Re: [PATCH v2] OMAP:GPTIMER:1ms tick generation correction

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



DebBarma, Tarun Kanti had written, on 06/21/2010 07:51 AM, the following:
Nishant,

-----Original Message-----
From: Nishanth Menon [mailto:menon.nishanth@xxxxxxxxx]
Sent: Monday, June 21, 2010 4:15 PM
To: DebBarma, Tarun Kanti
Cc: linux-omap@xxxxxxxxxxxxxxx; R, Sricharan
Subject: Re: [PATCH v2] OMAP:GPTIMER:1ms tick generation correction

NAK - my prev comments are not fixed here either.

On 06/21/2010 03:23 PM, Tarun Kanti DebBarma wrote:
Generation of 1ms granular GPTIMER events using 32KHz or system
clocks as inputs does not have whole number count value to load
into the register. This inaccurate count value with respect to 1ms
period leads to time drift subsequently. OMAP3 and later silicons
have dedicated registers for GPTIMER1, GPTIMER2 and GPTIMER10,
which can be programmed with computed values to keep this error
controlled within specified limit.

Version 2:
move this to diffstat section please.
(i) optimized omap_dm_timer_ms_correction() function and corrected
error in computing the positive and negative increments.
(ii) typo corrections in comment section and warning removal related
to 80-character limit

Tested on Zoom3 using Linus tree.

Signed-off-by: R Sricharan<r.sricharan@xxxxxx>
Signed-off-by: Tarun Kanti DebBarma<tarun.kanti@xxxxxx>
---
  arch/arm/plat-omap/dmtimer.c              |  131
+++++++++++++++++++++--------
  arch/arm/plat-omap/include/plat/dmtimer.h |    1 +
  2 files changed, 96 insertions(+), 36 deletions(-)

diff --git a/arch/arm/plat-omap/dmtimer.c b/arch/arm/plat-omap/dmtimer.c
index c64875f..42344a7
--- a/arch/arm/plat-omap/dmtimer.c
+++ b/arch/arm/plat-omap/dmtimer.c
@@ -160,6 +160,9 @@ struct omap_dm_timer {
  	unsigned reserved:1;
  	unsigned enabled:1;
  	unsigned posted:1;
+#ifdef CONFIG_ARCH_OMAP2PLUS
+	unsigned ms_correction:1;
+#endif
  };

  static int dm_timer_count;
@@ -185,18 +188,30 @@ static const int omap1_dm_timer_count =
ARRAY_SIZE(omap1_dm_timers);
  #ifdef CONFIG_ARCH_OMAP2
  static struct omap_dm_timer omap2_dm_timers[] = {
-	{ .phys_base = 0x48028000, .irq = INT_24XX_GPTIMER1 },
-	{ .phys_base = 0x4802a000, .irq = INT_24XX_GPTIMER2 },
-	{ .phys_base = 0x48078000, .irq = INT_24XX_GPTIMER3 },
-	{ .phys_base = 0x4807a000, .irq = INT_24XX_GPTIMER4 },
-	{ .phys_base = 0x4807c000, .irq = INT_24XX_GPTIMER5 },
-	{ .phys_base = 0x4807e000, .irq = INT_24XX_GPTIMER6 },
-	{ .phys_base = 0x48080000, .irq = INT_24XX_GPTIMER7 },
-	{ .phys_base = 0x48082000, .irq = INT_24XX_GPTIMER8 },
-	{ .phys_base = 0x48084000, .irq = INT_24XX_GPTIMER9 },
-	{ .phys_base = 0x48086000, .irq = INT_24XX_GPTIMER10 },
-	{ .phys_base = 0x48088000, .irq = INT_24XX_GPTIMER11 },
-	{ .phys_base = 0x4808a000, .irq = INT_24XX_GPTIMER12 },
+	{ .phys_base = 0x48028000, .irq = INT_24XX_GPTIMER1,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4802a000, .irq = INT_24XX_GPTIMER2,
+						.ms_correction = 0 },
+	{ .phys_base = 0x48078000, .irq = INT_24XX_GPTIMER3,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4807a000, .irq = INT_24XX_GPTIMER4,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4807c000, .irq = INT_24XX_GPTIMER5,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4807e000, .irq = INT_24XX_GPTIMER6,
+						.ms_correction = 0 },
+	{ .phys_base = 0x48080000, .irq = INT_24XX_GPTIMER7,
+						.ms_correction = 0 },
+	{ .phys_base = 0x48082000, .irq = INT_24XX_GPTIMER8,
+						.ms_correction = 0 },
+	{ .phys_base = 0x48084000, .irq = INT_24XX_GPTIMER9,
+						.ms_correction = 0 },
+	{ .phys_base = 0x48086000, .irq = INT_24XX_GPTIMER10,
+						.ms_correction = 0 },
+	{ .phys_base = 0x48088000, .irq = INT_24XX_GPTIMER11,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4808a000, .irq = INT_24XX_GPTIMER12,
+						.ms_correction = 0 },
ignore of previous comments on .ms_correction initialization.
try the output of the following code:
struct b {
         unsigned char z:1;
         unsigned char a:1;
};

static struct b b1[2]={
         {.z =1},
         {.a = 1},
};

unsigned int main()
{
         int i;
         for (i=0; i<2; i++)
         printf("%d: %d %d\n", i, b1[i].z, b1[i].a);
}
This comment is not clear to me.
Are you suggesting to make the initialization of ms_correction variable
in a separate function instead of doing the present way?

no. my suggestion is this:
you dont need to do .ms_correction = 0
when the variable is static - by C standards static variables are initialized to 0. so, .ms_correction = 1 alone is needed in the specific GPTimer cases where this is needed.

the example i posted shows that it works :).


