Re: ramdisk corruption problems - was: RE: pivot_root and initrdkern el panic woes

[Andrea Arcangeli]

On Sun, Dec 30, 2001 at 04:35:46PM -0800, Linus Torvalds wrote:
> 
> On Mon, 31 Dec 2001, Andrea Arcangeli wrote:
> >
> > actually bh_new is needed also to serialize with the buffercache, a new
> > bh mapped in pagecache must be dropped from the buffercache before we
> > can start using it (unmap_underlying_metadata).
> 
> You're right, although it's something of an optimization (ie we could as
> well just depend on the "mapped" bit and watch it change).

we could even hold the optimization (cache coherency only on new blocks)
by pushing the cache coherency into the lowlevel just like the bh
clearing, but the current buffer_new branch in the library code seems
clean (and potentially a little faster with big softblocksize due the
partial clearing).

Andrea