All of lore.kernel.org
 help / color / mirror / Atom feed
From: sashiko-bot@kernel.org
To: "Stefan Dösinger" <stefandoesinger@gmail.com>
Cc: devicetree@vger.kernel.org, mfd@lists.linux.dev,
	linux-phy@lists.infradead.org, robh@kernel.org,
	conor+dt@kernel.org, vkoul@kernel.org, neil.armstrong@linaro.org,
	olteanv@gmail.com, lee@kernel.org
Subject: Re: [PATCH 3/6] pmdomain: zte: Add a zx297520v3 power domain driver
Date: Fri, 20 Nov 2026 19:51:31 +0000	[thread overview]
Message-ID: <sashiko-outbox-225964@kernel.org> (raw)
In-Reply-To: <20261120-zx29usb-v1-3-b0433258e51b@gmail.com>

Thank you for your contribution! Sashiko AI review found 2 potential
issues to consider:
- [High] pmdomain: zte: child-to-parent device link deadlock in
  zx29_pd_probe()
- [High] pmdomain: zte: wrong supplier device in zx29_pd_attach_dev()
  device link
--

commit 65d36c75a56ee45de3b21629f8434fb30bf2339b
Author: Stefan Dösinger <stefandoesinger@gmail.com>

pmdomain: zte: Add a zx297520v3 power domain driver

This commit adds a power domain driver for the ZTE zx297520v3 SoC to
control the power, reset, and bus isolation registers for the USB
controller.

> diff --git a/drivers/pmdomain/zte/zx297520v3-pm-domain.c b/drivers/pmdomain/zte/zx297520v3-pm-domain.c
> new file mode 100644
> index 0000000000000..3df4cd5f1b229
> --- /dev/null
> +++ b/drivers/pmdomain/zte/zx297520v3-pm-domain.c

[ ... ]

> +static int zx29_pd_attach_dev(struct generic_pm_domain *domain, struct device *dev)
> +{
> +	/* The DT link to the consumer targets the (MFD) parent, which doesn't prevent unbinding the
> +	 * this driver or removing the module. Either would leave the consumer with a dangling
> +	 * reference.
> +	 */
> +	if (!device_link_add(dev, &domain->dev, DL_FLAG_AUTOREMOVE_CONSUMER))
> +		return -ENOMEM;
> +
> +	return 0;
> +}

[Severity: High]
Should the device link supplier here be the zx297520v3-socsys-power
platform device (&pdev->dev) rather than &domain->dev?

&domain->dev is the internal genpd provider device deleted by
of_genpd_del_provider().  Because no consumer link targets &pdev->dev,
unbinding this driver starts devres teardown while consumers are still
bound and only unbinds them inside zx29_pd_release_provider() ->
of_genpd_del_provider() -> device_del(&gpd->dev) under gpd_list_lock and
of_genpd_mutex.

That inverts the normal lock order against a probing consumer that already
holds device_lock(consumer) while calling dev_pm_domain_attach() ->
genpd_dev_pm_attach():

CPU0                              CPU1
----                              ----
lock(&consumer->mutex)
                                  lock(&gpd_list_lock)
                                  lock(&of_genpd_mutex)
lock(&gpd_list_lock)
                                  lock(&consumer->mutex)

 *** DEADLOCK ***

CPU0: really_probe(consumer) -> ... -> genpd_get_from_provider()
CPU1: zx29_pd_release_provider() -> of_genpd_del_provider() ->
      device_del(&gpd->dev) -> ... -> device_links_unbind_consumers()

In addition, if a consumer attaches via genpd_dev_pm_attach_by_id() or
genpd_dev_pm_attach_by_name(), dev is a driverless virtual device on
genpd_bus_type, so its link to &domain->dev stays in DL_STATE_AVAILABLE:
device_links_unbind_consumers() never detaches it,
device_links_driver_cleanup() hits WARN_ON(), and pm_genpd_remove() fails
with -EBUSY right before devres frees zx29_pd, leaving a freed
generic_pm_domain on gpd_list.

[ ... ]

> +static int zx29_pd_probe(struct platform_device *pdev)
> +{

[ ... ]

> +	res = of_genpd_add_provider_onecell(of_node, zx29_pd_data);
> +	if (res)
> +		return dev_err_probe(dev, res, "Cannot register power domains\n");
> +
> +	device_link_add(dev, dev->parent, DL_FLAG_AUTOREMOVE_CONSUMER);
> +
> +	dev_dbg(dev, "zx297520v3 power domain initialized.\n");
> +
> +	return devm_add_action_or_reset(dev, zx29_pd_release_provider, of_node);
> +}

[Severity: High]
Can adding a managed device link from dev to dev->parent deadlock when
zx297520v3_crm_probe() registers its MFD children after late_initcall
(for example on module load or driver rebind)?

