| Message ID | 20240611163525.4156688-2-matthew.gerlach@linux.intel.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Krzysztof Wilczyński |
| Headers | show |
| Series | [v6,1/2] dt-bindings: PCI: altera: Convert to YAML | expand |
On Tue, Jun 11, 2024 at 11:35:25AM -0500, matthew.gerlach@linux.intel.com wrote: > From: Matthew Gerlach <matthew.gerlach@linux.intel.com> > > Add support for the device tree binding update. As part of > converting the binding document from text to yaml, with schema > validation, a device tree subnode was added to properly map > legacy interrupts. Maintain backward compatibility with previous binding. If something was *added* to the binding, I think it would be helpful to split that into two patches: (1) convert to YAML with zero functional changes, (2) add the new stuff. Adding something at the same time as changing the format makes it hard to review. Then we could have a more specific subject and commit log for *this* patch. > Signed-off-by: Matthew Gerlach <matthew.gerlach@linux.intel.com> > --- > drivers/pci/controller/pcie-altera.c | 13 +++++++++++-- > 1 file changed, 11 insertions(+), 2 deletions(-) > > diff --git a/drivers/pci/controller/pcie-altera.c b/drivers/pci/controller/pcie-altera.c > index a9536dc4bf96..88511fa2f078 100644 > --- a/drivers/pci/controller/pcie-altera.c > +++ b/drivers/pci/controller/pcie-altera.c > @@ -667,11 +667,20 @@ static void altera_pcie_isr(struct irq_desc *desc) > static int altera_pcie_init_irq_domain(struct altera_pcie *pcie) > { > struct device *dev = &pcie->pdev->dev; > - struct device_node *node = dev->of_node; > + struct device_node *node, *child; > > /* Setup INTx */ > + child = of_get_next_child(dev->of_node, NULL); > + if (child) > + node = child; > + else > + node = dev->of_node; > + > pcie->irq_domain = irq_domain_add_linear(node, PCI_NUM_INTX, > - &intx_domain_ops, pcie); > + &intx_domain_ops, pcie); > + if (child) > + of_node_put(child); > + > if (!pcie->irq_domain) { > dev_err(dev, "Failed to get a INTx IRQ domain\n"); > return -ENOMEM; > -- > 2.34.1 >
On Tue, 11 Jun 2024, Bjorn Helgaas wrote: > On Tue, Jun 11, 2024 at 11:35:25AM -0500, matthew.gerlach@linux.intel.com wrote: >> From: Matthew Gerlach <matthew.gerlach@linux.intel.com> >> >> Add support for the device tree binding update. As part of >> converting the binding document from text to yaml, with schema >> validation, a device tree subnode was added to properly map >> legacy interrupts. Maintain backward compatibility with previous binding. > > If something was *added* to the binding, I think it would be helpful > to split that into two patches: (1) convert to YAML with zero > functional changes, (2) add the new stuff. Adding something at the > same time as changing the format makes it hard to review. Thanks for feedback. It was during the conversion to YAML that a problem with the original binding was discovered. As Rob Herring pointed out in https://patchwork.ozlabs.org/project/devicetree-bindings/patch/20240513205913.313592-1-matthew.gerlach@linux.intel.com/ "Making the PCI host the interrupt parent didn't even work in the kernel until somewhat recently (maybe a few years now). That's why a bunch of PCI hosts have an interrupt-controller child node." This was an attempt to fix the problem. I can resubmit a conversion to YAML with zero functional changes. Matthew Gerlach > > Then we could have a more specific subject and commit log for *this* > patch. > >> Signed-off-by: Matthew Gerlach <matthew.gerlach@linux.intel.com> >> --- >> drivers/pci/controller/pcie-altera.c | 13 +++++++++++-- >> 1 file changed, 11 insertions(+), 2 deletions(-) >> >> diff --git a/drivers/pci/controller/pcie-altera.c b/drivers/pci/controller/pcie-altera.c >> index a9536dc4bf96..88511fa2f078 100644 >> --- a/drivers/pci/controller/pcie-altera.c >> +++ b/drivers/pci/controller/pcie-altera.c >> @@ -667,11 +667,20 @@ static void altera_pcie_isr(struct irq_desc *desc) >> static int altera_pcie_init_irq_domain(struct altera_pcie *pcie) >> { >> struct device *dev = &pcie->pdev->dev; >> - struct device_node *node = dev->of_node; >> + struct device_node *node, *child; >> >> /* Setup INTx */ >> + child = of_get_next_child(dev->of_node, NULL); >> + if (child) >> + node = child; >> + else >> + node = dev->of_node; >> + >> pcie->irq_domain = irq_domain_add_linear(node, PCI_NUM_INTX, >> - &intx_domain_ops, pcie); >> + &intx_domain_ops, pcie); >> + if (child) >> + of_node_put(child); >> + >> if (!pcie->irq_domain) { >> dev_err(dev, "Failed to get a INTx IRQ domain\n"); >> return -ENOMEM; >> -- >> 2.34.1 >> >
