[1/2] drivers: amba: properly handle devices with power domains

Message ID 1448456289-31960-1-git-send-email-m.szyprowski@samsung.com
State New
Headers show

Commit Message

Marek Szyprowski Nov. 25, 2015, 12:58 p.m.
To read pid/cid registers, the probed device need to be properly turned on.
When it is inside a power domain, the bus code should ensure that the
given power domain is enabled before trying to access device's registers.

Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>

---
 drivers/amba/bus.c | 7 +++++++
 1 file changed, 7 insertions(+)

-- 
1.9.2

--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Comments

Marek Szyprowski Nov. 25, 2015, 1:34 p.m. | #1
Hello,

On 2015-11-25 14:24, Russell King - ARM Linux wrote:
> On Wed, Nov 25, 2015 at 01:58:09PM +0100, Marek Szyprowski wrote:

>> To read pid/cid registers, the probed device need to be properly turned on.

>> When it is inside a power domain, the bus code should ensure that the

>> given power domain is enabled before trying to access device's registers.

>>

>> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>

>> ---

>>   drivers/amba/bus.c | 7 +++++++

>>   1 file changed, 7 insertions(+)

>>

>> diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c

>> index f009936..25715cb 100644

>> --- a/drivers/amba/bus.c

>> +++ b/drivers/amba/bus.c

>> @@ -373,6 +373,12 @@ int amba_device_add(struct amba_device *dev, struct resource *parent)

>>   		goto err_release;

>>   	}

>>   

>> +	ret = dev_pm_domain_attach(&dev->dev, true);

>> +	if (ret) {

>> +		iounmap(tmp);

>> +		goto err_release;

>> +	}

>> +

> NAK.  If dev_pm_domain_attach() returns an error, even -EPROBE_DEFER,

> the result will be a missing device that has no way to be recovered.

> This is too fragile.


Then how the problem of accessing registers in turned-off device should 
be solved?

Is ignoring dev_pm_domain_attach() return value a solution for you?

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland

--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Ulf Hansson Nov. 25, 2015, 1:56 p.m. | #2
On 25 November 2015 at 14:34, Marek Szyprowski <m.szyprowski@samsung.com> wrote:
> Hello,

>

> On 2015-11-25 14:24, Russell King - ARM Linux wrote:

>>

>> On Wed, Nov 25, 2015 at 01:58:09PM +0100, Marek Szyprowski wrote:

>>>

>>> To read pid/cid registers, the probed device need to be properly turned

>>> on.

>>> When it is inside a power domain, the bus code should ensure that the

>>> given power domain is enabled before trying to access device's registers.

>>>

>>> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>

>>> ---

>>>   drivers/amba/bus.c | 7 +++++++

>>>   1 file changed, 7 insertions(+)

>>>

>>> diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c

>>> index f009936..25715cb 100644

>>> --- a/drivers/amba/bus.c

>>> +++ b/drivers/amba/bus.c

>>> @@ -373,6 +373,12 @@ int amba_device_add(struct amba_device *dev, struct

>>> resource *parent)

>>>                 goto err_release;

>>>         }

>>>   +     ret = dev_pm_domain_attach(&dev->dev, true);

>>> +       if (ret) {

>>> +               iounmap(tmp);

>>> +               goto err_release;

>>> +       }

>>> +

>>

>> NAK.  If dev_pm_domain_attach() returns an error, even -EPROBE_DEFER,

>> the result will be a missing device that has no way to be recovered.

>> This is too fragile.


I agree with Russell here, this isn't going to work.

>

>

> Then how the problem of accessing registers in turned-off device should be

> solved?


The long term solution has been discussed [1], please have a look and
feel free to hack on it if you like. If not, I will sooner or later
find time to pick up the work I have started, but can't give you any
promises when ready.

>

> Is ignoring dev_pm_domain_attach() return value a solution for you?


That's probably better than nothing, but I wonder if it in practice
will have any effect? It requires the PM domain to be initialized (and
having a DT provider for it) before you register the AMBA device, is
that so in your case?

Kind regards
Uffe

[1]
http://comments.gmane.org/gmane.linux.kernel.mmc/32587
--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Marek Szyprowski Nov. 25, 2015, 2:43 p.m. | #3
Hello,

On 2015-11-25 14:56, Ulf Hansson wrote:
> On 25 November 2015 at 14:34, Marek Szyprowski <m.szyprowski@samsung.com> wrote:

>> On 2015-11-25 14:24, Russell King - ARM Linux wrote:

>>> On Wed, Nov 25, 2015 at 01:58:09PM +0100, Marek Szyprowski wrote:

>>>> To read pid/cid registers, the probed device need to be properly turned

>>>> on.

