[PATCH 0/5] Add high resolution generic time support for S3C64xx

(no body)

(no body)

(no body)

(no body)

(no body)

(no body)

On Wed, Aug 31, 2011 at 02:34:15PM +0200, Tomasz Figa wrote:

This patch is a prerequisite to adding generic time support for S3C64xx.
It makes possible to differentiate SoCs, required to exclude timers 3 and 4
from PWM driver only on S3C64xx.

Now we have the cpu_is_foo() support for Samsung CPUs can we use that
instead?

2011/8/31 Mark Brown [off-list ref]:

On Wed, Aug 31, 2011 at 02:34:15PM +0200, Tomasz Figa wrote:

quoted

This patch is a prerequisite to adding generic time support for S3C64xx.
It makes possible to differentiate SoCs, required to exclude timers 3 and 4
from PWM driver only on S3C64xx.

Now we have the cpu_is_foo() support for Samsung CPUs can we use that
instead?

Well, I thought about it, but I based my patches on latest stable sources and
cpu_is_s3c64xx wasn't there yet. Should I rebase them to Linus' tree?

I'm not yet fully used to submitting patches here, so things like choosing the
base for patches are sometimes a bit confusing to me and I couldn't find any
guide. Should fixes be based on stable tree and new features implemented on
top of current Linus' tree? Could you give me some advice on this?

Tomasz Figa wrote:

2011/8/31 Mark Brown [off-list ref]:

quoted

On Wed, Aug 31, 2011 at 02:34:15PM +0200, Tomasz Figa wrote:

quoted

This patch is a prerequisite to adding generic time support for

S3C64xx.

quoted

quoted

It makes possible to differentiate SoCs, required to exclude timers 3

and 4

quoted

quoted

from PWM driver only on S3C64xx.

Now we have the cpu_is_foo() support for Samsung CPUs can we use that
instead?

Well, I thought about it, but I based my patches on latest stable sources

and

cpu_is_s3c64xx wasn't there yet. Should I rebase them to Linus' tree?

I'm not yet fully used to submitting patches here, so things like choosing

the

base for patches are sometimes a bit confusing to me and I couldn't find

any

guide. Should fixes be based on stable tree and new features implemented

on

top of current Linus' tree? Could you give me some advice on this?

Hi Tomasz,

You can use "soc_is_s3c64xx()" based on Samsung -next tree.
git://git.kernel.org/pub/scm/linux/kernel/git/kgene/linux-samsung.git
for-next

In addition, would be better to me if you could work about Samsung stuff
based on it.

As a note, your previous patches are in there :)

If any problems, please let me know...

Thanks.

Best regards,
Kgene.
--
Kukjin Kim [off-list ref], Senior Engineer,
SW Solution Development Team, Samsung Electronics Co., Ltd.

On Thu, Sep 1, 2011 at 11:26 AM, Kukjin Kim [off-list ref] wrote:

Tomasz Figa wrote:

quoted

2011/8/31 Mark Brown [off-list ref]:

quoted

On Wed, Aug 31, 2011 at 02:34:15PM +0200, Tomasz Figa wrote:

quoted

This patch is a prerequisite to adding generic time support for

S3C64xx.

quoted

quoted

quoted

It makes possible to differentiate SoCs, required to exclude timers 3

and 4

quoted

quoted

quoted

from PWM driver only on S3C64xx.

Now we have the cpu_is_foo() support for Samsung CPUs can we use that
instead?

Well, I thought about it, but I based my patches on latest stable sources

and

quoted

cpu_is_s3c64xx wasn't there yet. Should I rebase them to Linus' tree?

I'm not yet fully used to submitting patches here, so things like choosing

the

quoted

base for patches are sometimes a bit confusing to me and I couldn't find

any

quoted

guide. Should fixes be based on stable tree and new features implemented

on

quoted

top of current Linus' tree? Could you give me some advice on this?

Hi Tomasz,

You can use "soc_is_s3c64xx()" based on Samsung -next tree.
git://git.kernel.org/pub/scm/linux/kernel/git/kgene/linux-samsung.git
for-next

+static struct platform_device_id s3c_pwm_driver_ids[] = {
+       {
+               .name           = "s3c24xx-pwm",
+               .driver_data    = TYPE_GENERIC,
+       }, {
+               .name           = "s3c64xx-pwm",
+               .driver_data    = TYPE_S3C64XX,
+       },
+};
+MODULE_DEVICE_TABLE(platform, s3c_pwm_driver_ids);

doesn't it right way?
do you want to use the soc_is_* at device?

In addition, would be better to me if you could work about Samsung stuff
based on it.

As a note, your previous patches are in there :)

If any problems, please let me know...

Thanks.

Best regards,
Kgene.
--
Kukjin Kim [off-list ref], Senior Engineer,
SW Solution Development Team, Samsung Electronics Co., Ltd.


_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel at lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

On Thursday 01 of September 2011 at 12:33:10, Kyungmin Park wrote:

On Thu, Sep 1, 2011 at 11:26 AM, Kukjin Kim [off-list ref] wrote:

quoted

Tomasz Figa wrote:

quoted

2011/8/31 Mark Brown [off-list ref]:

quoted

On Wed, Aug 31, 2011 at 02:34:15PM +0200, Tomasz Figa wrote:

quoted

This patch is a prerequisite to adding generic time support for

S3C64xx.

quoted

quoted

quoted

It makes possible to differentiate SoCs, required to exclude timers 3

and 4

quoted

quoted

quoted

from PWM driver only on S3C64xx.

Now we have the cpu_is_foo() support for Samsung CPUs can we use that
instead?

