[GIT PULL] irq_work changes for 3.9

[GIT PULL] irq_work changes for 3.9

[Frederic Weisbecker]

Hi Ingo,

The printk changes for full dynticks support are still pending while
we don't know Linus's opinion about these. Meanwhile here is the
part of it that I think is uncontroversial. This way we can make
the next submission attempt to Linus a bit more simple.

These three patches are general fixes for irq work. The two first
patches fix tight races on global irq work claiming that prevent the irq work
subsystem from dropping a work enqueuing attempt because it thinks it's
already pending while it may be already executing or executed.

The last patch just extends irq work support everywhere. It's also a
kconfig cleanup.

The branch can be found at:

git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git
	core/irq_work

Head is: 6147a9d8070e1c9d16d57eb53a14942b95b28dc4

Thanks.

---
Frederic Weisbecker (3):
  irq_work: Fix racy IRQ_WORK_BUSY flag setting
  irq_work: Fix racy check on work pending flag
  irq_work: Remove CONFIG_HAVE_IRQ_WORK

 arch/alpha/Kconfig                  |    1 -
 arch/arm/Kconfig                    |    1 -
 arch/arm64/Kconfig                  |    1 -
 arch/blackfin/Kconfig               |    1 -
 arch/frv/Kconfig                    |    1 -
 arch/hexagon/Kconfig                |    1 -
 arch/mips/Kconfig                   |    1 -
 arch/parisc/Kconfig                 |    1 -
 arch/powerpc/Kconfig                |    1 -
 arch/s390/Kconfig                   |    1 -
 arch/sh/Kconfig                     |    1 -
 arch/sparc/Kconfig                  |    1 -
 arch/x86/Kconfig                    |    1 -
 drivers/staging/iio/trigger/Kconfig |    1 -
 init/Kconfig                        |    4 ----
 kernel/irq_work.c                   |   21 +++++++++++++++------
 16 files changed, 15 insertions(+), 24 deletions(-)

-- 
1.7.5.4

Re: [GIT PULL] irq_work changes for 3.9

[Andreas Mohr]

Hi,

> These three patches are general fixes for irq work. The two first
> patches fix tight races on global irq work claiming that prevent the irq work
> subsystem from dropping a work enqueuing attempt because it thinks it's
> already pending while it may be already executing or executed.

Would that one happen to be one that would fix my system dropping
dead on sound output (looping the existing buffer ad infinitum)
every couple idle minutes when using my sound card's
PulseAudio IRQ-less playback patch (to-be-submitted),
and which seems to have started (or got quite a bit worse
so as to let me notice?) once I added my
"quirks: enable APIC De-Assert Message bit for VIA 8235, too."
(to-be-submitted) commit?
(which reduces IRQs from duplicate to single - thereby probably
reducing the opportunities for the system to notice that work is pending)

Of course a lone kick to the mouse always fixes it...

(-rc2 here)

Hell yeah that does sound like a potential candidate to me.

Andreas Mohr

Re: [GIT PULL] irq_work changes for 3.9

[Frederic Weisbecker]

2013/1/16 Andreas Mohr <andi@lisas.de>:
> Hi,
>
>> These three patches are general fixes for irq work. The two first
>> patches fix tight races on global irq work claiming that prevent the irq work
>> subsystem from dropping a work enqueuing attempt because it thinks it's
>> already pending while it may be already executing or executed.
>
> Would that one happen to be one that would fix my system dropping
> dead on sound output (looping the existing buffer ad infinitum)
> every couple idle minutes when using my sound card's
> PulseAudio IRQ-less playback patch (to-be-submitted),
> and which seems to have started (or got quite a bit worse
> so as to let me notice?) once I added my
> "quirks: enable APIC De-Assert Message bit for VIA 8235, too."
> (to-be-submitted) commit?
> (which reduces IRQs from duplicate to single - thereby probably
> reducing the opportunities for the system to notice that work is pending)
>
> Of course a lone kick to the mouse always fixes it...
>
> (-rc2 here)
>
> Hell yeah that does sound like a potential candidate to me.
>
> Andreas Mohr

