x86, numa: always initialize all possible nodes
Subsystems (4)
cover x86, numa: always initialize all possible nodes
Hi,
this has been posted as an RFC previously [1]. There didn't seem to be any objections so I am reposting this for inclusion. I have added a debugging patch which prints the zonelist setup for each numa node for an easier debugging of a broken zonelist setup.
PATCH 1/2 x86, numa: always initialize all possible nodes
From: Michal Hocko <[email protected]>
Pingfan Liu has reported the following splat
[ 5.772742] BUG: unable to handle kernel paging request at 0000000000002088 [ 5.773618] PGD 0 P4D 0 [ 5.773618] Oops: 0000 [#1] SMP NOPTI [ 5.773618] CPU: 2 PID: 1 Comm: swapper/0 Not tainted 4.20.0-rc1+ #3 [ 5.773618] Hardware name: Dell Inc. PowerEdge R7425/02MJ3T, BIOS 1.4.3 06/29/2018 [ 5.773618] RIP: 0010:__alloc_pages_nodemask+0xe2/0x2a0 [ 5.773618] Code: 00 00 44 89 ea 80 ca 80 41 83 f8 01 44 0f 44 ea 89 da c1 ea 08 83 e2 01 88 54 24 20 48 8b 54 24 08 48 85 d2 0f 85 46 01 00 00 <3b> 77 08 0f 82 3d 01 00 00 48 89 f8 44 89 ea 48 89 e1 44 89 e6 89 [ 5.773618] RSP: 0018:ffffaa600005fb20 EFLAGS: 00010246 [ 5.773618] RAX: 0000000000000000 RBX: 00000000006012c0 RCX: 0000000000000000 [ 5.773618] RDX: 0000000000000000 RSI: 0000000000000002 RDI: 0000000000002080 [ 5.773618] RBP: 00000000006012c0 R08: 0000000000000000 R09: 0000000000000002 [ 5.773618] R10: 00000000006080c0 R11: 0000000000000002 R12: 0000000000000000 [ 5.773618] R13: 0000000000000001 R14: 0000000000000000 R15: 0000000000000002 [ 5.773618] FS: 0000000000000000(0000) GS:ffff8c69afe00000(0000) knlGS:0000000000000000 [ 5.773618] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 5.773618] CR2: 0000000000002088 CR3: 000000087e00a000 CR4: 00000000003406e0 [ 5.773618] Call Trace: [ 5.773618] new_slab+0xa9/0x570 [ 5.773618] ___slab_alloc+0x375/0x540 [ 5.773618] ? pinctrl_bind_pins+0x2b/0x2a0 [ 5.773618] __slab_alloc+0x1c/0x38 [ 5.773618] __kmalloc_node_track_caller+0xc8/0x270 [ 5.773618] ? pinctrl_bind_pins+0x2b/0x2a0 [ 5.773618] devm_kmalloc+0x28/0x60 [ 5.773618] pinctrl_bind_pins+0x2b/0x2a0 [ 5.773618] really_probe+0x73/0x420 [ 5.773618] driver_probe_device+0x115/0x130 [ 5.773618] __driver_attach+0x103/0x110 [ 5.773618] ? driver_probe_device+0x130/0x130 [ 5.773618] bus_for_each_dev+0x67/0xc0 [ 5.773618] ? klist_add_tail+0x3b/0x70 [ 5.773618] bus_add_driver+0x41/0x260 [ 5.773618] ? pcie_port_setup+0x4d/0x4d [ 5.773618] driver_register+0x5b/0xe0 [ 5.773618] ? pcie_port_setup+0x4d/0x4d [ 5.773618] do_one_initcall+0x4e/0x1d4 [ 5.773618] ? init_setup+0x25/0x28 [ 5.773618] kernel_init_freeable+0x1c1/0x26e [ 5.773618] ? loglevel+0x5b/0x5b [ 5.773618] ? rest_init+0xb0/0xb0 [ 5.773618] kernel_init+0xa/0x110 [ 5.773618] ret_from_fork+0x22/0x40 [ 5.773618] Modules linked in: [ 5.773618] CR2: 0000000000002088 [ 5.773618] ---[ end trace 1030c9120a03d081 ]---
with his AMD machine with the following topology
NUMA node0 CPU(s): 0,8,16,24 NUMA node1 CPU(s): 2,10,18,26 NUMA node2 CPU(s): 4,12,20,28 NUMA node3 CPU(s): 6,14,22,30 NUMA node4 CPU(s): 1,9,17,25 NUMA node5 CPU(s): 3,11,19,27 NUMA node6 CPU(s): 5,13,21,29 NUMA node7 CPU(s): 7,15,23,31
[ 0.007418] Early memory node ranges [ 0.007419] node 1: [mem 0x0000000000001000-0x000000000008efff] [ 0.007420] node 1: [mem 0x0000000000090000-0x000000000009ffff] [ 0.007422] node 1: [mem 0x0000000000100000-0x000000005c3d6fff] [ 0.007422] node 1: [mem 0x00000000643df000-0x0000000068ff7fff] [ 0.007423] node 1: [mem 0x000000006c528000-0x000000006fffffff] [ 0.007424] node 1: [mem 0x0000000100000000-0x000000047fffffff] [ 0.007425] node 5: [mem 0x0000000480000000-0x000000087effffff]
and nr_cpus set to 4. The underlying reason is tha the device is bound to node 2 which doesn't have any memory and init_cpu_to_node only initializes memory-less nodes for possible cpus which nr_cpus restrics. This in turn means that proper zonelists are not allocated and the page allocator blows up.