[snip]

Hi Tomasz,

You can use "soc_is_s3c64xx()" based on Samsung -next tree.
git://git.kernel.org/pub/scm/linux/kernel/git/kgene/linux-samsung.git
for-next

+static struct platform_device_id s3c_pwm_driver_ids[] = {
+       {
+               .name           = "s3c24xx-pwm",
+               .driver_data    = TYPE_GENERIC,
+       }, {
+               .name           = "s3c64xx-pwm",
+               .driver_data    = TYPE_S3C64XX,
+       },
+};
+MODULE_DEVICE_TABLE(platform, s3c_pwm_driver_ids);

doesn't it right way?
do you want to use the soc_is_* at device?

Well, now we have three alternatives:
- to leave it as is (i.e. changing platform device names)
- to use cpu_is_* (does it ever exist for S3C64xx or soc_is_s3c64xx is its equivalent?)
- to use soc_is_* 

Waiting for more comments then.

Best regards,
Tom

On Thu, Sep 01, 2011 at 01:18:32PM +0200, Tomasz Figa wrote:

On Thursday 01 of September 2011 at 12:33:10, Kyungmin Park wrote:

Well, now we have three alternatives:
- to leave it as is (i.e. changing platform device names)
- to use cpu_is_* (does it ever exist for S3C64xx or soc_is_s3c64xx is its equivalent?)
- to use soc_is_* 

soc_is is the same thing.

On Thursday 01 of September 2011 at 13:18:32, Tomasz Figa wrote:

On Thursday 01 of September 2011 at 12:33:10, Kyungmin Park wrote:

quoted

On Thu, Sep 1, 2011 at 11:26 AM, Kukjin Kim [off-list ref] wrote:

quoted

Tomasz Figa wrote:

quoted

2011/8/31 Mark Brown [off-list ref]:

quoted

On Wed, Aug 31, 2011 at 02:34:15PM +0200, Tomasz Figa wrote:

quoted

This patch is a prerequisite to adding generic time support for

S3C64xx.

quoted

quoted

quoted

It makes possible to differentiate SoCs, required to exclude timers 3

and 4

quoted

quoted

quoted

from PWM driver only on S3C64xx.

Now we have the cpu_is_foo() support for Samsung CPUs can we use that
instead?

[snip]

Hi Tomasz,

You can use "soc_is_s3c64xx()" based on Samsung -next tree.
git://git.kernel.org/pub/scm/linux/kernel/git/kgene/linux-samsung.git
for-next

+static struct platform_device_id s3c_pwm_driver_ids[] = {
+       {
+               .name           = "s3c24xx-pwm",
+               .driver_data    = TYPE_GENERIC,
+       }, {
+               .name           = "s3c64xx-pwm",
+               .driver_data    = TYPE_S3C64XX,
+       },
+};
+MODULE_DEVICE_TABLE(platform, s3c_pwm_driver_ids);

doesn't it right way?
do you want to use the soc_is_* at device?

Well, now we have three alternatives:
- to leave it as is (i.e. changing platform device names)
- to use cpu_is_* (does it ever exist for S3C64xx or soc_is_s3c64xx is its equivalent?)
- to use soc_is_* 

Waiting for more comments then.

Ping.

Should I keep it as is or rather change it to use soc_is_s3c64xx instead?

Best regards,
Tom

On Tue, Sep 06, 2011 at 12:41:07PM +0200, Tomasz Figa wrote:

On Thursday 01 of September 2011 at 13:18:32, Tomasz Figa wrote:

quoted

Well, now we have three alternatives:
- to leave it as is (i.e. changing platform device names)
- to use cpu_is_* (does it ever exist for S3C64xx or soc_is_s3c64xx is its equivalent?)
- to use soc_is_* 

Waiting for more comments then.

Ping.

Should I keep it as is or rather change it to use soc_is_s3c64xx instead?

I don't know what question you're asking (the questions don't make sense
in terms of the quoted context.)

If you're asking whether you should use a cpu_ prefix instead of a soc_
prefix for identifying s3c64xx, then consider the question.

What is the CPU?

What are S3C64xx / OMAP / PXA310 / SA-11x0 / AT91RM9200?  CPU or SoC ?
What are Cortex A8/A9 / Xscale / StrongARM / ARM920?  CPU or SoC ?

On Tuesday 06 of September 2011 at 12:03:11, Russell King - ARM Linux wrote:

On Tue, Sep 06, 2011 at 12:41:07PM +0200, Tomasz Figa wrote:

quoted

On Thursday 01 of September 2011 at 13:18:32, Tomasz Figa wrote:

quoted

Well, now we have three alternatives:
- to leave it as is (i.e. changing platform device names)
- to use cpu_is_* (does it ever exist for S3C64xx or soc_is_s3c64xx is its equivalent?)
- to use soc_is_* 

Waiting for more comments then.

Ping.

Should I keep it as is or rather change it to use soc_is_s3c64xx instead?

I don't know what question you're asking (the questions don't make sense
in terms of the quoted context.)

If you're asking whether you should use a cpu_ prefix instead of a soc_
prefix for identifying s3c64xx, then consider the question.

What is the CPU?

What are S3C64xx / OMAP / PXA310 / SA-11x0 / AT91RM9200?  CPU or SoC ?
What are Cortex A8/A9 / Xscale / StrongARM / ARM920?  CPU or SoC ?

Well, I believe that the context I quoted explains everything, but let me
explain that again:

The patch is about a driver being able to check to what SoC a device belongs,
but there are at least two ways to achieve that.

I can see many drivers using the way I used, i.e. distinct platform device names
for each SoC that has to be differentiated plus a generic one. However, I was
suggested that using cpu_is_*/soc_is_* might be a better approach, while at the
same time I received an opinion that the original method is the preferred one.

Best regards,
Tom

On Tue, Sep 06, 2011 at 01:35:58PM +0200, Tomasz Figa wrote:

Well, I believe that the context I quoted explains everything, but let me
explain that again:

The patch is about a driver being able to check to what SoC a device belongs,
but there are at least two ways to achieve that.

