Skip to contentSkip to conversation

PCI: imx: imx6_pcie_attach_pd

PATCH In review Leonard Crestez 3 participants 8 messages Jan 21, 2019 – Jan 25, 2019

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

cover PCI: imx: imx6_pcie_attach_pd

This fixes pci probing on imx6qp as reported by Lukas Hartmann.

The second patch fixes a check that shouldn't actually fail.

Leonard Crestez (2):
PCI: imx: Fix probe failure without power domain
PCI: imx: Fix checking pd_pcie_phy device_link

 drivers/pci/controller/dwc/pci-imx6.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)
Signature · 1 line
2.17.1
#1 lore raw

PATCH 1/2 PCI: imx: Fix probe failure without power domain

On chips without a separate power domain for PCI (such as 6q/6qp) the imx6_pcie_attach_pd function incorrectly returns an error.

Fix by returning 0 if dev_pm_domain_attach_by_name doesn't find anything.

Fixes 3f7cceeab895 ("PCI: imx: Add multi-pd support")
---
 drivers/pci/controller/dwc/pci-imx6.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/pci/controller/dwc/pci-imx6.c +3 -0
index 52e47dac028f..ac5f6ae0b254 100644
@@ -308,10 +308,13 @@ static int imx6_pcie_attach_pd(struct device *dev)
 		return 0;
 
 	imx6_pcie->pd_pcie = dev_pm_domain_attach_by_name(dev, "pcie");
 	if (IS_ERR(imx6_pcie->pd_pcie))
 		return PTR_ERR(imx6_pcie->pd_pcie);
+	/* Do nothing when power domain missing */
+	if (!imx6_pcie->pd_pcie)
+		return 0;
 	link = device_link_add(dev, imx6_pcie->pd_pcie,
 			DL_FLAG_STATELESS |
 			DL_FLAG_PM_RUNTIME |
 			DL_FLAG_RPM_ACTIVE);
 	if (!link) {
Signature · 1 line
2.17.1
#2 lore raw

Re: [PATCH 1/2] PCI: imx: Fix probe failure without power domain

Quoted 18 lines · Leonard Crestez

On chips without a separate power domain for PCI (such as 6q/6qp) the imx6_pcie_attach_pd function incorrectly returns an error.

Fix by returning 0 if dev_pm_domain_attach_by_name doesn't find anything.

Fixes 3f7cceeab895 ("PCI: imx: Add multi-pd support")
---
 drivers/pci/controller/dwc/pci-imx6.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/pci/controller/dwc/pci-imx6.c +0 -0
index 52e47dac028f..ac5f6ae0b254 100644

drivers/pci/controller/dwc/pci-imx6.c static int imx6_pcie_attach_pd(struct device *dev) · line 315

Quoted 8 lines
drivers/pci/controller/dwc/pci-imx6.c +3 -0
@@ -308,10 +308,13 @@ static int imx6_pcie_attach_pd(struct device *dev)
 		return 0;
 
 	imx6_pcie->pd_pcie = dev_pm_domain_attach_by_name(dev, "pcie");
 	if (IS_ERR(imx6_pcie->pd_pcie))
 		return PTR_ERR(imx6_pcie->pd_pcie);
+	/* Do nothing when power domain missing */
+	if (!imx6_pcie->pd_pcie)
+		return 0;

As I said in the mail that proposed this patch for testing: I think it would be better to change dev_pm_domain_attach_by_name to not return an error indication by returning NULL or an ERR_PTR value. (Or change device_link_add to accept NULL if NULL is a dummy value.)

Just repeating it here to have it near the actual patch.

Best regards
Uwe

Quoted 14 lines
drivers/pci/controller/dwc/pci-imx6.c +0 -0
 	link = device_link_add(dev, imx6_pcie->pd_pcie,
 			DL_FLAG_STATELESS |
 			DL_FLAG_PM_RUNTIME |
 			DL_FLAG_RPM_ACTIVE);
 	if (!link) {
Signature · 7 lines
2.17.1


_______________________________________________
linux-arm-kernel mailing list
[email protected]
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel
Signature · 2 lines
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
#3 lore raw

Re: [PATCH 1/2] PCI: imx: Fix probe failure without power domain

Quoted 25 lines · Uwe Kleine-König
Quoted 10 lines · Leonard Crestez

On chips without a separate power domain for PCI (such as 6q/6qp) the imx6_pcie_attach_pd function incorrectly returns an error.

Fix by returning 0 if dev_pm_domain_attach_by_name doesn't find anything.

