Skip to contentSkip to conversation

swiotlb: Some regression fixes

PATCH Applied Takashi Iwai 2 participants 4 messages Apr 15, 2018 – Apr 23, 2018 Library code

lore
Share & tools
lore
#0 lore raw

cover swiotlb: Some regression fixes

Hi,

this is a couple of small regression fixes I stumbled on recently. The first one is a trivial one, and it was already posted in another thread ("swiotlb: Fix unexpected swiotlb_alloc_coherent() failures").

thanks,

Takashi

===

Takashi Iwai (2):
dma-direct: Don't repeat allocation for no-op GFP_DMA
swiotlb: Fix dma_supported() to consider direct allocation

 lib/dma-direct.c | 3 ++-
 lib/swiotlb.c    | 4 ++++
 2 files changed, 6 insertions(+), 1 deletion(-)
Signature · 1 line
2.16.3
#1 lore raw

PATCH 1/2 dma-direct: Don't repeat allocation for no-op GFP_DMA

When an allocation with lower dma_coherent mask fails, dma_direct_alloc() retries the allocation with GFP_DMA. But, it's useless for architectures that has no ZONE_DMA, obviously.

Fix it by adding the check of CONFIG_ZONE_DMA before retrying the allocation.

Fixes 95f183916d4b ("dma-direct: retry allocations using GFP_DMA for small masks")
---
 lib/dma-direct.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
lib/dma-direct.c +2 -1
index c0bba30fef0a..bbfb229aa067 100644
@@ -84,7 +84,8 @@ void *dma_direct_alloc(struct device *dev, size_t size, dma_addr_t *dma_handle,
 		__free_pages(page, page_order);
 		page = NULL;
 
-		if (dev->coherent_dma_mask < DMA_BIT_MASK(32) &&
+		if (IS_ENABLED(CONFIG_ZONE_DMA) &&
+		    dev->coherent_dma_mask < DMA_BIT_MASK(32) &&
 		    !(gfp & GFP_DMA)) {
 			gfp = (gfp & ~GFP_DMA32) | GFP_DMA;
 			goto again;
Signature · 1 line
2.16.3
#2 lore raw

Re: [PATCH 1/2] dma-direct: Don't repeat allocation for no-op GFP_DMA

Thanks,

applied to the dma-mapping tree for 4.17.

#3 lore raw

PATCH 2/2 swiotlb: Fix dma_supported() to consider direct allocation

In reply to Takashi Iwai

Along with the recent change to use the common swiotlb dma_ops, x86 also takes over the swiotlb_dma_supported(). This caused a regression when a low bit DMA mask is set; e.g. parport_pc driver now fails to set the 24bit DMA mask:

parport_pc parport_pc.956: Unable to set coherent dma mask: disabling DMA

It's because swiotlb_dma_supported() only checks the swiotlb range, and the 24bit DMA mask is below io_tlb_end. OTOH, in the past kernel versions, x86's swiotlb dma_supported() was NULL, which was effectively evaluated as the direct DMA, hence the lower DMA mask was allowed.

This patch fixes the regression by extending swiotlb_dma_supported() to check the direct DMA at first for covering the primary allocation before swiotlb.

Fixes 6e4bf5867783 ("x86/dma: Use generic swiotlb_ops")
---
 lib/swiotlb.c | 4 ++++
 1 file changed, 4 insertions(+)
lib/swiotlb.c +4 -0
index de7cc540450f..a81502ea79e7 100644
@@ -1042,6 +1042,10 @@ swiotlb_dma_mapping_error(struct device *hwdev, dma_addr_t dma_addr)
 int
 swiotlb_dma_supported(struct device *hwdev, u64 mask)
 {
+#ifdef CONFIG_DMA_DIRECT_OPS
+	if (dma_direct_supported(hwdev, mask))
+		return 1;
+#endif
 	return __phys_to_dma(hwdev, io_tlb_end - 1) <= mask;
 }
 
Signature · 1 line
2.16.3