I can see many drivers using the way I used, i.e. distinct platform device names
for each SoC that has to be differentiated plus a generic one. However, I was
suggested that using cpu_is_*/soc_is_* might be a better approach, while at the
same time I received an opinion that the original method is the preferred one.

Ok.  There's pros and cons here:

Pros to soc_is_*:
1. Doesn't require additional storage in every driver but does require code.
2. Obvious what's going on for a specific SoC when reading the code.

Cons:
1. Needs to be updated each time a new SoC is added.
2. Doesn't scale well - number of soc_is_* will grow to unmanageable levels
   over time.

Pros to platform driver name:
1. Can re-use existing names if feature compatible without driver mods.
2. Scales well with an increasing number of SoCs.

Cons:
1. People will hate having SoC names which don't refer to their exact SoC.
2. Probably requires storage of a set of flags in driver private data
   to identify SoC specific features.
3. Requires additional string space to identify each driver name.

However, thinking about this more wrt DT, there's another aspect to this.
Rather than encoding into the driver "this SoC has features and quirks
X,Y,Z" maybe that information should be in the device tree itself.  For
example, SoC 1 has X and Z, SoC 2 has Y.  Then a new SoC 3 comes along
with X and Y but not Z.  If X, Y, Z are encoded into DT then there's no
need to touch the kernel to support SoC 3, not even to change driver
names or soc_is_xxx macros etc.

So maybe the answer is: platform driver name determines a set of feature
flags for the driver in lieu of DT support, and when DT support comes
along the flags are controlled by DT properties instead.

On Tue, Sep 06, 2011 at 01:04:30PM +0100, Russell King - ARM Linux wrote:

Pros to platform driver name:
1. Can re-use existing names if feature compatible without driver mods.
2. Scales well with an increasing number of SoCs.

Cons:
1. People will hate having SoC names which don't refer to their exact SoC.
2. Probably requires storage of a set of flags in driver private data
   to identify SoC specific features.
3. Requires additional string space to identify each driver name.

There's also an issue with actually getting the devices together to
register which causes fragility here.

However, thinking about this more wrt DT, there's another aspect to this.
Rather than encoding into the driver "this SoC has features and quirks
X,Y,Z" maybe that information should be in the device tree itself.  For
example, SoC 1 has X and Z, SoC 2 has Y.  Then a new SoC 3 comes along
with X and Y but not Z.  If X, Y, Z are encoded into DT then there's no
need to touch the kernel to support SoC 3, not even to change driver
names or soc_is_xxx macros etc.

This does depend pretty strongly on making sure that the SoC DT is
distributed separately to the board DT - that should be solved now.

Tomasz Figa wrote:

From d1fa581f09d8f8a3cc24d05420e9e65112fd8c12 Mon Sep 17 00:00:00 2001
From: Tomasz Figa <redacted>
Date: Sun, 28 Aug 2011 02:45:06 +0200
Subject: [PATCH 2/5] ARM: s3c64xx: Add generic high resolution time

support

 using PWM timers.

This patch adds a sys_timer implementing generic clock source and clock

event

device on S3C64xx using PWM timers 3 and 4. It can be enabled by a Kconfig
option, based on which either the new sys_timer is used as s3c64xx_timer

or

b/arch/arm/mach-s3c64xx/Makefile

timers.

+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Notes:
+ * - Frequency divisors are currently hardcoded to make both timers

operate

+ *   at 1/6 the frequency of PCLK, so with the usual PCLK frequency of 66

MHz,

+ *   the timers are clocked at 11 MHz giving us the operating range from
+ *   90 nsec to 386 sec.
+ * - PWM registers, especially TCON, are shared with PWM driver, but

since

+ *   all the clock event callbacks are run in atomic context

synchronization

+ *   is not needed.
+ */
+
+#include <linux/sched.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/err.h>
+#include <linux/clk.h>
+#include <linux/clockchips.h>
+#include <linux/platform_device.h>
+
+#include <mach/map.h>
+#include <plat/regs-timer.h>
+#include <asm/mach/time.h>
+
+#define PWM_EVENT	(3)
+#define PWM_SOURCE	(4)
+
+#define ONESHOT		(0)
+#define PERIODIC	(1)
+
+#define TCNT_MAX	(0xffffffff)
+
+static struct clk *timerclk;
+
+/*
+ * PWM timers setup code
+ */
+
+static inline void s3c64xx_pwm_stop(unsigned int pwm_id)
+{
+	unsigned long tcon = __raw_readl(S3C2410_TCON);
+
+	switch (pwm_id) {
+	case 3:
+		tcon &= ~S3C2410_TCON_T3START;
+		break;
+	case 4:
+		tcon &= ~S3C2410_TCON_T4START;
+		break;
+	}
+
+	__raw_writel(tcon, S3C2410_TCON);
+}
+
+static inline void s3c64xx_pwm_init(unsigned int pwm_id, unsigned long

tcnt)

