Skip to contentSkip to conversation

pmdomain: imx8m-blk-ctrl: fix suspend/resume order

PATCH In review Vitor Soares 6 participants 8 messages Apr 18, 2024 – Jun 25, 2024

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 pmdomain: imx8m-blk-ctrl: fix suspend/resume order

From: Vitor Soares <[email protected]>

During the probe, the genpd power_dev is added to the dpm_list after blk_ctrl due to its parent/child relationship. Making the blk_ctrl suspend after and resume before the genpd power_dev.

As a consequence, the system hangs when resuming the VPU due to the power domain dependency.

To ensure the proper suspend/resume order, add a device link betweem blk_ctrl and genpd power_dev. It guarantees genpd power_dev is suspended after and resumed before blk-ctrl.

---

This is a new patch, but is a follow-up of:
https://lore.kernel.org/all/[email protected]/

As suggested by Lucas, we are addressing this PM issue in the imx8m-blk-ctrl driver instead of in the imx8mm.dtsi.

 drivers/pmdomain/imx/imx8m-blk-ctrl.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/pmdomain/imx/imx8m-blk-ctrl.c +16 -0
index ca942d7929c2..cd0d2296080d 100644
@@ -283,6 +283,20 @@ static int imx8m_blk_ctrl_probe(struct platform_device *pdev)
 			goto cleanup_pds;
 		}
 
+		/*
+		 * Enforce suspend/resume ordering by making genpd power_dev a
+		 * provider of blk-ctrl. Genpd power_dev is suspended after and
+		 * resumed before blk-ctrl.
+		 */
+		if (!device_link_add(dev, domain->power_dev, DL_FLAG_STATELESS)) {
+			ret = -EINVAL;
+			dev_err_probe(dev, ret,
+				      "failed to link to %s\n", data->name);
+			pm_genpd_remove(&domain->genpd);
+			dev_pm_domain_detach(domain->power_dev, true);
+			goto cleanup_pds;
+		}
+
 		/*
 		 * We use runtime PM to trigger power on/off of the upstream GPC
 		 * domain, as a strict hierarchical parent/child power domain
@@ -324,6 +338,7 @@ static int imx8m_blk_ctrl_probe(struct platform_device *pdev)
 	of_genpd_del_provider(dev->of_node);
 cleanup_pds:
 	for (i--; i >= 0; i--) {
+		device_link_remove(dev, bc->domains[i].power_dev);
 		pm_genpd_remove(&bc->domains[i].genpd);
 		dev_pm_domain_detach(bc->domains[i].power_dev, true);
 	}
@@ -343,6 +358,7 @@ static void imx8m_blk_ctrl_remove(struct platform_device *pdev)
 	for (i = 0; bc->onecell_data.num_domains; i++) {
 		struct imx8m_blk_ctrl_domain *domain = &bc->domains[i];
 
+		device_link_remove(&pdev->dev, domain->power_dev);
 		pm_genpd_remove(&domain->genpd);
 		dev_pm_domain_detach(domain->power_dev, true);
 	}
Signature · 1 line
2.34.1
#2 lore raw

Hi Markus,

Quoted 5 lines · Markus Elfring
Quoted 2 lines

                                        …, add a device link betweem blk_ctrl and genpd power_dev. …

I hope that a typo will be avoided in this change description for the final commit.

Thanks for you feedback.
I will fix it next version.

Best regards,
Vitor Soares

Quoted 2 lines

Regards,
Markus

#3 lore raw
In reply to Vitor Soares
Quoted 6 lines · Vitor Soares

From: Vitor Soares <[email protected]>

During the probe, the genpd power_dev is added to the dpm_list after blk_ctrl due to its parent/child relationship. Making the blk_ctrl suspend after and resume before the genpd power_dev.

Quoted 7 lines

As a consequence, the system hangs when resuming the VPU due to the power domain dependency.

To ensure the proper suspend/resume order, add a device link betweem blk_ctrl and genpd power_dev. It guarantees genpd power_dev is suspended after and resumed before blk-ctrl.

Cc Adam Ford

Francesco

#4 lore raw
Quoted 11 lines · Francesco Dolcini
Quoted 9 lines · Vitor Soares

From: Vitor Soares <[email protected]>

During the probe, the genpd power_dev is added to the dpm_list after blk_ctrl due to its parent/child relationship. Making the blk_ctrl suspend after and resume before the genpd power_dev.

As a consequence, the system hangs when resuming the VPU due to the power domain dependency.

Quoted 6 lines
Quoted 4 lines

To ensure the proper suspend/resume order, add a device link betweem blk_ctrl and genpd power_dev. It guarantees genpd power_dev is suspended after and resumed before blk-ctrl.

Cc Adam Ford

Thanks! This is what I was thinking from the other thread. I just didn't see this one.

adam

Quoted 3 lines

Francesco

Reviewed-by on v1 1/1
#5 lore raw
In reply to Vitor Soares

Am Donnerstag, dem 18.04.2024 um 16:51 +0100 schrieb Vitor Soares:

Quoted 13 lines

From: Vitor Soares <[email protected]>

During the probe, the genpd power_dev is added to the dpm_list after blk_ctrl due to its parent/child relationship. Making the blk_ctrl suspend after and resume before the genpd power_dev.

