Skip to contentSkip to conversation

dmaengine: pch_dma: use generic power management

PATCH In review Vaibhav Gupta 3 participants 6 messages Jul 20, 2020 – Aug 17, 2020 dmaengine

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 dmaengine: pch_dma: use generic power management

Drivers using legacy PM have to manage PCI states and device's PM states themselves. They also need to take care of configuration registers.

With improved and powerful support of generic PM, PCI Core takes care of above mentioned, device-independent, jobs.

This driver makes use of PCI helper functions like pci_save/restore_state(), pci_enable/disable_device(),
and pci_set_power_state() to do required operations. In generic mode, they are no longer needed.

Change function parameter in both .suspend() and .resume() to "struct device*" type. Use dev_get_drvdata() to get drv data.

Compile-tested only.

---
 drivers/dma/pch_dma.c | 35 +++++++++--------------------------
 1 file changed, 9 insertions(+), 26 deletions(-)
drivers/dma/pch_dma.c +9 -26
index a3b0b4c56a19..e93005837e3f 100644
@@ -735,8 +735,7 @@ static irqreturn_t pd_irq(int irq, void *devid)
 	return ret0 | ret2;
 }
 
-#ifdef	CONFIG_PM
-static void pch_dma_save_regs(struct pch_dma *pd)
+static void __maybe_unused pch_dma_save_regs(struct pch_dma *pd)
 {
 	struct pch_dma_chan *pd_chan;
 	struct dma_chan *chan, *_c;
@@ -759,7 +758,7 @@ static void pch_dma_save_regs(struct pch_dma *pd)
 	}
 }
 
-static void pch_dma_restore_regs(struct pch_dma *pd)
+static void __maybe_unused pch_dma_restore_regs(struct pch_dma *pd)
 {
 	struct pch_dma_chan *pd_chan;
 	struct dma_chan *chan, *_c;
@@ -782,40 +781,25 @@ static void pch_dma_restore_regs(struct pch_dma *pd)
 	}
 }
 
-static int pch_dma_suspend(struct pci_dev *pdev, pm_message_t state)
+static int __maybe_unused pch_dma_suspend(struct device *dev)
 {
-	struct pch_dma *pd = pci_get_drvdata(pdev);
+	struct pch_dma *pd = dev_get_drvdata(dev);
 
 	if (pd)
 		pch_dma_save_regs(pd);
 
-	pci_save_state(pdev);
-	pci_disable_device(pdev);
-	pci_set_power_state(pdev, pci_choose_state(pdev, state));
-
 	return 0;
 }
 
-static int pch_dma_resume(struct pci_dev *pdev)
+static int __maybe_unused pch_dma_resume(struct device *dev)
 {
-	struct pch_dma *pd = pci_get_drvdata(pdev);
-	int err;
-
-	pci_set_power_state(pdev, PCI_D0);
-	pci_restore_state(pdev);
-
-	err = pci_enable_device(pdev);
-	if (err) {
-		dev_dbg(&pdev->dev, "failed to enable device\n");
-		return err;
-	}
+	struct pch_dma *pd = dev_get_drvdata(dev);
 
 	if (pd)
 		pch_dma_restore_regs(pd);
 
 	return 0;
 }
-#endif
 
 static int pch_dma_probe(struct pci_dev *pdev,
 				   const struct pci_device_id *id)
@@ -993,15 +977,14 @@ static const struct pci_device_id pch_dma_id_table[] = {
 	{ 0, },
 };
 
+static SIMPLE_DEV_PM_OPS(pch_dma_pm_ops, pch_dma_suspend, pch_dma_resume);
+
 static struct pci_driver pch_dma_driver = {
 	.name		= DRV_NAME,
 	.id_table	= pch_dma_id_table,
 	.probe		= pch_dma_probe,
 	.remove		= pch_dma_remove,
-#ifdef CONFIG_PM
-	.suspend	= pch_dma_suspend,
-	.resume		= pch_dma_resume,
-#endif
+	.driver.pm	= &pch_dma_pm_ops,
 };
 
 module_pci_driver(pch_dma_driver);
Signature · 1 line
2.27.0
maintainer
#1 lore raw

Hi Vaibhav,

Quoted 6 lines · Vaibhav Gupta

Drivers using legacy PM have to manage PCI states and device's PM states themselves. They also need to take care of configuration registers.

