by Yinghai Lu

[permalink] [raw]

Subject: Re: [PATCH] x86: only clear node_states for 64bit

Yinghai Lu wrote:
> Ingo Molnar wrote:
>> * Yinghai Lu <[email protected]> wrote:
>>
>>> Andrew Morton wrote:
>>>> On Mon, 22 Jun 2009 08:38:50 -0700
>>>> Yinghai Lu <[email protected]> wrote:
>>>>
>>>>> Nathan reported that
>>>>> | commit 73d60b7f747176dbdff826c4127d22e1fd3f9f74
>>>>> | Author: Yinghai Lu <[email protected]>
>>>>> | Date: Tue Jun 16 15:33:00 2009 -0700
>>>>> |
>>>>> | page-allocator: clear N_HIGH_MEMORY map before we set it again
>>>>> |
>>>>> | SRAT tables may contains nodes of very small size. The arch code may
>>>>> | decide to not activate such a node. However, currently the early boot
>>>>> | code sets N_HIGH_MEMORY for such nodes. These nodes therefore seem to be
>>>>> | active although these nodes have no present pages.
>>>>> |
>>>>> | For 64bit N_HIGH_MEMORY == N_NORMAL_MEMORY, so that works for 64 bit too
>>>>>
>>>>> the cpuset.mems cgroup attribute on an i386 kvm guest
>>>>>
>>>>> fix it by only clearing node_states[N_NORMAL_MEMORY] for 64bit only.
>>>>> and need to do save/restore for that in find_zone_movable_pfn
>>>>>
>>>> There appear to be some words omitted from this changelog - it doesn't
>>>> make sense.
>>>>
>>>> I think that perhaps a line got deleted before "the cpuset.mems cgroup
>>>> ...". That was the line which actualy describes the bug which we're
>>>> fixing. Or perhaps it was a single word? "zeroes".
>>>>
>>>>
>>>> I did this:
>>>>
>>>> Nathan reported that
>>>> :
>>>> : | commit 73d60b7f747176dbdff826c4127d22e1fd3f9f74
>>>> : | Author: Yinghai Lu <[email protected]>
>>>> : | Date: Tue Jun 16 15:33:00 2009 -0700
>>>> : |
>>>> : | page-allocator: clear N_HIGH_MEMORY map before we set it again
>>>> : |
>>>> : | SRAT tables may contains nodes of very small size. The arch code may
>>>> : | decide to not activate such a node. However, currently the early boot
>>>> : | code sets N_HIGH_MEMORY for such nodes. These nodes therefore seem to be
>>>> : | active although these nodes have no present pages.
>>>> : |
>>>> : | For 64bit N_HIGH_MEMORY == N_NORMAL_MEMORY, so that works for 64 bit too
>>>> :
>>> "
>>>> : unintentionally and incorrectly clears the cpuset.mems cgroup attribute on
>>>> : an i386 kvm guest
>>> "
>>> ==>
>>>
>>> 32bit assume NORMAL_MEMORY bit and HIGH_MEMORY bit are set for
>>> Node0 always.
>> Where in the code is this assumption?
>
> in mm/page_alloc.c
> /*
> * Array of node states.
> */
> nodemask_t node_states[NR_NODE_STATES] __read_mostly = {
> [N_POSSIBLE] = NODE_MASK_ALL,
> [N_ONLINE] = { { [0] = 1UL } },
> #ifndef CONFIG_NUMA
> [N_NORMAL_MEMORY] = { { [0] = 1UL } },
> #ifdef CONFIG_HIGHMEM
> [N_HIGH_MEMORY] = { { [0] = 1UL } },
> #endif
> [N_CPU] = { { [0] = 1UL } },
> #endif /* NUMA */
> };
> EXPORT_SYMBOL(node_states);
>
> for x86 64bit, we clear POSSIBLE and ONLINE in arch/x86/mm/numa_64.c::initmem_init
> and this patch clear NORMAL in arch/x86/mm/init_64.c::paging_init
>
> for x86 32bit: ONLINE get cleared in get_memcfg_from_srat()
> and NORMAL and HIGH_MEMORY are not cleared
> before try to set new in mm/page_alloc.c::free_area_init_nodes
>
>>> and some code only check if HIGH_MEMORY is there to know if
>>> NORMAL_MEMORY is there.
>> Which code is that exactly?
>>
> with grep:
> arch/x86/mm/init_64.c: nodes_clear(node_states[N_NORMAL_MEMORY]);
> drivers/base/node.c: return print_nodes_state(N_NORMAL_MEMORY, buf);
> include/linux/nodemask.h: N_NORMAL_MEMORY, /* The node has regular memory */
> include/linux/nodemask.h: N_HIGH_MEMORY = N_NORMAL_MEMORY,
> mm/memcontrol.c: if (!node_state(node, N_NORMAL_MEMORY))
> mm/page_alloc.c: [N_NORMAL_MEMORY] = { { [0] = 1UL } },
> mm/page_alloc.c: node_set_state(zone_to_nid(zone), N_NORMAL_MEMORY);
> mm/slub.c: for_each_node_state(node, N_NORMAL_MEMORY) {
> mm/slub.c: for_each_node_state(node, N_NORMAL_MEMORY) {
> mm/slub.c: for_each_node_state(node, N_NORMAL_MEMORY) {
> mm/slub.c: for_each_node_state(node, N_NORMAL_MEMORY) {
> mm/slub.c: for_each_node_state(node, N_NORMAL_MEMORY) {
> mm/slub.c: for_each_node_state(node, N_NORMAL_MEMORY) {
> mm/slub.c: for_each_node_state(node, N_NORMAL_MEMORY) {
