Re: [PATCH][RFC] asm-generic:remove calling flush_write_buffers() in dma_sync_*_for_cpu

[Arnd Bergmann <arnd@arndb.de>]

On Monday 29 June 2009, Alan Cox wrote:
> > Of course that raises the question of whether smp_wmb() is too weak in case of
> > !SMP or X86_PPRO_FENCE, but with the described scenario, I don't think
> > it does.
> 
> I'm pretty sure it is for the Winchip -but we don't care as there are no
> SMP ones. OTOH with DMA we do care

Ok. The Winchip also does not have an IOMMU or the need for SWIOTLB, so
I guess it would be ok to move the flush_write_buffers() out of the common
code into the x86 pci-nommu implementation. That one is also the only
place that calls flush_write_buffers() in dma_map_().

This would need to get revisited if we want to implement OOSTORE for
VIA Nano, which could then need it in swiotlb.
---
[PATCH] asm-generic: remove calling flush_write_buffers() in dma_sync_*

Flushing the write buffers is only needed on x86 processors using OOSTORE
(currently only pre-VIA Centaur processors) when giving a buffer to the
device, but not for handing it back. 

Based on an earlier patch from Ming Lei <tom.leiming@gmail.com>

Cc: Joerg Roedel <joerg.roedel@amd.com>
Cc: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Alan Cox <alan@lxorguk.ukuu.org.uk>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>

--- a/arch/x86/kernel/pci-nommu.c
+++ b/arch/x86/kernel/pci-nommu.c
@@ -79,12 +79,29 @@ static void nommu_free_coherent(struct device *dev, size_t size, void *vaddr,
 	free_pages((unsigned long)vaddr, get_order(size));
 }
 
+static void nommu_sync_single_for_device(struct device *dev,
+			dma_addr_t addr, size_t size,
+			enum dma_data_direction dir)
+{
+	flush_write_buffers();
+}
+
+
+static void nommu_sync_sg_for_device(struct device *dev,
+			struct scatterlist *sg, int nelems,
+			enum dma_data_direction dir)
+{
+	flush_write_buffers();
+}
+
 struct dma_map_ops nommu_dma_ops = {
-	.alloc_coherent	= dma_generic_alloc_coherent,
-	.free_coherent	= nommu_free_coherent,
-	.map_sg		= nommu_map_sg,
-	.map_page	= nommu_map_page,
-	.is_phys	= 1,
+	.alloc_coherent		= dma_generic_alloc_coherent,
+	.free_coherent		= nommu_free_coherent,
+	.map_sg			= nommu_map_sg,
+	.map_page		= nommu_map_page,
+	.sync_single_for_device = nommu_sync_single_for_device,
+	.sync_sg_for_device	= nommu_sync_sg_for_device,
+	.is_phys		= 1,
 };
 
 void __init no_iommu_init(void)
diff --git a/include/asm-generic/dma-mapping-common.h b/include/asm-generic/dma-mapping-common.h
index 5406a60..e694263 100644
--- a/include/asm-generic/dma-mapping-common.h
+++ b/include/asm-generic/dma-mapping-common.h
@@ -103,7 +103,6 @@ static inline void dma_sync_single_for_cpu(struct device *dev, dma_addr_t addr,
 	if (ops->sync_single_for_cpu)
 		ops->sync_single_for_cpu(dev, addr, size, dir);
 	debug_dma_sync_single_for_cpu(dev, addr, size, dir);
-	flush_write_buffers();
 }
 
 static inline void dma_sync_single_for_device(struct device *dev,
@@ -116,7 +115,6 @@ static inline void dma_sync_single_for_device(struct device *dev,
 	if (ops->sync_single_for_device)
 		ops->sync_single_for_device(dev, addr, size, dir);
 	debug_dma_sync_single_for_device(dev, addr, size, dir);
-	flush_write_buffers();
 }
 
 static inline void dma_sync_single_range_for_cpu(struct device *dev,
@@ -132,7 +130,6 @@ static inline void dma_sync_single_range_for_cpu(struct device *dev,
 		ops->sync_single_range_for_cpu(dev, addr, offset, size, dir);
 		debug_dma_sync_single_range_for_cpu(dev, addr, offset, size, dir);
 
-		flush_write_buffers();
 	} else
 		dma_sync_single_for_cpu(dev, addr, size, dir);
 }
@@ -150,7 +147,6 @@ static inline void dma_sync_single_range_for_device(struct device *dev,
 		ops->sync_single_range_for_device(dev, addr, offset, size, dir);
 		debug_dma_sync_single_range_for_device(dev, addr, offset, size, dir);
 
-		flush_write_buffers();
 	} else
 		dma_sync_single_for_device(dev, addr, size, dir);
 }
@@ -165,7 +161,6 @@ dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg,
 	if (ops->sync_sg_for_cpu)
 		ops->sync_sg_for_cpu(dev, sg, nelems, dir);
 	debug_dma_sync_sg_for_cpu(dev, sg, nelems, dir);
-	flush_write_buffers();
 }
 
 static inline void
@@ -179,7 +174,6 @@ dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg,
 		ops->sync_sg_for_device(dev, sg, nelems, dir);
 	debug_dma_sync_sg_for_device(dev, sg, nelems, dir);
 
-	flush_write_buffers();
 }
 
 #define dma_map_single(d, a, s, r) dma_map_single_attrs(d, a, s, r, NULL)