Re: [patch] fs: improved handling of page and buffer IO errors

[Matthew Wilcox <matthew@wil.cx>]

On Wed, Oct 22, 2008 at 04:45:24PM +0200, Miklos Szeredi wrote:
> On Wed, 22 Oct 2008, Matthew Wilcox wrote:
> > remap_file_pages() only hurts if you map the same page more than once
> > (which is permitted, but again, I don't think anyone actually does
> > that).
> 
> This is getting very offtopic... but remap_file_pages() is just like
> MAP_FIXED, that the address at which a page is mapped is determined by
> the caller, not the kernel.  So coherency with other, independent
> mapping of the file is basically up to chance.

Oh, right, I see.

> Do we care?  I very much hope not.  Why do PA-RISC and friends care at
> all about mmap coherecy?  Is it real-world problem driven or just to
> be safe?

One reason to care is performance.  If two tasks have libc mapped
coherently, then the addresses will collide in the cache and they'll
use the same cachelines.  I don't know how much applications care about
correctness with mmap -- they certainly do with sysv shm.  I probably
knew about some applications that broke when I was working on this code --
back in 2002.

-- 
Matthew Wilcox				Intel Open Source Technology Centre
"Bill, look, we understand that you're interested in selling us this
operating system, but compare it to ours.  We can't possibly take such
a retrograde step."