[PATCH] interconnect: Restore sync state by ignoring ipa-virt in provider count

[PATCH] interconnect: Restore sync state by ignoring ipa-virt in provider count

[Stephen Boyd]

Ignore compatible strings for the IPA virt drivers that were removed in
commits 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0
interconnects") and 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0
interconnects") so that the sync state logic can kick in again.
Otherwise all the interconnects in the system will stay pegged at max
speeds because 'providers_count' is always going to be one larger than
the number of drivers that will ever probe on sc7180 or sdx55. This
fixes suspend on sc7180 and sdx55 devices when you don't have a
devicetree patch to remove the ipa-virt compatible node.

Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
Cc: Doug Anderson <dianders@chromium.org>
Cc: Alex Elder <elder@linaro.org>
Cc: Taniya Das <quic_tdas@quicinc.com>
Cc: Mike Tipton <quic_mdtipton@quicinc.com>
Fixes: 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0 interconnects")
Fixes: 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0 interconnects")
Signed-off-by: Stephen Boyd <swboyd@chromium.org>
---
 drivers/interconnect/core.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/drivers/interconnect/core.c b/drivers/interconnect/core.c
index 9050ca1f4285..c52915a58b22 100644
--- a/drivers/interconnect/core.c
+++ b/drivers/interconnect/core.c
@@ -1087,9 +1087,15 @@ static int of_count_icc_providers(struct device_node *np)
 {
 	struct device_node *child;
 	int count = 0;
+	const struct of_device_id ignore_list[] = {
+		{ .compatible = "qcom,sc7180-ipa-virt" },
+		{ .compatible = "qcom,sdx55-ipa-virt" },
+		{}
+	};
 
 	for_each_available_child_of_node(np, child) {
-		if (of_property_read_bool(child, "#interconnect-cells"))
+		if (of_property_read_bool(child, "#interconnect-cells") &&
+		    likely(!of_match_node(ignore_list, child)))
 			count++;
 		count += of_count_icc_providers(child);
 	}

base-commit: 2fb251c265608636fc961b7d38e1a03937e57371
-- 
https://chromeos.dev

Re: [PATCH] interconnect: Restore sync state by ignoring ipa-virt in provider count

[Alex Elder]

On 4/26/22 8:32 PM, Stephen Boyd wrote:
> Ignore compatible strings for the IPA virt drivers that were removed in
> commits 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0
> interconnects") and 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0
> interconnects") so that the sync state logic can kick in again.
> Otherwise all the interconnects in the system will stay pegged at max
> speeds because 'providers_count' is always going to be one larger than
> the number of drivers that will ever probe on sc7180 or sdx55. This
> fixes suspend on sc7180 and sdx55 devices when you don't have a
> devicetree patch to remove the ipa-virt compatible node.
> 
> Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
> Cc: Doug Anderson <dianders@chromium.org>
> Cc: Alex Elder <elder@linaro.org>
> Cc: Taniya Das <quic_tdas@quicinc.com>
> Cc: Mike Tipton <quic_mdtipton@quicinc.com>
> Fixes: 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0 interconnects")
> Fixes: 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0 interconnects")
> Signed-off-by: Stephen Boyd <swboyd@chromium.org>

So of_count_icc_providers() counts the number of
interconnect providers defined in the DTB, regardless
of whether anything in the code supports it.

This seems to be a more general problem, but I
suppose in practice it's not likely to occur.

I think your solution looks fine, but I'm interested
in what Georgi has to say.

Reviewed-by: Alex Elder <elder@linaro.org>


> ---
>   drivers/interconnect/core.c | 8 +++++++-
>   1 file changed, 7 insertions(+), 1 deletion(-)
> 
> diff --git a/drivers/interconnect/core.c b/drivers/interconnect/core.c
> index 9050ca1f4285..c52915a58b22 100644
> --- a/drivers/interconnect/core.c
> +++ b/drivers/interconnect/core.c
> @@ -1087,9 +1087,15 @@ static int of_count_icc_providers(struct device_node *np)
>   {
>   	struct device_node *child;
>   	int count = 0;
> +	const struct of_device_id ignore_list[] = {
> +		{ .compatible = "qcom,sc7180-ipa-virt" },
> +		{ .compatible = "qcom,sdx55-ipa-virt" },
> +		{}
> +	};
>   
>   	for_each_available_child_of_node(np, child) {
> -		if (of_property_read_bool(child, "#interconnect-cells"))
> +		if (of_property_read_bool(child, "#interconnect-cells") &&
> +		    likely(!of_match_node(ignore_list, child)))
>   			count++;
>   		count += of_count_icc_providers(child);
>   	}
> 
> base-commit: 2fb251c265608636fc961b7d38e1a03937e57371

