On Tue, Jun 18, 2013 at 04:11:16PM +0100, Russell King - ARM Linux wrote: > On Tue, Jun 18, 2013 at 05:02:49PM +0200, Linus Walleij wrote: > > Nowadays I would do the above with regmap_update_bits(). > > Mutual exclusion for read-modify-write of individual bits in a > > register is one of those cases where doing a regmap over > > a memory-mapped register range makes a lot of sense. > > (drivers/mfd/syscon.c being a nice example) > So, for that solution we need to have some kind of global regmap per > register or somesuch. Then you run into regmap needing a struct > device - well, with a shared register, which struct device do you > use, or do you have to invent one? > That sounds more heavy-weight than is really necessary. Yes, regmap is far too heavyweight for a lot of these things - it shouldn't really go in performance critical at the CPU level paths, it's designed around things where I/O costs are considerable and worth avoiding. I do think that this is a very good idea - with the I2C/SPI drivers I was reviewing prior to regmap there were always just far too many simple bugs with read/modify/write cycles so factoring out the code really helped with review.
Attachment:
signature.asc
Description: Digital signature