With improved and powerful support of generic PM, PCI Core takes care of above mentioned, device-independent, jobs.

Quoted 7 lines

This driver makes use of PCI helper functions like pci_save/restore_state(), pci_enable/disable_device(),
and pci_set_power_state() to do required operations. In generic mode, they are no longer needed.

Change function parameter in both .suspend() and .resume() to "struct device*" type. Use dev_get_drvdata() to get drv data.

You are doing too many things in One patch. Do consider splitting them up to a change per patch. for example using __maybe could be one patch, removing code is suspend-resume callbacks would be other one.

Quoted 2 lines

Compile-tested only.

I would like to see some testing before we merge this

Quoted 95 lines
---
 drivers/dma/pch_dma.c | 35 +++++++++--------------------------
 1 file changed, 9 insertions(+), 26 deletions(-)
drivers/dma/pch_dma.c +9 -26
index a3b0b4c56a19..e93005837e3f 100644
@@ -735,8 +735,7 @@ static irqreturn_t pd_irq(int irq, void *devid)
 	return ret0 | ret2;
 }
 
-#ifdef	CONFIG_PM
-static void pch_dma_save_regs(struct pch_dma *pd)
+static void __maybe_unused pch_dma_save_regs(struct pch_dma *pd)
 {
 	struct pch_dma_chan *pd_chan;
 	struct dma_chan *chan, *_c;
@@ -759,7 +758,7 @@ static void pch_dma_save_regs(struct pch_dma *pd)
 	}
 }
 
-static void pch_dma_restore_regs(struct pch_dma *pd)
+static void __maybe_unused pch_dma_restore_regs(struct pch_dma *pd)
 {
 	struct pch_dma_chan *pd_chan;
 	struct dma_chan *chan, *_c;
@@ -782,40 +781,25 @@ static void pch_dma_restore_regs(struct pch_dma *pd)
 	}
 }
 
-static int pch_dma_suspend(struct pci_dev *pdev, pm_message_t state)
+static int __maybe_unused pch_dma_suspend(struct device *dev)
 {
-	struct pch_dma *pd = pci_get_drvdata(pdev);
+	struct pch_dma *pd = dev_get_drvdata(dev);
 
 	if (pd)
 		pch_dma_save_regs(pd);
 
-	pci_save_state(pdev);
-	pci_disable_device(pdev);
-	pci_set_power_state(pdev, pci_choose_state(pdev, state));
-
 	return 0;
 }
 
-static int pch_dma_resume(struct pci_dev *pdev)
+static int __maybe_unused pch_dma_resume(struct device *dev)
 {
-	struct pch_dma *pd = pci_get_drvdata(pdev);
-	int err;
-
-	pci_set_power_state(pdev, PCI_D0);
-	pci_restore_state(pdev);
-
-	err = pci_enable_device(pdev);
-	if (err) {
-		dev_dbg(&pdev->dev, "failed to enable device\n");
-		return err;
-	}
+	struct pch_dma *pd = dev_get_drvdata(dev);
 
 	if (pd)
 		pch_dma_restore_regs(pd);
 
 	return 0;
 }
-#endif
 
 static int pch_dma_probe(struct pci_dev *pdev,
 				   const struct pci_device_id *id)
@@ -993,15 +977,14 @@ static const struct pci_device_id pch_dma_id_table[] = {
 	{ 0, },
 };
 
+static SIMPLE_DEV_PM_OPS(pch_dma_pm_ops, pch_dma_suspend, pch_dma_resume);
+
 static struct pci_driver pch_dma_driver = {
 	.name		= DRV_NAME,
 	.id_table	= pch_dma_id_table,
 	.probe		= pch_dma_probe,
 	.remove		= pch_dma_remove,
-#ifdef CONFIG_PM
-	.suspend	= pch_dma_suspend,
-	.resume		= pch_dma_resume,
-#endif
+	.driver.pm	= &pch_dma_pm_ops,
 };
 
 module_pci_driver(pch_dma_driver);
Signature · 1 line
2.27.0
Signature · 1 line
~Vinod
#2 lore raw
Quoted 14 lines · Vinod Koul

Hi Vaibhav,

Quoted 11 lines · Vaibhav Gupta

Drivers using legacy PM have to manage PCI states and device's PM states themselves. They also need to take care of configuration registers.

