Skip to contentSkip to conversation

tvp5150 regression after commit 9f924169c035

DISCUSSION Javier Martinez Canillas 3 participants 24 messages Feb 3, 2016 – Apr 15, 2016

Subsystems (5)
lore
Share & tools
lore
#0 lore raw

Hello Wolfram,

I've a issue with a I2C video decoder driver (drivers/media/i2c/tvp5150.c).

In v4.5-rc1, the driver gets I2C read / writes timeouts when accessing the device I2C registers:

tvp5150 1-005c: i2c i/o error: rc == -110
tvp5150: probe of 1-005c failed with error -110

The driver used to work up to v4.4 so this is a regression in v4.5-rc1:

tvp5150 1-005c: tvp5151 (1.0) chip found @ 0xb8 (OMAP I2C adapter) tvp5150 1-005c: tvp5151 detected.

I tracked down to commit 9f924169c035 ("i2c: always enable RuntimePM for the adapter device") and reverting that commit makes things to work again.

The tvp5150 driver doesn't have runtime PM support but the I2C controller driver does (drivers/i2c/busses/i2c-omap.c) FWIW.

I tried adding runtime PM support to tvp5150 (basically calling pm_runtime enable/get on probe before the first I2C read to resume the controller) but that it did not work.

Not filling the OMAP I2C driver's runtime PM callbacks does not help either.

Any hints about the proper way to fix this issue?

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
#1 lore raw

Hi,

thanks for reporting the issue!

Quoted 3 lines

Not filling the OMAP I2C driver's runtime PM callbacks does not help either.

Any hints about the proper way to fix this issue?

Can the I2C device be en-/disabled in some way? Which board is this happening with? Any specs for it publicly available?

Regards,

Wolfram

#2 lore raw

Hello Wolfram,

Quoted 4 lines · Wolfram Sang

Hi,

thanks for reporting the issue!

Thanks for the quick response.

Quoted 7 lines
Quoted 3 lines

Not filling the OMAP I2C driver's runtime PM callbacks does not help either.

Any hints about the proper way to fix this issue?

Can the I2C device be en-/disabled in some way? Which board is this happening with? Any specs for it publicly available?

Yes, the device has a PDN (Power-Down) pin that puts the device in stand by mode but it is not used by the driver. The only thing that the driver does is to toggle the PDN and RESET pins to put the device in normal operation.

In fact, the behavior after commit 9f924169c035 is the same as when the power sequence does not happen. That confused me at the beginning and I was chasing a red herring by looking at the OMAP GPIO driver and GPIO core.

The datasheet of the tvp5151 is public in case you need more info [0].

This is happening on an OMAP3 DM3730 IGEPv2 board [1] using an expansion board [2] that has the tvp5151 video decoder among other peripherals.

I don't think the schematics for these boards are public but I've them in case you need me to look at something.

Quoted 4 lines

Regards,

Wolfram

[0]: http://www.ti.com/lit/ds/symlink/tvp5151.pdf
[1]: https://isee.biz/products/igep-processor-boards/igepv2-dm3730
[2]: https://isee.biz/products/igep-expansion-boards/igepv2-expansion

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
#3 lore raw
In reply to Javier Martinez Canillas
Quoted 21 lines · Javier Martinez Canillas

Hello Wolfram,

I've a issue with a I2C video decoder driver (drivers/media/i2c/tvp5150.c).

In v4.5-rc1, the driver gets I2C read / writes timeouts when accessing the device I2C registers:

tvp5150 1-005c: i2c i/o error: rc == -110
tvp5150: probe of 1-005c failed with error -110

The driver used to work up to v4.4 so this is a regression in v4.5-rc1:

tvp5150 1-005c: tvp5151 (1.0) chip found @ 0xb8 (OMAP I2C adapter) tvp5150 1-005c: tvp5151 detected.

I tracked down to commit 9f924169c035 ("i2c: always enable RuntimePM for the adapter device") and reverting that commit makes things to work again.

The tvp5150 driver doesn't have runtime PM support but the I2C controller driver does (drivers/i2c/busses/i2c-omap.c) FWIW.

Quoted 7 lines

I tried adding runtime PM support to tvp5150 (basically calling pm_runtime enable/get on probe before the first I2C read to resume the controller) but that it did not work.

