Skip to contentSkip to conversation

ARM: shmobile: r8a7779: Correct TMU clock support again

PATCH New Simon Horman 5 participants 17 messages Feb 14, 2013 – Feb 18, 2013

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

This partially reverts 58079fa7d54a0929d304054ee759187a2ccd3cdf (ARM: shmobile: r8a7779: Correct TMU clock support) and fixes a regression introduced by that patch.

Although the documentation I have indicates that the patch above is correct it appears that the change causes the Marzen board to fail to boot as follows.

...
hub 4-0:1.0: USB hub found
hub 4-0:1.0: 2 ports detected

INFO: rcu_sched self-detected stall on CPUINFO: rcu_sched detected stalls on CPUs/tasks: { 1} (detected by 2, t=279640 jiffies, g=4294967052, c=4294967051, q=38)

Task dump for CPU 1:
swapper/0 R running 0 1 0 0x00000002
[<c02b8f5c>] (__schedule+0x1b0/0x4c0) from [<c013c590>] (__loop_delay+0x4/0xc)

{ 1} (t=279640 jiffies g=4294967052 c=4294967052 q=37)
[<c000ef9c>] (unwind_backtrace+0x0/0xf8) from [<c0068488>] (rcu_check_callbacks+0x218/0x6b8) [<c0068488>] (rcu_check_callbacks+0x218/0x6b8) from [<c0026774>] (update_process_times+0x38/0x4c) [<c0026774>] (update_process_times+0x38/0x4c) from [<c00569e0>] (tick_nohz_handler+0xb4/0x11c) [<c00569e0>] (tick_nohz_handler+0xb4/0x11c) from [<c000e518>] (twd_handler+0x34/0x44) [<c000e518>] (twd_handler+0x34/0x44) from [<c0063484>] (handle_percpu_devid_irq+0x68/0x80) [<c0063484>] (handle_percpu_devid_irq+0x68/0x80) from [<c005febc>] (generic_handle_irq+0x20/0x30) [<c005febc>] (generic_handle_irq+0x20/0x30) from [<c000a5ec>] (handle_IRQ+0x40/0x90) [<c000a5ec>] (handle_IRQ+0x40/0x90) from [<c000934c>] (gic_handle_irq+0x2c/0x5c) [<c000934c>] (gic_handle_irq+0x2c/0x5c) from [<c0009a40>] (__irq_svc+0x40/0x50)

Exception stack(0xef03ddf8 to 0xef03de40)

dde0: 000001c1 ffffffff
de00: 000001d8 01bf01bf ef35ec40 ef35e800 ef35ec6c 0000002b ef35ec68 c013c560
de20: c0392994 60000113 00000000 ef03de40 c01a5d40 c013c590 20000113 ffffffff [<c0009a40>] (__irq_svc+0x40/0x50) from [<c013c590>] (__loop_delay+0x4/0xc)

Cc Denis Oliver Kropp <[email protected]>
Cc Magnus Damm <[email protected]>
---
 arch/arm/mach-shmobile/clock-r8a7779.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/arm/mach-shmobile/clock-r8a7779.c +1 -1
index 202370d..83369c9 100644
@@ -161,7 +161,7 @@ static struct clk_lookup lookups[] = {
 	CLKDEV_DEV_ID("ehci-platform.0", &mstp_clks[MSTP100]), /* USB EHCI port0/1 */
 	CLKDEV_DEV_ID("ohci-platform.0", &mstp_clks[MSTP100]), /* USB OHCI port0/1 */
 	CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
-	CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+	CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
 	CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */
 	CLKDEV_DEV_ID("i2c-rcar.0", &mstp_clks[MSTP030]), /* I2C0 */
 	CLKDEV_DEV_ID("i2c-rcar.1", &mstp_clks[MSTP029]), /* I2C1 */
Signature · 1 line
1.7.10.4
#1 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Hi Simon

Quoted 7 lines

This partially reverts 58079fa7d54a0929d304054ee759187a2ccd3cdf (ARM: shmobile: r8a7779: Correct TMU clock support) and fixes a regression introduced by that patch.