Re: [PATCH] interconnect: Restore sync state by ignoring ipa-virt in provider count

[Doug Anderson]

Hi,

On Tue, Apr 26, 2022 at 6:32 PM Stephen Boyd <swboyd@chromium.org> wrote:
>
> Ignore compatible strings for the IPA virt drivers that were removed in
> commits 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0
> interconnects") and 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0
> interconnects") so that the sync state logic can kick in again.
> Otherwise all the interconnects in the system will stay pegged at max
> speeds because 'providers_count' is always going to be one larger than
> the number of drivers that will ever probe on sc7180 or sdx55. This
> fixes suspend on sc7180 and sdx55 devices when you don't have a
> devicetree patch to remove the ipa-virt compatible node.
>
> Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
> Cc: Doug Anderson <dianders@chromium.org>
> Cc: Alex Elder <elder@linaro.org>
> Cc: Taniya Das <quic_tdas@quicinc.com>
> Cc: Mike Tipton <quic_mdtipton@quicinc.com>
> Fixes: 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0 interconnects")
> Fixes: 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0 interconnects")
> Signed-off-by: Stephen Boyd <swboyd@chromium.org>
> ---
>  drivers/interconnect/core.c | 8 +++++++-
>  1 file changed, 7 insertions(+), 1 deletion(-)

Ick, but seems like an expedient fix to get us back to working while
we try to come up with something better.

Reviewed-by: Douglas Anderson <dianders@chromium.org>

Re: [PATCH] interconnect: Restore sync state by ignoring ipa-virt in provider count

[Georgi Djakov]

On 27.04.22 15:00, Alex Elder wrote:
> On 4/26/22 8:32 PM, Stephen Boyd wrote:
>> Ignore compatible strings for the IPA virt drivers that were removed in
>> commits 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0
>> interconnects") and 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0
>> interconnects") so that the sync state logic can kick in again.
>> Otherwise all the interconnects in the system will stay pegged at max
>> speeds because 'providers_count' is always going to be one larger than
>> the number of drivers that will ever probe on sc7180 or sdx55. This
>> fixes suspend on sc7180 and sdx55 devices when you don't have a
>> devicetree patch to remove the ipa-virt compatible node.
>>
>> Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
>> Cc: Doug Anderson <dianders@chromium.org>
>> Cc: Alex Elder <elder@linaro.org>
>> Cc: Taniya Das <quic_tdas@quicinc.com>
>> Cc: Mike Tipton <quic_mdtipton@quicinc.com>
>> Fixes: 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0 interconnects")
>> Fixes: 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0 interconnects")
>> Signed-off-by: Stephen Boyd <swboyd@chromium.org>
> 
> So of_count_icc_providers() counts the number of
> interconnect providers defined in the DTB, regardless
> of whether anything in the code supports it.

Yes, that's the case currently. There could be multiple provider drivers
in different modules, and the modules may be loaded even not during boot,
but later. So we rely on DT.

