[RFC 0/5] Clean up VMD DMA Map Ops

[RFC 0/5] Clean up VMD DMA Map Ops

[Jon Derrick]

Inspired by Christoph's last set:
https://lkml.org/lkml/2019/8/28/667

VMD currently works with VT-d enabled by pointing DMA and IOMMU actions at the
VMD endpoint. The problem with this approach is that the VMD endpoint's
device-specific attributes, such as the dma mask, are used instead.

This set cleans up VMD by removing the override that redirects dma map
operations to the VMD endpoint. Instead it introduces a new dma alias mechanism
into the existing dma alias infrastructure.

Patch 1 and 2 are miscellaneous fixes discovered during development.
Patch 1 is ready, but 2 likely doesn't go far enough for proper teardown on
addition failure.

Jon Derrick (5):
  iommu: Remove device link to group on failure
  iommu/vt-d: Unlink device if failed to add to group
  x86/PCI: Expose VMD's device in pci_sysdata
  PCI: vmd: Stop overriding dma_map_ops
  x86/PCI: Remove unused X86_DEV_DMA_OPS

 arch/x86/Kconfig               |   3 -
 arch/x86/include/asm/device.h  |  10 ---
 arch/x86/include/asm/pci.h     |   4 +-
 arch/x86/pci/common.c          |  44 ++----------
 drivers/iommu/intel-iommu.c    |  26 ++++---
 drivers/iommu/iommu.c          |   1 +
 drivers/pci/controller/Kconfig |   1 -
 drivers/pci/controller/vmd.c   | 152 +----------------------------------------
 drivers/pci/pci.c              |   4 +-
 drivers/pci/search.c           |   6 ++
 include/linux/pci.h            |   1 +
 11 files changed, 37 insertions(+), 215 deletions(-)

-- 
1.8.3.1

_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

[RFC 1/5] iommu: Remove device link to group on failure

[Jon Derrick]

This adds the missing teardown step that removes the device link from
the group when the device addition fails.

Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
---
 drivers/iommu/iommu.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index d5174f0..3e35284 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c
@@ -768,6 +768,7 @@ int iommu_group_add_device(struct iommu_group *group, struct device *dev)
 	mutex_unlock(&group->mutex);
 	dev->iommu_group = NULL;
 	kobject_put(group->devices_kobj);
+	sysfs_remove_link(group->devices_kobj, device->name);
 err_free_name:
 	kfree(device->name);
 err_remove_link:
-- 
1.8.3.1

_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 1/5] iommu: Remove device link to group on failure

[Lu Baolu]

Hi,

On 1/1/20 4:24 AM, Jon Derrick wrote:
> This adds the missing teardown step that removes the device link from
> the group when the device addition fails.

This change looks good to me.

Reviewed-by: Lu Baolu <baolu.lu@linux.intel.com>

Best regards,
baolu

> 
> Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
> ---
>   drivers/iommu/iommu.c | 1 +
>   1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
> index d5174f0..3e35284 100644
> --- a/drivers/iommu/iommu.c
> +++ b/drivers/iommu/iommu.c
> @@ -768,6 +768,7 @@ int iommu_group_add_device(struct iommu_group *group, struct device *dev)
>   	mutex_unlock(&group->mutex);
>   	dev->iommu_group = NULL;
>   	kobject_put(group->devices_kobj);
> +	sysfs_remove_link(group->devices_kobj, device->name);
>   err_free_name:
>   	kfree(device->name);
>   err_remove_link:
> 
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

[RFC 2/5] iommu/vt-d: Unlink device if failed to add to group

[Jon Derrick]

If the device fails to be added to the group, make sure to unlink the
reference before returning.

Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
---
 drivers/iommu/intel-iommu.c | 13 ++++++++++---
 1 file changed, 10 insertions(+), 3 deletions(-)

diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
index b2526a4..978d502 100644
--- a/drivers/iommu/intel-iommu.c
+++ b/drivers/iommu/intel-iommu.c
@@ -5625,8 +5625,10 @@ static int intel_iommu_add_device(struct device *dev)
 
 	group = iommu_group_get_for_dev(dev);
 
-	if (IS_ERR(group))
-		return PTR_ERR(group);
+	if (IS_ERR(group)) {
+		ret = PTR_ERR(group);
+		goto unlink;
+	}
 
 	iommu_group_put(group);
 
@@ -5652,7 +5654,8 @@ static int intel_iommu_add_device(struct device *dev)
 				if (!get_private_domain_for_dev(dev)) {
 					dev_warn(dev,
 						 "Failed to get a private domain.\n");
-					return -ENOMEM;
+					ret = -ENOMEM;
+					goto unlink;
 				}
 
 				dev_info(dev,
@@ -5667,6 +5670,10 @@ static int intel_iommu_add_device(struct device *dev)
 	}
 
 	return 0;
+
+unlink:
+	iommu_device_unlink(&iommu->iommu, dev);
+	return ret;
 }
 
 static void intel_iommu_remove_device(struct device *dev)
-- 
1.8.3.1

_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 2/5] iommu/vt-d: Unlink device if failed to add to group

[Lu Baolu]

On 1/1/20 4:24 AM, Jon Derrick wrote:
> If the device fails to be added to the group, make sure to unlink the
> reference before returning.
> 
> Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>

This fix looks reasonable to me.

Acked-by: Lu Baolu <baolu.lu@linux.intel.com>

Best regards,
baolu

> ---
>   drivers/iommu/intel-iommu.c | 13 ++++++++++---
>   1 file changed, 10 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
> index b2526a4..978d502 100644
> --- a/drivers/iommu/intel-iommu.c
> +++ b/drivers/iommu/intel-iommu.c
> @@ -5625,8 +5625,10 @@ static int intel_iommu_add_device(struct device *dev)
>   
>   	group = iommu_group_get_for_dev(dev);
>   
> -	if (IS_ERR(group))
> -		return PTR_ERR(group);
> +	if (IS_ERR(group)) {
> +		ret = PTR_ERR(group);
> +		goto unlink;
> +	}
>   
>   	iommu_group_put(group);
>   
> @@ -5652,7 +5654,8 @@ static int intel_iommu_add_device(struct device *dev)
>   				if (!get_private_domain_for_dev(dev)) {
>   					dev_warn(dev,
>   						 "Failed to get a private domain.\n");
> -					return -ENOMEM;
> +					ret = -ENOMEM;
> +					goto unlink;
>   				}
>   
>   				dev_info(dev,
> @@ -5667,6 +5670,10 @@ static int intel_iommu_add_device(struct device *dev)
>   	}
>   
>   	return 0;
> +
> +unlink:
> +	iommu_device_unlink(&iommu->iommu, dev);
> +	return ret;
>   }
>   
>   static void intel_iommu_remove_device(struct device *dev)
> 
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 2/5] iommu/vt-d: Unlink device if failed to add to group

[Lu Baolu]

On 1/1/20 4:24 AM, Jon Derrick wrote:
> If the device fails to be added to the group, make sure to unlink the
> reference before returning.
> 
> Signed-off-by: Jon Derrick<jonathan.derrick@intel.com>

Queued for v5.6.

Best regards,
baolu
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 2/5] iommu/vt-d: Unlink device if failed to add to group

[Joerg Roedel]

On Sun, Jan 12, 2020 at 09:36:56AM +0800, Lu Baolu wrote:
> On 1/1/20 4:24 AM, Jon Derrick wrote:
> > If the device fails to be added to the group, make sure to unlink the
> > reference before returning.
> > 
> > Signed-off-by: Jon Derrick<jonathan.derrick@intel.com>
> 
> Queued for v5.6.

No need to do so, I sent it upstream with the last pile of iommu fixes.


Thanks,

	Joerg

_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 2/5] iommu/vt-d: Unlink device if failed to add to group

[Lu Baolu]

On 1/13/20 8:20 PM, Joerg Roedel wrote:
> On Sun, Jan 12, 2020 at 09:36:56AM +0800, Lu Baolu wrote:
>> On 1/1/20 4:24 AM, Jon Derrick wrote:
>>> If the device fails to be added to the group, make sure to unlink the
>>> reference before returning.
>>>
>>> Signed-off-by: Jon Derrick<jonathan.derrick@intel.com>
>>
>> Queued for v5.6.
> 
> No need to do so, I sent it upstream with the last pile of iommu fixes.

