Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1331851 > unrolled thread

Re: [PATCH V4 22/23] arm64, pci, acpi: Assign legacy IRQs once device is enable.

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH V4 22/23] arm64, pci, acpi: Assign legacy IRQs once  device is enable. Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> - 2016-02-11 13:00 +0100
    Re: [PATCH V4 22/23] arm64, pci, acpi: Assign legacy IRQs once  device is enable. Lorenzo Pieralisi <lorenzo.pieralisi@arm.com> - 2016-02-11 18:20 +0100
      Re: [PATCH V4 22/23] arm64, pci, acpi: Assign legacy IRQs once device  is enable. Tomasz Nowicki <tn@semihalf.com> - 2016-02-11 19:50 +0100
    Re: [PATCH V4 22/23] arm64, pci, acpi: Assign legacy IRQs once device  is enable. Tomasz Nowicki <tn@semihalf.com> - 2016-02-11 19:50 +0100

#1331851 — Re: [PATCH V4 22/23] arm64, pci, acpi: Assign legacy IRQs once device is enable.

On Thu, Feb 04, 2016 at 06:29:00PM +0100, Tomasz Nowicki wrote:
> This is the last step before enabling generic ACPI PCI host controller
> for ARM64. We need to take care of legacy IRQ mapping for non-MSI(X)
> PCI devices. pcibios_enable_device() boot order is not sensitive to
> ACPI device enumeration, so it is the best place to assign device's IRQs.

I guess you are referring to:

https://lists.linaro.org/pipermail/linaro-acpi/2015-October/005944.html

It is weird that the dependency can't be enforced, I will have a look
into this, it would be nice to have DT and ACPI legacy IRQs mapping
confined in pcibios_add_device() so that we can remove them in one go
when Matthew's series is merged.

As for the MSI check, by reading commit history its need I think
it harks back to bba6f6fc, which was supposed to be a quick hack and it
has been in the kernel for 9 years :), is it really needed ?

> NOTE: *This is going to be temporary solution*. There is ongoing work
> which aims for cleaning legacy IRQ allocation, see [1].
> We can consider this patch as the necessary evil which will be removed
> once [1] series hits mailnline in the near future.
> 
> 1. http://comments.gmane.org/gmane.linux.kernel.pci/46461
> 
> Signed-off-by: Tomasz Nowicki <tn@semihalf.com>
> ---
>  arch/arm64/kernel/pci.c | 14 ++++++++++++--
>  1 file changed, 12 insertions(+), 2 deletions(-)
> 
> diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c
> index 023b983..d1a701f 100644
> --- a/arch/arm64/kernel/pci.c
> +++ b/arch/arm64/kernel/pci.c
> @@ -39,16 +39,26 @@ resource_size_t pcibios_align_resource(void *data, const struct resource *res,
>  }
>  
>  /**
> - * pcibios_enable_device - Enable I/O and memory.
> + * pcibios_enable_device - Enable I/O, memory and legacy IRQs for ACPI.
>   * @dev: PCI device to be enabled
>   * @mask: bitmask of BARs to enable
>   */
>  int pcibios_enable_device(struct pci_dev *dev, int mask)
>  {
> +	int ret;
> +
>  	if (pci_has_flag(PCI_PROBE_ONLY))
>  		return 0;
>  
> -	return pci_enable_resources(dev, mask);
> +	ret = pci_enable_resources(dev, mask);
> +	if (ret < 0)
> +		return ret;
> +
> +#ifdef CONFIG_ACPI
> +	if (!pci_dev_msi_enabled(dev))
> +		return acpi_pci_irq_enable(dev);

You need a guard here anyway, we do not want to probe ACPI IRQ if
it is disabled (ie acpi_pci_disabled). Same goes for DT in
pcibios_add_device(), which is not making this code any nicer.

Lorenzo

> +#endif
> +	return 0;
>  }
>  
>  /*
> -- 
> 1.9.1
> 

[toc] | [next] | [standalone]

#1332254

On Thu, Feb 11, 2016 at 11:58:53AM +0000, Lorenzo Pieralisi wrote:
> On Thu, Feb 04, 2016 at 06:29:00PM +0100, Tomasz Nowicki wrote:
> > This is the last step before enabling generic ACPI PCI host controller
> > for ARM64. We need to take care of legacy IRQ mapping for non-MSI(X)
> > PCI devices. pcibios_enable_device() boot order is not sensitive to
> > ACPI device enumeration, so it is the best place to assign device's IRQs.
> 
> I guess you are referring to:
> 
> https://lists.linaro.org/pipermail/linaro-acpi/2015-October/005944.html
> 
> It is weird that the dependency can't be enforced, I will have a look
> into this, it would be nice to have DT and ACPI legacy IRQs mapping
> confined in pcibios_add_device() so that we can remove them in one go
> when Matthew's series is merged.