Not filling the OMAP I2C driver's runtime PM callbacks does not help either.

Any hints about the proper way to fix this issue?

Asking linux-pm for help:

The commit in question enables RuntimePM for the logical adapter device which sits between the HW I2C controller and the I2C client device. This adapter device has been used with pm_runtime_no_callbacks before enabling RPM. Now, Alan explained to me that "suspend events will propagate from the I2C clients all the way up to the adapter's parent." with RPM enabled for the adapter device. Which should be a no-op if the client doesn't do any PM at all? What do I miss?

Thanks,

Wolfram

#4 lore raw

Hello,

Quoted 40 lines · Wolfram Sang
Quoted 21 lines · Javier Martinez Canillas

Hello Wolfram,

I've a issue with a I2C video decoder driver (drivers/media/i2c/tvp5150.c).

In v4.5-rc1, the driver gets I2C read / writes timeouts when accessing the device I2C registers:

tvp5150 1-005c: i2c i/o error: rc == -110
tvp5150: probe of 1-005c failed with error -110

The driver used to work up to v4.4 so this is a regression in v4.5-rc1:

tvp5150 1-005c: tvp5151 (1.0) chip found @ 0xb8 (OMAP I2C adapter) tvp5150 1-005c: tvp5151 detected.

I tracked down to commit 9f924169c035 ("i2c: always enable RuntimePM for the adapter device") and reverting that commit makes things to work again.

The tvp5150 driver doesn't have runtime PM support but the I2C controller driver does (drivers/i2c/busses/i2c-omap.c) FWIW.

Quoted 7 lines

I tried adding runtime PM support to tvp5150 (basically calling pm_runtime enable/get on probe before the first I2C read to resume the controller) but that it did not work.

Not filling the OMAP I2C driver's runtime PM callbacks does not help either.

Any hints about the proper way to fix this issue?

Asking linux-pm for help:

The commit in question enables RuntimePM for the logical adapter device which sits between the HW I2C controller and the I2C client device. This adapter device has been used with pm_runtime_no_callbacks before enabling RPM. Now, Alan explained to me that "suspend events will propagate from the I2C clients all the way up to the adapter's parent." with RPM enabled for the adapter device. Which should be a no-op if the client doesn't do any PM at all? What do I miss?

Quoted 4 lines

Thanks,

Wolfram

I'm adding Tony Lindgren to the cc list as well since he is the OMAP maintainer and I see that has struggled lately with runtime PM issues so maybe he has more ideas.

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
maintainer
#5 lore raw

Hi,

Quoted 43 lines · Javier Martinez Canillas

Hello,

Quoted 32 lines · Wolfram Sang
Quoted 21 lines · Javier Martinez Canillas

Hello Wolfram,

I've a issue with a I2C video decoder driver (drivers/media/i2c/tvp5150.c).

In v4.5-rc1, the driver gets I2C read / writes timeouts when accessing the device I2C registers:

tvp5150 1-005c: i2c i/o error: rc == -110
tvp5150: probe of 1-005c failed with error -110

The driver used to work up to v4.4 so this is a regression in v4.5-rc1:

tvp5150 1-005c: tvp5151 (1.0) chip found @ 0xb8 (OMAP I2C adapter) tvp5150 1-005c: tvp5151 detected.

I tracked down to commit 9f924169c035 ("i2c: always enable RuntimePM for the adapter device") and reverting that commit makes things to work again.

The tvp5150 driver doesn't have runtime PM support but the I2C controller driver does (drivers/i2c/busses/i2c-omap.c) FWIW.

Quoted 7 lines

I tried adding runtime PM support to tvp5150 (basically calling pm_runtime enable/get on probe before the first I2C read to resume the controller) but that it did not work.

Not filling the OMAP I2C driver's runtime PM callbacks does not help either.

Any hints about the proper way to fix this issue?

Asking linux-pm for help:

Quoted 7 lines

The commit in question enables RuntimePM for the logical adapter device which sits between the HW I2C controller and the I2C client device. This adapter device has been used with pm_runtime_no_callbacks before enabling RPM. Now, Alan explained to me that "suspend events will propagate from the I2C clients all the way up to the adapter's parent." with RPM enabled for the adapter device. Which should be a no-op if the client doesn't do any PM at all? What do I miss?

