On 28-05-2013 4:09, Tejun Heo wrote:
Yeah, that's exactly where I'm confused. ATA_DMA_BOUNDARY is 64k which becomes both queue_segment_boundary and dma seg_boundary. AFAICS, both __blk_segment_map_sg() and dma mapping won't merge across seg_boundary and as each bvec is a single page at most, we shouldn't need to worry about getting sg's which cross 64k boundaries in bmdma controllers. Hmmmmm...... I gotta be missing something. What am I missing here?
Probably nothing... maybe Jeff just copied that from ATADRVR (which was his inspiration IIRC).
Maybe we should ask him instead of guessing. :-)
Yeah, maybe. I can't find any reason why ata_bmdma_fill_sg() would need to worry about 64k boundary. The dumb variant is putting an extra restriction on it but the plain one doesn't seem to be adding anything. Would you be interested in submitting a patch to remove those?
Hm, at least not right now. I'm too busy.
BTW, I've just done some experimentation with my R-Car target and ATA_DEBUG/ATA_VERBOSE_DEBUG #define'd and it turned out that block layer didn't merge any segments at all... :-/
Even the ones with contiguous physical addresses?
Exactly. All DMA segments were 4K in size, regardless of whether the addresses were adjacent or not.
It could just be that the physical pages are disjoint as we don't really do anything to allocate pages contiguously and iommus don't kick in / merge unless necessary (it's not like modern hardware struggle with sglists anyway), so it could just be that there's nothing to merge.
There were mergeable segments, for sure.
Thanks.
MBR, Sergei -- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