With improved and powerful support of generic PM, PCI Core takes care of above mentioned, device-independent, jobs.

This driver makes use of PCI helper functions like pci_save/restore_state(), pci_enable/disable_device(),
and pci_set_power_state() to do required operations. In generic mode, they are no longer needed.

Quoted 7 lines
Quoted 2 lines

Change function parameter in both .suspend() and .resume() to "struct device*" type. Use dev_get_drvdata() to get drv data.

You are doing too many things in One patch. Do consider splitting them up to a change per patch. for example using __maybe could be one patch, removing code is suspend-resume callbacks would be other one.

Sure. But I guess just marking of "__maybe_unused" is not some significant change. All the legacy PCI drivers have .suspend() and .resume() inside "#ifdef CONFIG_PM" container.
It is only when I am upgrading them one by one to generic, I remove the container and mark them with the attribute. So it is like a part of complete generic upgrade.

Thanks
Vaibhav Gupta

Quoted 96 lines
Quoted 1 line

Compile-tested only.

I would like to see some testing before we merge this

Quoted 92 lines
---
 drivers/dma/pch_dma.c | 35 +++++++++--------------------------
 1 file changed, 9 insertions(+), 26 deletions(-)
drivers/dma/pch_dma.c +9 -26
index a3b0b4c56a19..e93005837e3f 100644
@@ -735,8 +735,7 @@ static irqreturn_t pd_irq(int irq, void *devid)
 	return ret0 | ret2;
 }
 
-#ifdef	CONFIG_PM
-static void pch_dma_save_regs(struct pch_dma *pd)
+static void __maybe_unused pch_dma_save_regs(struct pch_dma *pd)
 {
 	struct pch_dma_chan *pd_chan;
 	struct dma_chan *chan, *_c;
@@ -759,7 +758,7 @@ static void pch_dma_save_regs(struct pch_dma *pd)
 	}
 }
 
-static void pch_dma_restore_regs(struct pch_dma *pd)
+static void __maybe_unused pch_dma_restore_regs(struct pch_dma *pd)
 {
 	struct pch_dma_chan *pd_chan;
 	struct dma_chan *chan, *_c;
@@ -782,40 +781,25 @@ static void pch_dma_restore_regs(struct pch_dma *pd)
 	}
 }
 
-static int pch_dma_suspend(struct pci_dev *pdev, pm_message_t state)
+static int __maybe_unused pch_dma_suspend(struct device *dev)
 {
-	struct pch_dma *pd = pci_get_drvdata(pdev);
+	struct pch_dma *pd = dev_get_drvdata(dev);
 
 	if (pd)
 		pch_dma_save_regs(pd);
 
-	pci_save_state(pdev);
-	pci_disable_device(pdev);
-	pci_set_power_state(pdev, pci_choose_state(pdev, state));
-
 	return 0;
 }
 
-static int pch_dma_resume(struct pci_dev *pdev)
+static int __maybe_unused pch_dma_resume(struct device *dev)
 {
-	struct pch_dma *pd = pci_get_drvdata(pdev);
-	int err;
-
-	pci_set_power_state(pdev, PCI_D0);
-	pci_restore_state(pdev);
-
-	err = pci_enable_device(pdev);
-	if (err) {
-		dev_dbg(&pdev->dev, "failed to enable device\n");
-		return err;
-	}
+	struct pch_dma *pd = dev_get_drvdata(dev);
 
 	if (pd)
 		pch_dma_restore_regs(pd);
 
 	return 0;
 }
-#endif
 
 static int pch_dma_probe(struct pci_dev *pdev,
 				   const struct pci_device_id *id)
@@ -993,15 +977,14 @@ static const struct pci_device_id pch_dma_id_table[] = {
 	{ 0, },
 };
 
+static SIMPLE_DEV_PM_OPS(pch_dma_pm_ops, pch_dma_suspend, pch_dma_resume);
+
 static struct pci_driver pch_dma_driver = {
 	.name		= DRV_NAME,
 	.id_table	= pch_dma_id_table,
 	.probe		= pch_dma_probe,
 	.remove		= pch_dma_remove,
-#ifdef CONFIG_PM
-	.suspend	= pch_dma_suspend,
-	.resume		= pch_dma_resume,
-#endif
+	.driver.pm	= &pch_dma_pm_ops,
 };
 