Although the documentation I have indicates that the patch above is correct it appears that the change causes the Marzen board to fail to boot as follows.

(snip)

arch/arm/mach-shmobile/clock-r8a7779.c static struct clk_lookup lookups[] = { · line 167

Quoted 11 lines
arch/arm/mach-shmobile/clock-r8a7779.c +1 -1
@@ -161,7 +161,7 @@ static struct clk_lookup lookups[] = {
 	CLKDEV_DEV_ID("ehci-platform.0", &mstp_clks[MSTP100]), /* USB EHCI port0/1 */
 	CLKDEV_DEV_ID("ohci-platform.0", &mstp_clks[MSTP100]), /* USB OHCI port0/1 */
 	CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
-	CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+	CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
 	CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */
 	CLKDEV_DEV_ID("i2c-rcar.0", &mstp_clks[MSTP030]), /* I2C0 */
 	CLKDEV_DEV_ID("i2c-rcar.1", &mstp_clks[MSTP029]), /* I2C1 */

Really ???
Is MSTP value of TMU01 same as TMU00 ?

Please wail, I ask it to hardware guys.

Best regards
---
Kuninori Morimoto

maintainer
#2 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Quoted 11 lines · Kuninori Morimoto

Hi Simon

Quoted 7 lines

This partially reverts 58079fa7d54a0929d304054ee759187a2ccd3cdf (ARM: shmobile: r8a7779: Correct TMU clock support) and fixes a regression introduced by that patch.

Although the documentation I have indicates that the patch above is correct it appears that the change causes the Marzen board to fail to boot as follows.

(snip)

Quoted 14 lines

arch/arm/mach-shmobile/clock-r8a7779.c static struct clk_lookup lookups[] = { · line 167

Quoted 11 lines
arch/arm/mach-shmobile/clock-r8a7779.c +1 -1
@@ -161,7 +161,7 @@ static struct clk_lookup lookups[] = {
      CLKDEV_DEV_ID("ehci-platform.0", &mstp_clks[MSTP100]), /* USB EHCI port0/1 */
      CLKDEV_DEV_ID("ohci-platform.0", &mstp_clks[MSTP100]), /* USB OHCI port0/1 */
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */
      CLKDEV_DEV_ID("i2c-rcar.0", &mstp_clks[MSTP030]), /* I2C0 */
      CLKDEV_DEV_ID("i2c-rcar.1", &mstp_clks[MSTP029]), /* I2C1 */

Really ???
Is MSTP value of TMU01 same as TMU00 ?

Usually, the TMU channels are bundled together. So TMU00 may be for channel 0->3 and TMU01 for 4->6.

Thanks,

/ magnus

#3 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Hi Magnus, Simon

Thank you for your explain

Quoted 15 lines
Quoted 6 lines
Quoted 6 lines
arch/arm/mach-shmobile/clock-r8a7779.c +0 -0
@@ -161,7 +161,7 @@ static struct clk_lookup lookups[] = {
      CLKDEV_DEV_ID("ehci-platform.0", &mstp_clks[MSTP100]), /* USB EHCI port0/1 */
      CLKDEV_DEV_ID("ohci-platform.0", &mstp_clks[MSTP100]), /* USB OHCI port0/1 */
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
Quoted 8 lines

arch/arm/mach-shmobile/clock-r8a7779.c static struct clk_lookup lookups[] = { · line 167

Quoted 5 lines
arch/arm/mach-shmobile/clock-r8a7779.c +1 -1
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */
      CLKDEV_DEV_ID("i2c-rcar.0", &mstp_clks[MSTP030]), /* I2C0 */
      CLKDEV_DEV_ID("i2c-rcar.1", &mstp_clks[MSTP029]), /* I2C1 */

Really ???
Is MSTP value of TMU01 same as TMU00 ?

Quoted 2 lines

Usually, the TMU channels are bundled together. So TMU00 may be for channel 0->3 and TMU01 for 4->6.

This means that current TMU02 numbering seems doubtful too ? How about just rever 58079fa7d54a0929d304054ee759187a2ccd3cdf ?