On Wed, Jun 12, 2024 at 08:12:05AM -0700, matthew.gerlach@linux.intel.com wrote: > > > On Tue, 11 Jun 2024, Bjorn Helgaas wrote: > > > On Tue, Jun 11, 2024 at 11:35:25AM -0500, matthew.gerlach@linux.intel.com wrote: > > > From: Matthew Gerlach <matthew.gerlach@linux.intel.com> > > > > > > Add support for the device tree binding update. As part of > > > converting the binding document from text to yaml, with schema > > > validation, a device tree subnode was added to properly map > > > legacy interrupts. Maintain backward compatibility with previous binding. > > > > If something was *added* to the binding, I think it would be helpful > > to split that into two patches: (1) convert to YAML with zero > > functional changes, (2) add the new stuff. Adding something at the > > same time as changing the format makes it hard to review. The policy for conversions is changes to match reality are fine, just need to be noted in the commit message. That generally implies no driver or dts changes which is not the case here. > Thanks for feedback. It was during the conversion to YAML that a problem > with the original binding was discovered. As Rob Herring pointed out in > https://patchwork.ozlabs.org/project/devicetree-bindings/patch/20240513205913.313592-1-matthew.gerlach@linux.intel.com/ > > "Making the PCI host the interrupt parent didn't even work in the kernel > until somewhat recently (maybe a few years now). That's why a bunch of PCI > hosts have an interrupt-controller child node." > > This was an attempt to fix the problem. I can resubmit a conversion to YAML > with zero functional changes. I wasn't suggesting you fix it. Just something I noticed looking at the other issue. If no one noticed or cared, why bother? It should work fine for recent kernels. Rob
On Thu, 13 Jun 2024, Rob Herring wrote: > On Wed, Jun 12, 2024 at 08:12:05AM -0700, matthew.gerlach@linux.intel.com wrote: >> >> >> On Tue, 11 Jun 2024, Bjorn Helgaas wrote: >> >>> On Tue, Jun 11, 2024 at 11:35:25AM -0500, matthew.gerlach@linux.intel.com wrote: >>>> From: Matthew Gerlach <matthew.gerlach@linux.intel.com> >>>> >>>> Add support for the device tree binding update. As part of >>>> converting the binding document from text to yaml, with schema >>>> validation, a device tree subnode was added to properly map >>>> legacy interrupts. Maintain backward compatibility with previous binding. >>> >>> If something was *added* to the binding, I think it would be helpful >>> to split that into two patches: (1) convert to YAML with zero >>> functional changes, (2) add the new stuff. Adding something at the >>> same time as changing the format makes it hard to review. > > The policy for conversions is changes to match reality are fine, just > need to be noted in the commit message. That generally implies no driver > or dts changes which is not the case here. > >> Thanks for feedback. It was during the conversion to YAML that a problem >> with the original binding was discovered. As Rob Herring pointed out in >> https://patchwork.ozlabs.org/project/devicetree-bindings/patch/20240513205913.313592-1-matthew.gerlach@linux.intel.com/ >> >> "Making the PCI host the interrupt parent didn't even work in the kernel >> until somewhat recently (maybe a few years now). That's why a bunch of PCI >> hosts have an interrupt-controller child node." >> >> This was an attempt to fix the problem. I can resubmit a conversion to YAML >> with zero functional changes. > > I wasn't suggesting you fix it. Just something I noticed looking at > the other issue. If no one noticed or cared, why bother? It should work > fine for recent kernels. Thanks for the feedback. I can resubmit a single conversion commit that passes the schema check by adding 3 address fields as you suggested. I will also mention this slight modification in the commit message. Matthew > > Rob >
diff --git a/drivers/pci/controller/pcie-altera.c b/drivers/pci/controller/pcie-altera.c index a9536dc4bf96..88511fa2f078 100644 --- a/drivers/pci/controller/pcie-altera.c +++ b/drivers/pci/controller/pcie-altera.c @@ -667,11 +667,20 @@ static void altera_pcie_isr(struct irq_desc *desc) static int altera_pcie_init_irq_domain(struct altera_pcie *pcie) { struct device *dev = &pcie->pdev->dev; - struct device_node *node = dev->of_node; + struct device_node *node, *child; /* Setup INTx */ + child = of_get_next_child(dev->of_node, NULL); + if (child) + node = child; + else + node = dev->of_node; + pcie->irq_domain = irq_domain_add_linear(node, PCI_NUM_INTX, - &intx_domain_ops, pcie); + &intx_domain_ops, pcie); + if (child) + of_node_put(child); + if (!pcie->irq_domain) { dev_err(dev, "Failed to get a INTx IRQ domain\n"); return -ENOMEM;