+{
+	unsigned long tcon = __raw_readl(S3C2410_TCON);
+
+	/* timers reload after counting zero, so reduce the count by 1 */
+	--tcnt;
+
+	/* stop the timer and ask it to load the new value */
+	switch (pwm_id) {
+	case 3:
+		tcon &= ~(0xf<<16);
+		tcon |= S3C2410_TCON_T3MANUALUPD;
+		break;
+	case 4:
+		tcon &= ~(7<<20);
+		tcon |= S3C2410_TCON_T4MANUALUPD;
+		break;
+	}
+
+	__raw_writel(tcnt, S3C2410_TCNTB(pwm_id));
+	__raw_writel(tcnt, S3C2410_TCMPB(pwm_id));
+	__raw_writel(tcon, S3C2410_TCON);
+}
+
+static inline void s3c64xx_pwm_start(unsigned int pwm_id, bool periodic)
+{
+	unsigned long tcon = __raw_readl(S3C2410_TCON);
+
+	switch (pwm_id) {
+	case 3:
+		tcon |= S3C2410_TCON_T3START;
+		tcon &= ~S3C2410_TCON_T3MANUALUPD;
+
+		if (periodic)
+			tcon |= S3C2410_TCON_T3RELOAD;
+		else
+			tcon &= ~S3C2410_TCON_T3RELOAD;
+		break;
+	case 4:
+		tcon |= S3C2410_TCON_T4START;
+		tcon &= ~S3C2410_TCON_T4MANUALUPD;
+
+		if (periodic)
+			tcon |= S3C2410_TCON_T4RELOAD;
+		else
+			tcon &= ~S3C2410_TCON_T4RELOAD;
+		break;
+	}
+
+	__raw_writel(tcon, S3C2410_TCON);
+}
+
+/*
+ * Clock event
+ */
+
+static struct clk *event_in;
+static struct clk *event_div;
+
+static unsigned long clock_count_per_tick;
+
+static int s3c64xx_clock_event_set_next_event(unsigned long cycles,
+						struct clock_event_device
*evt)
+{
+	s3c64xx_pwm_init(PWM_EVENT, cycles);
+	s3c64xx_pwm_start(PWM_EVENT, ONESHOT);
+	return 0;
+}
+
+static void s3c64xx_clock_event_resume(void)
+{
+	unsigned long pclk;
+	struct clk *tscaler;
+
+	pclk = clk_get_rate(timerclk);
+	tscaler = clk_get_parent(event_div);
+	clk_set_rate(tscaler, pclk / 3);
+
+	clk_set_rate(event_div, pclk / 6);
+	clk_set_parent(event_in, event_div);
+
+	s3c64xx_pwm_init(PWM_EVENT, clock_count_per_tick);
+	s3c64xx_pwm_start(PWM_EVENT, PERIODIC);
+}
+
+static void s3c64xx_clock_event_set_mode(enum clock_event_mode mode,
+						struct clock_event_device
*evt)
+{
+	s3c64xx_pwm_stop(PWM_EVENT);
+
+	switch (mode) {
+	case CLOCK_EVT_MODE_PERIODIC:
+		s3c64xx_pwm_init(PWM_EVENT, clock_count_per_tick);
+		s3c64xx_pwm_start(PWM_EVENT, PERIODIC);
+		break;
+	case CLOCK_EVT_MODE_ONESHOT:
+		break;
+	case CLOCK_EVT_MODE_UNUSED:
+	case CLOCK_EVT_MODE_SHUTDOWN:
+		break;
+	case CLOCK_EVT_MODE_RESUME:
+		s3c64xx_clock_event_resume();
+		break;
+	}
+}
+
+static struct clock_event_device s3c64xx_clock_event_device = {
+	.name		= "s3c64xx_clkevt",
+	.features       = CLOCK_EVT_FEAT_PERIODIC |
CLOCK_EVT_FEAT_ONESHOT,
+	.rating		= 200,
+	.shift		= 32,
+	.set_next_event	= s3c64xx_clock_event_set_next_event,
+	.set_mode	= s3c64xx_clock_event_set_mode,
+};
+
+static irqreturn_t s3c64xx_clock_event_isr(int irq, void *dev_id)
+{
+	struct clock_event_device *evt = &s3c64xx_clock_event_device;
+
+	evt->event_handler(evt);
+
+	return IRQ_HANDLED;
+}
+
+static struct irqaction s3c64xx_clock_event_irq = {
+	.name		= "s3c64xx_clkevt_irq",
+	.flags		= IRQF_DISABLED | IRQF_TIMER | IRQF_IRQPOLL,
+	.handler	= s3c64xx_clock_event_isr,
+};
+
+static void __init s3c64xx_clock_event_init(void)
+{
+	unsigned long clock_rate;
+
+	clock_rate = clk_get_rate(event_in);
+
+	clock_count_per_tick = clock_rate / HZ;
+
+	s3c64xx_clock_event_device.mult = div_sc(clock_rate,
+			NSEC_PER_SEC, s3c64xx_clock_event_device.shift);
+	s3c64xx_clock_event_device.max_delta_ns =
+			clockevent_delta2ns(-1,

&s3c64xx_clock_event_device);