Best regards
---
Kuninori Morimoto

#4 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Hi Magnus, Simon again

Quoted 5 lines
Quoted 2 lines

Usually, the TMU channels are bundled together. So TMU00 may be for channel 0->3 and TMU01 for 4->6.

This means that current TMU02 numbering seems doubtful too ? How about just rever 58079fa7d54a0929d304054ee759187a2ccd3cdf ?

I meant there is no tmu2 code on setup.c

Best regards
---
Kuninori Morimoto

#5 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

In reply to Kuninori Morimoto
Quoted 20 lines · Kuninori Morimoto

Hi Magnus, Simon

Thank you for your explain

Quoted 15 lines
Quoted 14 lines

arch/arm/mach-shmobile/clock-r8a7779.c static struct clk_lookup lookups[] = { · line 167

Quoted 11 lines
arch/arm/mach-shmobile/clock-r8a7779.c +1 -1
@@ -161,7 +161,7 @@ static struct clk_lookup lookups[] = {
      CLKDEV_DEV_ID("ehci-platform.0", &mstp_clks[MSTP100]), /* USB EHCI port0/1 */
      CLKDEV_DEV_ID("ohci-platform.0", &mstp_clks[MSTP100]), /* USB OHCI port0/1 */
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */
      CLKDEV_DEV_ID("i2c-rcar.0", &mstp_clks[MSTP030]), /* I2C0 */
      CLKDEV_DEV_ID("i2c-rcar.1", &mstp_clks[MSTP029]), /* I2C1 */

Really ???
Is MSTP value of TMU01 same as TMU00 ?

Quoted 5 lines
Quoted 2 lines

Usually, the TMU channels are bundled together. So TMU00 may be for channel 0->3 and TMU01 for 4->6.

This means that current TMU02 numbering seems doubtful too ? How about just rever 58079fa7d54a0929d304054ee759187a2ccd3cdf ?

Perhaps that is a good idea.

The original motivation for this patch was to add the TMU02 line. And "fixing" TMU01 was an afterthought. However, I am also now doubtful about the correctness of the TMU02 line and thus the usefulness of the original patch.

#6 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Hi Simon

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
arch/arm/mach-shmobile/clock-r8a7779.c +1 -1
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */

(snip)

Quoted 9 lines
Quoted 2 lines

This means that current TMU02 numbering seems doubtful too ? How about just rever 58079fa7d54a0929d304054ee759187a2ccd3cdf ?

Perhaps that is a good idea.

The original motivation for this patch was to add the TMU02 line. And "fixing" TMU01 was an afterthought. However, I am also now doubtful about the correctness of the TMU02 line and thus the usefulness of the original patch.

I think this comment out is creating confusion ?

/* TMU00 */ -> /* TMU0 channel 0 */
/* TMU01 */ -> /* TMU0 channel 1 */
/* TMU02 */ -> /* TMU0 channel 2 */

Best regards
---
Kuninori Morimoto

#7 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Quoted 18 lines · Kuninori Morimoto

Hi Simon

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
arch/arm/mach-shmobile/clock-r8a7779.c +1 -1
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */

(snip)

Quoted 9 lines
Quoted 2 lines

This means that current TMU02 numbering seems doubtful too ? How about just rever 58079fa7d54a0929d304054ee759187a2ccd3cdf ?

Perhaps that is a good idea.

The original motivation for this patch was to add the TMU02 line. And "fixing" TMU01 was an afterthought. However, I am also now doubtful about the correctness of the TMU02 line and thus the usefulness of the original patch.

Quoted 5 lines

I think this comment out is creating confusion ?

/* TMU00 */ -> /* TMU0 channel 0 */
/* TMU01 */ -> /* TMU0 channel 1 */
/* TMU02 */ -> /* TMU0 channel 2 */

That does make things a little clearer,
but I for one wasn't confused by the existing comments.

