Skip to contentSkip to conversation

AT91: SAM9G45 - add a separate clock entry for every single TC block

PATCH New Fabian Godehardt 4 participants 10 messages Sep 3, 2010 – Sep 8, 2010

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 AT91: SAM9G45 - add a separate clock entry for every single TC block

Without this patch you will not be able to register the first block because of the second association call on at91_add_device_tc().

---
 arch/arm/mach-at91/at91sam9g45.c         |   12 +++++++++---
 arch/arm/mach-at91/at91sam9g45_devices.c |    4 ++--
 2 files changed, 11 insertions(+), 5 deletions(-)
arch/arm/mach-at91/at91sam9g45.c +9 -3
index 753c0d3..52ef2d6 100644
@@ -121,8 +121,13 @@ static struct clk ssc1_clk = {
 	.pmc_mask	= 1 << AT91SAM9G45_ID_SSC1,
 	.type		= CLK_TYPE_PERIPHERAL,
 };
-static struct clk tcb_clk = {
-	.name		= "tcb_clk",
+static struct clk tcb0_clk = {
+	.name		= "tcb0_clk",
+	.pmc_mask	= 1 << AT91SAM9G45_ID_TCB,
+	.type		= CLK_TYPE_PERIPHERAL,
+};
+static struct clk tcb1_clk = {
+	.name		= "tcb1_clk",
 	.pmc_mask	= 1 << AT91SAM9G45_ID_TCB,
 	.type		= CLK_TYPE_PERIPHERAL,
 };
@@ -208,7 +213,8 @@ static struct clk *periph_clocks[] __initdata = {
 	&spi1_clk,
 	&ssc0_clk,
 	&ssc1_clk,
-	&tcb_clk,
+	&tcb0_clk,
+	&tcb1_clk,
 	&pwm_clk,
 	&tsc_clk,
 	&dma_clk,
arch/arm/mach-at91/at91sam9g45_devices.c +2 -2
index 809114d..4822019 100644
@@ -835,9 +835,9 @@ static struct platform_device at91sam9g45_tcb1_device = {
 static void __init at91_add_device_tc(void)
 {
 	/* this chip has one clock and irq for all six TC channels */
-	at91_clock_associate("tcb_clk", &at91sam9g45_tcb0_device.dev, "t0_clk");
+	at91_clock_associate("tcb0_clk", &at91sam9g45_tcb0_device.dev, "t0_clk");
 	platform_device_register(&at91sam9g45_tcb0_device);
-	at91_clock_associate("tcb_clk", &at91sam9g45_tcb1_device.dev, "t0_clk");
+	at91_clock_associate("tcb1_clk", &at91sam9g45_tcb1_device.dev, "t0_clk");
 	platform_device_register(&at91sam9g45_tcb1_device);
 }
 #else
Signature · 1 line
1.6.6.1
#1 lore raw

PATCH 1/1 AT91: SAM9G45 - add a separate clock entry for every single TC block

Quoted 4 lines · Fabian Godehardt

Without this patch you will not be able to register the first block because of the second association call on at91_add_device_tc().

duplicate code is not a good idea

this is one of the reason I'm think to drop the current clock assotion code of at91 to use the generic clkdev

here you can also do not specify the device as the clock is not device specific

Best Regards,
J.

#2 lore raw

PATCH 1/1 AT91: SAM9G45 - add a separate clock entry for every single TC block

In reply to Fabian Godehardt

Le 03/09/2010 14:31, Fabian Godehardt :

Quoted 2 lines

Without this patch you will not be able to register the first block because of the second association call on at91_add_device_tc().

Yes, I noticed that.

Quoted 1 line

Tell me if my little modification is ok for you. I will then sign it and send it to Russell patch tracking system: what do you think about it?