Fix the issue by reworking how x86 initializes the memory less nodes. The current implementation is hacked into the workflow and it doesn't allow any flexibility. There is init_memory_less_node called for each offline node that has a CPU as already mentioned above. This will make sure that we will have a new online node without any memory. Much later on we build a zone list for this node and things seem to work, except they do not (e.g. due to nr_cpus). Not to mention that it doesn't really make much sense to consider an empty node as online because we just consider this node whenever we want to iterate nodes to use and empty node is obviously not the best candidate. This is all just too fragile.
The new code relies on the arch specific initialization to allocate all possible NUMA nodes (including memory less) - numa_register_memblks in this case. Generic code then initializes both zonelists (__build_all_zonelists) and allocator internals (free_area_init_nodes) for all non-null pgdats rather than online ones.
For the x86 specific part also do not make new node online in alloc_node_data because this is too early to know that. numa_register_memblks knows that a node has some memory so it can make the node online appropriately. init_memory_less_node hack can be safely removed altogether now.
--- arch/x86/mm/numa.c | 27 +++------------------------ mm/page_alloc.c | 15 +++++++++------ 2 files changed, 12 insertions(+), 30 deletions(-)
arch/x86/mm/numa.c +3 -24
@@ -216,8 +216,6 @@ static void __init alloc_node_data(int nid) node_data[nid] = nd;memset (NODE_DATA (nid),0 ,sizeof (pg_data_t)); - -node_set_online (nid); }/**
@@ -570,7 +568,7 @@ static int __init numa_register_memblks(struct numa_meminfo *mi)return -EINVAL ;/* Finally register nodes. */ -for_each_node_mask (nid, node_possible_map) { +for_each_node_mask (nid, numa_nodes_parsed) {u64 start =PFN_PHYS (max_pfn);u64 end =0 ;
@@ -581,9 +579,6 @@ static int __init numa_register_memblks(struct numa_meminfo *mi) end =max (mi->blk[i].end, end); } -if (start >= end) -continue ; -/* * Don't confuse VM with a node that doesn't have the * minimum amount of memory:
@@ -592,6 +587,8 @@ static int __init numa_register_memblks(struct numa_meminfo *mi)continue ;alloc_node_data (nid); +if (end) +node_set_online (nid); }/* Dump memblock with node info and return. */
@@ -721,21 +718,6 @@ void __init x86_numa_init(void)numa_init (dummy_numa_init); } -static void __init init_memory_less_node (int nid) -{ -unsigned long zones_size[MAX_NR_ZONES ] = {0 }; -unsigned long zholes_size[MAX_NR_ZONES ] = {0 }; - -/* Allocate and initialize node data. Memory-less node is now online.*/ -alloc_node_data (nid); -free_area_init_node (nid, zones_size,0 , zholes_size); - -/* -* All zonelists will be built later in start_kernel() after per cpu -* areas are initialized. -*/ -} -/* * Setup early cpu_to_node. *
@@ -763,9 +745,6 @@ void __init init_cpu_to_node(void)if (node =NUMA_NO_NODE )continue ; -if (!node_online (node)) -init_memory_less_node (node); -numa_set_node (cpu, node); } }
mm/page_alloc.c +9 -6
@@ -5361,10 +5361,11 @@ static void __build_all_zonelists(void *data)if (self && !node_online (self->node_id)) {build_zonelists (self); }else { -for_each_online_node (nid) { +for_each_node (nid) {pg_data_t *pgdat =NODE_DATA (nid); -build_zonelists (pgdat); +if (pgdat) +build_zonelists (pgdat); }#ifdef CONFIG_HAVE_MEMORYLESS_NODES
@@ -6644,10 +6645,8 @@ static unsigned long __init find_min_pfn_for_node(int nid)for_each_mem_pfn_range (i, nid, &start_pfn,NULL ,NULL ) min_pfn =min (min_pfn, start_pfn); -if (min_pfn =ULONG_MAX ) { -pr_warn ("Could not find start_pfn for node %d\n" , nid); +if (min_pfn =ULONG_MAX )return 0 ; - }return min_pfn; }
@@ -6991,8 +6990,12 @@ void __init free_area_init_nodes(unsigned long *max_zone_pfn)mminit_verify_pageflags_layout ();setup_nr_node_ids ();zero_resv_unavail (); -for_each_online_node (nid) { +for_each_node (nid) {pg_data_t *pgdat =NODE_DATA (nid); + +if (!pgdat) +continue ; +free_area_init_node (nid,NULL ,find_min_pfn_for_node (nid),NULL );
Signature · 1 line
2.20.1
PATCH 2/2 mm: be more verbose about zonelist initialization
In reply to Michal HockoFrom: Michal Hocko <[email protected]>
We have seen several bugs where zonelists have not been initialized properly and it is not really straightforward to track those bugs down. One way to help a bit at least is to dump zonelists of each node when they are (re)initialized.
--- mm/page_alloc.c | 5 +++++ 1 file changed, 5 insertions(+)
mm/page_alloc.c +5 -0
@@ -5259,6 +5259,11 @@ static void build_zonelists(pg_data_t *pgdat)build_zonelists_in_node_order (pgdat, node_order, nr_nodes);build_thisnode_zonelists (pgdat); + +pr_info ("node[%d] zonelist: " , pgdat->node_id); +for_each_zone_zonelist (zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK ],MAX_NR_ZONES -1 ) +pr_cont ("%d:%s " ,zone_to_nid (zone), zone->name); +pr_cont ("\n" ); }#ifdef CONFIG_HAVE_MEMORYLESS_NODES
Signature · 1 line
2.20.1
Quoted 7 lines · Michal Hocko
Hi,
this has been posted as an RFC previously [1]. There didn't seem to be any objections so I am reposting this for inclusion. I have added a debugging patch which prints the zonelist setup for each numa node for an easier debugging of a broken zonelist setup.
FWIW,
for the series.
Signature · 2 lines
Sincerely yours, Mike.
kbuild test robot
build botRe: [PATCH 2/2] mm: be more verbose about zonelist initialization
In reply to Michal HockoI love your patch! Yet something to improve:
Hi Michal,
I love your patch! Yet something to improve:
[auto build test ERROR on linus/master]
[also build test ERROR on v5.0-rc4 next-20190212]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
url: https://github.com/0day-ci/linux/commits/Michal-Hocko/x86-numa-always-initialize-all-possible-nodes/20190213-071628
config: x86_64-randconfig-x016-201906 (attached as .config)
compiler: gcc-8 (Debian 8.2.0-20) 8.2.0
reproduce:
# save the attached .config to linux build tree
make ARCH=x86_64 All errors (new ones prefixed by >>):
In file included from include/linux/gfp.h:6,
from include/linux/mm.h:10,
from mm/page_alloc.c:18:
mm/page_alloc.c: In function 'build_zonelists':
>> mm/page_alloc.c:5423:31: error: 'z' undeclared (first use in this function)
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^
include/linux/mmzone.h:1036:7: note: in definition of macro 'for_each_zone_zonelist_nodemask'
for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
^
mm/page_alloc.c:5423:2: note: in expansion of macro 'for_each_zone_zonelist'
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~~~~~~~~~~~~~~~~~~~
mm/page_alloc.c:5423:31: note: each undeclared identifier is reported only once for each function it appears in
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^
include/linux/mmzone.h:1036:7: note: in definition of macro 'for_each_zone_zonelist_nodemask'
for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
^
mm/page_alloc.c:5423:2: note: in expansion of macro 'for_each_zone_zonelist'
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~~~~~~~~~~~~~~~~~~~
>> mm/page_alloc.c:5423:25: error: 'zone' undeclared (first use in this function)
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~
include/linux/mmzone.h:1036:59: note: in definition of macro 'for_each_zone_zonelist_nodemask'
for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
^~~~
mm/page_alloc.c:5423:2: note: in expansion of macro 'for_each_zone_zonelist'
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~~~~~~~~~~~~~~~~~~~
include/linux/mmzone.h:1036:57: warning: left-hand operand of comma expression has no effect [-Wunused-value]
for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
^
include/linux/mmzone.h:1058:2: note: in expansion of macro 'for_each_zone_zonelist_nodemask'
for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
mm/page_alloc.c:5423:2: note: in expansion of macro 'for_each_zone_zonelist'
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~~~~~~~~~~~~~~~~~~~
include/linux/mmzone.h:1038:50: warning: left-hand operand of comma expression has no effect [-Wunused-value]
z = next_zones_zonelist(++z, highidx, nodemask), \
^
include/linux/mmzone.h:1058:2: note: in expansion of macro 'for_each_zone_zonelist_nodemask'
for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
mm/page_alloc.c:5423:2: note: in expansion of macro 'for_each_zone_zonelist'
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~~~~~~~~~~~~~~~~~~~vim +/z +5423 mm/page_alloc.c
5382
5383 /*
5384 * Build zonelists ordered by zone and nodes within zones.
