On Tue, Dec 17, 2024 at 05:13:39PM +0200, Vladimir Oltean wrote: > On Sat, Dec 14, 2024 at 04:11:54PM +0100, Christian Marangi wrote: > > We can see that: > > - as suggested regmap doesn't cause any performance penality. It does > > even produce better result. > > - the read/set/restore implementation gives worse performance. > > > > So I guess I will follow the path of regmap + cache page. What do you > > think? > > I'm not seeing results with the "times" variable changed, but in > general, I guess the "switch regmap + page" and "switch regmap + phy > regmap" will remain neck and neck in terms of performance, surpassing > the "switch regmap restore" techniques more and more as "times" > increases. So going with a PHY regmap probably sounds good. Could you find a way to reuse Maxime's mdio-regmap.c driver? Either create separate regmaps for each PHY address, or make that driver accept a configuration which isn't limited to a single ctx->valid_addr?