Quoted 3 lines

I'm adding Tony Lindgren to the cc list as well since he is the OMAP maintainer and I see that has struggled lately with runtime PM issues so maybe he has more ideas.

Hmm yeah I wonder if this canned solution helps here too:

1. Check if the driver(s) are using pm_runtime_use_autosuspend()

2. If so, you must use pm_runtime_dont_use_autosuspend() before pm_runtime_put_sync() to make sure that pm_runtime_put_sync() works.

3. Or you can use pm_runtime_put_sync_suspend() instead of pm_runtime_put_sync() for sections of code where the clocks need to be stopped.

Regards,

Tony

#22 lore raw
In reply to Javier Martinez Canillas
Quoted 4 lines

I'm adding Tony Lindgren to the cc list as well since he is the OMAP maintainer and I see that has struggled lately with runtime PM issues so maybe he has more ideas.

Good idea. Did you try his patch which is in my for-current branch ("i2c: omap: Fix PM regression with deferred probe for pm_runtime_reinit")?

#23 lore raw

Hello Wolfram,

Quoted 9 lines · Wolfram Sang
Quoted 4 lines

I'm adding Tony Lindgren to the cc list as well since he is the OMAP maintainer and I see that has struggled lately with runtime PM issues so maybe he has more ideas.

Good idea. Did you try his patch which is in my for-current branch ("i2c: omap: Fix PM regression with deferred probe for pm_runtime_reinit")?

Yes I did but unfortunately it did not help in this case.

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
#6 lore raw
In reply to Tony Lindgren

Hello Tony,

Quoted 46 lines · Tony Lindgren

Hi,

Quoted 43 lines · Javier Martinez Canillas

Hello,

Quoted 39 lines · Wolfram Sang
Quoted 21 lines · Javier Martinez Canillas

Hello Wolfram,

I've a issue with a I2C video decoder driver (drivers/media/i2c/tvp5150.c).

In v4.5-rc1, the driver gets I2C read / writes timeouts when accessing the device I2C registers:

tvp5150 1-005c: i2c i/o error: rc == -110
tvp5150: probe of 1-005c failed with error -110

The driver used to work up to v4.4 so this is a regression in v4.5-rc1:

tvp5150 1-005c: tvp5151 (1.0) chip found @ 0xb8 (OMAP I2C adapter) tvp5150 1-005c: tvp5151 detected.

I tracked down to commit 9f924169c035 ("i2c: always enable RuntimePM for the adapter device") and reverting that commit makes things to work again.

The tvp5150 driver doesn't have runtime PM support but the I2C controller driver does (drivers/i2c/busses/i2c-omap.c) FWIW.

Quoted 7 lines

I tried adding runtime PM support to tvp5150 (basically calling pm_runtime enable/get on probe before the first I2C read to resume the controller) but that it did not work.

Not filling the OMAP I2C driver's runtime PM callbacks does not help either.

Any hints about the proper way to fix this issue?

Asking linux-pm for help:

The commit in question enables RuntimePM for the logical adapter device which sits between the HW I2C controller and the I2C client device. This adapter device has been used with pm_runtime_no_callbacks before enabling RPM. Now, Alan explained to me that "suspend events will propagate from the I2C clients all the way up to the adapter's parent." with RPM enabled for the adapter device. Which should be a no-op if the client doesn't do any PM at all? What do I miss?

Quoted 8 lines
Quoted 3 lines

I'm adding Tony Lindgren to the cc list as well since he is the OMAP maintainer and I see that has struggled lately with runtime PM issues so maybe he has more ideas.

Hmm yeah I wonder if this canned solution helps here too:

1. Check if the driver(s) are using pm_runtime_use_autosuspend()

By driver do you mean the OMAP GPIO driver or the tvp5150 I2C driver? The latter does not have runtime PM support.

Quoted 8 lines

2. If so, you must use pm_runtime_dont_use_autosuspend() before pm_runtime_put_sync() to make sure that pm_runtime_put_sync() works.

3. Or you can use pm_runtime_put_sync_suspend() instead of pm_runtime_put_sync() for sections of code where the clocks need to be stopped.

I can check if the OMAP GPIO is following these and give a try but don't have access to the board right now so I'll do it on Monday.

Quoted 4 lines

Regards,

Tony

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
maintainer
#7 lore raw
Quoted 8 lines · Javier Martinez Canillas
Quoted 4 lines · Tony Lindgren

Hmm yeah I wonder if this canned solution helps here too:

1. Check if the driver(s) are using pm_runtime_use_autosuspend()

By driver do you mean the OMAP GPIO driver or the tvp5150 I2C driver? The latter does not have runtime PM support.

Sounds like OMAP GPIO then.

Quoted 11 lines
Quoted 8 lines

2. If so, you must use pm_runtime_dont_use_autosuspend() before pm_runtime_put_sync() to make sure that pm_runtime_put_sync() works.

3. Or you can use pm_runtime_put_sync_suspend() instead of pm_runtime_put_sync() for sections of code where the clocks need to be stopped.

I can check if the OMAP GPIO is following these and give a try but don't have access to the board right now so I'll do it on Monday.

It does not seem to be using pm_runtime_autosuspend(). Did you try reverting commit de85b9d57ab ("PM / runtime: Re-init runtime PM states at probe error and driver unbind") and see if that helps?

If it does, then sounds like we may have some other regression as well.

Regards,

Tony

#8 lore raw

Hello Tony,

Quoted 4 lines · Tony Lindgren
Quoted 3 lines · Javier Martinez Canillas
Quoted 2 lines · Tony Lindgren

Hmm yeah I wonder if this canned solution helps here too:

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

1. Check if the driver(s) are using pm_runtime_use_autosuspend()

By driver do you mean the OMAP GPIO driver or the tvp5150 I2C driver? The latter does not have runtime PM support.

Sounds like OMAP GPIO then.

Ok.

Quoted 9 lines
Quoted 9 lines
Quoted 8 lines

2. If so, you must use pm_runtime_dont_use_autosuspend() before pm_runtime_put_sync() to make sure that pm_runtime_put_sync() works.

3. Or you can use pm_runtime_put_sync_suspend() instead of pm_runtime_put_sync() for sections of code where the clocks need to be stopped.

Quoted 8 lines
Quoted 2 lines

I can check if the OMAP GPIO is following these and give a try but don't have access to the board right now so I'll do it on Monday.

It does not seem to be using pm_runtime_autosuspend(). Did you try reverting commit de85b9d57ab ("PM / runtime: Re-init runtime PM states at probe error and driver unbind") and see if that helps?

Yes, that's the first thing I tried when I noticed your patch:

("i2c: omap: Fix PM regression with deferred probe for pm_runtime_reinit")

But neither reverting commit de85b9d57ab nor your fix made a difference.

Quoted 3 lines

If it does, then sounds like we may have some other regression as well.

It seems that is not related but I hope that given you were looking at the runtime PM core lately, maybe you can figure out what we are missing.

I'm far from being familiar with the runtime PM framework but I've looked and can't figure out why Wolfram's commit make this driver to fail and reverting his commit make its work again.

Quoted 4 lines

Regards,

Tony

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
maintainer
#9 lore raw
Quoted 35 lines · Javier Martinez Canillas

Hello Tony,

Quoted 4 lines · Tony Lindgren
Quoted 3 lines · Javier Martinez Canillas
Quoted 2 lines · Tony Lindgren

Hmm yeah I wonder if this canned solution helps here too:

Quoted 8 lines
Quoted 5 lines
Quoted 2 lines

1. Check if the driver(s) are using pm_runtime_use_autosuspend()

By driver do you mean the OMAP GPIO driver or the tvp5150 I2C driver? The latter does not have runtime PM support.

Sounds like OMAP GPIO then.

Ok.

Quoted 9 lines
Quoted 9 lines
Quoted 8 lines

2. If so, you must use pm_runtime_dont_use_autosuspend() before pm_runtime_put_sync() to make sure that pm_runtime_put_sync() works.

3. Or you can use pm_runtime_put_sync_suspend() instead of pm_runtime_put_sync() for sections of code where the clocks need to be stopped.

