AMD IOMMU: implement lazy IO/TLB flushing

The IO/TLB flushing on every unmaping operation is the most expensive
part in AMD IOMMU code and not strictly necessary. It is sufficient to
do the flush before any entries are reused. This is patch implements
lazy IO/TLB flushing which does exactly this.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

3 files changed, 31 insertions, 5 deletions

diff --git a/arch/x86/kernel/amd_iommu.c b/arch/x86/kernel/amd_iommu.c
index 691e023695ad..679f2a8e22ee 100644
--- a/arch/x86/kernel/amd_iommu.c
+++ b/arch/x86/kernel/amd_iommu.c
@@ -203,6 +203,14 @@ static int iommu_flush_pages(struct amd_iommu *iommu, u16 domid,
 	return 0;
 }
 
+/* Flush the whole IO/TLB for a given protection domain */
+static void iommu_flush_tlb(struct amd_iommu *iommu, u16 domid)
+{
+	u64 address = CMD_INV_IOMMU_ALL_PAGES_ADDRESS;
+
+	iommu_queue_inv_iommu_pages(iommu, address, domid, 0, 1);
+}
+
 /****************************************************************************
  *
  * The functions below are used the create the page table mappings for
@@ -386,14 +394,18 @@ static unsigned long dma_ops_alloc_addresses(struct device *dev,
 			PAGE_SIZE) >> PAGE_SHIFT;
 	limit = limit < size ? limit : size;
 
-	if (dom->next_bit >= limit)
+	if (dom->next_bit >= limit) {
 		dom->next_bit = 0;
+		dom->need_flush = true;
+	}
 
 	address = iommu_area_alloc(dom->bitmap, limit, dom->next_bit, pages,
 			0 , boundary_size, 0);
-	if (address == -1)
+	if (address == -1) {
 		address = iommu_area_alloc(dom->bitmap, limit, 0, pages,
 				0, boundary_size, 0);
+		dom->need_flush = true;
+	}
 
 	if (likely(address != -1)) {
 		dom->next_bit = address + pages;
@@ -553,6 +565,8 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu,
 	dma_dom->bitmap[0] = 1;
 	dma_dom->next_bit = 0;
 
+	dma_dom->need_flush = false;
+
 	/* Intialize the exclusion range if necessary */
 	if (iommu->exclusion_start &&
 	    iommu->exclusion_start < dma_dom->aperture_size) {
@@ -795,7 +809,10 @@ static dma_addr_t __map_single(struct device *dev,
 	}
 	address += offset;
 
-	if (unlikely(iommu_has_npcache(iommu)))
+	if (unlikely(dma_dom->need_flush && !iommu_fullflush)) {
+		iommu_flush_tlb(iommu, dma_dom->domain.id);
+		dma_dom->need_flush = false;
+	} else if (unlikely(iommu_has_npcache(iommu)))
 		iommu_flush_pages(iommu, dma_dom->domain.id, address, size);
 
 out:
@@ -829,7 +846,8 @@ static void __unmap_single(struct amd_iommu *iommu,
 
 	dma_ops_free_addresses(dma_dom, dma_addr, pages);
 
-	iommu_flush_pages(iommu, dma_dom->domain.id, dma_addr, size);
+	if (iommu_fullflush)
+		iommu_flush_pages(iommu, dma_dom->domain.id, dma_addr, size);
 }
 
 /*
diff --git a/arch/x86/kernel/amd_iommu_init.c b/arch/x86/kernel/amd_iommu_init.c
index a69cc0f52042..f2fa8dc81beb 100644
--- a/arch/x86/kernel/amd_iommu_init.c
+++ b/arch/x86/kernel/amd_iommu_init.c
@@ -995,6 +995,11 @@ int __init amd_iommu_init(void)
 	else
 		printk("disabled\n");
 
+	if (iommu_fullflush)
+		printk(KERN_INFO "AMD IOMMU: IO/TLB flush on unmap enabled\n");
+	else
+		printk(KERN_INFO "AMD IOMMU: Lazy IO/TLB flushing enabled\n");
+
 out:
 	return ret;
 
@@ -1057,7 +1062,7 @@ void __init amd_iommu_detect(void)
 static int __init parse_amd_iommu_options(char *str)
 {
 	for (; *str; ++str) {
-		if (strcmp(str, "isolate") == 0)
+		if (strncmp(str, "isolate", 7) == 0)
 			amd_iommu_isolate = 1;
 	}
 
diff --git a/include/asm-x86/amd_iommu_types.h b/include/asm-x86/amd_iommu_types.h
index dcc812067394..dcc472445ffd 100644
--- a/include/asm-x86/amd_iommu_types.h
+++ b/include/asm-x86/amd_iommu_types.h
@@ -196,6 +196,9 @@ struct dma_ops_domain {
 	 * just calculate its address in constant time.
 	 */
 	u64 **pte_pages;
+
+	/* This will be set to true when TLB needs to be flushed */
+	bool need_flush;
 };
 
 /*