2015-12-13 19:27:56

by Jeff Merkey

[permalink] [raw]
Subject: Linux v3.11 and v3.12 bugs - history question

Hi Greg,

You might know the answer to this and save me days of running down
bugs in old kernels.

I am completing full regression testing of the mdb debugger on all
linux versions back to 2.6.37 and I am am almost done. I've
encountered some nasty bugs in these two linux release lines. These
bugs are documented all over the web about these versions.

https://forum.blackmagicdesign.com/viewtopic.php?f=12&t=37186
https://bugzilla.redhat.com/show_bug.cgi?id=990955
https://lkml.org/lkml/2013/8/12/31

I would like to put a fix into my patches for them to make them stable
enough to use. The bug is:

BUG: scheduling while atomic: swapper/0/0/0x10000100

across several functions which shows up when I have held the
processsors in the debugger for several minutes but only if I take a
breakpoint AT INTERRUPT. It does not show up otherwise. It shows
up when I exit the debugger on either v3.11 or v3.12.

I've tracked it down to a linux bug, and it does not affect any other
versions.

Do you recall what caused this bug (I have seen comments its in
several drivers) and point me to the commit that fixed it or thread
because as near as I can tell, it affects these versions ONLY. I have
seen the "BUG scheduling while atomic" fire off on other functions on
these versions but only on these two releases.

Any help would be appreciated.


2015-12-13 20:31:04

by Jeff Merkey

[permalink] [raw]
Subject: Re: Linux v3.11 and v3.12 bugs - history question

Hi Greg,

I think I found it. Going back through all the commits on
linux-stable. I love this git thing, it's totally fucking cool. What
a great idea, it's such a time and labor saving tool, it only took me
ten years to start using it.

Jeff


On 12/13/15, Jeff Merkey <[email protected]> wrote:
> Hi Greg,
>
> You might know the answer to this and save me days of running down
> bugs in old kernels.
>
> I am completing full regression testing of the mdb debugger on all
> linux versions back to 2.6.37 and I am am almost done. I've
> encountered some nasty bugs in these two linux release lines. These
> bugs are documented all over the web about these versions.
>
> https://forum.blackmagicdesign.com/viewtopic.php?f=12&t=37186
> https://bugzilla.redhat.com/show_bug.cgi?id=990955
> https://lkml.org/lkml/2013/8/12/31
>
> I would like to put a fix into my patches for them to make them stable
> enough to use. The bug is:
>
> BUG: scheduling while atomic: swapper/0/0/0x10000100
>
> across several functions which shows up when I have held the
> processsors in the debugger for several minutes but only if I take a
> breakpoint AT INTERRUPT. It does not show up otherwise. It shows
> up when I exit the debugger on either v3.11 or v3.12.
>
> I've tracked it down to a linux bug, and it does not affect any other
> versions.
>
> Do you recall what caused this bug (I have seen comments its in
> several drivers) and point me to the commit that fixed it or thread
> because as near as I can tell, it affects these versions ONLY. I have
> seen the "BUG scheduling while atomic" fire off on other functions on
> these versions but only on these two releases.
>
> Any help would be appreciated.
>

2015-12-13 20:58:23

by Jeff Merkey

[permalink] [raw]
Subject: Re: Linux v3.11 and v3.12 bugs - history question

I found it. Added to the patches -- fixed. Different incarnation of
the bug in 3.12 vs. 3.11 -- same bug though ...



On 12/13/15, Jeff Merkey <[email protected]> wrote:
> Hi Greg,
>
> I think I found it. Going back through all the commits on
> linux-stable. I love this git thing, it's totally fucking cool. What
> a great idea, it's such a time and labor saving tool, it only took me
> ten years to start using it.
>
> Jeff
>
>
> On 12/13/15, Jeff Merkey <[email protected]> wrote:
>> Hi Greg,
>>
>> You might know the answer to this and save me days of running down
>> bugs in old kernels.
>>
>> I am completing full regression testing of the mdb debugger on all
>> linux versions back to 2.6.37 and I am am almost done. I've
>> encountered some nasty bugs in these two linux release lines. These
>> bugs are documented all over the web about these versions.
>>
>> https://forum.blackmagicdesign.com/viewtopic.php?f=12&t=37186
>> https://bugzilla.redhat.com/show_bug.cgi?id=990955
>> https://lkml.org/lkml/2013/8/12/31
>>
>> I would like to put a fix into my patches for them to make them stable
>> enough to use. The bug is:
>>
>> BUG: scheduling while atomic: swapper/0/0/0x10000100
>>
>> across several functions which shows up when I have held the
>> processsors in the debugger for several minutes but only if I take a
>> breakpoint AT INTERRUPT. It does not show up otherwise. It shows
>> up when I exit the debugger on either v3.11 or v3.12.
>>
>> I've tracked it down to a linux bug, and it does not affect any other
>> versions.
>>
>> Do you recall what caused this bug (I have seen comments its in
>> several drivers) and point me to the commit that fixed it or thread
>> because as near as I can tell, it affects these versions ONLY. I have
>> seen the "BUG scheduling while atomic" fire off on other functions on
>> these versions but only on these two releases.
>>
>> Any help would be appreciated.
>>
>