Re: [RFC PATCH v2 6/8] pci/arm: don't do iommu call for phantom functions

[Stewart Hildebrand]

On 5/12/23 03:28, Jan Beulich wrote:
> On 11.05.2023 21:16, Stewart Hildebrand wrote:
>> It's not necessary to add/remove/assign/deassign pci phantom functions
>> for the ARM SMMU drivers. All associated AXI stream IDs are added during
>> the iommu call for the base PCI device/function.
>>
>> However, the ARM SMMU drivers can cope with the extra/unnecessary calls just
>> fine, so this patch is RFC as it's not strictly required.
> 
> Tying the skipping to IS_ENABLED(CONFIG_HAS_DEVICE_TREE) goes against
> one of Julien's earlier comments, towards DT and ACPI wanting to
> co-exist at some point. So I think keeping the supposedly unnecessary
> calls is going to be unavoidable, unless you have a runtime property
> that you could check instead.

I'll drop this patch in v3

Re: [RFC PATCH v2 6/8] pci/arm: don't do iommu call for phantom functions

[Jan Beulich]

On 11.05.2023 21:16, Stewart Hildebrand wrote:
> It's not necessary to add/remove/assign/deassign pci phantom functions
> for the ARM SMMU drivers. All associated AXI stream IDs are added during
> the iommu call for the base PCI device/function.
> 
> However, the ARM SMMU drivers can cope with the extra/unnecessary calls just
> fine, so this patch is RFC as it's not strictly required.

Tying the skipping to IS_ENABLED(CONFIG_HAS_DEVICE_TREE) goes against
one of Julien's earlier comments, towards DT and ACPI wanting to
co-exist at some point. So I think keeping the supposedly unnecessary
calls is going to be unavoidable, unless you have a runtime property
that you could check instead.

Jan

[RFC PATCH v2 6/8] pci/arm: don't do iommu call for phantom functions

[Stewart Hildebrand]

It's not necessary to add/remove/assign/deassign pci phantom functions
for the ARM SMMU drivers. All associated AXI stream IDs are added during
the iommu call for the base PCI device/function.

However, the ARM SMMU drivers can cope with the extra/unnecessary calls just
fine, so this patch is RFC as it's not strictly required.

Signed-off-by: Stewart Hildebrand 
---
I'm aware the indentation is wrong. I just wanted to keep the diffstat small
while this particular patch is RFC.

v1->v2:
* new patch
---
 xen/drivers/passthrough/pci.c | 5 -
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/xen/drivers/passthrough/pci.c b/xen/drivers/passthrough/pci.c
index 6dbaae682773..3823edf096eb 100644
--- a/xen/drivers/passthrough/pci.c
+++ b/xen/drivers/passthrough/pci.c
@@ -871,6 +871,7 @@ static int deassign_device(struct domain *d, uint16_t seg, 
uint8_t bus,
 else
 target = hardware_domain;
 
+if ( !IS_ENABLED(CONFIG_HAS_DEVICE_TREE) )
 while ( pdev->phantom_stride )
 {
 devfn += pdev->phantom_stride;
@@ -1335,7 +1336,7 @@ static int iommu_add_device(struct pci_dev *pdev)
 return 0;
 
 rc = iommu_call(hd->platform_ops, add_device, devfn, pci_to_dev(pdev));
-if ( rc || !pdev->phantom_stride )
+if ( rc || !pdev->phantom_stride || IS_ENABLED(CONFIG_HAS_DEVICE_TREE) )
 return rc;
 
 for ( ; ; )
@@ -1379,6 +1380,7 @@ static int iommu_remove_device(struct pci_dev *pdev)
 if ( !is_iommu_enabled(pdev->domain) )
 return 0;
 
+if ( !IS_ENABLED(CONFIG_HAS_DEVICE_TREE) )
 for ( devfn = pdev->devfn ; pdev->phantom_stride; )
 {
 int rc;
@@ -1464,6 +1466,7 @@ static int assign_device(struct domain *d, u16 seg, u8 
bus, u8 devfn, u32 flag)
   pci_to_dev(pdev), flag)) )
 goto done;
 
+if ( !IS_ENABLED(CONFIG_HAS_DEVICE_TREE) )
 for ( ; pdev->phantom_stride; rc = 0 )
 {
 devfn += pdev->phantom_stride;
-- 
2.40.1