powerpc/powernv: Fix IOMMU table for VFIO dev

On PHB3, PCI devices can bypass IOMMU for DMA access. If we pass
through one PCI device, whose hose driver ever enable the bypass
mode, pdev->dev.archdata.dma_data.iommu_table_base isn't IOMMU
table. However, EEH needs access the IOMMU table when the device
is owned by guest.

The patch fixes pdev->dev.archdata.dma_data.iommu_table when
passing through the device to guest in pnv_pci_ioda2_set_bypass().

Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

1 files changed, 21 insertions, 9 deletions

diff --git a/arch/powerpc/platforms/powernv/pci-ioda.c b/arch/powerpc/platforms/powernv/pci-ioda.c
index 866692b851f2..035db476be0e 100644
--- a/arch/powerpc/platforms/powernv/pci-ioda.c
+++ b/arch/powerpc/platforms/powernv/pci-ioda.c
@@ -495,14 +495,22 @@ static int pnv_pci_ioda_dma_set_mask(struct pnv_phb *phb,
 	return 0;
 }
 
-static void pnv_ioda_setup_bus_dma(struct pnv_ioda_pe *pe, struct pci_bus *bus)
+static void pnv_ioda_setup_bus_dma(struct pnv_ioda_pe *pe,
+				   struct pci_bus *bus,
+				   bool add_to_iommu_group)
 {
 	struct pci_dev *dev;
 
 	list_for_each_entry(dev, &bus->devices, bus_list) {
-		set_iommu_table_base_and_group(&dev->dev, &pe->tce32_table);
+		if (add_to_iommu_group)
+			set_iommu_table_base_and_group(&dev->dev,
+						       &pe->tce32_table);
+		else
+			set_iommu_table_base(&dev->dev, &pe->tce32_table);
+
 		if (dev->subordinate)
-			pnv_ioda_setup_bus_dma(pe, dev->subordinate);
+			pnv_ioda_setup_bus_dma(pe, dev->subordinate,
+					       add_to_iommu_group);
 	}
 }
 
@@ -680,7 +688,7 @@ static void pnv_pci_ioda_setup_dma_pe(struct pnv_phb *phb,
 	if (pe->pdev)
 		set_iommu_table_base_and_group(&pe->pdev->dev, tbl);
 	else
-		pnv_ioda_setup_bus_dma(pe, pe->pbus);
+		pnv_ioda_setup_bus_dma(pe, pe->pbus, true);
 
 	return;
  fail:
@@ -716,11 +724,15 @@ static void pnv_pci_ioda2_set_bypass(struct iommu_table *tbl, bool enable)
 						     0);
 
 		/*
-		 * We might want to reset the DMA ops of all devices on
-		 * this PE. However in theory, that shouldn't be necessary
-		 * as this is used for VFIO/KVM pass-through and the device
-		 * hasn't yet been returned to its kernel driver
+		 * EEH needs the mapping between IOMMU table and group
+		 * of those VFIO/KVM pass-through devices. We can postpone
+		 * resetting DMA ops until the DMA mask is configured in
+		 * host side.
 		 */
+		if (pe->pdev)
+			set_iommu_table_base(&pe->pdev->dev, tbl);
+		else
+			pnv_ioda_setup_bus_dma(pe, pe->pbus, false);
 	}
 	if (rc)
 		pe_err(pe, "OPAL error %lld configuring bypass window\n", rc);
@@ -809,7 +821,7 @@ static void pnv_pci_ioda2_setup_dma_pe(struct pnv_phb *phb,
 	if (pe->pdev)
 		set_iommu_table_base_and_group(&pe->pdev->dev, tbl);
 	else
-		pnv_ioda_setup_bus_dma(pe, pe->pbus);
+		pnv_ioda_setup_bus_dma(pe, pe->pbus, true);
 
 	/* Also create a bypass window */
 	pnv_pci_ioda2_setup_bypass_pe(phb, pe);