diff mbox

[1/2] RTC: Add core rtc support for Gemini Soc devices

Message ID f22e65faa7d86074cda27973f83d7a52@advem.lv
State New
Headers show

Commit Message

Roman Yeryomin July 8, 2015, 10:29 p.m. UTC 
On 2015-05-08 19:08, Arnd Bergmann wrote:
> On Thursday 07 May 2015 12:03:25 Roman Yeryomin wrote:
>> On 2015-05-06 22:22, Arnd Bergmann wrote:
>> > On Wednesday 06 May 2015 19:21:21 Roman Yeryomin wrote:
>> >>
>> >> > Thanks to point that, I'll merge that one which seems clean enough:
>> >> > http://git.openwrt.org/?p=openwrt.git;a=blob;f=target/linux/gemini/files/drivers/rtc/rtc-gemini.c;h=587d8122b2fbb1230437eadcce4789a53aa60ee5;hb=4c637410a2a1ab45988e8ca6202554a502102039
>> >>
>> >> For 3.18 (and up) to work on gemini 160-gemini-timers.patch is vital.
>> >> Without that timers are broken and cpu is super slow.
>> >
>> > Do you know what part of the patch is the actual bug fix? We should
>> > probably
>> > merge that separately and mark it for stable backports, while the bulk
>> > of that
>> > patch seems to just rearrange code.
>> >
>> >       Arnd
>> 
>> Sorry, didn't try to extract the exact lines it but I guess it's that
>> part which touches the scheduler clock code.
>> Because without that patch sched clock runs at 100Hz instead of 25MHz.
>> I can dive into this deeper if you want, I just didn't see the need.
> 
> I think that would be helpful, yes.
> 

OK, here are the minimal changes required (see attachment). Tested on 
4.1.1
Let me know if you want me to submit it in some other way.

Regards,
Roman

Comments

Arnd Bergmann July 9, 2015, 8:04 a.m. UTC | #1 
On Thursday 09 July 2015 01:29:39 Roman Yeryomin wrote:
> 
> OK, here are the minimal changes required (see attachment). Tested on 
> 4.1.1
> Let me know if you want me to submit it in some other way.
> 

The changes look ok to me. Please submit them as two separate
patches (1. set up and use timer3, 2. register sched_clock) according
to the Documentation/SubmittingPatches description.

Thanks,

	Arnd
Hans Ulli Kroll July 9, 2015, 11:58 a.m. UTC | #2 
Hi

On Thu, 9 Jul 2015, Arnd Bergmann wrote:

> On Thursday 09 July 2015 01:29:39 Roman Yeryomin wrote:
> > 
> > OK, here are the minimal changes required (see attachment). Tested on 
> > 4.1.1
> > Let me know if you want me to submit it in some other way.
> > 
> 
> The changes look ok to me. Please submit them as two separate
> patches (1. set up and use timer3, 2. register sched_clock) according
> to the Documentation/SubmittingPatches description.
> 
> Thanks,
> 
> 	Arnd
> 

have you seen this thread ...
[PATCH 00/18] ARM: Migrate clockevent drivers to 'set-state
http://permalink.gmane.org/gmane.linux.ports.arm.kernel/423992

arch/arm/mach-gemini/time.c              | 69 ++++++++++++------------ 

Two or more weeks ago I send a patchset fixing this problem.
At this time Arnd was in holiday and until today I was very busy at work.

I can rework my patches on top of Viresh's patchset

and Roman please fix the subject.

