I suspect we have some ugly (hard trackable) regression between
mainline and wireless-testing. It sounds impossible, but this seems to
affect Broadcom cards and looks like not related to ssb or b43.
Today I've compiled 2.6.38 and 2.6.39-rc6, both work GOOD.
I've compiled wireless-testing and it failed, BAD.
I've reverted all recent ssb patches from top and it still fails BAD.
What would be the easiest way to try bisecting? I can see
wireless-testing merges mainline quite often. Can I use some trick on
checkouted wireless-testing? Do I have to checkout mainline and merge
wireless-testing into it? Something even different?
--
Rafał
2011/5/7 Larry Finger <[email protected]>:
> On 05/06/2011 07:34 PM, Ben Greear wrote:
>>
>> On 05/06/2011 04:51 PM, Rafał Miłecki wrote:
>>>
>>> I suspect we have some ugly (hard trackable) regression between
>>> mainline and wireless-testing. It sounds impossible, but this seems to
>>> affect Broadcom cards and looks like not related to ssb or b43.
>>>
>>> Today I've compiled 2.6.38 and 2.6.39-rc6, both work GOOD.
>>> I've compiled wireless-testing and it failed, BAD.
>>> I've reverted all recent ssb patches from top and it still fails BAD.
>>>
>>> What would be the easiest way to try bisecting? I can see
>>> wireless-testing merges mainline quite often. Can I use some trick on
>>> checkouted wireless-testing? Do I have to checkout mainline and merge
>>> wireless-testing into it? Something even different?
>>>
>>
>> git bisect can handle it...I've spent some quality time finding
>> bugs in ath9k with that lately...
>
> The other thing to do is to see if branching wire-testing to 2.6.38 results
> in a workable system.
I've beed testing clean 2.6.38, so this should be GOOD. Hm, maybe this
is a good idea? Probably a few patches more to test, but should not be
more than 1 or 2 bisect steps.
> What card fails? I've been busy with Realtek hardware lately and have not
> run very much with Broadcom, but my 4311 worked earlier this week.
This is really ugly. We are performing following scenario:
modprobe wl; connect; download sth small; rmmod wl;
modprobe b43; connect; try to download 2GB
With 2.6.38 and 2.6.39 usually whole 2GB gets downloaded, sometimes
card disconnects after 1GB.
With wireless-testing it fails after a minute or earlier. I've tried
reverting every ssb patch that is present in wireless-testing only
(including first patch: "serdes workaround").
Card is 14e4:4315 (BCM4312).
--
Rafał
On 05/06/2011 07:34 PM, Ben Greear wrote:
> On 05/06/2011 04:51 PM, Rafał Miłecki wrote:
>> I suspect we have some ugly (hard trackable) regression between
>> mainline and wireless-testing. It sounds impossible, but this seems to
>> affect Broadcom cards and looks like not related to ssb or b43.
>>
>> Today I've compiled 2.6.38 and 2.6.39-rc6, both work GOOD.
>> I've compiled wireless-testing and it failed, BAD.
>> I've reverted all recent ssb patches from top and it still fails BAD.
>>
>> What would be the easiest way to try bisecting? I can see
>> wireless-testing merges mainline quite often. Can I use some trick on
>> checkouted wireless-testing? Do I have to checkout mainline and merge
>> wireless-testing into it? Something even different?
>>
>
> git bisect can handle it...I've spent some quality time finding
> bugs in ath9k with that lately...
The other thing to do is to see if branching wire-testing to 2.6.38 results in a
workable system.
What card fails? I've been busy with Realtek hardware lately and have not run
very much with Broadcom, but my 4311 worked earlier this week.
Larry
On 05/06/2011 04:51 PM, Rafał Miłecki wrote:
> I suspect we have some ugly (hard trackable) regression between
> mainline and wireless-testing. It sounds impossible, but this seems to
> affect Broadcom cards and looks like not related to ssb or b43.
>
> Today I've compiled 2.6.38 and 2.6.39-rc6, both work GOOD.
> I've compiled wireless-testing and it failed, BAD.
> I've reverted all recent ssb patches from top and it still fails BAD.
>
> What would be the easiest way to try bisecting? I can see
> wireless-testing merges mainline quite often. Can I use some trick on
> checkouted wireless-testing? Do I have to checkout mainline and merge
> wireless-testing into it? Something even different?
>
git bisect can handle it...I've spent some quality time finding
bugs in ath9k with that lately...
Ben
--
Ben Greear <[email protected]>
Candela Technologies Inc http://www.candelatech.com
W dniu 7 maja 2011 02:34 użytkownik Ben Greear
<[email protected]> napisał:
> On 05/06/2011 04:51 PM, Rafał Miłecki wrote:
>>
>> I suspect we have some ugly (hard trackable) regression between
>> mainline and wireless-testing. It sounds impossible, but this seems to
>> affect Broadcom cards and looks like not related to ssb or b43.
>>
>> Today I've compiled 2.6.38 and 2.6.39-rc6, both work GOOD.
>> I've compiled wireless-testing and it failed, BAD.
>> I've reverted all recent ssb patches from top and it still fails BAD.
>>
>> What would be the easiest way to try bisecting? I can see
>> wireless-testing merges mainline quite often. Can I use some trick on
>> checkouted wireless-testing? Do I have to checkout mainline and merge
>> wireless-testing into it? Something even different?
>>
>
> git bisect can handle it...I've spent some quality time finding
> bugs in ath9k with that lately...
Sure it can ;) But what should I match as GOOD for first bisect?
wireless-testing merges mainline all the time. So there are
wireless-testing specific commits between John's merge of 2.6.39-rc5
and 2.6.39-rc6. I can not match Linus's 2.6.39-rc6 as GOOD, because I
won't test wireless-testing specific changes done earlier (these that
didn't go info mainline).
Maybe I should call GOOD a commit that wireless-testing is based on?
Which one would it be?
Or maybe I could rebase wireless-testing on mainline and then I could
use 2.6.39-rc6 as GOOD?
--
Rafał