[PATCH] arch_timer: Do not register arch_sys_counter twice

[PATCH] arch_timer: Do not register arch_sys_counter twice

[Thierry Reding]

Commit 65cd4f6 (arch_timer: Move to generic sched_clock framework) added
code to register the arch_sys_counter in arch_timer_register() but it is
already registered in arch_counter_register(). This results in the timer
being added to the clocksource list twice, therefore causing an infinite
loop in the list.

Remove the duplicate registration and register the scheduler clock after
the original registration instead.

This fixes a hang during boot on Tegra114 (Cortex-A15).

Signed-off-by: Thierry Reding <treding-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
---
While I've only tested this on Tegra114, I suspect the same hang during
boot happens for all processors that use this clock source.

 drivers/clocksource/arm_arch_timer.c | 12 +++---------
 1 file changed, 3 insertions(+), 9 deletions(-)

diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c
index f655036..95fb944 100644
--- a/drivers/clocksource/arm_arch_timer.c
+++ b/drivers/clocksource/arm_arch_timer.c
@@ -436,6 +436,9 @@ static void __init arch_counter_register(unsigned type)
 	cyclecounter.mult = clocksource_counter.mult;
 	cyclecounter.shift = clocksource_counter.shift;
 	timecounter_init(&timecounter, &cyclecounter, start_count);
+
+	/* 56 bits minimum, so we assume worst case rollover */
+	sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
 }
 
 static void arch_timer_stop(struct clock_event_device *clk)
@@ -515,15 +518,6 @@ static int __init arch_timer_register(void)
 		goto out;
 	}
 
-	clocksource_register_hz(&clocksource_counter, arch_timer_rate);
-	cyclecounter.mult = clocksource_counter.mult;
-	cyclecounter.shift = clocksource_counter.shift;
-	timecounter_init(&timecounter, &cyclecounter,
-			 arch_counter_get_cntvct());
-
-	/* 56 bits minimum, so we assume worst case rollover */
-	sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
-
 	if (arch_timer_use_virtual) {
 		ppi = arch_timer_ppi[VIRT_PPI];
 		err = request_percpu_irq(ppi, arch_timer_handler_virt,
-- 
1.8.4

Re: [PATCH] arch_timer: Do not register arch_sys_counter twice

[Will Deacon]

On Tue, Oct 15, 2013 at 02:31:51PM +0100, Thierry Reding wrote:
> Commit 65cd4f6 (arch_timer: Move to generic sched_clock framework) added
> code to register the arch_sys_counter in arch_timer_register() but it is
> already registered in arch_counter_register(). This results in the timer
> being added to the clocksource list twice, therefore causing an infinite
> loop in the list.
> 
> Remove the duplicate registration and register the scheduler clock after
> the original registration instead.
> 
> This fixes a hang during boot on Tegra114 (Cortex-A15).
> 
> Signed-off-by: Thierry Reding <treding-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> ---
> While I've only tested this on Tegra114, I suspect the same hang during
> boot happens for all processors that use this clock source.
> 
>  drivers/clocksource/arm_arch_timer.c | 12 +++---------
>  1 file changed, 3 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c
> index f655036..95fb944 100644
> --- a/drivers/clocksource/arm_arch_timer.c
> +++ b/drivers/clocksource/arm_arch_timer.c
> @@ -436,6 +436,9 @@ static void __init arch_counter_register(unsigned type)
>  	cyclecounter.mult = clocksource_counter.mult;
>  	cyclecounter.shift = clocksource_counter.shift;
>  	timecounter_init(&timecounter, &cyclecounter, start_count);
> +
> +	/* 56 bits minimum, so we assume worst case rollover */
> +	sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
>  }
>  
>  static void arch_timer_stop(struct clock_event_device *clk)
> @@ -515,15 +518,6 @@ static int __init arch_timer_register(void)
>  		goto out;
>  	}
>  
> -	clocksource_register_hz(&clocksource_counter, arch_timer_rate);
> -	cyclecounter.mult = clocksource_counter.mult;
> -	cyclecounter.shift = clocksource_counter.shift;
> -	timecounter_init(&timecounter, &cyclecounter,
> -			 arch_counter_get_cntvct());
> -
> -	/* 56 bits minimum, so we assume worst case rollover */
> -	sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
> -
>  	if (arch_timer_use_virtual) {
>  		ppi = arch_timer_ppi[VIRT_PPI];
>  		err = request_percpu_irq(ppi, arch_timer_handler_virt,

Excuse my ignorance, but I'm failing to apply either this patch or the one
that Stephen Boyd proposed:

  http://lists.infradead.org/pipermail/linux-arm-kernel/2013-October/204665.html

The second hunk (deletions) doesn't apply at all, and if I just apply the
first hunk then things won't compile. Which tree is this against?