Thanks,
Georgi

> This seems to be a more general problem, but I
> suppose in practice it's not likely to occur.
> 
> I think your solution looks fine, but I'm interested
> in what Georgi has to say.
> 
> Reviewed-by: Alex Elder <elder@linaro.org>
> 
> 
>> ---
>>   drivers/interconnect/core.c | 8 +++++++-
>>   1 file changed, 7 insertions(+), 1 deletion(-)
>>
>> diff --git a/drivers/interconnect/core.c b/drivers/interconnect/core.c
>> index 9050ca1f4285..c52915a58b22 100644
>> --- a/drivers/interconnect/core.c
>> +++ b/drivers/interconnect/core.c
>> @@ -1087,9 +1087,15 @@ static int of_count_icc_providers(struct device_node *np)
>>   {
>>       struct device_node *child;
>>       int count = 0;
>> +    const struct of_device_id ignore_list[] = {
>> +        { .compatible = "qcom,sc7180-ipa-virt" },
>> +        { .compatible = "qcom,sdx55-ipa-virt" },
>> +        {}
>> +    };
>>       for_each_available_child_of_node(np, child) {
>> -        if (of_property_read_bool(child, "#interconnect-cells"))
>> +        if (of_property_read_bool(child, "#interconnect-cells") &&
>> +            likely(!of_match_node(ignore_list, child)))
>>               count++;
>>           count += of_count_icc_providers(child);
>>       }
>>
>> base-commit: 2fb251c265608636fc961b7d38e1a03937e57371
> 

Re: [PATCH] interconnect: Restore sync state by ignoring ipa-virt in provider count

[Alex Elder]

On 4/27/22 3:52 PM, Georgi Djakov wrote:
> On 27.04.22 15:00, Alex Elder wrote:
>> On 4/26/22 8:32 PM, Stephen Boyd wrote:
>>> Ignore compatible strings for the IPA virt drivers that were removed in
>>> commits 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0
>>> interconnects") and 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0
>>> interconnects") so that the sync state logic can kick in again.
>>> Otherwise all the interconnects in the system will stay pegged at max
>>> speeds because 'providers_count' is always going to be one larger than
>>> the number of drivers that will ever probe on sc7180 or sdx55. This
>>> fixes suspend on sc7180 and sdx55 devices when you don't have a
>>> devicetree patch to remove the ipa-virt compatible node.
>>>
>>> Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
>>> Cc: Doug Anderson <dianders@chromium.org>
>>> Cc: Alex Elder <elder@linaro.org>
>>> Cc: Taniya Das <quic_tdas@quicinc.com>
>>> Cc: Mike Tipton <quic_mdtipton@quicinc.com>
>>> Fixes: 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0 
>>> interconnects")
>>> Fixes: 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0 
>>> interconnects")
>>> Signed-off-by: Stephen Boyd <swboyd@chromium.org>
>>
>> So of_count_icc_providers() counts the number of
>> interconnect providers defined in the DTB, regardless
>> of whether anything in the code supports it.
> 
> Yes, that's the case currently. There could be multiple provider drivers
> in different modules, and the modules may be loaded even not during boot,
> but later. So we rely on DT.

Georgi, do you have any other ideas about how to improve this?

What you say makes sense--you figure out the total number of
providers at init time based on what's defined in DT.

But when all is said and done, what if even *one* of those
has no provider driver?  Isn't that what causes the problem
here, that icc_sync_state assumes all providers specified in
DT will eventually have a driver that calls icc_sync_state()?

(I don't claim to know the interconnect code in detail, so
I might not be quite understanding how this works.)

					-Alex

> Thanks,
> Georgi
> 
>> This seems to be a more general problem, but I
>> suppose in practice it's not likely to occur.
>>
>> I think your solution looks fine, but I'm interested
>> in what Georgi has to say.
>>
>> Reviewed-by: Alex Elder <elder@linaro.org>
>>
>>
>>> ---
>>>   drivers/interconnect/core.c | 8 +++++++-
>>>   1 file changed, 7 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/interconnect/core.c b/drivers/interconnect/core.c
>>> index 9050ca1f4285..c52915a58b22 100644
>>> --- a/drivers/interconnect/core.c
>>> +++ b/drivers/interconnect/core.c
>>> @@ -1087,9 +1087,15 @@ static int of_count_icc_providers(struct 
>>> device_node *np)
>>>   {
>>>       struct device_node *child;
>>>       int count = 0;
>>> +    const struct of_device_id ignore_list[] = {
>>> +        { .compatible = "qcom,sc7180-ipa-virt" },
>>> +        { .compatible = "qcom,sdx55-ipa-virt" },
>>> +        {}
>>> +    };
>>>       for_each_available_child_of_node(np, child) {
>>> -        if (of_property_read_bool(child, "#interconnect-cells"))
>>> +        if (of_property_read_bool(child, "#interconnect-cells") &&
>>> +            likely(!of_match_node(ignore_list, child)))
>>>               count++;
>>>           count += of_count_icc_providers(child);
>>>       }
>>>
>>> base-commit: 2fb251c265608636fc961b7d38e1a03937e57371
>>
> 

