Re: [PATCH] x86: split e820 reserved entries record to late v4

[Ingo Molnar <mingo@elte.hu>]

* Yinghai Lu <yhlu.kernel@gmail.com> wrote:

> On Thu, Aug 28, 2008 at 11:58 PM, Ingo Molnar <mingo@elte.hu> wrote:
> >
> > * Yinghai Lu <yhlu.kernel@gmail.com> wrote:
> >
> >> >  BIOS-e820: 0000000077ff0000 - 0000000078000000 (reserved)
> >> >  BIOS-e820: 00000000e0000000 - 00000000f0000000 (reserved)
> >> >  BIOS-e820: 00000000fec00000 - 0000000100000000 (reserved)
> >> >
> >> > which overlaps with the chipset PCI BAR (hpet) resource:
> >> >
> >> >   pci 0000:00:14.0: BAR has HPET at fed00000-fed003ff
> >> >
> >> > so due to this 1K conflict we take the full e820-reserved entry out and
> >> > give the range 0xfec00000-0x100000000 as 'free'.
> >>
> >> you will get
> >> fec00000 - ffffffff reserved
> >>    fed0000 - fed003ff hpet
> >>      fed0000 - fed003ff 0000:00:14.0
> >
> > ok - because it's fully contained insert_resource() will succeed? I
> > thought it would only succeed if the new resource was smaller than (a
> > subset of) the existing resource. In the other direction, when a newly
> > inserted resource is a superset of the existing resource, i thought we'd
> > fail.
> >
> > hypothetical scenario, what if we had neither a superset nor a subset
> > scenario, but a partial overlap, between:
> >
> >> >  BIOS-e820: 00000000fec00000 - 0000000100000000 (reserved)
> >
> > and:
> >
> >> >   pci 0000:00:14.0: BAR has HPET at feb0f000-fec01000
> >
> > i.e. we have:
> >
> >      [... PCI BAR ...]
> >            [... e820 reservation ...]
> >
> > in that case the insert_resource() will fail due to the conflict. Can we
> > declare it in that case that the e820 reserved entry is mortally broken
> > and we just ignore it?
> 
> yes, that will fail to insert ...
> 
> expand to 0xfeb0f000 - 0xfffffff and try again.?
> 
> may need to update insert_resource to return conflict resource ...

yes, that sounds an excellent idea - i was thinking of something 
muchmore complex like breaking up the reserved entry - but indeed just 
creating a large enough superset should be perfect. I.e. extend both 
start and end until we fit fully. [or reach some natural boundary such 
as 0 or 4GB]

> > At least we should emit a prominent warning if insert_resource() fails,
> > and add in an mdelay(2000) so that the user sees it.
>
> right

btw., perhaps we should try this: first try a request_resource(). If 
that fails it means we overlap with something - then we should already 
printk a warning. (e820 reserved entries should never conflict with PCI 
resources, should they?)

then try an insert_resource(). If that too fails it means a partial 
overlap - printk another warning. Try the extension (within reasonable 
limits) and retry.

Does that sound worthwile?

	Ingo