Hi Andy,

On Mon, 5 Aug 2024 22:13:38 +0200
Andy Shevchenko [off-list ref] wrote:

On Mon, Aug 5, 2024 at 12:19 PM Herve Codina [off-list ref] wrote:

quoted

Add a PCI driver that handles the LAN966x PCI device using a device-tree
overlay. This overlay is applied to the PCI device DT node and allows to
describe components that are present in the device.

The memory from the device-tree is remapped to the BAR memory thanks to
"ranges" properties computed at runtime by the PCI core during the PCI
enumeration.

The PCI device itself acts as an interrupt controller and is used as the
parent of the internal LAN966x interrupt controller to route the
interrupts to the assigned PCI INTx interrupt.  

...

+ device.h

Will be added.

quoted

+#include <linux/irq.h>
+#include <linux/irqdomain.h>
+#include <linux/module.h>
+#include <linux/of_platform.h>  

quoted

+#include <linux/pci.h>  

quoted

+#include <linux/pci_ids.h>  

AFAIU pci_ids..h is guaranteed to be included by pci.h, but having it
here explicitly doesn't make it worse, so up to you.

I will keep pci_ids.h

quoted

+#include <linux/slab.h>  

...

quoted

+static irqreturn_t pci_dev_irq_handler(int irq, void *data)
+{
+       struct pci_dev_intr_ctrl *intr_ctrl = data;
+       int ret;
+
+       ret = generic_handle_domain_irq(intr_ctrl->irq_domain, 0);
+       return IRQ_RETVAL(!ret);  

Hmm... I dunno if it was me who suggested IRQ_RETVAL() here, but it
usually makes sense for the cases where ret is not inverted.

Perhaps

  if (ret)
    return NONE;
  return HANDLED;

is slightly better in this case?

Right. I will use a more compact version:
  return ret ? IRQ_NONE : IRQ_HANDLED;

quoted

+}  

...

quoted

+static struct pci_dev_intr_ctrl *pci_dev_create_intr_ctrl(struct pci_dev *pdev)
+{
+       struct pci_dev_intr_ctrl *intr_ctrl;
+       struct fwnode_handle *fwnode;
+       int ret;  

quoted

+       if (!pdev->irq)
+               return ERR_PTR(-EOPNOTSUPP);  

Before even trying to get it via APIs? (see below as well)
Also, when is it possible to have 0 here?

pdev->irq can be 0 if the PCI device did not request any IRQ
(i.e. PCI_INTERRUPT_PIN in PCI config header is 0).

I use that to check whether or not INTx is supported.

Even if this code is present in the LAN966x PCI driver, it can be use as a
starting point for other drivers and may be moved to a common part in the
future.

Do you think I should remove it ?

If keeping it is fine, I will add a comment.

quoted

+       fwnode = dev_fwnode(&pdev->dev);
+       if (!fwnode)
+               return ERR_PTR(-ENODEV);
+
+       intr_ctrl = kmalloc(sizeof(*intr_ctrl), GFP_KERNEL);  

Hmm... Why not use __free()?

Well, just because I am not used to using __free() and I didn't think
about it.

I will use it in the next operation.

quoted

+       if (!intr_ctrl)
+               return ERR_PTR(-ENOMEM);
+
+       intr_ctrl->pci_dev = pdev;
+
+       intr_ctrl->irq_domain = irq_domain_create_linear(fwnode, 1, &pci_dev_irq_domain_ops,
+                                                        intr_ctrl);
+       if (!intr_ctrl->irq_domain) {
+               pci_err(pdev, "Failed to create irqdomain\n");
+               ret = -ENOMEM;
+               goto err_free_intr_ctrl;
+       }  

quoted

+       ret = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_INTX);
+       if (ret < 0) {
+               pci_err(pdev, "Unable alloc irq vector (%d)\n", ret);
+               goto err_remove_domain;
+       }  

I am wondering if you even need this in case you want solely INTx.

I have the feeling that it is needed.
pci_alloc_irq_vectors() will call pci_intx() which in turn enables INT
clearing PCI_COMMAND_INTX_DISABLE flag in the PCI_COMMAND config word.

quoted

+       intr_ctrl->irq = pci_irq_vector(pdev, 0);  

Don't remember documentation by heart for this, but the implementation
suggests that it can be called without the above for retrieving INTx.

