← Back to team overview

linux-traipu team mailing list archive

[Bug 855199] Re: "irqfixup" and "irqpoll" broken since 2.6.39

 

** Description changed:

  In all kernels since 2.6.39, the "irqfixup" and "irqpoll" options are no
- longer taking effect.  I have several machines that need to boot with
- one of these options, and I see reports of this regression from multiple
- other users.
+ longer taking effect.
  
  The interrupts now generate the same errors seen without these kernel
  options, e.g.:
  
    irq 19: nobody cared (try booting with the "irqpoll" option)
  
  even though irqpoll, or irqfixup, is used.
  
  The Linux irq code was reworked during the 2.6.39 cycle.  I have gone
- through the patches, and this is the commit where the regression
- arrives:
+ through the patches, isolated two regressions, and submitted patches for
+ each.
  
-   commit d05c65fff0
+ The first bug effectively disabled the bad-irq handling routines, by a
+ test condition be accidentally reversed.  This is the commit where the
+ regression arrives:
+ 
+   commit d05c65fff0 , genirq: spurious: Run only one poller at a time
  
  http://git.kernel.org/?p=linux/kernel/git/torvalds/linux.git;a=commit;h=d05c65fff0ef672be75429266751f0e015b54d94
  
- I have a patch that fixes most of my machones, and the maintainer has
- accepted it:
+ The maintainer has accepted my patch for that, and it will be in Linux
+ 3.2.  I'm keeping an eye on the stable tree.  At this point, I'd ask
+ Ubuntu to apply it.  It's attached here, and is in Linus' tree at:
  
-   https://lkml.org/lkml/2011/11/2/260
+ http://git.kernel.org/?p=linux/kernel/git/torvalds/linux.git;a=commit;h=c75d720fca8a91ce99196d33adea383621027bf2
  
+ That bug disabled irqfixup & irqpoll for everybody.  But with that
+ fixed, I had some machines where those kernel options still failed.  I
+ bisected that problem to here:
  
- Any testing of the current patch would be great.
- 
- I wish I could say that's the end of it, but I have found configurations
- on my machines where a bad irq still goes uncaught, even with my patch.
- 
- Focusing on those cases, it comes down to this patch:
- 
-   commit fa27271bc , genirq: Fixup poll handling
+   commit fa27271bc , genirq: Fixup poll handling
  
  http://git.kernel.org/?p=linux/kernel/git/torvalds/linux.git;a=commitdiff;h=fa27271bc8d230355c1f24ddea103824fdc12de6
  
- This patch changes the approach in multiple ways, and I haven't figured
- out where it drops the ball.  I've toyed around with the changes, but
- have not produced a clear fix.  I'll keep trying, and will be in touch
- with the maintainer again.
+ I've just submitted another patch, for that regression. (So far, no one
+ else seeing this bug report has hit that second regression, so there may
+ be no one here in a position to test it.  I have found a previous post
+ to LKML, from a user who bisected his problem to the same commit,
+ fa2727, so I may try to contact him: https://lkml.org/lkml/2011/7/24/54
+ .)
+ 
+ At any rate, my second submission to Thomas can be watched here:
+ 
+   https://lkml.org/lkml/2011/11/19/189
+ 
+ And I'll report back.

-- 
You received this bug notification because you are a member of UBUNTU -
AL - BR, which is subscribed to Mythbuntu.
https://bugs.launchpad.net/bugs/855199

Title:
  "irqfixup" and "irqpoll" broken since 2.6.39

Status in The Linux Kernel:
  Confirmed
Status in Mythbuntu, Ubuntu derivative focused upon MythTV:
  Confirmed
Status in “linux” package in Ubuntu:
  Triaged

Bug description:
  In all kernels since 2.6.39, the "irqfixup" and "irqpoll" options are
  no longer taking effect.

  The interrupts now generate the same errors seen without these kernel
  options, e.g.:

    irq 19: nobody cared (try booting with the "irqpoll" option)

  even though irqpoll, or irqfixup, is used.

  The Linux irq code was reworked during the 2.6.39 cycle.  I have gone
  through the patches, isolated two regressions, and submitted patches
  for each.

  The first bug effectively disabled the bad-irq handling routines, by a
  test condition be accidentally reversed.  This is the commit where the
  regression arrives:

    commit d05c65fff0 , genirq: spurious: Run only one poller at a time

  http://git.kernel.org/?p=linux/kernel/git/torvalds/linux.git;a=commit;h=d05c65fff0ef672be75429266751f0e015b54d94

  The maintainer has accepted my patch for that, and it will be in Linux
  3.2.  I'm keeping an eye on the stable tree.  At this point, I'd ask
  Ubuntu to apply it.  It's attached here, and is in Linus' tree at:

  http://git.kernel.org/?p=linux/kernel/git/torvalds/linux.git;a=commit;h=c75d720fca8a91ce99196d33adea383621027bf2

  That bug disabled irqfixup & irqpoll for everybody.  But with that
  fixed, I had some machines where those kernel options still failed.  I
  bisected that problem to here:

    commit fa27271bc , genirq: Fixup poll handling

  http://git.kernel.org/?p=linux/kernel/git/torvalds/linux.git;a=commitdiff;h=fa27271bc8d230355c1f24ddea103824fdc12de6

  I've just submitted another patch, for that regression. (So far, no
  one else seeing this bug report has hit that second regression, so
  there may be no one here in a position to test it.  I have found a
  previous post to LKML, from a user who bisected his problem to the
  same commit, fa2727, so I may try to contact him:
  https://lkml.org/lkml/2011/7/24/54 .)

  At any rate, my second submission to Thomas can be watched here:

    https://lkml.org/lkml/2011/11/19/189

  And I'll report back.

To manage notifications about this bug go to:
https://bugs.launchpad.net/linux/+bug/855199/+subscriptions