Hi Ulf/Arend, > > > On Wed, Nov 26, 2014 at 11:40 PM, Arend van Spriel <ar...@broadcom.com> > > wrote: > > On 11/26/14 14:43, Ulf Hansson wrote: > > > On 26 November 2014 at 12:07, Avinash Patil<pat...@marvell.com> wrote: > > > From: Bing Zhao<bz...@marvell.com> > > > > This patch adds new API to handle scatter gather aggregation. > > > Why is this needed? > > One thing I can imagine is that the receiving device has specific > scatter-gather alignment requirements that are not covered in > mmc_io_rw_extended(). So the new API does not so much handle scatter- > gather, but leaves that up to the caller. Question is whether > mmc_io_rw_extended() could be changed to accommodate those requirements > instead of duplicating big part of it. Without knowing the exact > requirements that is difficult to answer. > > Regards, > Arend
My apologies for opening discussion on this topic after long time. Earlier we had dropped this patch, because we didn't see significant performance improvement with it. Now we have optimized code in our WLAN driver and are planning to have this patch. Existing mmc_io_rw_extended() API expects caller to pass single contiguous buffer. It will be split if it exceeds segment size, SG table is prepared and used it for reading/writing the data. Our intention for defining new API here is to facilitate caller to provide ready SG table(scattered buffer list). On one of our platforms maximum segments are 256 with each segment of 4k size. So this new API would be really beneficial there for performance improvement. Let me know your suggestions/thoughts. Regards, Amitkumar
