Re: HPET regression in 2.6.26 versus 2.6.25 -- revert for 2.6.26-rc1 failed

[Bill Fink <billfink@mindspring.com>]

On Thu, 14 Aug 2008, Yinghai Lu wrote:

> On Thu, Aug 14, 2008 at 3:04 AM, Bill Fink <billfink@mindspring.com> wrote:
> > Hi David,
> >
> > On Wed, 13 Aug 2008, David Witbrodt wrote:
> >
> >> [Yinghai, please note that I did not request a patch to revert the
> >> problem commit.  I was merely experimenting -- on my own time, so
> >> you folks would not have to bother -- to see if I could make it
> >> work.  I should have made that more clear!  Having said that, I am
> >> glad to test changes of any kind on my machine:  reverts, code for
> >> debugging or info, experiments, etc.]
> >
> > I'm not sure Yinghai's revert patch is completely equivalent to
> > a revert of the original problematic commit, by a side-by-side
> > comparison of the original commit with his recent revert patch,
> > but then I don't really know that code at all.
> >
> > In the original code there was a section (in e820_reserve_resources()):
> >
> > #ifdef CONFIG_KEXEC
> >                   if (crashk_res.start != crashk_res.end)
> >                           request_resource(res, &crashk_res);
> > #endif
> >
> > If you don't have CONFIG_KEXEC defined in your .config, which is
> > probably the case, then you would never request a crashk_res resource.
> > But in the code after the original commit, it unconditionally calls
> > (in reserve_crashkernel()):
> >
> >           crashk_res.start = crash_base;
> >           crashk_res.end   = crash_base + crash_size - 1;
> >           insert_resource(&iomem_resource, &crashk_res);
> >
> > And after Yinghai's revert patch it still does (in reserve_crashkernel()):
> >
> >        crashk_res.start = crash_base;
> >        crashk_res.end   = crash_base + crash_size - 1;
> >        crashk_res_ptr = &crashk_res;
> >
> > and (in setup_arch()):
> >
> >        num_res = 3;
> >        if (crashk_res_ptr) {
> >                res_kernel[num_res] = crashk_res_ptr;
> >                num_res++;
> >        }
> >        e820_reserve_resources(res_kernel, num_res);
> >
> > then (in e820_reserve_resources()):
> >
> >                        for (j = 0; j < nr_res_k; j++) {
> >                                if (!res_kernel[j])
> >                                        continue;
> >                                request_resource(res, res_kernel[j]);
> >                        }
> >
> > which for j == 3 is:
> >
> >        request_resource(res, &crashk_res);
> >
> > Now it would appear that the new:
> >
> >        insert_resource(&iomem_resource, &crashk_res);
> >
> > or new:
> >
> >        request_resource(res, &crashk_res);
> >
> > should be noops.  But if for any reason crash_size is not zero,
> > then there could be a difference.  I have no idea if this is at all
> > significant, but I thought I'd point it out just in case.
> 
> why oops ?
> if not valid crash kernel size etc is input, crashk_res_ptr will be null
> 
> >        if (crashk_res_ptr) {
> >                res_kernel[num_res] = crashk_res_ptr;
> >                num_res++;
> >        }
> 
> it that is not appended to res_kernel...

You're right.  Looking just at the diffs, I didn't realize that all
of reserve_crashkernel() is inside "#ifdef CONFIG_KEXEC" and thus
crashk_res_ptr is probably null in David's kernel.  Unless of course,
in the unlikey event that the memory location for crashk_res_ptr was
being corrupted somehow.

						-Bill