As a consequence, the system hangs when resuming the VPU due to the power domain dependency.

To ensure the proper suspend/resume order, add a device link betweem blk_ctrl and genpd power_dev. It guarantees genpd power_dev is suspended after and resumed before blk-ctrl.

Quoted 6 lines
Quoted 52 lines

---

This is a new patch, but is a follow-up of:
https://lore.kernel.org/all/[email protected]/

As suggested by Lucas, we are addressing this PM issue in the imx8m-blk-ctrl driver instead of in the imx8mm.dtsi.

 drivers/pmdomain/imx/imx8m-blk-ctrl.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/pmdomain/imx/imx8m-blk-ctrl.c +16 -0
index ca942d7929c2..cd0d2296080d 100644
@@ -283,6 +283,20 @@ static int imx8m_blk_ctrl_probe(struct platform_device *pdev)
 			goto cleanup_pds;
 		}
 
+		/*
+		 * Enforce suspend/resume ordering by making genpd power_dev a
+		 * provider of blk-ctrl. Genpd power_dev is suspended after and
+		 * resumed before blk-ctrl.
+		 */
+		if (!device_link_add(dev, domain->power_dev, DL_FLAG_STATELESS)) {
+			ret = -EINVAL;
+			dev_err_probe(dev, ret,
+				      "failed to link to %s\n", data->name);
+			pm_genpd_remove(&domain->genpd);
+			dev_pm_domain_detach(domain->power_dev, true);
+			goto cleanup_pds;
+		}
+
 		/*
 		 * We use runtime PM to trigger power on/off of the upstream GPC
 		 * domain, as a strict hierarchical parent/child power domain
@@ -324,6 +338,7 @@ static int imx8m_blk_ctrl_probe(struct platform_device *pdev)
 	of_genpd_del_provider(dev->of_node);
 cleanup_pds:
 	for (i--; i >= 0; i--) {
+		device_link_remove(dev, bc->domains[i].power_dev);
 		pm_genpd_remove(&bc->domains[i].genpd);
 		dev_pm_domain_detach(bc->domains[i].power_dev, true);
 	}
@@ -343,6 +358,7 @@ static void imx8m_blk_ctrl_remove(struct platform_device *pdev)
 	for (i = 0; bc->onecell_data.num_domains; i++) {
 		struct imx8m_blk_ctrl_domain *domain = &bc->domains[i];
 
+		device_link_remove(&pdev->dev, domain->power_dev);
 		pm_genpd_remove(&domain->genpd);
 		dev_pm_domain_detach(domain->power_dev, true);
 	}
#6 lore raw
In reply to Vitor Soares
Quoted 7 lines · Vitor Soares

From: Vitor Soares <[email protected]>

During the probe, the genpd power_dev is added to the dpm_list after blk_ctrl due to its parent/child relationship. Making the blk_ctrl suspend after and resume before the genpd power_dev.

Quoted 6 lines

As a consequence, the system hangs when resuming the VPU due to the power domain dependency.

To ensure the proper suspend/resume order, add a device link betweem blk_ctrl and genpd power_dev. It guarantees genpd power_dev is suspended after and resumed before blk-ctrl.

Before discussing $subject patch, would you mind explaining to me why imx8m-blk-ctrl needs to use the ->suspend() callback at all?

Looking closer at that code (imx8m_blk_ctrl_suspend()), it calls pm_runtime_get_sync() for devices to power on "everything". Why isn't that managed by the consumer drivers (on a case by case basis) that are managing the devices that are attached to the genpds instead?

Kind regards
Uffe

Quoted 62 lines

---

This is a new patch, but is a follow-up of:
https://lore.kernel.org/all/[email protected]/

As suggested by Lucas, we are addressing this PM issue in the imx8m-blk-ctrl driver instead of in the imx8mm.dtsi.

 drivers/pmdomain/imx/imx8m-blk-ctrl.c | 16 ++++++++++++++++
 1 file changed, 16 insertions(+)
drivers/pmdomain/imx/imx8m-blk-ctrl.c +16 -1
index ca942d7929c2..cd0d2296080d 100644
@@ -283,6 +283,20 @@ static int imx8m_blk_ctrl_probe(struct platform_device *pdev)
                        goto cleanup_pds;
                }
 
