On Wed, Mar 01 2006, Douglas Gilbert wrote: > James Bottomley wrote: > > On Wed, 2006-03-01 at 16:29 +0100, Jens Axboe wrote: > > > >>Strictly speaking, the clustering bit is unrelated. I seem to recall > >>Gerard years ago talking about some sym chips that did not like > >>clustering, hence it was disabled. > > > > > > Yes, I remember that too ... I've never been able to find out which > > chip, though ... the scripts all seem happily coded for variable size sg > > segments. > > > > However, given the new way 2.6 does memory allocations, > > ENABLE_CLUSTERING will probably make quite a difference to the size of > > the sg list ... since we try to allocate contiguous pages, physical > > merging becomes much more of a possibility (I think I last measured it > > at around 30% of all SG tables, as opposed to <1% with the old > > allocation method). > > James, > So the maximum data carrying size of a scatter gather list > is not deterministic? Is the worst case (page_size * SG_ALL)? It's block layer restricted (well not because of the block layer, but the limits that the hardware tolds us that it has), typically page_size * max_sg_entries is the correct answer but it may of course be more. -- Jens Axboe - : send the line "unsubscribe linux-scsi" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
