Patchwork 2.6.28-rc9 panics with crashkernel=256M while booting

login
register
mail settings
Submitter Andrew Morton
Date Dec. 25, 2008, 7:35 a.m.
Message ID <20081224233536.b067c9da.akpm@linux-foundation.org>
Download mbox | patch
Permalink /patch/15561/
State Not Applicable
Headers show

Comments

Andrew Morton - Dec. 25, 2008, 7:35 a.m.
(cc's added)

On Wed, 24 Dec 2008 13:25:49 +0530 Chandru <chandru@in.ibm.com> wrote:

> On a ppc machine booting linux-2.6.28-rc9 with crashkernel=256M@32M boot 
> parameter causes the kernel to panic while booting. __Following are the console 
> messages...

- Please put [patch] in the Subject: line of patches

- Please choose a suitable title, as per
  Documentation/SubmittingPatches, section 15.

- Please cc suitable mailing lists and maintainers on bug reports and
  on patches.



From: Chandru <chandru@in.ibm.com>

When booted with crashkernel=224M@32M or any memory size less than this,
the system boots properly.  The following was the observation..  The
system comes up with two nodes (0-256M and 256M-4GB).  _The crashkernel
memory reservation spans across these two nodes.  _The
mark_reserved_regions_for_nid() in arch/powerpc/mm/numa.c resizes the
reserved part of the memory within it as:

_ _ _ _ _ _ if (end_pfn > node_ar.end_pfn)
_ _ _ _ _ _ _ _ reserve_size = (node_ar.end_pfn << PAGE_SHIFT)
_ _ _ _ _ _ _ _ _ _ - (start_pfn << PAGE_SHIFT);


but the reserve_bootmem_node() in mm/bootmem.c raises the pfn value of end 

_ _ end = PFN_UP(physaddr + size);

This causes end to get a value past the last page in the 0-256M node. 
_Again when reserve_bootmem_node() returns,
_mark_reserved_regions_for_nid() loops around to set the rest of the
crashkernel memory in the next node as reserved.  _ It references
NODE_DATA(node_ar.nid) and this causes another 'Oops: kernel access of bad
area' problem.  The following changes made the system to boot with any
amount of crashkernel memory size.

Signed-off-by: Chandru S <chandru@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
---

 arch/powerpc/mm/numa.c |    7 ++++---
 mm/bootmem.c           |    4 ++++
 2 files changed, 8 insertions(+), 3 deletions(-)
Paul Mackerras - Dec. 26, 2008, 12:59 a.m.
Andrew Morton writes:

> On Wed, 24 Dec 2008 13:25:49 +0530 Chandru <chandru@in.ibm.com> wrote:
> 
> > On a ppc machine booting linux-2.6.28-rc9 with crashkernel=256M@32M boot 
> > parameter causes the kernel to panic while booting. __Following are the console 
> > messages...
> 
> - Please put [patch] in the Subject: line of patches
> 
> - Please choose a suitable title, as per
>   Documentation/SubmittingPatches, section 15.
> 
> - Please cc suitable mailing lists and maintainers on bug reports and
>   on patches.

Dave Hansen was working on this code recently, and this looks a bit
similar to some changes he was making.  Dave, what's your opinion on
this patch?

Paul.