Ulli
Linus Walleij July 9, 2015, 3:48 p.m. UTC | #3 
On Thu, Jul 9, 2015 at 3:02 PM, Roman Yeryomin <roman@advem.lv> wrote:
> On 2015-07-09 14:58, Hans Ulli Kroll wrote:
>>
>> Hi
>>
>> On Thu, 9 Jul 2015, Arnd Bergmann wrote:
>>
>>> On Thursday 09 July 2015 01:29:39 Roman Yeryomin wrote:
>>> >
>>> > OK, here are the minimal changes required (see attachment). Tested on
>>> > 4.1.1
>>> > Let me know if you want me to submit it in some other way.
>>> >
>>>
>>> The changes look ok to me. Please submit them as two separate
>>> patches (1. set up and use timer3, 2. register sched_clock) according
>>> to the Documentation/SubmittingPatches description.
>>>
>>> Thanks,
>>>
>>>         Arnd
>>>
>>
>> have you seen this thread ...
>> [PATCH 00/18] ARM: Migrate clockevent drivers to 'set-state
>> http://permalink.gmane.org/gmane.linux.ports.arm.kernel/423992
>>
>> arch/arm/mach-gemini/time.c              | 69 ++++++++++++------------
>>
>> Two or more weeks ago I send a patchset fixing this problem.
>> At this time Arnd was in holiday and until today I was very busy at work.
>
>
> Sorry, I didn't see your patches.
> Arnd requested minimal changes and until yesterday I was very busy too.
>
>> I can rework my patches on top of Viresh's patchset
>
>
> It looks that this patchset doesn't conflict with the fix needed.

Can any of you guys just look at this thing from OpenWRT:
http://git.openwrt.org/?p=openwrt.git;a=blob;f=target/linux/gemini/patches-3.18/160-gemini-timers.patch;h=4b0edb45d162f751bfb7475f5b3158f28d69fe84;hb=HEAD

As usual they are just keeping it out of mainline, and this
is adding sched_clock() and other goodies so should
probably be upstreamed ASAP.

Yours,
Linus Walleij
Roman Yeryomin July 9, 2015, 4:48 p.m. UTC | #4 
On 2015-07-09 18:48, Linus Walleij wrote:
> On Thu, Jul 9, 2015 at 3:02 PM, Roman Yeryomin <roman@advem.lv> wrote:
>> On 2015-07-09 14:58, Hans Ulli Kroll wrote:
>>> 
>>> Hi
>>> 
>>> On Thu, 9 Jul 2015, Arnd Bergmann wrote:
>>> 
>>>> On Thursday 09 July 2015 01:29:39 Roman Yeryomin wrote:
>>>> >
>>>> > OK, here are the minimal changes required (see attachment). Tested on
>>>> > 4.1.1
>>>> > Let me know if you want me to submit it in some other way.
>>>> >
>>>> 
>>>> The changes look ok to me. Please submit them as two separate
>>>> patches (1. set up and use timer3, 2. register sched_clock) 
>>>> according
>>>> to the Documentation/SubmittingPatches description.
>>>> 
>>>> Thanks,
>>>> 
>>>>         Arnd
>>>> 
>>> 
>>> have you seen this thread ...
>>> [PATCH 00/18] ARM: Migrate clockevent drivers to 'set-state
>>> http://permalink.gmane.org/gmane.linux.ports.arm.kernel/423992
>>> 
>>> arch/arm/mach-gemini/time.c              | 69 
>>> ++++++++++++------------
>>> 
>>> Two or more weeks ago I send a patchset fixing this problem.
>>> At this time Arnd was in holiday and until today I was very busy at 
>>> work.
>> 
>> 
>> Sorry, I didn't see your patches.
>> Arnd requested minimal changes and until yesterday I was very busy 
>> too.
>> 
>>> I can rework my patches on top of Viresh's patchset
>> 
>> 
>> It looks that this patchset doesn't conflict with the fix needed.
> 
> Can any of you guys just look at this thing from OpenWRT:
> http://git.openwrt.org/?p=openwrt.git;a=blob;f=target/linux/gemini/patches-3.18/160-gemini-timers.patch;h=4b0edb45d162f751bfb7475f5b3158f28d69fe84;hb=HEAD
> 
> As usual they are just keeping it out of mainline, and this
> is adding sched_clock() and other goodies so should
> probably be upstreamed ASAP.

