Re: [parisc-linux] rmap: parisc __flush_dcache_page

[James Bottomley <James.Bottomley@steeleye.com>]

On Thu, 2004-04-08 at 13:18, Andrea Arcangeli wrote:
> it enterely depends on the workload. On a desktop machine there may be
> only some hundred entries in those lists at maximum with glibc being the
> biggest offender:
> 
> cat /proc/*/maps | grep libc.so.6 | wc -l
> 
> with shared memory on some server there can be easily several thousand
> entries for some inode even on 64bit, but a timeslice was probably
> exaggerated (the timeslice was for the walking of the ptes in each
> mapping too, I don't think you need to look at every pte).

So you're worried about our code?  OK, if you look, you'll see we only
have to flush one address in the mmap_shared list, (which is usually the
long list).

I'd constructed it on the predicate that flushing a non-current space is
more expensive than finding a current one, but I can alter it to flush
the first vma it comes to with a present translation.

The mmap list is usually empty.  We only excite that case for multiple
private mappings of a file which for some reason gets updated.

I'd be very surprised if flush_dcache_page executes more than a few
hundred instructions all told...that's certainly nowhere close to a
timeslice.

James