Re: [Patch] x86 : set numa_nodes_parsed at acpi_numa_memory_affinity_init

[Jiang Liu <liuj97@gmail.com>]

On 05/31/2012 07:25 PM, KOSAKI Motohiro wrote:
> Hi Ishimatsu-san,
> 
> 
> (5/29/12 3:45 AM), Yasuaki Ishimatsu wrote:
>> When hot-adding a CPU into my system, the system showed following message
>> since node_to_cpumask_map[2] was not allocated memory.
>>
>> Booting Node 2 Processor 32 APIC 0xc0
>> node_to_cpumask_map[2] NULL
>> Pid: 0, comm: swapper/32 Tainted: G       A     3.3.5-acd #21
>> Call Trace:
>>   [<ffffffff81048845>] debug_cpumask_set_cpu+0x155/0x160
>>   [<ffffffff8105e28a>] ? add_timer_on+0xaa/0x120
>>   [<ffffffff8150665f>] numa_add_cpu+0x1e/0x22
>>   [<ffffffff815020bb>] identify_cpu+0x1df/0x1e4
>>   [<ffffffff815020d6>] identify_econdary_cpu+0x16/0x1d
>>   [<ffffffff81504614>] smp_store_cpu_info+0x3c/0x3e
>>   [<ffffffff81505263>] smp_callin+0x139/0x1be
>>   [<ffffffff815052fb>] start_secondary+0x13/0xeb
>>
>> In my investigation, the root cause is that the bit of node 2 was not set at
>> numa_nodes_parsed since hot-added cpu was not written in ACPI SRAT Table.
>> But hot-added memory which is same PXM as the cpu is written in here.
> 
> I don't understand this explanation. If cpu is not written in srat table,
> it is firmware bug. isn't it? srat table should describe all possible cpus.
> Why doesn't your system describe cpu info in acpi table?
> 
>>
>> ACPI Spec Rev 5.0 says about ACPI SRAT talbe as follows:
>> This optional table provides information that allows OSPM to associate
>> processors and memory ranges, including ranges of memory provided by
>> hot-added memory devices, with system localities / proximity domains
>> and clock domains.
>>
>> So I think the ACPI SRAT Table is not wrong. Thus numa_nodes_parsed should be set
>> by not only acpi_numa_processor_affinity_init / acpi_numa_x2apic_affinity_init
>> but also acpi_numa_memory_affinity_init for the case.
>>
>> Signed-off-by: Yasuaki Ishimatsu<isimatu.yasuaki@jp.fujitsu.com>
>> ---
>>   arch/x86/mm/srat.c |    2 ++
>>   1 file changed, 2 insertions(+)
>>
>> Index: linux-3.4/arch/x86/mm/srat.c
>> ===================================================================
>> --- linux-3.4.orig/arch/x86/mm/srat.c	2012-05-21 07:29:13.000000000 +0900
>> +++ linux-3.4/arch/x86/mm/srat.c	2012-05-29 05:28:27.353916902 +0900
>> @@ -176,6 +176,8 @@  acpi_numa_memory_affinity_init(struct ac
>>   		return;
>>   	}
>>
>> +	node_set(node, numa_nodes_parsed);
>> +
> 
> But...
> I think this patch itself is correct. If a system have cpuless memory node, 
> acpi_numa_processor_affinity_init / acpi_numa_x2apic_affinity_init can't
> find any cpu description. but we need memory node allocation.
SRAT table only provides information for CPUs present at boot time. 
For hot-added CPUs, the _PXM and _SLI methods provide the same information for them
as SRAT table.

> 
> 
>>   	printk(KERN_INFO "SRAT: Node %u PXM %u %Lx-%Lx\n", node, pxm,
>>   	       start, end);
>>   }
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>> Please read the FAQ at  http://www.tux.org/lkml/
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/