|| ||Andrew Morton <firstname.lastname@example.org>|
|| ||Andi Kleen <email@example.com>|
|| ||Re: [PATCH prototype] [0/8] Predictive bitmaps for ELF executables|
|| ||Wed, 19 Mar 2008 02:04:40 -0700|
|| ||firstname.lastname@example.org, email@example.com|
On Wed, 19 Mar 2008 09:32:28 +0100 Andi Kleen <firstname.lastname@example.org> wrote:
> On Tue, Mar 18, 2008 at 10:44:37AM -0700, Andrew Morton wrote:
> > On Tue, 18 Mar 2008 18:20:45 +0100 Andi Kleen <email@example.com> wrote:
> > > > What's the permission problem? executable-but-not-readable files? Could
> > >
> > > Not writable.
> > Oh.
> > I doubt if a userspace implementation would even try to alter the ELF
> > files, really - there seems to be no point in it. This is just complexity
> Well the information has to be somewhere and i think the ELF file
> is the best location for it. It makes it the most user transparent.
Adopt a standard, stick with it.
Assuming that all users have the same access pattern might be inefficient,
a little bit. There might be some advantage to making it per-user, dunno.
The requirement to write to an executable sounds like a bit of a
> > > Yes it could, but i dont even want to thi nk about all the issues of
> > > doing such an interface. It is basically an microkernelish approach.
> > > I prefer monolithic simplicity.
> > It's not complex at all. Pass a null-terminated pathname to the server and
> > keep running. The server will asynchronously read your pages for you.
> But how do you update the bitmap in your scheme?
will write the bitmap to ~/.bitmaps/usr/lib64/firefox-220.127.116.11/firefox-bin ?
if it proves useful, build it all into libc..
I'm assuming that the per-page minor fault cost is relatively low and that
the major benefit is in disk scheduling[*]. If that's false then we'd need
kernel support I guess - some sort of gang-fault syscall?
* solid-state disks are going to put a lot of code out of a job.