Quoted 8 lines
Quoted 2 lines

I can check if the OMAP GPIO is following these and give a try but don't have access to the board right now so I'll do it on Monday.

It does not seem to be using pm_runtime_autosuspend(). Did you try reverting commit de85b9d57ab ("PM / runtime: Re-init runtime PM states at probe error and driver unbind") and see if that helps?

Quoted 9 lines

Yes, that's the first thing I tried when I noticed your patch:

("i2c: omap: Fix PM regression with deferred probe for pm_runtime_reinit")

But neither reverting commit de85b9d57ab nor your fix made a difference.

Quoted 2 lines

If it does, then sounds like we may have some other regression as well.

OK I doubt it's the GPIO driver if reverting 9f924169c035 helps.

I do see some GPIO regressions in current Linux next though, but sounds like you're using v4.5-rc series.

Quoted 8 lines

It seems that is not related but I hope that given you were looking at the runtime PM core lately, maybe you can figure out what we are missing.

I'm far from being familiar with the runtime PM framework but I've looked and can't figure out why Wolfram's commit make this driver to fail and reverting his commit make its work again.

No idea. What kind of PM runtime use case has that one been tested with?

Regards,

Tony

#10 lore raw

Hello Tony,

On 02/12/2016 08:46 PM, Tony Lindgren wrote:

[snip]

Quoted 4 lines

OK I doubt it's the GPIO driver if reverting 9f924169c035 helps.

Right, sorry. I first thought the problem was with the OMAP GPIO driver but then found that reverting 9f924169c035 caused the issue to go away which indicates the problem is somehow related to I2C.

I forgot that and got confused again when answering to your email.

Quoted 3 lines

I do see some GPIO regressions in current Linux next though, but sounds like you're using v4.5-rc series.

Yes, I'm testing with v4.5-rc and not linux-next.

Quoted 4 lines
Quoted 4 lines

It seems that is not related but I hope that given you were looking at the runtime PM core lately, maybe you can figure out what we are missing.

Quoted 8 lines
Quoted 4 lines

I'm far from being familiar with the runtime PM framework but I've looked and can't figure out why Wolfram's commit make this driver to fail and reverting his commit make its work again.

No idea. What kind of PM runtime use case has that one been tested with?

Nothing special, the driver just fails to probe. What the driver does is to toggle some pins (Power Down and Reset) in the tvp5150 chip to put it in an operational state and then attempts to read an I2C register to detect the chip version and this is what fails.

Quoted 4 lines

Regards,

Tony

I'll try to find some time next week to dig deeper on this. Just thought that may be related to the issue you found but it seems that's not the case.

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
#11 lore raw
Quoted 4 lines

I'll try to find some time next week to dig deeper on this. Just thought that may be related to the issue you found but it seems that's not the case.

Any updates on this?

Thanks,

Wolfram

#12 lore raw

Hello Wolfram,

Quoted 7 lines · Wolfram Sang
Quoted 4 lines

I'll try to find some time next week to dig deeper on this. Just thought that may be related to the issue you found but it seems that's not the case.

Any updates on this?

Sorry, I've been sick and then busy with other stuff so I didn't have time to dig deeper on this.

Quoted 4 lines

Thanks,

Wolfram

I'll write what I found so far in case someone with better knowledge about the runtime PM API and the OMAP I2C controller driver can have an idea of what could be causing this.

The problem is that after commit 9f924169c035 ("i2c: always enable RuntimePM for the adapter device"), i2c_smbus_read_byte_data() fails and returns -110 (ETIMEDOUT) error in the tvp5150 I2C driver.

The call trace is the following:

i2c_device_probe
tvp5150_probe

    tvp5150_read
      i2c_smbus_read_byte_data
        i2c_smbus_xfer
	  i2c_transfer
	    omap_i2c_xfer
              omap_i2c_xfer_msg

The fail in omap_i2c_xfer_msg() is due the wait_for_completion_timeout() timeout, so it seems the omap_i2c_isr_thread() is not calling complete() for the cmd_complete completion when runtime PM is enabled (no idea why).