Got it. Thank you!

Best regards,
baolu
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

[RFC 3/5] x86/PCI: Expose VMD's device in pci_sysdata

[Jon Derrick]

To be used by intel-iommu code to find the correct domain.

Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
---
 arch/x86/include/asm/pci.h   | 4 ++--
 drivers/pci/controller/vmd.c | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/arch/x86/include/asm/pci.h b/arch/x86/include/asm/pci.h
index 90d0731..7656807 100644
--- a/arch/x86/include/asm/pci.h
+++ b/arch/x86/include/asm/pci.h
@@ -25,7 +25,7 @@ struct pci_sysdata {
 	void		*fwnode;	/* IRQ domain for MSI assignment */
 #endif
 #if IS_ENABLED(CONFIG_VMD)
-	bool vmd_domain;		/* True if in Intel VMD domain */
+	struct device *vmd_dev;		/* Non-null if in Intel VMD domain */
 #endif
 };
 
@@ -65,7 +65,7 @@ static inline bool is_vmd(struct pci_bus *bus)
 #if IS_ENABLED(CONFIG_VMD)
 	struct pci_sysdata *sd = bus->sysdata;
 
-	return sd->vmd_domain;
+	return !!sd->vmd_dev;
 #else
 	return false;
 #endif
diff --git a/drivers/pci/controller/vmd.c b/drivers/pci/controller/vmd.c
index 2128422..907b5bd 100644
--- a/drivers/pci/controller/vmd.c
+++ b/drivers/pci/controller/vmd.c
@@ -679,7 +679,7 @@ static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
 		.parent = res,
 	};
 
-	sd->vmd_domain = true;
+	sd->vmd_dev = &vmd->dev->dev;
 	sd->domain = vmd_find_free_domain();
 	if (sd->domain < 0)
 		return sd->domain;
-- 
1.8.3.1

_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 3/5] x86/PCI: Expose VMD's device in pci_sysdata

[Christoph Hellwig]

On Tue, Dec 31, 2019 at 01:24:21PM -0700, Jon Derrick wrote:
> To be used by intel-iommu code to find the correct domain.

Any reason to prefer this version over my patches 2 and 3 from the
series in August?
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 3/5] x86/PCI: Expose VMD's device in pci_sysdata

[Derrick, Jonathan]

On Thu, 2020-01-09 at 15:33 +0100, Christoph Hellwig wrote:
> On Tue, Dec 31, 2019 at 01:24:21PM -0700, Jon Derrick wrote:
> > To be used by intel-iommu code to find the correct domain.
> 
> Any reason to prefer this version over my patches 2 and 3 from the
> series in August?

Mine uses the correct device's dma mask
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 3/5] x86/PCI: Expose VMD's device in pci_sysdata

[Derrick, Jonathan]

On Thu, 2020-01-09 at 15:33 +0100, Christoph Hellwig wrote:
> On Tue, Dec 31, 2019 at 01:24:21PM -0700, Jon Derrick wrote:
> > To be used by intel-iommu code to find the correct domain.
> 
> Any reason to prefer this version over my patches 2 and 3 from the
> series in August?

2 & 3 of your set is fine.
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

[RFC 4/5] PCI: vmd: Stop overriding dma_map_ops

[Jon Derrick]

Devices on the VMD domain use the VMD endpoint's requester-id and have
been relying on the VMD endpoint's dma operations. The downside of this
was that VMD domain devices would use the VMD endpoint's attributes when
doing DMA and IOMMU mapping. We can be smarter about this by only using
the VMD endpoint when mapping and providing the correct child device's
attributes during dma operations.

This patch adds a new dma alias mechanism by adding a hint to a pci_dev
to point to a singular DMA requester's pci_dev. This integrates into the
existing dma alias infrastructure to reduce the impact of the changes
required to support this mode.

Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
---
 arch/x86/pci/common.c          |   6 +-
 drivers/iommu/intel-iommu.c    |  13 ++--
 drivers/pci/controller/Kconfig |   1 -
 drivers/pci/controller/vmd.c   | 150 -----------------------------------------
 drivers/pci/pci.c              |   4 +-
 drivers/pci/search.c           |   6 ++
 include/linux/pci.h            |   1 +
 7 files changed, 23 insertions(+), 158 deletions(-)

diff --git a/arch/x86/pci/common.c b/arch/x86/pci/common.c
index 1e59df0..4022609 100644
--- a/arch/x86/pci/common.c
+++ b/arch/x86/pci/common.c
@@ -664,8 +664,12 @@ static void set_dma_domain_ops(struct pci_dev *pdev) {}
 
 static void set_dev_domain_options(struct pci_dev *pdev)
 {
-	if (is_vmd(pdev->bus))
+	if (is_vmd(pdev->bus)) {
+		struct pci_sysdata *sd = pdev->bus->sysdata;
+
+		pdev->dma_parent = to_pci_dev(sd->vmd_dev);
 		pdev->hotplug_user_indicators = 1;
+	}
 }
 
 int pcibios_add_device(struct pci_dev *dev)
diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
index 978d502..5aee648 100644
--- a/drivers/iommu/intel-iommu.c
+++ b/drivers/iommu/intel-iommu.c
@@ -776,11 +776,8 @@ static struct intel_iommu *device_to_iommu(struct device *dev, u8 *bus, u8 *devf
 
 		pdev = to_pci_dev(dev);
 
-#ifdef CONFIG_X86
-		/* VMD child devices currently cannot be handled individually */
-		if (is_vmd(pdev->bus))
-			return NULL;
-#endif
+		if (pdev->dma_parent)
+			pdev = pdev->dma_parent;
 
 		/* VFs aren't listed in scope tables; we need to look up
 		 * the PF instead to find the IOMMU. */
@@ -2428,6 +2425,12 @@ static struct dmar_domain *find_domain(struct device *dev)
 		     dev->archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO))
 		return NULL;
 
+	if (dev_is_pci(dev)) {
+		struct pci_dev *pdev = to_pci_dev(dev);
+		if (pdev->dma_parent)
+			dev = &pdev->dma_parent->dev;
+	}
+
 	/* No lock here, assumes no domain exit in normal case */
 	info = dev->archdata.iommu;
 	if (likely(info))
diff --git a/drivers/pci/controller/Kconfig b/drivers/pci/controller/Kconfig
index c77069c..55671429 100644
--- a/drivers/pci/controller/Kconfig
+++ b/drivers/pci/controller/Kconfig
@@ -239,7 +239,6 @@ config PCIE_TANGO_SMP8759
 
 config VMD
 	depends on PCI_MSI && X86_64 && SRCU
-	select X86_DEV_DMA_OPS
 	tristate "Intel Volume Management Device Driver"
 	---help---
 	  Adds support for the Intel Volume Management Device (VMD). VMD is a
diff --git a/drivers/pci/controller/vmd.c b/drivers/pci/controller/vmd.c
index 907b5bd..5824a39 100644
--- a/drivers/pci/controller/vmd.c
+++ b/drivers/pci/controller/vmd.c
@@ -98,9 +98,6 @@ struct vmd_dev {
 	struct irq_domain	*irq_domain;
 	struct pci_bus		*bus;
 	u8			busn_start;
-
-	struct dma_map_ops	dma_ops;
-	struct dma_domain	dma_domain;
 };
 
 static inline struct vmd_dev *vmd_from_bus(struct pci_bus *bus)
@@ -295,151 +292,6 @@ static void vmd_set_desc(msi_alloc_info_t *arg, struct msi_desc *desc)
 	.chip		= &vmd_msi_controller,
 };
 
