Improve timer Resolution on imx28

Improve timer Resolution on imx28

[Torben Hohn]

Hi,

this Series increases the timer frequency and decreases min_delta_ns.
I saw pretty high max latencies with cyclictest: ~600us.

The cause was the min_delta_ns value of 0xf 32kHz clockcycles.
But with the 32kHz Clock, the period is still around 30us.
So we just increase the clock freqeuncy driving the timers
here.

[PATCH 1/2] mxs timer: use ahbx bus clock to drive the timers on timrotv2

[Torben Hohn]

timer resolution of ~32us is pretty low.
v2 has 32bits resolution, so we have quite some headroom, and
can use the 24MHz clock.
v1 has only 16bits, so we only increase v2.

So we just exchange the timrot clock in imx28.
On imx23 we have timrotv1 and everything stays the same.

Signed-off-by: Torben Hohn <torbenh@linutronix.de>
---
 arch/arm/mach-mxs/timer.c   |    9 +++++----
 drivers/clk/mxs/clk-imx28.c |    2 +-
 2 files changed, 6 insertions(+), 5 deletions(-)

diff --git a/arch/arm/mach-mxs/timer.c b/arch/arm/mach-mxs/timer.c
index 856f4c7..689a2b5 100644
--- a/arch/arm/mach-mxs/timer.c
+++ b/arch/arm/mach-mxs/timer.c
@@ -72,8 +72,9 @@
 #define BM_TIMROT_TIMCTRLn_IRQ_EN	(1 << 14)
 #define BM_TIMROT_TIMCTRLn_IRQ		(1 << 15)
 #define BP_TIMROT_TIMCTRLn_SELECT	0
-#define BV_TIMROTv1_TIMCTRLn_SELECT__32KHZ_XTAL	0x8
-#define BV_TIMROTv2_TIMCTRLn_SELECT__32KHZ_XTAL	0xb
+#define BV_TIMROTv1_TIMCTRLn_SELECT__32KHZ_XTAL		0x8
+#define BV_TIMROTv2_TIMCTRLn_SELECT__32KHZ_XTAL		0xb
+#define BV_TIMROTv2_TIMCTRLn_SELECT__TICK_ALWAYS	0xf
 
 static struct clock_event_device mxs_clockevent_device;
 static enum clock_event_mode mxs_clockevent_mode = CLOCK_EVT_MODE_UNUSED;
@@ -288,7 +289,7 @@ void __init mxs_timer_init(void)
 	/* one for clock_event */
 	__raw_writel((timrot_is_v1() ?
 			BV_TIMROTv1_TIMCTRLn_SELECT__32KHZ_XTAL :
-			BV_TIMROTv2_TIMCTRLn_SELECT__32KHZ_XTAL) |
+			BV_TIMROTv2_TIMCTRLn_SELECT__TICK_ALWAYS) |
 			BM_TIMROT_TIMCTRLn_UPDATE |
 			BM_TIMROT_TIMCTRLn_IRQ_EN,
 			mxs_timrot_base + HW_TIMROT_TIMCTRLn(0));
@@ -296,7 +297,7 @@ void __init mxs_timer_init(void)
 	/* another for clocksource */
 	__raw_writel((timrot_is_v1() ?
 			BV_TIMROTv1_TIMCTRLn_SELECT__32KHZ_XTAL :
-			BV_TIMROTv2_TIMCTRLn_SELECT__32KHZ_XTAL) |
+			BV_TIMROTv2_TIMCTRLn_SELECT__TICK_ALWAYS) |
 			BM_TIMROT_TIMCTRLn_RELOAD,
 			mxs_timrot_base + HW_TIMROT_TIMCTRLn(1));
 
diff --git a/drivers/clk/mxs/clk-imx28.c b/drivers/clk/mxs/clk-imx28.c
index db3af08..0174270 100644
--- a/drivers/clk/mxs/clk-imx28.c
+++ b/drivers/clk/mxs/clk-imx28.c
@@ -238,7 +238,7 @@ int __init mx28_clocks_init(void)
 		of_clk_add_provider(np, of_clk_src_onecell_get, &clk_data);
 	}
 
-	clk_register_clkdev(clks[clk32k], NULL, "timrot");
+	clk_register_clkdev(clks[xbus], NULL, "timrot");
 	clk_register_clkdev(clks[enet_out], NULL, "enet_out");
 
 	for (i = 0; i < ARRAY_SIZE(clks_init_on); i++)
-- 
1.7.10.4

[PATCH 2/2] mxs timer: decrease mxs_clockevent_device.min_delta_ns to 2 clock cycles

[Torben Hohn]

Signed-off-by: Torben Hohn <torbenh@linutronix.de>
---
 arch/arm/mach-mxs/timer.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/arch/arm/mach-mxs/timer.c b/arch/arm/mach-mxs/timer.c
index 689a2b5..cf45e0a 100644
--- a/arch/arm/mach-mxs/timer.c
+++ b/arch/arm/mach-mxs/timer.c
@@ -219,7 +219,7 @@ static int __init mxs_clockevent_init(struct clk *timer_clk)
 		mxs_clockevent_device.max_delta_ns =
 			clockevent_delta2ns(0xfffffffe, &mxs_clockevent_device);
 		mxs_clockevent_device.min_delta_ns =
-			clockevent_delta2ns(0xf, &mxs_clockevent_device);
+			clockevent_delta2ns(0x2, &mxs_clockevent_device);
 	}
 
 	clockevents_register_device(&mxs_clockevent_device);
-- 
1.7.10.4

Improve timer Resolution on imx28

[Marek Vasut]

Dear Torben Hohn,

> Hi,
> 
> this Series increases the timer frequency and decreases min_delta_ns.
> I saw pretty high max latencies with cyclictest: ~600us.
> 
> The cause was the min_delta_ns value of 0xf 32kHz clockcycles.
> But with the 32kHz Clock, the period is still around 30us.
> So we just increase the clock freqeuncy driving the timers
> here.

Thanks for the set. Just a really quick question -- won't this under no 
circumstance cause interrupt storm?

Best regards,
Marek Vasut

[PATCH 1/2] mxs timer: use ahbx bus clock to drive the timers on timrotv2

[Shawn Guo]

On Fri, Dec 21, 2012 at 03:06:15PM +0100, Torben Hohn wrote:
> timer resolution of ~32us is pretty low.
> v2 has 32bits resolution, so we have quite some headroom, and
> can use the 24MHz clock.
> v1 has only 16bits, so we only increase v2.
> 
> So we just exchange the timrot clock in imx28.
> On imx23 we have timrotv1 and everything stays the same.
> 
> Signed-off-by: Torben Hohn <torbenh@linutronix.de>

1) s/ahbx/apbx on patch subject.
2) The patch prefix should look like "ARM: mxs: ..."

Fixed them and applied both patches.  Thanks.

Shawn