#8 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Quoted 19 lines · Simon Horman
Quoted 18 lines · Kuninori Morimoto

Hi Simon

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
(quoted diff) +1 -1
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */

(snip)

Quoted 9 lines
Quoted 2 lines

This means that current TMU02 numbering seems doubtful too ? How about just rever 58079fa7d54a0929d304054ee759187a2ccd3cdf ?

Perhaps that is a good idea.

The original motivation for this patch was to add the TMU02 line. And "fixing" TMU01 was an afterthought. However, I am also now doubtful about the correctness of the TMU02 line and thus the usefulness of the original patch.

Quoted 8 lines
Quoted 5 lines

I think this comment out is creating confusion ?

/* TMU00 */ -> /* TMU0 channel 0 */
/* TMU01 */ -> /* TMU0 channel 1 */
/* TMU02 */ -> /* TMU0 channel 2 */

That does make things a little clearer,
but I for one wasn't confused by the existing comments.

Ok, now I think I see what you are getting at.

MSTP016 is for channel 0..n of TMU0
MSTP015 is for channel 1..n of TMU1
MSTP014 is for channel 2..n of TMU2

And all the TMU values set by CLKDEV_DEV_ID statements above relate to channels of TMU0.

That does seem to be a valid interpretation of the documentation now that I read it again.

maintainer
#9 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Quoted 35 lines · Simon Horman
Quoted 19 lines · Simon Horman
Quoted 18 lines · Kuninori Morimoto

Hi Simon

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
(quoted diff) +1 -1
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */

(snip)

Quoted 9 lines
Quoted 2 lines

This means that current TMU02 numbering seems doubtful too ? How about just rever 58079fa7d54a0929d304054ee759187a2ccd3cdf ?

Perhaps that is a good idea.

The original motivation for this patch was to add the TMU02 line. And "fixing" TMU01 was an afterthought. However, I am also now doubtful about the correctness of the TMU02 line and thus the usefulness of the original patch.

Quoted 8 lines
Quoted 5 lines

I think this comment out is creating confusion ?

/* TMU00 */ -> /* TMU0 channel 0 */
/* TMU01 */ -> /* TMU0 channel 1 */
/* TMU02 */ -> /* TMU0 channel 2 */

That does make things a little clearer,
but I for one wasn't confused by the existing comments.

Ok, now I think I see what you are getting at.

MSTP016 is for channel 0..n of TMU0
MSTP015 is for channel 1..n of TMU1
MSTP014 is for channel 2..n of TMU2

Quoted 5 lines

And all the TMU values set by CLKDEV_DEV_ID statements above relate to channels of TMU0.

That does seem to be a valid interpretation of the documentation now that I read it again.

Hm, your channel assignment above seems a tad different from what I believe:

MSTP016 is for channel 0..n of TMU0
MSTP015 is for channel 0..n of TMU1
MSTP014 is for channel 0..n of TMU2

Actually, you can check how the TMU channels are bundled together based on the I/O memory and interrupts.
The TMU driver configuration in setup-sh73a0 contains these:

static struct sh_timer_config tmu00_platform_data = {
	.name = "TMU00",
	.channel_offset = 0x4,
	.timer_bit = 0,
	.clockevent_rating = 200,
};
static struct resource tmu00_resources[] = {
	[0] = {
		.name	= "TMU00",
		.start	= 0xfff60008,
		.end	= 0xfff60013,
		.flags	= IORESOURCE_MEM,
	},
	[1] = {
		.start	= intcs_evt2irq(0x0e80), /* TMU0_TUNI00 */
		.flags	= IORESOURCE_IRQ,
	},
};

and

static struct sh_timer_config tmu01_platform_data = {
	.name = "TMU01",
	.channel_offset = 0x10,
	.timer_bit = 1,
	.clocksource_rating = 200,
};
static struct resource tmu01_resources[] = {
	[0] = {
		.name	= "TMU01",
		.start	= 0xfff60014,
		.end	= 0xfff6001f,
		.flags	= IORESOURCE_MEM,
	},
	[1] = {
		.start	= intcs_evt2irq(0x0ea0), /* TMU0_TUNI01 */
		.flags	= IORESOURCE_IRQ,
	},
};

Note how the IORESOURCE_MEM ranges for tmu00 and tmu01 are closely located. Also check the channel_offset platform data that point out the shared start-stop register. The timer bit is the per-channel start-stop bit. Another hint is in the IORESOURCE_IRQ and the TMU0_TUNI00 + TMU0_TUNI01 comments that both start with TMU0_.