If I revert commit 9f924169c035 then things works again or if I use the following patch that basically disables runtime PM for the OMAP adapter after it has been enabled by the I2C core due commit 9f924169c035:

drivers/i2c/busses/i2c-omap.c +2 -0
index 37a88f0ae179..0b72b21b379d 100644
@@ -1445,6 +1445,8 @@ omap_i2c_probe(struct platform_device *pdev)
 		goto err_unuse_clocks;
 	}
 
+	pm_runtime_disable(&adap->dev);
+
 	dev_info(omap->dev, "bus %d rev%d.%d at %d kHz\n", adap->nr,
 		 major, minor, omap->speed);
 

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
#13 lore raw
Quoted 4 lines

I'll write what I found so far in case someone with better knowledge about the runtime PM API and the OMAP I2C controller driver can have an idea of what could be causing this.

Thanks for the summary. I got no other reports like this, I wonder about that. That being said, can you try this patch if it makes a change?

http://patchwork.ozlabs.org/patch/609280/

#14 lore raw

Hello Wolfram,

Quoted 6 lines · Wolfram Sang
Quoted 4 lines

I'll write what I found so far in case someone with better knowledge about the runtime PM API and the OMAP I2C controller driver can have an idea of what could be causing this.

Thanks for the summary. I got no other reports like this, I wonder about

Yes, I also wonder why I'm the only one facing this issue... maybe no one else is using the tvp5150 driver on an OMAP board with mainline?

Quoted 4 lines

that. That being said, can you try this patch if it makes a change?

http://patchwork.ozlabs.org/patch/609280/

Unfortunately it doesn't help, I just tried today's -next (next-20160414) that already has the mentioned patch but the issue is still present.

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
#15 lore raw
Quoted 3 lines

Yes, I also wonder why I'm the only one facing this issue... maybe no one else is using the tvp5150 driver on an OMAP board with mainline?

I wonder why it only affects tvp5150. I don't see the connection yet.

#16 lore raw

Hello Wofram,

Quoted 6 lines · Wolfram Sang
Quoted 3 lines

Yes, I also wonder why I'm the only one facing this issue... maybe no one else is using the tvp5150 driver on an OMAP board with mainline?

I wonder why it only affects tvp5150. I don't see the connection yet.

Yes, me neither. All other I2C devices are working properly on this board.

The only thing I can think, is that the tvp5150 needs a reset sequence in order to be operative. It basically toggles two pins in the chip, this is done in tvp5150_init() [0] and is needed before accessing I2C registers.

Maybe runtime pm has an effect on this and the chip is not reset correctly?

[0]: https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/tree/drivers/media/i2c/tvp5150.c#n1311

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
maintainer
#17 lore raw

Hi,

Quoted 10 lines · Javier Martinez Canillas

Hello Wofram,

Quoted 6 lines · Wolfram Sang
Quoted 3 lines

Yes, I also wonder why I'm the only one facing this issue... maybe no one else is using the tvp5150 driver on an OMAP board with mainline?

I wonder why it only affects tvp5150. I don't see the connection yet.

Quoted 7 lines

Yes, me neither. All other I2C devices are working properly on this board.

The only thing I can think, is that the tvp5150 needs a reset sequence in order to be operative. It basically toggles two pins in the chip, this is done in tvp5150_init() [0] and is needed before accessing I2C registers.

Maybe runtime pm has an effect on this and the chip is not reset correctly?

Is this with omap3 and does tvp5150 have a reset GPIO pin?

If so, you could be hitting the GPIO errata where a glitch can happen when restoring the GPIO state coming back from off mode in idle. This happes for GPIO pins that are not in GPIO bank1 and have an external pull down resistor on the line.

The short term workaround is to mux the reset pin to use the internal pulls by using PIN_INPUT_PULLUP | MUX_MODE7, or depending on the direction, PIN_INPUT_PULLDOWN | MUX_MODE7.

The long term workaround is tho have gpio-omap.c do this dynamically with pinctrl-single.c using gpio-ranges, but that's going to take a while.. You can search for erratum 1.158 for more info.

Regards,

Tony

Quoted 1 line

[0]: https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/tree/drivers/media/i2c/tvp5150.c#n1311

#18 lore raw

Hello Tony,

Quoted 15 lines · Tony Lindgren

Hi,

Quoted 12 lines · Javier Martinez Canillas

Hello Wofram,

Quoted 6 lines · Wolfram Sang
Quoted 3 lines

Yes, I also wonder why I'm the only one facing this issue... maybe no one else is using the tvp5150 driver on an OMAP board with mainline?