-/*
- * VMD replaces the requester ID with its own.  DMA mappings for devices in a
- * VMD domain need to be mapped for the VMD, not the device requiring
- * the mapping.
- */
-static struct device *to_vmd_dev(struct device *dev)
-{
-	struct pci_dev *pdev = to_pci_dev(dev);
-	struct vmd_dev *vmd = vmd_from_bus(pdev->bus);
-
-	return &vmd->dev->dev;
-}
-
-static void *vmd_alloc(struct device *dev, size_t size, dma_addr_t *addr,
-		       gfp_t flag, unsigned long attrs)
-{
-	return dma_alloc_attrs(to_vmd_dev(dev), size, addr, flag, attrs);
-}
-
-static void vmd_free(struct device *dev, size_t size, void *vaddr,
-		     dma_addr_t addr, unsigned long attrs)
-{
-	return dma_free_attrs(to_vmd_dev(dev), size, vaddr, addr, attrs);
-}
-
-static int vmd_mmap(struct device *dev, struct vm_area_struct *vma,
-		    void *cpu_addr, dma_addr_t addr, size_t size,
-		    unsigned long attrs)
-{
-	return dma_mmap_attrs(to_vmd_dev(dev), vma, cpu_addr, addr, size,
-			attrs);
-}
-
-static int vmd_get_sgtable(struct device *dev, struct sg_table *sgt,
-			   void *cpu_addr, dma_addr_t addr, size_t size,
-			   unsigned long attrs)
-{
-	return dma_get_sgtable_attrs(to_vmd_dev(dev), sgt, cpu_addr, addr, size,
-			attrs);
-}
-
-static dma_addr_t vmd_map_page(struct device *dev, struct page *page,
-			       unsigned long offset, size_t size,
-			       enum dma_data_direction dir,
-			       unsigned long attrs)
-{
-	return dma_map_page_attrs(to_vmd_dev(dev), page, offset, size, dir,
-			attrs);
-}
-
-static void vmd_unmap_page(struct device *dev, dma_addr_t addr, size_t size,
-			   enum dma_data_direction dir, unsigned long attrs)
-{
-	dma_unmap_page_attrs(to_vmd_dev(dev), addr, size, dir, attrs);
-}
-
-static int vmd_map_sg(struct device *dev, struct scatterlist *sg, int nents,
-		      enum dma_data_direction dir, unsigned long attrs)
-{
-	return dma_map_sg_attrs(to_vmd_dev(dev), sg, nents, dir, attrs);
-}
-
-static void vmd_unmap_sg(struct device *dev, struct scatterlist *sg, int nents,
-			 enum dma_data_direction dir, unsigned long attrs)
-{
-	dma_unmap_sg_attrs(to_vmd_dev(dev), sg, nents, dir, attrs);
-}
-
-static void vmd_sync_single_for_cpu(struct device *dev, dma_addr_t addr,
-				    size_t size, enum dma_data_direction dir)
-{
-	dma_sync_single_for_cpu(to_vmd_dev(dev), addr, size, dir);
-}
-
-static void vmd_sync_single_for_device(struct device *dev, dma_addr_t addr,
-				       size_t size, enum dma_data_direction dir)
-{
-	dma_sync_single_for_device(to_vmd_dev(dev), addr, size, dir);
-}
-
-static void vmd_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
-				int nents, enum dma_data_direction dir)
-{
-	dma_sync_sg_for_cpu(to_vmd_dev(dev), sg, nents, dir);
-}
-
-static void vmd_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
-				   int nents, enum dma_data_direction dir)
-{
-	dma_sync_sg_for_device(to_vmd_dev(dev), sg, nents, dir);
-}
-
-static int vmd_dma_supported(struct device *dev, u64 mask)
-{
-	return dma_supported(to_vmd_dev(dev), mask);
-}
-
-static u64 vmd_get_required_mask(struct device *dev)
-{
-	return dma_get_required_mask(to_vmd_dev(dev));
-}
-
-static void vmd_teardown_dma_ops(struct vmd_dev *vmd)
-{
-	struct dma_domain *domain = &vmd->dma_domain;
-
-	if (get_dma_ops(&vmd->dev->dev))
-		del_dma_domain(domain);
-}
-
-#define ASSIGN_VMD_DMA_OPS(source, dest, fn)	\
-	do {					\
-		if (source->fn)			\
-			dest->fn = vmd_##fn;	\
-	} while (0)
-
-static void vmd_setup_dma_ops(struct vmd_dev *vmd)
-{
-	const struct dma_map_ops *source = get_dma_ops(&vmd->dev->dev);
-	struct dma_map_ops *dest = &vmd->dma_ops;
-	struct dma_domain *domain = &vmd->dma_domain;
-
-	domain->domain_nr = vmd->sysdata.domain;
-	domain->dma_ops = dest;
-
-	if (!source)
-		return;
-	ASSIGN_VMD_DMA_OPS(source, dest, alloc);
-	ASSIGN_VMD_DMA_OPS(source, dest, free);
-	ASSIGN_VMD_DMA_OPS(source, dest, mmap);
-	ASSIGN_VMD_DMA_OPS(source, dest, get_sgtable);
-	ASSIGN_VMD_DMA_OPS(source, dest, map_page);
-	ASSIGN_VMD_DMA_OPS(source, dest, unmap_page);
-	ASSIGN_VMD_DMA_OPS(source, dest, map_sg);
-	ASSIGN_VMD_DMA_OPS(source, dest, unmap_sg);
-	ASSIGN_VMD_DMA_OPS(source, dest, sync_single_for_cpu);
-	ASSIGN_VMD_DMA_OPS(source, dest, sync_single_for_device);
-	ASSIGN_VMD_DMA_OPS(source, dest, sync_sg_for_cpu);
-	ASSIGN_VMD_DMA_OPS(source, dest, sync_sg_for_device);
-	ASSIGN_VMD_DMA_OPS(source, dest, dma_supported);
-	ASSIGN_VMD_DMA_OPS(source, dest, get_required_mask);
-	add_dma_domain(domain);
-}
-#undef ASSIGN_VMD_DMA_OPS
-
 static char __iomem *vmd_cfg_addr(struct vmd_dev *vmd, struct pci_bus *bus,
 				  unsigned int devfn, int reg, int len)
 {
@@ -709,7 +561,6 @@ static int vmd_enable_domain(struct vmd_dev *vmd, unsigned long features)
 	}
 