So basically, in this case both tmu00 and tmu01 should be assigned to MSTP016 aka TMU00.

This line must be correct from my point of view:

CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU00 */

Cheers,

/ magnus

#10 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Quoted 39 lines · Magnus Damm
Quoted 38 lines · Simon Horman
Quoted 19 lines · Simon Horman
Quoted 18 lines · Kuninori Morimoto

Hi Simon

Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
(quoted diff) +1 -1
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */

(snip)

Quoted 9 lines
Quoted 2 lines

This means that current TMU02 numbering seems doubtful too ? How about just rever 58079fa7d54a0929d304054ee759187a2ccd3cdf ?

Perhaps that is a good idea.

The original motivation for this patch was to add the TMU02 line. And "fixing" TMU01 was an afterthought. However, I am also now doubtful about the correctness of the TMU02 line and thus the usefulness of the original patch.

Quoted 8 lines
Quoted 5 lines

I think this comment out is creating confusion ?

/* TMU00 */ -> /* TMU0 channel 0 */
/* TMU01 */ -> /* TMU0 channel 1 */
/* TMU02 */ -> /* TMU0 channel 2 */

That does make things a little clearer,
but I for one wasn't confused by the existing comments.

Ok, now I think I see what you are getting at.

MSTP016 is for channel 0..n of TMU0
MSTP015 is for channel 1..n of TMU1
MSTP014 is for channel 2..n of TMU2

And all the TMU values set by CLKDEV_DEV_ID statements above relate to channels of TMU0.

Quoted 8 lines
Quoted 2 lines

That does seem to be a valid interpretation of the documentation now that I read it again.

Hm, your channel assignment above seems a tad different from what I believe:

MSTP016 is for channel 0..n of TMU0
MSTP015 is for channel 0..n of TMU1
MSTP014 is for channel 0..n of TMU2

Yes, thats is what I wanted to say too.

Quoted 52 lines

Actually, you can check how the TMU channels are bundled together based on the I/O memory and interrupts.
The TMU driver configuration in setup-sh73a0 contains these:

static struct sh_timer_config tmu00_platform_data = {
	.name = "TMU00",
	.channel_offset = 0x4,
	.timer_bit = 0,
	.clockevent_rating = 200,
};
static struct resource tmu00_resources[] = {
	[0] = {
		.name	= "TMU00",
		.start	= 0xfff60008,
		.end	= 0xfff60013,
		.flags	= IORESOURCE_MEM,
	},
	[1] = {
		.start	= intcs_evt2irq(0x0e80), /* TMU0_TUNI00 */
		.flags	= IORESOURCE_IRQ,
	},
};

and

static struct sh_timer_config tmu01_platform_data = {
	.name = "TMU01",
	.channel_offset = 0x10,
	.timer_bit = 1,
	.clocksource_rating = 200,
};
static struct resource tmu01_resources[] = {
	[0] = {
		.name	= "TMU01",
		.start	= 0xfff60014,
		.end	= 0xfff6001f,
		.flags	= IORESOURCE_MEM,
	},
	[1] = {
		.start	= intcs_evt2irq(0x0ea0), /* TMU0_TUNI01 */
		.flags	= IORESOURCE_IRQ,
	},
};

Note how the IORESOURCE_MEM ranges for tmu00 and tmu01 are closely located. Also check the channel_offset platform data that point out the shared start-stop register. The timer bit is the per-channel start-stop bit. Another hint is in the IORESOURCE_IRQ and the TMU0_TUNI00 + TMU0_TUNI01 comments that both start with TMU0_.

Quoted 6 lines

So basically, in this case both tmu00 and tmu01 should be assigned to MSTP016 aka TMU00.

