A basic device-specific linear memory mapping was introduced back in
commit ("dma: Take into account dma_pfn_offset") as a single-valued offset
preserved in the device.dma_pfn_offset field, which was initialized for
instance by means of the "dma-ranges" DT property. Afterwards the
functionality was extended to support more than one device-specific region
defined in the device.dma_range_map list of maps. But all of these
improvements concerned a single pointer, page or sg DMA-mapping methods,
while the system resource mapping function turned to miss the
corresponding modification. Thus the dma_direct_map_resource() method now
just casts the CPU physical address to the device DMA address with no
dma-ranges-based mapping taking into account, which is obviously wrong.
Let's fix it by using the phys_to_dma_direct() method to get the
device-specific bus address from the passed memory resource for the case
of the directly mapped DMA.
Fixes: 25f1e1887088 ("dma: Take into account dma_pfn_offset")
Signed-off-by: Serge Semin <[email protected]>
---
kernel/dma/direct.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c
index 9743c6ccce1a..bc06db74dfdb 100644
--- a/kernel/dma/direct.c
+++ b/kernel/dma/direct.c
@@ -497,7 +497,7 @@ int dma_direct_map_sg(struct device *dev, struct scatterlist *sgl, int nents,
dma_addr_t dma_direct_map_resource(struct device *dev, phys_addr_t paddr,
size_t size, enum dma_data_direction dir, unsigned long attrs)
{
- dma_addr_t dma_addr = paddr;
+ dma_addr_t dma_addr = phys_to_dma_direct(dev, paddr);
if (unlikely(!dma_capable(dev, dma_addr, size, false))) {
dev_err_once(dev,
--
2.35.1
So I think the big problem pointed out by Robin is that existing DTs
might not take this into account. So I think we need to do a little
research and at least Cc all maintainers and lists for relevant in-tree
DTs for drivers that use dma_map_resource to discuss this.
On Mon, May 09, 2022 at 08:15:52AM +0200, Christoph Hellwig wrote:
> So I think the big problem pointed out by Robin is that existing DTs
> might not take this into account.
I'd say that the biggest problem isn't in the DT part but in the
drivers using the dma_map_resource() method since they don't expect
the non-uniform DMA-direct mapping can be available.
> So I think we need to do a little
> research and at least Cc all maintainers and lists for relevant in-tree
> DTs for drivers that use dma_map_resource to discuss this.
Right. I'll send the next patchset revision out with all possibly
interested maintainers and lists in Cc of this patch.
-Sergey
>