On Wed, Jun 20, 2012 at 5:07 PM, Andrew Morton
<akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> On Wed, 20 Jun 2012 17:21:53 +0800
> Jiang Liu <jiang.liu@xxxxxxxxxx> wrote:
>
>> Function kswapd_stop() will be called to destroy the kswapd work thread
>> when all memory of a NUMA node has been offlined. But kswapd_stop() only
>> terminates the work thread without resetting NODE_DATA(nid)->kswapd to NULL.
>> The stale pointer will prevent kswapd_run() from creating a new work thread
>> when adding memory to the memory-less NUMA node again. Eventually the stale
>> pointer may cause invalid memory access.
>
> whoops.
>
>>
>> ...
>>
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -2961,8 +2961,10 @@ void kswapd_stop(int nid)
>> {
>> struct task_struct *kswapd = NODE_DATA(nid)->kswapd;
>>
>> - if (kswapd)
>> + if (kswapd) {
>> kthread_stop(kswapd);
>> + NODE_DATA(nid)->kswapd = NULL;
>> + }
>> }
>>
>> static int __init kswapd_init(void)
>
> OK.
>
> This function is full of races (ones which we'll never hit ;)) unless
> the caller provides locking. It appears that lock_memory_hotplug() is
> the locking, so I propose this addition:
>
> --- a/mm/vmscan.c~memory-hotplug-fix-invalid-memory-access-caused-by-stale-kswapd-pointer-fix
> +++ a/mm/vmscan.c
> @@ -2955,7 +2955,8 @@ int kswapd_run(int nid)
> }
>
> /*
> - * Called by memory hotplug when all memory in a node is offlined.
> + * Called by memory hotplug when all memory in a node is offlined. Caller must
> + * hold lock_memory_hotplug().
> */
> void kswapd_stop(int nid)
> {
> --- a/include/linux/mmzone.h~memory-hotplug-fix-invalid-memory-access-caused-by-stale-kswapd-pointer-fix
> +++ a/include/linux/mmzone.h
> @@ -693,7 +693,7 @@ typedef struct pglist_data {
> range, including holes */
> int node_id;
> wait_queue_head_t kswapd_wait;
> - struct task_struct *kswapd;
> + struct task_struct *kswapd; /* Protected by lock_memory_hotplug() */
except
"system_state == SYSTEM_BOOTING"?
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href