iommu sg merging: swiotlb: respect the segment boundary limits
authorFUJITA Tomonori <tomof@acm.org>
Tue, 5 Feb 2008 06:28:16 +0000 (22:28 -0800)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Tue, 5 Feb 2008 17:44:12 +0000 (09:44 -0800)
This patch makes swiotlb not allocate a memory area spanning LLD's segment
boundary.

is_span_boundary() judges whether a memory area spans LLD's segment boundary.
If map_single finds such a area, map_single tries to find the next available
memory area.

Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: James Bottomley <James.Bottomley@steeleye.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Greg KH <greg@kroah.com>
Cc: Jeff Garzik <jeff@garzik.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
lib/swiotlb.c

index 1a8050ade86159877fd6eb4486954c7719bfc28d..4bb5a11e18a289e6bcf3a89ca41b0a13fb0b40cc 100644 (file)
@@ -282,6 +282,15 @@ address_needs_mapping(struct device *hwdev, dma_addr_t addr)
        return (addr & ~mask) != 0;
 }
 
+static inline unsigned int is_span_boundary(unsigned int index,
+                                           unsigned int nslots,
+                                           unsigned long offset_slots,
+                                           unsigned long max_slots)
+{
+       unsigned long offset = (offset_slots + index) & (max_slots - 1);
+       return offset + nslots > max_slots;
+}
+
 /*
  * Allocates bounce buffer and returns its kernel virtual address.
  */
@@ -292,6 +301,16 @@ map_single(struct device *hwdev, char *buffer, size_t size, int dir)
        char *dma_addr;
        unsigned int nslots, stride, index, wrap;
        int i;
+       unsigned long start_dma_addr;
+       unsigned long mask;
+       unsigned long offset_slots;
+       unsigned long max_slots;
+
+       mask = dma_get_seg_boundary(hwdev);
+       start_dma_addr = virt_to_bus(io_tlb_start) & mask;
+
+       offset_slots = ALIGN(start_dma_addr, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT;
+       max_slots = ALIGN(mask + 1, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT;
 
        /*
         * For mappings greater than a page, we limit the stride (and
@@ -311,10 +330,17 @@ map_single(struct device *hwdev, char *buffer, size_t size, int dir)
         */
        spin_lock_irqsave(&io_tlb_lock, flags);
        {
-               wrap = index = ALIGN(io_tlb_index, stride);
-
+               index = ALIGN(io_tlb_index, stride);
                if (index >= io_tlb_nslabs)
-                       wrap = index = 0;
+                       index = 0;
+
+               while (is_span_boundary(index, nslots, offset_slots,
+                                       max_slots)) {
+                       index += stride;
+                       if (index >= io_tlb_nslabs)
+                               index = 0;
+               }
+               wrap = index;
 
                do {
                        /*
@@ -341,9 +367,12 @@ map_single(struct device *hwdev, char *buffer, size_t size, int dir)
 
                                goto found;
                        }
-                       index += stride;
-                       if (index >= io_tlb_nslabs)
-                               index = 0;
+                       do {
+                               index += stride;
+                               if (index >= io_tlb_nslabs)
+                                       index = 0;
+                       } while (is_span_boundary(index, nslots, offset_slots,
+                                                 max_slots));
                } while (index != wrap);
 
                spin_unlock_irqrestore(&io_tlb_lock, flags);