  };

  static const char *omap2_dm_source_names[] __initdata = {
@@ -218,18 +233,30 @@ static const int omap2_dm_timer_count =
ARRAY_SIZE(omap2_dm_timers);
  #ifdef CONFIG_ARCH_OMAP3
  static struct omap_dm_timer omap3_dm_timers[] = {
-	{ .phys_base = 0x48318000, .irq = INT_24XX_GPTIMER1 },
-	{ .phys_base = 0x49032000, .irq = INT_24XX_GPTIMER2 },
-	{ .phys_base = 0x49034000, .irq = INT_24XX_GPTIMER3 },
-	{ .phys_base = 0x49036000, .irq = INT_24XX_GPTIMER4 },
-	{ .phys_base = 0x49038000, .irq = INT_24XX_GPTIMER5 },
-	{ .phys_base = 0x4903A000, .irq = INT_24XX_GPTIMER6 },
-	{ .phys_base = 0x4903C000, .irq = INT_24XX_GPTIMER7 },
-	{ .phys_base = 0x4903E000, .irq = INT_24XX_GPTIMER8 },
-	{ .phys_base = 0x49040000, .irq = INT_24XX_GPTIMER9 },
-	{ .phys_base = 0x48086000, .irq = INT_24XX_GPTIMER10 },
-	{ .phys_base = 0x48088000, .irq = INT_24XX_GPTIMER11 },
-	{ .phys_base = 0x48304000, .irq = INT_34XX_GPT12_IRQ },
+	{ .phys_base = 0x48318000, .irq = INT_24XX_GPTIMER1,
+						.ms_correction = 1 },
+	{ .phys_base = 0x49032000, .irq = INT_24XX_GPTIMER2,
+						.ms_correction = 1 },
+	{ .phys_base = 0x49034000, .irq = INT_24XX_GPTIMER3,
+						.ms_correction = 0 },
+	{ .phys_base = 0x49036000, .irq = INT_24XX_GPTIMER4,
+						.ms_correction = 0 },
+	{ .phys_base = 0x49038000, .irq = INT_24XX_GPTIMER5,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4903A000, .irq = INT_24XX_GPTIMER6,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4903C000, .irq = INT_24XX_GPTIMER7,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4903E001, .irq = INT_24XX_GPTIMER8,
+						.ms_correction = 0 },
+	{ .phys_base = 0x49040000, .irq = INT_24XX_GPTIMER9,
+						.ms_correction = 0 },
+	{ .phys_base = 0x48086000, .irq = INT_24XX_GPTIMER10,
+						.ms_correction = 1 },
+	{ .phys_base = 0x48088000, .irq = INT_24XX_GPTIMER11,
+						.ms_correction = 0 },
+	{ .phys_base = 0x48304000, .irq = INT_34XX_GPT12_IRQ,
+
NAK
Will be taken care.
					.ms_correction = 0 },
  };

