mm: Avoid calling build_all_zonelists_init under hotplug context
Subsystems (6)
Oscar Salvador
maintainerPATCH 1/1 mm: Avoid calling build_all_zonelists_init under hotplug context
Recently a customer of ours experienced a crash when booting the system while enabling memory-hotplug.
The problem is that Normal zones on different nodes don't get their private zone->pageset allocated, and keep sharing the initial boot_pageset. The sharing between zones is normally safe as explained by the comment for boot_pageset - it's a percpu structure, and manipulations are done with disabled interrupts, and boot_pageset is set up in a way that any page placed on its pcplist is immediately flushed to shared zone's freelist, because pcp->high == 1.
However, the hotplug operation updates pcp->high to a higher value as it expects to be operating on a private pageset.
The problem is in build_all_zonelists(), which is called when the first range of pages is onlined for the Normal zone of node X or Y:
if (system_state == SYSTEM_BOOTING) {
build_all_zonelists_init();
} else {
#ifdef CONFIG_MEMORY_HOTPLUG
if (zone)
setup_zone_pageset(zone);
#endif
/* we have to stop all cpus to guarantee there is no user
of zonelist */
stop_machine(__build_all_zonelists, pgdat, NULL);
/* cpuset refresh routine should be here */
}When called during hotplug, it should execute the setup_zone_pageset(zone) which allocates the private pageset.
However, with memhp_default_state=online, this happens early while system_state == SYSTEM_BOOTING is still true, hence this step is skipped. (and build_all_zonelists_init() is probably unsafe anyway at this point).
Another hotplug operation on the same zone then leads to zone_pcp_update(zone) called from online_pages(), which updates the pcp->high for the shared boot_pageset to a value higher than 1.
At that point, pages freed from Node X and Y Normal zones can end up on the same pcplist and from there they can be freed to the wrong zone's freelist, leading to the corruption and crashes.
Please, note that upstream has fixed that differently (and unintentionally) by adding another boot state (SYSTEM_SCHEDULING), which is set before smp_init(). That should happen before memory hotplug events even with memhp_default_state=online. Backporting that would be too intrusive.
--- include/linux/mmzone.h | 3 ++- init/main.c | 2 +- mm/memory_hotplug.c | 10 +++++----- mm/page_alloc.c | 7 ++++--- 4 files changed, 12 insertions(+), 10 deletions(-)
include/linux/mmzone.h +2 -1
@@ -756,7 +756,8 @@ static inline bool is_dev_zone(const struct zone *zone)#include <linux/memory_hotplug.h> extern struct mutex zonelists_mutex; -void build_all_zonelists (pg_data_t *pgdat,struct zone *zone); +void build_all_zonelists (pg_data_t *pgdat,struct zone *zone, +bool hotplug_context);void wakeup_kswapd (struct zone *zone,int order,enum zone_type classzone_idx); bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, int classzone_idx, unsigned int alloc_flags,
init/main.c +1 -1
@@ -512,7 +512,7 @@ asmlinkage __visible void __init start_kernel(void)smp_prepare_boot_cpu ();/* arch-specific boot-cpu hooks */ boot_cpu_hotplug_init (); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , false);page_alloc_init ();pr_notice ("Kernel command line: %s\n" , boot_command_line);
mm/memory_hotplug.c +5 -5
@@ -1125,7 +1125,7 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_typmutex_lock (&zonelists_mutex);if (!populated_zone (zone)) { need_zonelists_rebuild =1 ; -build_all_zonelists (NULL , zone); +build_all_zonelists (NULL , zone, true); } ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
@@ -1146,7 +1146,7 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_typif (onlined_pages) {node_states_set_node (nid, &arg);if (need_zonelists_rebuild) -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , true);else zone_pcp_update (zone); }