Quoted 6 lines
Quoted 3 lines
drivers/dma/pch_dma.c +0 -0
 module_pci_driver(pch_dma_driver);
Signature · 1 line
2.27.0
Signature · 1 line
~Vinod

_______________________________________________
Linux-kernel-mentees mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/linux-kernel-mentees

Reviewed-by
#3 lore raw
In reply to Vinod Koul
Quoted 12 lines · Vinod Koul
Quoted 11 lines · Vaibhav Gupta

Drivers using legacy PM have to manage PCI states and device's PM states themselves. They also need to take care of configuration registers.

With improved and powerful support of generic PM, PCI Core takes care of above mentioned, device-independent, jobs.

This driver makes use of PCI helper functions like pci_save/restore_state(), pci_enable/disable_device(),
and pci_set_power_state() to do required operations. In generic mode, they are no longer needed.

Quoted 6 lines
Quoted 2 lines

Change function parameter in both .suspend() and .resume() to "struct device*" type. Use dev_get_drvdata() to get drv data.

You are doing too many things in One patch. Do consider splitting them up to a change per patch. for example using __maybe could be one patch, removing code is suspend-resume callbacks would be other one.

Vinod, while I completely agree with you in general, in this case it would make more unnecessary churn and will be rather unhelpful in all
ways: for the contributor to do this work, for the reader to collect all the pieces. It also will be a bisectability issue, because the #ifdeffery replacement (IIRC you need to move from CONFIG_PM to CONFIG_PM_SLEEP). I really don't see any advantages of the splitting here.

Quoted 3 lines
Quoted 1 line

Compile-tested only.

I would like to see some testing before we merge this

I have hardware to test (Intel Minnowboard v1) but have no time. And taking into account that I did similar changes for many other drivers, I can give my

and take responsibility if somebody complains in the future (I don't believe it will be one).

P.S. Another scenario if Vaibhav can find that board (there were dozens of thousands at least produced and floating on the second hand market) and test himself. It may be good since he touches the full lot of PCH (EGT20) drivers.

Quoted 97 lines
Quoted 93 lines
---
 drivers/dma/pch_dma.c | 35 +++++++++--------------------------
 1 file changed, 9 insertions(+), 26 deletions(-)
drivers/dma/pch_dma.c +9 -27
index a3b0b4c56a19..e93005837e3f 100644
@@ -735,8 +735,7 @@ static irqreturn_t pd_irq(int irq, void *devid)
      return ret0 | ret2;
 }
 
-#ifdef       CONFIG_PM
-static void pch_dma_save_regs(struct pch_dma *pd)
+static void __maybe_unused pch_dma_save_regs(struct pch_dma *pd)
 {
      struct pch_dma_chan *pd_chan;
      struct dma_chan *chan, *_c;
@@ -759,7 +758,7 @@ static void pch_dma_save_regs(struct pch_dma *pd)
      }
 }
 
-static void pch_dma_restore_regs(struct pch_dma *pd)
+static void __maybe_unused pch_dma_restore_regs(struct pch_dma *pd)
 {
      struct pch_dma_chan *pd_chan;
      struct dma_chan *chan, *_c;
@@ -782,40 +781,25 @@ static void pch_dma_restore_regs(struct pch_dma *pd)
      }
 }
 
-static int pch_dma_suspend(struct pci_dev *pdev, pm_message_t state)
+static int __maybe_unused pch_dma_suspend(struct device *dev)
 {
-     struct pch_dma *pd = pci_get_drvdata(pdev);
+     struct pch_dma *pd = dev_get_drvdata(dev);
 
      if (pd)
              pch_dma_save_regs(pd);
 
-     pci_save_state(pdev);
-     pci_disable_device(pdev);
-     pci_set_power_state(pdev, pci_choose_state(pdev, state));
-
      return 0;
 }
 
