[v2] Improvements to pcie_bandwidth_available() for eGPUs

[PATCH v2 5/9] PCI: pciehp: Move check for is_thunderbolt into a quirk

Posted by Mario Limonciello 2 years, 1 month ago

commit 493fb50e958c ("PCI: pciehp: Assume NoCompl+ for Thunderbolt
ports") added a check into pciehp code to explicitly set NoCompl+
for all Intel Thunderbolt controllers, including those that don't
need it.

This overloaded the purpose of the `is_thunderbolt` member of
`struct pci_device` because that means that any controller that
identifies as thunderbolt would set NoCompl+ even if it doesn't
suffer this deficiency. As that commit helpfully specifies all the
controllers with the problem, move them into a PCI quirk.

Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
---
 drivers/pci/hotplug/pciehp_hpc.c |  6 +-----
 drivers/pci/quirks.c             | 20 ++++++++++++++++++++
 include/linux/pci.h              |  1 +
 3 files changed, 22 insertions(+), 5 deletions(-)

diff --git a/drivers/pci/hotplug/pciehp_hpc.c b/drivers/pci/hotplug/pciehp_hpc.c
index fd713abdfb9f..23a92d681d1c 100644
--- a/drivers/pci/hotplug/pciehp_hpc.c
+++ b/drivers/pci/hotplug/pciehp_hpc.c
@@ -991,11 +991,7 @@ struct controller *pcie_init(struct pcie_device *dev)
 	if (pdev->hotplug_user_indicators)
 		slot_cap &= ~(PCI_EXP_SLTCAP_AIP | PCI_EXP_SLTCAP_PIP);
 
-	/*
-	 * We assume no Thunderbolt controllers support Command Complete events,
-	 * but some controllers falsely claim they do.
-	 */
-	if (pdev->is_thunderbolt)
+	if (pdev->no_command_complete)
 		slot_cap |= PCI_EXP_SLTCAP_NCCS;
 
 	ctrl->slot_cap = slot_cap;
diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
index eeec1d6f9023..4bbf6e33ca11 100644
--- a/drivers/pci/quirks.c
+++ b/drivers/pci/quirks.c
@@ -3807,6 +3807,26 @@ DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_CACTUS_RIDGE_4C
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_PORT_RIDGE,
 			quirk_thunderbolt_hotplug_msi);
 
+/*
+ * We assume no Thunderbolt controllers support Command Complete events,
+ * but some controllers falsely claim they do.
+ */
+static void quirk_thunderbolt_command_complete(struct pci_dev *pdev)
+{
+	pdev->no_command_complete = 1;
+}
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LIGHT_RIDGE,
+			quirk_thunderbolt_command_complete);
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_EAGLE_RIDGE,
+			quirk_thunderbolt_command_complete);
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LIGHT_PEAK,
+			quirk_thunderbolt_command_complete);
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_CACTUS_RIDGE_4C,
+			quirk_thunderbolt_command_complete);
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_CACTUS_RIDGE_2C,
+			quirk_thunderbolt_command_complete);
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_PORT_RIDGE,
+			quirk_thunderbolt_command_complete);
 #ifdef CONFIG_ACPI
 /*
  * Apple: Shutdown Cactus Ridge Thunderbolt controller.
diff --git a/include/linux/pci.h b/include/linux/pci.h
index 530b0a360514..439c2dac8a3e 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -441,6 +441,7 @@ struct pci_dev {
 	unsigned int	is_hotplug_bridge:1;
 	unsigned int	shpc_managed:1;		/* SHPC owned by shpchp */
 	unsigned int	is_thunderbolt:1;	/* Thunderbolt controller */
+	unsigned int	no_command_complete:1;	/* No command completion */
 	/*
 	 * Devices marked being untrusted are the ones that can potentially
 	 * execute DMA attacks and similar. They are typically connected
-- 
2.34.1

Re: [PATCH v2 5/9] PCI: pciehp: Move check for is_thunderbolt into a quirk

Posted by Ilpo Järvinen 2 years, 1 month ago

On Fri, 3 Nov 2023, Mario Limonciello wrote:

> commit 493fb50e958c ("PCI: pciehp: Assume NoCompl+ for Thunderbolt
> ports") added a check into pciehp code to explicitly set NoCompl+
> for all Intel Thunderbolt controllers, including those that don't
> need it.
> 
> This overloaded the purpose of the `is_thunderbolt` member of
> `struct pci_device` because that means that any controller that
> identifies as thunderbolt would set NoCompl+ even if it doesn't
> suffer this deficiency. As that commit helpfully specifies all the
> controllers with the problem, move them into a PCI quirk.
> 
> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
> ---
>  drivers/pci/hotplug/pciehp_hpc.c |  6 +-----
>  drivers/pci/quirks.c             | 20 ++++++++++++++++++++
>  include/linux/pci.h              |  1 +
>  3 files changed, 22 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/pci/hotplug/pciehp_hpc.c b/drivers/pci/hotplug/pciehp_hpc.c
> index fd713abdfb9f..23a92d681d1c 100644
> --- a/drivers/pci/hotplug/pciehp_hpc.c
> +++ b/drivers/pci/hotplug/pciehp_hpc.c
> @@ -991,11 +991,7 @@ struct controller *pcie_init(struct pcie_device *dev)
>  	if (pdev->hotplug_user_indicators)
>  		slot_cap &= ~(PCI_EXP_SLTCAP_AIP | PCI_EXP_SLTCAP_PIP);
>  
> -	/*
> -	 * We assume no Thunderbolt controllers support Command Complete events,
> -	 * but some controllers falsely claim they do.
> -	 */
> -	if (pdev->is_thunderbolt)
> +	if (pdev->no_command_complete)
>  		slot_cap |= PCI_EXP_SLTCAP_NCCS;
>  
>  	ctrl->slot_cap = slot_cap;
> diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
> index eeec1d6f9023..4bbf6e33ca11 100644
> --- a/drivers/pci/quirks.c
> +++ b/drivers/pci/quirks.c
> @@ -3807,6 +3807,26 @@ DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_CACTUS_RIDGE_4C
>  DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_PORT_RIDGE,
>  			quirk_thunderbolt_hotplug_msi);
>  
> +/*
> + * We assume no Thunderbolt controllers support Command Complete events,
> + * but some controllers falsely claim they do.

