On Tue, Aug 31, 2021 at 9:59 AM Saravana Kannan [off-list ref] wrote:

On Tue, Aug 31, 2021 at 6:56 AM Ulf Hansson [off-list ref] wrote:

quoted

Hi Saravana, Rafael, Stephen,

I have observed an issue with a driver's  ->sync_state() callback that
doesn't get called, even if consumers/supplier device links conditions
should have been satisfied. I have narrowed down the problem, but I am
not sure what is the best solution to fix it, hence I am turning to
you for some discussion/advice.

I have two test platform drivers, one that matches on the
"test,pm-domains-test" compatible string (this driver also has the
->sync_state() callback assigned) and another driver that matches on
"test,runtime-pm-test".

This is the relevant part in my DTS file:

       pm_domain_test {
                compatible = "test,pm-domains-test";

                pdParent: power-domain-parent {
                        #power-domain-cells = <0>;
                };

                pdChild: power-domain-child {
                        #power-domain-cells = <0>;
                        power-domains = <&pdParent>;
                };
        };

        soctest: soctest {
                compatible = "simple-bus";

                rpmtest0 {
                        compatible = "test,runtime-pm-test";
                        power-domains = <&pdParent>;
                };
        };

During boot the fw_devlinks are being created and their corresponding
links. With some debug enabled this shows some of the interesting
parts that are being printed to the log:

[    0.041539] device: 'pm_domain_test': device_add
[    0.041629] OF: Not linking pm_domain_test to pm_domain_test - is descendant
[    0.041718] device: 'soctest': device_add
[    0.041803] OF: Linking rpmtest0 (consumer) to pm_domain_test (supplier)
[    0.041829] device: 'platform:pm_domain_test--platform:soctest': device_add
[    0.041892] platform soctest: Linked as a sync state only consumer
to pm_domain_test
[    0.041957] OF:    create child: /soctest/rpmtest0
[    0.041995] device: 'soctest:rpmtest0': device_add
[    0.042072] device:
'platform:pm_domain_test--platform:soctest:rpmtest0': device_add
[    0.042132] devices_kset: Moving soctest:rpmtest0 to end of list
[    0.042141] platform soctest:rpmtest0: Linked as a consumer to pm_domain_test

The interesting thing here is the "sync state only" link that gets
created. I assume there are good reasons for creating this link, even
if I fail to understand exactly why.

In general there's a good reason for creating these links from parent
devices of the consumer to the supplier. It is documented in the code
under __fw_devlink_link_to_consumers().

/*
 * If consumer device is not available yet, make a "proxy"
 * SYNC_STATE_ONLY link from the consumer's parent device to
 * the supplier device. This is necessary to make sure the
 * supplier doesn't get a sync_state() callback before the real
 * consumer can create a device link to the supplier.
 *
 * This proxy link step is needed to handle the case where the
 * consumer's parent device is added before the supplier.
 */

and under __fw_devlink_link_to_suppliers().

/*
 * Make "proxy" SYNC_STATE_ONLY device links to represent the needs of
 * all the descendants. This proxy link step is needed to handle the
 * case where the supplier is added before the consumer's parent device
 * (@dev).
 */

quoted

In any case, the sync state only link never gets dropped, which I
assume is because there is no driver getting bound for the "soctest"
device (it has only the "simple-bus" compatible).

Yeah, you've identified the problem correctly. I've been thinking
about this possibility (and all the side effects a fix might have). I
can send out a fix for this soon (within a week or so).

quoted

In other words, it doesn't matter that both the rpmtest0 and the
pm_domain_test devices are probed, thus satisfying the
supplier/consumer conditions, the ->sync_state() callback doesn't get
called anyway.

Can you perhaps help to point me in a direction of how to best fix this problem?

I hope you are okay with me sending a fix.

Ulf,

Could you give this a shot please?

https://lore.kernel.org/lkml/20210831224510.703253-1-saravanak@google.com/T/#u (local)

-Saravana