-static int pch_dma_resume(struct pci_dev *pdev)
+static int __maybe_unused pch_dma_resume(struct device *dev)
 {
-     struct pch_dma *pd = pci_get_drvdata(pdev);
-     int err;
-
-     pci_set_power_state(pdev, PCI_D0);
-     pci_restore_state(pdev);
-
-     err = pci_enable_device(pdev);
-     if (err) {
-             dev_dbg(&pdev->dev, "failed to enable device\n");
-             return err;
-     }
+     struct pch_dma *pd = dev_get_drvdata(dev);
 
      if (pd)
              pch_dma_restore_regs(pd);
 
      return 0;
 }
-#endif
 
 static int pch_dma_probe(struct pci_dev *pdev,
                                 const struct pci_device_id *id)
@@ -993,15 +977,14 @@ static const struct pci_device_id pch_dma_id_table[] = {
      { 0, },
 };
 
+static SIMPLE_DEV_PM_OPS(pch_dma_pm_ops, pch_dma_suspend, pch_dma_resume);
+
 static struct pci_driver pch_dma_driver = {
      .name           = DRV_NAME,
      .id_table       = pch_dma_id_table,
      .probe          = pch_dma_probe,
      .remove         = pch_dma_remove,
-#ifdef CONFIG_PM
-     .suspend        = pch_dma_suspend,
-     .resume         = pch_dma_resume,
-#endif
+     .driver.pm      = &pch_dma_pm_ops,
 };
 
 module_pci_driver(pch_dma_driver);
--
Quoted 1 line

2.27.0

--
~Vinod

Signature · 2 lines
With Best Regards,
Andy Shevchenko
#4 lore raw
Quoted 30 lines · Andy Shevchenko
Quoted 12 lines · Vinod Koul
Quoted 11 lines · Vaibhav Gupta

Drivers using legacy PM have to manage PCI states and device's PM states themselves. They also need to take care of configuration registers.

With improved and powerful support of generic PM, PCI Core takes care of above mentioned, device-independent, jobs.

This driver makes use of PCI helper functions like pci_save/restore_state(), pci_enable/disable_device(),
and pci_set_power_state() to do required operations. In generic mode, they are no longer needed.

Quoted 6 lines
Quoted 2 lines

Change function parameter in both .suspend() and .resume() to "struct device*" type. Use dev_get_drvdata() to get drv data.

You are doing too many things in One patch. Do consider splitting them up to a change per patch. for example using __maybe could be one patch, removing code is suspend-resume callbacks would be other one.

Vinod, while I completely agree with you in general, in this case it would make more unnecessary churn and will be rather unhelpful in all
ways: for the contributor to do this work, for the reader to collect all the pieces. It also will be a bisectability issue, because the #ifdeffery replacement (IIRC you need to move from CONFIG_PM to CONFIG_PM_SLEEP). I really don't see any advantages of the splitting here.

Quoted 2 lines
Quoted 1 line

Compile-tested only.

Quoted 9 lines
Quoted 1 line

I would like to see some testing before we merge this

I have hardware to test (Intel Minnowboard v1) but have no time. And taking into account that I did similar changes for many other drivers, I can give my

and take responsibility if somebody complains in the future (I don't believe it will be one).

Thanks! :D

Quoted 5 lines

P.S. Another scenario if Vaibhav can find that board (there were dozens of thousands at least produced and floating on the second hand market) and test himself. It may be good since he touches the full lot of PCH (EGT20) drivers.

I cannot promise, but surely will try to get my hands on it :)

Thanks
Vaibhav Gupta

Quoted 103 lines
Quoted 97 lines
Quoted 93 lines
---
 drivers/dma/pch_dma.c | 35 +++++++++--------------------------
 1 file changed, 9 insertions(+), 26 deletions(-)
drivers/dma/pch_dma.c +9 -27
index a3b0b4c56a19..e93005837e3f 100644
@@ -735,8 +735,7 @@ static irqreturn_t pd_irq(int irq, void *devid)
      return ret0 | ret2;
 }
 
-#ifdef       CONFIG_PM
-static void pch_dma_save_regs(struct pch_dma *pd)
+static void __maybe_unused pch_dma_save_regs(struct pch_dma *pd)
 {
      struct pch_dma_chan *pd_chan;
      struct dma_chan *chan, *_c;
@@ -759,7 +758,7 @@ static void pch_dma_save_regs(struct pch_dma *pd)
      }
 }
 