This is what I initially submitted here.
But Arnd wanted to see a smaller patch which only fixes gemini timer 
issue (t.i. shed clock being 100Hz instead of 25MHz).
Also I've submitted 4.1 support yesterday, but patches didn't change 
much comparing to 3.18
I would love to see all those patches/drivers in mainline - tell me 
where do I submit all that :)

Regards,
Roman
Roman Yeryomin July 9, 2015, 4:51 p.m. UTC | #5 
On 2015-07-09 11:04, Arnd Bergmann wrote:
> On Thursday 09 July 2015 01:29:39 Roman Yeryomin wrote:
>> 
>> OK, here are the minimal changes required (see attachment). Tested on
>> 4.1.1
>> Let me know if you want me to submit it in some other way.
>> 
> 
> The changes look ok to me. Please submit them as two separate
> patches (1. set up and use timer3, 2. register sched_clock) according
> to the Documentation/SubmittingPatches description.
> 

So what do we do?
Do you still want me to split this as you said and submit or you want 
Hans's patches rebased/resubmitted?

Regards,
Roman
Linus Walleij July 10, 2015, 8:48 a.m. UTC | #6 
On Thu, Jul 9, 2015 at 6:48 PM, Roman Yeryomin <roman@advem.lv> wrote:
> [Me]
>> As usual they are just keeping it out of mainline, and this
>> is adding sched_clock() and other goodies so should
>> probably be upstreamed ASAP.
>
>
> This is what I initially submitted here.

Ah. I missed part of the start of the discussion it seems.

> But Arnd wanted to see a smaller patch which only fixes gemini timer issue
> (t.i. shed clock being 100Hz instead of 25MHz).
> Also I've submitted 4.1 support yesterday, but patches didn't change much
> comparing to 3.18

On the other hand I think the bigger patch is OK actually.

I rewrote the timer support some time back and did not get
anyone to test it properly at the time, so the patch went in
untested. So this is a fix.

I think this should be applied as-is since you can actually
test the whole thing on real hardware, and also add:

Fixes: f3372c01816e "ARM: gemini: convert to GENERIC_CLOCKEVENTS"
Reviewed-by: Linus Walleij <linus.walleij@linaro.org>

> I would love to see all those patches/drivers in mainline - tell me where do
> I submit all that :)

I recently looked at some of the patch stacks in OpenWRT and
get a bit sad that the stuff did not go upstream, so I am ready
to help in any way I can.

Anything GPIO related I am ready to help in reviewing/applying
and I will also help out with ARM32 if I can. Sadly I do not have
a Gemini system.

Yours,
Linus Walleij
Arnd Bergmann July 12, 2015, 9:06 p.m. UTC | #7 
On Friday 10 July 2015 10:48:26 Linus Walleij wrote:
> On Thu, Jul 9, 2015 at 6:48 PM, Roman Yeryomin <roman@advem.lv> wrote:
> > [Me]
> >> As usual they are just keeping it out of mainline, and this
> >> is adding sched_clock() and other goodies so should
> >> probably be upstreamed ASAP.
> >
> >
> > This is what I initially submitted here.
> 
> Ah. I missed part of the start of the discussion it seems.
> 
> > But Arnd wanted to see a smaller patch which only fixes gemini timer issue
> > (t.i. shed clock being 100Hz instead of 25MHz).
> > Also I've submitted 4.1 support yesterday, but patches didn't change much
> > comparing to 3.18
> 
> On the other hand I think the bigger patch is OK actually.
> 
> I rewrote the timer support some time back and did not get
> anyone to test it properly at the time, so the patch went in
> untested. So this is a fix.

The bigger patch seemed a little too verbose to have it backported
to stable kernels, so I asked for a minimal fix that could be marked
as stable, with the other changes applied on top for the current
arm-soc tree.

