sparc/iommu: use __sbus_iommu_map_page to implement the map_sg path
authorChristoph Hellwig <hch@lst.de>
Tue, 16 Apr 2019 18:23:46 +0000 (20:23 +0200)
committerDavid S. Miller <davem@davemloft.net>
Thu, 9 May 2019 00:11:57 +0000 (17:11 -0700)
This means we handle > PAGE_SIZE offsets fine, and grow the size check
so far only performed in the map_page path.  We lose the optimization
to not double flush a page if it apears in multiple consecutive SG list
entries.  But at least for block I/O those don't happen anymore since
we properly merge in higher layers anyway.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
arch/sparc/mm/iommu.c

index 37b5ce7657f61bc2d4f4bd68ab61a024e37c9ee7..8fbc08d148361fcdf6ce65ec8ea3978a96129fef 100644 (file)
@@ -217,6 +217,11 @@ static dma_addr_t __sbus_iommu_map_page(struct device *dev, struct page *page,
        if (!len || len > 256 * 1024)
                return DMA_MAPPING_ERROR;
 
+       /*
+        * We expect unmapped highmem pages to be not in the cache.
+        * XXX Is this a good assumption?
+        * XXX What if someone else unmaps it here and races us?
+        */
        if (per_page_flush && !PageHighMem(page)) {
                unsigned long vaddr, p;
 
@@ -247,30 +252,14 @@ static int __sbus_iommu_map_sg(struct device *dev, struct scatterlist *sgl,
                int nents, enum dma_data_direction dir, unsigned long attrs,
                bool per_page_flush)
 {
-       unsigned long page, oldpage = 0;
        struct scatterlist *sg;
-       int i, j, n;
+       int j;
 
        for_each_sg(sgl, sg, nents, j) {
-               n = (sg->length + sg->offset + PAGE_SIZE-1) >> PAGE_SHIFT;
-
-               /*
-                * We expect unmapped highmem pages to be not in the cache.
-                * XXX Is this a good assumption?
-                * XXX What if someone else unmaps it here and races us?
-                */
-               if (per_page_flush && !PageHighMem(sg_page(sg))) {
-                       page = (unsigned long)page_address(sg_page(sg));
-                       for (i = 0; i < n; i++) {
-                               if (page != oldpage) {  /* Already flushed? */
-                                       flush_page_for_dma(page);
-                                       oldpage = page;
-                               }
-                               page += PAGE_SIZE;
-                       }
-               }
-
-               sg->dma_address = iommu_get_one(dev, sg_phys(sg), n) + sg->offset;
+               sg->dma_address =__sbus_iommu_map_page(dev, sg_page(sg),
+                               sg->offset, sg->length, per_page_flush);
+               if (sg->dma_address == DMA_MAPPING_ERROR)
+                       return 0;
                sg->dma_length = sg->length;
        }