>>>> When it is inside a power domain, the bus code should ensure that the

>>>> given power domain is enabled before trying to access device's registers.

>>>>

>>>> Signed-off-by: Marek Szyprowski <m.szyprowski@samsung.com>

>>>> ---

>>>>    drivers/amba/bus.c | 7 +++++++

>>>>    1 file changed, 7 insertions(+)

>>>>

>>>> diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c

>>>> index f009936..25715cb 100644

>>>> --- a/drivers/amba/bus.c

>>>> +++ b/drivers/amba/bus.c

>>>> @@ -373,6 +373,12 @@ int amba_device_add(struct amba_device *dev, struct

>>>> resource *parent)

>>>>                  goto err_release;

>>>>          }

>>>>    +     ret = dev_pm_domain_attach(&dev->dev, true);

>>>> +       if (ret) {

>>>> +               iounmap(tmp);

>>>> +               goto err_release;

>>>> +       }

>>>> +

>>> NAK.  If dev_pm_domain_attach() returns an error, even -EPROBE_DEFER,

>>> the result will be a missing device that has no way to be recovered.

>>> This is too fragile.

> I agree with Russell here, this isn't going to work.

>

>>

>> Then how the problem of accessing registers in turned-off device should be

>> solved?

> The long term solution has been discussed [1], please have a look and

> feel free to hack on it if you like. If not, I will sooner or later

> find time to pick up the work I have started, but can't give you any

> promises when ready.


Frankly, I still don't get how do you want to solve the problem of
incomplete/not-yet-probed pm domain during the amba_device_add() and reading
pid/cid identifiers, so I will leave it for you.

>> Is ignoring dev_pm_domain_attach() return value a solution for you?

> That's probably better than nothing, but I wonder if it in practice

> will have any effect? It requires the PM domain to be initialized (and

> having a DT provider for it) before you register the AMBA device, is

> that so in your case?


Yes, this will work for me, as power domain driver is registered very early
on Exynos platform (this was already needed to let it working with IOMMU
driver).

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland

--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Marek Szyprowski Nov. 26, 2015, 8:39 a.m. | #4
Hello,

On 2015-11-25 19:09, Russell King - ARM Linux wrote:
> On Wed, Nov 25, 2015 at 02:56:10PM +0100, Ulf Hansson wrote:

>> On 25 November 2015 at 14:34, Marek Szyprowski <m.szyprowski@samsung.com> wrote:

>>> Is ignoring dev_pm_domain_attach() return value a solution for you?

>> That's probably better than nothing, but I wonder if it in practice

>> will have any effect?

> If the PM domain is down, then trying to tread the ID is likely to oops

> the kernel.


In my case kernel simply hangs (no single message, even when earlyprintk 
is enabled)
instead of oopsing, that's why I've submitted this patch.

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland

--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Ulf Hansson Nov. 26, 2015, 10:24 a.m. | #5
On 26 November 2015 at 09:39, Marek Szyprowski <m.szyprowski@samsung.com> wrote:
> Hello,

>

> On 2015-11-25 19:09, Russell King - ARM Linux wrote:

>>

>> On Wed, Nov 25, 2015 at 02:56:10PM +0100, Ulf Hansson wrote:

>>>

>>> On 25 November 2015 at 14:34, Marek Szyprowski <m.szyprowski@samsung.com>

>>> wrote:

>>>>

>>>> Is ignoring dev_pm_domain_attach() return value a solution for you?

>>>

>>> That's probably better than nothing, but I wonder if it in practice

>>> will have any effect?

>>

>> If the PM domain is down, then trying to tread the ID is likely to oops

>> the kernel.

>

>

> In my case kernel simply hangs (no single message, even when earlyprintk is

> enabled)

> instead of oopsing, that's why I've submitted this patch.

>


Okay.

I suggest we go ahead and try that approach (ignoring the return
value). It's "quick fix", but the easiest way forward to solve the
problem.

My only concern is if such change impacts the boot time. We should do
some tests to see if the change is negligible, if not we should
probably think of something else (like keeping the PM domain powered
until late_init).

Russell, what do you think?

Kind regards
Uffe
--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Marek Szyprowski Nov. 26, 2015, 12:33 p.m. | #6
Hello,

On 2015-11-26 11:59, Russell King - ARM Linux wrote:
> On Thu, Nov 26, 2015 at 11:24:50AM +0100, Ulf Hansson wrote:

>> On 26 November 2015 at 09:39, Marek Szyprowski <m.szyprowski@samsung.com> wrote:

>>> Hello,

>>>

>>> On 2015-11-25 19:09, Russell King - ARM Linux wrote:

>>>> On Wed, Nov 25, 2015 at 02:56:10PM +0100, Ulf Hansson wrote:

>>>>> On 25 November 2015 at 14:34, Marek Szyprowski <m.szyprowski@samsung.com>

>>>>> wrote:

>>>>>> Is ignoring dev_pm_domain_attach() return value a solution for you?

>>>>> That's probably better than nothing, but I wonder if it in practice

>>>>> will have any effect?

>>>> If the PM domain is down, then trying to tread the ID is likely to oops

>>>> the kernel.

>>>

>>> In my case kernel simply hangs (no single message, even when earlyprintk is

>>> enabled)

>>> instead of oopsing, that's why I've submitted this patch.

>>>

>> Okay.

>>

>> I suggest we go ahead and try that approach (ignoring the return

>> value). It's "quick fix", but the easiest way forward to solve the

>> problem.

>>

>> My only concern is if such change impacts the boot time. We should do

>> some tests to see if the change is negligible, if not we should

>> probably think of something else (like keeping the PM domain powered

>> until late_init).

>>

>> Russell, what do you think?

> I thought someone had a patch to read the ID during the match callback?


Okay, I've reused that patch and it seems to solve mthisproblem. I will 
send v2
in a few minutes.

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland

--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Marek Szyprowski Nov. 26, 2015, 12:55 p.m. | #7
Hello,

On 2015-11-26 11:24, Ulf Hansson wrote:
> On 26 November 2015 at 09:39, Marek Szyprowski <m.szyprowski@samsung.com> wrote:

>> On 2015-11-25 19:09, Russell King - ARM Linux wrote:

>>> On Wed, Nov 25, 2015 at 02:56:10PM +0100, Ulf Hansson wrote:

>>>> On 25 November 2015 at 14:34, Marek Szyprowski <m.szyprowski@samsung.com>

>>>> wrote:

>>>>> Is ignoring dev_pm_domain_attach() return value a solution for you?

>>>> That's probably better than nothing, but I wonder if it in practice

>>>> will have any effect?

>>> If the PM domain is down, then trying to tread the ID is likely to oops

>>> the kernel.

>>

>> In my case kernel simply hangs (no single message, even when earlyprintk is

>> enabled)

>> instead of oopsing, that's why I've submitted this patch.

>>

> Okay.

>

> I suggest we go ahead and try that approach (ignoring the return

> value). It's "quick fix", but the easiest way forward to solve the

> problem.

>

> My only concern is if such change impacts the boot time. We should do

> some tests to see if the change is negligible, if not we should

> probably think of something else (like keeping the PM domain powered

> until late_init).

>

> Russell, what do you think?


Keeping PM domains powered until late_init would be a good idea 
regardless of
the problem discussed here, because I often see that pm domains are 
turned on
and off a dozen of times during typical boot of Exynos based systems.

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland

--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Mathieu Poirier Nov. 26, 2015, 3:04 p.m. | #8
On 26 November 2015 at 03:24, Ulf Hansson <ulf.hansson@linaro.org> wrote:
> On 26 November 2015 at 09:39, Marek Szyprowski <m.szyprowski@samsung.com> wrote:

>> Hello,

>>

>> On 2015-11-25 19:09, Russell King - ARM Linux wrote:

>>>

>>> On Wed, Nov 25, 2015 at 02:56:10PM +0100, Ulf Hansson wrote:

>>>>

>>>> On 25 November 2015 at 14:34, Marek Szyprowski <m.szyprowski@samsung.com>

>>>> wrote:

>>>>>

>>>>> Is ignoring dev_pm_domain_attach() return value a solution for you?

>>>>

>>>> That's probably better than nothing, but I wonder if it in practice

>>>> will have any effect?

>>>

>>> If the PM domain is down, then trying to tread the ID is likely to oops

>>> the kernel.

