Re: [PATCH 16/35] x86: make 64 bit use early_res instead of bootmembefore slab

[Tejun Heo]

Hello, Yinghai.

On 02/25/2010 08:32 AM, Yinghai Lu wrote:
>>> Kernel panic - not syncing: free_early on not reserved area: 2f80000-2f9ffff!
>>> Pid: 0, comm: swapper Not tainted 2.6.33-rc8-tip-00918-gb34d361 #159
>>> Call Trace:	
>>>  [<ffffffff815d12e2>] panic+0x84/0x147
>>>  [<ffffffff81cceecb>] free_early+0x64/0x6d
>>>  [<ffffffff81cd2231>] free_bootmem+0xc/0xe
>>>  [<ffffffff81cc1cd6>] pcpu_fc_free+0x1d/0x1f
>>>  [<ffffffff81cd6056>] pcpu_embed_first_chunk+0x13a/0x27f
>>>  [<ffffffff81cc1cd8>] ? pcpu_fc_alloc+0x0/0xac
>>>  [<ffffffff81cc1cb9>] ? pcpu_fc_free+0x0/0x1f
>>>  [<ffffffff81cc1af7>] setup_per_cpu_areas+0x82/0x239
>>>  [<ffffffff81cb5b3c>] start_kernel+0x1b8/0x43b
>>>  [<ffffffff81cb52bc>] x86_64_start_reservations+0xa7/0xab
>>>  [<ffffffff81cb53b8>] x86_64_start_kernel+0xf8/0x107
>>>
>>
>> we can not handle partial free, and current only user for that is pcpu_setup...
> 
> any plan to pcpu_setup to use slab allocation instead of bootmem?
> that code is only one user that do partial bootmem free.

The problem is that the pcpu_embed_first_chunk() allocator depends on
having a large contiguous slab of memory to overlay the percpu
variables and then giving back the unused parts (due to layout and
cpu_possible_mask not being contiguous).  There's no way to do partial
alloc/free anymore?

Thanks.

-- 
tejun
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/