Fixes 3f7cceeab895 ("PCI: imx: Add multi-pd support")

drivers/pci/controller/dwc/pci-imx6.c static int imx6_pcie_attach_pd(struct device *dev) · line 315

Quoted 8 lines
drivers/pci/controller/dwc/pci-imx6.c +3 -0
@@ -308,10 +308,13 @@ static int imx6_pcie_attach_pd(struct device *dev)
  		return 0;
 
  	imx6_pcie->pd_pcie = dev_pm_domain_attach_by_name(dev, "pcie");
  	if (IS_ERR(imx6_pcie->pd_pcie))
  		return PTR_ERR(imx6_pcie->pd_pcie);
+	/* Do nothing when power domain missing */
+	if (!imx6_pcie->pd_pcie)
+		return 0;

As I said in the mail that proposed this patch for testing: I think it would be better to change dev_pm_domain_attach_by_name to not return an error indication by returning NULL or an ERR_PTR value. (Or change device_link_add to accept NULL if NULL is a dummy value.)

Quoted 1 line

Just repeating it here to have it near the actual patch.

Link to that previous discussion:
http://lists.infradead.org/pipermail/linux-arm-kernel/2019-January/626859.html

I'm not sure what you mean. Should dev_pm_domain_attach_by_name return ERR_PTR(-ENOENT) for name not found? It would still require special handling for callers.

The device_link_add function already returns NULL if the consumer or supplier is NULL but that's also the only way it signals failure. Maybe that function should be adjusted to return ERR_PTR instead?

Changing core API for a driver bug seems inappropriate.

The current imx6_pcie_attach_pd code is complicated because it tries to distinguish between "multi pd" and "no pd / single pd". I wish there was a way for these PM_RUNTIME device_links to be setup automatically in core and have multi-PD behave the same as a single PD.

There aren't many multi-PD users but commit 6494a9ad86de ("usb: xhci:
tegra: Add genpd support") seems to do pretty much the same thing.

--
Regards,
Leonard

#4 lore raw

Re: [PATCH 1/2] PCI: imx: Fix probe failure without power domain

Quoted 26 lines · Leonard Crestez
Quoted 25 lines · Uwe Kleine-König
Quoted 10 lines · Leonard Crestez

On chips without a separate power domain for PCI (such as 6q/6qp) the imx6_pcie_attach_pd function incorrectly returns an error.

Fix by returning 0 if dev_pm_domain_attach_by_name doesn't find anything.

Fixes 3f7cceeab895 ("PCI: imx: Add multi-pd support")

drivers/pci/controller/dwc/pci-imx6.c static int imx6_pcie_attach_pd(struct device *dev) · line 315

Quoted 8 lines
drivers/pci/controller/dwc/pci-imx6.c +3 -0
@@ -308,10 +308,13 @@ static int imx6_pcie_attach_pd(struct device *dev)
  		return 0;
 
  	imx6_pcie->pd_pcie = dev_pm_domain_attach_by_name(dev, "pcie");
  	if (IS_ERR(imx6_pcie->pd_pcie))
  		return PTR_ERR(imx6_pcie->pd_pcie);
+	/* Do nothing when power domain missing */
+	if (!imx6_pcie->pd_pcie)
+		return 0;

As I said in the mail that proposed this patch for testing: I think it would be better to change dev_pm_domain_attach_by_name to not return an error indication by returning NULL or an ERR_PTR value. (Or change device_link_add to accept NULL if NULL is a dummy value.)

Quoted 8 lines
Quoted 1 line

Just repeating it here to have it near the actual patch.

Link to that previous discussion:
http://lists.infradead.org/pipermail/linux-arm-kernel/2019-January/626859.html

I'm not sure what you mean. Should dev_pm_domain_attach_by_name return ERR_PTR(-ENOENT) for name not found? It would still require special handling for callers.

Yes, callers would still need to handle an error, but they only need to check for IS_ERR(retval) instead of IS_ERR(retval) && retval != NULL.

Quoted 5 lines

The device_link_add function already returns NULL if the consumer or supplier is NULL but that's also the only way it signals failure. Maybe that function should be adjusted to return ERR_PTR instead?

Changing core API for a driver bug seems inappropriate.

Well, if the driver bug is there because the API is used in a wrong way and the right way is strange, fixing the API and checking other users for similar mistakes is the right thing to do.

Quoted 7 lines

The current imx6_pcie_attach_pd code is complicated because it tries to distinguish between "multi pd" and "no pd / single pd". I wish there was a way for these PM_RUNTIME device_links to be setup automatically in core and have multi-PD behave the same as a single PD.

There aren't many multi-PD users but commit 6494a9ad86de ("usb: xhci:
tegra: Add genpd support") seems to do pretty much the same thing.

