1 files changed, 23 insertions, 22 deletions

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index a8fd9ebdc8e..5662e226b0c 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -50,7 +50,7 @@ static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
 /* returns !0 if the IOMMU is caching non-present entries in its TLB */
 static int iommu_has_npcache(struct amd_iommu *iommu)
 {
-	return iommu->cap & IOMMU_CAP_NPCACHE;
+	return iommu->cap & (1UL << IOMMU_CAP_NPCACHE);
 }
 
 /****************************************************************************
@@ -187,6 +187,8 @@ static int iommu_queue_command(struct amd_iommu *iommu, struct iommu_cmd *cmd)
 
 	spin_lock_irqsave(&iommu->lock, flags);
 	ret = __iommu_queue_command(iommu, cmd);
+	if (!ret)
+		iommu->need_sync = 1;
 	spin_unlock_irqrestore(&iommu->lock, flags);
 
 	return ret;
@@ -210,10 +212,13 @@ static int iommu_completion_wait(struct amd_iommu *iommu)
 	cmd.data[0] = CMD_COMPL_WAIT_INT_MASK;
 	CMD_SET_TYPE(&cmd, CMD_COMPL_WAIT);
 
-	iommu->need_sync = 0;
-
 	spin_lock_irqsave(&iommu->lock, flags);
 
+	if (!iommu->need_sync)
+		goto out;
+
+	iommu->need_sync = 0;
+
 	ret = __iommu_queue_command(iommu, &cmd);
 
 	if (ret)
@@ -254,8 +259,6 @@ static int iommu_queue_inv_dev_entry(struct amd_iommu *iommu, u16 devid)
 
 	ret = iommu_queue_command(iommu, &cmd);
 
-	iommu->need_sync = 1;
-
 	return ret;
 }
 
@@ -281,8 +284,6 @@ static int iommu_queue_inv_iommu_pages(struct amd_iommu *iommu,
 
 	ret = iommu_queue_command(iommu, &cmd);
 
-	iommu->need_sync = 1;
-
 	return ret;
 }
 
@@ -536,6 +537,9 @@ static void dma_ops_free_addresses(struct dma_ops_domain *dom,
 {
 	address >>= PAGE_SHIFT;
 	iommu_area_free(dom->bitmap, address, pages);
+
+	if (address >= dom->next_bit)
+		dom->need_flush = true;
 }
 
 /****************************************************************************
@@ -759,8 +763,6 @@ static void set_device_domain(struct amd_iommu *iommu,
 	write_unlock_irqrestore(&amd_iommu_devtable_lock, flags);
 
 	iommu_queue_inv_dev_entry(iommu, devid);
-
-	iommu->need_sync = 1;
 }
 
 /*****************************************************************************
@@ -855,6 +857,9 @@ static int get_device_resources(struct device *dev,
 		print_devid(_bdf, 1);
 	}
 
+	if (domain_for_device(_bdf) == NULL)
+		set_device_domain(*iommu, *domain, _bdf);
+
 	return 1;
 }
 
@@ -992,8 +997,10 @@ static void __unmap_single(struct amd_iommu *iommu,
 
 	dma_ops_free_addresses(dma_dom, dma_addr, pages);
 
-	if (amd_iommu_unmap_flush)
+	if (amd_iommu_unmap_flush || dma_dom->need_flush) {
 		iommu_flush_pages(iommu, dma_dom->domain.id, dma_addr, size);
+		dma_dom->need_flush = false;
+	}
 }
 
 /*
@@ -1026,8 +1033,7 @@ static dma_addr_t map_single(struct device *dev, phys_addr_t paddr,
 	if (addr == bad_dma_address)
 		goto out;
 
-	if (unlikely(iommu->need_sync))
-		iommu_completion_wait(iommu);
+	iommu_completion_wait(iommu);
 
 out:
 	spin_unlock_irqrestore(&domain->lock, flags);
@@ -1055,8 +1061,7 @@ static void unmap_single(struct device *dev, dma_addr_t dma_addr,
 
 	__unmap_single(iommu, domain->priv, dma_addr, size, dir);
 
-	if (unlikely(iommu->need_sync))
-		iommu_completion_wait(iommu);
+	iommu_completion_wait(iommu);
 
 	spin_unlock_irqrestore(&domain->lock, flags);
 }
@@ -1122,8 +1127,7 @@ static int map_sg(struct device *dev, struct scatterlist *sglist,
 			goto unmap;
 	}
 
-	if (unlikely(iommu->need_sync))
-		iommu_completion_wait(iommu);
+	iommu_completion_wait(iommu);
 
 out:
 	spin_unlock_irqrestore(&domain->lock, flags);
@@ -1168,8 +1172,7 @@ static void unmap_sg(struct device *dev, struct scatterlist *sglist,
 		s->dma_address = s->dma_length = 0;
 	}
 
-	if (unlikely(iommu->need_sync))
-		iommu_completion_wait(iommu);
+	iommu_completion_wait(iommu);
 
 	spin_unlock_irqrestore(&domain->lock, flags);
 }
@@ -1220,8 +1223,7 @@ static void *alloc_coherent(struct device *dev, size_t size,
 		goto out;
 	}
 
-	if (unlikely(iommu->need_sync))
-		iommu_completion_wait(iommu);
+	iommu_completion_wait(iommu);
 
 out:
 	spin_unlock_irqrestore(&domain->lock, flags);
@@ -1252,8 +1254,7 @@ static void free_coherent(struct device *dev, size_t size,
 
 	__unmap_single(iommu, domain->priv, dma_addr, size, DMA_BIDIRECTIONAL);
 
-	if (unlikely(iommu->need_sync))
-		iommu_completion_wait(iommu);
+	iommu_completion_wait(iommu);
 
 	spin_unlock_irqrestore(&domain->lock, flags);