94504531839f # bad: [8aec7af99b1e4594c4bb9e1c48005e6111f97e8e] ath5k: Support synth-only channel change for AR2413/AR5413 git bisect bad 8aec7af99b1e4594c4bb9e1c48005e6111f97e8e # good: [b2b4c69f682a2868411899a77842061dd745884f] ath5k: Tweak power detector delays on RF5111/RF5112 git bisect good b2b4c69f682a2868411899a77842061dd745884f # good: [f08fbf6cf4a31c8df52b21440c7a7e6fbe474b28] ath5k: Update PLL programming for turbo/half/quarter git bisect good f08fbf6cf4a31c8df52b21440c7a7e6fbe474b28 # good: [4c57581d939fd0f8f244b9730812069f4dac308a] ath5k: Skip powertable setting when we are on the same channel git bisect good 4c57581d939fd0f8f244b9730812069f4dac308a I may have done something wrong, I guess. I was testing with compat-wireless and not applying patches uniformly really, so they may have mucked things up, but I was getting some success with commits from Nov. 23, 2010 as well as some failures. I will try bisecting again soon and will probably just end up rebooting and testing each kernel individually that way, which is a seriously lame pain in the rear way to test just one driver.