[v2,1/2] PCI: dra7xx: Fix legacy INTD IRQ handling
diff mbox

Message ID 493a4f80-677c-b213-d13a-e922b28c38e0@ti.com
State New
Headers show

Commit Message

Vignesh Raghavendra Jan. 8, 2018, 1:48 p.m. UTC
On Thursday 04 January 2018 09:42 PM, Lorenzo Pieralisi wrote:
> On Thu, Jan 04, 2018 at 07:04:30PM +0530, Kishon Vijay Abraham I wrote:
>> Hi,
>>
>> On Thursday 04 January 2018 11:38 AM, Kishon Vijay Abraham I wrote:
>>>
>>>
>>> On Friday 29 December 2017 05:11 PM, Vignesh R wrote:
>>>> Legacy INTD IRQ handling is broken on dra7xx due to fact that driver
>>>> uses hwirq in range of 1-4 for INTA, INTD whereas IRQ domain is of size
>>>> 4 which is numbered 0-3. Therefore when INTD IRQ line is used with
>>>> pci-dra7xx driver following warning is seen:
>>>>
>>>>        WARNING: CPU: 0 PID: 1 at kernel/irq/irqdomain.c:342 irq_domain_associate+0x12c/0x1c4
>>>>        error: hwirq 0x4 is too large for dummy
>>>>
>>>> Fix this by using pci_irqd_intx_xlate() helper to translate the INTx 1-4
>>>> range into the 0-3 as done in other PCIe drivers.
>>>>
>>>> Suggested-by: Bjorn Helgaas <bhelgaas@google.com>
>>>> Reported-by: Chris Welch <Chris.Welch@viavisolutions.com>
>>>> Signed-off-by: Vignesh R <vigneshr@ti.com>
>>>
>>> Acked-by: Kishon Vijay Abraham I <kishon@ti.com>
>>>> ---
>>>>  drivers/pci/dwc/pci-dra7xx.c | 3 ++-
>>>>  1 file changed, 2 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/drivers/pci/dwc/pci-dra7xx.c b/drivers/pci/dwc/pci-dra7xx.c
>>>> index 58aed0896468..892f93910012 100644
>>>> --- a/drivers/pci/dwc/pci-dra7xx.c
>>>> +++ b/drivers/pci/dwc/pci-dra7xx.c
>>>> @@ -226,6 +226,7 @@ static int dra7xx_pcie_intx_map(struct irq_domain *domain, unsigned int irq,
>>>>  
>>>>  static const struct irq_domain_ops intx_domain_ops = {
>>>>  	.map = dra7xx_pcie_intx_map,
>>>> +	.xlate = pci_irqd_intx_xlate,
>>
>> Looking at this again, I think the mapping should be done in device tree using
>> the interrupt-map property.
>>
>> Lorenzo, can you hold merging this patch till we verify using the dt approach?
> 
> Yes and I agree that's a DT mapping bug.

I did some testing by changing interrupt-map property of PCIe node as as
below:
call the correct handler with above change:

static int dra7xx_pcie_init_irq_domain(struct pcie_port *pp)
@@ -269,7 +270,7 @@ static irqreturn_t dra7xx_pcie_msi_irq_handler(int
irq, void *arg)
 	case INTC:
 	case INTD:
 		generic_handle_irq(irq_find_mapping(dra7xx->irq_domain,
-						    ffs(reg)));
+						    ffs(reg) - 1));
 		break;
 	}


If we merge the driver change then the DT backward compatibility is
broken. Right now INTA/B/C keep working even with buggy interrupt-map
settings. But, if we fix  interrupt-map and merge above driver change,
then we will be *breaking* DT backward compatibility and INTA/B/C will
no longer work with older DTs.
Therefore, IMO, this patch should be merged in order to keep DT backward
compatibility and live with wrong DT representation.