+	s3c64xx_clock_event_device.min_delta_ns =
+			clockevent_delta2ns(1, &s3c64xx_clock_event_device);
+
+	s3c64xx_clock_event_device.cpumask = cpumask_of(0);
+	clockevents_register_device(&s3c64xx_clock_event_device);
+
+	setup_irq(IRQ_TIMER3, &s3c64xx_clock_event_irq);
+}
+
+/*
+ * Clock source
+ */
+
+static struct clk *source_in;
+static struct clk *source_div;
+
+static cycle_t s3c64xx_clocksource_read(struct clocksource *cs)
+{
+	return (cycle_t) ~__raw_readl(S3C2410_TCNTO(PWM_SOURCE));
+}
+
+static void s3c64xx_clocksource_resume(struct clocksource *cs)
+{
+	unsigned long pclk;
+	struct clk *tscaler;
+
+	pclk = clk_get_rate(timerclk);
+	tscaler = clk_get_parent(source_div);
+	clk_set_rate(tscaler, pclk / 3);
+
+	clk_set_rate(source_div, pclk / 6);
+	clk_set_parent(source_in, source_div);
+
+	s3c64xx_pwm_init(PWM_SOURCE, TCNT_MAX);
+	s3c64xx_pwm_start(PWM_SOURCE, PERIODIC);
+}
+
+static struct clocksource pwm_clocksource = {
+	.name		= "s3c64xx_clksrc",
+	.rating		= 250,
+	.read		= s3c64xx_clocksource_read,
+	.mask		= CLOCKSOURCE_MASK(32),
+	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
+	.resume		= s3c64xx_clocksource_resume,
+};
+
+static void __init s3c64xx_clocksource_init(void)
+{
+	unsigned long clock_rate;
+
+	clock_rate = clk_get_rate(source_in);
+
+	s3c64xx_pwm_init(PWM_SOURCE, TCNT_MAX);
+	s3c64xx_pwm_start(PWM_SOURCE, PERIODIC);
+
+	if (clocksource_register_hz(&pwm_clocksource, clock_rate))
+		panic("%s: can't register clocksource\n",
pwm_clocksource.name);
+}
+
+static void __init s3c64xx_timer_init_common(void)
+{
+	struct platform_device tmpdev;
+	unsigned long pclk;
+	struct clk *tscaler;
+
+	tmpdev.dev.bus = &platform_bus_type;
+
+	timerclk = clk_get(NULL, "timers");
+	if (IS_ERR(timerclk))
+		panic("failed to get timers clock for system timer");
+
+	tmpdev.id = PWM_EVENT;
+	event_in = clk_get(&tmpdev.dev, "pwm-tin");
+	if (IS_ERR(event_in))
+		panic("failed to get pwm-event_in clock for system timer");
+
+	event_div = clk_get(&tmpdev.dev, "pwm-tdiv");
+	if (IS_ERR(event_div))
+		panic("failed to get pwm-event_div clock for system timer");
+
+	tmpdev.id = PWM_SOURCE;
+	source_in = clk_get(&tmpdev.dev, "pwm-tin");
+	if (IS_ERR(source_in))
+		panic("failed to get pwm-source_in clock for system timer");
+
+	source_div = clk_get(&tmpdev.dev, "pwm-tdiv");
+	if (IS_ERR(source_div))
+		panic("failed to get pwm-source_div clock for system

timer");

Hi Tomasz Figa,

Sorry for late response, probably I'm missing this......and I couldn't
review this in detail :(

But as you know above mach-s3c64xx/time.c is similar with
plat-s5p/s5p-time.c so I'd prefer that you could consolidate them,
samsung_timer into plat-samsung directory. In addition, if you need to
implement SoC specific feature, you can use soc_is_xxxx() like
soc_is_s3c64xx() there.

If any problems, please let me know.

Thanks.

Best regards,
Kgene.
--
Kukjin Kim [off-list ref], Senior Engineer,
SW Solution Development Team, Samsung Electronics Co., Ltd.

Mark Brown wrote:

On Tue, Sep 06, 2011 at 01:04:30PM +0100, Russell King - ARM Linux wrote:

quoted

Pros to platform driver name:
1. Can re-use existing names if feature compatible without driver mods.
2. Scales well with an increasing number of SoCs.

quoted

Cons:
1. People will hate having SoC names which don't refer to their exact

SoC.

quoted

2. Probably requires storage of a set of flags in driver private data
   to identify SoC specific features.
3. Requires additional string space to identify each driver name.

There's also an issue with actually getting the devices together to
register which causes fragility here.

quoted

However, thinking about this more wrt DT, there's another aspect to

this.

quoted

Rather than encoding into the driver "this SoC has features and quirks
X,Y,Z" maybe that information should be in the device tree itself.  For
example, SoC 1 has X and Z, SoC 2 has Y.  Then a new SoC 3 comes along
with X and Y but not Z.  If X, Y, Z are encoded into DT then there's no
need to touch the kernel to support SoC 3, not even to change driver
names or soc_is_xxx macros etc.

This does depend pretty strongly on making sure that the SoC DT is
distributed separately to the board DT - that should be solved now.

Yes, I agree. So we can keep the current way to support each SoCs' device
with one driver until DT world.

Thanks.

Best regards,
Kgene.
--
Kukjin Kim [off-list ref], Senior Engineer,
SW Solution Development Team, Samsung Electronics Co., Ltd.

On Monday 10 of October 2011 at 18:57:58, Kukjin Kim wrote:

Tomasz Figa wrote:

quoted

From d1fa581f09d8f8a3cc24d05420e9e65112fd8c12 Mon Sep 17 00:00:00 2001
From: Tomasz Figa <redacted>
Date: Sun, 28 Aug 2011 02:45:06 +0200
Subject: [PATCH 2/5] ARM: s3c64xx: Add generic high resolution time

support

quoted

 using PWM timers.

This patch adds a sys_timer implementing generic clock source and clock

event

quoted

device on S3C64xx using PWM timers 3 and 4. It can be enabled by a Kconfig
option, based on which either the new sys_timer is used as s3c64xx_timer

or

quoted

s3c64xx_timer is defined to s3c24xx_timer which is the old tick timer.

Signed-off-by: Tomasz Figa <redacted>
---
 arch/arm/Kconfig                         |    1 -
 arch/arm/mach-s3c64xx/Kconfig            |   12 ++
 arch/arm/mach-s3c64xx/Makefile           |    3 +
 arch/arm/mach-s3c64xx/time.c             |  320
++++++++++++++++++++++++++++++
 arch/arm/plat-samsung/include/plat/cpu.h |    7 +
 5 files changed, 342 insertions(+), 1 deletions(-)
 create mode 100644 arch/arm/mach-s3c64xx/time.c

diff --git a/arch/arm/Kconfig b/arch/arm/Kconfig
index 9adc278..05941f6 100644
--- a/arch/arm/Kconfig
+++ b/arch/arm/Kconfig

@@ -700,7 +700,6 @@ config ARCH_S3C64XX
 	select ARM_VIC
 	select HAVE_CLK
 	select NO_IOPORT
-	select ARCH_USES_GETTIMEOFFSET
 	select ARCH_HAS_CPUFREQ
 	select ARCH_REQUIRE_GPIOLIB
 	select SAMSUNG_CLKSRC

diff --git a/arch/arm/mach-s3c64xx/Kconfig b/arch/arm/mach-s3c64xx/Kconfig
index e4177e2..2db76fd 100644
--- a/arch/arm/mach-s3c64xx/Kconfig
+++ b/arch/arm/mach-s3c64xx/Kconfig

