From: Paul McKenney <Paul.McKenney@us...> - 2001-11-02 14:19:34
> > My guess is that the read-side code would be greatly simplified if the
> > size of the array/bitflag was in the same memory block as the
> > array/bitflag,
> > so that there would only be the dependency on the pointer. Maneesh,
> > thoughts?
> Right now size fields for the array and bitmap are not poniers, are you
> that we should make them integer pointer and allocate them whith the
> Also I couldnot get how doing that will ensure ordering?
> Actually putting rmb() on read side is not very complicated as we have
> inline functions (fcheck() and fcheck_files()) and putting rmb() in those
> should do for most of the cases.
I am suggesting creating a struct that contains both the size and the
array, and having a pointer to this struct. This would allow the code
to pick up the pointer to this combined size&array struct, do an rmb(),
and then be sure that the size and the array are guaranteed to match.
This would add some lines to the patch, but would simplify the assumptions
the code is making.
Get latest updates about Open Source Projects, Conferences and News.