Re: [PATCH 07/11] powerpc: make dma_cache_sync a no-op

2017-10-04 Thread Christophe LEROY 



Le 03/10/2017 à 13:43, Christoph Hellwig a écrit :

On Tue, Oct 03, 2017 at 01:24:57PM +0200, Christophe LEROY wrote:

powerpc does not implement DMA_ATTR_NON_CONSISTENT allocations, so it
doesn't make any sense to do any work in dma_cache_sync given that it
must be a no-op when dma_alloc_attrs returns coherent memory.

What about arch/powerpc/mm/dma-noncoherent.c ?

Powerpc 8xx doesn't have coherent memory.


It doesn't implement the DMA_ATTR_NON_CONSISTENT interface either,
so if it really doesn't have a way to provide dma coherent allocation
(although the code in __dma_alloc_coherent suggests it does provide
dma coherent allocations) I have no idea how it could ever have
worked.



Yes indeed it provides coherent memory by allocation non cached memory.

And drivers aiming at using non coherent memory do it by using kmalloc() 
with GFP_DMA then dma_map_single().
Then they use dma_sync_single_for_xxx(), which calls __dma_sync() on the 
8xx and is a nop on other powerpcs.


Christophe
___
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [PATCH 07/11] powerpc: make dma_cache_sync a no-op

2017-10-03 Thread Christophe LEROY 



Le 03/10/2017 à 12:43, Christoph Hellwig a écrit :

powerpc does not implement DMA_ATTR_NON_CONSISTENT allocations, so it
doesn't make any sense to do any work in dma_cache_sync given that it
must be a no-op when dma_alloc_attrs returns coherent memory.

What about arch/powerpc/mm/dma-noncoherent.c ?

Powerpc 8xx doesn't have coherent memory.

Christophe



Signed-off-by: Christoph Hellwig 
---
  arch/powerpc/include/asm/dma-mapping.h | 2 --
  1 file changed, 2 deletions(-)

diff --git a/arch/powerpc/include/asm/dma-mapping.h 
b/arch/powerpc/include/asm/dma-mapping.h
index eaece3d3e225..320846442bfb 100644
--- a/arch/powerpc/include/asm/dma-mapping.h
+++ b/arch/powerpc/include/asm/dma-mapping.h
@@ -144,8 +144,6 @@ static inline phys_addr_t dma_to_phys(struct device *dev, 
dma_addr_t daddr)
  static inline void dma_cache_sync(struct device *dev, void *vaddr, size_t 
size,
enum dma_data_direction direction)
  {
-   BUG_ON(direction == DMA_NONE);
-   __dma_sync(vaddr, size, (int)direction);
  }
  
  #endif /* __KERNEL__ */



___
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [PATCH 07/11] powerpc: make dma_cache_sync a no-op

2017-10-03 Thread Christoph Hellwig 
On Tue, Oct 03, 2017 at 01:24:57PM +0200, Christophe LEROY wrote:
>> powerpc does not implement DMA_ATTR_NON_CONSISTENT allocations, so it
>> doesn't make any sense to do any work in dma_cache_sync given that it
>> must be a no-op when dma_alloc_attrs returns coherent memory.
> What about arch/powerpc/mm/dma-noncoherent.c ?
>
> Powerpc 8xx doesn't have coherent memory.

It doesn't implement the DMA_ATTR_NON_CONSISTENT interface either,
so if it really doesn't have a way to provide dma coherent allocation
(although the code in __dma_alloc_coherent suggests it does provide
dma coherent allocations) I have no idea how it could ever have
worked.
___
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu

Re: [PATCH 07/11] powerpc: make dma_cache_sync a no-op

2017-10-03 Thread Robin Murphy 
On 03/10/17 12:24, Christophe LEROY wrote:
> 
> 
> Le 03/10/2017 à 12:43, Christoph Hellwig a écrit :
>> powerpc does not implement DMA_ATTR_NON_CONSISTENT allocations, so it
>> doesn't make any sense to do any work in dma_cache_sync given that it
>> must be a no-op when dma_alloc_attrs returns coherent memory.
> What about arch/powerpc/mm/dma-noncoherent.c ?

AFAICS, just like the ARM code from which it is derived, that will
always return a non-cacheable remap of the allocation, which should thus
not require explicit maintenance after CPU accesses. dma_cache_sync()
would only matter if dma_alloc_attrs(..., DMA_ATTR_NON_CONSISTENT) got
special treatment and was allowed to return a cacheable address, but PPC
doesn't even propagate the attrs to its internal __dma_alloc_coherent()
implementations.

Robin.

> Powerpc 8xx doesn't have coherent memory.
> 
> Christophe
> 
>>
>> Signed-off-by: Christoph Hellwig 
>> ---
>>   arch/powerpc/include/asm/dma-mapping.h | 2 --
>>   1 file changed, 2 deletions(-)
>>
>> diff --git a/arch/powerpc/include/asm/dma-mapping.h
>> b/arch/powerpc/include/asm/dma-mapping.h
>> index eaece3d3e225..320846442bfb 100644
>> --- a/arch/powerpc/include/asm/dma-mapping.h
>> +++ b/arch/powerpc/include/asm/dma-mapping.h
>> @@ -144,8 +144,6 @@ static inline phys_addr_t dma_to_phys(struct
>> device *dev, dma_addr_t daddr)
>>   static inline void dma_cache_sync(struct device *dev, void *vaddr,
>> size_t size,
>>   enum dma_data_direction direction)
>>   {
>> -    BUG_ON(direction == DMA_NONE);
>> -    __dma_sync(vaddr, size, (int)direction);
>>   }
>>     #endif /* __KERNEL__ */
>>

___
iommu mailing list
iommu@lists.linux-foundation.org
https://lists.linuxfoundation.org/mailman/listinfo/iommu