two dma-direct fixes for 4.18

When a device has a DMA offset the dma capable result will change due to the difference between the physical and DMA address. Take that into account.

Signed-off-by Christoph Hellwig <[email protected]>

Reviewed-by Benjamin Herrenschmidt <[email protected]>

---
  kernel/dma/direct.c | 4 ++--
  1 file changed, 2 insertions(+), 2 deletions(-)

kernel/dma/direct.c +2 -2

index 1c35b7b945d0..de87b0282e74 100644
@@ -168,7 +168,7 @@ int dma_direct_map_sg(struct device *dev, struct scatterlist *sgl, int nents,
  int dma_direct_supported(struct device *dev, u64 mask)
  {
  #ifdef CONFIG_ZONE_DMA
-	if (mask < DMA_BIT_MASK(ARCH_ZONE_DMA_BITS))
+	if (mask < phys_to_dma(dev, DMA_BIT_MASK(ARCH_ZONE_DMA_BITS)))
  		return 0;
  #else
  	/*

@@ -177,7 +177,7 @@ int dma_direct_supported(struct device *dev, u64 mask)
  	 * memory, or by providing a ZONE_DMA32.  If neither is the case, the
  	 * architecture needs to use an IOMMU instead of the direct mapping.
  	 */
-	if (mask < DMA_BIT_MASK(32))
+	if (mask < phys_to_dma(dev, DMA_BIT_MASK(32)))
  		return 0;
  #endif
  	/*

Quoted 3 lines · Christoph Hellwig

When a device has a DMA offset the dma capable result will change due to the difference between the physical and DMA address. Take that into account.

The "phys_to_dma(..., DMA_BIT_MASK(...))" idiom always looks like a glaring error at first glance, but this whole function is fairly unintuitive anyway, and ultimately I think the change does work out to be correct.

It might be nicer if we could reference max_zone_pfns[] for a bit more clarity, but I guess that's not arch-independent.

We need to take the DMA offset and encryption bit into account when selecting a zone. Add a helper that takes those into account and use it.

Signed-off-by Christoph Hellwig <[email protected]>

---
  kernel/dma/direct.c | 16 ++++++++++++----
  1 file changed, 12 insertions(+), 4 deletions(-)

kernel/dma/direct.c +10 -2

index de87b0282e74..3f8b596ea393 100644
@@ -58,6 +58,14 @@ static bool dma_coherent_ok(struct device *dev, phys_addr_t phys, size_t size)
  	return addr + size - 1 <= dev->coherent_dma_mask;
  }
 
+static bool dma_coherent_below(struct device *dev, u64 mask)
+{
+	dma_addr_t addr = force_dma_unencrypted() ?
+		__phys_to_dma(dev, mask) : phys_to_dma(dev, mask);
+
+	return dev->coherent_dma_mask <= addr;
+}
+
  void *dma_direct_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
  		gfp_t gfp, unsigned long attrs)
  {

@@ -70,9 +78,9 @@ void *dma_direct_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
  	gfp &= ~__GFP_ZERO;
 
  	/* GFP_DMA32 and GFP_DMA are no ops without the corresponding zones: */
-	if (dev->coherent_dma_mask <= DMA_BIT_MASK(ARCH_ZONE_DMA_BITS))
+	if (dma_coherent_below(dev, DMA_BIT_MASK(ARCH_ZONE_DMA_BITS)))
  		gfp |= GFP_DMA;
-	if (dev->coherent_dma_mask <= DMA_BIT_MASK(32) && !(gfp & GFP_DMA))
+	if (dma_coherent_below(dev, DMA_BIT_MASK(32) && !(gfp & GFP_DMA)))
  		gfp |= GFP_DMA32;
 
  again:

kernel/dma/direct.c +0 -0

  		    !(gfp & GFP_DMA)) {
  			gfp = (gfp & ~GFP_DMA32) | GFP_DMA;
  			goto again;

Quoted 41 lines · Christoph Hellwig

We need to take the DMA offset and encryption bit into account when selecting a zone.  Add a helper that takes those into account and use it.

Signed-off-by Christoph Hellwig <[email protected]>

---
  kernel/dma/direct.c | 16 ++++++++++++----

  1 file changed, 12 insertions(+), 4 deletions(-)

kernel/dma/direct.c +10 -2

index de87b0282e74..3f8b596ea393 100644
@@ -58,6 +58,14 @@ static bool dma_coherent_ok(struct device *dev, 
 phys_addr_t phys, size_t size)
       return addr + size - 1 <= dev->coherent_dma_mask;
   }
+static bool dma_coherent_below(struct device *dev, u64 mask)
+{
+    dma_addr_t addr = force_dma_unencrypted() ?
+        __phys_to_dma(dev, mask) : phys_to_dma(dev, mask);
+
+    return dev->coherent_dma_mask <= addr;
+}
+
   void *dma_direct_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
           gfp_t gfp, unsigned long attrs)
   {

@@ -70,9 +78,9 @@ void *dma_direct_alloc(struct device *dev, size_t 
 size, dma_addr_t *dma_handle,
       gfp &= ~__GFP_ZERO;
       /* GFP_DMA32 and GFP_DMA are no ops without the corresponding zones: */
-    if (dev->coherent_dma_mask <= DMA_BIT_MASK(ARCH_ZONE_DMA_BITS))
+    if (dma_coherent_below(dev, DMA_BIT_MASK(ARCH_ZONE_DMA_BITS)))
           gfp |= GFP_DMA;
-    if (dev->coherent_dma_mask <= DMA_BIT_MASK(32) && !(gfp & GFP_DMA))
+    if (dma_coherent_below(dev, DMA_BIT_MASK(32) && !(gfp & GFP_DMA)))
           gfp |= GFP_DMA32;

  again:

kernel/dma/direct.c +0 -0

@@ -93,14 +101,14 @@ void *dma_direct_alloc(struct device *dev, size_t 
 size, dma_addr_t *dma_handle,
           page = NULL;

Quoted 2 lines

          if (IS_ENABLED(CONFIG_ZONE_DMA32) &&
-            dev->coherent_dma_mask < DMA_BIT_MASK(64) &&

We're effectively changing the "<" into a "<="...

Quoted 1 line

+            dma_coherent_below(dev, DMA_BIT_MASK(64)) &&

...which I think means this check now *always* returns true...