This line must be correct from my point of view:

CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU00 */

That seems sane. But we are still left guessing about sh_tmu.2. Can we assume it is channel 2 or TMU0 ?

#11 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

In reply to Magnus Damm
Quoted 13 lines · Magnus Damm
Quoted 11 lines · Kuninori Morimoto

Hi Simon

Quoted 7 lines

This partially reverts 58079fa7d54a0929d304054ee759187a2ccd3cdf (ARM: shmobile: r8a7779: Correct TMU clock support) and fixes a regression introduced by that patch.

Although the documentation I have indicates that the patch above is correct it appears that the change causes the Marzen board to fail to boot as follows.

(snip)

Quoted 13 lines
Quoted 5 lines
Quoted 5 lines
arch/arm/mach-shmobile/clock-r8a7779.c +0 -0
@@ -161,7 +161,7 @@ static struct clk_lookup lookups[] = {
      CLKDEV_DEV_ID("ehci-platform.0", &mstp_clks[MSTP100]), /* USB EHCI port0/1 */
      CLKDEV_DEV_ID("ohci-platform.0", &mstp_clks[MSTP100]), /* USB OHCI port0/1 */
Quoted 8 lines

arch/arm/mach-shmobile/clock-r8a7779.c static struct clk_lookup lookups[] = { · line 167

Quoted 6 lines
arch/arm/mach-shmobile/clock-r8a7779.c +1 -1
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */
      CLKDEV_DEV_ID("i2c-rcar.0", &mstp_clks[MSTP030]), /* I2C0 */
      CLKDEV_DEV_ID("i2c-rcar.1", &mstp_clks[MSTP029]), /* I2C1 */

Really ???

I am surprised too!

Quoted 1 line
Quoted 1 line

Is MSTP value of TMU01 same as TMU00 ?

With the patch above the board appears to boot.
Without the patch, it doesn't.

Quoted 2 lines

Usually, the TMU channels are bundled together. So TMU00 may be for channel 0->3 and TMU01 for 4->6.

Ok, so in that case the correct value for "sh_tmu.2" might actually be MSTP016 or MSTP015 depending on the bundle size. I am unsure how to test sh_tmu.2.

However, the documentation I have seems to talk about channels being TMU0..TMU2 and control of them being via MSTP016..MSTP014. I.e. what the code currently is but not what seems to work for TMU1 in practice.

#12 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Quoted 38 lines · Simon Horman
Quoted 13 lines · Magnus Damm
Quoted 11 lines · Kuninori Morimoto

Hi Simon

Quoted 7 lines

This partially reverts 58079fa7d54a0929d304054ee759187a2ccd3cdf (ARM: shmobile: r8a7779: Correct TMU clock support) and fixes a regression introduced by that patch.

Although the documentation I have indicates that the patch above is correct it appears that the change causes the Marzen board to fail to boot as follows.

(snip)

Quoted 13 lines
Quoted 5 lines
Quoted 5 lines
arch/arm/mach-shmobile/clock-r8a7779.c +0 -0
@@ -161,7 +161,7 @@ static struct clk_lookup lookups[] = {
      CLKDEV_DEV_ID("ehci-platform.0", &mstp_clks[MSTP100]), /* USB EHCI port0/1 */
      CLKDEV_DEV_ID("ohci-platform.0", &mstp_clks[MSTP100]), /* USB OHCI port0/1 */
Quoted 8 lines

arch/arm/mach-shmobile/clock-r8a7779.c static struct clk_lookup lookups[] = { · line 167

Quoted 6 lines
arch/arm/mach-shmobile/clock-r8a7779.c +1 -1
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */
      CLKDEV_DEV_ID("i2c-rcar.0", &mstp_clks[MSTP030]), /* I2C0 */
      CLKDEV_DEV_ID("i2c-rcar.1", &mstp_clks[MSTP029]), /* I2C1 */

Really ???

I am surprised too!

Quoted 1 line
Quoted 1 line

Is MSTP value of TMU01 same as TMU00 ?

With the patch above the board appears to boot.
Without the patch, it doesn't.

Quoted 2 lines

Usually, the TMU channels are bundled together. So TMU00 may be for channel 0->3 and TMU01 for 4->6.