Re: [PATCH] interconnect: Restore sync state by ignoring ipa-virt in provider count

[Nathan Chancellor]

Hi Stephen,

On Tue, Apr 26, 2022 at 06:32:26PM -0700, Stephen Boyd wrote:
> Ignore compatible strings for the IPA virt drivers that were removed in
> commits 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0
> interconnects") and 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0
> interconnects") so that the sync state logic can kick in again.
> Otherwise all the interconnects in the system will stay pegged at max
> speeds because 'providers_count' is always going to be one larger than
> the number of drivers that will ever probe on sc7180 or sdx55. This
> fixes suspend on sc7180 and sdx55 devices when you don't have a
> devicetree patch to remove the ipa-virt compatible node.
> 
> Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
> Cc: Doug Anderson <dianders@chromium.org>
> Cc: Alex Elder <elder@linaro.org>
> Cc: Taniya Das <quic_tdas@quicinc.com>
> Cc: Mike Tipton <quic_mdtipton@quicinc.com>
> Fixes: 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0 interconnects")
> Fixes: 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0 interconnects")
> Signed-off-by: Stephen Boyd <swboyd@chromium.org>

This patch as commit ad3cc2f05fda ("interconnect: Restore sync state by
ignoring ipa-virt in provider count") in -next causes the following
build warning when CONFIG_OF is disabled because of_match_node is just
NULL:

drivers/interconnect/core.c:1090:28: warning: unused variable 'ignore_list' [-Wunused-variable]
        const struct of_device_id ignore_list[] = {
                                  ^
1 warning generated.

Should it just be marked as __maybe_unused or is there a different fix
that would be more appropriate?

Cheers,
Nathan

Re: [PATCH] interconnect: Restore sync state by ignoring ipa-virt in provider count

[Stephen Boyd]

Quoting Nathan Chancellor (2022-05-02 08:18:24)
> Hi Stephen,
>
> On Tue, Apr 26, 2022 at 06:32:26PM -0700, Stephen Boyd wrote:
> > Ignore compatible strings for the IPA virt drivers that were removed in
> > commits 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0
> > interconnects") and 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0
> > interconnects") so that the sync state logic can kick in again.
> > Otherwise all the interconnects in the system will stay pegged at max
> > speeds because 'providers_count' is always going to be one larger than
> > the number of drivers that will ever probe on sc7180 or sdx55. This
> > fixes suspend on sc7180 and sdx55 devices when you don't have a
> > devicetree patch to remove the ipa-virt compatible node.
> >
> > Cc: Bjorn Andersson <bjorn.andersson@linaro.org>
> > Cc: Doug Anderson <dianders@chromium.org>
> > Cc: Alex Elder <elder@linaro.org>
> > Cc: Taniya Das <quic_tdas@quicinc.com>
> > Cc: Mike Tipton <quic_mdtipton@quicinc.com>
> > Fixes: 2fb251c26560 ("interconnect: qcom: sdx55: Drop IP0 interconnects")
> > Fixes: 2f3724930eb4 ("interconnect: qcom: sc7180: Drop IP0 interconnects")
> > Signed-off-by: Stephen Boyd <swboyd@chromium.org>
>
> This patch as commit ad3cc2f05fda ("interconnect: Restore sync state by
> ignoring ipa-virt in provider count") in -next causes the following
> build warning when CONFIG_OF is disabled because of_match_node is just
> NULL:
>
> drivers/interconnect/core.c:1090:28: warning: unused variable 'ignore_list' [-Wunused-variable]
>         const struct of_device_id ignore_list[] = {
>                                   ^
> 1 warning generated.
>
> Should it just be marked as __maybe_unused or is there a different fix
> that would be more appropriate?

Yeah I think __maybe_unused would work. The other approach would be to
make of_match_node() a static inline that returns false with CONFIG_OF=n
but that's probably a super large change. I hope the compiler would be
able to optimize that whole thing away because the arguments are const
pointers but I don't know.