[PATCH] x86/efi: Only pass mapped RAM regions to free_bootmem_late()

[PATCH] x86/efi: Only pass mapped RAM regions to free_bootmem_late()

[Matt Fleming]

From: Matt Fleming <matt.fleming@intel.com>

free_bootmem_late() expects to only be passed RAM regions that the
kernel can access, and that have a corresponding 'struct page'. It's
possible for regions in the EFI memory map to reside in address ranges
for which pfn_to_page() doesn't work, for instance when running on an
i386 system with CONFIG_HIGHMEM=n.

This is in fact the case for one of Alan's machines where some of the
EFI boot services pages live in highmem, and running a kernel without
CONFIG_HIGHMEM enabled results in the following oops,

 BUG: unable to handle kernel paging request at f7f1f080
 IP: [<c17fba96>] __free_pages_bootmem+0x5a/0xb8
 *pdpt = 0000000001887001 *pde = 0000000001984067 *pte = 000000000 0000000
 Oops: 0002 [#1] SMP

[...]

 Call Trace:
  [<c17feacc>] free_bootmem_late+0x2d/0x3d
  [<c17f1013>] efi_free_boot_services+0x48/0x5b
  [<c17ddc12>] start_kernel+0x3ad/0x3cf
  [<c17dd654>] ? set_init_arg+0x49/0x49
  [<c17dd380>] i386_start_kernel+0x12e/0x131

Reported-by: Alan Cox <alan@lxorguk.ukuu.org.uk>
Cc: Matthew Garrett <mjg59@srcf.ucam.org>
Cc: Dave Hansen <dave.hansen@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Signed-off-by: Matt Fleming <matt.fleming@intel.com>
---

Alan, could you double check that this patch also fixes the oops on your
CONFIG_HIGHMEM=n machine?

 arch/x86/platform/efi/efi.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/arch/x86/platform/efi/efi.c b/arch/x86/platform/efi/efi.c
index 3781dd39e8bd..893d183ee98f 100644
--- a/arch/x86/platform/efi/efi.c
+++ b/arch/x86/platform/efi/efi.c
@@ -451,6 +451,9 @@ void __init efi_free_boot_services(void)
 		if (!size)
 			continue;
 
+		if (!e820_all_mapped(start, start + size, E820_RAM))
+			continue;
+
 		free_bootmem_late(start, size);
 	}
 
-- 
1.9.0

Re: [PATCH] x86/efi: Only pass mapped RAM regions to free_bootmem_late()

[Dave Hansen]

On 06/05/2014 06:27 AM, Matt Fleming wrote:
> free_bootmem_late() expects to only be passed RAM regions that the
> kernel can access, and that have a corresponding 'struct page'. It's
> possible for regions in the EFI memory map to reside in address ranges
> for which pfn_to_page() doesn't work, for instance when running on an
> i386 system with CONFIG_HIGHMEM=n.

FWIW, I think there are also a bunch of ways to hit this if someone uses
mem=, even on 64-bit.

Re: [PATCH] x86/efi: Only pass mapped RAM regions to free_bootmem_late()

[Alan Cox]

On Thu,  5 Jun 2014 14:27:34 +0100
Matt Fleming <matt@console-pimps.org> wrote:

> From: Matt Fleming <matt.fleming@intel.com>
> 
> free_bootmem_late() expects to only be passed RAM regions that the
> kernel can access, and that have a corresponding 'struct page'. It's
> possible for regions in the EFI memory map to reside in address ranges
> for which pfn_to_page() doesn't work, for instance when running on an
> i386 system with CONFIG_HIGHMEM=n.
> 
> This is in fact the case for one of Alan's machines where some of the
> EFI boot services pages live in highmem, and running a kernel without
> CONFIG_HIGHMEM enabled results in the following oops,
> 
>  BUG: unable to handle kernel paging request at f7f1f080
>  IP: [<c17fba96>] __free_pages_bootmem+0x5a/0xb8
>  *pdpt = 0000000001887001 *pde = 0000000001984067 *pte = 000000000 0000000
>  Oops: 0002 [#1] SMP

No improvement, and the if case is never true 8(

Re: [PATCH] x86/efi: Only pass mapped RAM regions to free_bootmem_late()

[Matt Fleming]

On 5 June 2014 16:01, Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:
> On Thu,  5 Jun 2014 14:27:34 +0100
> Matt Fleming <matt@console-pimps.org> wrote:
>
>> From: Matt Fleming <matt.fleming@intel.com>
>>
>> free_bootmem_late() expects to only be passed RAM regions that the
>> kernel can access, and that have a corresponding 'struct page'. It's
>> possible for regions in the EFI memory map to reside in address ranges
>> for which pfn_to_page() doesn't work, for instance when running on an
>> i386 system with CONFIG_HIGHMEM=n.
>>
>> This is in fact the case for one of Alan's machines where some of the
>> EFI boot services pages live in highmem, and running a kernel without
>> CONFIG_HIGHMEM enabled results in the following oops,
>>
>>  BUG: unable to handle kernel paging request at f7f1f080
>>  IP: [<c17fba96>] __free_pages_bootmem+0x5a/0xb8
>>  *pdpt = 0000000001887001 *pde = 0000000001984067 *pte = 000000000 0000000
>>  Oops: 0002 [#1] SMP
>
> No improvement, and the if case is never true 8(

Argh, dammit. OK, back to the drawing board. Thanks for testing.