Quoted 8 lines

Ok, so in that case the correct value for "sh_tmu.2" might actually be MSTP016 or MSTP015 depending on the bundle size. I am unsure how to test sh_tmu.2.

However, the documentation I have seems to talk about channels being TMU0..TMU2 and control of them being via MSTP016..MSTP014. I.e. what the code currently is but not what seems to work for TMU1 in practice.

I think you are getting caught up on naming semantics. The MSTP bit controls a block, which in turn usually has multiple TMU channels associated with it, so it's quite reasonable to register the same bit multiple times for each channel it impacts (that's what refcounting is for, after all).

The TMU driver only handles individual channels, so sh_tmu.0/1/2 are just channel numbers, you don't have any insight in to which block they are a subset of or how any of their MSTP bits relate to each other, nor should you, as this gets handled by the clk API.

Consider the following example for SH7786:

        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.0", &mstp_clks[MSTP008]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.1", &mstp_clks[MSTP008]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.2", &mstp_clks[MSTP008]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.3", &mstp_clks[MSTP009]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.4", &mstp_clks[MSTP009]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.5", &mstp_clks[MSTP009]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.6", &mstp_clks[MSTP010]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.7", &mstp_clks[MSTP010]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.8", &mstp_clks[MSTP010]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.9", &mstp_clks[MSTP011]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.10", &mstp_clks[MSTP011]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.11", &mstp_clks[MSTP011]),

registers a tmu_fck lookup that maps back appropriately for each channel, regardless of MSTP overlap. In practice these channels are split between 4 TMU blocks, but the driver doesn't know or care.

#13 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Quoted 74 lines · Paul Mundt
Quoted 38 lines · Simon Horman
Quoted 13 lines · Magnus Damm
Quoted 11 lines · Kuninori Morimoto

Hi Simon

Quoted 7 lines

This partially reverts 58079fa7d54a0929d304054ee759187a2ccd3cdf (ARM: shmobile: r8a7779: Correct TMU clock support) and fixes a regression introduced by that patch.

Although the documentation I have indicates that the patch above is correct it appears that the change causes the Marzen board to fail to boot as follows.

(snip)

Quoted 13 lines
Quoted 5 lines
Quoted 5 lines
arch/arm/mach-shmobile/clock-r8a7779.c +0 -0
@@ -161,7 +161,7 @@ static struct clk_lookup lookups[] = {
      CLKDEV_DEV_ID("ehci-platform.0", &mstp_clks[MSTP100]), /* USB EHCI port0/1 */
      CLKDEV_DEV_ID("ohci-platform.0", &mstp_clks[MSTP100]), /* USB OHCI port0/1 */
Quoted 8 lines

arch/arm/mach-shmobile/clock-r8a7779.c static struct clk_lookup lookups[] = { · line 167

Quoted 6 lines
arch/arm/mach-shmobile/clock-r8a7779.c +1 -1
      CLKDEV_DEV_ID("sh_tmu.0", &mstp_clks[MSTP016]), /* TMU00 */
-     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP015]), /* TMU01 */
+     CLKDEV_DEV_ID("sh_tmu.1", &mstp_clks[MSTP016]), /* TMU01 */
      CLKDEV_DEV_ID("sh_tmu.2", &mstp_clks[MSTP014]), /* TMU02 */
      CLKDEV_DEV_ID("i2c-rcar.0", &mstp_clks[MSTP030]), /* I2C0 */
      CLKDEV_DEV_ID("i2c-rcar.1", &mstp_clks[MSTP029]), /* I2C1 */

Really ???

I am surprised too!

Quoted 1 line
Quoted 1 line

Is MSTP value of TMU01 same as TMU00 ?

With the patch above the board appears to boot.
Without the patch, it doesn't.

Quoted 2 lines

Usually, the TMU channels are bundled together. So TMU00 may be for channel 0->3 and TMU01 for 4->6.

Quoted 8 lines

Ok, so in that case the correct value for "sh_tmu.2" might actually be MSTP016 or MSTP015 depending on the bundle size. I am unsure how to test sh_tmu.2.

