MTK_IOMMU driver writes one page entry and does tlb flush at a time
currently. More optimal would be to aggregate the writes and flush
BUS buffer in the end.
For 50MB buffer mapping, if mtk_iommu driver use iotlb_sync_range()
instead of tlb_add_range() and tlb_flush_walk/leaf(), it can increase
50% performance or more(depending on size of every page size) in
comparison to flushing after each page entry update. So we prefer to
use iotlb_sync_range() to replace iotlb_sync(), tlb_add_range() and
tlb_flush_walk/leaf() for MTK platforms.
Signed-off-by: Chao Hao <chao.hao@xxxxxxxxxxxx>
---
drivers/iommu/mtk_iommu.c | 6 ++++++
1 file changed, 6 insertions(+)
diff --git a/drivers/iommu/mtk_iommu.c b/drivers/iommu/mtk_iommu.c
index 785b228d39a6..d3400c15ff7b 100644
--- a/drivers/iommu/mtk_iommu.c
+++ b/drivers/iommu/mtk_iommu.c
@@ -224,6 +224,11 @@ static void mtk_iommu_tlb_flush_range_sync(unsigned long iova, size_t size,
}
}
+static void __mtk_iommu_tlb_flush_range_sync(unsigned long iova, size_t size)
+{
+ mtk_iommu_tlb_flush_range_sync(iova, size, 0, NULL)
+}
+
static void mtk_iommu_tlb_flush_page_nosync(struct iommu_iotlb_gather *gather,
unsigned long iova, size_t granule,
void *cookie)
@@ -536,6 +541,7 @@ static const struct iommu_ops mtk_iommu_ops = {
.map = mtk_iommu_map,
.unmap = mtk_iommu_unmap,
.flush_iotlb_all = mtk_iommu_flush_iotlb_all,
+ .iotlb_sync_range = __mtk_iommu_tlb_flush_range_sync,
.iotlb_sync = mtk_iommu_iotlb_sync,
.iova_to_phys = mtk_iommu_iova_to_phys,
.probe_device = mtk_iommu_probe_device,