Hi Andrew, [I cc'ed the people relevant to this issue] On Sat, Oct 12, 2002 at 10:26:44AM -0700, Andrew Morton wrote: > Ingo Oeser wrote: > > Stupid question: Would you accept a patch that extends > > get_user_pages() to accept an additional "struct scatterlist vector[]"? > > It's not really my area Ingo. But I can wave such a patch about > on the mailing lists, generally get it some review and attention > I guess. I had waved an example on what is really needed instead of kiobuf crap some time ago[1]. This raised a discussion on linux-scsi[2] (but I'm not subscribed there) and someone[2] actually successfully tested this. > Such nfrastructure would need something which used it, as a proof-of-concept, > testbed, etc... I would love to test my ideas out, but the special purpose device where I need it for has bit-errors on its big SDRAM chips and I can only use a small 32K area of storage for testing, which is not expected to reveal any noticable performance from that method, due to the high setup overhead. I think you know the numbers from direct-io. The video hardware, where you (or Geert?) basically implemented the things, we proposed[3] would be a perfect testbed for this. Thanks & Regards Ingo Oeser [1] <20020720003918.G758@nightmaster.csn.tu-chemnitz.de> on lkml [2] on linux-scsi [3] Attached here. -- Science is what we can tell a computer. Art is everything else. --- D.E.Knuth