5385 * This results in conserving DMA zone[s] until all Normal memory is
5386 * exhausted, but results in overflowing to remote node while memory
5387 * may still exist in local DMA zone.
5388 */
5389
5390 static void build_zonelists(pg_data_t *pgdat)
5391 {
5392 static int node_order[MAX_NUMNODES];
5393 int node, load, nr_nodes = 0;
5394 nodemask_t used_mask;
5395 int local_node, prev_node;
5396
5397 /* NUMA-aware ordering of nodes */
5398 local_node = pgdat->node_id;
5399 load = nr_online_nodes;
5400 prev_node = local_node;
5401 nodes_clear(used_mask);
5402
5403 memset(node_order, 0, sizeof(node_order));
5404 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
5405 /*
5406 * We don't want to pressure a particular node.
5407 * So adding penalty to the first node in same
5408 * distance group to make it round-robin.
5409 */
5410 if (node_distance(local_node, node) !=
5411 node_distance(local_node, prev_node))
5412 node_load[node] = load;
5413
5414 node_order[nr_nodes++] = node;
5415 prev_node = node;
5416 load--;
5417 }
5418
5419 build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
5420 build_thisnode_zonelists(pgdat);
5421
5422 pr_info("node[%d] zonelist: ", pgdat->node_id);
> 5423 for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
5424 pr_cont("%d:%s ", zone_to_nid(zone), zone->name);
5425 pr_cont("\n");
5426 }
5427
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
kbuild test robot
build botRe: [PATCH 2/2] mm: be more verbose about zonelist initialization
In reply to Michal HockoI love your patch! Perhaps something to improve:
Hi Michal,
I love your patch! Perhaps something to improve:
[auto build test WARNING on linus/master]
[also build test WARNING on v5.0-rc4 next-20190212]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system]
url: https://github.com/0day-ci/linux/commits/Michal-Hocko/x86-numa-always-initialize-all-possible-nodes/20190213-071628
config: x86_64-kexec (attached as .config)
compiler: gcc-7 (Debian 7.3.0-1) 7.3.0 reproduce:
# save the attached .config to linux build tree
make ARCH=x86_64 All warnings (new ones prefixed by >>):
In file included from include/linux/gfp.h:6:0,
from include/linux/mm.h:10,
from mm/page_alloc.c:18:
mm/page_alloc.c: In function 'build_zonelists':
mm/page_alloc.c:5423:31: error: 'z' undeclared (first use in this function)
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^
include/linux/mmzone.h:1036:7: note: in definition of macro 'for_each_zone_zonelist_nodemask'
for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
^
>> mm/page_alloc.c:5423:2: note: in expansion of macro 'for_each_zone_zonelist'
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~~~~~~~~~~~~~~~~~~~
mm/page_alloc.c:5423:31: note: each undeclared identifier is reported only once for each function it appears in
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^
include/linux/mmzone.h:1036:7: note: in definition of macro 'for_each_zone_zonelist_nodemask'
for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
^
>> mm/page_alloc.c:5423:2: note: in expansion of macro 'for_each_zone_zonelist'
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~~~~~~~~~~~~~~~~~~~
mm/page_alloc.c:5423:25: error: 'zone' undeclared (first use in this function)
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^
include/linux/mmzone.h:1036:59: note: in definition of macro 'for_each_zone_zonelist_nodemask'
for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
^~~~
>> mm/page_alloc.c:5423:2: note: in expansion of macro 'for_each_zone_zonelist'
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~~~~~~~~~~~~~~~~~~~
include/linux/mmzone.h:1036:57: warning: left-hand operand of comma expression has no effect [-Wunused-value]
for (z = first_zones_zonelist(zlist, highidx, nodemask), zone = zonelist_zone(z); \
^
>> include/linux/mmzone.h:1058:2: note: in expansion of macro 'for_each_zone_zonelist_nodemask'
for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> mm/page_alloc.c:5423:2: note: in expansion of macro 'for_each_zone_zonelist'
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~~~~~~~~~~~~~~~~~~~
include/linux/mmzone.h:1038:50: warning: left-hand operand of comma expression has no effect [-Wunused-value]
z = next_zones_zonelist(++z, highidx, nodemask), \
^
>> include/linux/mmzone.h:1058:2: note: in expansion of macro 'for_each_zone_zonelist_nodemask'
for_each_zone_zonelist_nodemask(zone, z, zlist, highidx, NULL)
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
>> mm/page_alloc.c:5423:2: note: in expansion of macro 'for_each_zone_zonelist'
for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
^~~~~~~~~~~~~~~~~~~~~~vim +/for_each_zone_zonelist +5423 mm/page_alloc.c
5382
5383 /*
5384 * Build zonelists ordered by zone and nodes within zones.