@@ -1220,7 +1220,7 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)* to access not-initialized zonelist, build here. */ mutex_lock (&zonelists_mutex); -build_all_zonelists (pgdat,NULL ); +build_all_zonelists (pgdat,NULL , true);mutex_unlock (&zonelists_mutex);/*
@@ -1276,7 +1276,7 @@ int try_online_node(int nid)if (pgdat->node_zonelists->_zonerefs->zone ==NULL ) {mutex_lock (&zonelists_mutex); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , true);mutex_unlock (&zonelists_mutex); }
@@ -2016,7 +2016,7 @@ static int __ref __offline_pages(unsigned long start_pfn,if (!populated_zone (zone)) {zone_pcp_reset (zone);mutex_lock (&zonelists_mutex); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , true);mutex_unlock (&zonelists_mutex); }else zone_pcp_update (zone);
mm/page_alloc.c +4 -3
@@ -4608,7 +4608,7 @@ int numa_zonelist_order_handler(struct ctl_table *table, int write, user_zonelist_order = oldval; }else if (oldval != user_zonelist_order) {mutex_lock (&zonelists_mutex); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , false);mutex_unlock (&zonelists_mutex); } }
@@ -4988,11 +4988,12 @@ build_all_zonelists_init(void)* (2) call of __init annotated helper build_all_zonelists_init * [protected by SYSTEM_BOOTING]. */ -void __ref build_all_zonelists (pg_data_t *pgdat,struct zone *zone) +void __ref build_all_zonelists (pg_data_t *pgdat,struct zone *zone, +bool hotplug_context) {set_zonelist_order (); -if (system_state ==SYSTEM_BOOTING ) { +if (system_state ==SYSTEM_BOOTING && !hotplug_context) {build_all_zonelists_init (); }else {#ifdef CONFIG_MEMORY_HOTPLUG
Signature · 1 line
2.26.2
Quoted 42 lines · Oscar Salvador
Recently a customer of ours experienced a crash when booting the system while enabling memory-hotplug.
The problem is that Normal zones on different nodes don't get their private
zone->pagesetallocated, and keep sharing the initialboot_pageset. The sharing between zones is normally safe as explained by the comment forboot_pageset- it's a percpu structure, and manipulations are done with disabled interrupts, andboot_pagesetis set up in a way that any page placed on its pcplist is immediately flushed to shared zone's freelist, becausepcp->high== 1.
However, the hotplug operation updatespcp->highto a higher value as it expects to be operating on a private pageset.The problem is in
build_all_zonelists(), which is called when the first range of pages is onlined for the Normal zone of node X or Y:if (system_state == SYSTEM_BOOTING) { build_all_zonelists_init(); } else { #ifdef CONFIG_MEMORY_HOTPLUG if (zone) setup_zone_pageset(zone); #endif /* we have to stop all cpus to guarantee there is no user of zonelist */ stop_machine(__build_all_zonelists, pgdat, NULL); /* cpuset refresh routine should be here */ }When called during hotplug, it should execute the
setup_zone_pageset(zone)which allocates the private pageset.
However, with memhp_default_state=online, this happens early whilesystem_state==SYSTEM_BOOTINGis still true, hence this step is skipped. (andbuild_all_zonelists_init()is probably unsafe anyway at this point).Another hotplug operation on the same zone then leads to
zone_pcp_update(zone)called fromonline_pages(), which updates thepcp->highfor the sharedboot_pagesetto a value higher than 1.
At that point, pages freed from Node X and Y Normal zones can end up on the same pcplist and from there they can be freed to the wrong zone's freelist, leading to the corruption and crashes.
Quoted 7 lines
Please, note that upstream has fixed that differently (and unintentionally) by adding another boot state (
SYSTEM_SCHEDULING), which is set beforesmp_init(). That should happen before memory hotplug events even with memhp_default_state=online. Backporting that would be too intrusive.Signed-off-by Oscar Salvador <[email protected]>Debugged-by Vlastimil Babka <[email protected]>
Yes, I believe this is the easiest and the least scary way to fix the issue for stable kernel users. Feel free to add
for that purpose.
