[PATCH v5] vfio/type1: Adopt fast IOTLB flush interface when unmap IOVAs

[Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>]

VFIO IOMMU type1 currently upmaps IOVA pages synchronously, which requires
IOTLB flushing for every unmapping. This results in large IOTLB flushing
overhead when handling pass-through devices has a large number of mapped
IOVAs. This can be avoided by using the new IOTLB flushing interface.

Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Joerg Roedel <joro@8bytes.org>
Signed-off-by: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
---

Changes from v4 (https://lkml.org/lkml/2018/1/31/153)
 * Change return type from ssize_t back to size_t since we no longer
   changing IOMMU API. Also update error handling logic accordingly.
 * In unmap_unpin_fast(), also sync when failing to allocate entry.
 * Some code restructuring and variable renaming.

 drivers/vfio/vfio_iommu_type1.c | 128 ++++++++++++++++++++++++++++++++++++----
 1 file changed, 117 insertions(+), 11 deletions(-)

diff --git a/drivers/vfio/vfio_iommu_type1.c b/drivers/vfio/vfio_iommu_type1.c
index e30e29a..6041530 100644
--- a/drivers/vfio/vfio_iommu_type1.c
+++ b/drivers/vfio/vfio_iommu_type1.c
@@ -102,6 +102,13 @@ struct vfio_pfn {
 	atomic_t		ref_count;
 };
 
+struct vfio_regions {
+	struct list_head list;
+	dma_addr_t iova;
+	phys_addr_t phys;
+	size_t len;
+};
+
 #define IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu)	\
 					(!list_empty(&iommu->domain_list))
 
@@ -648,11 +655,102 @@ static int vfio_iommu_type1_unpin_pages(void *iommu_data,
 	return i > npage ? npage : (i > 0 ? i : -EINVAL);
 }
 
+static long vfio_sync_unpin(struct vfio_dma *dma, struct vfio_domain *domain,
+				struct list_head *regions)
+{
+	long unlocked = 0;
+	struct vfio_regions *entry, *next;
+
+	iommu_tlb_sync(domain->domain);
+
+	list_for_each_entry_safe(entry, next, regions, list) {
+		unlocked += vfio_unpin_pages_remote(dma,
+						    entry->iova,
+						    entry->phys >> PAGE_SHIFT,
+						    entry->len >> PAGE_SHIFT,
+						    false);
+		list_del(&entry->list);
+		kfree(entry);
+	}
+
+	cond_resched();
+
+	return unlocked;
+}
+
+/*
+ * Generally, VFIO needs to unpin remote pages after each IOTLB flush.
+ * Therefore, when using IOTLB flush sync interface, VFIO need to keep track
+ * of these regions (currently using a list).
+ *
+ * This value specifies maximum number of regions for each IOTLB flush sync.
+ */
+#define VFIO_IOMMU_TLB_SYNC_MAX		512
+
+static size_t unmap_unpin_fast(struct vfio_domain *domain,
+			       struct vfio_dma *dma, dma_addr_t *iova,
+			       size_t len, phys_addr_t phys, long *unlocked,
+			       struct list_head *unmapped_list,
+			       int *unmapped_cnt)
+{
+	size_t unmapped = 0;
+	struct vfio_regions *entry = kzalloc(sizeof(*entry), GFP_KERNEL);
+
+	if (entry) {
+		unmapped = iommu_unmap_fast(domain->domain, *iova, len);
+
+		if (!unmapped) {
+			kfree(entry);
+		} else {
+			iommu_tlb_range_add(domain->domain, *iova, unmapped);
+			entry->iova = *iova;
+			entry->phys = phys;
+			entry->len  = unmapped;
+			list_add_tail(&entry->list, unmapped_list);
+
+			*iova += unmapped;
+			(*unmapped_cnt)++;
+		}
+	}
+
+	/*
+	 * Sync if the number of fast-unmap regions hits the limit
+	 * or in case of errors.
+	 */
+	if (*unmapped_cnt >= VFIO_IOMMU_TLB_SYNC_MAX || !unmapped) {
+		*unlocked += vfio_sync_unpin(dma, domain,
+					     unmapped_list);
+		*unmapped_cnt = 0;
+	}
+
+	return unmapped;
+}
+
+static size_t unmap_unpin_slow(struct vfio_domain *domain,
+			       struct vfio_dma *dma, dma_addr_t *iova,
+			       size_t len, phys_addr_t phys,
+			       long *unlocked)
+{
+	size_t unmapped = iommu_unmap(domain->domain, *iova, len);
+
+	if (unmapped) {
+		*unlocked += vfio_unpin_pages_remote(dma, *iova,
+						     phys >> PAGE_SHIFT,
+						     unmapped >> PAGE_SHIFT,
+						     false);
+		*iova += unmapped;
+		cond_resched();
+	}
+	return unmapped;
+}
+
 static long vfio_unmap_unpin(struct vfio_iommu *iommu, struct vfio_dma *dma,
 			     bool do_accounting)
 {
 	dma_addr_t iova = dma->iova, end = dma->iova + dma->size;
 	struct vfio_domain *domain, *d;
+	struct list_head unmapped_region_list;
+	int unmapped_region_cnt = 0;
 	long unlocked = 0;
 
 	if (!dma->size)
@@ -661,6 +759,8 @@ static long vfio_unmap_unpin(struct vfio_iommu *iommu, struct vfio_dma *dma,
 	if (!IS_IOMMU_CAP_DOMAIN_IN_CONTAINER(iommu))
 		return 0;
 
+	INIT_LIST_HEAD(&unmapped_region_list);
+
 	/*
 	 * We use the IOMMU to track the physical addresses, otherwise we'd
 	 * need a much more complicated tracking system.  Unfortunately that
@@ -698,20 +798,26 @@ static long vfio_unmap_unpin(struct vfio_iommu *iommu, struct vfio_dma *dma,
 				break;
 		}
 
-		unmapped = iommu_unmap(domain->domain, iova, len);
-		if (WARN_ON(!unmapped))
-			break;
-
-		unlocked += vfio_unpin_pages_remote(dma, iova,
-						    phys >> PAGE_SHIFT,
-						    unmapped >> PAGE_SHIFT,
-						    false);
-		iova += unmapped;
-
-		cond_resched();
+		/*
+		 * First, try to use fast unmap/unpin. In case of failure,
+		 * switch to slow unmap/unpin path.
+		 */
+		unmapped = unmap_unpin_fast(domain, dma, &iova, len, phys,
+					    &unlocked, &unmapped_region_list,
+					    &unmapped_region_cnt);
+		if (!unmapped) {
+			unmapped = unmap_unpin_slow(domain, dma, &iova, len,
+						    phys, &unlocked);
+			if (WARN_ON(!unmapped))
+				break;
+		}
 	}
 
 	dma->iommu_mapped = false;
+
+	if (unmapped_region_cnt)
+		unlocked += vfio_sync_unpin(dma, domain, &unmapped_region_list);
+
 	if (do_accounting) {
 		vfio_lock_acct(dma->task, -unlocked, NULL);
 		return 0;
-- 
1.8.3.1