On Tue, Feb 07, 2017 at 09:40:04AM +0000, David Howells wrote:

Matthew Wilcox [off-list ref] wrote:

quoted

[adding linux-arch to see if anyone there wants to do an optimised
version of memfill for their CPU]

For mn10300, this is superfluous since the memset() implementation will do
optimised filling of up to 8 x 4 bytes per loop if the alignments suit.

This is also superfluous for frv as that will do up to 8 x 8 bytes per loop.

So on both those arches, memfill() should probably just wrap memset().

You've misunderstood the purpose of memfill.  memfill allows the caller
to specify a pattern which is not a single byte in size, eg memfill(addr,
0x12345678, 64) would result in 0x12345678 being reproduced 16 times.
memset(addr, 0x12345678, 64) would result in 0x78 being reproduced
64 times.