 	vmd_attach_resources(vmd);
-	vmd_setup_dma_ops(vmd);
 	dev_set_msi_domain(&vmd->bus->dev, vmd->irq_domain);
 
 	pci_scan_child_bus(vmd->bus);
@@ -824,7 +675,6 @@ static void vmd_remove(struct pci_dev *dev)
 	pci_stop_root_bus(vmd->bus);
 	pci_remove_root_bus(vmd->bus);
 	vmd_cleanup_srcu(vmd);
-	vmd_teardown_dma_ops(vmd);
 	vmd_detach_resources(vmd);
 	irq_domain_remove(vmd->irq_domain);
 }
diff --git a/drivers/pci/pci.c b/drivers/pci/pci.c
index ad746d9..ef7f219 100644
--- a/drivers/pci/pci.c
+++ b/drivers/pci/pci.c
@@ -6034,7 +6034,9 @@ bool pci_devs_are_dma_aliases(struct pci_dev *dev1, struct pci_dev *dev2)
 	return (dev1->dma_alias_mask &&
 		test_bit(dev2->devfn, dev1->dma_alias_mask)) ||
 	       (dev2->dma_alias_mask &&
-		test_bit(dev1->devfn, dev2->dma_alias_mask));
+		test_bit(dev1->devfn, dev2->dma_alias_mask)) ||
+	       (dev1->dma_parent && dev1->dma_parent == dev2) ||
+	       (dev2->dma_parent && dev2->dma_parent == dev1);
 }
 
 bool pci_device_is_present(struct pci_dev *pdev)
diff --git a/drivers/pci/search.c b/drivers/pci/search.c
index bade140..a1d4899 100644
--- a/drivers/pci/search.c
+++ b/drivers/pci/search.c
@@ -32,6 +32,12 @@ int pci_for_each_dma_alias(struct pci_dev *pdev,
 	struct pci_bus *bus;
 	int ret;
 
+	if (unlikely(pdev->dma_parent)) {
+		pdev = pdev->dma_parent;
+		return fn(pdev, PCI_DEVID(pdev->bus->number, pdev->devfn),
+			  data);
+	}
+
 	ret = fn(pdev, pci_dev_id(pdev), data);
 	if (ret)
 		return ret;
diff --git a/include/linux/pci.h b/include/linux/pci.h
index c393dff..0a10d1e 100644
--- a/include/linux/pci.h
+++ b/include/linux/pci.h
@@ -316,6 +316,7 @@ struct pci_dev {
 	u8		pin;		/* Interrupt pin this device uses */
 	u16		pcie_flags_reg;	/* Cached PCIe Capabilities Register */
 	unsigned long	*dma_alias_mask;/* Mask of enabled devfn aliases */
+	struct pci_dev	*dma_parent;	/* DMA requester on another bus */
 
 	struct pci_driver *driver;	/* Driver bound to this device */
 	u64		dma_mask;	/* Mask of the bits of bus address this
-- 
1.8.3.1

