mm, memory_hotplug: handle unitialized numa node gracefully
Subsystems (5)
Michal Hocko
maintainercover mm, memory_hotplug: handle unitialized numa node gracefully
Hi,
the previous version of this patchset has been posted her:
http://lkml.kernel.org/r/[email protected]
I have ended up with 6 patches in the end. The core of the fix is patch 2 which also links existing bug reports. The high level goal is to have all possible numa nodes have their pgdat allocated and initialized so
for_each_possible_node(nid)
NODE_DATA(nid)
will never return a garbage. This has proven to be problem in several places when an offline numa node is used for an allocation just to realize that node_data and therefore allocation fallback zonelists are not initialized and such an allocation request blows up.
There were attempts to address that by checking node_online in several places including the page allocator. This patchset approaches the problem from a different perspective and instead of special casing, which just adds a runtime overhead, it allocates pglist_data for each possible node. This can add some memory overhead for platforms with high number of possible nodes if they do not contain any memory. This should be a rather rare configuration though.
How to test this? David has provided and excellent howto:
http://lkml.kernel.org/r/[email protected]
Patches 1 and 3-6 are mostly cleanups. The patchset has been reviewed by Rafael (thanks!) and the core fix tested by Rafael and Alexey (thanks to both). David has tested as per instructions above and hasn't found any fallouts in the memory hotplug scenarios.
Please consider for inclusion.
Thanks!
Diffstat:
arch/ia64/mm/discontig.c | 11 +--- include/linux/memory_hotplug.h | 118 ++++++++++++++++++++--------------------- mm/memcontrol.c | 14 +---- mm/memory_hotplug.c | 55 ++++--------------- mm/page_alloc.c | 70 +++++++++++++++++++++--- 5 files changed, 133 insertions(+), 135 deletions(-)
Michal Hocko
maintainerPATCH 1/6 mm, memory_hotplug: make arch_alloc_nodedata independent on CONFIG_MEMORY_HOTPLUG
From: Michal Hocko <[email protected]>
This is a preparatory patch and it doesn't introduce any functional change. It merely pulls out arch_alloc_nodedata (and co) outside of CONFIG_MEMORY_HOTPLUG because the following patch will need to call this from the generic MM code.
--- arch/ia64/mm/discontig.c | 2 - include/linux/memory_hotplug.h | 119 ++++++++++++++++----------------- 2 files changed, 59 insertions(+), 62 deletions(-)
arch/ia64/mm/discontig.c +0 -2
@@ -608,7 +608,6 @@ void __init paging_init(void) zero_page_memmap_ptr =virt_to_page (ia64_imva (empty_zero_page)); } -#ifdef CONFIG_MEMORY_HOTPLUG pg_data_t *arch_alloc_nodedata (int nid) {unsigned long size =compute_pernodesize (nid);
@@ -626,7 +625,6 @@ void arch_refresh_nodedata(int update_node, pg_data_t *update_pgdat) pgdat_list[update_node] = update_pgdat;scatter_node_data (); } -#endif #ifdef CONFIG_SPARSEMEM_VMEMMAP int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,
include/linux/memory_hotplug.h +59 -60
131 lines not on this page · raw message
Signature · 1 line
2.30.2
Michal Hocko
maintainerPATCH 2/6 mm: handle uninitialized numa nodes gracefully
In reply to Michal HockoFrom: Michal Hocko <[email protected]>
We have had several reports [1][2][3] that page allocator blows up when an allocation from a possible node is requested. The underlying reason is that NODE_DATA for the specific node is not allocated.
NUMA specific initialization is arch specific and it can vary a lot. E.g. x86 tries to initialize all nodes that have some cpu affinity (see init_cpu_to_node) but this can be insufficient because the node might be cpuless for example.
One way to address this problem would be to check for !node_online nodes when trying to get a zonelist and silently fall back to another node. That is unfortunately adding a branch into allocator hot path and it doesn't handle any other potential NODE_DATA users.
This patch takes a different approach (following a lead of [3]) and it pre allocates pgdat for all possible nodes in an arch indipendent code
- free_area_init. All uninitialized nodes are treated as memoryless nodes. node_state of the node is not changed because that would lead to other side effects - e.g. sysfs representation of such a node and from past discussions [4] it is known that some tools might have problems digesting that.
Newly allocated pgdat only gets a minimal initialization and the rest of the work is expected to be done by the memory hotplug - hotadd_new_pgdat (renamed to hotadd_init_pgdat).
generic_alloc_nodedata is changed to use the memblock allocator because neither page nor slab allocators are available at the stage when all pgdats are allocated. Hotplug doesn't allocate pgdat anymore so we can use the early boot allocator. The only arch specific implementation is ia64 and that is changed to use the early allocator as well.
[1] http://lkml.kernel.org/r/[email protected]
[2] http://lkml.kernel.org/r/[email protected]
[3] http://lkml.kernel.org/r/[email protected]
[4] http://lkml.kernel.org/r/[email protected]
--- arch/ia64/mm/discontig.c | 4 ++-- include/linux/memory_hotplug.h | 2 +- mm/memory_hotplug.c | 21 +++++++++------------ mm/page_alloc.c | 34 +++++++++++++++++++++++++++++++--- 4 files changed, 43 insertions(+), 18 deletions(-)
arch/ia64/mm/discontig.c +2 -2
@@ -608,11 +608,11 @@ void __init paging_init(void) zero_page_memmap_ptr =virt_to_page (ia64_imva (empty_zero_page)); } -pg_data_t *arch_alloc_nodedata (int nid) +pg_data_t *__init arch_alloc_nodedata (int nid) {unsigned long size =compute_pernodesize (nid); -return kzalloc (size,GFP_KERNEL ); +return memblock_alloc (size,SMP_CACHE_BYTES ); }void arch_free_nodedata (pg_data_t *pgdat)
include/linux/memory_hotplug.h +1 -1
@@ -44,7 +44,7 @@ extern void arch_refresh_nodedata(int nid, pg_data_t *pgdat);*/ #define generic_alloc_nodedata (nid) \ ({ \ - kzalloc(sizeof(pg_data_t), GFP_KERNEL); \ + memblock_alloc(sizeof(*pgdat), SMP_CACHE_BYTES); \ })/* * This definition is just for error path in node hotadd.
mm/memory_hotplug.c +9 -12
@@ -1162,19 +1162,21 @@ static void reset_node_present_pages(pg_data_t *pgdat) }/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ -static pg_data_t __ref *hotadd_new_pgdat (int nid) +static pg_data_t __ref *hotadd_init_pgdat (int nid) {struct pglist_data *pgdat; pgdat =NODE_DATA (nid); -if (!pgdat) { - pgdat =arch_alloc_nodedata (nid); -if (!pgdat) -return NULL ; +/* +* NODE_DATA is preallocated (free_area_init) but its internal +* state is not allocated completely. Add missing pieces. +* Completely offline nodes stay around and they just need +* reintialization. +*/ +if (pgdat->per_cpu_nodestats == &boot_nodestats) { pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); -arch_refresh_nodedata (nid, pgdat); }else {int cpu;/*
@@ -1193,8 +1195,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid) } } -/* we can use NODE_DATA(nid) from here */ - pgdat->node_id = nid; pgdat->node_start_pfn =0 ;/* init node's zones as empty zones, we don't have any present pages.*/
@@ -1246,7 +1246,7 @@ static int __try_online_node(int nid, bool set_node_online)if (node_online (nid))return 0 ; - pgdat =hotadd_new_pgdat (nid); + pgdat =hotadd_init_pgdat (nid);if (!pgdat) {pr_err ("Cannot online node %d due to NULL pgdat\n" , nid); ret = -ENOMEM ;
@@ -1445,9 +1445,6 @@ int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)return ret; error: -/* rollback pgdat allocation and others */ -if (new_node) -rollback_node_hotadd (nid);if (IS_ENABLED (CONFIG_ARCH_KEEP_MEMBLOCK ))memblock_remove (start, size); error_mem_hotplug_end:
mm/page_alloc.c +31 -3
@@ -6402,7 +6402,11 @@ static void __build_all_zonelists(void *data)if (self && !node_online (self->node_id)) {build_zonelists (self); }else { -for_each_online_node (nid) { +/* +* All possible nodes have pgdat preallocated +* free_area_init +*/ +for_each_node (nid) {pg_data_t *pgdat =NODE_DATA (nid);build_zonelists (pgdat);
@@ -8096,8 +8100,32 @@ void __init free_area_init(unsigned long *max_zone_pfn)/* Initialise every node */ mminit_verify_pageflags_layout ();setup_nr_node_ids (); -for_each_online_node (nid) { -pg_data_t *pgdat =NODE_DATA (nid); +for_each_node (nid) { +pg_data_t *pgdat; + +if (!node_online (nid)) { +pr_warn ("Node %d uninitialized by the platform. Please report with boot dmesg.\n" , nid); + +/* Allocator not initialized yet */ + pgdat =arch_alloc_nodedata (nid); +if (!pgdat) { +pr_err ("Cannot allocate %zuB for node %d.\n" , +sizeof (*pgdat), nid); +continue ; + } +arch_refresh_nodedata (nid, pgdat); +free_area_init_memoryless_node (nid); +/* +* not marking this node online because we do not want to +* confuse userspace by sysfs files/directories for node +* without any memory attached to it (see topology_init) +* The pgdat will get fully initialized when a memory is +* hotpluged into it by hotadd_init_pgdat +*/ +continue ; + } + + pgdat =NODE_DATA (nid);free_area_init_node (nid);/* Any memory on that node */
Signature · 1 line
2.30.2
Michal Hocko
maintainerPATCH 3/6 mm, memory_hotplug: drop arch_free_nodedata
In reply to Michal HockoFrom: Michal Hocko <[email protected]>
Prior to "mm: handle uninitialized numa nodes gracefully" memory hotplug used to allocate pgdat when memory has been added to a node (hotadd_init_pgdat) arch_free_nodedata has been only used in the failure path because once the pgdat is exported (to be visible by NODA_DATA(nid)) it cannot really be freed because there is no synchronization available for that.
pgdat is allocated for each possible nodes now so the memory hotplug doesn't need to do the ever use arch_free_nodedata so drop it.
This patch doesn't introduce any functional change.
--- arch/ia64/mm/discontig.c | 5 ----- include/linux/memory_hotplug.h | 3 --- mm/memory_hotplug.c | 10 ---------- 3 files changed, 18 deletions(-)
arch/ia64/mm/discontig.c +0 -5
@@ -615,11 +615,6 @@ pg_data_t * __init arch_alloc_nodedata(int nid)return memblock_alloc (size,SMP_CACHE_BYTES ); } -void arch_free_nodedata (pg_data_t *pgdat) -{ -kfree (pgdat); -} -void arch_refresh_nodedata (int update_node,pg_data_t *update_pgdat) { pgdat_list[update_node] = update_pgdat;
include/linux/memory_hotplug.h +0 -3
@@ -24,17 +24,14 @@ struct vmem_altmap;* node_data[nid] = kzalloc() works well. But it depends on the architecture. * * In general, generic_alloc_nodedata() is used. -* Now, arch_free_nodedata() is just defined for error path of node_hot_add. * */ extern pg_data_t *arch_alloc_nodedata (int nid); -extern void arch_free_nodedata (pg_data_t *pgdat);extern void arch_refresh_nodedata (int nid,pg_data_t *pgdat);#else /* CONFIG_HAVE_ARCH_NODEDATA_EXTENSION */ #define arch_alloc_nodedata (nid) generic_alloc_nodedata(nid) -#define arch_free_nodedata (pgdat) generic_free_nodedata(pgdat)#ifdef CONFIG_NUMA /*
mm/memory_hotplug.c +0 -10
@@ -1217,16 +1217,6 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid)return pgdat; } -static void rollback_node_hotadd (int nid) -{ -pg_data_t *pgdat =NODE_DATA (nid); - -arch_refresh_nodedata (nid,NULL ); -free_percpu (pgdat->per_cpu_nodestats); -arch_free_nodedata (pgdat); -} - -/* * __try_online_node - online a node if offlined * @nid: the node ID
Signature · 1 line
2.30.2
Michal Hocko
maintainerPATCH 4/6 mm, memory_hotplug: reorganize new pgdat initialization
In reply to Michal HockoFrom: Michal Hocko <[email protected]>
When a !node_online node is brought up it needs a hotplug specific initialization because the node could be either uninitialized yet or it could have been recycled after previous hotremove. hotadd_init_pgdat is responsible for that.
Internal pgdat state is initialized at two places currently
- hotadd_init_pgdat
- free_area_init_core_hotplug
There is no real clear cut what should go where but this patch's chosen to move the whole internal state initialization into free_area_init_core_hotplug. hotadd_init_pgdat is still responsible to pull all the parts together - most notably to initialize zonelists because those depend on the overall topology.
This patch doesn't introduce any functional change.
--- include/linux/memory_hotplug.h | 2 +- mm/memory_hotplug.c | 28 +++------------------------- mm/page_alloc.c | 25 +++++++++++++++++++++++-- 3 files changed, 27 insertions(+), 28 deletions(-)
include/linux/memory_hotplug.h +1 -1
@@ -319,7 +319,7 @@ extern void set_zone_contiguous(struct zone *zone);extern void clear_zone_contiguous (struct zone *zone);#ifdef CONFIG_MEMORY_HOTPLUG -extern void __ref free_area_init_core_hotplug (int nid); +extern void __ref free_area_init_core_hotplug (struct pglist_data *pgdat);extern int __add_memory (int nid,u64 start,u64 size,mhp_t mhp_flags);extern int add_memory (int nid,u64 start,u64 size,mhp_t mhp_flags); extern int add_memory_resource(int nid, struct resource *resource,
mm/memory_hotplug.c +3 -25
@@ -1166,39 +1166,16 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid) {struct pglist_data *pgdat; - pgdat =NODE_DATA (nid); -/* * NODE_DATA is preallocated (free_area_init) but its internal * state is not allocated completely. Add missing pieces. * Completely offline nodes stay around and they just need * reintialization. */ -if (pgdat->per_cpu_nodestats == &boot_nodestats) { - pgdat->per_cpu_nodestats = - alloc_percpu(struct per_cpu_nodestat); - }else { -int cpu; -/* -* Reset the nr_zones, order and highest_zoneidx before reuse. -* Note that kswapd will init kswapd_highest_zoneidx properly -* when it starts in the near future. -*/ - pgdat->nr_zones =0 ; - pgdat->kswapd_order =0 ; - pgdat->kswapd_highest_zoneidx =0 ; -for_each_online_cpu (cpu) { -struct per_cpu_nodestat *p; - - p =per_cpu_ptr (pgdat->per_cpu_nodestats, cpu); -memset (p,0 ,sizeof (*p)); - } - } - - pgdat->node_start_pfn =0 ; + pgdat =NODE_DATA (nid);/* init node's zones as empty zones, we don't have any present pages.*/ -free_area_init_core_hotplug (nid); +free_area_init_core_hotplug (pgdat);/* * The node we allocated has no zone fallback lists. For avoiding
@@ -1210,6 +1187,7 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid)* When memory is hot-added, all the memory is in offline state. So * clear all zones' present_pages because they will be updated in * online_pages() and offline_pages(). +* TODO: should be in free_area_init_core_hotplug? */ reset_node_managed_pages (pgdat);reset_node_present_pages (pgdat);
mm/page_alloc.c +23 -2
@@ -7506,12 +7506,33 @@ static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx,* NOTE: this function is only called during memory hotplug */ #ifdef CONFIG_MEMORY_HOTPLUG -void __ref free_area_init_core_hotplug (int nid) +void __ref free_area_init_core_hotplug (struct pglist_data *pgdat) { +int nid = pgdat->node_id;enum zone_type z; -pg_data_t *pgdat =NODE_DATA (nid); +int cpu;pgdat_init_internals (pgdat); + +if (pgdat->per_cpu_nodestats == &boot_nodestats) + pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); + +/* +* Reset the nr_zones, order and highest_zoneidx before reuse. +* Note that kswapd will init kswapd_highest_zoneidx properly +* when it starts in the near future. +*/ + pgdat->nr_zones =0 ; + pgdat->kswapd_order =0 ; + pgdat->kswapd_highest_zoneidx =0 ; + pgdat->node_start_pfn =0 ; +for_each_online_cpu (cpu) { +struct per_cpu_nodestat *p; + + p =per_cpu_ptr (pgdat->per_cpu_nodestats, cpu); +memset (p,0 ,sizeof (*p)); + } +for (z =0 ; z <MAX_NR_ZONES ; z++)zone_init_internals (&pgdat->node_zones[z], z, nid,0 ); }
Signature · 1 line
2.30.2
Michal Hocko
maintainerPATCH 5/6 mm: make free_area_init_node aware of memory less nodes
In reply to Michal HockoFrom: Michal Hocko <[email protected]>
free_area_init_node is also called from memory less node initialization path (free_area_init_memoryless_node). It doesn't really make much sense to display the physical memory range for those nodes:
Initmem setup node XX [mem 0x0000000000000000-0x0000000000000000]
Instead be explicit that the node is memoryless:
Initmem setup node XX as memoryless
--- mm/page_alloc.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)
mm/page_alloc.c +8 -3
@@ -7682,9 +7682,14 @@ static void __init free_area_init_node(int nid) pgdat->node_start_pfn = start_pfn; pgdat->per_cpu_nodestats =NULL ; -pr_info ("Initmem setup node %d [mem %#018Lx-%#018Lx]\n" , nid, - (u64 )start_pfn <<PAGE_SHIFT , - end_pfn ? ((u64 )end_pfn <<PAGE_SHIFT ) -1 :0 ); +if (start_pfn != end_pfn) { +pr_info ("Initmem setup node %d [mem %#018Lx-%#018Lx]\n" , nid, + (u64 )start_pfn <<PAGE_SHIFT , + end_pfn ? ((u64 )end_pfn <<PAGE_SHIFT ) -1 :0 ); + }else { +pr_info ("Initmem setup node %d as memoryless\n" , nid); + } +calculate_node_totalpages (pgdat, start_pfn, end_pfn);alloc_node_mem_map (pgdat);
Signature · 1 line
2.30.2
Michal Hocko
maintainerPATCH 6/6 memcg: do not tweak node in alloc_mem_cgroup_per_node_info
In reply to Michal HockoFrom: Wei Yang <[email protected]>
alloc_mem_cgroup_per_node_info is allocated for each possible node and this used to be a problem because not !node_online nodes didn't have appropriate data structure allocated. This has changed by "mm: handle uninitialized numa nodes gracefully" so we can drop the special casing here.
--- mm/memcontrol.c | 14 ++------------ 1 file changed, 2 insertions(+), 12 deletions(-)
mm/memcontrol.c +2 -12
@@ -5067,18 +5067,8 @@ struct mem_cgroup *mem_cgroup_from_id(unsigned short id)static int alloc_mem_cgroup_per_node_info (struct mem_cgroup *memcg,int node) {struct mem_cgroup_per_node *pn; -int tmp = node; -/* -* This routine is called against possible nodes. -* But it's BUG to call kmalloc() against offline node. -* -* TODO: this routine can waste much memory for nodes which will -* never be onlined. It's better to use memory hotplug callback -* function. -*/ -if (!node_state (node,N_NORMAL_MEMORY )) - tmp =-1 ; - pn =kzalloc_node (sizeof (*pn),GFP_KERNEL , tmp); + + pn =kzalloc_node (sizeof (*pn),GFP_KERNEL , node);if (!pn)return 1 ;
Signature · 1 line
2.30.2
Re: [PATCH 1/6] mm, memory_hotplug: make arch_alloc_nodedata independent on CONFIG_MEMORY_HOTPLUG
In reply to Michal HockoQuoted 7 lines · Michal Hocko
From: Michal Hocko <[email protected]>
This is a preparatory patch and it doesn't introduce any functional change. It merely pulls out
arch_alloc_nodedata(and co) outside ofCONFIG_MEMORY_HOTPLUGbecause the following patch will need to call this from the generic MM code.
I would have hoped we'd have dropped support for ia64 by now, to get rid of this special case ...
Quoted 6 lines
Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>--- arch/ia64/mm/discontig.c | 2 - include/linux/memory_hotplug.h | 119 ++++++++++++++++----------------- 2 files changed, 59 insertions(+), 62 deletions(-)
Signature · 3 lines
Thanks, David / dhildenb
Re: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to Michal HockoQuoted 57 lines · Michal Hocko
From: Michal Hocko <[email protected]>
We have had several reports [1][2][3] that page allocator blows up when an allocation from a possible node is requested. The underlying reason is that
NODE_DATAfor the specific node is not allocated.NUMA specific initialization is arch specific and it can vary a lot. E.g. x86 tries to initialize all nodes that have some cpu affinity (see
init_cpu_to_node) but this can be insufficient because the node might be cpuless for example.One way to address this problem would be to check for !node_online nodes when trying to get a zonelist and silently fall back to another node. That is unfortunately adding a branch into allocator hot path and it doesn't handle any other potential
NODE_DATAusers.This patch takes a different approach (following a lead of [3]) and it pre allocates pgdat for all possible nodes in an arch indipendent code
-free_area_init. All uninitialized nodes are treated as memoryless nodes.node_stateof the node is not changed because that would lead to other side effects - e.g. sysfs representation of such a node and from past discussions [4] it is known that some tools might have problems digesting that.Newly allocated pgdat only gets a minimal initialization and the rest of the work is expected to be done by the memory hotplug -
hotadd_new_pgdat(renamed tohotadd_init_pgdat).
generic_alloc_nodedatais changed to use the memblock allocator because neither page nor slab allocators are available at the stage when all pgdats are allocated. Hotplug doesn't allocate pgdat anymore so we can use the early boot allocator. The only arch specific implementation is ia64 and that is changed to use the early allocator as well.Reported-by Alexey Makhalov <[email protected]>Tested-by Alexey Makhalov <[email protected]>Reported-by Nico Pache <[email protected]>Acked-by Rafael Aquini <[email protected]>Tested-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>[1] http://lkml.kernel.org/r/[email protected]
[2] http://lkml.kernel.org/r/[email protected]
[3] http://lkml.kernel.org/r/[email protected]
[4] http://lkml.kernel.org/r/[email protected]--- arch/ia64/mm/discontig.c | 4 ++-- include/linux/memory_hotplug.h | 2 +- mm/memory_hotplug.c | 21 +++++++++------------ mm/page_alloc.c | 34 +++++++++++++++++++++++++++++++--- 4 files changed, 43 insertions(+), 18 deletions(-)arch/ia64/mm/discontig.c +0
-0
arch/ia64/mm/discontig.c void __init paging_init(void) · line 615
Quoted 10 lines
arch/ia64/mm/discontig.c +2
-2@@ -608,11 +608,11 @@ void __init paging_init(void) zero_page_memmap_ptr =virt_to_page (ia64_imva (empty_zero_page)); } -pg_data_t *arch_alloc_nodedata (int nid) +pg_data_t *__init arch_alloc_nodedata (int nid) {unsigned long size =compute_pernodesize (nid); -return kzalloc (size,GFP_KERNEL ); +return memblock_alloc (size,SMP_CACHE_BYTES );
I feel like we should have
long arch_pgdat_size(void) instead and have a generic allocation function.
But we can clean that up in the future.
Quoted 67 lines
arch/ia64/mm/discontig.c +0
-0}void arch_free_nodedata (pg_data_t *pgdat)include/linux/memory_hotplug.h +1
-1@@ -44,7 +44,7 @@ extern void arch_refresh_nodedata(int nid, pg_data_t *pgdat);*/ #define generic_alloc_nodedata (nid) \ ({ \ - kzalloc(sizeof(pg_data_t), GFP_KERNEL); \ + memblock_alloc(sizeof(*pgdat), SMP_CACHE_BYTES); \ })/* * This definition is just for error path in node hotadd. mm/memory_hotplug.c +9
-9@@ -1162,19 +1162,21 @@ static void reset_node_present_pages(pg_data_t *pgdat) }/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ -static pg_data_t __ref *hotadd_new_pgdat (int nid) +static pg_data_t __ref *hotadd_init_pgdat (int nid) {struct pglist_data *pgdat; pgdat =NODE_DATA (nid); -if (!pgdat) { - pgdat =arch_alloc_nodedata (nid); -if (!pgdat) -return NULL ; +/* +* NODE_DATA is preallocated (free_area_init) but its internal +* state is not allocated completely. Add missing pieces. +* Completely offline nodes stay around and they just need +* reintialization. +*/ +if (pgdat->per_cpu_nodestats == &boot_nodestats) { pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); -arch_refresh_nodedata (nid, pgdat); }else {int cpu;/* @@ -1193,8 +1195,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid) } } -/* we can use NODE_DATA(nid) from here */ - pgdat->node_id = nid; pgdat->node_start_pfn =0 ;/* init node's zones as empty zones, we don't have any present pages.*/ @@ -1246,7 +1246,7 @@ static int __try_online_node(int nid, bool set_node_online)if (node_online (nid))return 0 ; - pgdat =hotadd_new_pgdat (nid); + pgdat =hotadd_init_pgdat (nid);if (!pgdat) {pr_err ("Cannot online node %d due to NULL pgdat\n" , nid); ret = -ENOMEM ;
mm/memory_hotplug.c int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags) · line 1450
Quoted 6 lines
mm/memory_hotplug.c +0
-3@@ -1445,9 +1445,6 @@ int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)return ret; error: -/* rollback pgdat allocation and others */ -if (new_node) -rollback_node_hotadd (nid);
As static rollback_node_hotadd() is unused in this patch, doesn't this trigger a warning? IOW, maybe merge at least the rollback_node_hotadd() removal into this patch. The arch_free_nodedata() removal can stay separate.
Quoted 8 lines
mm/memory_hotplug.c +0
-0if (IS_ENABLED (CONFIG_ARCH_KEEP_MEMBLOCK ))memblock_remove (start, size); error_mem_hotplug_end:mm/page_alloc.c +0
-0
mm/page_alloc.c static void __build_all_zonelists(void *data) · line 6406
Quoted 6 lines
mm/page_alloc.c +2
-1@@ -6402,7 +6402,11 @@ static void __build_all_zonelists(void *data)if (self && !node_online (self->node_id)) {build_zonelists (self); }else { -for_each_online_node (nid) { +/* +* All possible nodes have pgdat preallocated
... in free_area_init() ?
Quoted 16 lines
mm/page_alloc.c +7
-2+* free_area_init +*/ +for_each_node (nid) {pg_data_t *pgdat =NODE_DATA (nid);build_zonelists (pgdat);@@ -8096,8 +8100,32 @@ void __init free_area_init(unsigned long *max_zone_pfn)/* Initialise every node */ mminit_verify_pageflags_layout ();setup_nr_node_ids (); -for_each_online_node (nid) { -pg_data_t *pgdat =NODE_DATA (nid); +for_each_node (nid) { +pg_data_t *pgdat; + +if (!node_online (nid)) {
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8116
Quoted 10 lines
mm/page_alloc.c +10
-0+pr_warn ("Node %d uninitialized by the platform. Please report with boot dmesg.\n" , nid); + +/* Allocator not initialized yet */ + pgdat =arch_alloc_nodedata (nid); +if (!pgdat) { +pr_err ("Cannot allocate %zuB for node %d.\n" , +sizeof (*pgdat), nid); +continue ; + } +arch_refresh_nodedata (nid, pgdat);
We could get rid of arch_refresh_nodedata() now and simply merge that into arch_alloc_nodedata(). But depends on how we want to proceed with arch_alloc_nodedata() eventually.
Signature · 3 lines
Thanks, David / dhildenb
Re: [PATCH 3/6] mm, memory_hotplug: drop arch_free_nodedata
In reply to Michal HockoQuoted 66 lines · Michal Hocko
From: Michal Hocko <[email protected]>
Prior to "mm: handle uninitialized numa nodes gracefully" memory hotplug used to allocate pgdat when memory has been added to a node (
hotadd_init_pgdat)arch_free_nodedatahas been only used in the failure path because once the pgdat is exported (to be visible byNODA_DATA(nid)) it cannot really be freed because there is no synchronization available for that.pgdat is allocated for each possible nodes now so the memory hotplug doesn't need to do the ever use
arch_free_nodedataso drop it.This patch doesn't introduce any functional change.
Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>--- arch/ia64/mm/discontig.c | 5 ----- include/linux/memory_hotplug.h | 3 --- mm/memory_hotplug.c | 10 ---------- 3 files changed, 18 deletions(-)arch/ia64/mm/discontig.c +0
-5@@ -615,11 +615,6 @@ pg_data_t * __init arch_alloc_nodedata(int nid)return memblock_alloc (size,SMP_CACHE_BYTES ); } -void arch_free_nodedata (pg_data_t *pgdat) -{ -kfree (pgdat); -} -void arch_refresh_nodedata (int update_node,pg_data_t *update_pgdat) { pgdat_list[update_node] = update_pgdat;include/linux/memory_hotplug.h +0
-3@@ -24,17 +24,14 @@ struct vmem_altmap;* node_data[nid] = kzalloc() works well. But it depends on the architecture. * * In general, generic_alloc_nodedata() is used. -* Now, arch_free_nodedata() is just defined for error path of node_hot_add. * */ extern pg_data_t *arch_alloc_nodedata (int nid); -extern void arch_free_nodedata (pg_data_t *pgdat);extern void arch_refresh_nodedata (int nid,pg_data_t *pgdat);#else /* CONFIG_HAVE_ARCH_NODEDATA_EXTENSION */ #define arch_alloc_nodedata (nid) generic_alloc_nodedata(nid) -#define arch_free_nodedata (pgdat) generic_free_nodedata(pgdat)#ifdef CONFIG_NUMA /* mm/memory_hotplug.c +0
-0@@ -1217,16 +1217,6 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid)return pgdat;
mm/memory_hotplug.c static pg_data_t __ref *hotadd_init_pgdat(int nid) · line 1227
Quoted 10 lines
mm/memory_hotplug.c +0
-8} -static void rollback_node_hotadd (int nid) -{ -pg_data_t *pgdat =NODE_DATA (nid); - -arch_refresh_nodedata (nid,NULL ); -free_percpu (pgdat->per_cpu_nodestats); -arch_free_nodedata (pgdat); -}
As mentioned, maybe we want to rip out rollback_node_hotadd() in the previous patch.
Signature · 3 lines
Thanks, David / dhildenb
Re: [PATCH 4/6] mm, memory_hotplug: reorganize new pgdat initialization
In reply to Michal HockoQuoted 122 lines · Michal Hocko
From: Michal Hocko <[email protected]>
When a !node_online node is brought up it needs a hotplug specific initialization because the node could be either uninitialized yet or it could have been recycled after previous hotremove.
hotadd_init_pgdatis responsible for that.Internal pgdat state is initialized at two places currently
-hotadd_init_pgdat
-free_area_init_core_hotplug
There is no real clear cut what should go where but this patch's chosen to move the whole internal state initialization intofree_area_init_core_hotplug.hotadd_init_pgdatis still responsible to pull all the parts together - most notably to initialize zonelists because those depend on the overall topology.This patch doesn't introduce any functional change.
Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>--- include/linux/memory_hotplug.h | 2 +- mm/memory_hotplug.c | 28 +++------------------------- mm/page_alloc.c | 25 +++++++++++++++++++++++-- 3 files changed, 27 insertions(+), 28 deletions(-)include/linux/memory_hotplug.h +1
-1@@ -319,7 +319,7 @@ extern void set_zone_contiguous(struct zone *zone);extern void clear_zone_contiguous (struct zone *zone);#ifdef CONFIG_MEMORY_HOTPLUG -extern void __ref free_area_init_core_hotplug (int nid); +extern void __ref free_area_init_core_hotplug (struct pglist_data *pgdat);extern int __add_memory (int nid,u64 start,u64 size,mhp_t mhp_flags);extern int add_memory (int nid,u64 start,u64 size,mhp_t mhp_flags); extern int add_memory_resource(int nid, struct resource *resource,mm/memory_hotplug.c +3
-25@@ -1166,39 +1166,16 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid) {struct pglist_data *pgdat; - pgdat =NODE_DATA (nid); -/* * NODE_DATA is preallocated (free_area_init) but its internal * state is not allocated completely. Add missing pieces. * Completely offline nodes stay around and they just need * reintialization. */ -if (pgdat->per_cpu_nodestats == &boot_nodestats) { - pgdat->per_cpu_nodestats = - alloc_percpu(struct per_cpu_nodestat); - }else { -int cpu; -/* -* Reset the nr_zones, order and highest_zoneidx before reuse. -* Note that kswapd will init kswapd_highest_zoneidx properly -* when it starts in the near future. -*/ - pgdat->nr_zones =0 ; - pgdat->kswapd_order =0 ; - pgdat->kswapd_highest_zoneidx =0 ; -for_each_online_cpu (cpu) { -struct per_cpu_nodestat *p; - - p =per_cpu_ptr (pgdat->per_cpu_nodestats, cpu); -memset (p,0 ,sizeof (*p)); - } - } - - pgdat->node_start_pfn =0 ; + pgdat =NODE_DATA (nid);/* init node's zones as empty zones, we don't have any present pages.*/ -free_area_init_core_hotplug (nid); +free_area_init_core_hotplug (pgdat);/* * The node we allocated has no zone fallback lists. For avoiding @@ -1210,6 +1187,7 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid)* When memory is hot-added, all the memory is in offline state. So * clear all zones' present_pages because they will be updated in * online_pages() and offline_pages(). +* TODO: should be in free_area_init_core_hotplug? */ reset_node_managed_pages (pgdat);reset_node_present_pages (pgdat);mm/page_alloc.c +16
-2@@ -7506,12 +7506,33 @@ static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx,* NOTE: this function is only called during memory hotplug */ #ifdef CONFIG_MEMORY_HOTPLUG -void __ref free_area_init_core_hotplug (int nid) +void __ref free_area_init_core_hotplug (struct pglist_data *pgdat) { +int nid = pgdat->node_id;enum zone_type z; -pg_data_t *pgdat =NODE_DATA (nid); +int cpu;pgdat_init_internals (pgdat); + +if (pgdat->per_cpu_nodestats == &boot_nodestats) + pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); + +/* +* Reset the nr_zones, order and highest_zoneidx before reuse. +* Note that kswapd will init kswapd_highest_zoneidx properly +* when it starts in the near future. +*/ + pgdat->nr_zones =0 ; + pgdat->kswapd_order =0 ; + pgdat->kswapd_highest_zoneidx =0 ; + pgdat->node_start_pfn =0 ;
mm/page_alloc.c static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, · line 7538
Quoted 10 lines
mm/page_alloc.c +7
-0+for_each_online_cpu (cpu) { +struct per_cpu_nodestat *p; + + p =per_cpu_ptr (pgdat->per_cpu_nodestats, cpu); +memset (p,0 ,sizeof (*p)); + } +for (z =0 ; z <MAX_NR_ZONES ; z++)zone_init_internals (&pgdat->node_zones[z], z, nid,0 ); }
I feel like we should be initializing all of this only once, just after allocating the node. There should be no difference between a node we're reusing and a "fresh" node. IOW, memory offlining should be resetting all state accordingly when the last memory goes away.
But I might be wrong and this patch looks like an improvement, as you say, without functional change
Signature · 3 lines
Thanks, David / dhildenb
Re: [PATCH 5/6] mm: make free_area_init_node aware of memory less nodes
In reply to Michal HockoQuoted 28 lines · Michal Hocko
From: Michal Hocko <[email protected]>
free_area_init_nodeis also called from memory less node initialization path (free_area_init_memoryless_node). It doesn't really make much sense to display the physical memory range for those nodes:
Initmem setup node XX [mem 0x0000000000000000-0x0000000000000000]Instead be explicit that the node is memoryless:
Initmem setup node XX as memorylessAcked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>--- mm/page_alloc.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)mm/page_alloc.c +1
-3@@ -7682,9 +7682,14 @@ static void __init free_area_init_node(int nid) pgdat->node_start_pfn = start_pfn; pgdat->per_cpu_nodestats =NULL ; -pr_info ("Initmem setup node %d [mem %#018Lx-%#018Lx]\n" , nid, - (u64 )start_pfn <<PAGE_SHIFT , - end_pfn ? ((u64 )end_pfn <<PAGE_SHIFT ) -1 :0 ); +if (start_pfn != end_pfn) {
mm/page_alloc.c static void __init free_area_init_node(int nid) · line 7695
Quoted 10 lines
mm/page_alloc.c +7
-0+pr_info ("Initmem setup node %d [mem %#018Lx-%#018Lx]\n" , nid, + (u64 )start_pfn <<PAGE_SHIFT , + end_pfn ? ((u64 )end_pfn <<PAGE_SHIFT ) -1 :0 ); + }else { +pr_info ("Initmem setup node %d as memoryless\n" , nid); + } +calculate_node_totalpages (pgdat, start_pfn, end_pfn);alloc_node_mem_map (pgdat);
Signature · 3 lines
Thanks, David / dhildenb
Re: [PATCH 6/6] memcg: do not tweak node in alloc_mem_cgroup_per_node_info
In reply to Michal HockoQuoted 29 lines · Michal Hocko
From: Wei Yang <[email protected]>
alloc_mem_cgroup_per_node_infois allocated for each possible node and this used to be a problem because not !node_online nodes didn't have appropriate data structure allocated. This has changed by "mm: handle uninitialized numa nodes gracefully" so we can drop the special casing here.Signed-off-by Wei Yang <[email protected]>Signed-off-by Michal Hocko <[email protected]>--- mm/memcontrol.c | 14 ++------------ 1 file changed, 2 insertions(+), 12 deletions(-)mm/memcontrol.c +0
-7@@ -5067,18 +5067,8 @@ struct mem_cgroup *mem_cgroup_from_id(unsigned short id)static int alloc_mem_cgroup_per_node_info (struct mem_cgroup *memcg,int node) {struct mem_cgroup_per_node *pn; -int tmp = node; -/* -* This routine is called against possible nodes. -* But it's BUG to call kmalloc() against offline node. -* -* TODO: this routine can waste much memory for nodes which will -* never be onlined. It's better to use memory hotplug callback
mm/memcontrol.c struct mem_cgroup *mem_cgroup_from_id(unsigned short id) · line 5074
Quoted 10 lines
mm/memcontrol.c +2
-5-* function. -*/ -if (!node_state (node,N_NORMAL_MEMORY )) - tmp =-1 ; - pn =kzalloc_node (sizeof (*pn),GFP_KERNEL , tmp); + + pn =kzalloc_node (sizeof (*pn),GFP_KERNEL , node);if (!pn)return 1 ;
Signature · 3 lines
Thanks, David / dhildenb
Re: [PATCH 5/6] mm: make free_area_init_node aware of memory less nodes
In reply to Michal HockoQuoted 7 lines · Michal Hocko
From: Michal Hocko <[email protected]>
free_area_init_nodeis also called from memory less node initialization path (free_area_init_memoryless_node). It doesn't really make much sense to display the physical memory range for those nodes:
Initmem setup node XX [mem 0x0000000000000000-0x0000000000000000]
Quoted 5 lines
Instead be explicit that the node is memoryless:
Initmem setup node XX as memorylessAcked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Quoted 29 lines
--- mm/page_alloc.c | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)mm/page_alloc.c +8
-3@@ -7682,9 +7682,14 @@ static void __init free_area_init_node(int nid) pgdat->node_start_pfn = start_pfn; pgdat->per_cpu_nodestats =NULL ; -pr_info ("Initmem setup node %d [mem %#018Lx-%#018Lx]\n" , nid, - (u64 )start_pfn <<PAGE_SHIFT , - end_pfn ? ((u64 )end_pfn <<PAGE_SHIFT ) -1 :0 ); +if (start_pfn != end_pfn) { +pr_info ("Initmem setup node %d [mem %#018Lx-%#018Lx]\n" , nid, + (u64 )start_pfn <<PAGE_SHIFT , + end_pfn ? ((u64 )end_pfn <<PAGE_SHIFT ) -1 :0 ); + }else { +pr_info ("Initmem setup node %d as memoryless\n" , nid); + } +calculate_node_totalpages (pgdat, start_pfn, end_pfn);alloc_node_mem_map (pgdat);Signature · 1 line
2.30.2
Signature · 2 lines
Sincerely yours, Mike.
Re: [PATCH 1/6] mm, memory_hotplug: make arch_alloc_nodedata independent on CONFIG_MEMORY_HOTPLUG
In reply to Michal HockoQuoted 9 lines · Michal Hocko
From: Michal Hocko <[email protected]>
This is a preparatory patch and it doesn't introduce any functional change. It merely pulls out
arch_alloc_nodedata(and co) outside ofCONFIG_MEMORY_HOTPLUGbecause the following patch will need to call this from the generic MM code.Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Quoted 165 lines
--- arch/ia64/mm/discontig.c | 2 - include/linux/memory_hotplug.h | 119 ++++++++++++++++----------------- 2 files changed, 59 insertions(+), 62 deletions(-)arch/ia64/mm/discontig.c +0
-2@@ -608,7 +608,6 @@ void __init paging_init(void) zero_page_memmap_ptr =virt_to_page (ia64_imva (empty_zero_page)); } -#ifdef CONFIG_MEMORY_HOTPLUG pg_data_t *arch_alloc_nodedata (int nid) {unsigned long size =compute_pernodesize (nid);@@ -626,7 +625,6 @@ void arch_refresh_nodedata(int update_node, pg_data_t *update_pgdat) pgdat_list[update_node] = update_pgdat;scatter_node_data (); } -#endif #ifdef CONFIG_SPARSEMEM_VMEMMAP int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node,include/linux/memory_hotplug.h +59
-60@@ -16,6 +16,65 @@ struct memory_group;struct resource ;struct vmem_altmap ; +#ifdef CONFIG_HAVE_ARCH_NODEDATA_EXTENSION +/* +* For supporting node-hotadd, we have to allocate a new pgdat. +* +* If an arch has generic style NODE_DATA(), +* node_data[nid] = kzalloc() works well. But it depends on the architecture. +* +* In general, generic_alloc_nodedata() is used. +* Now, arch_free_nodedata() is just defined for error path of node_hot_add. +* +*/ +extern pg_data_t *arch_alloc_nodedata (int nid); +extern void arch_free_nodedata (pg_data_t *pgdat); +extern void arch_refresh_nodedata (int nid,pg_data_t *pgdat); + +#else /* CONFIG_HAVE_ARCH_NODEDATA_EXTENSION */ + +#define arch_alloc_nodedata (nid) generic_alloc_nodedata(nid) +#define arch_free_nodedata (pgdat) generic_free_nodedata(pgdat) + +#ifdef CONFIG_NUMA +/* +* XXX: node aware allocation can't work well to get new node's memory at this time. +* Because, pgdat for the new node is not allocated/initialized yet itself. +* To use new node's memory, more consideration will be necessary. +*/ +#define generic_alloc_nodedata (nid) \ +({ \ + kzalloc(sizeof(pg_data_t), GFP_KERNEL); \ +}) +/* +* This definition is just for error path in node hotadd. +* For node hotremove, we have to replace this. +*/ +#define generic_free_nodedata (pgdat) kfree(pgdat) + +extern pg_data_t *node_data[]; +static inline void arch_refresh_nodedata (int nid,pg_data_t *pgdat) +{ + node_data[nid] = pgdat; +} + +#else /* !CONFIG_NUMA */ + +/* never called */ +static inline pg_data_t *generic_alloc_nodedata (int nid) +{ +BUG (); +return NULL ; +} +static inline void generic_free_nodedata (pg_data_t *pgdat) +{ +} +static inline void arch_refresh_nodedata (int nid,pg_data_t *pgdat) +{ +} +#endif /* CONFIG_NUMA */ +#endif /* CONFIG_HAVE_ARCH_NODEDATA_EXTENSION */ +#ifdef CONFIG_MEMORY_HOTPLUG struct page *pfn_to_online_page (unsigned long pfn);@@ -154,66 +213,6 @@ int add_pages(int nid, unsigned long start_pfn, unsigned long nr_pages, struct mhp_params *params);#endif /* ARCH_HAS_ADD_PAGES */ -#ifdef CONFIG_HAVE_ARCH_NODEDATA_EXTENSION -/* -* For supporting node-hotadd, we have to allocate a new pgdat. -* -* If an arch has generic style NODE_DATA(), -* node_data[nid] = kzalloc() works well. But it depends on the architecture. -* -* In general, generic_alloc_nodedata() is used. -* Now, arch_free_nodedata() is just defined for error path of node_hot_add. -* -*/ -extern pg_data_t *arch_alloc_nodedata (int nid); -extern void arch_free_nodedata (pg_data_t *pgdat); -extern void arch_refresh_nodedata (int nid,pg_data_t *pgdat); - -#else /* CONFIG_HAVE_ARCH_NODEDATA_EXTENSION */ - -#define arch_alloc_nodedata (nid) generic_alloc_nodedata(nid) -#define arch_free_nodedata (pgdat) generic_free_nodedata(pgdat) - -#ifdef CONFIG_NUMA -/* -* If ARCH_HAS_NODEDATA_EXTENSION=n, this func is used to allocate pgdat. -* XXX: kmalloc_node() can't work well to get new node's memory at this time. -* Because, pgdat for the new node is not allocated/initialized yet itself. -* To use new node's memory, more consideration will be necessary. -*/ -#define generic_alloc_nodedata (nid) \ -({ \ - kzalloc(sizeof(pg_data_t), GFP_KERNEL); \ -}) -/* -* This definition is just for error path in node hotadd. -* For node hotremove, we have to replace this. -*/ -#define generic_free_nodedata (pgdat) kfree(pgdat) - -extern pg_data_t *node_data[]; -static inline void arch_refresh_nodedata (int nid,pg_data_t *pgdat) -{ - node_data[nid] = pgdat; -} - -#else /* !CONFIG_NUMA */ - -/* never called */ -static inline pg_data_t *generic_alloc_nodedata (int nid) -{ -BUG (); -return NULL ; -} -static inline void generic_free_nodedata (pg_data_t *pgdat) -{ -} -static inline void arch_refresh_nodedata (int nid,pg_data_t *pgdat) -{ -} -#endif /* CONFIG_NUMA */ -#endif /* CONFIG_HAVE_ARCH_NODEDATA_EXTENSION */ -void get_online_mems (void );void put_online_mems (void );Signature · 1 line
2.30.2
Signature · 2 lines
Sincerely yours, Mike.
Re: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to Michal HockoQuoted 34 lines · Michal Hocko
From: Michal Hocko <[email protected]>
We have had several reports [1][2][3] that page allocator blows up when an allocation from a possible node is requested. The underlying reason is that
NODE_DATAfor the specific node is not allocated.NUMA specific initialization is arch specific and it can vary a lot. E.g. x86 tries to initialize all nodes that have some cpu affinity (see
init_cpu_to_node) but this can be insufficient because the node might be cpuless for example.One way to address this problem would be to check for !node_online nodes when trying to get a zonelist and silently fall back to another node. That is unfortunately adding a branch into allocator hot path and it doesn't handle any other potential
NODE_DATAusers.This patch takes a different approach (following a lead of [3]) and it pre allocates pgdat for all possible nodes in an arch indipendent code
-free_area_init. All uninitialized nodes are treated as memoryless nodes.node_stateof the node is not changed because that would lead to other side effects - e.g. sysfs representation of such a node and from past discussions [4] it is known that some tools might have problems digesting that.Newly allocated pgdat only gets a minimal initialization and the rest of the work is expected to be done by the memory hotplug -
hotadd_new_pgdat(renamed tohotadd_init_pgdat).
generic_alloc_nodedatais changed to use the memblock allocator because neither page nor slab allocators are available at the stage when all pgdats are allocated. Hotplug doesn't allocate pgdat anymore so we can use the early boot allocator. The only arch specific implementation is ia64 and that is changed to use the early allocator as well.
Quoted 6 lines
Reported-by Alexey Makhalov <[email protected]>Tested-by Alexey Makhalov <[email protected]>Reported-by Nico Pache <[email protected]>Acked-by Rafael Aquini <[email protected]>Tested-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Some minor nits below, other than that
Quoted 108 lines
[1] http://lkml.kernel.org/r/[email protected]
[2] http://lkml.kernel.org/r/[email protected]
[3] http://lkml.kernel.org/r/[email protected]
[4] http://lkml.kernel.org/r/[email protected]--- arch/ia64/mm/discontig.c | 4 ++-- include/linux/memory_hotplug.h | 2 +- mm/memory_hotplug.c | 21 +++++++++------------ mm/page_alloc.c | 34 +++++++++++++++++++++++++++++++--- 4 files changed, 43 insertions(+), 18 deletions(-)arch/ia64/mm/discontig.c +2
-2@@ -608,11 +608,11 @@ void __init paging_init(void) zero_page_memmap_ptr =virt_to_page (ia64_imva (empty_zero_page)); } -pg_data_t *arch_alloc_nodedata (int nid) +pg_data_t *__init arch_alloc_nodedata (int nid) {unsigned long size =compute_pernodesize (nid); -return kzalloc (size,GFP_KERNEL ); +return memblock_alloc (size,SMP_CACHE_BYTES ); }void arch_free_nodedata (pg_data_t *pgdat)include/linux/memory_hotplug.h +1
-1@@ -44,7 +44,7 @@ extern void arch_refresh_nodedata(int nid, pg_data_t *pgdat);*/ #define generic_alloc_nodedata (nid) \ ({ \ - kzalloc(sizeof(pg_data_t), GFP_KERNEL); \ + memblock_alloc(sizeof(*pgdat), SMP_CACHE_BYTES); \ })/* * This definition is just for error path in node hotadd. mm/memory_hotplug.c +9
-12@@ -1162,19 +1162,21 @@ static void reset_node_present_pages(pg_data_t *pgdat) }/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ -static pg_data_t __ref *hotadd_new_pgdat (int nid) +static pg_data_t __ref *hotadd_init_pgdat (int nid) {struct pglist_data *pgdat; pgdat =NODE_DATA (nid); -if (!pgdat) { - pgdat =arch_alloc_nodedata (nid); -if (!pgdat) -return NULL ; +/* +* NODE_DATA is preallocated (free_area_init) but its internal +* state is not allocated completely. Add missing pieces. +* Completely offline nodes stay around and they just need +* reintialization. +*/ +if (pgdat->per_cpu_nodestats == &boot_nodestats) { pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); -arch_refresh_nodedata (nid, pgdat); }else {int cpu;/* @@ -1193,8 +1195,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid) } } -/* we can use NODE_DATA(nid) from here */ - pgdat->node_id = nid; pgdat->node_start_pfn =0 ;/* init node's zones as empty zones, we don't have any present pages.*/ @@ -1246,7 +1246,7 @@ static int __try_online_node(int nid, bool set_node_online)if (node_online (nid))return 0 ; - pgdat =hotadd_new_pgdat (nid); + pgdat =hotadd_init_pgdat (nid);if (!pgdat) {pr_err ("Cannot online node %d due to NULL pgdat\n" , nid); ret = -ENOMEM ;@@ -1445,9 +1445,6 @@ int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)return ret; error: -/* rollback pgdat allocation and others */ -if (new_node) -rollback_node_hotadd (nid);if (IS_ENABLED (CONFIG_ARCH_KEEP_MEMBLOCK ))memblock_remove (start, size); error_mem_hotplug_end:mm/page_alloc.c +0
-0
mm/page_alloc.c static void __build_all_zonelists(void *data) · line 6407
Quoted 7 lines
mm/page_alloc.c +3
-1@@ -6402,7 +6402,11 @@ static void __build_all_zonelists(void *data)if (self && !node_online (self->node_id)) {build_zonelists (self); }else { -for_each_online_node (nid) { +/* +* All possible nodes have pgdat preallocated +* free_area_init
Nit: "by free_area_init"
Quoted 6 lines
mm/page_alloc.c +2
-0+*/ +for_each_node (nid) {pg_data_t *pgdat =NODE_DATA (nid);build_zonelists (pgdat);
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8107
Quoted 10 lines
mm/page_alloc.c +5
-2@@ -8096,8 +8100,32 @@ void __init free_area_init(unsigned long *max_zone_pfn)/* Initialise every node */ mminit_verify_pageflags_layout ();setup_nr_node_ids (); -for_each_online_node (nid) { -pg_data_t *pgdat =NODE_DATA (nid); +for_each_node (nid) { +pg_data_t *pgdat; + +if (!node_online (nid)) { +pr_warn ("Node %d uninitialized by the platform. Please report with boot dmesg.\n" , nid);
Do we assume that platform code must allocate node data for all nodes in the system? Because if we don't this warning is misleading.
Quoted 8 lines
mm/page_alloc.c +8
-0+ +/* Allocator not initialized yet */ + pgdat =arch_alloc_nodedata (nid); +if (!pgdat) { +pr_err ("Cannot allocate %zuB for node %d.\n" , +sizeof (*pgdat), nid); +continue ; + }
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8125
Quoted 10 lines
mm/page_alloc.c +10
-0+arch_refresh_nodedata (nid, pgdat); +free_area_init_memoryless_node (nid); +/* +* not marking this node online because we do not want to +* confuse userspace by sysfs files/directories for node +* without any memory attached to it (see topology_init) +* The pgdat will get fully initialized when a memory is +* hotpluged into it by hotadd_init_pgdat +*/ +continue ;
This can be made slightly more concise if we fall through after arch_refresh_nodedata(), e.g. something like
... arch_refresh_nodedata(nid, pgdat); } pgdat = NODE_DATA(nid); free_area_init_node(nid); /* * Do not mark memoryless node online because we do not want to * confuse userspace by sysfs files/directories for node * without any memory attached to it (see topology_init) * The pgdat will get fully initialized when a memory is * hotpluged into it by hotadd_init_pgdat */ if (!pgdat->node_present_pages) continue;
but I don't feel strongly about it.
Quoted 9 lines
mm/page_alloc.c +3
-0+ } + + pgdat =NODE_DATA (nid);free_area_init_node (nid);/* Any memory on that node */ Signature · 1 line
2.30.2
Signature · 2 lines
Sincerely yours, Mike.
Re: [PATCH 3/6] mm, memory_hotplug: drop arch_free_nodedata
In reply to Michal HockoQuoted 9 lines · Michal Hocko
From: Michal Hocko <[email protected]>
Prior to "mm: handle uninitialized numa nodes gracefully" memory hotplug used to allocate pgdat when memory has been added to a node (
hotadd_init_pgdat)arch_free_nodedatahas been only used in the failure path because once the pgdat is exported (to be visible byNODA_DATA(nid)) it cannot really be freed because there is no synchronization available for that.
Quoted 7 lines
pgdat is allocated for each possible nodes now so the memory hotplug doesn't need to do the ever use
arch_free_nodedataso drop it.This patch doesn't introduce any functional change.
Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Quoted 68 lines
--- arch/ia64/mm/discontig.c | 5 ----- include/linux/memory_hotplug.h | 3 --- mm/memory_hotplug.c | 10 ---------- 3 files changed, 18 deletions(-)arch/ia64/mm/discontig.c +0
-5@@ -615,11 +615,6 @@ pg_data_t * __init arch_alloc_nodedata(int nid)return memblock_alloc (size,SMP_CACHE_BYTES ); } -void arch_free_nodedata (pg_data_t *pgdat) -{ -kfree (pgdat); -} -void arch_refresh_nodedata (int update_node,pg_data_t *update_pgdat) { pgdat_list[update_node] = update_pgdat;include/linux/memory_hotplug.h +0
-3@@ -24,17 +24,14 @@ struct vmem_altmap;* node_data[nid] = kzalloc() works well. But it depends on the architecture. * * In general, generic_alloc_nodedata() is used. -* Now, arch_free_nodedata() is just defined for error path of node_hot_add. * */ extern pg_data_t *arch_alloc_nodedata (int nid); -extern void arch_free_nodedata (pg_data_t *pgdat);extern void arch_refresh_nodedata (int nid,pg_data_t *pgdat);#else /* CONFIG_HAVE_ARCH_NODEDATA_EXTENSION */ #define arch_alloc_nodedata (nid) generic_alloc_nodedata(nid) -#define arch_free_nodedata (pgdat) generic_free_nodedata(pgdat)#ifdef CONFIG_NUMA /* mm/memory_hotplug.c +0
-10@@ -1217,16 +1217,6 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid)return pgdat; } -static void rollback_node_hotadd (int nid) -{ -pg_data_t *pgdat =NODE_DATA (nid); - -arch_refresh_nodedata (nid,NULL ); -free_percpu (pgdat->per_cpu_nodestats); -arch_free_nodedata (pgdat); -} - -/* * __try_online_node - online a node if offlined * @nid: the node ID Signature · 1 line
2.30.2
Signature · 2 lines
Sincerely yours, Mike.
Mike Rapoport
reviewerRe: [PATCH 4/6] mm, memory_hotplug: reorganize new pgdat initialization
In reply to Michal HockoQuoted 118 lines · Michal Hocko
From: Michal Hocko <[email protected]>
When a !node_online node is brought up it needs a hotplug specific initialization because the node could be either uninitialized yet or it could have been recycled after previous hotremove.
hotadd_init_pgdatis responsible for that.Internal pgdat state is initialized at two places currently
-hotadd_init_pgdat
-free_area_init_core_hotplug
There is no real clear cut what should go where but this patch's chosen to move the whole internal state initialization intofree_area_init_core_hotplug.hotadd_init_pgdatis still responsible to pull all the parts together - most notably to initialize zonelists because those depend on the overall topology.This patch doesn't introduce any functional change.
Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>--- include/linux/memory_hotplug.h | 2 +- mm/memory_hotplug.c | 28 +++------------------------- mm/page_alloc.c | 25 +++++++++++++++++++++++-- 3 files changed, 27 insertions(+), 28 deletions(-)include/linux/memory_hotplug.h +1
-1@@ -319,7 +319,7 @@ extern void set_zone_contiguous(struct zone *zone);extern void clear_zone_contiguous (struct zone *zone);#ifdef CONFIG_MEMORY_HOTPLUG -extern void __ref free_area_init_core_hotplug (int nid); +extern void __ref free_area_init_core_hotplug (struct pglist_data *pgdat);extern int __add_memory (int nid,u64 start,u64 size,mhp_t mhp_flags);extern int add_memory (int nid,u64 start,u64 size,mhp_t mhp_flags); extern int add_memory_resource(int nid, struct resource *resource,mm/memory_hotplug.c +3
-25@@ -1166,39 +1166,16 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid) {struct pglist_data *pgdat; - pgdat =NODE_DATA (nid); -/* * NODE_DATA is preallocated (free_area_init) but its internal * state is not allocated completely. Add missing pieces. * Completely offline nodes stay around and they just need * reintialization. */ -if (pgdat->per_cpu_nodestats == &boot_nodestats) { - pgdat->per_cpu_nodestats = - alloc_percpu(struct per_cpu_nodestat); - }else { -int cpu; -/* -* Reset the nr_zones, order and highest_zoneidx before reuse. -* Note that kswapd will init kswapd_highest_zoneidx properly -* when it starts in the near future. -*/ - pgdat->nr_zones =0 ; - pgdat->kswapd_order =0 ; - pgdat->kswapd_highest_zoneidx =0 ; -for_each_online_cpu (cpu) { -struct per_cpu_nodestat *p; - - p =per_cpu_ptr (pgdat->per_cpu_nodestats, cpu); -memset (p,0 ,sizeof (*p)); - } - } - - pgdat->node_start_pfn =0 ; + pgdat =NODE_DATA (nid);/* init node's zones as empty zones, we don't have any present pages.*/ -free_area_init_core_hotplug (nid); +free_area_init_core_hotplug (pgdat);/* * The node we allocated has no zone fallback lists. For avoiding @@ -1210,6 +1187,7 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid)* When memory is hot-added, all the memory is in offline state. So * clear all zones' present_pages because they will be updated in * online_pages() and offline_pages(). +* TODO: should be in free_area_init_core_hotplug? */ reset_node_managed_pages (pgdat);reset_node_present_pages (pgdat);mm/page_alloc.c +12
-2@@ -7506,12 +7506,33 @@ static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx,* NOTE: this function is only called during memory hotplug */ #ifdef CONFIG_MEMORY_HOTPLUG -void __ref free_area_init_core_hotplug (int nid) +void __ref free_area_init_core_hotplug (struct pglist_data *pgdat) { +int nid = pgdat->node_id;enum zone_type z; -pg_data_t *pgdat =NODE_DATA (nid); +int cpu;pgdat_init_internals (pgdat); + +if (pgdat->per_cpu_nodestats == &boot_nodestats) + pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); + +/* +* Reset the nr_zones, order and highest_zoneidx before reuse. +* Note that kswapd will init kswapd_highest_zoneidx properly +* when it starts in the near future. +*/
mm/page_alloc.c static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, · line 7534
Quoted 10 lines
mm/page_alloc.c +10
-0+ pgdat->nr_zones =0 ; + pgdat->kswapd_order =0 ; + pgdat->kswapd_highest_zoneidx =0 ; + pgdat->node_start_pfn =0 ; +for_each_online_cpu (cpu) { +struct per_cpu_nodestat *p; + + p =per_cpu_ptr (pgdat->per_cpu_nodestats, cpu); +memset (p,0 ,sizeof (*p)); + }
It seems to me that most of this is already done by free_area_init_node() at boot time anyway. Do I miss something?
Quoted 7 lines
mm/page_alloc.c +1
-0+for (z =0 ; z <MAX_NR_ZONES ; z++)zone_init_internals (&pgdat->node_zones[z], z, nid,0 ); }Signature · 1 line
2.30.2
Signature · 2 lines
Sincerely yours, Mike.
Michal Hocko
reviewerRe: [PATCH 4/6] mm, memory_hotplug: reorganize new pgdat initialization
In reply to David HildenbrandQuoted 139 lines · David Hildenbrand
Quoted 122 lines · Michal Hocko
From: Michal Hocko <[email protected]>
When a !node_online node is brought up it needs a hotplug specific initialization because the node could be either uninitialized yet or it could have been recycled after previous hotremove.
hotadd_init_pgdatis responsible for that.Internal pgdat state is initialized at two places currently
-hotadd_init_pgdat
-free_area_init_core_hotplug
There is no real clear cut what should go where but this patch's chosen to move the whole internal state initialization intofree_area_init_core_hotplug.hotadd_init_pgdatis still responsible to pull all the parts together - most notably to initialize zonelists because those depend on the overall topology.This patch doesn't introduce any functional change.
Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>--- include/linux/memory_hotplug.h | 2 +- mm/memory_hotplug.c | 28 +++------------------------- mm/page_alloc.c | 25 +++++++++++++++++++++++-- 3 files changed, 27 insertions(+), 28 deletions(-)include/linux/memory_hotplug.h +1
-1@@ -319,7 +319,7 @@ extern void set_zone_contiguous(struct zone *zone);extern void clear_zone_contiguous (struct zone *zone);#ifdef CONFIG_MEMORY_HOTPLUG -extern void __ref free_area_init_core_hotplug (int nid); +extern void __ref free_area_init_core_hotplug (struct pglist_data *pgdat);extern int __add_memory (int nid,u64 start,u64 size,mhp_t mhp_flags);extern int add_memory (int nid,u64 start,u64 size,mhp_t mhp_flags); extern int add_memory_resource(int nid, struct resource *resource,mm/memory_hotplug.c +3
-25@@ -1166,39 +1166,16 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid) {struct pglist_data *pgdat; - pgdat =NODE_DATA (nid); -/* * NODE_DATA is preallocated (free_area_init) but its internal * state is not allocated completely. Add missing pieces. * Completely offline nodes stay around and they just need * reintialization. */ -if (pgdat->per_cpu_nodestats == &boot_nodestats) { - pgdat->per_cpu_nodestats = - alloc_percpu(struct per_cpu_nodestat); - }else { -int cpu; -/* -* Reset the nr_zones, order and highest_zoneidx before reuse. -* Note that kswapd will init kswapd_highest_zoneidx properly -* when it starts in the near future. -*/ - pgdat->nr_zones =0 ; - pgdat->kswapd_order =0 ; - pgdat->kswapd_highest_zoneidx =0 ; -for_each_online_cpu (cpu) { -struct per_cpu_nodestat *p; - - p =per_cpu_ptr (pgdat->per_cpu_nodestats, cpu); -memset (p,0 ,sizeof (*p)); - } - } - - pgdat->node_start_pfn =0 ; + pgdat =NODE_DATA (nid);/* init node's zones as empty zones, we don't have any present pages.*/ -free_area_init_core_hotplug (nid); +free_area_init_core_hotplug (pgdat);/* * The node we allocated has no zone fallback lists. For avoiding @@ -1210,6 +1187,7 @@ static pg_data_t __ref *hotadd_init_pgdat(int nid)* When memory is hot-added, all the memory is in offline state. So * clear all zones' present_pages because they will be updated in * online_pages() and offline_pages(). +* TODO: should be in free_area_init_core_hotplug? */ reset_node_managed_pages (pgdat);reset_node_present_pages (pgdat);mm/page_alloc.c +16
-2@@ -7506,12 +7506,33 @@ static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx,* NOTE: this function is only called during memory hotplug */ #ifdef CONFIG_MEMORY_HOTPLUG -void __ref free_area_init_core_hotplug (int nid) +void __ref free_area_init_core_hotplug (struct pglist_data *pgdat) { +int nid = pgdat->node_id;enum zone_type z; -pg_data_t *pgdat =NODE_DATA (nid); +int cpu;pgdat_init_internals (pgdat); + +if (pgdat->per_cpu_nodestats == &boot_nodestats) + pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); + +/* +* Reset the nr_zones, order and highest_zoneidx before reuse. +* Note that kswapd will init kswapd_highest_zoneidx properly +* when it starts in the near future. +*/ + pgdat->nr_zones =0 ; + pgdat->kswapd_order =0 ; + pgdat->kswapd_highest_zoneidx =0 ; + pgdat->node_start_pfn =0 ;mm/page_alloc.c static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, · line 7538
Quoted 10 lines
mm/page_alloc.c +7
-0+for_each_online_cpu (cpu) { +struct per_cpu_nodestat *p; + + p =per_cpu_ptr (pgdat->per_cpu_nodestats, cpu); +memset (p,0 ,sizeof (*p)); + } +for (z =0 ; z <MAX_NR_ZONES ; z++)zone_init_internals (&pgdat->node_zones[z], z, nid,0 ); }I feel like we should be initializing all of this only once, just after allocating the node. There should be no difference between a node we're reusing and a "fresh" node. IOW, memory offlining should be resetting all state accordingly when the last memory goes away.
Quoted 2 lines
But I might be wrong and this patch looks like an improvement, as you say, without functional change
Yeah, I really wanted to have this simple and straightforward. To be completely honest I am not even sure this is necessary. Something really woth looking at.
Quoted 1 line
Thanks!
Signature · 2 lines
Michal Hocko SUSE Labs
Michal Hocko
reviewerRe: [PATCH 4/6] mm, memory_hotplug: reorganize new pgdat initialization
In reply to Mike RapoportQuoted 1 line · Mike Rapoport
On Thu, Jan 27, 2022 at 09:53:03AM +0100, Michal Hocko wrote:
[...]
Quoted 5 lines
Quoted 5 lines
mm/page_alloc.c +5
-0+ pgdat->nr_zones =0 ; + pgdat->kswapd_order =0 ; + pgdat->kswapd_highest_zoneidx =0 ; + pgdat->node_start_pfn =0 ; +for_each_online_cpu (cpu) {
Quoted 8 lines
mm/page_alloc.c static void __meminit zone_init_internals(struct zone *zone, enum zone_type idx, · line 7534
Quoted 5 lines
mm/page_alloc.c +5
-0+struct per_cpu_nodestat *p; + + p =per_cpu_ptr (pgdat->per_cpu_nodestats, cpu); +memset (p,0 ,sizeof (*p)); + }It seems to me that most of this is already done by
free_area_init_node()at boot time anyway. Do I miss something?
As already replied to David I really didn't want to change the existing logic.
Signature · 2 lines
Michal Hocko SUSE Labs
Michal Hocko
reviewerRe: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to Mike RapoportQuoted 1 line · Mike Rapoport
On Thu, Jan 27, 2022 at 09:53:01AM +0100, Michal Hocko wrote:
[...]
Quoted 5 lines
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8107
Quoted 2 lines
mm/page_alloc.c +2
-0+if (!node_online (nid)) { +pr_warn ("Node %d uninitialized by the platform. Please report with boot dmesg.\n" , nid);Do we assume that platform code must allocate node data for all nodes in the system? Because if we don't this warning is misleading.
At least x86 does that (init_cpu_to_node). Now that you brought that up I guess you are right that this could be more misleading than helpful. What about
pr_info("Initializing node %d as memoryless\n", nid);Is this better?
Quoted 40 lines
Quoted 8 lines
mm/page_alloc.c +8
-0+ +/* Allocator not initialized yet */ + pgdat =arch_alloc_nodedata (nid); +if (!pgdat) { +pr_err ("Cannot allocate %zuB for node %d.\n" , +sizeof (*pgdat), nid); +continue ; + }mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8125
Quoted 10 lines
mm/page_alloc.c +10
-0+arch_refresh_nodedata (nid, pgdat); +free_area_init_memoryless_node (nid); +/* +* not marking this node online because we do not want to +* confuse userspace by sysfs files/directories for node +* without any memory attached to it (see topology_init) +* The pgdat will get fully initialized when a memory is +* hotpluged into it by hotadd_init_pgdat +*/ +continue ;This can be made slightly more concise if we fall through after
arch_refresh_nodedata(), e.g. something like... arch_refresh_nodedata(nid, pgdat); } pgdat = NODE_DATA(nid); free_area_init_node(nid); /* * Do not mark memoryless node online because we do not want to * confuse userspace by sysfs files/directories for node * without any memory attached to it (see topology_init) * The pgdat will get fully initialized when a memory is * hotpluged into it by hotadd_init_pgdat */ if (!pgdat->node_present_pages) continue;
Quoted 1 line
but I don't feel strongly about it.
I do not have strong preference either way. Unless this is considered better by more people I would stick with what I have.
Signature · 2 lines
Michal Hocko SUSE Labs
Michal Hocko
reviewerRe: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to David HildenbrandQuoted 1 line · David Hildenbrand
On 27.01.22 09:53, Michal Hocko wrote:
[...]
Quoted 13 lines
Quoted 13 lines
arch/ia64/mm/discontig.c +1
-1@@ -608,11 +608,11 @@ void __init paging_init(void) zero_page_memmap_ptr =virt_to_page (ia64_imva (empty_zero_page)); } -pg_data_t *arch_alloc_nodedata (int nid) +pg_data_t *__init arch_alloc_nodedata (int nid) {unsigned long size =compute_pernodesize (nid);
Quoted 8 lines
arch/ia64/mm/discontig.c void __init paging_init(void) · line 615
Quoted 2 lines
arch/ia64/mm/discontig.c +1
-1-return kzalloc (size,GFP_KERNEL ); +return memblock_alloc (size,SMP_CACHE_BYTES );I feel like we should have
long
arch_pgdat_size(void)instead and have a generic allocation function.But we can clean that up in the future.
I can have a look later (unless somebody beat me to it). I am not even sure this whole ia64 weirdness is useful these days.
[...]
Quoted 11 lines
arch/ia64/mm/discontig.c int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags) · line 1450
Quoted 7 lines
arch/ia64/mm/discontig.c +0
-3@@ -1445,9 +1445,6 @@ int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)return ret; error: -/* rollback pgdat allocation and others */ -if (new_node) -rollback_node_hotadd (nid);As static
rollback_node_hotadd()is unused in this patch, doesn't this trigger a warning? IOW, maybe merge at least therollback_node_hotadd()removal into this patch. Thearch_free_nodedata()removal can stay separate.
It is my slight preference to have this patch smaller to be easier to review and therefore I have removed all parts in a follow up. If a warning is a real deal at this stage then I can change that of course.
Quoted 17 lines
Quoted 9 lines
arch/ia64/mm/discontig.c +0
-0if (IS_ENABLED (CONFIG_ARCH_KEEP_MEMBLOCK ))memblock_remove (start, size); error_mem_hotplug_end:mm/page_alloc.c +0
-0mm/page_alloc.c static void __build_all_zonelists(void *data) · line 6406
Quoted 6 lines
mm/page_alloc.c +2
-1@@ -6402,7 +6402,11 @@ static void __build_all_zonelists(void *data)if (self && !node_online (self->node_id)) {build_zonelists (self); }else { -for_each_online_node (nid) { +/* +* All possible nodes have pgdat preallocated ... in
free_area_init()?
Will fix it up.
Quoted 22 lines
Quoted 22 lines
mm/page_alloc.c +13
-2+* free_area_init +*/ +for_each_node (nid) {pg_data_t *pgdat =NODE_DATA (nid);build_zonelists (pgdat);@@ -8096,8 +8100,32 @@ void __init free_area_init(unsigned long *max_zone_pfn)/* Initialise every node */ mminit_verify_pageflags_layout ();setup_nr_node_ids (); -for_each_online_node (nid) { -pg_data_t *pgdat =NODE_DATA (nid); +for_each_node (nid) { +pg_data_t *pgdat; + +if (!node_online (nid)) { +pr_warn ("Node %d uninitialized by the platform. Please report with boot dmesg.\n" , nid); + +/* Allocator not initialized yet */ + pgdat =arch_alloc_nodedata (nid); +if (!pgdat) { + pr_err("Cannot allocate %zuB for node %d.\n",
Quoted 8 lines
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8116
Quoted 4 lines
mm/page_alloc.c +4
-0+ sizeof(*pgdat), nid); +continue ; + } +arch_refresh_nodedata (nid, pgdat);We could get rid of
arch_refresh_nodedata()now and simply merge that intoarch_alloc_nodedata(). But depends on how we want to proceed witharch_alloc_nodedata()eventually.
yeah, I will postpone that to a later follow ups.
Quoted 1 line
Thanks!
Btw now I have realized that I have lost the following hung somewhere:
mm/internal.h +2 -0
@@ -718,4 +718,6 @@ void vunmap_range_noflush(unsigned long start, unsigned long end);int numa_migrate_prep (struct page *page,struct vm_area_struct *vma,unsigned long addr,int page_nid,int *flags); +DECLARE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); +#endif /* __MM_INTERNAL_H */
mm/page_alloc.c +1 -1
@@ -6380,7 +6380,7 @@ static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonesta#define BOOT_PAGESET_BATCH 1 static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset); static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats); -static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); +DEFINE_PER_CPU (struct per_cpu_nodestat , boot_nodestats);static void __build_all_zonelists (void *data) {
Signature · 2 lines
Michal Hocko SUSE Labs
Mike Rapoport
reviewerRe: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to Michal HockoQuoted 11 lines · Michal Hocko
Quoted 1 line · Mike Rapoport
On Thu, Jan 27, 2022 at 09:53:01AM +0100, Michal Hocko wrote:
[...]
Quoted 5 lines
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8107
Quoted 2 lines
mm/page_alloc.c +2
-0+if (!node_online (nid)) { +pr_warn ("Node %d uninitialized by the platform. Please report with boot dmesg.\n" , nid);Do we assume that platform code must allocate node data for all nodes in the system? Because if we don't this warning is misleading.
At least x86 does that (
init_cpu_to_node). Now that you brought that up I guess you are right that this could be more misleading than helpful.
I'm not sure if other architectures allocate memoryless nodes, but for sure only x86 initializes it with free_area_init_memoryless_node().
Quoted 3 lines
What about
pr_info("Initializing node %d as memoryless\n", nid);Is this better?
I think yes.
Quoted 47 lines
Quoted 40 lines
Quoted 8 lines
mm/page_alloc.c +8
-0+ +/* Allocator not initialized yet */ + pgdat =arch_alloc_nodedata (nid); +if (!pgdat) { +pr_err ("Cannot allocate %zuB for node %d.\n" , +sizeof (*pgdat), nid); +continue ; + }mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8125
Quoted 10 lines
mm/page_alloc.c +10
-0+arch_refresh_nodedata (nid, pgdat); +free_area_init_memoryless_node (nid); +/* +* not marking this node online because we do not want to +* confuse userspace by sysfs files/directories for node +* without any memory attached to it (see topology_init) +* The pgdat will get fully initialized when a memory is +* hotpluged into it by hotadd_init_pgdat +*/ +continue ;This can be made slightly more concise if we fall through after
arch_refresh_nodedata(), e.g. something like... arch_refresh_nodedata(nid, pgdat); } pgdat = NODE_DATA(nid); free_area_init_node(nid); /* * Do not mark memoryless node online because we do not want to * confuse userspace by sysfs files/directories for node * without any memory attached to it (see topology_init) * The pgdat will get fully initialized when a memory is * hotpluged into it by hotadd_init_pgdat */ if (!pgdat->node_present_pages) continue;Quoted 1 line
but I don't feel strongly about it.
I do not have strong preference either way. Unless this is considered better by more people I would stick with what I have.
Signature · 2 lines
Michal Hocko SUSE Labs
Signature · 2 lines
Sincerely yours, Mike.
Re: [PATCH 1/6] mm, memory_hotplug: make arch_alloc_nodedata independent on CONFIG_MEMORY_HOTPLUG
In reply to Michal HockoQuoted 9 lines · Michal Hocko
From: Michal Hocko <[email protected]>
This is a preparatory patch and it doesn't introduce any functional change. It merely pulls out
arch_alloc_nodedata(and co) outside ofCONFIG_MEMORY_HOTPLUGbecause the following patch will need to call this from the generic MM code.Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Signature · 2 lines
Oscar Salvador SUSE Labs
Re: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to Michal HockoQuoted 34 lines · Michal Hocko
From: Michal Hocko <[email protected]>
We have had several reports [1][2][3] that page allocator blows up when an allocation from a possible node is requested. The underlying reason is that
NODE_DATAfor the specific node is not allocated.NUMA specific initialization is arch specific and it can vary a lot. E.g. x86 tries to initialize all nodes that have some cpu affinity (see
init_cpu_to_node) but this can be insufficient because the node might be cpuless for example.One way to address this problem would be to check for !node_online nodes when trying to get a zonelist and silently fall back to another node. That is unfortunately adding a branch into allocator hot path and it doesn't handle any other potential
NODE_DATAusers.This patch takes a different approach (following a lead of [3]) and it pre allocates pgdat for all possible nodes in an arch indipendent code
-free_area_init. All uninitialized nodes are treated as memoryless nodes.node_stateof the node is not changed because that would lead to other side effects - e.g. sysfs representation of such a node and from past discussions [4] it is known that some tools might have problems digesting that.Newly allocated pgdat only gets a minimal initialization and the rest of the work is expected to be done by the memory hotplug -
hotadd_new_pgdat(renamed tohotadd_init_pgdat).
generic_alloc_nodedatais changed to use the memblock allocator because neither page nor slab allocators are available at the stage when all pgdats are allocated. Hotplug doesn't allocate pgdat anymore so we can use the early boot allocator. The only arch specific implementation is ia64 and that is changed to use the early allocator as well.
Quoted 6 lines
Reported-by Alexey Makhalov <[email protected]>Tested-by Alexey Makhalov <[email protected]>Reported-by Nico Pache <[email protected]>Acked-by Rafael Aquini <[email protected]>Tested-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
With the mentioned fixups:
Signature · 2 lines
Oscar Salvador SUSE Labs
Re: [PATCH 3/6] mm, memory_hotplug: drop arch_free_nodedata
In reply to Michal HockoQuoted 9 lines · Michal Hocko
From: Michal Hocko <[email protected]>
Prior to "mm: handle uninitialized numa nodes gracefully" memory hotplug used to allocate pgdat when memory has been added to a node (
hotadd_init_pgdat)arch_free_nodedatahas been only used in the failure path because once the pgdat is exported (to be visible byNODA_DATA(nid)) it cannot really be freed because there is no synchronization available for that.
Quoted 7 lines
pgdat is allocated for each possible nodes now so the memory hotplug doesn't need to do the ever use
arch_free_nodedataso drop it.This patch doesn't introduce any functional change.
Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Signature · 2 lines
Oscar Salvador SUSE Labs
Re: [PATCH 4/6] mm, memory_hotplug: reorganize new pgdat initialization
In reply to Michal HockoQuoted 16 lines · Michal Hocko
From: Michal Hocko <[email protected]>
When a !node_online node is brought up it needs a hotplug specific initialization because the node could be either uninitialized yet or it could have been recycled after previous hotremove.
hotadd_init_pgdatis responsible for that.Internal pgdat state is initialized at two places currently
-hotadd_init_pgdat
-free_area_init_core_hotplug
There is no real clear cut what should go where but this patch's chosen to move the whole internal state initialization intofree_area_init_core_hotplug.hotadd_init_pgdatis still responsible to pull all the parts together - most notably to initialize zonelists because those depend on the overall topology.
Quoted 4 lines
This patch doesn't introduce any functional change.
Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
I have some comments below:
Quoted 11 lines
mm/memory_hotplug.c +1
-0/* * The node we allocated has no zone fallback lists. For avoiding @@ -1210,6 +1187,7 @@ static pg_data_t __ref *hotadd_init_pgdat(intnid) * When memory is hot-added, all the memory is in offline state. So * clear all zones' present_pages because they will be updated in * online_pages() and offline_pages(). +* TODO: should be in free_area_init_core_hotplug? */ reset_node_managed_pages (pgdat);
reset_node_present_pages(pgdat);
Most likely yes, but more below.
Quoted 28 lines
mm/page_alloc.c +13
-2@@ -7506,12 +7506,33 @@ static void __meminitzone_init_internals(struct zone *zone, enum zone_type idx, * NOTE: this function is only called during memory hotplug */ #ifdef CONFIG_MEMORY_HOTPLUG -void __ref free_area_init_core_hotplug (int nid) +void __ref free_area_init_core_hotplug (struct pglist_data *pgdat) { +int nid = pgdat->node_id;enum zone_type z; -pg_data_t *pgdat =NODE_DATA (nid); +int cpu;pgdat_init_internals (pgdat); + +if (pgdat->per_cpu_nodestats == &boot_nodestats) + pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); + +/* +* Reset the nr_zones, order and highest_zoneidx before reuse. +* Note that kswapd will init kswapd_highest_zoneidx properly +* when it starts in the near future. +*/ + pgdat->nr_zones =0 ;
mm/page_alloc.c static void __meminit · line 7536
Quoted 10 lines
mm/page_alloc.c +10
-0+ pgdat->kswapd_order =0 ; + pgdat->kswapd_highest_zoneidx =0 ; + pgdat->node_start_pfn =0 ; +for_each_online_cpu (cpu) { +struct per_cpu_nodestat *p; + + p =per_cpu_ptr (pgdat->per_cpu_nodestats, cpu); +memset (p,0 ,sizeof (*p)); + } +
I am with David that the cleanest way would be to initialize all of this (pgdat internals) only once at boot time now that we are initializing the node anyway, and memory-offlining would have to wipe it out once last pgdat's chunk of memory goes away.
I guess free_area_init_core_hotplug() would still need to allocate alloc_percpu(struct per_cpu_nodestat) for the "new" node since doing it at boot time is not desirable because of the memory waste on unused nodes.
As you said, probably worth looking at but not really something to cover on this patchset at all.
Signature · 2 lines
Oscar Salvador SUSE Labs
Re: [PATCH 5/6] mm: make free_area_init_node aware of memory less nodes
In reply to Michal HockoQuoted 7 lines · Michal Hocko
From: Michal Hocko <[email protected]>
free_area_init_nodeis also called from memory less node initialization path (free_area_init_memoryless_node). It doesn't really make much sense to display the physical memory range for those nodes:
Initmem setup node XX [mem 0x0000000000000000-0x0000000000000000]
Quoted 5 lines
Instead be explicit that the node is memoryless:
Initmem setup node XX as memorylessAcked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Signature · 2 lines
Oscar Salvador SUSE Labs
Re: [PATCH 6/6] memcg: do not tweak node in alloc_mem_cgroup_per_node_info
In reply to Michal HockoQuoted 10 lines · Michal Hocko
From: Wei Yang <[email protected]>
alloc_mem_cgroup_per_node_infois allocated for each possible node and this used to be a problem because not !node_online nodes didn't have appropriate data structure allocated. This has changed by "mm: handle uninitialized numa nodes gracefully" so we can drop the special casing here.Signed-off-by Wei Yang <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Signature · 2 lines
Oscar Salvador SUSE Labs
Michal Hocko
reviewer Tested-by Alexey Makhalov Acked-by Rafael Aquini Tested-by Rafael Aquini Acked-by David Hildenbrand Reviewed-by Oscar Salvador Acked-by Mike Rapoport on v1 2/6Re: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to Michal HockoThis is an updated version of the patch including acks and reviews. All other patches in this series have only got acks and reviews at this stage so I am not reposting the full bunch. Please add to the MM tree Andrew. Thanks!
---
From ad93c2641c4073213a0f9f0d0cd5286aff0de8e0 Mon Sep 17 00:00:00 2001
From: Michal Hocko <[email protected]>
Date: Thu, 9 Dec 2021 10:00:02 +0100
Subject: [PATCH] mm: handle uninitialized numa nodes gracefully
We have had several reports [1][2][3] that page allocator blows up when an allocation from a possible node is requested. The underlying reason is that NODE_DATA for the specific node is not allocated.
NUMA specific initialization is arch specific and it can vary a lot. E.g. x86 tries to initialize all nodes that have some cpu affinity (see init_cpu_to_node) but this can be insufficient because the node might be cpuless for example.
One way to address this problem would be to check for !node_online nodes when trying to get a zonelist and silently fall back to another node. That is unfortunately adding a branch into allocator hot path and it doesn't handle any other potential NODE_DATA users.
This patch takes a different approach (following a lead of [3]) and it pre allocates pgdat for all possible nodes in an arch indipendent code
- free_area_init. All uninitialized nodes are treated as memoryless nodes. node_state of the node is not changed because that would lead to other side effects - e.g. sysfs representation of such a node and from past discussions [4] it is known that some tools might have problems digesting that.
Newly allocated pgdat only gets a minimal initialization and the rest of the work is expected to be done by the memory hotplug - hotadd_new_pgdat (renamed to hotadd_init_pgdat).
generic_alloc_nodedata is changed to use the memblock allocator because neither page nor slab allocators are available at the stage when all pgdats are allocated. Hotplug doesn't allocate pgdat anymore so we can use the early boot allocator. The only arch specific implementation is ia64 and that is changed to use the early allocator as well.
[1] http://lkml.kernel.org/r/[email protected]
[2] http://lkml.kernel.org/r/[email protected]
[3] http://lkml.kernel.org/r/[email protected]
[4] http://lkml.kernel.org/r/[email protected]
--- arch/ia64/mm/discontig.c | 4 ++-- include/linux/memory_hotplug.h | 2 +- mm/internal.h | 2 ++ mm/memory_hotplug.c | 21 +++++++++----------- mm/page_alloc.c | 36 ++++++++++++++++++++++++++++++---- 5 files changed, 46 insertions(+), 19 deletions(-)
arch/ia64/mm/discontig.c +2 -2
@@ -608,11 +608,11 @@ void __init paging_init(void) zero_page_memmap_ptr =virt_to_page (ia64_imva (empty_zero_page)); } -pg_data_t *arch_alloc_nodedata (int nid) +pg_data_t *__init arch_alloc_nodedata (int nid) {unsigned long size =compute_pernodesize (nid); -return kzalloc (size,GFP_KERNEL ); +return memblock_alloc (size,SMP_CACHE_BYTES ); }void arch_free_nodedata (pg_data_t *pgdat)
include/linux/memory_hotplug.h +1 -1
@@ -44,7 +44,7 @@ extern void arch_refresh_nodedata(int nid, pg_data_t *pgdat);*/ #define generic_alloc_nodedata (nid) \ ({ \ - kzalloc(sizeof(pg_data_t), GFP_KERNEL); \ + memblock_alloc(sizeof(*pgdat), SMP_CACHE_BYTES); \ })/* * This definition is just for error path in node hotadd.
mm/internal.h +2 -0
@@ -718,4 +718,6 @@ void vunmap_range_noflush(unsigned long start, unsigned long end);int numa_migrate_prep (struct page *page,struct vm_area_struct *vma,unsigned long addr,int page_nid,int *flags); +DECLARE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); +#endif /* __MM_INTERNAL_H */
mm/memory_hotplug.c +9 -12
@@ -1162,19 +1162,21 @@ static void reset_node_present_pages(pg_data_t *pgdat) }/* we are OK calling __meminit stuff here - we have CONFIG_MEMORY_HOTPLUG */ -static pg_data_t __ref *hotadd_new_pgdat (int nid) +static pg_data_t __ref *hotadd_init_pgdat (int nid) {struct pglist_data *pgdat; pgdat =NODE_DATA (nid); -if (!pgdat) { - pgdat =arch_alloc_nodedata (nid); -if (!pgdat) -return NULL ; +/* +* NODE_DATA is preallocated (free_area_init) but its internal +* state is not allocated completely. Add missing pieces. +* Completely offline nodes stay around and they just need +* reintialization. +*/ +if (pgdat->per_cpu_nodestats == &boot_nodestats) { pgdat->per_cpu_nodestats = alloc_percpu(struct per_cpu_nodestat); -arch_refresh_nodedata (nid, pgdat); }else {int cpu;/*
@@ -1193,8 +1195,6 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid) } } -/* we can use NODE_DATA(nid) from here */ - pgdat->node_id = nid; pgdat->node_start_pfn =0 ;/* init node's zones as empty zones, we don't have any present pages.*/
@@ -1246,7 +1246,7 @@ static int __try_online_node(int nid, bool set_node_online)if (node_online (nid))return 0 ; - pgdat =hotadd_new_pgdat (nid); + pgdat =hotadd_init_pgdat (nid);if (!pgdat) {pr_err ("Cannot online node %d due to NULL pgdat\n" , nid); ret = -ENOMEM ;
@@ -1445,9 +1445,6 @@ int __ref add_memory_resource(int nid, struct resource *res, mhp_t mhp_flags)return ret; error: -/* rollback pgdat allocation and others */ -if (new_node) -rollback_node_hotadd (nid);if (IS_ENABLED (CONFIG_ARCH_KEEP_MEMBLOCK ))memblock_remove (start, size); error_mem_hotplug_end:
mm/page_alloc.c +32 -4
@@ -6380,7 +6380,7 @@ static void per_cpu_pages_init(struct per_cpu_pages *pcp, struct per_cpu_zonesta#define BOOT_PAGESET_BATCH 1 static DEFINE_PER_CPU(struct per_cpu_pages, boot_pageset); static DEFINE_PER_CPU(struct per_cpu_zonestat, boot_zonestats); -static DEFINE_PER_CPU(struct per_cpu_nodestat, boot_nodestats); +DEFINE_PER_CPU (struct per_cpu_nodestat , boot_nodestats);static void __build_all_zonelists (void *data) {
@@ -6402,7 +6402,11 @@ static void __build_all_zonelists(void *data)if (self && !node_online (self->node_id)) {build_zonelists (self); }else { -for_each_online_node (nid) { +/* +* All possible nodes have pgdat preallocated +* in free_area_init +*/ +for_each_node (nid) {pg_data_t *pgdat =NODE_DATA (nid);build_zonelists (pgdat);
@@ -8096,8 +8100,32 @@ void __init free_area_init(unsigned long *max_zone_pfn)/* Initialise every node */ mminit_verify_pageflags_layout ();setup_nr_node_ids (); -for_each_online_node (nid) { -pg_data_t *pgdat =NODE_DATA (nid); +for_each_node (nid) { +pg_data_t *pgdat; + +if (!node_online (nid)) { +pr_info ("Initializing node %d as memoryless\n" , nid); + +/* Allocator not initialized yet */ + pgdat =arch_alloc_nodedata (nid); +if (!pgdat) { +pr_err ("Cannot allocate %zuB for node %d.\n" , +sizeof (*pgdat), nid); +continue ; + } +arch_refresh_nodedata (nid, pgdat); +free_area_init_memoryless_node (nid); +/* +* not marking this node online because we do not want to +* confuse userspace by sysfs files/directories for node +* without any memory attached to it (see topology_init) +* The pgdat will get fully initialized when a memory is +* hotpluged into it by hotadd_init_pgdat +*/ +continue ; + } + + pgdat =NODE_DATA (nid);free_area_init_node (nid);/* Any memory on that node */
Signature · 5 lines
2.30.2 -- Michal Hocko SUSE Labs
Re: [PATCH 1/6] mm, memory_hotplug: make arch_alloc_nodedata independent on CONFIG_MEMORY_HOTPLUG
In reply to Michal HockoQuoted 9 lines · Michal Hocko
From: Michal Hocko <[email protected]>
This is a preparatory patch and it doesn't introduce any functional change. It merely pulls out
arch_alloc_nodedata(and co) outside ofCONFIG_MEMORY_HOTPLUGbecause the following patch will need to call this from the generic MM code.Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Signature · 2 lines
Wei Yang Help you, Help me
Re: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to Michal HockoQuoted 11 lines · Michal Hocko
[...]
Quoted 10 lines
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8117
Quoted 10 lines
mm/page_alloc.c +10
-0+ +/* Allocator not initialized yet */ + pgdat =arch_alloc_nodedata (nid); +if (!pgdat) { +pr_err ("Cannot allocate %zuB for node %d.\n" , +sizeof (*pgdat), nid); +continue ; + } +arch_refresh_nodedata (nid, pgdat); +free_area_init_memoryless_node (nid);
free_area_init_memoryless_node() seems to be defined used out side page_alloc.c? It just call free_area_init_node() directly. We want to use the name to reflect the effect?
Quoted 7 lines
Quoted 7 lines
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8124
Quoted 7 lines
mm/page_alloc.c +7
-0+/* +* not marking this node online because we do not want to +* confuse userspace by sysfs files/directories for node +* without any memory attached to it (see topology_init) +* The pgdat will get fully initialized when a memory is +* hotpluged into it by hotadd_init_pgdat +*/
Hmm... which following step would mark the node online? On x86, the node is onlined in alloc_node_date(). This is not onlined here.
Quoted 23 lines
Quoted 23 lines
Review
Quoted 1 line
(quoted diff) +1
-0+ continue;This can be made slightly more concise if we fall through after
arch_refresh_nodedata(), e.g. something like... arch_refresh_nodedata(nid, pgdat); } pgdat = NODE_DATA(nid); free_area_init_node(nid); /* * Do not mark memoryless node online because we do not want to * confuse userspace by sysfs files/directories for node * without any memory attached to it (see topology_init) * The pgdat will get fully initialized when a memory is * hotpluged into it by hotadd_init_pgdat */ if (!pgdat->node_present_pages) continue;
Quoted 4 lines
Quoted 1 line
but I don't feel strongly about it.
I do not have strong preference either way. Unless this is considered better by more people I would stick with what I have.
I agree with Mike.
Quoted 3 lines
Signature · 2 lines
Michal Hocko SUSE Labs
Signature · 2 lines
Wei Yang Help you, Help me
Re: [PATCH 3/6] mm, memory_hotplug: drop arch_free_nodedata
In reply to Michal HockoQuoted 9 lines · Michal Hocko
From: Michal Hocko <[email protected]>
Prior to "mm: handle uninitialized numa nodes gracefully" memory hotplug used to allocate pgdat when memory has been added to a node (
hotadd_init_pgdat)arch_free_nodedatahas been only used in the failure path because once the pgdat is exported (to be visible byNODA_DATA(nid)) it cannot really be freed because there is no synchronization available for that.
Quoted 7 lines
pgdat is allocated for each possible nodes now so the memory hotplug doesn't need to do the ever use
arch_free_nodedataso drop it.This patch doesn't introduce any functional change.
Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Signature · 2 lines
Wei Yang Help you, Help me
Re: [PATCH 4/6] mm, memory_hotplug: reorganize new pgdat initialization
In reply to Michal HockoQuoted 15 lines · Michal Hocko
From: Michal Hocko <[email protected]>
When a !node_online node is brought up it needs a hotplug specific initialization because the node could be either uninitialized yet or it could have been recycled after previous hotremove.
hotadd_init_pgdatis responsible for that.Internal pgdat state is initialized at two places currently
-hotadd_init_pgdat
-free_area_init_core_hotplug
There is no real clear cut what should go where but this patch's chosen to move the whole internal state initialization intofree_area_init_core_hotplug.hotadd_init_pgdatis still responsible to pull all the parts together - most notably to initialize zonelists because those depend on the overall topology.
Quoted 4 lines
This patch doesn't introduce any functional change.
Acked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Signature · 2 lines
Wei Yang Help you, Help me
Re: [PATCH 5/6] mm: make free_area_init_node aware of memory less nodes
In reply to Michal HockoQuoted 7 lines · Michal Hocko
From: Michal Hocko <[email protected]>
free_area_init_nodeis also called from memory less node initialization path (free_area_init_memoryless_node). It doesn't really make much sense to display the physical memory range for those nodes:
Initmem setup node XX [mem 0x0000000000000000-0x0000000000000000]
Quoted 5 lines
Instead be explicit that the node is memoryless:
Initmem setup node XX as memorylessAcked-by Rafael Aquini <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Signature · 2 lines
Wei Yang Help you, Help me
Re: [PATCH 6/6] memcg: do not tweak node in alloc_mem_cgroup_per_node_info
In reply to Michal HockoQuoted 4 lines · Michal Hocko
From: Wei Yang <[email protected]>
alloc_mem_cgroup_per_node_infois allocated for each possible node and this used to be a problem because not !node_online nodes didn't have
^^^ a typo here?
!node_online nodes is enough?
Quoted 6 lines
appropriate data structure allocated. This has changed by "mm: handle uninitialized numa nodes gracefully" so we can drop the special casing here.
Signed-off-by Wei Yang <[email protected]>Signed-off-by Michal Hocko <[email protected]>
Signature · 2 lines
Wei Yang Help you, Help me
Michal Hocko
reviewerRe: [PATCH 6/6] memcg: do not tweak node in alloc_mem_cgroup_per_node_info
Quoted 8 lines · Wei Yang
Quoted 4 lines · Michal Hocko
From: Wei Yang <[email protected]>
alloc_mem_cgroup_per_node_infois allocated for each possible node and this used to be a problem because not !node_online nodes didn't have^^^ a typo here?
!node_online nodes is enough?
Yes. Fixed up. Thx!
Quoted 10 lines
Quoted 6 lines
appropriate data structure allocated. This has changed by "mm: handle uninitialized numa nodes gracefully" so we can drop the special casing here.
Signed-off-by Wei Yang <[email protected]>Signed-off-by Michal Hocko <[email protected]>Signature · 2 lines
Wei Yang Help you, Help me
Signature · 2 lines
Michal Hocko SUSE Labs
Michal Hocko
reviewerRe: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to Wei YangQuoted 16 lines · Wei Yang
Quoted 11 lines · Michal Hocko
[...]
Quoted 10 lines
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8117
Quoted 10 lines
mm/page_alloc.c +10
-0+ +/* Allocator not initialized yet */ + pgdat =arch_alloc_nodedata (nid); +if (!pgdat) { +pr_err ("Cannot allocate %zuB for node %d.\n" , +sizeof (*pgdat), nid); +continue ; + } +arch_refresh_nodedata (nid, pgdat); +free_area_init_memoryless_node (nid);
free_area_init_memoryless_node()seems to be defined used out sidepage_alloc.c? It just callfree_area_init_node()directly. We want to use the name to reflect the effect?
yes.
Quoted 10 lines
Quoted 7 lines
Quoted 7 lines
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8124
Quoted 7 lines
mm/page_alloc.c +7
-0+/* +* not marking this node online because we do not want to +* confuse userspace by sysfs files/directories for node +* without any memory attached to it (see topology_init) +* The pgdat will get fully initialized when a memory is +* hotpluged into it by hotadd_init_pgdat +*/ Hmm... which following step would mark the node online? On x86, the node is onlined in
alloc_node_date(). This is not onlined here.
The comment tries to explain that this happens during the memory hotplug. Or maybe I have missed your question?
[...]
Signature · 2 lines
Michal Hocko SUSE Labs
Re: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
Quoted 14 lines · Michal Hocko
Quoted 12 lines · Wei Yang
On Thu, Jan 27, 2022 at 03:47:40PM +0100, Michal Hocko wrote:
[...]Quoted 7 lines
Quoted 7 lines
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8124
Quoted 7 lines
mm/page_alloc.c +7
-0+/* +* not marking this node online because we do not want to +* confuse userspace by sysfs files/directories for node +* without any memory attached to it (see topology_init) +* The pgdat will get fully initialized when a memory is +* hotpluged into it by hotadd_init_pgdat +*/ Hmm... which following step would mark the node online? On x86, the node is onlined in
alloc_node_date(). This is not onlined here.
Quoted 3 lines
The comment tries to explain that this happens during the memory hotplug. Or maybe I have missed your question?
I am not sure for others, while the comment confused me a little.
Currently in kernel, there are two situations for node online:
* during memory hotplug
* during sys-init
For memory hotplug, we allocate pgdat and online node. And current hot-add process has already put them in two steps:
1. __try_online_node()
2. node_set_online()
So emphasize "not online" node here, confuse me a little. It is a natural thing to not online node until it has memory.
But from another point of view, the comment here is reasonable. During sys-init, we online node at the same time when creating pgdat. And even for memory-less node on x86, we online them too.
Well, this is all about the comment. I have tried to grab may head but not come up with a better idea.
Or maybe this is just my personal feeling, don't bother if no-one else feel like this.
Quoted 4 lines
[...]
Signature · 2 lines
Michal Hocko SUSE Labs
Signature · 2 lines
Wei Yang Help you, Help me
Mike Rapoport
reviewerRe: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
Quoted 39 lines · Wei Yang
Quoted 14 lines · Michal Hocko
Quoted 12 lines · Wei Yang
On Thu, Jan 27, 2022 at 03:47:40PM +0100, Michal Hocko wrote:
[...]Quoted 7 lines
Quoted 7 lines
mm/page_alloc.c void __init free_area_init(unsigned long *max_zone_pfn) · line 8124
Quoted 7 lines
mm/page_alloc.c +7
-0+/* +* not marking this node online because we do not want to +* confuse userspace by sysfs files/directories for node +* without any memory attached to it (see topology_init) +* The pgdat will get fully initialized when a memory is +* hotpluged into it by hotadd_init_pgdat +*/ Hmm... which following step would mark the node online? On x86, the node is onlined in
alloc_node_date(). This is not onlined here.Quoted 3 lines
The comment tries to explain that this happens during the memory hotplug. Or maybe I have missed your question?
I am not sure for others, while the comment confused me a little.
Currently in kernel, there are two situations for node online:
* during memory hotplug
* during sys-initFor memory hotplug, we allocate pgdat and online node. And current hot-add process has already put them in two steps:
1.
__try_online_node()
2.node_set_online()So emphasize "not online" node here, confuse me a little. It is a natural thing to not online node until it has memory.
But from another point of view, the comment here is reasonable. During sys-init, we online node at the same time when creating pgdat. And even for memory-less node on x86, we online them too.
Quoted 5 lines
Well, this is all about the comment. I have tried to grab may head but not come up with a better idea.
Or maybe this is just my personal feeling, don't bother if no-one else feel like this.
I shuffled the words a bit, maybe this sounds better not only to me :)
/*
* We do not want to confuse userspace by sysfs files/directories for node
* without any memory attached to it, so this node is not marked as
* N_MEMORY and not marked online so that topology_init() won't create
* sysfs hierarchy for this node. The pgdat will get fully initialized by
* hotadd_init_pgdat() when memory is hotpluged into this node */
Signature · 2 lines
Sincerely yours, Mike.
Re: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
Quoted 48 lines · Mike Rapoport
Quoted 39 lines · Wei Yang
Quoted 14 lines · Michal Hocko
Quoted 12 lines · Wei Yang
On Thu, Jan 27, 2022 at 03:47:40PM +0100, Michal Hocko wrote:
[...]Quoted 7 lines
Quoted 7 lines
Review
Quoted 7 lines
(quoted diff) +7
-0+ /* + * not marking this node online because we do not want to + * confuse userspace by sysfs files/directories for node + * without any memory attached to it (see topology_init) + * The pgdat will get fully initialized when a memory is + * hotpluged into it by hotadd_init_pgdat + */Hmm... which following step would mark the node online? On x86, the node is onlined in
alloc_node_date(). This is not onlined here.Quoted 3 lines
The comment tries to explain that this happens during the memory hotplug. Or maybe I have missed your question?
I am not sure for others, while the comment confused me a little.
Currently in kernel, there are two situations for node online:
* during memory hotplug
* during sys-initFor memory hotplug, we allocate pgdat and online node. And current hot-add process has already put them in two steps:
1.
__try_online_node()
2.node_set_online()So emphasize "not online" node here, confuse me a little. It is a natural thing to not online node until it has memory.
But from another point of view, the comment here is reasonable. During sys-init, we online node at the same time when creating pgdat. And even for memory-less node on x86, we online them too.
Quoted 5 lines
Well, this is all about the comment. I have tried to grab may head but not come up with a better idea.
Or maybe this is just my personal feeling, don't bother if no-one else feel like this.
I shuffled the words a bit, maybe this sounds better not only to me :)
Quoted 8 lines
/*
* We do not want to confuse userspace by sysfs files/directories for node
* without any memory attached to it, so this node is not marked as
*N_MEMORYand not marked online so thattopology_init()won't create
* sysfs hierarchy for this node. The pgdat will get fully initialized by
*hotadd_init_pgdat()when memory is hotpluged into this node */
Note that the topology_init() part might change soon [1] so maybe we want to rephrase that to "so that no sysfs hierarchy will be created via register_one_node() for this node." right away.
or (because there is also /sys/devices/system/node/online )
"so that the node won't be indicated as online to user space, for example, via register_one_node()."
[1] https://lkml.kernel.org/r/[email protected]
Signature · 3 lines
Thanks, David / dhildenb
Michal Hocko
reviewerRe: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to Mike RapoportQuoted 48 lines · Mike Rapoport
Quoted 39 lines · Wei Yang
Quoted 14 lines · Michal Hocko
Quoted 12 lines · Wei Yang
On Thu, Jan 27, 2022 at 03:47:40PM +0100, Michal Hocko wrote:
[...]Quoted 7 lines
Quoted 7 lines
Review
Quoted 7 lines
(quoted diff) +7
-0+ /* + * not marking this node online because we do not want to + * confuse userspace by sysfs files/directories for node + * without any memory attached to it (see topology_init) + * The pgdat will get fully initialized when a memory is + * hotpluged into it by hotadd_init_pgdat + */Hmm... which following step would mark the node online? On x86, the node is onlined in
alloc_node_date(). This is not onlined here.Quoted 3 lines
The comment tries to explain that this happens during the memory hotplug. Or maybe I have missed your question?
I am not sure for others, while the comment confused me a little.
Currently in kernel, there are two situations for node online:
* during memory hotplug
* during sys-initFor memory hotplug, we allocate pgdat and online node. And current hot-add process has already put them in two steps:
1.
__try_online_node()
2.node_set_online()So emphasize "not online" node here, confuse me a little. It is a natural thing to not online node until it has memory.
But from another point of view, the comment here is reasonable. During sys-init, we online node at the same time when creating pgdat. And even for memory-less node on x86, we online them too.
Quoted 5 lines
Well, this is all about the comment. I have tried to grab may head but not come up with a better idea.
Or maybe this is just my personal feeling, don't bother if no-one else feel like this.
I shuffled the words a bit, maybe this sounds better not only to me :)
Quoted 7 lines
/*
* We do not want to confuse userspace by sysfs files/directories for node
* without any memory attached to it, so this node is not marked as
*N_MEMORYand not marked online so thattopology_init()won't create
* sysfs hierarchy for this node. The pgdat will get fully initialized by
*hotadd_init_pgdat()when memory is hotpluged into this node */
Yes, I like this comment more than mine. Andrew could you replace it in your tree or should I repost the patch?
Thanks!
Signature · 2 lines
Michal Hocko SUSE Labs
Mike Rapoport
reviewerRe: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to David HildenbrandQuoted 58 lines · David Hildenbrand
Quoted 48 lines · Mike Rapoport
Quoted 39 lines · Wei Yang
Quoted 14 lines · Michal Hocko
Quoted 12 lines · Wei Yang
On Thu, Jan 27, 2022 at 03:47:40PM +0100, Michal Hocko wrote:
[...]Quoted 7 lines
Quoted 7 lines
Review
Quoted 7 lines
(quoted diff) +7
-0+ /* + * not marking this node online because we do not want to + * confuse userspace by sysfs files/directories for node + * without any memory attached to it (see topology_init) + * The pgdat will get fully initialized when a memory is + * hotpluged into it by hotadd_init_pgdat + */Hmm... which following step would mark the node online? On x86, the node is onlined in
alloc_node_date(). This is not onlined here.Quoted 3 lines
The comment tries to explain that this happens during the memory hotplug. Or maybe I have missed your question?
I am not sure for others, while the comment confused me a little.
Currently in kernel, there are two situations for node online:
* during memory hotplug
* during sys-initFor memory hotplug, we allocate pgdat and online node. And current hot-add process has already put them in two steps:
1.
__try_online_node()
2.node_set_online()So emphasize "not online" node here, confuse me a little. It is a natural thing to not online node until it has memory.
But from another point of view, the comment here is reasonable. During sys-init, we online node at the same time when creating pgdat. And even for memory-less node on x86, we online them too.
Quoted 5 lines
Well, this is all about the comment. I have tried to grab may head but not come up with a better idea.
Or maybe this is just my personal feeling, don't bother if no-one else feel like this.
I shuffled the words a bit, maybe this sounds better not only to me :)
Quoted 8 lines
/*
* We do not want to confuse userspace by sysfs files/directories for node
* without any memory attached to it, so this node is not marked as
*N_MEMORYand not marked online so thattopology_init()won't create
* sysfs hierarchy for this node. The pgdat will get fully initialized by
*hotadd_init_pgdat()when memory is hotpluged into this node */
Quoted 3 lines
Note that the
topology_init()part might change soon [1] so maybe we want to rephrase that to "so that no sysfs hierarchy will be created viaregister_one_node()for this node." right away.
Heh, this will be your responsibility to update the comment here when you post non-RFC version ;-)
Quoted 12 lines
or (because there is also
/sys/devices/system/node/online)"so that the node won't be indicated as online to user space, for example, via
register_one_node()."[1] https://lkml.kernel.org/r/[email protected]
Signature · 3 lines
Thanks, David / dhildenb
Signature · 2 lines
Sincerely yours, Mike.
Re: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
Quoted 59 lines · Mike Rapoport
Quoted 58 lines · David Hildenbrand
Quoted 56 lines · Mike Rapoport
Quoted 39 lines · Wei Yang
Quoted 14 lines · Michal Hocko
Quoted 12 lines · Wei Yang
On Thu, Jan 27, 2022 at 03:47:40PM +0100, Michal Hocko wrote:
[...]Quoted 7 lines
Quoted 7 lines
Review
Quoted 7 lines
(quoted diff) +7
-0+ /* + * not marking this node online because we do not want to + * confuse userspace by sysfs files/directories for node + * without any memory attached to it (see topology_init) + * The pgdat will get fully initialized when a memory is + * hotpluged into it by hotadd_init_pgdat + */Hmm... which following step would mark the node online? On x86, the node is onlined in
alloc_node_date(). This is not onlined here.Quoted 3 lines
The comment tries to explain that this happens during the memory hotplug. Or maybe I have missed your question?
I am not sure for others, while the comment confused me a little.
Currently in kernel, there are two situations for node online:
* during memory hotplug
* during sys-initFor memory hotplug, we allocate pgdat and online node. And current hot-add process has already put them in two steps:
1.
__try_online_node()
2.node_set_online()So emphasize "not online" node here, confuse me a little. It is a natural thing to not online node until it has memory.
But from another point of view, the comment here is reasonable. During sys-init, we online node at the same time when creating pgdat. And even for memory-less node on x86, we online them too.
Quoted 5 lines
Well, this is all about the comment. I have tried to grab may head but not come up with a better idea.
Or maybe this is just my personal feeling, don't bother if no-one else feel like this.
I shuffled the words a bit, maybe this sounds better not only to me :)
/*
* We do not want to confuse userspace by sysfs files/directories for node
* without any memory attached to it, so this node is not marked as
*N_MEMORYand not marked online so thattopology_init()won't create
* sysfs hierarchy for this node. The pgdat will get fully initialized by
*hotadd_init_pgdat()when memory is hotpluged into this node */
Quoted 6 lines
Quoted 3 lines
Note that the
topology_init()part might change soon [1] so maybe we want to rephrase that to "so that no sysfs hierarchy will be created viaregister_one_node()for this node." right away.Heh, this will be your responsibility to update the comment here when you post non-RFC version ;-)
I'm usually sending patches against Linus' tree. And I'll post non-RFC most probably today (after testing on aarch64) ;)
So I'd appreciate if we could just phrase it more generically, as I tried. But of course, we can try making my life harder ;)
Signature · 3 lines
Thanks, David / dhildenb
Michal Hocko
reviewerRe: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
So we will go with the following.
/*
* We do not want to confuse userspace by sysfs files/directories for node
* without any memory attached to it, so this node is not marked as
* N_MEMORY and not marked online so that no sysfs hierarchy will be
* created via register_one_node for this node. The pgdat will get fully
* initialized by hotadd_init_pgdat() when memory is hotpluged into this
* node
*/
OK?
Signature · 2 lines
Michal Hocko SUSE Labs
Re: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
Quoted 11 lines · Michal Hocko
So we will go with the following.
/*
* We do not want to confuse userspace by sysfs files/directories for node
* without any memory attached to it, so this node is not marked as
*N_MEMORYand not marked online so that no sysfs hierarchy will be
* created viaregister_one_nodefor this node. The pgdat will get fully
* initialized byhotadd_init_pgdat()when memory is hotpluged into this
* node
*/OK?
LGTM, thanks!
Signature · 3 lines
Thanks, David / dhildenb
Andrew Morton
maintainerRe: [PATCH 2/6] mm: handle uninitialized numa nodes gracefully
In reply to Michal HockoQuoted 10 lines · On Thu, 3 Feb 2022 09:39:32 +0100 Michal Hocko <[email protected]> wrote:
Quoted 9 lines
I shuffled the words a bit, maybe this sounds better not only to me :)
/*
* We do not want to confuse userspace by sysfs files/directories for node
* without any memory attached to it, so this node is not marked as
*N_MEMORYand not marked online so thattopology_init()won't create
* sysfs hierarchy for this node. The pgdat will get fully initialized by
*hotadd_init_pgdat()when memory is hotpluged into this node */
Quoted 2 lines
Yes, I like this comment more than mine. Andrew could you replace it in your tree or should I repost the patch?
I did this.
mm/page_alloc.c renamed from mm/page_alloc.c~mm-handle-uninitialized-numa-nodes-gracefully-fix +9 -5
@@ -8155,12 +8155,16 @@ void __init free_area_init(unsigned long }arch_refresh_nodedata (nid, pgdat);free_area_init_memoryless_node (nid); +/* -* not marking this node online because we do not want to -* confuse userspace by sysfs files/directories for node -* without any memory attached to it (see topology_init) -* The pgdat will get fully initialized when a memory is -* hotpluged into it by hotadd_init_pgdat +* We do not want to confuse userspace by sysfs +* files/directories for node without any memory +* attached to it, so this node is not marked as +* N_MEMORY and not marked online so that no sysfs +* hierarchy will be created via register_one_node for +* it. The pgdat will get fully initialized by +* hotadd_init_pgdat() when memory is hotplugged into +* this node. */ continue ; }
_
Oscar Salvador
maintainerRe: [PATCH 4/6] mm, memory_hotplug: reorganize new pgdat initialization
In reply to Michal HockoQuoted 11 lines · Michal Hocko
Quoted 7 lines
I feel like we should be initializing all of this only once, just after allocating the node. There should be no difference between a node we're reusing and a "fresh" node. IOW, memory offlining should be resetting all state accordingly when the last memory goes away.
But I might be wrong and this patch looks like an improvement, as you say, without functional change
Yeah, I really wanted to have this simple and straightforward. To be completely honest I am not even sure this is necessary. Something really woth looking at.
Seizing the opportunity that I had to look at this code and at x86's numa init code again I am preparing something to further sort this out and simplify it a bit.
So I will have something soon unless someone beats me to it first.
Signature · 2 lines
Oscar Salvador SUSE Labs