Comments

Lorenzo Pieralisi Jan. 8, 2018, 6:24 p.m. UTC | #1
On Mon, Jan 08, 2018 at 07:18:59PM +0530, Vignesh R wrote:
> 
> 
> On Thursday 04 January 2018 09:42 PM, Lorenzo Pieralisi wrote:
> > On Thu, Jan 04, 2018 at 07:04:30PM +0530, Kishon Vijay Abraham I wrote:
> >> Hi,
> >>
> >> On Thursday 04 January 2018 11:38 AM, Kishon Vijay Abraham I wrote:
> >>>
> >>>
> >>> On Friday 29 December 2017 05:11 PM, Vignesh R wrote:
> >>>> Legacy INTD IRQ handling is broken on dra7xx due to fact that driver
> >>>> uses hwirq in range of 1-4 for INTA, INTD whereas IRQ domain is of size
> >>>> 4 which is numbered 0-3. Therefore when INTD IRQ line is used with
> >>>> pci-dra7xx driver following warning is seen:
> >>>>
> >>>>        WARNING: CPU: 0 PID: 1 at kernel/irq/irqdomain.c:342 irq_domain_associate+0x12c/0x1c4
> >>>>        error: hwirq 0x4 is too large for dummy
> >>>>
> >>>> Fix this by using pci_irqd_intx_xlate() helper to translate the INTx 1-4
> >>>> range into the 0-3 as done in other PCIe drivers.
> >>>>
> >>>> Suggested-by: Bjorn Helgaas <bhelgaas@google.com>
> >>>> Reported-by: Chris Welch <Chris.Welch@viavisolutions.com>
> >>>> Signed-off-by: Vignesh R <vigneshr@ti.com>
> >>>
> >>> Acked-by: Kishon Vijay Abraham I <kishon@ti.com>
> >>>> ---
> >>>>  drivers/pci/dwc/pci-dra7xx.c | 3 ++-
> >>>>  1 file changed, 2 insertions(+), 1 deletion(-)
> >>>>
> >>>> diff --git a/drivers/pci/dwc/pci-dra7xx.c b/drivers/pci/dwc/pci-dra7xx.c
> >>>> index 58aed0896468..892f93910012 100644
> >>>> --- a/drivers/pci/dwc/pci-dra7xx.c
> >>>> +++ b/drivers/pci/dwc/pci-dra7xx.c
> >>>> @@ -226,6 +226,7 @@ static int dra7xx_pcie_intx_map(struct irq_domain *domain, unsigned int irq,
> >>>>  
> >>>>  static const struct irq_domain_ops intx_domain_ops = {
> >>>>  	.map = dra7xx_pcie_intx_map,
> >>>> +	.xlate = pci_irqd_intx_xlate,
> >>
> >> Looking at this again, I think the mapping should be done in device tree using
> >> the interrupt-map property.
> >>
> >> Lorenzo, can you hold merging this patch till we verify using the dt approach?
> > 
> > Yes and I agree that's a DT mapping bug.
> 
> I did some testing by changing interrupt-map property of PCIe node as as
> below:
> diff --git a/arch/arm/boot/dts/dra7.dtsi b/arch/arm/boot/dts/dra7.dtsi
> index 261b5deb2456..2b1a2c5303bb 100644
> --- a/arch/arm/boot/dts/dra7.dtsi
> +++ b/arch/arm/boot/dts/dra7.dtsi
> @@ -332,10 +332,10 @@
>                                 phys = <&pcie1_phy>;
>                                 phy-names = "pcie-phy0";
>                                 interrupt-map-mask = <0 0 0 7>;
> -                               interrupt-map = <0 0 0 1 &pcie1_intc 1>,
> -                                               <0 0 0 2 &pcie1_intc 2>,
> -                                               <0 0 0 3 &pcie1_intc 3>,
> -                                               <0 0 0 4 &pcie1_intc 4>;
> +                               interrupt-map = <0 0 0 1 &pcie1_intc 0>,
> +                                               <0 0 0 2 &pcie1_intc 1>,
> +                                               <0 0 0 3 &pcie1_intc 2>,
> +                                               <0 0 0 4 &pcie1_intc 3>;
> 
> 
> Although this eliminates need for xlate, we still need driver change to
> call the correct handler with above change:
> 
> static int dra7xx_pcie_init_irq_domain(struct pcie_port *pp)
> @@ -269,7 +270,7 @@ static irqreturn_t dra7xx_pcie_msi_irq_handler(int
> irq, void *arg)
>  	case INTC:
>  	case INTD:
>  		generic_handle_irq(irq_find_mapping(dra7xx->irq_domain,
> -						    ffs(reg)));
> +						    ffs(reg) - 1));
>  		break;
>  	}
> 
> 
> If we merge the driver change then the DT backward compatibility is
> broken. Right now INTA/B/C keep working even with buggy interrupt-map
> settings. But, if we fix  interrupt-map and merge above driver change,
> then we will be *breaking* DT backward compatibility and INTA/B/C will
> no longer work with older DTs.
> Therefore, IMO, this patch should be merged in order to keep DT backward
> compatibility and live with wrong DT representation.

Yes, this is a mess. I am not fond at all of this fix but I will end
up merging it for the same reasons you provided - I can't break
backward DT compatibility.

Lorenzo
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch
diff mbox

diff --git a/arch/arm/boot/dts/dra7.dtsi b/arch/arm/boot/dts/dra7.dtsi
index 261b5deb2456..2b1a2c5303bb 100644
--- a/arch/arm/boot/dts/dra7.dtsi
+++ b/arch/arm/boot/dts/dra7.dtsi
@@ -332,10 +332,10 @@ 
                                phys = <&pcie1_phy>;
                                phy-names = "pcie-phy0";
                                interrupt-map-mask = <0 0 0 7>;
-                               interrupt-map = <0 0 0 1 &pcie1_intc 1>,
-                                               <0 0 0 2 &pcie1_intc 2>,
-                                               <0 0 0 3 &pcie1_intc 3>,
-                                               <0 0 0 4 &pcie1_intc 4>;
+                               interrupt-map = <0 0 0 1 &pcie1_intc 0>,
+                                               <0 0 0 2 &pcie1_intc 1>,
+                                               <0 0 0 3 &pcie1_intc 2>,
+                                               <0 0 0 4 &pcie1_intc 3>;


Although this eliminates need for xlate, we still need driver change to