Quoted 14 lines
---
 arch/arm/mach-at91/at91sam9g45.c         |   12 +++++++++---
 arch/arm/mach-at91/at91sam9g45_devices.c |    4 ++--
 2 files changed, 11 insertions(+), 5 deletions(-)
arch/arm/mach-at91/at91sam9g45.c +0 -1
index 753c0d3..52ef2d6 100644
@@ -121,8 +121,13 @@ static struct clk ssc1_clk = {
 	.pmc_mask	= 1 << AT91SAM9G45_ID_SSC1,
 	.type		= CLK_TYPE_PERIPHERAL,
 };
-static struct clk tcb_clk = {

arch/arm/mach-at91/at91sam9g45.c static struct clk ssc1_clk = { · line 132

Quoted 10 lines
arch/arm/mach-at91/at91sam9g45.c +7 -1
-	.name		= "tcb_clk",
+static struct clk tcb0_clk = {
+	.name		= "tcb0_clk",
+	.pmc_mask	= 1 << AT91SAM9G45_ID_TCB,
+	.type		= CLK_TYPE_PERIPHERAL,
+};
+static struct clk tcb1_clk = {
+	.name		= "tcb1_clk",
 	.pmc_mask	= 1 << AT91SAM9G45_ID_TCB,
 	.type		= CLK_TYPE_PERIPHERAL,

I prefer to add a "fake" clock that depend on the first one instead of doubling the definition: like this:

/* One additional fake clock for second TC block */ static struct clk tcb1_clk = {

        .name           = "tcb1_clk",
        .pmc_mask       = 0,
        .type           = CLK_TYPE_PERIPHERAL,
        .parent         = &tcb0_clk,

};

This way we keep the single clock entry but we provide a fake entry for registering the second TC block.

Quoted 17 lines
arch/arm/mach-at91/at91sam9g45.c +2 -1
 };
@@ -208,7 +213,8 @@ static struct clk *periph_clocks[] __initdata = {
 	&spi1_clk,
 	&ssc0_clk,
 	&ssc1_clk,
-	&tcb_clk,
+	&tcb0_clk,
+	&tcb1_clk,
 	&pwm_clk,
 	&tsc_clk,
 	&dma_clk,
arch/arm/mach-at91/at91sam9g45_devices.c +0 -0
index 809114d..4822019 100644
@@ -835,9 +835,9 @@ static struct platform_device at91sam9g45_tcb1_device = {
 static void __init at91_add_device_tc(void)

arch/arm/mach-at91/at91sam9g45_devices.c static struct platform_device at91sam9g45_tcb1_device = { · line 843

Quoted 10 lines
arch/arm/mach-at91/at91sam9g45_devices.c +2 -2
 {
 	/* this chip has one clock and irq for all six TC channels */
-	at91_clock_associate("tcb_clk", &at91sam9g45_tcb0_device.dev, "t0_clk");
+	at91_clock_associate("tcb0_clk", &at91sam9g45_tcb0_device.dev, "t0_clk");
 	platform_device_register(&at91sam9g45_tcb0_device);
-	at91_clock_associate("tcb_clk", &at91sam9g45_tcb1_device.dev, "t0_clk");
+	at91_clock_associate("tcb1_clk", &at91sam9g45_tcb1_device.dev, "t0_clk");
 	platform_device_register(&at91sam9g45_tcb1_device);
 }
 #else

Ok for other modifications,

Best regards,

Signature · 1 line
Nicolas Ferre
#3 lore raw

PATCH 1/1 AT91: SAM9G45 - add a separate clock entry for every single TC block

Quoted 39 lines · Nicolas Ferre

Le 03/09/2010 14:31, Fabian Godehardt :

Quoted 2 lines

Without this patch you will not be able to register the first block because of the second association call on at91_add_device_tc().

Yes, I noticed that.

Quoted 1 line

Tell me if my little modification is ok for you. I will then sign it and send it to Russell patch tracking system: what do you think about it?

