Re: [PATCH] x86_64: setup saved_max_pfn correctly (kdump)

["Magnus Damm" <magnus.damm@gmail.com>]

On 11/3/06, Vivek Goyal <vgoyal@in.ibm.com> wrote:
> On Fri, Nov 03, 2006 at 03:05:08AM +0900, Magnus Damm wrote:
> > Hi Mel,
> >
> > Thanks for your input! Great work with the add_active_range() code.
> >
> > On 11/3/06, Mel Gorman <mel@csn.ul.ie> wrote:
> > >Hey Magnus,
> > >
> > >I see what you are doing and why. However if you look in
> > >arch/x86_64/kernel/setup.c, you'll see
> > >
> > >         parse_early_param();
> > >
> > >         finish_e820_parsing();
> > >
> > >         e820_register_active_regions(0, 0, -1UL);
> > >
> > >If you just called e820_register_active_regions(0, 0, -1UL) before
> > >parse_early_param(), would it still fix the problem without having to call
> > >e820_register_active_regions(0, 0, -1UL) twice?
> >
> > Well, I guess it is possible to move the
> > e820_register_active_regions() up, but I'm not sure if that would give
> > us anything.
> >
> > We need to call e820_register_active_regions() before e820_end_of_ram,
> > that's for sure, but the "exactmap" code in parse_memmap_opt() sets
> > e820.nr_map to 0 after the call to e820_end_of_ram(). Then it adds a
> > new set of user-supplied ranges to the e820 map which then need to be
> > registered using e820_register_active_regions().
> >
> > So yeah, we can move the function up above parse_early_param() but
> > then we need to insert another call to e820_register_active_regions()
> > somewhere after all user-supplied ranges have been added.
> >
> > Another solution could be to rewrite e820_end_of_ram() to instead scan
> > e820.map[] backwards from e820.nr_map - 1 to locate the last ram page.
> > But can you do that in two lines of code? =)
> >
>
> Is there are reason that why e820_end_of_ram() should be looking at active
> regions instead of dicrectly probing e820 memory map? If no, then modifying
> e820_end_of_ram() wil save us extra calls.

No special reason. I was just happy that I was able to solve the
problem by using two lines of new kernel functions. =) Also, the code
is not timing critical at all so I see no reason to write up something
else when we already have a working solution.

Feel free to hack up with a better patch and I'll test and report
before end of next week.

Thanks!

/ magnus