5385 * This results in conserving DMA zone[s] until all Normal memory is
5386 * exhausted, but results in overflowing to remote node while memory
5387 * may still exist in local DMA zone.
5388 */
5389
5390 static void build_zonelists(pg_data_t *pgdat)
5391 {
5392 static int node_order[MAX_NUMNODES];
5393 int node, load, nr_nodes = 0;
5394 nodemask_t used_mask;
5395 int local_node, prev_node;
5396
5397 /* NUMA-aware ordering of nodes */
5398 local_node = pgdat->node_id;
5399 load = nr_online_nodes;
5400 prev_node = local_node;
5401 nodes_clear(used_mask);
5402
5403 memset(node_order, 0, sizeof(node_order));
5404 while ((node = find_next_best_node(local_node, &used_mask)) >= 0) {
5405 /*
5406 * We don't want to pressure a particular node.
5407 * So adding penalty to the first node in same
5408 * distance group to make it round-robin.
5409 */
5410 if (node_distance(local_node, node) !=
5411 node_distance(local_node, prev_node))
5412 node_load[node] = load;
5413
5414 node_order[nr_nodes++] = node;
5415 prev_node = node;
5416 load--;
5417 }
5418
5419 build_zonelists_in_node_order(pgdat, node_order, nr_nodes);
5420 build_thisnode_zonelists(pgdat);
5421
5422 pr_info("node[%d] zonelist: ", pgdat->node_id);
> 5423 for_each_zone_zonelist(zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK], MAX_NR_ZONES-1)
5424 pr_cont("%d:%s ", zone_to_nid(zone), zone->name);
5425 pr_cont("\n");
5426 }
5427
---
0-DAY kernel test infrastructure Open Source Technology Center
https://lists.01.org/pipermail/kbuild-all Intel Corporation
PATCH v2 2/2 mm: be more verbose about zonelist initialization
In reply to Michal HockoFrom: Michal Hocko <[email protected]>
We have seen several bugs where zonelists have not been initialized properly and it is not really straightforward to track those bugs down. One way to help a bit at least is to dump zonelists of each node when they are (re)initialized.
--- mm/page_alloc.c | 6 ++++++ 1 file changed, 6 insertions(+)
mm/page_alloc.c +6 -0
@@ -5234,6 +5234,7 @@ static void build_zonelists(pg_data_t *pgdat)int node, load, nr_nodes =0 ;nodemask_t used_mask;int local_node, prev_node; +struct zone *zone;/* NUMA-aware ordering of nodes */ local_node = pgdat->node_id;
@@ -5259,6 +5260,11 @@ static void build_zonelists(pg_data_t *pgdat)build_zonelists_in_node_order (pgdat, node_order, nr_nodes);build_thisnode_zonelists (pgdat); + +pr_info ("node[%d] zonelist: " , pgdat->node_id); +for_each_zone_zonelist (zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK ],MAX_NR_ZONES -1 ) +pr_cont ("%d:%s " ,zone_to_nid (zone), zone->name); +pr_cont ("\n" ); }#ifdef CONFIG_HAVE_MEMORYLESS_NODES
Signature · 1 line
2.20.1
PATCH v3 2/2 mm: be more verbose about zonelist initialization
In reply to Michal HockoFrom: Michal Hocko <[email protected]>
We have seen several bugs where zonelists have not been initialized properly and it is not really straightforward to track those bugs down. One way to help a bit at least is to dump zonelists of each node when they are (re)initialized.
---
Sorry for spamming. I have screwed up ammending the previous version.
mm/page_alloc.c | 7 +++++++ 1 file changed, 7 insertions(+)
mm/page_alloc.c +7 -0
@@ -5234,6 +5234,8 @@ static void build_zonelists(pg_data_t *pgdat)int node, load, nr_nodes =0 ;nodemask_t used_mask;int local_node, prev_node; +struct zone *zone; +struct zoneref *z;/* NUMA-aware ordering of nodes */ local_node = pgdat->node_id;
@@ -5259,6 +5261,11 @@ static void build_zonelists(pg_data_t *pgdat)build_zonelists_in_node_order (pgdat, node_order, nr_nodes);build_thisnode_zonelists (pgdat); + +pr_info ("node[%d] zonelist: " , pgdat->node_id); +for_each_zone_zonelist (zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK ],MAX_NR_ZONES -1 ) +pr_cont ("%d:%s " ,zone_to_nid (zone), zone->name); +pr_cont ("\n" ); }#ifdef CONFIG_HAVE_MEMORYLESS_NODES
Signature · 1 line
2.20.1
Peter Zijlstra
maintainerRe: [PATCH v3 2/2] mm: be more verbose about zonelist initialization
mm/page_alloc.c static void build_zonelists(pg_data_t *pgdat) · line 5269
Quoted 10 lines · Michal Hocko
mm/page_alloc.c +5
-0@@ -5259,6 +5261,11 @@ static void build_zonelists(pg_data_t *pgdat)build_zonelists_in_node_order (pgdat, node_order, nr_nodes);build_thisnode_zonelists (pgdat); + +pr_info ("node[%d] zonelist: " , pgdat->node_id); +for_each_zone_zonelist (zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK ],MAX_NR_ZONES -1 ) +pr_cont ("%d:%s " ,zone_to_nid (zone), zone->name); +pr_cont ("\n" ); }
Have you ran this by the SGI and other stupid large machine vendors? Traditionally they tend to want to remove such things instead of adding them.
Re: [PATCH v3 2/2] mm: be more verbose about zonelist initialization
Quoted 13 lines · Peter Zijlstra
mm/page_alloc.c static void build_zonelists(pg_data_t *pgdat) · line 5269
Quoted 10 lines · Michal Hocko
mm/page_alloc.c +5
-0@@ -5259,6 +5261,11 @@ static void build_zonelists(pg_data_t *pgdat)build_zonelists_in_node_order (pgdat, node_order, nr_nodes);build_thisnode_zonelists (pgdat); + +pr_info ("node[%d] zonelist: " , pgdat->node_id); +for_each_zone_zonelist (zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK ],MAX_NR_ZONES -1 ) +pr_cont ("%d:%s " ,zone_to_nid (zone), zone->name); +pr_cont ("\n" ); }Have you ran this by the SGI and other stupid large machine vendors?
I do not have such a large machine handy. The biggest I have has handfull (say dozen) of NUMA nodes.
Quoted 2 lines
Traditionally they tend to want to remove such things instead of adding them.
I do not insist on this patch but I find it handy. If there is an opposition I will not miss it much.
Signature · 2 lines
Michal Hocko SUSE Labs
Peter Zijlstra
maintainerRe: [PATCH v3 2/2] mm: be more verbose about zonelist initialization
Quoted 15 lines · Michal Hocko
Quoted 13 lines · Peter Zijlstra
mm/page_alloc.c static void build_zonelists(pg_data_t *pgdat) · line 5269
Quoted 10 lines · Michal Hocko
mm/page_alloc.c +5
-0@@ -5259,6 +5261,11 @@ static void build_zonelists(pg_data_t *pgdat)build_zonelists_in_node_order (pgdat, node_order, nr_nodes);build_thisnode_zonelists (pgdat); + +pr_info ("node[%d] zonelist: " , pgdat->node_id); +for_each_zone_zonelist (zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK ],MAX_NR_ZONES -1 ) +pr_cont ("%d:%s " ,zone_to_nid (zone), zone->name); +pr_cont ("\n" ); }Have you ran this by the SGI and other stupid large machine vendors?
Quoted 8 lines
I do not have such a large machine handy. The biggest I have has handfull (say dozen) of NUMA nodes.
Quoted 2 lines
Traditionally they tend to want to remove such things instead of adding them.
I do not insist on this patch but I find it handy. If there is an opposition I will not miss it much.
Well, I don't have machines like that either and don't mind the patch. Just raising the issue; I've had the big iron boys complain about similar things (typically printing something for every CPU, which gets out of hand much faster than zones, but still).
Re: [PATCH v3 2/2] mm: be more verbose about zonelist initialization
Quoted 22 lines · Peter Zijlstra
Quoted 21 lines · Michal Hocko
Quoted 13 lines · Peter Zijlstra
mm/page_alloc.c static void build_zonelists(pg_data_t *pgdat) · line 5269
Quoted 10 lines · Michal Hocko
mm/page_alloc.c +5
-0@@ -5259,6 +5261,11 @@ static void build_zonelists(pg_data_t *pgdat)build_zonelists_in_node_order (pgdat, node_order, nr_nodes);build_thisnode_zonelists (pgdat); + +pr_info ("node[%d] zonelist: " , pgdat->node_id); +for_each_zone_zonelist (zone, z, &pgdat->node_zonelists[ZONELIST_FALLBACK ],MAX_NR_ZONES -1 ) +pr_cont ("%d:%s " ,zone_to_nid (zone), zone->name); +pr_cont ("\n" ); }Have you ran this by the SGI and other stupid large machine vendors?
I do not have such a large machine handy. The biggest I have has handfull (say dozen) of NUMA nodes.
Quoted 2 lines
Traditionally they tend to want to remove such things instead of adding them.
Quoted 7 lines
Quoted 2 lines
I do not insist on this patch but I find it handy. If there is an opposition I will not miss it much.
Well, I don't have machines like that either and don't mind the patch. Just raising the issue; I've had the big iron boys complain about similar things (typically printing something for every CPU, which gets out of hand much faster than zones, but still).