Quoted 14 lines
---
 arch/arm/mach-at91/at91sam9g45.c         |   12 +++++++++---
 arch/arm/mach-at91/at91sam9g45_devices.c |    4 ++--
 2 files changed, 11 insertions(+), 5 deletions(-)
arch/arm/mach-at91/at91sam9g45.c +0 -1
index 753c0d3..52ef2d6 100644
@@ -121,8 +121,13 @@ static struct clk ssc1_clk = {
 	.pmc_mask	= 1 << AT91SAM9G45_ID_SSC1,
 	.type		= CLK_TYPE_PERIPHERAL,
 };
-static struct clk tcb_clk = {

arch/arm/mach-at91/at91sam9g45.c static struct clk ssc1_clk = { · line 132

Quoted 10 lines
arch/arm/mach-at91/at91sam9g45.c +7 -1
-	.name		= "tcb_clk",
+static struct clk tcb0_clk = {
+	.name		= "tcb0_clk",
+	.pmc_mask	= 1 << AT91SAM9G45_ID_TCB,
+	.type		= CLK_TYPE_PERIPHERAL,
+};
+static struct clk tcb1_clk = {
+	.name		= "tcb1_clk",
 	.pmc_mask	= 1 << AT91SAM9G45_ID_TCB,
 	.type		= CLK_TYPE_PERIPHERAL,

I prefer to add a "fake" clock that depend on the first one instead of doubling the definition: like this:

Quoted 8 lines

/* One additional fake clock for second TC block */ static struct clk tcb1_clk = {

        .name           = "tcb1_clk",
        .pmc_mask       = 0,
        .type           = CLK_TYPE_PERIPHERAL,
        .parent         = &tcb0_clk,

};

this modification will also work when we enable and disable the clock the fact of duplicating the clock will result that if we disable the clock we will disable it for both device

in the Nico implementation the cloc will be disable only when no-one use it

it's not the best solution but for now on it's before switching to the clkdev

Best Regards,
J.

#4 lore raw

PATCH 1/1 AT91: SAM9G45 - add a separate clock entry for every single TC block

In reply to Nicolas Ferre

Hi,

Am Freitag 03 September 2010 14:05:43 schrieben Sie:

Quoted 41 lines

Le 03/09/2010 14:31, Fabian Godehardt :

Quoted 2 lines

Without this patch you will not be able to register the first block because of the second association call on at91_add_device_tc().

Yes, I noticed that.

Quoted 1 line

Tell me if my little modification is ok for you. I will then sign it and send it to Russell patch tracking system: what do you think about it?

Quoted 14 lines
---
 arch/arm/mach-at91/at91sam9g45.c         |   12 +++++++++---
 arch/arm/mach-at91/at91sam9g45_devices.c |    4 ++--
 2 files changed, 11 insertions(+), 5 deletions(-)
(quoted diff) +0 -0

b/arch/arm/mach-at91/at91sam9g45.c index 753c0d3..52ef2d6 100644

arch/arm/mach-at91/at91sam9g45.c +0 -1
@@ -121,8 +121,13 @@ static struct clk ssc1_clk = {
 	.pmc_mask	= 1 << AT91SAM9G45_ID_SSC1,
 	.type		= CLK_TYPE_PERIPHERAL,
 };
-static struct clk tcb_clk = {

arch/arm/mach-at91/at91sam9g45.c static struct clk ssc1_clk = { · line 132

Quoted 10 lines
arch/arm/mach-at91/at91sam9g45.c +7 -1
-	.name		= "tcb_clk",
+static struct clk tcb0_clk = {
+	.name		= "tcb0_clk",
+	.pmc_mask	= 1 << AT91SAM9G45_ID_TCB,
+	.type		= CLK_TYPE_PERIPHERAL,
+};
+static struct clk tcb1_clk = {
+	.name		= "tcb1_clk",
 	.pmc_mask	= 1 << AT91SAM9G45_ID_TCB,
 	.type		= CLK_TYPE_PERIPHERAL,

I prefer to add a "fake" clock that depend on the first one instead of doubling the definition: like this:

/* One additional fake clock for second TC block */ static struct clk tcb1_clk = {

Quoted 8 lines
        .name           = "tcb1_clk",
        .pmc_mask       = 0,
        .type           = CLK_TYPE_PERIPHERAL,
        .parent         = &tcb0_clk,

};

This way we keep the single clock entry but we provide a fake entry for registering the second TC block.