IMO, this wording makes little sense with the new code. How about taking 
some text from the original commit's changelog:

/*
 * Certain Thunderbolt 1 controllers falsely claim to support Command 
 * Completed events.
 */

The code change looks fine.

-- 
 i.

> + */
> +static void quirk_thunderbolt_command_complete(struct pci_dev *pdev)
> +{
> +	pdev->no_command_complete = 1;
> +}
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LIGHT_RIDGE,
> +			quirk_thunderbolt_command_complete);
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_EAGLE_RIDGE,
> +			quirk_thunderbolt_command_complete);
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_LIGHT_PEAK,
> +			quirk_thunderbolt_command_complete);
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_CACTUS_RIDGE_4C,
> +			quirk_thunderbolt_command_complete);
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_CACTUS_RIDGE_2C,
> +			quirk_thunderbolt_command_complete);
> +DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_PORT_RIDGE,
> +			quirk_thunderbolt_command_complete);
>  #ifdef CONFIG_ACPI
>  /*
>   * Apple: Shutdown Cactus Ridge Thunderbolt controller.
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index 530b0a360514..439c2dac8a3e 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -441,6 +441,7 @@ struct pci_dev {
>  	unsigned int	is_hotplug_bridge:1;
>  	unsigned int	shpc_managed:1;		/* SHPC owned by shpchp */
>  	unsigned int	is_thunderbolt:1;	/* Thunderbolt controller */
> +	unsigned int	no_command_complete:1;	/* No command completion */
>  	/*
>  	 * Devices marked being untrusted are the ones that can potentially
>  	 * execute DMA attacks and similar. They are typically connected
>

Re: [PATCH v2 5/9] PCI: pciehp: Move check for is_thunderbolt into a quirk

Posted by Mario Limonciello 2 years, 1 month ago

On 11/6/2023 06:41, Ilpo Järvinen wrote:
> On Fri, 3 Nov 2023, Mario Limonciello wrote:
> 
>> commit 493fb50e958c ("PCI: pciehp: Assume NoCompl+ for Thunderbolt
>> ports") added a check into pciehp code to explicitly set NoCompl+
>> for all Intel Thunderbolt controllers, including those that don't
>> need it.
>>
>> This overloaded the purpose of the `is_thunderbolt` member of
>> `struct pci_device` because that means that any controller that
>> identifies as thunderbolt would set NoCompl+ even if it doesn't
>> suffer this deficiency. As that commit helpfully specifies all the
>> controllers with the problem, move them into a PCI quirk.
>>
>> Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
>> ---
>>   drivers/pci/hotplug/pciehp_hpc.c |  6 +-----
>>   drivers/pci/quirks.c             | 20 ++++++++++++++++++++
>>   include/linux/pci.h              |  1 +
>>   3 files changed, 22 insertions(+), 5 deletions(-)
>>
>> diff --git a/drivers/pci/hotplug/pciehp_hpc.c b/drivers/pci/hotplug/pciehp_hpc.c
>> index fd713abdfb9f..23a92d681d1c 100644
>> --- a/drivers/pci/hotplug/pciehp_hpc.c
>> +++ b/drivers/pci/hotplug/pciehp_hpc.c
>> @@ -991,11 +991,7 @@ struct controller *pcie_init(struct pcie_device *dev)
>>   	if (pdev->hotplug_user_indicators)
>>   		slot_cap &= ~(PCI_EXP_SLTCAP_AIP | PCI_EXP_SLTCAP_PIP);
>>   
>> -	/*
>> -	 * We assume no Thunderbolt controllers support Command Complete events,
>> -	 * but some controllers falsely claim they do.
>> -	 */
>> -	if (pdev->is_thunderbolt)
>> +	if (pdev->no_command_complete)
>>   		slot_cap |= PCI_EXP_SLTCAP_NCCS;
>>   
>>   	ctrl->slot_cap = slot_cap;
>> diff --git a/drivers/pci/quirks.c b/drivers/pci/quirks.c
>> index eeec1d6f9023..4bbf6e33ca11 100644
>> --- a/drivers/pci/quirks.c
>> +++ b/drivers/pci/quirks.c
>> @@ -3807,6 +3807,26 @@ DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_CACTUS_RIDGE_4C
>>   DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_INTEL, PCI_DEVICE_ID_INTEL_PORT_RIDGE,
>>   			quirk_thunderbolt_hotplug_msi);
>>   
>> +/*
>> + * We assume no Thunderbolt controllers support Command Complete events,
>> + * but some controllers falsely claim they do.
> 
> IMO, this wording makes little sense with the new code. How about taking
> some text from the original commit's changelog:
> 
> /*
>   * Certain Thunderbolt 1 controllers falsely claim to support Command
>   * Completed events.
>   */
> 
> The code change looks fine.
> 

Ack, will change.