+               /*
+                * Enforce suspend/resume ordering by making genpd power_dev a
+                * provider of blk-ctrl. Genpd power_dev is suspended after and
+                * resumed before blk-ctrl.
+                */
+               if (!device_link_add(dev, domain->power_dev, DL_FLAG_STATELESS)) {
+                       ret = -EINVAL;
+                       dev_err_probe(dev, ret,
+                                     "failed to link to %s\n", data->name);
+                       pm_genpd_remove(&domain->genpd);
+                       dev_pm_domain_detach(domain->power_dev, true);
+                       goto cleanup_pds;
+               }
+
                /*
                 * We use runtime PM to trigger power on/off of the upstream GPC
                 * domain, as a strict hierarchical parent/child power domain
@@ -324,6 +338,7 @@ static int imx8m_blk_ctrl_probe(struct platform_device *pdev)
        of_genpd_del_provider(dev->of_node);
 cleanup_pds:
        for (i--; i >= 0; i--) {
+               device_link_remove(dev, bc->domains[i].power_dev);
                pm_genpd_remove(&bc->domains[i].genpd);
                dev_pm_domain_detach(bc->domains[i].power_dev, true);
        }
@@ -343,6 +358,7 @@ static void imx8m_blk_ctrl_remove(struct platform_device *pdev)
        for (i = 0; bc->onecell_data.num_domains; i++) {
                struct imx8m_blk_ctrl_domain *domain = &bc->domains[i];
 
+               device_link_remove(&pdev->dev, domain->power_dev);
                pm_genpd_remove(&domain->genpd);
                dev_pm_domain_detach(domain->power_dev, true);
        }
--

2.34.1

#7 lore raw
Quoted 18 lines · Ulf Hansson
Quoted 7 lines · Vitor Soares

From: Vitor Soares <[email protected]>

During the probe, the genpd power_dev is added to the dpm_list after blk_ctrl due to its parent/child relationship. Making the blk_ctrl suspend after and resume before the genpd power_dev.

Quoted 6 lines

As a consequence, the system hangs when resuming the VPU due to the power domain dependency.

To ensure the proper suspend/resume order, add a device link betweem blk_ctrl and genpd power_dev. It guarantees genpd power_dev is suspended after and resumed before blk-ctrl.

Before discussing $subject patch, would you mind explaining to me why imx8m-blk-ctrl needs to use the ->suspend() callback at all?

Quoted 8 lines

Looking closer at that code (imx8m_blk_ctrl_suspend()), it calls pm_runtime_get_sync() for devices to power on "everything". Why isn't that managed by the consumer drivers (on a case by case basis) that are managing the devices that are attached to the genpds instead?

Kind regards
Uffe

Hi Lucas, you know the driver architecture better than I do. Would you mind addressing Ulf concerns?

In addition to that, I would say that given that this is a bug fix, it would be nice to have it merged. Having a buggy driver in the kernel is not solving this kind of architectural question.

Best regards,
Vitor Soares

Quoted 72 lines
Quoted 72 lines

Cc: <[email protected]>
Closes:
https://lore.kernel.org/all/[email protected]/

Fixes 2684ac05a8c4 ("soc: imx: add i.MX8M blk-ctrl driver")

---

This is a new patch, but is a follow-up of:
https://lore.kernel.org/all/[email protected]/

As suggested by Lucas, we are addressing this PM issue in the imx8m-blk-ctrl driver instead of in the imx8mm.dtsi.

 drivers/pmdomain/imx/imx8m-blk-ctrl.c | 16 ++++++++++++++++

 1 file changed, 16 insertions(+)
drivers/pmdomain/imx/imx8m-blk-ctrl.c +2 -0
index ca942d7929c2..cd0d2296080d 100644
@@ -283,6 +283,20 @@ static int imx8m_blk_ctrl_probe(struct platform_device
 *pdev)
                         goto cleanup_pds;
                 }
 
+               /*
+                * Enforce suspend/resume ordering by making genpd power_dev

a
+                * provider of blk-ctrl. Genpd power_dev is suspended after and
+                * resumed before blk-ctrl.
+                */
+               if (!device_link_add(dev, domain->power_dev, DL_FLAG_STATELESS)) {
+                       ret = -EINVAL;
+                       dev_err_probe(dev, ret,
+                                     "failed to link to %s\n", data->name); +                       pm_genpd_remove(&domain->genpd);
+                       dev_pm_domain_detach(domain->power_dev, true); +                       goto cleanup_pds;
+               }
+
                /*
                 * We use runtime PM to trigger power on/off of the upstream GPC
                 * domain, as a strict hierarchical parent/child power domain

drivers/pmdomain/imx/imx8m-blk-ctrl.c +0 -0
@@ -324,6 +338,7 @@ static int imx8m_blk_ctrl_probe(struct platform_device
 *pdev)
         of_genpd_del_provider(dev->of_node);

 cleanup_pds:
        for (i--; i >= 0; i--) {
+               device_link_remove(dev, bc->domains[i].power_dev); pm_genpd_remove(&bc->domains[i].genpd); dev_pm_domain_detach(bc->domains[i].power_dev, true); }

drivers/pmdomain/imx/imx8m-blk-ctrl.c +1 -0
@@ -343,6 +358,7 @@ static void imx8m_blk_ctrl_remove(struct platform_device
 *pdev)
         for (i = 0; bc->onecell_data.num_domains; i++) {
                 struct imx8m_blk_ctrl_domain *domain = &bc->domains[i];
 
+               device_link_remove(&pdev->dev, domain->power_dev);
                 pm_genpd_remove(&domain->genpd);
                 dev_pm_domain_detach(domain->power_dev, true);

        }
--
2.34.1