It's ok and works for me - thanks.

Best regards,
Fabian

#5 lore raw

PATCH 1/1 AT91: SAM9G45 - add a separate clock entry for every single TC block

From: Fabian Godehardt <[email protected]>

Without this patch you will not be able to register the first block because of the second association call on at91_add_device_tc().

---
 arch/arm/mach-at91/at91sam9g45.c         |   15 ++++++++++++---
 arch/arm/mach-at91/at91sam9g45_devices.c |    4 ++--
 2 files changed, 14 insertions(+), 5 deletions(-)
arch/arm/mach-at91/at91sam9g45.c +12 -3
index 753c0d3..c67b47f 100644
@@ -121,8 +121,8 @@ static struct clk ssc1_clk = {
 	.pmc_mask	= 1 << AT91SAM9G45_ID_SSC1,
 	.type		= CLK_TYPE_PERIPHERAL,
 };
-static struct clk tcb_clk = {
-	.name		= "tcb_clk",
+static struct clk tcb0_clk = {
+	.name		= "tcb0_clk",
 	.pmc_mask	= 1 << AT91SAM9G45_ID_TCB,
 	.type		= CLK_TYPE_PERIPHERAL,
 };
@@ -192,6 +192,14 @@ static struct clk ohci_clk = {
 	.parent		= &uhphs_clk,
 };
 
+/* One additional fake clock for second TC block */
+static struct clk tcb1_clk = {
+	.name		= "tcb1_clk",
+	.pmc_mask	= 0,
+	.type		= CLK_TYPE_PERIPHERAL,
+	.parent		= &tcb0_clk,
+};
+
 static struct clk *periph_clocks[] __initdata = {
 	&pioA_clk,
 	&pioB_clk,
@@ -208,7 +216,7 @@ static struct clk *periph_clocks[] __initdata = {
 	&spi1_clk,
 	&ssc0_clk,
 	&ssc1_clk,
-	&tcb_clk,
+	&tcb0_clk,
 	&pwm_clk,
 	&tsc_clk,
 	&dma_clk,
@@ -221,6 +229,7 @@ static struct clk *periph_clocks[] __initdata = {
 	&mmc1_clk,
 	// irq0
 	&ohci_clk,
+	&tcb1_clk,
 };
 
 /*
arch/arm/mach-at91/at91sam9g45_devices.c +2 -2
index 809114d..4822019 100644
@@ -835,9 +835,9 @@ static struct platform_device at91sam9g45_tcb1_device = {
 static void __init at91_add_device_tc(void)
 {
 	/* this chip has one clock and irq for all six TC channels */
-	at91_clock_associate("tcb_clk", &at91sam9g45_tcb0_device.dev, "t0_clk");
+	at91_clock_associate("tcb0_clk", &at91sam9g45_tcb0_device.dev, "t0_clk");
 	platform_device_register(&at91sam9g45_tcb0_device);
-	at91_clock_associate("tcb_clk", &at91sam9g45_tcb1_device.dev, "t0_clk");
+	at91_clock_associate("tcb1_clk", &at91sam9g45_tcb1_device.dev, "t0_clk");
 	platform_device_register(&at91sam9g45_tcb1_device);
 }
 #else
Signature · 1 line
1.5.6.5
#7 lore raw
In reply to Nicolas Ferre

hi,

arch/arm/mach-at91/at91sam9g45.c static struct clk ohci_clk = { · line 202

Quoted 8 lines
arch/arm/mach-at91/at91sam9g45.c +8 -0
+/* One additional fake clock for second TC block */
+static struct clk tcb1_clk = {
+ ? ? ? .name ? ? ? ? ? = "tcb1_clk",
+ ? ? ? .pmc_mask ? ? ? = 0,
+ ? ? ? .type ? ? ? ? ? = CLK_TYPE_PERIPHERAL,
+ ? ? ? .parent ? ? ? ? = &tcb0_clk,
+};
+

Looking at this again... since type is CLK_TYPE_PERIPHERAL, when you call clk_register() the "parent" is changed to the master clock.