@@ -25,6 +25,18 @@ config CPU_S3C6410
 	help
 	  Enable S3C6410 CPU support

+config S3C64XX_GENERIC_CLOCKEVENTS
+	bool "Enable high resolution generic time support using PWM timers"
+	select GENERIC_CLOCKEVENTS
+	help
+	  This option enables high resolution generic time support
+	  on S3C6400/S3C6410 SoCs using PWM timers 3 and 4 instead
+	  of standard periodic tick using PWM timer 4.
+
+config S3C64XX_ARCH_USES_GETTIMEOFFSET
+	def_bool y if !S3C64XX_GENERIC_CLOCKEVENTS
+	select ARCH_USES_GETTIMEOFFSET
+
 config S3C64XX_DMA
 	bool "S3C64XX DMA"
 	select S3C_DMA

diff --git a/arch/arm/mach-s3c64xx/Makefile

b/arch/arm/mach-s3c64xx/Makefile

quoted

index 4657363..a152002 100644

--- a/arch/arm/mach-s3c64xx/Makefile
+++ b/arch/arm/mach-s3c64xx/Makefile

@@ -15,6 +15,9 @@ obj-y				+= cpu.o
 obj-y				+= clock.o
 obj-y				+= gpiolib.o

+# Generic clockevents
+obj-$(CONFIG_S3C64XX_GENERIC_CLOCKEVENTS) += time.o
+
 # Core support for S3C6400 system

 obj-$(CONFIG_CPU_S3C6400)	+= s3c6400.o

diff --git a/arch/arm/mach-s3c64xx/time.c b/arch/arm/mach-s3c64xx/time.c
new file mode 100644
index 0000000..4fa57f0
--- /dev/null
+++ b/arch/arm/mach-s3c64xx/time.c

@@ -0,0 +1,320 @@
+/* linux/arch/arm/mach-s3c64xx/time.c
+ *
+ * Copyright (c) 2011 Tomasz Figa <tomasz.figa@gmail.com>
+ *
+ * based on linux/arch/arm/plat-samsung/s5p-time.c
+ *
+ * S3C64XX generic high resolution time support using PWM 3 and PWM 4

timers.

quoted

+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Notes:
+ * - Frequency divisors are currently hardcoded to make both timers

operate

quoted

+ *   at 1/6 the frequency of PCLK, so with the usual PCLK frequency of 66

MHz,

quoted

+ *   the timers are clocked at 11 MHz giving us the operating range from
+ *   90 nsec to 386 sec.
+ * - PWM registers, especially TCON, are shared with PWM driver, but

since

quoted

+ *   all the clock event callbacks are run in atomic context

synchronization

quoted

+ *   is not needed.
+ */
+
+#include <linux/sched.h>
+#include <linux/interrupt.h>
+#include <linux/irq.h>
+#include <linux/err.h>
+#include <linux/clk.h>
+#include <linux/clockchips.h>
+#include <linux/platform_device.h>
+
+#include <mach/map.h>
+#include <plat/regs-timer.h>
+#include <asm/mach/time.h>
+
+#define PWM_EVENT	(3)
+#define PWM_SOURCE	(4)
+
+#define ONESHOT		(0)
+#define PERIODIC	(1)
+
+#define TCNT_MAX	(0xffffffff)
+
+static struct clk *timerclk;
+
+/*
+ * PWM timers setup code
+ */
+
+static inline void s3c64xx_pwm_stop(unsigned int pwm_id)
+{
+	unsigned long tcon = __raw_readl(S3C2410_TCON);
+
+	switch (pwm_id) {
+	case 3:
+		tcon &= ~S3C2410_TCON_T3START;
+		break;
+	case 4:
+		tcon &= ~S3C2410_TCON_T4START;
+		break;
+	}
+
+	__raw_writel(tcon, S3C2410_TCON);
+}
+
+static inline void s3c64xx_pwm_init(unsigned int pwm_id, unsigned long

tcnt)

quoted

+{
+	unsigned long tcon = __raw_readl(S3C2410_TCON);
+
+	/* timers reload after counting zero, so reduce the count by 1 */
+	--tcnt;
+
+	/* stop the timer and ask it to load the new value */
+	switch (pwm_id) {
+	case 3:
+		tcon &= ~(0xf<<16);
+		tcon |= S3C2410_TCON_T3MANUALUPD;
+		break;
+	case 4:
+		tcon &= ~(7<<20);
+		tcon |= S3C2410_TCON_T4MANUALUPD;
+		break;
+	}
+
+	__raw_writel(tcnt, S3C2410_TCNTB(pwm_id));
+	__raw_writel(tcnt, S3C2410_TCMPB(pwm_id));
+	__raw_writel(tcon, S3C2410_TCON);
+}
+
+static inline void s3c64xx_pwm_start(unsigned int pwm_id, bool periodic)
+{
+	unsigned long tcon = __raw_readl(S3C2410_TCON);
+
+	switch (pwm_id) {
+	case 3:
+		tcon |= S3C2410_TCON_T3START;
+		tcon &= ~S3C2410_TCON_T3MANUALUPD;
+
+		if (periodic)
+			tcon |= S3C2410_TCON_T3RELOAD;
+		else
+			tcon &= ~S3C2410_TCON_T3RELOAD;
+		break;
+	case 4:
+		tcon |= S3C2410_TCON_T4START;
+		tcon &= ~S3C2410_TCON_T4MANUALUPD;
+
+		if (periodic)
+			tcon |= S3C2410_TCON_T4RELOAD;
+		else
+			tcon &= ~S3C2410_TCON_T4RELOAD;
+		break;
+	}
+
+	__raw_writel(tcon, S3C2410_TCON);
+}
+
+/*
+ * Clock event
+ */
+
+static struct clk *event_in;
+static struct clk *event_div;
+
+static unsigned long clock_count_per_tick;
+
+static int s3c64xx_clock_event_set_next_event(unsigned long cycles,
+						struct clock_event_device
*evt)
+{
+	s3c64xx_pwm_init(PWM_EVENT, cycles);
+	s3c64xx_pwm_start(PWM_EVENT, ONESHOT);
+	return 0;
+}
+
+static void s3c64xx_clock_event_resume(void)
+{
+	unsigned long pclk;
+	struct clk *tscaler;
+
+	pclk = clk_get_rate(timerclk);
+	tscaler = clk_get_parent(event_div);
+	clk_set_rate(tscaler, pclk / 3);
+
+	clk_set_rate(event_div, pclk / 6);
+	clk_set_parent(event_in, event_div);
+
+	s3c64xx_pwm_init(PWM_EVENT, clock_count_per_tick);
+	s3c64xx_pwm_start(PWM_EVENT, PERIODIC);
+}
+
+static void s3c64xx_clock_event_set_mode(enum clock_event_mode mode,
+						struct clock_event_device
*evt)
+{
+	s3c64xx_pwm_stop(PWM_EVENT);
+
+	switch (mode) {
+	case CLOCK_EVT_MODE_PERIODIC:
+		s3c64xx_pwm_init(PWM_EVENT, clock_count_per_tick);
+		s3c64xx_pwm_start(PWM_EVENT, PERIODIC);
+		break;
+	case CLOCK_EVT_MODE_ONESHOT:
+		break;
+	case CLOCK_EVT_MODE_UNUSED:
+	case CLOCK_EVT_MODE_SHUTDOWN:
+		break;
+	case CLOCK_EVT_MODE_RESUME:
+		s3c64xx_clock_event_resume();
+		break;
+	}
+}
+
+static struct clock_event_device s3c64xx_clock_event_device = {
+	.name		= "s3c64xx_clkevt",
+	.features       = CLOCK_EVT_FEAT_PERIODIC |
CLOCK_EVT_FEAT_ONESHOT,
+	.rating		= 200,
+	.shift		= 32,
+	.set_next_event	= s3c64xx_clock_event_set_next_event,
+	.set_mode	= s3c64xx_clock_event_set_mode,
+};
+
+static irqreturn_t s3c64xx_clock_event_isr(int irq, void *dev_id)
+{
+	struct clock_event_device *evt = &s3c64xx_clock_event_device;
+
+	evt->event_handler(evt);
+
+	return IRQ_HANDLED;
+}
+
+static struct irqaction s3c64xx_clock_event_irq = {
+	.name		= "s3c64xx_clkevt_irq",
+	.flags		= IRQF_DISABLED | IRQF_TIMER | IRQF_IRQPOLL,
+	.handler	= s3c64xx_clock_event_isr,
+};
+
+static void __init s3c64xx_clock_event_init(void)
+{
+	unsigned long clock_rate;
+
+	clock_rate = clk_get_rate(event_in);
+
+	clock_count_per_tick = clock_rate / HZ;
+
+	s3c64xx_clock_event_device.mult = div_sc(clock_rate,
+			NSEC_PER_SEC, s3c64xx_clock_event_device.shift);
+	s3c64xx_clock_event_device.max_delta_ns =
+			clockevent_delta2ns(-1,

&s3c64xx_clock_event_device);

quoted

+	s3c64xx_clock_event_device.min_delta_ns =
+			clockevent_delta2ns(1, &s3c64xx_clock_event_device);
+
+	s3c64xx_clock_event_device.cpumask = cpumask_of(0);
+	clockevents_register_device(&s3c64xx_clock_event_device);
+
+	setup_irq(IRQ_TIMER3, &s3c64xx_clock_event_irq);
+}
+
+/*
+ * Clock source
+ */
+
+static struct clk *source_in;
+static struct clk *source_div;
+
+static cycle_t s3c64xx_clocksource_read(struct clocksource *cs)
+{
+	return (cycle_t) ~__raw_readl(S3C2410_TCNTO(PWM_SOURCE));
+}
+
+static void s3c64xx_clocksource_resume(struct clocksource *cs)
+{
+	unsigned long pclk;
+	struct clk *tscaler;
+
+	pclk = clk_get_rate(timerclk);
+	tscaler = clk_get_parent(source_div);
+	clk_set_rate(tscaler, pclk / 3);
+
+	clk_set_rate(source_div, pclk / 6);
+	clk_set_parent(source_in, source_div);
+
+	s3c64xx_pwm_init(PWM_SOURCE, TCNT_MAX);
+	s3c64xx_pwm_start(PWM_SOURCE, PERIODIC);
+}
+
+static struct clocksource pwm_clocksource = {
+	.name		= "s3c64xx_clksrc",
+	.rating		= 250,
+	.read		= s3c64xx_clocksource_read,
+	.mask		= CLOCKSOURCE_MASK(32),
+	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
+	.resume		= s3c64xx_clocksource_resume,
+};
+
+static void __init s3c64xx_clocksource_init(void)
+{
+	unsigned long clock_rate;
+
+	clock_rate = clk_get_rate(source_in);
+
+	s3c64xx_pwm_init(PWM_SOURCE, TCNT_MAX);
+	s3c64xx_pwm_start(PWM_SOURCE, PERIODIC);
+
+	if (clocksource_register_hz(&pwm_clocksource, clock_rate))
+		panic("%s: can't register clocksource\n",
pwm_clocksource.name);
+}
+
+static void __init s3c64xx_timer_init_common(void)
+{
+	struct platform_device tmpdev;
+	unsigned long pclk;
+	struct clk *tscaler;
+
+	tmpdev.dev.bus = &platform_bus_type;
+
+	timerclk = clk_get(NULL, "timers");
+	if (IS_ERR(timerclk))
+		panic("failed to get timers clock for system timer");
+
+	tmpdev.id = PWM_EVENT;
+	event_in = clk_get(&tmpdev.dev, "pwm-tin");
+	if (IS_ERR(event_in))
+		panic("failed to get pwm-event_in clock for system timer");
+
+	event_div = clk_get(&tmpdev.dev, "pwm-tdiv");
+	if (IS_ERR(event_div))
+		panic("failed to get pwm-event_div clock for system timer");
+
+	tmpdev.id = PWM_SOURCE;
+	source_in = clk_get(&tmpdev.dev, "pwm-tin");
+	if (IS_ERR(source_in))
+		panic("failed to get pwm-source_in clock for system timer");
+
+	source_div = clk_get(&tmpdev.dev, "pwm-tdiv");
+	if (IS_ERR(source_div))
+		panic("failed to get pwm-source_div clock for system