  static const char *omap3_dm_source_names[] __initdata = {
@@ -250,18 +277,30 @@ static const int omap3_dm_timer_count =
ARRAY_SIZE(omap3_dm_timers);
  #ifdef CONFIG_ARCH_OMAP4
  static struct omap_dm_timer omap4_dm_timers[] = {
-	{ .phys_base = 0x4a318000, .irq = OMAP44XX_IRQ_GPT1 },
-	{ .phys_base = 0x48032000, .irq = OMAP44XX_IRQ_GPT2 },
-	{ .phys_base = 0x48034000, .irq = OMAP44XX_IRQ_GPT3 },
-	{ .phys_base = 0x48036000, .irq = OMAP44XX_IRQ_GPT4 },
-	{ .phys_base = 0x40138000, .irq = OMAP44XX_IRQ_GPT5 },
-	{ .phys_base = 0x4013a000, .irq = OMAP44XX_IRQ_GPT6 },
-	{ .phys_base = 0x4013a000, .irq = OMAP44XX_IRQ_GPT7 },
-	{ .phys_base = 0x4013e000, .irq = OMAP44XX_IRQ_GPT8 },
-	{ .phys_base = 0x4803e000, .irq = OMAP44XX_IRQ_GPT9 },
-	{ .phys_base = 0x48086000, .irq = OMAP44XX_IRQ_GPT10 },
-	{ .phys_base = 0x48088000, .irq = OMAP44XX_IRQ_GPT11 },
-	{ .phys_base = 0x4a320000, .irq = OMAP44XX_IRQ_GPT12 },
+	{ .phys_base = 0x4a318000, .irq = OMAP44XX_IRQ_GPT1,
+						.ms_correction = 1 },
+	{ .phys_base = 0x48032000, .irq = OMAP44XX_IRQ_GPT2,
+						.ms_correction = 1 },
+	{ .phys_base = 0x48034000, .irq = OMAP44XX_IRQ_GPT3,
+						.ms_correction = 0 },
+	{ .phys_base = 0x48036000, .irq = OMAP44XX_IRQ_GPT4,
+						.ms_correction = 0 },
+	{ .phys_base = 0x40138000, .irq = OMAP44XX_IRQ_GPT5,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4013a000, .irq = OMAP44XX_IRQ_GPT6,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4013a000, .irq = OMAP44XX_IRQ_GPT7,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4013e000, .irq = OMAP44XX_IRQ_GPT8,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4803e000, .irq = OMAP44XX_IRQ_GPT9,
+						.ms_correction = 0 },
+	{ .phys_base = 0x48086000, .irq = OMAP44XX_IRQ_GPT10,
+						.ms_correction = 1 },
+	{ .phys_base = 0x48088000, .irq = OMAP44XX_IRQ_GPT11,
+						.ms_correction = 0 },
+	{ .phys_base = 0x4a320000, .irq = OMAP44XX_IRQ_GPT12,
+
						.ms_correction = 0 },
NAK
Will be taken care.
  };
  static const char *omap4_dm_source_names[] __initdata = {
  	"sys_clkin_ck",
@@ -612,6 +651,10 @@ void omap_dm_timer_set_load_start(struct
omap_dm_timer *timer, int autoreload,
  {
  	u32 l;

+#ifdef CONFIG_ARCH_OMAP2PLUS
+	if (timer->ms_correction)
+		omap_dm_timer_ms_correction(timer, load);
+#endif
  	l = omap_dm_timer_read_reg(timer, OMAP_TIMER_CTRL_REG);
  	if (autoreload) {
  		l |= OMAP_TIMER_CTRL_AR;
@@ -733,6 +776,22 @@ int omap_dm_timers_active(void)
  }
  EXPORT_SYMBOL_GPL(omap_dm_timers_active);

+/**
+ * omap_dm_timer_ms_correction - hardware correction for millisecond
timer
+ * @timer: GPTIMER on which the correction support is to be applied
+ * @load: timer load value, used here to extract the expiry count
+ */
+void omap_dm_timer_ms_correction(struct omap_dm_timer *timer, u32 load)
does this function need to be exposed to the world? why?
omap_dm_timer_set_load_start seems to automatically use it. and
omap_dm_timer_set_load_start is private within dmtimer.c
This will be changed to static void omap_dm_timer_ms_correction (...)

+{
+	int pos_increment, neg_increment;
+	unsigned int count = (0xFFFFFFFF - load)*1024;
+
+	pos_increment = (((count/1000)+1)*1000000) - (count*1000);
+	neg_increment = ((count/1000)*1000000) - (count*1000);
NAK. comments from last time not fixed! - fix check patch AND use
kernel.h ROUND functions.
I did not quite understand the comment. With regard to using round() macro in kernel.h is there any advantage?

yes, ROUND macros are safe from a perspective of overflow and truncation - further this is a standard mechanism used elsewhere in the kernel. so why define your own macros?

Also scripts/checkpatch.pl --strict has not been run on this patch, you definitely need space between operators such as / and * ..


+	omap_dm_timer_write_reg(timer, OMAP_TIMER_TICK_POS_REG,
pos_increment);
+	omap_dm_timer_write_reg(timer, OMAP_TIMER_TICK_NEG_REG,
neg_increment);
+}
+
  int __init omap_dm_timer_init(void)
  {
  	struct omap_dm_timer *timer;
diff --git a/arch/arm/plat-omap/include/plat/dmtimer.h b/arch/arm/plat-
omap/include/plat/dmtimer.h
index 20f1054..ad37bcc 100644
--- a/arch/arm/plat-omap/include/plat/dmtimer.h
+++ b/arch/arm/plat-omap/include/plat/dmtimer.h
@@ -80,5 +80,6 @@ void omap_dm_timer_write_counter(struct omap_dm_timer
*timer, unsigned int value
  int omap_dm_timers_active(void);

+void omap_dm_timer_ms_correction(struct omap_dm_timer *timer, u32
load);
as mentioned offline: why does this function need to be exposed to the
world?
This will be removed.
  #endif /* __ASM_ARCH_DMTIMER_H */

Regards,
Nishanth Menon
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


--
Regards,
Nishanth Menon
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html


[Index of Archives]     [Linux Arm (vger)]     [ARM Kernel]     [ARM MSM]     [Linux Tegra]     [Linux WPAN Networking]     [Linux Wireless Networking]     [Maemo Users]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite Trails]     [Linux Kernel]     [Linux SCSI]

  Powered by Linux