>>

>>

>> In my case kernel simply hangs (no single message, even when earlyprintk is

>> enabled)

>> instead of oopsing, that's why I've submitted this patch.

>>

>

> Okay.

>

> I suggest we go ahead and try that approach (ignoring the return

> value). It's "quick fix", but the easiest way forward to solve the

> problem.

>

> My only concern is if such change impacts the boot time. We should do

> some tests to see if the change is negligible, if not we should

> probably think of something else (like keeping the PM domain powered

> until late_init).

>

> Russell, what do you think?


That's the approach I took for coresight on Juno - simply initialise
the power domain very early on and keep it on for things like amba
probing to be successful.  When the boot process is done unused genPDs
are switched off automatically.  That way we don't need to add extra
code to amba.

>

> Kind regards

> Uffe

>

> _______________________________________________

> linux-arm-kernel mailing list

> linux-arm-kernel@lists.infradead.org

> http://lists.infradead.org/mailman/listinfo/linux-arm-kernel

--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Marek Szyprowski Nov. 27, 2015, 7:42 a.m. | #9
Hello,

On 2015-11-26 16:04, Mathieu Poirier wrote:
> On 26 November 2015 at 03:24, Ulf Hansson <ulf.hansson@linaro.org> wrote:

>> On 26 November 2015 at 09:39, Marek Szyprowski <m.szyprowski@samsung.com> wrote:

>>> On 2015-11-25 19:09, Russell King - ARM Linux wrote:

>>>> On Wed, Nov 25, 2015 at 02:56:10PM +0100, Ulf Hansson wrote:

>>>>> On 25 November 2015 at 14:34, Marek Szyprowski <m.szyprowski@samsung.com>

>>>>> wrote:

>>>>>> Is ignoring dev_pm_domain_attach() return value a solution for you?

>>>>> That's probably better than nothing, but I wonder if it in practice

>>>>> will have any effect?

>>>> If the PM domain is down, then trying to tread the ID is likely to oops

>>>> the kernel.

>>> In my case kernel simply hangs (no single message, even when earlyprintk is

>>> enabled)

>>> instead of oopsing, that's why I've submitted this patch.

>>>

>> Okay.

>>

>> I suggest we go ahead and try that approach (ignoring the return

>> value). It's "quick fix", but the easiest way forward to solve the

>> problem.

>>

>> My only concern is if such change impacts the boot time. We should do

>> some tests to see if the change is negligible, if not we should

>> probably think of something else (like keeping the PM domain powered

>> until late_init).

>>

>> Russell, what do you think?

> That's the approach I took for coresight on Juno - simply initialise

> the power domain very early on and keep it on for things like amba

> probing to be successful.  When the boot process is done unused genPDs

> are switched off automatically.  That way we don't need to add extra

> code to amba.


Frankly I don't get the point that 'you don't need to add extra code to 
amba'.
Turning power domains on very early on boot is rationale, but then 
gen_pd core
often turns them off and on, depending on the sequence of the registered 
devices
and their runtime pm status. Amba bus should support proper registration of
devices in both cases, when power domain for the registered device is 
turn on
and off.

In my opinion it should be a gen_pd core responsibility to keep power domain
turned on until late init. Right now it turns it off when all devices, which
have been registered so far, have been turned off. Usually a moment later
another device get registered to the given domain and this requires the 
domain
to get enabled again. This problem can be observed only when there are more
than on device in a power domain.

Best regards
-- 
Marek Szyprowski, PhD
Samsung R&D Institute Poland

--
To unsubscribe from this list: send the line "unsubscribe linux-samsung-soc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/drivers/amba/bus.c b/drivers/amba/bus.c
index f009936..25715cb 100644
--- a/drivers/amba/bus.c
+++ b/drivers/amba/bus.c
@@ -373,6 +373,12 @@  int amba_device_add(struct amba_device *dev, struct resource *parent)
 		goto err_release;
 	}
 
+	ret = dev_pm_domain_attach(&dev->dev, true);
+	if (ret) {
+		iounmap(tmp);
+		goto err_release;
+	}
+
 	ret = amba_get_enable_pclk(dev);
 	if (ret == 0) {
 		u32 pid, cid;
@@ -398,6 +404,7 @@  int amba_device_add(struct amba_device *dev, struct resource *parent)
 	}
 
 	iounmap(tmp);
+	dev_pm_domain_detach(&dev->dev, true);
 
 	if (ret)
 		goto err_release;