Messages in this thread | | | Subject | RE: [RFC v3 4/7] dmaengine: Add slave DMA interface | Date | Tue, 19 Feb 2008 10:46:54 -0800 |
| |
>From: Haavard Skinnemoen [mailto:hskinnemoen@atmel.com] >Sent: Monday, February 18, 2008 5:30 AM >To: Nelson, Shannon >Cc: Haavard Skinnemoen; Williams, Dan J; >linux-kernel@vger.kernel.org; David Brownell; >kernel@avr32linux.org; Francis Moreau; Paul Mundt; Vladimir A. >Barinov; Pierre Ossman >Subject: Re: [RFC v3 4/7] dmaengine: Add slave DMA interface > >On Fri, 15 Feb 2008 09:12:35 -0800 >"Nelson, Shannon" <shannon.nelson@intel.com> wrote: > >> I'll jump in here briefly - I'm okay with the direction this >is going, >> but I want to be protective of ioatdma performance. As used >in struct >> ioat_desc_sw, the cookie and ack elements end up very close >to the end >> of a cache line and I'd like them to not get pushed out across the >> boundry. I don't think this proposal changes the layout, I'm just >> bringing up my concern. > >Sure, performance is very important, and it's good to see that you're >critical about the changes I'm proposing. That said, the memory layout >doesn't change at all with this patch -- the fields that didn't go into >the generic dma descriptor were at the end of the struct to begin with. > >I can add a comment saying that cookie and ack must always come first. >Any other fields that we need to be careful about? > >Haavard >
Those are the only two that I'm worried about at the moment. I'm just hoping that a quirk in some compiler's struct packing doesn't push them over that edge.
Thanks, sln
| |