Maybe we can try to push this through and revert if somebody complains about an excessive output.
Signature · 2 lines
Michal Hocko SUSE Labs
Re: [PATCH v3 2/2] mm: be more verbose about zonelist initialization
In reply to Michal HockoQuoted 5 lines · Michal Hocko
We have seen several bugs where zonelists have not been initialized properly and it is not really straightforward to track those bugs down. One way to help a bit at least is to dump zonelists of each node when they are (re)initialized.
Were you thinking of boot-time bugs and crashes, or just stuff going wonky after boot?
We don't have the zonelists dumped in /proc anywhere, do we? Would that help?
Re: [PATCH v3 2/2] mm: be more verbose about zonelist initialization
Quoted 9 lines · Dave Hansen
Quoted 5 lines · Michal Hocko
We have seen several bugs where zonelists have not been initialized properly and it is not really straightforward to track those bugs down. One way to help a bit at least is to dump zonelists of each node when they are (re)initialized.
Were you thinking of boot-time bugs and crashes, or just stuff going wonky after boot?
Mostly boot time. I haven't seen hotplug related bugs in this direction. All the issues I have seen so far is that we forget a node altogether and it ends up with no zonelists at all. But who knows maybe we have some hidden bugs where zonelists is initialized only partially for some reason and there is no real way to find out.
Quoted 2 lines
We don't have the zonelists dumped in /proc anywhere, do we? Would that help?
I would prefer to not export such an implementation detail into proc
Signature · 2 lines
Michal Hocko SUSE Labs
Quoted 7 lines · Michal Hocko
Hi,
this has been posted as an RFC previously [1]. There didn't seem to be any objections so I am reposting this for inclusion. I have added a debugging patch which prints the zonelist setup for each numa node for an easier debugging of a broken zonelist setup.
Friendly ping. I haven't heard any complains so can we route this via tip/x86/mm or should we go via mmotm.
Signature · 2 lines
Michal Hocko SUSE Labs
Quoted 11 lines · Michal Hocko
Quoted 7 lines · Michal Hocko
Hi,
this has been posted as an RFC previously [1]. There didn't seem to be any objections so I am reposting this for inclusion. I have added a debugging patch which prints the zonelist setup for each numa node for an easier debugging of a broken zonelist setup.Friendly ping. I haven't heard any complains so can we route this via tip/x86/mm or should we go via mmotm.
It seems that Dave is busy. Let's add Andrew. Can we get this [1] merged finally, please?
[1] http://lkml.kernel.org/r/[email protected]
Signature · 2 lines
Michal Hocko SUSE Labs
Dave Hansen
Reviewed-byQuoted 4 lines · Michal Hocko
Quoted 2 lines
Friendly ping. I haven't heard any complains so can we route this via tip/x86/mm or should we go via mmotm.
It seems that Dave is busy. Let's add Andrew. Can we get this [1] merged finally, please?
Sorry these slipped through the cracks.
These look sane to me. Because it pokes around mm/page_alloc.c a bit, and could impact other architectures, my preference would be for Andrew to pick these up for -mm. But, I don't feel that strongly about it.
Forgot to cc Andrew. Now for real.
Andrew please note that Dave has reviewed the patch
http://lkml.kernel.org/r/[email protected]
Or do you want me to resubmit?
Quoted 17 lines · Michal Hocko
Quoted 11 lines · Michal Hocko
Quoted 7 lines · Michal Hocko
Hi,
this has been posted as an RFC previously [1]. There didn't seem to be any objections so I am reposting this for inclusion. I have added a debugging patch which prints the zonelist setup for each numa node for an easier debugging of a broken zonelist setup.Friendly ping. I haven't heard any complains so can we route this via tip/x86/mm or should we go via mmotm.
It seems that Dave is busy. Let's add Andrew. Can we get this [1] merged finally, please?
Signature · 2 lines
Michal Hocko SUSE Labs
Hi -
This patch triggered an oops for me (more below).
On 2/12/19 4:53 AM, Michal Hocko wrote:
[snip]
Quoted 10 lines
Fix the issue by reworking how x86 initializes the memory less nodes. The current implementation is hacked into the workflow and it doesn't allow any flexibility. There is
init_memory_less_nodecalled for each offline node that has a CPU as already mentioned above. This will make sure that we will have a new online node without any memory. Much later on we build a zone list for this node and things seem to work, except they do not (e.g. due tonr_cpus). Not to mention that it doesn't really make much sense to consider an empty node as online because we just consider this node whenever we want to iterate nodes to use and empty node is obviously not the best candidate. This is all just too fragile.
The problem might be in here - I have a case with a 'memoryless' node that has CPUs that get onlined during SMP boot, but that onlining triggers a page fault during device registration.
I'm running on a NUMA machine but I marked all of the memory on node 1 as type 12 (PRAM), using the memmap arg. That makes node 1 appear to have no memory.
