Re: [Qemu-devel] [PATCH 05/10] dma-helpers: add dma_buf_read and dma_buf_write

[Kevin Wolf <kwolf@redhat.com>]

Am 11.08.2011 16:24, schrieb Paolo Bonzini:
> On 08/11/2011 03:29 PM, Stefan Hajnoczi wrote:
>>>>
>>>>  What do you think about passing the residual bytes for short transfers?
>>>>    Should I look into updating BlockDriverCompletionFunc, or is the approach
>>>>  of patch 2 okay?  If I have an excuse to learn more about Coccinelle, that
>>>>  can be fun.:)
>> The bdrv_aio_readv() and bdrv_aio_writev() functions don't have the
>> concept of residual bytes.  They only work on fully completed I/O
>> operations.  If there is an error they pass -errno.
> 
> But if a transfer was split due to failure of cpu_physical_memory_map, 
> and only the second part fails, you can have a short transfer and you 
> need to pass residual bytes back.  The only way out of this is to make a 
> bounce buffer as big as all the unmappable parts of the S/G list, which 
> is undesirable of course.  So the residual bytes are a general DMA 
> concept, not specific to SCSI.
> 
>> Therefore I don't think BlockDriverCompletionFunc is the right type
>> to add residual bytes to.
> 
> Right, I would rather update BlockDriverCompletionFunc to pass the AIOCB 
> as a third parameter, and store the residual bytes in the DMAAIOCB (with 
> a getter that the completion function can use).

Isn't the DMAAIOCB already passed as opaque to the callback?

Kevin