One option, that is not ideal but has the merit of setting the stage
for pcibios_enable_device() AND pcibios_add_device() removal, is to add
IRQ mapping (by adding the call) in a pcibios_alloc_irq() callback (if
Bjorn does not remove it from core code before we manage to add it,
I think he is only reverting the x86 version).

https://lkml.org/lkml/2016/2/9/648

That's called at device probe time, it should not change DT probing path
(unless we use the irq number before the device is probed, which I doubt)
and should allow the ACPI scan handlers to be installed so that the ACPI
IRQ mapping can be effectively carried out.

pcibios_alloc_irq() replaces pcibios_add_device().

When Matthew's patchset lands in mainline, pcibios_alloc_irq() will
be removed too (at least we have a place where all legacy IRQ mappings
are carried out and I can obliterate it easily).

Other option is to change ACPI core code, I see no other way.

Tested on KVM PCI host generic (that uses pci_fixup_irqs() so it does
not really count for DT).

Here (on top of your series), ready for flak.

Lorenzo

-- >8 --
diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c
index 0b53262..26ee291 100644
--- a/arch/arm64/kernel/pci.c
+++ b/arch/arm64/kernel/pci.c
@@ -45,28 +45,23 @@ resource_size_t pcibios_align_resource(void *data, const struct resource *res,
  */
 int pcibios_enable_device(struct pci_dev *dev, int mask)
 {
-	int ret;
-
 	if (pci_has_flag(PCI_PROBE_ONLY))
 		return 0;
 
-	ret = pci_enable_resources(dev, mask);
-	if (ret < 0)
-		return ret;
-
-#ifdef CONFIG_ACPI
-	if (!pci_dev_msi_enabled(dev))
-		return acpi_pci_irq_enable(dev);
-#endif
-	return 0;
+	return pci_enable_resources(dev, mask);
 }
 
 /*
- * Try to assign the IRQ number from DT when adding a new device
+ * Try to assign the IRQ number when probing a new device
  */
-int pcibios_add_device(struct pci_dev *dev)
+int pcibios_alloc_irq(struct pci_dev *dev)
 {
-	dev->irq = of_irq_parse_and_map_pci(dev, 0, 0);
+	if (acpi_disabled)
+		dev->irq = of_irq_parse_and_map_pci(dev, 0, 0);
+#ifdef CONFIG_ACPI
+	else
+		return acpi_pci_irq_enable(dev);
+#endif
 
 	return 0;
 }

[toc] | [prev] | [next] | [standalone]

#1332305 — Re: [PATCH V4 22/23] arm64, pci, acpi: Assign legacy IRQs once device is enable.

On 11.02.2016 18:17, Lorenzo Pieralisi wrote:
> Here (on top of your series), ready for flak.
>
> Lorenzo
>
> -- >8 --
> diff --git a/arch/arm64/kernel/pci.c b/arch/arm64/kernel/pci.c
> index 0b53262..26ee291 100644
> --- a/arch/arm64/kernel/pci.c
> +++ b/arch/arm64/kernel/pci.c
> @@ -45,28 +45,23 @@ resource_size_t pcibios_align_resource(void *data, const struct resource *res,
>    */
>   int pcibios_enable_device(struct pci_dev *dev, int mask)
>   {
> -	int ret;
> -
>   	if (pci_has_flag(PCI_PROBE_ONLY))
>   		return 0;
>
> -	ret = pci_enable_resources(dev, mask);
> -	if (ret < 0)
> -		return ret;
> -
> -#ifdef CONFIG_ACPI
> -	if (!pci_dev_msi_enabled(dev))
> -		return acpi_pci_irq_enable(dev);
> -#endif
> -	return 0;
> +	return pci_enable_resources(dev, mask);
>   }
>
>   /*
> - * Try to assign the IRQ number from DT when adding a new device
> + * Try to assign the IRQ number when probing a new device
>    */
> -int pcibios_add_device(struct pci_dev *dev)
> +int pcibios_alloc_irq(struct pci_dev *dev)
>   {
> -	dev->irq = of_irq_parse_and_map_pci(dev, 0, 0);
> +	if (acpi_disabled)
> +		dev->irq = of_irq_parse_and_map_pci(dev, 0, 0);
> +#ifdef CONFIG_ACPI
> +	else
> +		return acpi_pci_irq_enable(dev);
> +#endif
>
>   	return 0;
>   }

I miss that way of solving the problem. OK lets try this way. I will 
integrate it.

Tomasz

[toc] | [prev] | [next] | [standalone]

#1332310 — Re: [PATCH V4 22/23] arm64, pci, acpi: Assign legacy IRQs once device is enable.

On 11.02.2016 12:58, Lorenzo Pieralisi wrote:
> As for the MSI check, by reading commit history its need I think
> it harks back to bba6f6fc, which was supposed to be a quick hack and it
> has been in the kernel for 9 years:), is it really needed ?

IMO, yes it is needed. We need to chose MSI vs IRQ, and MSI is 
preferable option.

Tomasz

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web