[PATCH] D25467: [ELF] - Alternative fix to prevent possible crash on large output.

[Rui Ueyama via llvm-commits]

IIRC, Windows can learn disk access patterns for process startup sequences
so that it look-aheads executable files. I can imagine that it could even
(un-)defrag executable files on disk so that scattered disk access pattern
for a process boot becomes sequential accesses on an actual disk.

On Mon, Oct 17, 2016 at 3:30 PM, Davide Italiano <davide at freebsd.org> wrote:

> On Mon, Oct 17, 2016 at 3:21 PM, Rui Ueyama <ruiu at google.com> wrote:
> > On Mon, Oct 17, 2016 at 3:06 PM, Davide Italiano <davide at freebsd.org>
> wrote:
> >>
> >> On Mon, Oct 17, 2016 at 2:26 PM, Rui Ueyama via llvm-commits
> >> <llvm-commits at lists.llvm.org> wrote:
> >> > This is a list of items that popped in my mind. Not exhaustive but
> >> > probably
> >> > good enough to start discussing what we should do next.
> >> >
> >> >  - FreeBSD: what is the status of building the entire FreeBSD system
> >> > with
> >> > LLD? Can it build everything including the kernel?
> >> >
> >>
> >> Another idea, which I discussed with Rafael and Michael a bit ago.
> >> Implement support for order file. For starters, you don't need anything
> >> fancy.
> >> Just get a list of symbols (which can be generated by systap or a
> >> profiler) and pass that to the linker, so that sections are laid out
> >> according to the partial order defined in the file (you can always
> >> reconstruct a total order starting from a partial order using some
> >> heuristic). This is particularly useful to improve startup time as you
> >> reduce the amount of paging. Firefox uses this for sure, and
> >> apparently ld64 implements this algorithm (or a slight variation of
> >> it). I'm confident this is quite effective (see Martin Liska's thesis
> >> which contains a more detailed description of the problem and
> >> references https://arxiv.org/pdf/1403.6997.pdf)
> >
> >
> > That's an effective technique for spinning disks for sure, but how about
> on
> > SSDs?
> >
>
> I'm not sure, maybe not, but I think still it's worth a try as
> implementing it shouldn't be a herculean task.
> When I spoke about it with Bob Wilson he mentioned that it works for
> them (ld64). They feed the output of a dtrace script to the linker in
> particular. Your (our?) mileage may vary.
>
> --
> Davide
>
> "There are no solved problems; there are only problems that are more
> or less solved" -- Henri Poincare
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20161017/a1322a61/attachment.html>