I'm still on parental leave and not dealing with patches directly,
so whichever patch we want should get routed through Ulli to arm@kernel.org

	Arnd
Hans Ulli Kroll July 19, 2015, 4:28 p.m. UTC | #8 
Hi

On Thu, 9 Jul 2015, Roman Yeryomin wrote:

> On 2015-07-09 11:04, Arnd Bergmann wrote:
> > On Thursday 09 July 2015 01:29:39 Roman Yeryomin wrote:
> > > 
> > > OK, here are the minimal changes required (see attachment). Tested on
> > > 4.1.1
> > > Let me know if you want me to submit it in some other way.
> > > 
> > 
> > The changes look ok to me. Please submit them as two separate
> > patches (1. set up and use timer3, 2. register sched_clock) according
> > to the Documentation/SubmittingPatches description.
> > 
> 
> So what do we do?
> Do you still want me to split this as you said and submit or you want Hans's
> patches rebased/resubmitted?
> 
> Regards,
> Roman
> 

can you rebase your patches againt my branch at

git://github.com/ulli-kroll/linux.git gemini/clockevents/4.3

I've added here the clockevents changes for 4.3 and patch fory
arch/arm/mach-gemini 
This changes will then go arm@kernel.org.

Hans Ulli
diff mbox

Patch

--- a/arch/arm/mach-gemini/time.c
+++ b/arch/arm/mach-gemini/time.c
@@ -15,6 +15,7 @@ 
 #include <asm/mach/time.h>
 #include <linux/clockchips.h>
 #include <linux/clocksource.h>
+#include <linux/sched_clock.h>
 
 /*
  * Register definitions for the timers
@@ -34,9 +35,17 @@ 
 #define TIMER_3_CR_ENABLE		(1 << 6)
 #define TIMER_3_CR_CLOCK		(1 << 7)
 #define TIMER_3_CR_INT			(1 << 8)
+#define TIMER_1_CR_UPDOWN		(1 << 9)
+#define TIMER_2_CR_UPDOWN		(1 << 10)
+#define TIMER_3_CR_UPDOWN		(1 << 11)
 
 static unsigned int tick_rate;
 
+static u64 notrace gemini_read_sched_clock(void)
+{
+	return readl(TIMER_COUNT(IO_ADDRESS(GEMINI_TIMER3_BASE)));
+}
+
 static int gemini_timer_set_next_event(unsigned long cycles,
 				       struct clock_event_device *evt)
 {
@@ -155,14 +164,18 @@  void __init gemini_timer_init(void)
 	 */
 	setup_irq(IRQ_TIMER2, &gemini_timer_irq);
 
-	/* Enable and use TIMER1 as clock source */
-	writel(0xffffffff, TIMER_COUNT(IO_ADDRESS(GEMINI_TIMER1_BASE)));
-	writel(0xffffffff, TIMER_LOAD(IO_ADDRESS(GEMINI_TIMER1_BASE)));
-	writel(TIMER_1_CR_ENABLE, TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
-	if (clocksource_mmio_init(TIMER_COUNT(IO_ADDRESS(GEMINI_TIMER1_BASE)),
-				  "TIMER1", tick_rate, 300, 32,
+	/* Enable and use TIMER3 as clock source */
+	writel(0, TIMER_COUNT(IO_ADDRESS(GEMINI_TIMER3_BASE)));
+	writel(0, TIMER_LOAD(IO_ADDRESS(GEMINI_TIMER3_BASE)));
+	writel(0, TIMER_MATCH1(IO_ADDRESS(GEMINI_TIMER3_BASE)));
+	writel(0, TIMER_MATCH2(IO_ADDRESS(GEMINI_TIMER3_BASE)));
+	writel(TIMER_3_CR_ENABLE | TIMER_3_CR_UPDOWN,
+		TIMER_CR(IO_ADDRESS(GEMINI_TIMER_BASE)));
+	if (clocksource_mmio_init(TIMER_COUNT(IO_ADDRESS(GEMINI_TIMER3_BASE)),
+				  "TIMER3", tick_rate, 300, 32,
 				  clocksource_mmio_readl_up))
 		pr_err("timer: failed to initialize gemini clock source\n");
+	sched_clock_register(gemini_read_sched_clock, 32, tick_rate);
 
 	/* Configure and register the clockevent */
 	clockevents_config_and_register(&gemini_clockevent, tick_rate,