x86-64 2.6.15-rc2-git5 fails to boot with 4GB memory

x86-64 2.6.15-rc2-git5 fails to boot with 4GB memory

[Matti Aarnio]

With 2 GB in place, the kernel boots just fine, but with
4 GB, it reports:

 kernel direct mapping tables upto ffff 8101 5000 000 @ 8000-f000
 PANIC: early exception rip  ffff ffff 8016 f002 error 0 cr2 4230
 PANIC: early exception rip  ffff ffff 8011 d1fe error 0 cr2 ffff ffff f5ff d023

and some other lines, which I didn't jot down on paper...
These were copied from some Fedora Core development kernel version
after 2.6.15-rc1 (last working one) in a box with 4 GB memory.

Those hex values didn't have intermediate spaces in them, though.
That was me trying to understand 64 bit values.

Last working kernel with all 4 GB memory in the box was  2.6.15-rc1
Since then the kernels have failed to boot at all, unless machine
PHYSICAL memory is stripped down to 2 GB.  Command-line options
(e.g. "mem=2G") don't help at all.

  /Matti Aarnio

Re: x86-64 2.6.15-rc2-git5 fails to boot with 4GB memory

[Andi Kleen]

Matti Aarnio <matti.aarnio@zmailer.org> writes:

> With 2 GB in place, the kernel boots just fine, but with
> 4 GB, it reports:

Works for me on several machines.

I even have a fix for the Asus wrong MCFG problem now that
broke the IOMMU on these boards (workaround is pci=nommconf) 

> 
>  kernel direct mapping tables upto ffff 8101 5000 000 @ 8000-f000
>  PANIC: early exception rip  ffff ffff 8016 f002 error 0 cr2 4230
>  PANIC: early exception rip  ffff ffff 8011 d1fe error 0 cr2 ffff ffff f5ff d023
> 
> and some other lines, which I didn't jot down on paper...

Can you please look up the RIP values in your System.map? 

> These were copied from some Fedora Core development kernel version
> after 2.6.15-rc1 (last working one) in a box with 4 GB memory.

Please try vanilla 2.6.15rc2 as a reference at least.

-Andi

Re: x86-64 2.6.15-rc2-git5 fails to boot with 4GB memory

[Matti Aarnio]

On Tue, Nov 29, 2005 at 07:01:12AM -0700, Andi Kleen wrote:
> Matti Aarnio <matti.aarnio@zmailer.org> writes:
> 
> > With 2 GB in place, the kernel boots just fine, but with
> > 4 GB, it reports:
> 
> Works for me on several machines.
> 
> I even have a fix for the Asus wrong MCFG problem now that
> broke the IOMMU on these boards (workaround is pci=nommconf) 
> 
> > 
> >  kernel direct mapping tables upto ffff 8101 5000 000 @ 8000-f000
> >  PANIC: early exception rip  ffff ffff 8016 f002 error 0 cr2 4230
> >  PANIC: early exception rip  ffff ffff 8011 d1fe error 0 cr2 ffff ffff f5ff d023
> > 
> > and some other lines, which I didn't jot down on paper...
> 
> Can you please look up the RIP values in your System.map? 
> 
> > These were copied from some Fedora Core development kernel version
> > after 2.6.15-rc1 (last working one) in a box with 4 GB memory.
> 
> Please try vanilla 2.6.15rc2 as a reference at least.

Tried.  Crashes with 4 GB memory present in the box.
Boots and runs nicely with 2 GB memory populated in.

After adding  -g  to  *CFLAGS of top-level  Makefile, and
trying to determine WHERE those PANICs happened in rc2:

(gdb) list *0xffffffff80163a43
0xffffffff80163a43 is in memmap_init_zone (mm/page_alloc.c:1687).
1682            for (pfn = start_pfn; pfn < end_pfn; pfn++, page++) {
1683                    if (!early_pfn_valid(pfn))
1684                            continue;
1685                    if (!early_pfn_in_nid(pfn, nid))
1686                            continue;
1687                    page = pfn_to_page(pfn);
1688                    set_page_links(page, zone, nid, pfn);
1689                    set_page_count(page, 1);
1690                    reset_page_mapcount(page);
1691                    SetPageReserved(page);

(gdb) list *0xffffffff801196fa
0xffffffff801196fa is in safe_smp_processor_id (include/asm/smp.h:77).
72      #define raw_smp_processor_id() read_pda(cpunumber)
73
74      static inline int hard_smp_processor_id(void)
75      {
76              /* we don't want to mark this access volatile - bad code generation */
77              return GET_APIC_ID(*(unsigned int *)(APIC_BASE+APIC_ID));
78      }
79
80      extern int safe_smp_processor_id(void);
81      extern int __cpu_disable(void);


Not that those explain all that much...


> -Andi

/Matti Aarnio

Re: x86-64 2.6.15-rc2-git5 fails to boot with 4GB memory

[Andi Kleen]

> Not that those explain all that much...

Can you send me your .config? If you have SPARSEMEM enabled can you
disable it?

-Andi

Re: x86-64 2.6.15-rc2-git5 fails to boot with 4GB memory

[Keith Mannthey]

On 11/29/05, Andi Kleen <ak@suse.de> wrote:
> > Not that those explain all that much...
>
> Can you send me your .config? If you have SPARSEMEM enabled can you
 > disable it?

This looks just like the sparsemem troubles.  There is a patch around
somwhere....  I thought a patch was being pushed into mainline but I
guess not.

Thanks,
 Keith

Re: x86-64 2.6.15-rc2-git5 fails to boot with 4GB memory

[Andi Kleen]

On Tue, Nov 29, 2005 at 06:26:28PM -0800, Keith Mannthey wrote:
> On 11/29/05, Andi Kleen <ak@suse.de> wrote:
> > > Not that those explain all that much...
> >
> > Can you send me your .config? If you have SPARSEMEM enabled can you
>  > disable it?
> 
> This looks just like the sparsemem troubles.  There is a patch around
> somwhere....  I thought a patch was being pushed into mainline but I
> guess not.

It was I think. But I still don't trust it.

-Andi