On Tue, 2020-09-08 at 11:43 +0200, Christoph Hellwig wrote:
> And because I like replying to myself so much, here is a link to the
> version with the arm cleanup patch applied.  Unlike the previous two
> attempts this has at least survived very basic sanity testing:
> 
> http://git.infradead.org/users/hch/misc.git/shortlog/refs/heads/dma-ranges.2
> 
> Note that we'll still need to sort out the arm/keystone warnings from
> the original patch.  Do we have anyone on the CC list who knows that
> platform a little better to figure out if the ifdef solution would work?

Had to do the following to boot without errors:

diff --git a/include/linux/dma-direct.h b/include/linux/dma-direct.h
index ef61a33c47bc..7dd88a0b6d0b 100644
--- a/include/linux/dma-direct.h
+++ b/include/linux/dma-direct.h
@@ -97,6 +97,9 @@ static inline bool dma_capable(struct device *dev, dma_addr_t addr, size_t size,
 {
        dma_addr_t end = addr + size - 1;
 
+       if (addr == DMA_MAPPING_ERROR)
+               return false;
+
        if (!dev->dma_mask)
                return false;
 
diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c
index 90f1ecb6baaf..25809703a5bf 100644
--- a/kernel/dma/direct.c
+++ b/kernel/dma/direct.c
@@ -71,7 +71,12 @@ static gfp_t dma_direct_optimal_gfp_mask(struct device *dev, u64 dma_mask,
 
 static bool dma_coherent_ok(struct device *dev, phys_addr_t phys, size_t size)
 {
-       return phys_to_dma_direct(dev, phys) + size - 1 <=
+       dma_addr_t dma_addr = phys_to_dma_direct(dev, phys);
+
+       if (dma_addr == DMA_MAPPING_ERROR)
+               return false;
+
+       return dma_addr + size - 1 <=
                        min_not_zero(dev->coherent_dma_mask, dev->bus_dma_limit);
 }


Regards,
Nicolas