Messages in this thread | | | From | Linus Torvalds <> | Date | Sat, 23 Apr 2022 14:02:36 -0700 | Subject | Re: [fs/pipe] 5a519c8fe4: WARNING:at_mm/page_alloc.c:#__alloc_pages |
| |
On Sat, Apr 23, 2022 at 1:23 PM Andrei Vagin <avagin@gmail.com> wrote: > > vmsplice & splice is the best. /proc/pid/mem is 30% slower. > process_vm_readv is 20% slower.
Hmm.
I wonder how much low-hanging fruit there would be to optimize there.
Because process_vm_readv() in particular is pretty much _designed_ for what you do, so it would be much nicer to basically work on that being optimal, rather than doing vmsplice hackery.
(And yes, vmsplice is designed for efficient copies from the VM too, of course, but that "limited to kernel pipe buffer" part is arguably even more fundamental to it)
Linus
| |