During SMP boot, the fault is in bus_add_device():
error = sysfs_create_link(&bus->p->devices_kset->kobj,
bus->p is NULL.
That p is the subsys_private struct, and it should have been set in
postcore_initcall(register_node_type);
But that happens after SMP boot. This fault happens during SMP boot.
The old code had set this node online via alloc_node_data(), so when it came time to do_cpu_up() -> try_online_node(), the node was already up and nothing happened.
Now, it attempts to online the node, which registers the node with sysfs, but that can't happen before the 'node' subsystem is registered.
My modified e820 map looks like this:
Quoted 6 lines
[ 0.000000] user: [mem 0x0000000000000100-0x000000000009c7ff] usable [ 0.000000] user: [mem 0x000000000009c800-0x000000000009ffff] reserved [ 0.000000] user: [mem 0x00000000000e0000-0x00000000000fffff] reserved [ 0.000000] user: [mem 0x0000000000100000-0x0000000073216fff] usable [ 0.000000] user: [mem 0x0000000073217000-0x0000000075316fff] reserved [ 0.000000] user: [mem 0x0000000075317000-0x00000000754f8fff] ACPI data
Quoted 8 lines
[ 0.000000] user: [mem 0x00000000754f9000-0x0000000076057fff] ACPI NVS [ 0.000000] user: [mem 0x0000000076058000-0x0000000077ae9fff] reserved [ 0.000000] user: [mem 0x0000000077aea000-0x0000000077ffffff] usable [ 0.000000] user: [mem 0x0000000078000000-0x000000008fffffff] reserved [ 0.000000] user: [mem 0x00000000fd000000-0x00000000fe7fffff] reserved [ 0.000000] user: [mem 0x00000000ff000000-0x00000000ffffffff] reserved [ 0.000000] user: [mem 0x0000000100000000-0x00000004ffffffff] usable [ 0.000000] user: [mem 0x0000000500000000-0x000000603fffffff] persistent (type 12)
Which leads to an empty zone 1:
Quoted 2 lines
[ 0.016060] Initmem setup node 0 [mem 0x0000000000001000-0x00000004ffffffff] [ 0.073310] Initmem setup node 1 [mem 0x0000000000000000-0x0000000000000000]
The backtrace:
Quoted 6 lines
[ 2.175327] Call Trace: [ 2.175327] device_add+0x43e/0x690 [ 2.175327] device_register+0x107/0x110 [ 2.175327] __register_one_node+0x72/0x150 [ 2.175327] __try_online_node+0x8f/0xd0 [ 2.175327] try_online_node+0x2b/0x50
Quoted 8 lines
[ 2.175327] do_cpu_up+0x46/0xf0 [ 2.175327] cpu_up+0x13/0x20 [ 2.175327] smp_init+0x6e/0xd0 [ 2.175327] kernel_init_freeable+0xe5/0x21f [ 2.175327] ? rest_init+0xb0/0xb0 [ 2.175327] kernel_init+0xf/0x180 [ 2.175327] ? rest_init+0xb0/0xb0 [ 2.175327] ret_from_fork+0x1f/0x30
To get it booting again, I unconditionally node_set_online:
arch/x86/mm/numa.c
@@ -583,7 +583,7 @@ static int __init numa_register_memblks(struct numa_meminfo *mi)
continue;
alloc_node_data(nid);
- if (end)
+ //if (end)
node_set_online(nid);
}A more elegant solution may be to avoid registering with sysfs during early boot, or something else entirely. But I figured I'd ask for help at this point. =)
Thanks,
Barret
On Wed 01-05-19 15:12:32, Barret Rhoden wrote:
[...]
Quoted 3 lines
A more elegant solution may be to avoid registering with sysfs during early boot, or something else entirely. But I figured I'd ask for help at this point. =)
Thanks for the report and an excellent analysis! This is really helpful. I will think about this some more but I am traveling this week. It seems really awkward to register a sysfs file for an empty range. That looks like a bug to me.
Signature · 2 lines
Michal Hocko SUSE Labs
Quoted 10 lines · Michal Hocko
On Wed 01-05-19 15:12:32, Barret Rhoden wrote:
[...]Quoted 3 lines
A more elegant solution may be to avoid registering with sysfs during early boot, or something else entirely. But I figured I'd ask for help at this point. =)
Thanks for the report and an excellent analysis! This is really helpful. I will think about this some more but I am traveling this week. It seems really awkward to register a sysfs file for an empty range. That looks like a bug to me.
I am sorry, but I didn't get to this for a long time and I am still busy. The patch has been dropped from the mm tree (thus linux-next). I hope I can revisit this or somebody else will take over and finish this work. This is much more trickier than I anticipated unfortunately.
Signature · 2 lines
Michal Hocko SUSE Labs