Will

Re: [PATCH] arch_timer: Do not register arch_sys_counter twice

[Thierry Reding]

[-- Attachment #1: Type: text/plain, Size: 3511 bytes --]

On Tue, Oct 15, 2013 at 03:23:14PM +0100, Will Deacon wrote:
> On Tue, Oct 15, 2013 at 02:31:51PM +0100, Thierry Reding wrote:
> > Commit 65cd4f6 (arch_timer: Move to generic sched_clock framework) added
> > code to register the arch_sys_counter in arch_timer_register() but it is
> > already registered in arch_counter_register(). This results in the timer
> > being added to the clocksource list twice, therefore causing an infinite
> > loop in the list.
> > 
> > Remove the duplicate registration and register the scheduler clock after
> > the original registration instead.
> > 
> > This fixes a hang during boot on Tegra114 (Cortex-A15).
> > 
> > Signed-off-by: Thierry Reding <treding-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
> > ---
> > While I've only tested this on Tegra114, I suspect the same hang during
> > boot happens for all processors that use this clock source.
> > 
> >  drivers/clocksource/arm_arch_timer.c | 12 +++---------
> >  1 file changed, 3 insertions(+), 9 deletions(-)
> > 
> > diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c
> > index f655036..95fb944 100644
> > --- a/drivers/clocksource/arm_arch_timer.c
> > +++ b/drivers/clocksource/arm_arch_timer.c
> > @@ -436,6 +436,9 @@ static void __init arch_counter_register(unsigned type)
> >  	cyclecounter.mult = clocksource_counter.mult;
> >  	cyclecounter.shift = clocksource_counter.shift;
> >  	timecounter_init(&timecounter, &cyclecounter, start_count);
> > +
> > +	/* 56 bits minimum, so we assume worst case rollover */
> > +	sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
> >  }
> >  
> >  static void arch_timer_stop(struct clock_event_device *clk)
> > @@ -515,15 +518,6 @@ static int __init arch_timer_register(void)
> >  		goto out;
> >  	}
> >  
> > -	clocksource_register_hz(&clocksource_counter, arch_timer_rate);
> > -	cyclecounter.mult = clocksource_counter.mult;
> > -	cyclecounter.shift = clocksource_counter.shift;
> > -	timecounter_init(&timecounter, &cyclecounter,
> > -			 arch_counter_get_cntvct());
> > -
> > -	/* 56 bits minimum, so we assume worst case rollover */
> > -	sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
> > -
> >  	if (arch_timer_use_virtual) {
> >  		ppi = arch_timer_ppi[VIRT_PPI];
> >  		err = request_percpu_irq(ppi, arch_timer_handler_virt,
> 
> Excuse my ignorance, but I'm failing to apply either this patch or the one
> that Stephen Boyd proposed:
> 
>   http://lists.infradead.org/pipermail/linux-arm-kernel/2013-October/204665.html

Hehe, that patch is exactly the same as this one. =)

> The second hunk (deletions) doesn't apply at all, and if I just apply the
> first hunk then things won't compile. Which tree is this against?

This is on top of today's linux-next[0] and will probably apply to
yesterday's linux-next too. I think that perhaps the patch or the one
that Stephen proposed himself should be squashed with the original
because that caused the breakage in the first place. From what I gather
from the mailing thread you linked to above it seems like John took an
earlier patch from Stephen and rebased it on top of something more
recent and that didn't work out as expected.

Ingo, any chance you could take this patch and squash it into the patch
mentioned above? Applying it as a separate fix will break bisectability
inbetween both patches.

Thierry

[0]: git://gitorious.org/thierryreding/linux-next.git

[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]

Re: [PATCH] arch_timer: Do not register arch_sys_counter twice

[John Stultz]

On 10/15/2013 08:03 AM, Thierry Reding wrote:
> On Tue, Oct 15, 2013 at 03:23:14PM +0100, Will Deacon wrote:
>> On Tue, Oct 15, 2013 at 02:31:51PM +0100, Thierry Reding wrote:
>>> Commit 65cd4f6 (arch_timer: Move to generic sched_clock framework) added
>>> code to register the arch_sys_counter in arch_timer_register() but it is
>>> already registered in arch_counter_register(). This results in the timer
>>> being added to the clocksource list twice, therefore causing an infinite
>>> loop in the list.
>>>
>>> Remove the duplicate registration and register the scheduler clock after
>>> the original registration instead.
>>>
>>> This fixes a hang during boot on Tegra114 (Cortex-A15).
>>>
>>> Signed-off-by: Thierry Reding <treding-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>
>>> ---
>>> While I've only tested this on Tegra114, I suspect the same hang during
>>> boot happens for all processors that use this clock source.
>>>
>>>  drivers/clocksource/arm_arch_timer.c | 12 +++---------
>>>  1 file changed, 3 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/drivers/clocksource/arm_arch_timer.c
b/drivers/clocksource/arm_arch_timer.c
>>> index f655036..95fb944 100644
>>> --- a/drivers/clocksource/arm_arch_timer.c
>>> +++ b/drivers/clocksource/arm_arch_timer.c
>>> @@ -436,6 +436,9 @@ static void __init
arch_counter_register(unsigned type)
>>>      cyclecounter.mult = clocksource_counter.mult;
>>>      cyclecounter.shift = clocksource_counter.shift;
>>>      timecounter_init(&timecounter, &cyclecounter, start_count);
>>> +
>>> +    /* 56 bits minimum, so we assume worst case rollover */
>>> +    sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
>>>  }
>>> 
>>>  static void arch_timer_stop(struct clock_event_device *clk)
>>> @@ -515,15 +518,6 @@ static int __init arch_timer_register(void)
>>>          goto out;
>>>      }
>>> 
>>> -    clocksource_register_hz(&clocksource_counter, arch_timer_rate);
>>> -    cyclecounter.mult = clocksource_counter.mult;
>>> -    cyclecounter.shift = clocksource_counter.shift;
>>> -    timecounter_init(&timecounter, &cyclecounter,
>>> -             arch_counter_get_cntvct());
>>> -
>>> -    /* 56 bits minimum, so we assume worst case rollover */
>>> -    sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
>>> -
>>>      if (arch_timer_use_virtual) {
>>>          ppi = arch_timer_ppi[VIRT_PPI];
>>>          err = request_percpu_irq(ppi, arch_timer_handler_virt,
>>
>> Excuse my ignorance, but I'm failing to apply either this patch or
the one
>> that Stephen Boyd proposed:
>>
>>  
http://lists.infradead.org/pipermail/linux-arm-kernel/2013-October/204665.html
>
> Hehe, that patch is exactly the same as this one. =)
>
>> The second hunk (deletions) doesn't apply at all, and if I just apply the
>> first hunk then things won't compile. Which tree is this against?

