Messages in this thread | | | Date | Thu, 7 Jan 2021 18:55:59 +0000 | From | Al Viro <> | Subject | Re: [x86] d55564cfc2: will-it-scale.per_thread_ops -5.8% regression |
| |
On Thu, Jan 07, 2021 at 06:40:58PM +0000, Al Viro wrote: > do_sys_poll(): do the wholesale copyout > > Don't bother with patching up just one field - 16 bits out of each 64. > The amount of memory traffic is not going to be greater (might be > smaller, actually) and the loop in copy_to_user() is optimized for > bulk copy.
BTW, considering the access pattern, I would expect it to be considerably cheaper in a lot of cases; basically, we have a copy of userland array of 64bit values, then we do a non-trivial amount of work and modify 16 bits out of each 64. Then we want that propagated back to the original array. I suspect that copying just those 16bit fields out is going to cost more that a bulk copy of the entire thing, and not just on s390 and similar oddball cases.
Comments?
| |