> From: Chandru <chandru@in.ibm.com>
> 
> When booted with crashkernel=224M@32M or any memory size less than this,
> the system boots properly.  The following was the observation..  The
> system comes up with two nodes (0-256M and 256M-4GB).  _The crashkernel
> memory reservation spans across these two nodes.  _The
> mark_reserved_regions_for_nid() in arch/powerpc/mm/numa.c resizes the
> reserved part of the memory within it as:
> 
> _ _ _ _ _ _ if (end_pfn > node_ar.end_pfn)
> _ _ _ _ _ _ _ _ reserve_size = (node_ar.end_pfn << PAGE_SHIFT)
> _ _ _ _ _ _ _ _ _ _ - (start_pfn << PAGE_SHIFT);
> 
> 
> but the reserve_bootmem_node() in mm/bootmem.c raises the pfn value of end 
> 
> _ _ end = PFN_UP(physaddr + size);
> 
> This causes end to get a value past the last page in the 0-256M node. 
> _Again when reserve_bootmem_node() returns,
> _mark_reserved_regions_for_nid() loops around to set the rest of the
> crashkernel memory in the next node as reserved.  _ It references
> NODE_DATA(node_ar.nid) and this causes another 'Oops: kernel access of bad
> area' problem.  The following changes made the system to boot with any
> amount of crashkernel memory size.
> 
> Signed-off-by: Chandru S <chandru@linux.vnet.ibm.com>
> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
> Cc: Paul Mackerras <paulus@samba.org>
> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
> ---
> 
>  arch/powerpc/mm/numa.c |    7 ++++---
>  mm/bootmem.c           |    4 ++++
>  2 files changed, 8 insertions(+), 3 deletions(-)
> 
> diff -puN arch/powerpc/mm/numa.c~2628-rc9-panics-with-crashkernel=256m-while-booting arch/powerpc/mm/numa.c
> --- a/arch/powerpc/mm/numa.c~2628-rc9-panics-with-crashkernel=256m-while-booting
> +++ a/arch/powerpc/mm/numa.c
> @@ -995,10 +995,11 @@ void __init do_init_bootmem(void)
>  				  start_pfn, end_pfn);
>  
>  		free_bootmem_with_active_regions(nid, end_pfn);
> +	}
> +
> +	for_each_online_node(nid) {
>  		/*
> -		 * Be very careful about moving this around.  Future
> -		 * calls to careful_allocation() depend on this getting
> -		 * done correctly.
> +		 * Be very careful about moving this around.
>  		 */
>  		mark_reserved_regions_for_nid(nid);
>  		sparse_memory_present_with_active_regions(nid);
> diff -puN mm/bootmem.c~2628-rc9-panics-with-crashkernel=256m-while-booting mm/bootmem.c
> --- a/mm/bootmem.c~2628-rc9-panics-with-crashkernel=256m-while-booting
> +++ a/mm/bootmem.c
> @@ -375,10 +375,14 @@ int __init reserve_bootmem_node(pg_data_
>  				 unsigned long size, int flags)
>  {
>  	unsigned long start, end;
> +	bootmem_data_t *bdata = pgdat->bdata;
>  
>  	start = PFN_DOWN(physaddr);
>  	end = PFN_UP(physaddr + size);
>  
> +	if (end > bdata->node_low_pfn)
> +		end = bdata->node_low_pfn;
> +
>  	return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
>  }
>  
> _
Dave Hansen - Dec. 29, 2008, 9:36 p.m.
On Fri, 2008-12-26 at 11:59 +1100, Paul Mackerras wrote:
> 
> > diff -puN arch/powerpc/mm/numa.c~2628-rc9-panics-with-crashkernel=256m-while-booting arch/powerpc/mm/numa.c
> > --- a/arch/powerpc/mm/numa.c~2628-rc9-panics-with-crashkernel=256m-while-booting
> > +++ a/arch/powerpc/mm/numa.c
> > @@ -995,10 +995,11 @@ void __init do_init_bootmem(void)
> >                                 start_pfn, end_pfn);
> >  
> >               free_bootmem_with_active_regions(nid, end_pfn);
> > +     }
> > +
> > +     for_each_online_node(nid) {
> >               /*
> > -              * Be very careful about moving this around.  Future
> > -              * calls to careful_allocation() depend on this getting
> > -              * done correctly.
> > +              * Be very careful about moving this around.
> >                */
> >               mark_reserved_regions_for_nid(nid);
> >               sparse_memory_present_with_active_regions(nid);

I think this reintroduces one of the bugs that I squashed.  You *have*
to call mark_reserved_regions_for_nid() right after you do
free_bootmem_with_active_regions().  Otherwise, someone else can
bootmem_alloc() a reserved region from that node.

Perhaps I need to make that comment a bit more forceful. :)

/*
 * Don't break this loop out.  Period.  Never.  Ever.
 * No, seriously.  Don't do it.  I mean it.  Really!
 */

-- Dave

Patch

diff -puN arch/powerpc/mm/numa.c~2628-rc9-panics-with-crashkernel=256m-while-booting arch/powerpc/mm/numa.c
--- a/arch/powerpc/mm/numa.c~2628-rc9-panics-with-crashkernel=256m-while-booting
+++ a/arch/powerpc/mm/numa.c
@@ -995,10 +995,11 @@  void __init do_init_bootmem(void)
 				  start_pfn, end_pfn);
 
 		free_bootmem_with_active_regions(nid, end_pfn);
+	}
+
+	for_each_online_node(nid) {
 		/*
-		 * Be very careful about moving this around.  Future
-		 * calls to careful_allocation() depend on this getting
-		 * done correctly.
+		 * Be very careful about moving this around.
 		 */
 		mark_reserved_regions_for_nid(nid);
 		sparse_memory_present_with_active_regions(nid);
diff -puN mm/bootmem.c~2628-rc9-panics-with-crashkernel=256m-while-booting mm/bootmem.c
--- a/mm/bootmem.c~2628-rc9-panics-with-crashkernel=256m-while-booting
+++ a/mm/bootmem.c
@@ -375,10 +375,14 @@  int __init reserve_bootmem_node(pg_data_
 				 unsigned long size, int flags)
 {
 	unsigned long start, end;
+	bootmem_data_t *bdata = pgdat->bdata;
 
 	start = PFN_DOWN(physaddr);
 	end = PFN_UP(physaddr + size);
 
+	if (end > bdata->node_low_pfn)
+		end = bdata->node_low_pfn;
+
 	return mark_bootmem_node(pgdat->bdata, start, end, 1, flags);
 }