-static void pch_dma_restore_regs(struct pch_dma *pd)
+static void __maybe_unused pch_dma_restore_regs(struct pch_dma *pd)
 {
      struct pch_dma_chan *pd_chan;
      struct dma_chan *chan, *_c;
@@ -782,40 +781,25 @@ static void pch_dma_restore_regs(struct pch_dma *pd)
      }
 }
 
-static int pch_dma_suspend(struct pci_dev *pdev, pm_message_t state)
+static int __maybe_unused pch_dma_suspend(struct device *dev)
 {
-     struct pch_dma *pd = pci_get_drvdata(pdev);
+     struct pch_dma *pd = dev_get_drvdata(dev);
 
      if (pd)
              pch_dma_save_regs(pd);
 
-     pci_save_state(pdev);
-     pci_disable_device(pdev);
-     pci_set_power_state(pdev, pci_choose_state(pdev, state));
-
      return 0;
 }
 
-static int pch_dma_resume(struct pci_dev *pdev)
+static int __maybe_unused pch_dma_resume(struct device *dev)
 {
-     struct pch_dma *pd = pci_get_drvdata(pdev);
-     int err;
-
-     pci_set_power_state(pdev, PCI_D0);
-     pci_restore_state(pdev);
-
-     err = pci_enable_device(pdev);
-     if (err) {
-             dev_dbg(&pdev->dev, "failed to enable device\n");
-             return err;
-     }
+     struct pch_dma *pd = dev_get_drvdata(dev);
 
      if (pd)
              pch_dma_restore_regs(pd);
 
      return 0;
 }
-#endif
 
 static int pch_dma_probe(struct pci_dev *pdev,
                                 const struct pci_device_id *id)
@@ -993,15 +977,14 @@ static const struct pci_device_id pch_dma_id_table[] = {
      { 0, },
 };
 
+static SIMPLE_DEV_PM_OPS(pch_dma_pm_ops, pch_dma_suspend, pch_dma_resume);
+
 static struct pci_driver pch_dma_driver = {
      .name           = DRV_NAME,
      .id_table       = pch_dma_id_table,
      .probe          = pch_dma_probe,
      .remove         = pch_dma_remove,
-#ifdef CONFIG_PM
-     .suspend        = pch_dma_suspend,
-     .resume         = pch_dma_resume,
-#endif
+     .driver.pm      = &pch_dma_pm_ops,
 };
 
 module_pci_driver(pch_dma_driver);
--
Quoted 1 line

2.27.0

--
~Vinod

Signature · 2 lines
With Best Regards,
Andy Shevchenko
maintainer
#5 lore raw
In reply to Andy Shevchenko
Quoted 35 lines · Andy Shevchenko
Quoted 12 lines · Vinod Koul
Quoted 11 lines · Vaibhav Gupta

Drivers using legacy PM have to manage PCI states and device's PM states themselves. They also need to take care of configuration registers.

With improved and powerful support of generic PM, PCI Core takes care of above mentioned, device-independent, jobs.

This driver makes use of PCI helper functions like pci_save/restore_state(), pci_enable/disable_device(),
and pci_set_power_state() to do required operations. In generic mode, they are no longer needed.

Quoted 6 lines
Quoted 2 lines

Change function parameter in both .suspend() and .resume() to "struct device*" type. Use dev_get_drvdata() to get drv data.

You are doing too many things in One patch. Do consider splitting them up to a change per patch. for example using __maybe could be one patch, removing code is suspend-resume callbacks would be other one.

Vinod, while I completely agree with you in general, in this case it would make more unnecessary churn and will be rather unhelpful in all
ways: for the contributor to do this work, for the reader to collect all the pieces. It also will be a bisectability issue, because the #ifdeffery replacement (IIRC you need to move from CONFIG_PM to CONFIG_PM_SLEEP). I really don't see any advantages of the splitting here.

Quoted 3 lines
Quoted 1 line

Compile-tested only.

I would like to see some testing before we merge this

I have hardware to test (Intel Minnowboard v1) but have no time. And taking into account that I did similar changes for many other drivers, I can give my

Quoted 8 lines

and take responsibility if somebody complains in the future (I don't believe it will be one).

P.S. Another scenario if Vaibhav can find that board (there were dozens of thousands at least produced and floating on the second hand market) and test himself. It may be good since he touches the full lot of PCH (EGT20) drivers.

Applied now, thanks

Signature · 1 line
~Vinod