Re: Problems booting 3.0.3 kernel on Octeon CN58XX board

[Jason Kwon <jason.kwon@ericsson.com>]

On 08/19/2011 10:00 AM, Guenter Roeck wrote:
> On Fri, 2011-08-19 at 12:32 -0400, David Daney wrote:
>> On 08/18/2011 05:10 PM, Jason Kwon wrote:
>>> Attempting to boot a 3.0.3 kernel on a CN58XX board produced the
>>> following oops:
>>>
>>> CPU 4 Unable to handle kernel paging request at virtual address
>>> 0000000001c00000, epc == ffffffff811aa9f4, ra == ffffffff811aaa98
>>> Oops[#1]:
>>> Cpu 4
>>> $ 0 : 0000000000000000 0000000010008ce0 ffffffff821d2b80 0000000001c00000
>>> $ 4 : 0000000001c00038 000000000000017c 0000000000080000 0000000000080072
>>> $ 8 : 0000000000000008 0000000000000002 0000000000000003 a800000002284520
>>> $12 : 0000000000000002 ffffffff8186ee80 ffffffffffffff80 0000000000000030
>>> $16 : 0000000000080072 0000000000000001 0000000001bfa8f0 0000000001bfa928
>>> $20 : a800000003aff8f0 00000000000f0000 ffffffff8186ee80 ffffffff821d2a80
>>> $24 : 0000000000000001 0000000000000038
>>> $28 : a80000041fc48000 a80000041fc4bd90 fffffffffffffffc ffffffff811aaa98
>>> Hi : 0000000000000000
>>> Lo : 0000000000000000
>>> epc : ffffffff811aa9f4 setup_per_zone_wmarks+0x19c/0x2d8
>>> Not tainted
>>> ra : ffffffff811aaa98 setup_per_zone_wmarks+0x240/0x2d8
>>> Status: 10008ce2 KX SX UX KERNEL EXL
>>> Cause : 40808408
>>> BadVA : 0000000001c00000
>>> PrId : 000d0301 (Cavium Octeon+)
>>> Modules linked in:
>>> Process swapper (pid: 1, threadinfo=a80000041fc48000,
>>> task=a80000041fc44038, tls=0000000000000000)
>>> Stack : 0000000000000000 000000000006f75d ffffffff8186eec0 0000000000000001
>>> 0000000000000547 ffffffff81825598 ffffffff81a80000 ffffffff818b3e68
>>> ffffffff818a40ac 0000000000000000 ffffffff81a80000 0000000000000000
>>> 0000000000000000 0000000000000000 0000000000000000 ffffffff818a40f0
>>> ffffffff81a80000 ffffffff81100438 ffffffff818b4198 ffffffff818b3e68
>>> ffffffff818b46c8 0000000000000000 0000000000000000 ffffffff818721d0
>>> 0000000000000000 0000000000000000 0000000000000000 ffffffff81109bb0
>>> 0000000000000000 0000000000000000 0000000000000000 0000000000000000
>>> 0000000000000000 0000000000000000 0000000000000000 ffffffff818720f8
>>> 0000000000000000 0000000000000000 0000000000000000 0000000000000000
>>> ...
>>> Call Trace:
>>> [<ffffffff811aa9f4>] setup_per_zone_wmarks+0x19c/0x2d8
>>> [<ffffffff818a40f0>] init_per_zone_wmark_min+0x44/0xe0
>>> [<ffffffff81100438>] do_one_initcall+0x38/0x160
>>> [<ffffffff818721d0>] kernel_init+0xd8/0x178
>>> [<ffffffff81109bb0>] kernel_thread_helper+0x10/0x18
>>>
>> It appears to be related to use of physical memory above the 16GB
>> barrier.  You could try reducing the amount of memory allocated to the
>> kernel by passing 'mem=1700M' on the kernel command line.
>>
> Hi David,
>
> are you sure ?
>
> This is what I see with our own boards (not the reference design board):
>
> Works:
>
> Linux version 3.0.3-423-gfa07d39 (groeck@rbos-pc-13) (gcc version 4.4.1
> (Debian 4.4.1-1) ) #2 SMP PREEMPT Thu Aug 18 14:09:53 PDT 2011
> [ ... ]
> CPU revision is: 000d030b (Cavium Octeon+)
> Checking for the multiply/shift bug... no.
> Checking for the daddiu bug... no.
> Determined physical RAM map:
>   memory: 00000000001fa000 @ 000000000160b000 (usable)
>   memory: 000000000e400000 @ 0000000001900000 (usable)
>   memory: 00000000d0000000 @ 0000000020000000 (usable)
>   memory: 000000000ffff000 @ 00000000f0001000 (usable)
>   memory: 0000000010000000 @ 0000000410000000 (usable)
>
> Crashes:
>
> Linux version 3.0.3-423-gfa07d39 (groeck@rbos-pc-13) (gcc version 4.4.1
> (Debian 4.4.1-1) ) #2 SMP PREEMPT Thu Aug 18 14:09:53 PDT 2011
> [ ... ]
> CPU revision is: 000d0003 (Cavium Octeon)
> Checking for the multiply/shift bug... no.
> Checking for the daddiu bug... no.
> Determined physical RAM map:
>   memory: 00000000001fa000 @ 000000000160b000 (usable)
>   memory: 000000000e400000 @ 0000000001900000 (usable)
>   memory: 0000000060000000 @ 0000000020000000 (usable)
>   memory: 0000000010000000 @ 0000000410000000 (usable)
>
> The memory at 0000000410000000 is there for both CPUs, yet the crash is
> only seen on the board with CN38xx. From a SW perspective, only
> difference besides the CPU type is that the working board has more
> memory.
>
> Thanks,
> Guenter
>
>
Well, I can confirm that setting mem=1700m on my CN5860 board allowed it 
to boot, at least:

Linux version 3.0.3-Cavium-Octeon+ (jkwon@xc5-pc2) (gcc version 4.3.3 
(Cavium Networks Version: 2_0_0 build 95) ) #2 SMP Thu Aug 18 15:16:58 
PDT 2011
[ ... ]
CPU revision is: 000d0301 (Cavium Octeon+)
Checking for the multiply/shift bug... no.
Checking for the daddiu bug... no.
Determined physical RAM map:
  memory: 0000000000208000 @ 0000000001872000 (usable)
  memory: 000000000dc00000 @ 0000000002200000 (usable)
  memory: 000000005c800000 @ 0000000020000000 (usable)

I also tried the memory restriction on a CN3860 board that was also 
hitting the same oops, and it then hit a different problem, so the 
restriction did seem to work on both boards.

Jason