zx297520v3_crm_probe() calls devm_mfd_add_devices() while really_probe()
holds device_lock(dev->parent).  Because of_genpd_add_provider_onecell()
sets genpd_sync_state on dev->parent, finishing dev's probe queues
sync_state on dev->parent and tries to lock dev->parent again on the same
thread:

really_probe(dev->parent)
  device_lock(dev->parent)
  zx297520v3_crm_probe()
    devm_mfd_add_devices()
      ...
        really_probe(dev)
          zx29_pd_probe()
            of_genpd_add_provider_onecell(dev->parent->of_node, ...)
            device_link_add(dev, dev->parent, DL_FLAG_AUTOREMOVE_CONSUMER)
          driver_bound(dev)
            device_links_driver_bound(dev)
              device_links_flush_sync_list(..., dev)
                device_lock(dev->parent)  <-- self-deadlock

Since dev is a child device registered via devm_mfd_add_devices() on
dev->parent, the driver core already orders suspend/resume and removes dev
when dev->parent unbinds without this link.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261120-zx29usb-v1-0-b0433258e51b@gmail.com?part=3

WARNING: multiple messages have this Message-ID (diff)
From: sashiko-bot@kernel.org
To: "Stefan Dösinger" <stefandoesinger@gmail.com>
Cc: devicetree@vger.kernel.org, mfd@lists.linux.dev,
	linux-phy@lists.infradead.org, robh@kernel.org,
	conor+dt@kernel.org, vkoul@kernel.org, neil.armstrong@linaro.org,
	olteanv@gmail.com, lee@kernel.org
Subject: Re: [PATCH 3/6] pmdomain: zte: Add a zx297520v3 power domain driver
Date: Fri, 20 Nov 2026 19:51:31 +0000	[thread overview]
Message-ID: <sashiko-outbox-225964@kernel.org> (raw)
In-Reply-To: <20261120-zx29usb-v1-3-b0433258e51b@gmail.com>

Thank you for your contribution! Sashiko AI review found 2 potential
issues to consider:
- [High] pmdomain: zte: child-to-parent device link deadlock in
  zx29_pd_probe()
- [High] pmdomain: zte: wrong supplier device in zx29_pd_attach_dev()
  device link
--

commit 65d36c75a56ee45de3b21629f8434fb30bf2339b
Author: Stefan Dösinger <stefandoesinger@gmail.com>

pmdomain: zte: Add a zx297520v3 power domain driver

This commit adds a power domain driver for the ZTE zx297520v3 SoC to
control the power, reset, and bus isolation registers for the USB
controller.

> diff --git a/drivers/pmdomain/zte/zx297520v3-pm-domain.c b/drivers/pmdomain/zte/zx297520v3-pm-domain.c
> new file mode 100644
> index 0000000000000..3df4cd5f1b229
> --- /dev/null
> +++ b/drivers/pmdomain/zte/zx297520v3-pm-domain.c

[ ... ]

> +static int zx29_pd_attach_dev(struct generic_pm_domain *domain, struct device *dev)
> +{
> +	/* The DT link to the consumer targets the (MFD) parent, which doesn't prevent unbinding the
> +	 * this driver or removing the module. Either would leave the consumer with a dangling
> +	 * reference.
> +	 */
> +	if (!device_link_add(dev, &domain->dev, DL_FLAG_AUTOREMOVE_CONSUMER))
> +		return -ENOMEM;
> +
> +	return 0;
> +}

[Severity: High]
Should the device link supplier here be the zx297520v3-socsys-power
platform device (&pdev->dev) rather than &domain->dev?

&domain->dev is the internal genpd provider device deleted by
of_genpd_del_provider().  Because no consumer link targets &pdev->dev,
unbinding this driver starts devres teardown while consumers are still
bound and only unbinds them inside zx29_pd_release_provider() ->
of_genpd_del_provider() -> device_del(&gpd->dev) under gpd_list_lock and
of_genpd_mutex.

That inverts the normal lock order against a probing consumer that already
holds device_lock(consumer) while calling dev_pm_domain_attach() ->
genpd_dev_pm_attach():

CPU0                              CPU1
----                              ----
lock(&consumer->mutex)
                                  lock(&gpd_list_lock)
                                  lock(&of_genpd_mutex)
lock(&gpd_list_lock)
                                  lock(&consumer->mutex)

 *** DEADLOCK ***

CPU0: really_probe(consumer) -> ... -> genpd_get_from_provider()
CPU1: zx29_pd_release_provider() -> of_genpd_del_provider() ->
      device_del(&gpd->dev) -> ... -> device_links_unbind_consumers()

In addition, if a consumer attaches via genpd_dev_pm_attach_by_id() or
genpd_dev_pm_attach_by_name(), dev is a driverless virtual device on
genpd_bus_type, so its link to &domain->dev stays in DL_STATE_AVAILABLE:
device_links_unbind_consumers() never detaches it,
device_links_driver_cleanup() hits WARN_ON(), and pm_genpd_remove() fails
with -EBUSY right before devres frees zx29_pd, leaving a freed
generic_pm_domain on gpd_list.

