dma_unmap_single() lacking cache sync on some archs?
[Resending with CC to affected parties]
Hi,
I would expect cache synchronization for DMA_TO_DEVICE and DMA_BIDIRECTIONAL when dma_map_single() is called, and for DMA_FROM_DEVICE and DMA_BIDIRECTIONAL when dma_unmap_single() is called.
However, on some architechtures (at least avr32, blackfin, ...), cache synchronization only happens when dma_map_single() is called (and then irrespective of DMA direction). dma_unmap_single() is a no-op for these archs.
See e.g. http://lxr.linux.no/#linux+v3.0.4/arch/avr32/include/asm/dma-mapping.h#L117
Isn't this a bug?
(Please CC me in responses.)
Signature · 2 lines
Arvid Brodin Enea Services Stockholm AB
Hi,
Quoted 8 lines · Arvid Brodin
[Resending with CC to affected parties]
Hi,
I would expect cache synchronization for
DMA_TO_DEVICEandDMA_BIDIRECTIONALwhendma_map_single()is called, and forDMA_FROM_DEVICEandDMA_BIDIRECTIONALwhendma_unmap_single()is called.
Quoted 7 lines
However, on some architechtures (at least avr32, blackfin, ...), cache synchronization only happens when
dma_map_single()is called (and then irrespective of DMA direction).dma_unmap_single()is a no-op for these archs.See e.g. http://lxr.linux.no/#linux+v3.0.4/arch/avr32/include/asm/dma-mapping.h#L117
Isn't this a bug?
I don't think so. What do other architectures do?
We always need to sync before the transfer because if there is dirty data in the cache, it might get written to RAM during the transfer, which would be bad. Then, since the relevant cache lines are already clean and invalid, and the CPU is not allowed to access the buffer during the transfer, there's no need to sync again when the transfer is complete.
Havard
Quoted 18 lines · On Tuesday 27 September 2011 09:55:02 Håvard Skinnemoen wrote:
Quoted 8 lines · Arvid Brodin
[Resending with CC to affected parties]
Hi,
I would expect cache synchronization for
DMA_TO_DEVICEandDMA_BIDIRECTIONALwhendma_map_single()is called, and forDMA_FROM_DEVICEandDMA_BIDIRECTIONALwhendma_unmap_single()is called.Quoted 7 lines
However, on some architechtures (at least avr32, blackfin, ...), cache synchronization only happens when
dma_map_single()is called (and then irrespective of DMA direction).dma_unmap_single()is a no-op for these archs.See e.g. http://lxr.linux.no/#linux+v3.0.4/arch/avr32/include/asm/dma-mapping.h#L117
Isn't this a bug?
Quoted 8 lines
I don't think so. What do other architectures do?
We always need to sync before the transfer because if there is dirty data in the cache, it might get written to RAM during the transfer, which would be bad. Then, since the relevant cache lines are already clean and invalid, and the CPU is not allowed to access the buffer during the transfer, there's no need to sync again when the transfer is complete.
On some architectures, e.g. ARMv6 and higher, a speculative prefetch might cause cache lines to be read again while an inbound DMA is on its way. On those architectures you need to discard cache lines before reading from the buffer. In fact also for DMA_FROM_DEVICE you need to flush or invalidate the cache for the buffer before the transfer and invalidate the cache again after the transfer.
Most architectures however do not require this.
Arnd
Quoted 19 lines · Håvard Skinnemoen
Hi,
Quoted 8 lines · Arvid Brodin
[Resending with CC to affected parties]
Hi,
I would expect cache synchronization for
DMA_TO_DEVICEandDMA_BIDIRECTIONALwhendma_map_single()is called, and forDMA_FROM_DEVICEandDMA_BIDIRECTIONALwhendma_unmap_single()is called.Quoted 7 lines
However, on some architechtures (at least avr32, blackfin, ...), cache synchronization only happens when
dma_map_single()is called (and then irrespective of DMA direction).dma_unmap_single()is a no-op for these archs.See e.g. http://lxr.linux.no/#linux+v3.0.4/arch/avr32/include/asm/dma-mapping.h#L117
Isn't this a bug?
Quoted 8 lines
I don't think so. What do other architectures do?
We always need to sync before the transfer because if there is dirty data in the cache, it might get written to RAM during the transfer, which would be bad. Then, since the relevant cache lines are already clean and invalid, and the CPU is not allowed to access the buffer during the transfer, there's no need to sync again when the transfer is complete.
I see. Thanks for the explanation!
Quoted 2 lines
Havard
Signature · 2 lines
Arvid Brodin Enea Services Stockholm AB