timer");

quoted

+
+	pclk = clk_get_rate(timerclk);
+	tscaler = clk_get_parent(event_div);
+	clk_set_rate(tscaler, pclk / 3);
+
+	clk_set_rate(event_div, pclk / 6);
+	clk_set_parent(event_in, event_div);
+
+	clk_set_rate(source_div, pclk / 6);
+	clk_set_parent(source_in, source_div);
+
+	clk_enable(timerclk);
+	clk_enable(event_in);
+	clk_enable(source_in);
+}
+
+static void __init s3c64xx_timer_init(void)
+{
+	s3c64xx_timer_init_common();
+	s3c64xx_clock_event_init();
+	s3c64xx_clocksource_init();
+}
+
+struct sys_timer s3c64xx_timer = {
+	.init		= s3c64xx_timer_init,
+};

diff --git a/arch/arm/plat-samsung/include/plat/cpu.h b/arch/arm/plat-

samsung/include/plat/cpu.h
index c0a5741..e1fcad0 100644

--- a/arch/arm/plat-samsung/include/plat/cpu.h
+++ b/arch/arm/plat-samsung/include/plat/cpu.h

@@ -74,6 +74,13 @@ extern struct syscore_ops s3c2416_pm_syscore_ops;
 extern struct syscore_ops s3c244x_pm_syscore_ops;
 extern struct syscore_ops s3c64xx_irq_syscore_ops;

+/* timer for 64xx */
+#ifdef CONFIG_S3C64XX_GENERIC_CLOCKEVENTS
+extern struct sys_timer s3c64xx_timer;
+#else
+#define s3c64xx_timer s3c24xx_timer
+#endif
+
 /* system device classes */

 extern struct sysdev_class s3c2410_sysclass;
--

1.7.6.1

Hi Tomasz Figa,

Sorry for late response, probably I'm missing this......and I couldn't
review this in detail :(

Hi,

Same on my side, sorry for my absence on the list. I've been very busy, with
no time for mainline submissions.

But as you know above mach-s3c64xx/time.c is similar with
plat-s5p/s5p-time.c so I'd prefer that you could consolidate them,
samsung_timer into plat-samsung directory. In addition, if you need to
implement SoC specific feature, you can use soc_is_xxxx() like
soc_is_s3c64xx() there.

OK, I will do that.

If any problems, please let me know.

OK, thanks.

Best regards,
Tom

Thanks.

Best regards,
Kgene.
--
Kukjin Kim [off-list ref], Senior Engineer,
SW Solution Development Team, Samsung Electronics Co., Ltd.

On Wed, Aug 31, 2011 at 02:34:52PM +0200, Tomasz Figa wrote:

+/*
+ * Clock source
+ */
+
+static struct clk *source_in;
+static struct clk *source_div;
+
+static cycle_t s3c64xx_clocksource_read(struct clocksource *cs)
+{
+	return (cycle_t) ~__raw_readl(S3C2410_TCNTO(PWM_SOURCE));
+}

This looks like a very simple MMIO clocksource...  This looks the
same as clocksource_mmio_readl_down().

+
+static void s3c64xx_clocksource_resume(struct clocksource *cs)
+{
+	unsigned long pclk;
+	struct clk *tscaler;
+
+	pclk = clk_get_rate(timerclk);
+	tscaler = clk_get_parent(source_div);
+	clk_set_rate(tscaler, pclk / 3);
+
+	clk_set_rate(source_div, pclk / 6);
+	clk_set_parent(source_in, source_div);
+
+	s3c64xx_pwm_init(PWM_SOURCE, TCNT_MAX);
+	s3c64xx_pwm_start(PWM_SOURCE, PERIODIC);
+}
+
+static struct clocksource pwm_clocksource = {
+	.name		= "s3c64xx_clksrc",
+	.rating		= 250,
+	.read		= s3c64xx_clocksource_read,
+	.mask		= CLOCKSOURCE_MASK(32),
+	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
+	.resume		= s3c64xx_clocksource_resume,
+};
+
+static void __init s3c64xx_clocksource_init(void)
+{
+	unsigned long clock_rate;
+
+	clock_rate = clk_get_rate(source_in);
+
+	s3c64xx_pwm_init(PWM_SOURCE, TCNT_MAX);
+	s3c64xx_pwm_start(PWM_SOURCE, PERIODIC);
+
+	if (clocksource_register_hz(&pwm_clocksource, clock_rate))
+		panic("%s: can't register clocksource\n", pwm_clocksource.name);

Apart from the resume entry, this looks like it could use
drivers/clocksource/mmio.c - and so avoid yet another clocksource
creation.

Maybe adding some kind of resume support to mmio.c would be a good
idea?