[ ... ]

> +static int zx29_pd_probe(struct platform_device *pdev)
> +{

[ ... ]

> +	res = of_genpd_add_provider_onecell(of_node, zx29_pd_data);
> +	if (res)
> +		return dev_err_probe(dev, res, "Cannot register power domains\n");
> +
> +	device_link_add(dev, dev->parent, DL_FLAG_AUTOREMOVE_CONSUMER);
> +
> +	dev_dbg(dev, "zx297520v3 power domain initialized.\n");
> +
> +	return devm_add_action_or_reset(dev, zx29_pd_release_provider, of_node);
> +}

[Severity: High]
Can adding a managed device link from dev to dev->parent deadlock when
zx297520v3_crm_probe() registers its MFD children after late_initcall
(for example on module load or driver rebind)?

zx297520v3_crm_probe() calls devm_mfd_add_devices() while really_probe()
holds device_lock(dev->parent).  Because of_genpd_add_provider_onecell()
sets genpd_sync_state on dev->parent, finishing dev's probe queues
sync_state on dev->parent and tries to lock dev->parent again on the same
thread:

really_probe(dev->parent)
  device_lock(dev->parent)
  zx297520v3_crm_probe()
    devm_mfd_add_devices()
      ...
        really_probe(dev)
          zx29_pd_probe()
            of_genpd_add_provider_onecell(dev->parent->of_node, ...)
            device_link_add(dev, dev->parent, DL_FLAG_AUTOREMOVE_CONSUMER)
          driver_bound(dev)
            device_links_driver_bound(dev)
              device_links_flush_sync_list(..., dev)
                device_lock(dev->parent)  <-- self-deadlock

Since dev is a child device registered via devm_mfd_add_devices() on
dev->parent, the driver core already orders suspend/resume and removes dev
when dev->parent unbinds without this link.

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20261120-zx29usb-v1-0-b0433258e51b@gmail.com?part=3

-- 
linux-phy mailing list
linux-phy@lists.infradead.org
https://lists.infradead.org/mailman/listinfo/linux-phy

  reply	other threads:[~2026-11-20 19:51 UTC|newest]

Thread overview: 20+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-20 19:20 [PATCH 0/6] ZTE zx297520v3 USB support Stefan Dösinger
2026-11-20 19:20 ` Stefan Dösinger
2026-11-20 19:20 ` [PATCH 1/6] dt-bindings: soc: zte: Document zx297520v3 socsys controller Stefan Dösinger
2026-11-20 19:20   ` Stefan Dösinger
2026-11-20 19:38   ` sashiko-bot
2026-11-20 19:38     ` sashiko-bot
2026-11-20 19:20 ` [PATCH 2/6] mfd: zx297520v3: Add socsys to the mfd host driver Stefan Dösinger
2026-11-20 19:20   ` Stefan Dösinger
2026-11-20 19:20 ` [PATCH 3/6] pmdomain: zte: Add a zx297520v3 power domain driver Stefan Dösinger
2026-11-20 19:20   ` Stefan Dösinger
2026-11-20 19:51   ` sashiko-bot [this message]
2026-11-20 19:51     ` sashiko-bot
2026-11-20 19:20 ` [PATCH 4/6] phy: zte: Add a zx29 USB PHY driver Stefan Dösinger
2026-11-20 19:20   ` Stefan Dösinger
2026-11-20 19:47   ` sashiko-bot
2026-11-20 19:47     ` sashiko-bot
2026-11-20 19:20 ` [PATCH 5/6] ARM: dts: zte: Declare nodes to enable the board's dwc2 controller Stefan Dösinger
2026-11-20 19:20   ` Stefan Dösinger
2026-11-20 19:20 ` [PATCH 6/6] ARM: dts: zte: Declare the HSIC controller node Stefan Dösinger
2026-11-20 19:20   ` Stefan Dösinger

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=sashiko-outbox-225964@kernel.org \
    --to=sashiko-bot@kernel.org \
    --cc=conor+dt@kernel.org \
    --cc=devicetree@vger.kernel.org \
    --cc=lee@kernel.org \
    --cc=linux-phy@lists.infradead.org \
    --cc=mfd@lists.linux.dev \
    --cc=neil.armstrong@linaro.org \
    --cc=olteanv@gmail.com \
    --cc=robh@kernel.org \
    --cc=sashiko-reviews@lists.linux.dev \
    --cc=stefandoesinger@gmail.com \
    --cc=vkoul@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.