_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 4/5] PCI: vmd: Stop overriding dma_map_ops

[Christoph Hellwig]

On Tue, Dec 31, 2019 at 01:24:22PM -0700, Jon Derrick wrote:
> Devices on the VMD domain use the VMD endpoint's requester-id and have
> been relying on the VMD endpoint's dma operations. The downside of this
> was that VMD domain devices would use the VMD endpoint's attributes when
> doing DMA and IOMMU mapping. We can be smarter about this by only using
> the VMD endpoint when mapping and providing the correct child device's
> attributes during dma operations.
> 
> This patch adds a new dma alias mechanism by adding a hint to a pci_dev
> to point to a singular DMA requester's pci_dev. This integrates into the
> existing dma alias infrastructure to reduce the impact of the changes
> required to support this mode.

If we want to lift this check into common code I think it should go
into struct device, as that is what DMA operates on normally.  That
being said given that this insane hack only exists for braindamage in
Intel hardware I'd rather keep it as isolated as possible.
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 4/5] PCI: vmd: Stop overriding dma_map_ops

[Derrick, Jonathan]

On Thu, 2020-01-09 at 15:36 +0100, Christoph Hellwig wrote:
> On Tue, Dec 31, 2019 at 01:24:22PM -0700, Jon Derrick wrote:
> > Devices on the VMD domain use the VMD endpoint's requester-id and have
> > been relying on the VMD endpoint's dma operations. The downside of this
> > was that VMD domain devices would use the VMD endpoint's attributes when
> > doing DMA and IOMMU mapping. We can be smarter about this by only using
> > the VMD endpoint when mapping and providing the correct child device's
> > attributes during dma operations.
> > 
> > This patch adds a new dma alias mechanism by adding a hint to a pci_dev
> > to point to a singular DMA requester's pci_dev. This integrates into the
> > existing dma alias infrastructure to reduce the impact of the changes
> > required to support this mode.
> 
> If we want to lift this check into common code I think it should go
> into struct device, as that is what DMA operates on normally.
I thought about that too, but the dma alias mechanism was in pci_dev. I
can prepare a new version with struct device.

>   That
> being said given that this insane hack only exists for braindamage in
> Intel hardware I'd rather keep it as isolated as possible. 
jmho but the footprint of the new set is pretty minimal and removes a
lot of dubious code in vmd.c.
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

[RFC 5/5] x86/PCI: Remove unused X86_DEV_DMA_OPS

[Jon Derrick]

VMD was the only user of device dma operations. Now that the IOMMU has
been made aware of direct DMA aliases, VMD domain devices can reference
the VMD endpoint directly and the VMD device dma operations has been
made obsolete.

Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
---
 arch/x86/Kconfig              |  3 ---
 arch/x86/include/asm/device.h | 10 ----------
 arch/x86/pci/common.c         | 38 --------------------------------------
 3 files changed, 51 deletions(-)

diff --git a/arch/x86/Kconfig b/arch/x86/Kconfig
index 5e89499..77f9426 100644
--- a/arch/x86/Kconfig
+++ b/arch/x86/Kconfig
@@ -2955,9 +2955,6 @@ config HAVE_ATOMIC_IOMAP
 	def_bool y
 	depends on X86_32
 
-config X86_DEV_DMA_OPS
-	bool
-
 source "drivers/firmware/Kconfig"
 
 source "arch/x86/kvm/Kconfig"
diff --git a/arch/x86/include/asm/device.h b/arch/x86/include/asm/device.h
index 5e12c63..7e31f7f 100644
--- a/arch/x86/include/asm/device.h
+++ b/arch/x86/include/asm/device.h
@@ -8,16 +8,6 @@ struct dev_archdata {
 #endif
 };
 
