On Fri, Jun 03, 2022 at 09:09:06AM -0400, Mike Snitzer wrote: > I think the point was to keep the biosets embedded in struct > mapped_device to avoid any possible cacheline bouncing due to the > bioset memory being disjoint. Probably. > But preserving that micro-optimization isn't something I've ever > quantified (at least not with focus). But we can either avoid the pointer or do the preallocation. Doing both doesn't really work. OTOH given that we've surived quite a while without the preallocation actually working I wonder if it actually is needed. > Looks fine, did you want to send an official patch with a proper > header and Signed-off-by? I plan to send a series with a few follow on after testing it a bit more.