So, with the above said, I will keep both pci_alloc_irq_vectors() and
pci_irq_vector() calls.

quoted

+       ret = request_irq(intr_ctrl->irq, pci_dev_irq_handler, IRQF_SHARED,
+                         dev_name(&pdev->dev), intr_ctrl);  

pci_name() ? (IIRC the macro name)

Indeed, will be changed.

quoted

+       if (ret) {
+               pci_err(pdev, "Unable to request irq %d (%d)\n", intr_ctrl->irq, ret);
+               goto err_free_irq_vector;
+       }
+
+       return intr_ctrl;
+
+err_free_irq_vector:
+       pci_free_irq_vectors(pdev);
+err_remove_domain:
+       irq_domain_remove(intr_ctrl->irq_domain);
+err_free_intr_ctrl:
+       kfree(intr_ctrl);
+       return ERR_PTR(ret);
+}  

...

quoted

+static void devm_pci_dev_remove_intr_ctrl(void *data)
+{  

quoted

+       struct pci_dev_intr_ctrl *intr_ctrl = data;  

It can be eliminated

static void devm_pci_...(void *intr_ctrl)

I will update.

quoted

+       pci_dev_remove_intr_ctrl(intr_ctrl);
+}  

...

quoted

+static int lan966x_pci_load_overlay(struct lan966x_pci *data)
+{
+       u32 dtbo_size = __dtbo_lan966x_pci_end - __dtbo_lan966x_pci_begin;
+       void *dtbo_start = __dtbo_lan966x_pci_begin;
+       int ret;
+
+       ret = of_overlay_fdt_apply(dtbo_start, dtbo_size, &data->ovcs_id, dev_of_node(data->dev));
+       if (ret)
+               return ret;
+
+       return 0;  

return of_overlay_fdt_apply() ?

Yes indeed.

quoted

+}  

...

quoted

+static int lan966x_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id)
+{
+       struct device *dev = &pdev->dev;
+       struct lan966x_pci *data;
+       int ret;
+
+       /*
+        * On ACPI system, fwnode can point to the ACPI node.
+        * This driver needs an of_node to be used as the device-tree overlay
+        * target. This of_node should be set by the PCI core if it succeeds in
+        * creating it (CONFIG_PCI_DYNAMIC_OF_NODES feature).
+        * Check here for the validity of this of_node.
+        */
+       if (!dev_of_node(dev)) {  

quoted

+               dev_err(dev, "Missing of_node for device\n");
+               return -EINVAL;  

return dev_err_probe() ?

Yes, I will update.

quoted

+       }
+
+       /* Need to be done before devm_pci_dev_create_intr_ctrl.
+        * It allocates an IRQ and so pdev->irq is updated.
+        */
+       ret = pcim_enable_device(pdev);
+       if (ret)
+               return ret;
+
+       ret = devm_pci_dev_create_intr_ctrl(pdev);
+       if (ret)
+               return ret;
+
+       data = devm_kzalloc(dev, sizeof(*data), GFP_KERNEL);
+       if (!data)
+               return -ENOMEM;
+
+       pci_set_drvdata(pdev, data);
+       data->dev = dev;
+
+       ret = lan966x_pci_load_overlay(data);
+       if (ret)
+               return ret;
+
+       pci_set_master(pdev);
+
+       ret = of_platform_default_populate(dev_of_node(dev), NULL, dev);
+       if (ret)
+               goto err_unload_overlay;
+
+       return 0;
+
+err_unload_overlay:
+       lan966x_pci_unload_overlay(data);
+       return ret;
+}  

...

quoted

+#include <dt-bindings/clock/microchip,lan966x.h>
+#include <dt-bindings/interrupt-controller/irq.h>
+#include <dt-bindings/mfd/atmel-flexcom.h>
+#include <dt-bindings/phy/phy-lan966x-serdes.h>  

quoted

+#include <dt-bindings/gpio/gpio.h>  

Alphabetical order?

Yes indeed.

Thanks for the review.

Best regards,
Hervé