-#if defined(CONFIG_X86_DEV_DMA_OPS) && defined(CONFIG_PCI_DOMAINS)
-struct dma_domain {
-	struct list_head node;
-	const struct dma_map_ops *dma_ops;
-	int domain_nr;
-};
-void add_dma_domain(struct dma_domain *domain);
-void del_dma_domain(struct dma_domain *domain);
-#endif
-
 struct pdev_archdata {
 };
 
diff --git a/arch/x86/pci/common.c b/arch/x86/pci/common.c
index 4022609..fcf03da 100644
--- a/arch/x86/pci/common.c
+++ b/arch/x86/pci/common.c
@@ -625,43 +625,6 @@ unsigned int pcibios_assign_all_busses(void)
 	return (pci_probe & PCI_ASSIGN_ALL_BUSSES) ? 1 : 0;
 }
 
-#if defined(CONFIG_X86_DEV_DMA_OPS) && defined(CONFIG_PCI_DOMAINS)
-static LIST_HEAD(dma_domain_list);
-static DEFINE_SPINLOCK(dma_domain_list_lock);
-
-void add_dma_domain(struct dma_domain *domain)
-{
-	spin_lock(&dma_domain_list_lock);
-	list_add(&domain->node, &dma_domain_list);
-	spin_unlock(&dma_domain_list_lock);
-}
-EXPORT_SYMBOL_GPL(add_dma_domain);
-
-void del_dma_domain(struct dma_domain *domain)
-{
-	spin_lock(&dma_domain_list_lock);
-	list_del(&domain->node);
-	spin_unlock(&dma_domain_list_lock);
-}
-EXPORT_SYMBOL_GPL(del_dma_domain);
-
-static void set_dma_domain_ops(struct pci_dev *pdev)
-{
-	struct dma_domain *domain;
-
-	spin_lock(&dma_domain_list_lock);
-	list_for_each_entry(domain, &dma_domain_list, node) {
-		if (pci_domain_nr(pdev->bus) == domain->domain_nr) {
-			pdev->dev.dma_ops = domain->dma_ops;
-			break;
-		}
-	}
-	spin_unlock(&dma_domain_list_lock);
-}
-#else
-static void set_dma_domain_ops(struct pci_dev *pdev) {}
-#endif
-
 static void set_dev_domain_options(struct pci_dev *pdev)
 {
 	if (is_vmd(pdev->bus)) {
@@ -701,7 +664,6 @@ int pcibios_add_device(struct pci_dev *dev)
 		pa_data = data->next;
 		memunmap(data);
 	}
-	set_dma_domain_ops(dev);
 	set_dev_domain_options(dev);
 	return 0;
 }
-- 
1.8.3.1

_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 5/5] x86/PCI: Remove unused X86_DEV_DMA_OPS

[Christoph Hellwig]

On Tue, Dec 31, 2019 at 01:24:23PM -0700, Jon Derrick wrote:
> VMD was the only user of device dma operations. Now that the IOMMU has
> been made aware of direct DMA aliases, VMD domain devices can reference
> the VMD endpoint directly and the VMD device dma operations has been
> made obsolete.
> 
> Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>

This seems to be a 1:1 copy of my patch from August?
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 5/5] x86/PCI: Remove unused X86_DEV_DMA_OPS

[Derrick, Jonathan]

On Thu, 2020-01-09 at 15:37 +0100, Christoph Hellwig wrote:
> On Tue, Dec 31, 2019 at 01:24:23PM -0700, Jon Derrick wrote:
> > VMD was the only user of device dma operations. Now that the IOMMU has
> > been made aware of direct DMA aliases, VMD domain devices can reference
> > the VMD endpoint directly and the VMD device dma operations has been
> > made obsolete.
> > 
> > Signed-off-by: Jon Derrick <jonathan.derrick@intel.com>
> 
> This seems to be a 1:1 copy of my patch from August?
Sorry I didn't notice that. I'll give you attributions.
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [RFC 0/5] Clean up VMD DMA Map Ops

[Joerg Roedel]

On Tue, Dec 31, 2019 at 01:24:18PM -0700, Jon Derrick wrote:
> Jon Derrick (5):
>   iommu: Remove device link to group on failure
>   iommu/vt-d: Unlink device if failed to add to group

Added 'Fixes:' tags to these two and applied them for v5.5, thanks.
_______________________________________________
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu