Introduce pcibios_free_irq() to free irq in pci_device_probe() and
pci_device_remove() that in drivers/pci/pci-driver.c.
Signed-off-by: Yajun Deng <[email protected]>
---
arch/arm64/kernel/pci.c | 10 +++++++++-
1 file changed, 9 insertions(+), 1 deletion(-)
diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c
index 1006ed2d7c60..40da5aff4548 100644
--- a/arch/arm64/kernel/pci.c
+++ b/arch/arm64/kernel/pci.c
@@ -25,10 +25,18 @@
int pcibios_alloc_irq(struct pci_dev *dev)
{
if (!acpi_disabled)
- acpi_pci_irq_enable(dev);
+ return acpi_pci_irq_enable(dev);
return 0;
}
+
+void pcibios_free_irq(struct pci_dev *dev)
+{
+ if (!acpi_disabled)
+ acpi_pci_irq_disable(dev);
+
+}
+
#endif
/*
--
2.32.0
[+Lorenzo]
On Wed, Aug 25, 2021 at 03:16:12PM +0800, Yajun Deng wrote:
> Introduce pcibios_free_irq() to free irq in pci_device_probe() and
> pci_device_remove() that in drivers/pci/pci-driver.c.
Please can you describe the problem you're solving?
> Signed-off-by: Yajun Deng <[email protected]>
> ---
> arch/arm64/kernel/pci.c | 10 +++++++++-
> 1 file changed, 9 insertions(+), 1 deletion(-)
>
> diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c
> index 1006ed2d7c60..40da5aff4548 100644
> --- a/arch/arm64/kernel/pci.c
> +++ b/arch/arm64/kernel/pci.c
> @@ -25,10 +25,18 @@
> int pcibios_alloc_irq(struct pci_dev *dev)
> {
> if (!acpi_disabled)
> - acpi_pci_irq_enable(dev);
> + return acpi_pci_irq_enable(dev);
This means we'll now fail device probe if we can't enable the GSI. Is that a
problem?
Will
August 25, 2021 6:05 PM, "Will Deacon" <[email protected]> wrote:
> [+Lorenzo]
>
> On Wed, Aug 25, 2021 at 03:16:12PM +0800, Yajun Deng wrote:
>
>> Introduce pcibios_free_irq() to free irq in pci_device_probe() and
>> pci_device_remove() that in drivers/pci/pci-driver.c.
>
> Please can you describe the problem you're solving?
>
pcibios_alloc_irq() will be called in pci_device_probe(), but there hasn't pcibios_free_irq()
in arm64 architecture correspond it. pcibios_free_irq() is an empty function in
drivers/pci/pci-driver.c.So pcibios_alloc_irq() and pcibios_free_irq() don't correspond.
>> Signed-off-by: Yajun Deng <[email protected]>
>> ---
>> arch/arm64/kernel/pci.c | 10 +++++++++-
>> 1 file changed, 9 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c
>> index 1006ed2d7c60..40da5aff4548 100644
>> --- a/arch/arm64/kernel/pci.c
>> +++ b/arch/arm64/kernel/pci.c
>> @@ -25,10 +25,18 @@
>> int pcibios_alloc_irq(struct pci_dev *dev)
>> {
>> if (!acpi_disabled)
>> - acpi_pci_irq_enable(dev);
>> + return acpi_pci_irq_enable(dev);
>
> This means we'll now fail device probe if we can't enable the GSI. Is that a
> problem?
>
Oh, It would be better that hasn't return.
> Will
On Wed, Aug 25, 2021 at 03:16:12PM +0800, Yajun Deng wrote:
> Introduce pcibios_free_irq() to free irq in pci_device_probe() and
> pci_device_remove() that in drivers/pci/pci-driver.c.
Add a rationale - it is just code inspection or you are fixing a bug ?
> Signed-off-by: Yajun Deng <[email protected]>
> ---
> arch/arm64/kernel/pci.c | 10 +++++++++-
> 1 file changed, 9 insertions(+), 1 deletion(-)
>
> diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c
> index 1006ed2d7c60..40da5aff4548 100644
> --- a/arch/arm64/kernel/pci.c
> +++ b/arch/arm64/kernel/pci.c
> @@ -25,10 +25,18 @@
> int pcibios_alloc_irq(struct pci_dev *dev)
> {
> if (!acpi_disabled)
> - acpi_pci_irq_enable(dev);
> + return acpi_pci_irq_enable(dev);
This is an unrelated change and it is potentially introducing
regressions. I need to page in the reasons why on arm64 we had to resort
to pcibios_alloc_irq() (probe ordering IIRC) - in the meanwhile
this function stays as it is.
> return 0;
> }
> +
> +void pcibios_free_irq(struct pci_dev *dev)
> +{
> + if (!acpi_disabled)
> + acpi_pci_irq_disable(dev);
> +
> +}
Adding pcibios_free_irq() makes sense and I believe it is a genuine
"fix".
Please add any information in the commit log that explains the
run-time condition you are fixing.
Thanks,
Lorenzo