[PATCH] powerpc: dma-mapping: Return dma_direct_ops variable when dev == NULL

[Chunhe Lan]

On 01/14/2014 06:14 PM, Benjamin Herrenschmidt wrote:
> On Tue, 2014-01-14 at 17:44 +0800, Chunhe Lan wrote:
>> Without this patch, kind of below error will be dumped if
>> 'insmod ixgbevf.ko' is executed:
>>
>>      ixgbevf: Intel(R) 10 Gigabit PCI Express Virtual Function
>>               Network Driver - version 2.7.12-k
>>      ixgbevf: Copyright (c) 2009 - 2012 Intel Corporation.
>>      ixgbevf 0000:01:10.0: enabling device (0000 -> 0002)
>>      ixgbevf 0000:01:10.0: No usable DMA configuration, aborting
>>      ixgbevf: probe of 0000:01:10.0 failed with error -5
>>      ......
>>      ......
> That's not right. The DMA ops must be set properly for the VF somewhere
> in the arch code instead. When creating VFs, is there a hook allowing
> the arch to fix things up ?
>
> (Also adding linux-pci on CC)
>
> Ben.
>
>> Signed-off-by: Chunhe Lan <Chunhe.Lan at freescale.com>
>> Cc: Benjamin Herrenschmidt <benh at kernel.crashing.org>
>> Tested-by: Chunhe Lan <Chunhe.Lan at freescale.com>
>> ---
>>   arch/powerpc/include/asm/dma-mapping.h |   13 +++++++++----
>>   1 files changed, 9 insertions(+), 4 deletions(-)
>>
>> diff --git a/arch/powerpc/include/asm/dma-mapping.h b/arch/powerpc/include/asm/dma-mapping.h
>> index e27e9ad..b8c10de 100644
>> --- a/arch/powerpc/include/asm/dma-mapping.h
>> +++ b/arch/powerpc/include/asm/dma-mapping.h
>> @@ -84,10 +84,15 @@ static inline struct dma_map_ops *get_dma_ops(struct device *dev)
         I see the get_dma_ops function in 
arch/*x86*/include/asm/dma-mapping.h as the following:

  32 static inline struct dma_map_ops *get_dma_ops(struct device *dev)
  33 {
  34 #ifndef CONFIG_X86_DEV_DMA_OPS
  35         return dma_ops;
  36 #else
  37         if (unlikely(!dev) || !dev->archdata.dma_ops)
  38                 return dma_ops;
  39         else
  40                 return dev->archdata.dma_ops;
  41 #endif
  42 }

         And also  see the get_dma_ops function in 
arch/*arm*/include/asm/dma-mapping.h as the following:

  18 static inline struct dma_map_ops *get_dma_ops(struct device *dev)
  19 {
  20         if (dev && dev->archdata.dma_ops)
  21                 return dev->archdata.dma_ops;
  22         return &arm_dma_ops;
  23 }

       Why not powerpc use this method to process dev == NULL ?

Thanks,
-Chunhe

>>   	 * only ISA DMA device we support is the floppy and we have a hack
>>   	 * in the floppy driver directly to get a device for us.
>>   	 */
>> -	if (unlikely(dev == NULL))
>> -		return NULL;
>> -
>> -	return dev->archdata.dma_ops;
>> +	if (dev && dev->archdata.dma_ops)
>> +		return dev->archdata.dma_ops;
>> +	/*
>> +	 * In some cases (for example, use the Intel(R) 10 Gigabit PCI
>> +	 * expression Virtual Function Network Driver -- ixgbevf.ko),
>> +	 * their value of dev is the NULL. If return NULL, the driver is
>> +	 * aborting. So return dma_direct_ops variable when dev == NULL.
>> +	 */
>> +	return &dma_direct_ops;
>>   }
>>   
>>   static inline void set_dma_ops(struct device *dev, struct dma_map_ops *ops)
>
>
>