Re: [PATCH 4/5] I/OAT: Tighten descriptor setup performance

[Andrew Morton <akpm@linux-foundation.org>]

On Wed, 17 Oct 2007 17:14:33 -0700
Shannon Nelson <shannon.nelson@intel.com> wrote:

> The change to the async_tx interface cost this driver some performance by
> spreading the descriptor setup across several functions, including multiple
> passes over the new descriptor chain.  Here we bring the work back into one
> primary function and only do one pass.
> 
> Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
> ---
> 
>  drivers/dma/ioat_dma.c |  170 +++++++++++++++++++++++++-----------------------
>  drivers/dma/ioatdma.h  |    6 +-
>  2 files changed, 93 insertions(+), 83 deletions(-)
> 
> diff --git a/drivers/dma/ioat_dma.c b/drivers/dma/ioat_dma.c
> index c44f551..117ac38 100644
> --- a/drivers/dma/ioat_dma.c
> +++ b/drivers/dma/ioat_dma.c
> @@ -46,9 +46,12 @@
>  /* internal functions */
>  static void ioat_dma_start_null_desc(struct ioat_dma_chan *ioat_chan);
>  static void ioat_dma_memcpy_cleanup(struct ioat_dma_chan *ioat_chan);
> +static inline struct ioat_desc_sw *ioat_dma_get_next_descriptor(
> +					      struct ioat_dma_chan *ioat_chan);

A forward-declared static inline is pretty weird.  Does gcc actually do the
right thing with it?

This function is far too large to be inlined anyway.