Re: [PATCH v1 1/4] driver core: Rely on supplier driver binding at link creation

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 4 potential issue(s) to consider:
- [High] driver core: AB-BA deadlock in device_link_add and device_release_driver_internal
- [High] driver core: fw_devlink broken by rejecting unbound suppliers in device_link_add
- [Low] driver core: mixed goto cleanup and scoped_guard violates cleanup guidelines
- [Low] driver core: dead code in PM runtime check due to device_is_bound
--

commit 1f69a7873c55c75da1b5f2ed9898c41c813921cc
Author: Alejandro Lucero <alucerop@amd.com>

driver core: Rely on supplier driver binding at link creation

This commit modifies device link creation to rely on supplier driver binding.
It evaluates device_is_bound() for the supplier upfront and conditionally
checks PM initialization based on the DL_FLAG_PM_RUNTIME flag.

> diff --git a/drivers/base/core.c b/drivers/base/core.c
> index 4c0c373998a19..eb6d87e35d76c 100644
> --- a/drivers/base/core.c
> +++ b/drivers/base/core.c
> @@ -834,15 +834,26 @@ struct device_link *device_link_add(struct device *consumer,
>  	device_pm_lock();
>  
>  	/*
> -	 * If the supplier has not been fully registered yet or there is a
> -	 * reverse (non-SYNC_STATE_ONLY) dependency between the consumer and
> -	 * the supplier already in the graph, return NULL. If the link is a
> -	 * SYNC_STATE_ONLY link, we don't check for reverse dependencies
> -	 * because it only affects sync_state() callbacks.
> +	 * If the supplier has not been fully registered yet with a driver
> +	 * return NULL.
>  	 */
> -	if (!device_pm_initialized(supplier)
> -	    || (!(flags & DL_FLAG_SYNC_STATE_ONLY) &&
> -		  device_is_dependent(consumer, supplier))) {
> +	scoped_guard(device, supplier) {

[Severity: High]
Can this lead to an AB-BA deadlock between device_links_write_lock() and
the supplier device lock?

The device_link_add() function holds device_links_write_lock() before
acquiring the supplier's device lock via scoped_guard(device, supplier).

However, during device unbinding, device_release_driver_internal() takes the
device lock via __device_driver_lock() first. It then calls
__device_release_driver() which invokes device_links_driver_cleanup(),
attempting to acquire device_links_write_lock().

Could this result in a lock inversion?

> +		if (!device_is_bound(supplier)) {

[Severity: High]
Does rejecting unbound suppliers break fw_devlink's dependency tracking?

The fw_devlink_create_devlink() routine actively tries to create dormant
links to supplier devices before they are probed or bound to a driver. This
ensures proper probe ordering.

Rejecting these links unconditionally because device_is_bound() requires a
driver to be attached appears to break deferred probing guarantees.

Should we permit links when the supplier is not bound yet?

> +			link = NULL;
> +			goto out;

[Severity: Low]
Does jumping out of a scope-based cleanup block violate the cleanup
guidelines?

The kernel subsystem cleanup guidelines state that the usage of goto and
cleanup helpers should never be mixed in the same function.

Using a goto to jump out of a scoped_guard() block to a manual cleanup label
mixes the two paradigms.

Could this be converted completely to scope-based cleanup, or would it be
better to avoid using scoped_guard() here?

> +		}
> +	}
> +	/*
> +	 * If consumer asks for PM to use the link and the supplier has not
> +	 * PM initialized, or if there is a reverse (non-SYNC_STATE_ONLY)
> +	 * dependency between the consumer and the supplier already in the
> +	 * graph, return NULL. If the link is a SYNC_STATE_ONLY link, we
> +	 * don't check for reverse dependencies because it only affects
> +	 * sync_state() callbacks.
> +	 */
> +	if (((flags & DL_FLAG_PM_RUNTIME) && !device_pm_initialized(supplier)) ||

[Severity: Low]
Is the !device_pm_initialized(supplier) condition reachable anymore?

The device_is_bound() check earlier in the function fundamentally
guarantees that a driver is attached. This is only possible if the
device has already been registered.

Since device_pm_initialized() directly evaluates to device_is_registered(),
!device_pm_initialized(supplier) will evaluate to false under all possible
code paths passing the new bound check.

Is this check now dead code?

> +	    (!(flags & DL_FLAG_SYNC_STATE_ONLY) &&
> +	     device_is_dependent(consumer, supplier))) {
>  		link = NULL;
>  		goto out;
>  	}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260921191239.4249-1-alucerop@amd.com?part=1