However, the documentation I have seems to talk about channels being TMU0..TMU2 and control of them being via MSTP016..MSTP014. I.e. what the code currently is but not what seems to work for TMU1 in practice.

I think you are getting caught up on naming semantics. The MSTP bit controls a block, which in turn usually has multiple TMU channels associated with it, so it's quite reasonable to register the same bit multiple times for each channel it impacts (that's what refcounting is for, after all).

The TMU driver only handles individual channels, so sh_tmu.0/1/2 are just channel numbers, you don't have any insight in to which block they are a subset of or how any of their MSTP bits relate to each other, nor should you, as this gets handled by the clk API.

Consider the following example for SH7786:

        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.0", &mstp_clks[MSTP008]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.1", &mstp_clks[MSTP008]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.2", &mstp_clks[MSTP008]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.3", &mstp_clks[MSTP009]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.4", &mstp_clks[MSTP009]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.5", &mstp_clks[MSTP009]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.6", &mstp_clks[MSTP010]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.7", &mstp_clks[MSTP010]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.8", &mstp_clks[MSTP010]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.9", &mstp_clks[MSTP011]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.10", &mstp_clks[MSTP011]),
        CLKDEV_ICK_ID("tmu_fck", "sh_tmu.11", &mstp_clks[MSTP011]),
Quoted 3 lines

registers a tmu_fck lookup that maps back appropriately for each channel, regardless of MSTP overlap. In practice these channels are split between 4 TMU blocks, but the driver doesn't know or care.

Right. But the above does know which MSTP000X to use for each sh_tmu.X. Getting that information correct for sh_tmu.2 on the r8a7779 problem I'm not completely sure of the correct answer to. Though I now strongly suspect it is MSTP016.

#14 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Quoted 5 lines · Simon Horman
Quoted 4 lines · Paul Mundt
Quoted 3 lines · Simon Horman

Ok, so in that case the correct value for "sh_tmu.2" might actually be MSTP016 or MSTP015 depending on the bundle size. I am unsure how to test sh_tmu.2.

[snip]

Quoted 3 lines

Getting that information correct for sh_tmu.2 on the r8a7779 problem I'm not completely sure of the correct answer to. Though I now strongly suspect it is MSTP016.

Testing shouldn't be that difficult, just switch your clocksource to it and make sure all TMU-related MSTP bits are first cleared?

#15 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Quoted 7 lines · Paul Mundt
Quoted 5 lines · Simon Horman
Quoted 4 lines · Paul Mundt
Quoted 3 lines · Simon Horman

Ok, so in that case the correct value for "sh_tmu.2" might actually be MSTP016 or MSTP015 depending on the bundle size. I am unsure how to test sh_tmu.2.

Quoted 8 lines

[snip]

Quoted 3 lines

Getting that information correct for sh_tmu.2 on the r8a7779 problem I'm not completely sure of the correct answer to. Though I now strongly suspect it is MSTP016.

Testing shouldn't be that difficult, just switch your clocksource to it and make sure all TMU-related MSTP bits are first cleared?

Thanks, I will try that.

#16 lore raw

PATCH 1/1 ARM: shmobile: r8a7779: Correct TMU clock support again

Quoted 10 lines · Simon Horman
Quoted 9 lines · Paul Mundt
Quoted 5 lines · Simon Horman
Quoted 4 lines · Paul Mundt
Quoted 3 lines · Simon Horman

Ok, so in that case the correct value for "sh_tmu.2" might actually be MSTP016 or MSTP015 depending on the bundle size. I am unsure how to test sh_tmu.2.

[snip]

Quoted 8 lines
Quoted 6 lines
Quoted 3 lines

Getting that information correct for sh_tmu.2 on the r8a7779 problem I'm not completely sure of the correct answer to. Though I now strongly suspect it is MSTP016.

Testing shouldn't be that difficult, just switch your clocksource to it and make sure all TMU-related MSTP bits are first cleared?

Thanks, I will try that.

Thanks again. I did try this and it appears to work.