> mm/slub.c: for_each_node_state(node, N_NORMAL_MEMORY) {
> mm/slub.c: for_each_node_state(node, N_NORMAL_MEMORY)
>
> Documentation/cgroups/cpusets.txt:automatically tracks the value of node_states[N_HIGH_MEMORY]--i.e.,
> Documentation/memory-hotplug.txt:status_change_nid is set node id when N_HIGH_MEMORY of nodemask is (will be)
> arch/ia64/kernel/uncached.c: if (!node_state(nid, N_HIGH_MEMORY))
> drivers/base/node.c: return print_nodes_state(N_HIGH_MEMORY, buf);
> include/linux/cpuset.h:#define cpuset_current_mems_allowed (node_states[N_HIGH_MEMORY])
> include/linux/nodemask.h: N_HIGH_MEMORY, /* The node has regular or high memory */
> include/linux/nodemask.h: N_HIGH_MEMORY = N_NORMAL_MEMORY,
> kernel/cpuset.c: * found any online mems, return node_states[N_HIGH_MEMORY].
> kernel/cpuset.c: * of node_states[N_HIGH_MEMORY].
> kernel/cpuset.c: node_states[N_HIGH_MEMORY]))
> kernel/cpuset.c: node_states[N_HIGH_MEMORY]);
> kernel/cpuset.c: *pmask = node_states[N_HIGH_MEMORY];
> kernel/cpuset.c: BUG_ON(!nodes_intersects(*pmask, node_states[N_HIGH_MEMORY]));
> kernel/cpuset.c: * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY];
> kernel/cpuset.c: node_states[N_HIGH_MEMORY]))
> kernel/cpuset.c: nodes_subset(cp->mems_allowed, node_states[N_HIGH_MEMORY]))
> kernel/cpuset.c: node_states[N_HIGH_MEMORY]);
> kernel/cpuset.c: * Keep top_cpuset.mems_allowed tracking node_states[N_HIGH_MEMORY].
> kernel/cpuset.c: * Call this routine anytime after node_states[N_HIGH_MEMORY] changes.
> kernel/cpuset.c: top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
> kernel/cpuset.c: top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
> kernel/cpuset.c: * subset of node_states[N_HIGH_MEMORY], even if this means going outside the
> mm/memcontrol.c: for_each_node_state(node, N_HIGH_MEMORY) {
> mm/memory_hotplug.c: node_set_state(zone_to_nid(zone), N_HIGH_MEMORY);
> mm/mempolicy.c: if (!nodes_subset(new, node_states[N_HIGH_MEMORY])) {
> mm/mempolicy.c: for_each_node_state(nid, N_HIGH_MEMORY) {
> mm/mempolicy.c: if (!nodes_subset(nodes, node_states[N_HIGH_MEMORY]))
> mm/mempolicy.c: nodes = node_states[N_HIGH_MEMORY];
> mm/mempolicy.c: &node_states[N_HIGH_MEMORY], MPOL_MF_STATS, md);
> mm/mempolicy.c: for_each_node_state(n, N_HIGH_MEMORY)
> mm/migrate.c: if (!node_state(node, N_HIGH_MEMORY))
> mm/oom_kill.c: nodemask_t nodes = node_states[N_HIGH_MEMORY];
> mm/page-writeback.c: for_each_node_state(node, N_HIGH_MEMORY) {
> mm/page_alloc.c: [N_HIGH_MEMORY] = { { [0] = 1UL } },
> mm/page_alloc.c: * tasks mems_allowed, or node_states[N_HIGH_MEMORY].)
> mm/page_alloc.c: &node_states[N_HIGH_MEMORY];
> mm/page_alloc.c: for_each_node_state(n, N_HIGH_MEMORY) {
> mm/page_alloc.c: (nodes_weight(node_states[N_HIGH_MEMORY]) + 1);
> mm/page_alloc.c: * Populate N_HIGH_MEMORY for calculating usable_nodes.
> mm/page_alloc.c: node_set_state(early_node_map[i].nid, N_HIGH_MEMORY);
> mm/page_alloc.c: nodemask_t saved_node_state = node_states[N_HIGH_MEMORY];
> mm/page_alloc.c: int usable_nodes = nodes_weight(node_states[N_HIGH_MEMORY]);
> mm/page_alloc.c: for_each_node_state(nid, N_HIGH_MEMORY) {
> mm/page_alloc.c: node_states[N_HIGH_MEMORY] = saved_node_state;
> mm/page_alloc.c: node_set_state(nid, N_HIGH_MEMORY);
> mm/vmalloc.c: for_each_node_state(nr, N_HIGH_MEMORY)
> mm/vmscan.c: for_each_node_state(nid, N_HIGH_MEMORY) {
> mm/vmscan.c: for_each_node_state(nid, N_HIGH_MEMORY)
> mm/vmstat.c: if (!node_state(pgdat->node_id, N_HIGH_MEMORY))
>
> for 64bit N_HIGH_MEMORY == NORMAL_MEMORY
>
> for 32bit, there are more reference to N_HIGH_MEMORY...
>

- Why is this patch good/desired?

fix the broken with cpuset.mems cgroup attribute on an i386 kvm guest

- What did prior code do, and why was that wrong?

clear node_states[N_HIGH_MEMORY] for 32 bit and 64bit.
actually we only need clear that for 64bit to make that right for some strange
case like small range in one node.

- What were the bad effects. (crash, right?)

cpuset.mems can not be used ...

- What does this patch do to achieve that good status?

fix the problem with cpuset.mems and only keep the clearing for x86 64bit.

YH