I wonder why it only affects tvp5150. I don't see the connection yet.

Yes, me neither. All other I2C devices are working properly on this board.

Quoted 8 lines
Quoted 5 lines

The only thing I can think, is that the tvp5150 needs a reset sequence in order to be operative. It basically toggles two pins in the chip, this is done in tvp5150_init() [0] and is needed before accessing I2C registers.

Maybe runtime pm has an effect on this and the chip is not reset correctly?

Is this with omap3 and does tvp5150 have a reset GPIO pin?

Yes, it's a DM3730 (OMAP3) and yes the tvp5150 (actually it's a tvp5151) has a reset pin that has to be toggled, along with a power-down pin for the chip to be in an operative state before accessing the I2C registers. That is the power/reset sequence I mentioned before.

Quoted 5 lines

If so, you could be hitting the GPIO errata where a glitch can happen when restoring the GPIO state coming back from off mode in idle. This happes for GPIO pins that are not in GPIO bank1 and have an external pull down resistor on the line.

The GPIO lines connected to these pins are:

GPIO126 (bank4 pin 30) -> tvp5150 power-down pin
GPIO167 (bank6 pin 7) -> tvp5150 reset pin

Neither are in GPIO bank1 so they could be affected by the errata you mention but there isn't external pull down (or up) resistors on these lines AFAICT by looking at the board schematics. I've added to the cc list to other people that are familiar with the board in case I missed something.

Quoted 4 lines

The short term workaround is to mux the reset pin to use the internal pulls by using PIN_INPUT_PULLUP | MUX_MODE7, or depending on the direction, PIN_INPUT_PULLDOWN | MUX_MODE7.

I guess you meant MUX_MODE4 here since the pin has to be in GPIO mode?

Also, I wonder how the issue could be related to the GPIO controller since is when enabling runtime PM for the I2C controller that things fail. IOW, disabling runtime PM for the I2C adapter shouldn't make things to work if the problem was caused by the mentioned GPIO errata.

In any case, I've tried to use the internal pulls as you suggested but that didn't solve the issue.

Quoted 9 lines

The long term workaround is tho have gpio-omap.c do this dynamically with pinctrl-single.c using gpio-ranges, but that's going to take a while.. You can search for erratum 1.158 for more info.

Regards,

Tony

Quoted 1 line

[0]: https://git.kernel.org/cgit/linux/kernel/git/next/linux-next.git/tree/drivers/media/i2c/tvp5150.c#n1311

I'm really out of ideas on this, I don't think that Wolfram change has to be reverted since it seems I'm the only one facing this issue and also the DTS for the board I'm using is not even in mainline yet.

I'm waiting for some discussions about the Media Controller input connectors DT bindings to settle before posting the board DTS patches.

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
maintainer
#19 lore raw
Quoted 25 lines · Javier Martinez Canillas
Quoted 1 line · Tony Lindgren

Is this with omap3 and does tvp5150 have a reset GPIO pin?

Yes, it's a DM3730 (OMAP3) and yes the tvp5150 (actually it's a tvp5151) has a reset pin that has to be toggled, along with a power-down pin for the chip to be in an operative state before accessing the I2C registers. That is the power/reset sequence I mentioned before.

Quoted 5 lines

If so, you could be hitting the GPIO errata where a glitch can happen when restoring the GPIO state coming back from off mode in idle. This happes for GPIO pins that are not in GPIO bank1 and have an external pull down resistor on the line.

The GPIO lines connected to these pins are:

GPIO126 (bank4 pin 30) -> tvp5150 power-down pin
GPIO167 (bank6 pin 7) -> tvp5150 reset pin

Neither are in GPIO bank1 so they could be affected by the errata you mention but there isn't external pull down (or up) resistors on these lines AFAICT by looking at the board schematics. I've added to the cc list to other people that are familiar with the board in case I missed something.

Quoted 5 lines
Quoted 3 lines

The short term workaround is to mux the reset pin to use the internal pulls by using PIN_INPUT_PULLUP | MUX_MODE7, or depending on the direction, PIN_INPUT_PULLDOWN | MUX_MODE7.