Which means, then later you call clk_enable() the "pmc_mask" is still 0, so 0 gets written (in pmc_periph_mode) to AT91_PMC_PCER. So the TCB clock won't be enabled.

Or am I missing something?

Regards,
Andrew Victor

#8 lore raw

Le 07/09/2010 21:42, avictor.za at gmail.com :

Quoted 14 lines

hi,

Review

Quoted 8 lines
(quoted diff) +8 -0
+/* One additional fake clock for second TC block */
+static struct clk tcb1_clk = {
+       .name           = "tcb1_clk",
+       .pmc_mask       = 0,
+       .type           = CLK_TYPE_PERIPHERAL,
+       .parent         = &tcb0_clk,
+};
+

Looking at this again... since type is CLK_TYPE_PERIPHERAL, when you call clk_register() the "parent" is changed to the master clock.

Quoted 5 lines

Which means, then later you call clk_enable() the "pmc_mask" is still 0, so 0 gets written (in pmc_periph_mode) to AT91_PMC_PCER. So the TCB clock won't be enabled.

Or am I missing something?

You are absolutely right!

What do you think about this modification of clk_register() function?

arch/arm/mach-at91/clock.c +2 -1
@@ -501,7 +501,8 @@ postcore_initcall(at91_clk_debugfs_init);
 int __init clk_register(struct clk *clk)
 {
        if (clk_is_peripheral(clk)) {
-               clk->parent = &mck;
+               if (!clk->parent)
+                       clk->parent = &mck;
                clk->mode = pmc_periph_mode;
                list_add_tail(&clk->node, &clocks);
        }

It is a very little modification which implements what I had in mind while creating a kind of "child peripheral" clock.

Best regards,

Signature · 1 line
Nicolas Ferre
#9 lore raw
Quoted 38 lines · Nicolas Ferre

Le 07/09/2010 21:42, avictor.za at gmail.com :

Quoted 14 lines

hi,

Review

Quoted 8 lines
(quoted diff) +8 -0
+/* One additional fake clock for second TC block */
+static struct clk tcb1_clk = {
+       .name           = "tcb1_clk",
+       .pmc_mask       = 0,
+       .type           = CLK_TYPE_PERIPHERAL,
+       .parent         = &tcb0_clk,
+};
+

Looking at this again... since type is CLK_TYPE_PERIPHERAL, when you call clk_register() the "parent" is changed to the master clock.

Quoted 5 lines

Which means, then later you call clk_enable() the "pmc_mask" is still 0, so 0 gets written (in pmc_periph_mode) to AT91_PMC_PCER. So the TCB clock won't be enabled.

Or am I missing something?

You are absolutely right!

What do you think about this modification of clk_register() function?

arch/arm/mach-at91/clock.c +2 -1
@@ -501,7 +501,8 @@ postcore_initcall(at91_clk_debugfs_init);
 int __init clk_register(struct clk *clk)
 {
        if (clk_is_peripheral(clk)) {
-               clk->parent = &mck;
+               if (!clk->parent)
+                       clk->parent = &mck;
                clk->mode = pmc_periph_mode;
                list_add_tail(&clk->node, &clocks);
        }
Quoted 2 lines

It is a very little modification which implements what I had in mind while creating a kind of "child peripheral" clock.

so so

but until we switch to clkdev it will solve the issue

Best Regards,
J.