The problem is from my merge to -tip.

>
> This is on top of today's linux-next[0] and will probably apply to
> yesterday's linux-next too. I think that perhaps the patch or the one
> that Stephen proposed himself should be squashed with the original
> because that caused the breakage in the first place. From what I gather
> from the mailing thread you linked to above it seems like John took an
> earlier patch from Stephen and rebased it on top of something more
> recent and that didn't work out as expected.

Yes, again my apologies. When I applied Stephen's original patch to the
recent tree, I mis-resolved the conflict.


> Ingo, any chance you could take this patch and squash it into the patch
> mentioned above? Applying it as a separate fix will break bisectability
> inbetween both patches.

I suspect Ingo won't squish down the fix onto the misresolved patch,
since -tip usually preserves history. But hopefully we can get this
merged quickly.

Again, sorry for the trouble!
-john

Re: [PATCH] arch_timer: Do not register arch_sys_counter twice

[John Stultz]

On 10/15/2013 06:31 AM, Thierry Reding wrote:
> Commit 65cd4f6 (arch_timer: Move to generic sched_clock framework) added
> code to register the arch_sys_counter in arch_timer_register() but it is
> already registered in arch_counter_register(). This results in the timer
> being added to the clocksource list twice, therefore causing an infinite
> loop in the list.
>
> Remove the duplicate registration and register the scheduler clock after
> the original registration instead.
>
> This fixes a hang during boot on Tegra114 (Cortex-A15).
>
> Signed-off-by: Thierry Reding <treding-DDmLM1+adcrQT0dZR+AlfA@public.gmane.org>

Acked-by: John Stultz <john.stultz-QSEj5FYQhm4dnm+yROfE0A@public.gmane.org>

Ingo/Thomas, apologies here, I miss-resolved the conflict updating
Stephens' patch (and also didn't add a note in the commit about it!),
which is causing issues in -next.

Would you mind applying this to tip/timers/core?

thanks
-john


> ---
> While I've only tested this on Tegra114, I suspect the same hang during
> boot happens for all processors that use this clock source.
>
>  drivers/clocksource/arm_arch_timer.c | 12 +++---------
>  1 file changed, 3 insertions(+), 9 deletions(-)
>
> diff --git a/drivers/clocksource/arm_arch_timer.c b/drivers/clocksource/arm_arch_timer.c
> index f655036..95fb944 100644
> --- a/drivers/clocksource/arm_arch_timer.c
> +++ b/drivers/clocksource/arm_arch_timer.c
> @@ -436,6 +436,9 @@ static void __init arch_counter_register(unsigned type)
>  	cyclecounter.mult = clocksource_counter.mult;
>  	cyclecounter.shift = clocksource_counter.shift;
>  	timecounter_init(&timecounter, &cyclecounter, start_count);
> +
> +	/* 56 bits minimum, so we assume worst case rollover */
> +	sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
>  }
>  
>  static void arch_timer_stop(struct clock_event_device *clk)
> @@ -515,15 +518,6 @@ static int __init arch_timer_register(void)
>  		goto out;
>  	}
>  
> -	clocksource_register_hz(&clocksource_counter, arch_timer_rate);
> -	cyclecounter.mult = clocksource_counter.mult;
> -	cyclecounter.shift = clocksource_counter.shift;
> -	timecounter_init(&timecounter, &cyclecounter,
> -			 arch_counter_get_cntvct());
> -
> -	/* 56 bits minimum, so we assume worst case rollover */
> -	sched_clock_register(arch_timer_read_counter, 56, arch_timer_rate);
> -
>  	if (arch_timer_use_virtual) {
>  		ppi = arch_timer_ppi[VIRT_PPI];
>  		err = request_percpu_irq(ppi, arch_timer_handler_virt,