I guess you meant MUX_MODE4 here since the pin has to be in GPIO mode?

No, the glitch affects the GPIO mode, so that's why direction + pull + safe mode is needed.

Quoted 4 lines

Also, I wonder how the issue could be related to the GPIO controller since is when enabling runtime PM for the I2C controller that things fail. IOW, disabling runtime PM for the I2C adapter shouldn't make things to work if the problem was caused by the mentioned GPIO errata.

If you block PM runtime for I2C, then it blocks deeper idle states for the whole device. Note that you can disable off mode during idle and suspend with:

# echo 0 > /sys/kernel/debug/pm_debug/enable_off_mode
Quoted 2 lines

In any case, I've tried to use the internal pulls as you suggested but that didn't solve the issue.

OK. Just to be sure.. Did you use the safe mode mux option instead of the GPIO mux option?

Note that in some cases the internal pull is not strong enough to keep the reset line up if there's an external pull down resistor.

Regards,

Tony

#20 lore raw

Hello Tony,

Quoted 10 lines · Tony Lindgren
Quoted 6 lines
Quoted 4 lines

The short term workaround is to mux the reset pin to use the internal pulls by using PIN_INPUT_PULLUP | MUX_MODE7, or depending on the direction, PIN_INPUT_PULLDOWN | MUX_MODE7.

I guess you meant MUX_MODE4 here since the pin has to be in GPIO mode?

No, the glitch affects the GPIO mode, so that's why direction + pull + safe mode is needed.

Ah ok, thanks for the explanation and sorry for the confusion.

Quoted 7 lines
Quoted 4 lines

Also, I wonder how the issue could be related to the GPIO controller since is when enabling runtime PM for the I2C controller that things fail. IOW, disabling runtime PM for the I2C adapter shouldn't make things to work if the problem was caused by the mentioned GPIO errata.

If you block PM runtime for I2C, then it blocks deeper idle states for the whole device. Note that you can disable off mode during idle

Thanks again for this clarification.

Quoted 4 lines

and suspend with:

# echo 0 > /sys/kernel/debug/pm_debug/enable_off_mode

I see thought that enable_off_mode is 0 by default when booting the board:

# cat /sys/kernel/debug/pm_debug/enable_off_mode 
0

So if I understood your explanation correctly, that means that the glitch should not happen for the GPIO pins since the machine doesn't enter into deeper idle states that could cause the glitch from erratum 1.158?

Quoted 6 lines
Quoted 2 lines

In any case, I've tried to use the internal pulls as you suggested but that didn't solve the issue.

OK. Just to be sure.. Did you use the safe mode mux option instead of the GPIO mux option?

No sorry, I tested with the GPIO mux mode before since I misunderstood your previous email. But now I've tested with safe mode mux and didn't make a difference (which was expected since off mode is disabled AFAIU).

Quoted 3 lines

Note that in some cases the internal pull is not strong enough to keep the reset line up if there's an external pull down resistor.

Yes, I got that from your previous email but as I mentioned, the lines in the board don't have external pull resistors.

Quoted 4 lines

Regards,

Tony

Best regards,

Signature · 3 lines
Javier Martinez Canillas
Open Source Group
Samsung Research America
maintainer
#21 lore raw
Quoted 8 lines · Javier Martinez Canillas
Quoted 2 lines · Tony Lindgren

If you block PM runtime for I2C, then it blocks deeper idle states for the whole device. Note that you can disable off mode during idle

Thanks again for this clarification.

Quoted 2 lines

and suspend with:

Quoted 7 lines
Quoted 2 lines
# echo 0 > /sys/kernel/debug/pm_debug/enable_off_mode

I see thought that enable_off_mode is 0 by default when booting the board:

# cat /sys/kernel/debug/pm_debug/enable_off_mode 
0

OK so you're not hitting off mode then.

Quoted 3 lines

So if I understood your explanation correctly, that means that the glitch should not happen for the GPIO pins since the machine doesn't enter into deeper idle states that could cause the glitch from erratum 1.158?

Correct. But you could still have a dependency to some other device driver that stays active if I2C keeps the whole system from hitting retention mode during idle.

Regards,

Tony