Re: 2.6.26-rc1 regression: ISA DMA broken (bisected)

[Takashi Iwai <tiwai@suse.de>]

At Fri, 09 May 2008 03:37:54 +0200,
Rene Herman wrote:
> 
> Good day.
> 
> commit 8779f2fc3b84ebb6c5181fb13d702e9944c16069
> 
> "x86: don't try to allocate from DMA zone at first"
> 
> breaks all of ISA DMA. Or all of ALSA ISA DMA at least. All
> ISA soundcards are silent following that commit -- no error
> messages, everything appears fine, just silence.
> 
> It won't just revert due to 32/64 merge.
> 
> Rene.

Thanks for catching it.  Yeah, the patch looks buggy.  We had an
implicit assumption that dev = NULL for ISA devices that require 24bit
DMA.

How about the patch below?  It's against the latest Linus git tree.


thanks,

Takashi


[PATCH] x86: Fix dma_alloc_coherent() for ISA devices

The recent work on x86 dma_alloc_coherent() breaks the ISA DMA buffer
allocation, which is represented by "dev = NULL" and requires 24bit
DMA implicitly.

Signed-off-by: Takashi Iwai <tiwai@suse.de>

---

diff --git a/arch/x86/kernel/pci-dma.c b/arch/x86/kernel/pci-dma.c
index 0c37f16..c5ef1af 100644
--- a/arch/x86/kernel/pci-dma.c
+++ b/arch/x86/kernel/pci-dma.c
@@ -385,11 +385,13 @@ dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
 	if (dma_alloc_from_coherent_mem(dev, size, dma_handle, &memory))
 		return memory;
 
-	if (!dev)
+	if (!dev) {
 		dev = &fallback_dev;
+		gfp |= GFP_DMA;
+	}
 	dma_mask = dev->coherent_dma_mask;
 	if (dma_mask == 0)
-		dma_mask = DMA_32BIT_MASK;
+		dma_mask = (gfp & GFP_DMA) ? DMA_24BIT_MASK : DMA_32BIT_MASK;
 
 	/* Device not DMA able */
 	if (dev->dma_mask == NULL)
@@ -403,7 +405,7 @@ dma_alloc_coherent(struct device *dev, size_t size, dma_addr_t *dma_handle,
 	   larger than 16MB and in this case we have a chance of
 	   finding fitting memory in the next higher zone first. If
 	   not retry with true GFP_DMA. -AK */
-	if (dma_mask <= DMA_32BIT_MASK)
+	if (dma_mask <= DMA_32BIT_MASK && !(gfp & GFP_DMA))
 		gfp |= GFP_DMA32;
 #endif