I doubt it. I don't see a sound driver using struct irq_work:

$git grep -F "struct irq_work" drivers/ sound/

drivers/acpi/apei/ghes.c:static struct irq_work ghes_proc_irq_work;
drivers/acpi/apei/ghes.c:static void ghes_proc_in_irq(struct irq_work *irq_work)
drivers/extcon/extcon-max77693.c:       struct work_struct irq_work;
drivers/extcon/extcon-max8997.c:        struct work_struct irq_work;
drivers/net/ethernet/microchip/enc28j60.c:      struct work_struct irq_work;
drivers/net/wireless/ti/wl1251/wl1251.h:        struct work_struct irq_work;
drivers/staging/iio/trigger/iio-trig-sysfs.c:   struct irq_work work;
drivers/staging/iio/trigger/iio-trig-sysfs.c:static void
iio_sysfs_trigger_work(struct irq_work *work)
drivers/thermal/exynos_thermal.c:       struct work_struct irq_work;

Re: [GIT PULL] irq_work changes for 3.9

[Andreas Mohr]

Hi,

[trimmed recipient list]

On Wed, Jan 16, 2013 at 11:12:09PM +0100, Frederic Weisbecker wrote:
> 2013/1/16 Andreas Mohr <andi@lisas.de>:
> > Hell yeah that does sound like a potential candidate to me.
> >
> > Andreas Mohr
> 
> I doubt it. I don't see a sound driver using struct irq_work:
> 
> $git grep -F "struct irq_work" drivers/ sound/

Even a more liberal search comes up similar.
I didn't know that irq_work is a somewhat more specific mechanism,
thank you for the clarification!

Leaves me wondering what actually could be responsible for this...
(perhaps even a userland bug i.e. PA buffer replenishing issues)

Oh well...

Andreas Mohr

Re: [GIT PULL] irq_work changes for 3.9

[Ingo Molnar]

* Frederic Weisbecker <fweisbec@gmail.com> wrote:

> Hi Ingo,
> 
> The printk changes for full dynticks support are still pending while
> we don't know Linus's opinion about these. Meanwhile here is the
> part of it that I think is uncontroversial. This way we can make
> the next submission attempt to Linus a bit more simple.
> 
> These three patches are general fixes for irq work. The two first
> patches fix tight races on global irq work claiming that prevent the irq work
> subsystem from dropping a work enqueuing attempt because it thinks it's
> already pending while it may be already executing or executed.
> 
> The last patch just extends irq work support everywhere. It's also a
> kconfig cleanup.
> 
> The branch can be found at:
> 
> git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks.git
> 	core/irq_work
> 
> Head is: 6147a9d8070e1c9d16d57eb53a14942b95b28dc4
> 
> Thanks.
> 
> ---
> Frederic Weisbecker (3):
>   irq_work: Fix racy IRQ_WORK_BUSY flag setting
>   irq_work: Fix racy check on work pending flag
>   irq_work: Remove CONFIG_HAVE_IRQ_WORK
> 
>  arch/alpha/Kconfig                  |    1 -
>  arch/arm/Kconfig                    |    1 -
>  arch/arm64/Kconfig                  |    1 -
>  arch/blackfin/Kconfig               |    1 -
>  arch/frv/Kconfig                    |    1 -
>  arch/hexagon/Kconfig                |    1 -
>  arch/mips/Kconfig                   |    1 -
>  arch/parisc/Kconfig                 |    1 -
>  arch/powerpc/Kconfig                |    1 -
>  arch/s390/Kconfig                   |    1 -
>  arch/sh/Kconfig                     |    1 -
>  arch/sparc/Kconfig                  |    1 -
>  arch/x86/Kconfig                    |    1 -
>  drivers/staging/iio/trigger/Kconfig |    1 -
>  init/Kconfig                        |    4 ----
>  kernel/irq_work.c                   |   21 +++++++++++++++------
>  16 files changed, 15 insertions(+), 24 deletions(-)

Pulled into tip:irq/core, thanks Frederic!

	Ingo