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....@mediatek.com> --- 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, -- 2.18.0 _______________________________________________ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu