On Wed, Dec 11, 2019 at 9:41 PM Robin Murphy robin.murphy@arm.com wrote:

On 11/12/2019 10:41 am, Chen-Yu Tsai wrote:

...

From: Chen-Yu Tsai wens@csie.org

max_pfn, as set in arch/arm/mm/init.c:

 static void __init find_limits(unsigned long *min,
                             unsigned long *max_low,
                             unsigned long *max_high)
 {
      *max_low = PFN_DOWN(memblock_get_current_limit());
      *min = PFN_UP(memblock_start_of_DRAM());
      *max_high = PFN_DOWN(memblock_end_of_DRAM());
 }

with memblock_end_of_DRAM() pointing to the next byte after DRAM. As such, max_pfn points to the PFN after the end of DRAM.

Thus when using max_pfn to check DMA masks, we should subtract one when checking DMA ranges against it.

Commit 8bf1268f48ad ("ARM: dma-api: fix off-by-one error in __dma_supported()") fixed the same issue, but missed this spot.

This issue was found while working on the sun4i-csi v4l2 driver on the Allwinner R40 SoC. On Allwinner SoCs, DRAM is offset at 0x40000000, and we are starting to use of_dma_configure() with the "dma-ranges" property in the device tree to have the DMA API handle the offset.

In this particular instance, dma-ranges was set to the same range as the actual available (2 GiB) DRAM. The following error appeared when the driver attempted to allocate a buffer:

 sun4i-csi 1c09000.csi: Coherent DMA mask 0x7fffffff (pfn 0x40000-0xc0000)
 covers a smaller range of system memory than the DMA zone pfn 0x0-0xc0001
 sun4i-csi 1c09000.csi: dma_alloc_coherent of size 307200 failed

Fixing the off-by-one error makes things work.

Fixes: 11a5aa32562e ("ARM: dma-mapping: check DMA mask against available memory") Fixes: 9f28cde0bc64 ("ARM: another fix for the DMA mapping checks") Fixes: ab746573c405 ("ARM: dma-mapping: allow larger DMA mask than supported") Cc: stable@vger.kernel.org Signed-off-by: Chen-Yu Tsai wens@csie.org

---

arch/arm/mm/dma-mapping.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c index e822af0d9219..f4daafdbac56 100644 --- a/arch/arm/mm/dma-mapping.c +++ b/arch/arm/mm/dma-mapping.c @@ -227,12 +227,12 @@ static int __dma_supported(struct device *dev, u64 mask, bool warn) * Translate the device's DMA mask to a PFN limit. This * PFN number includes the page which we can DMA to. */

• if (dma_to_pfn(dev, mask) < max_dma_pfn) {
  

• if (dma_to_pfn(dev, mask) < max_dma_pfn - 1) {
  

I think this correction actually wants to happen a couple of lines up in the definition:

    unsigned long max_dma_pfn = min(max_pfn, arm_dma_pfn_limit);

max_pfn is indeed an exclusive limit, but AFAICS arm_dma_pfn_limit is inclusive, so none of these "+1"s and "-1"s can be entirely right for both cases.

You're absolutely right. I'll fix it and send a v2 out.

Thanks

ChenYu

Robin.

...

          if (warn)
                  dev_warn(dev, "Coherent DMA mask %#llx (pfn %#lx-%#lx) covers a smaller range of system memory than the DMA zone pfn 0x0-%#lx\n",
                           mask,
                           dma_to_pfn(dev, 0), dma_to_pfn(dev, mask) + 1,

•                          max_dma_pfn + 1);
  

•                          max_dma_pfn);
          return 0;
  }