Re: PATCH] mm: Use alloc_bootmem_node_nopanic on really needed path

[Yinghai Lu <yinghai@kernel.org>]

On 05/06/2011 03:39 PM, Andrew Morton wrote:
> On Fri, 06 May 2011 15:29:41 -0700
> Yinghai Lu <yinghai@kernel.org> wrote:
> 
>> Stefan found nobootmem does not work on his system that only have 8M ram.
> 
> What does "does not work" mean?  From the patch, it appears that the
> machine paniced?

yes, early panic.

Linux version 2.6.36 (stefan@hel-stefan.lan) (gcc version 4.6.0 20110428
> (Red Hat 4.6.0-6) (GCC) ) #5 Thu May 5 21:49:09 CEST 2011
> BIOS-provided physical RAM map:
>  BIOS-88: 0000000000000000 - 000000000009f000 (usable)
>  BIOS-88: 0000000000100000 - 0000000000840000 (usable)
> bootconsole [earlyser0] enabled
> Notice: NX (Execute Disable) protection missing in CPU or disabled in BIOS!
> DMI not present or invalid.
> last_pfn = 0x840 max_arch_pfn = 0x100000
> init_memory_mapping: 0000000000000000-0000000000840000
> 8MB LOWMEM available.
>   mapped low ram: 0 - 00840000
>   low ram: 0 - 00840000
> Zone PFN ranges:
>   DMA      0x00000001 -> 0x00001000
>   Normal   empty
> Movable zone start PFN for each node
> early_node_map[2] active PFN ranges
>     0: 0x00000001 -> 0x0000009f
>     0: 0x00000100 -> 0x00000840
> BUG: Int 6: CR2 (null)
>      EDI c034663c  ESI (null)  EBP c0329f38  ESP c0329ef4
>      EBX c0346380  EDX 00000006  ECX ffffffff  EAX fffffff4
>      err (null)  EIP c0353191   CS c0320060  flg 00010082
> Stack: (null) c030c533 000007cd (null) c030c533 00000001 (null) (null)
>        00000003 0000083f 00000018 00000002 00000002 c0329f6c c03534d6 (null)
>        (null) 00000100 00000840 (null) c0329f64 00000001 00001000 (null)
> Pid: 0, comm: swapper Not tainted 2.6.36 #5
> Call Trace:
>  [<c02e3707>] ? 0xc02e3707
>  [<c035e6e5>] 0xc035e6e5
>  [<c0353191>] ? 0xc0353191
>  [<c03534d6>] 0xc03534d6
>  [<c034f1cd>] 0xc034f1cd
>  [<c034a824>] 0xc034a824
>  [<c03513cb>] ? 0xc03513cb
>  [<c0349432>] 0xc0349432
>  [<c0349066>] 0xc0349066


> 
> This is important because without a recognisable description of the
> failure it is hard for people to determine whether your patch might fix
> a problem which they are observing.
> 
> The patch applies to 2.6.36, 37 and 38 (at least).  Do you think it
> should be backported?  If so, how far back in time is it applicable?

Since we introduced CONFIG_NO_BOOTMEM... 
like we need it from 2.6.34.

Thanks

Yinghai