I don't know enough about power domains to follow this argumentation in the time I have available.

Best regards
Uwe

Signature · 2 lines
Pengutronix e.K.                           | Uwe Kleine-König            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
#5 lore raw

Re: [PATCH 1/2] PCI: imx: Fix probe failure without power domain

Quoted 27 lines · Uwe Kleine-König
Quoted 26 lines · Leonard Crestez
Quoted 24 lines · Uwe Kleine-König
Quoted 10 lines · Leonard Crestez

On chips without a separate power domain for PCI (such as 6q/6qp) the imx6_pcie_attach_pd function incorrectly returns an error.

Fix by returning 0 if dev_pm_domain_attach_by_name doesn't find anything.

Fixes 3f7cceeab895 ("PCI: imx: Add multi-pd support")

drivers/pci/controller/dwc/pci-imx6.c static int imx6_pcie_attach_pd(struct device *dev) · line 315

Quoted 8 lines
drivers/pci/controller/dwc/pci-imx6.c +3 -0
@@ -308,10 +308,13 @@ static int imx6_pcie_attach_pd(struct device *dev)
   		return 0;
 
   imx6_pcie->pd_pcie = dev_pm_domain_attach_by_name(dev, "pcie");
   if (IS_ERR(imx6_pcie->pd_pcie))
   	return PTR_ERR(imx6_pcie->pd_pcie);
+	/* Do nothing when power domain missing */
+	if (!imx6_pcie->pd_pcie)
+		return 0;

As I said in the mail that proposed this patch for testing: I think it would be better to change dev_pm_domain_attach_by_name to not return an error indication by returning NULL or an ERR_PTR value. (Or change device_link_add to accept NULL if NULL is a dummy value.)

Quoted 6 lines
Quoted 3 lines

I'm not sure what you mean. Should dev_pm_domain_attach_by_name return ERR_PTR(-ENOENT) for name not found? It would still require special handling for callers.

Yes, callers would still need to handle an error, but they only need to check for IS_ERR(retval) instead of IS_ERR(retval) && retval != NULL.

But "not found" is not treated as an error here, it's what happens when no PD is assigned. The fix makes imx6_pcie_attach_pd return success in this case.

Making dev_pm_domain_attach_by_name return ENOENT would still require a special case for ENOENT here.

--
Regards,
Leonard

#6 lore raw

PATCH 2/2 PCI: imx: Fix checking pd_pcie_phy device_link

In reply to Leonard Crestez

This check was completely wrong. It doesn't break anything specific because adding the second device_link can't fail.

Fixes 3f7cceeab895 ("PCI: imx: Add multi-pd support")
---
 drivers/pci/controller/dwc/pci-imx6.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
drivers/pci/controller/dwc/pci-imx6.c +4 -4
index ac5f6ae0b254..80f843030e36 100644
@@ -324,17 +324,17 @@ static int imx6_pcie_attach_pd(struct device *dev)
 
 	imx6_pcie->pd_pcie_phy = dev_pm_domain_attach_by_name(dev, "pcie_phy");
 	if (IS_ERR(imx6_pcie->pd_pcie_phy))
 		return PTR_ERR(imx6_pcie->pd_pcie_phy);
 
-	device_link_add(dev, imx6_pcie->pd_pcie_phy,
+	link = device_link_add(dev, imx6_pcie->pd_pcie_phy,
 			DL_FLAG_STATELESS |
 			DL_FLAG_PM_RUNTIME |
 			DL_FLAG_RPM_ACTIVE);
-	if (IS_ERR(link)) {
-		dev_err(dev, "Failed to add device_link to pcie_phy pd: %ld\n", PTR_ERR(link));
-		return PTR_ERR(link);
+	if (!link) {
+		dev_err(dev, "Failed to add device_link to pcie_phy pd.\n");
+		return -EINVAL;
 	}
 
 	return 0;
 }
 
Signature · 1 line
2.17.1
#7 lore raw
In reply to Leonard Crestez
Quoted 10 lines · Leonard Crestez

This fixes pci probing on imx6qp as reported by Lukas Hartmann.

The second patch fixes a check that shouldn't actually fail.

Leonard Crestez (2):
PCI: imx: Fix probe failure without power domain
PCI: imx: Fix checking pd_pcie_phy device_link

 drivers/pci/controller/dwc/pci-imx6.c | 11 +++++++----
 1 file changed, 7 insertions(+), 4 deletions(-)

I have applied them to pci/controller-fixes and Bjorn already queued them for one of the upcoming -rc*.

Thanks,
Lorenzo