Re: [PATCH 004 of 9] md: Factor out part of raid10d into a separate function.

On Tuesday August 1, davidsen@tmr.com wrote:

don't think this is better, NeilBrown wrote:

quoted

raid10d has toooo many nested block, so take the fix_read_error
functionality out into a separate function.
 

Definite improvement in readability. Will all versions of the compiler 
do something appropriate WRT inlining or not?

As the separated function is called about once in a blue moon, it
hardly matters.  I'd probably rather it wasn't inlined so as to be
sure it doesn't clutter the L-1 cache when it isn't needed, but that's
the sort of thing I really want to leave to the compiler.

Maybe it would be good to stick an 'unlikely' or 'likely' in raid10d
to tell the compiler how likely a read error is...

NeilBrown