Thanks a lot!
Quoted 114 lines
--- include/linux/mmzone.h | 3 ++- init/main.c | 2 +- mm/memory_hotplug.c | 10 +++++----- mm/page_alloc.c | 7 ++++--- 4 files changed, 12 insertions(+), 10 deletions(-)include/linux/mmzone.h +2
-1@@ -756,7 +756,8 @@ static inline bool is_dev_zone(const struct zone *zone)#include <linux/memory_hotplug.h> extern struct mutex zonelists_mutex; -void build_all_zonelists (pg_data_t *pgdat,struct zone *zone); +void build_all_zonelists (pg_data_t *pgdat,struct zone *zone, +bool hotplug_context);void wakeup_kswapd (struct zone *zone,int order,enum zone_type classzone_idx); bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, int classzone_idx, unsigned int alloc_flags,init/main.c +1
-1@@ -512,7 +512,7 @@ asmlinkage __visible void __init start_kernel(void)smp_prepare_boot_cpu ();/* arch-specific boot-cpu hooks */ boot_cpu_hotplug_init (); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , false);page_alloc_init ();pr_notice ("Kernel command line: %s\n" , boot_command_line);mm/memory_hotplug.c +5
-5@@ -1125,7 +1125,7 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_typmutex_lock (&zonelists_mutex);if (!populated_zone (zone)) { need_zonelists_rebuild =1 ; -build_all_zonelists (NULL , zone); +build_all_zonelists (NULL , zone, true); } ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,@@ -1146,7 +1146,7 @@ int __ref online_pages(unsigned long pfn, unsigned long nr_pages, int online_typif (onlined_pages) {node_states_set_node (nid, &arg);if (need_zonelists_rebuild) -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , true);else zone_pcp_update (zone); }@@ -1220,7 +1220,7 @@ static pg_data_t __ref *hotadd_new_pgdat(int nid, u64 start)* to access not-initialized zonelist, build here. */ mutex_lock (&zonelists_mutex); -build_all_zonelists (pgdat,NULL ); +build_all_zonelists (pgdat,NULL , true);mutex_unlock (&zonelists_mutex);/* @@ -1276,7 +1276,7 @@ int try_online_node(int nid)if (pgdat->node_zonelists->_zonerefs->zone ==NULL ) {mutex_lock (&zonelists_mutex); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , true);mutex_unlock (&zonelists_mutex); }@@ -2016,7 +2016,7 @@ static int __ref __offline_pages(unsigned long start_pfn,if (!populated_zone (zone)) {zone_pcp_reset (zone);mutex_lock (&zonelists_mutex); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , true);mutex_unlock (&zonelists_mutex); }else zone_pcp_update (zone);mm/page_alloc.c +4
-3@@ -4608,7 +4608,7 @@ int numa_zonelist_order_handler(struct ctl_table *table, int write, user_zonelist_order = oldval; }else if (oldval != user_zonelist_order) {mutex_lock (&zonelists_mutex); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , false);mutex_unlock (&zonelists_mutex); } }@@ -4988,11 +4988,12 @@ build_all_zonelists_init(void)* (2) call of __init annotated helper build_all_zonelists_init * [protected by SYSTEM_BOOTING]. */ -void __ref build_all_zonelists (pg_data_t *pgdat,struct zone *zone) +void __ref build_all_zonelists (pg_data_t *pgdat,struct zone *zone, +bool hotplug_context) {set_zonelist_order (); -if (system_state ==SYSTEM_BOOTING ) { +if (system_state ==SYSTEM_BOOTING && !hotplug_context) {build_all_zonelists_init (); }else {#ifdef CONFIG_MEMORY_HOTPLUG Signature · 1 line
2.26.2
Signature · 2 lines
Michal Hocko SUSE Labs
Greg KH
maintainerQuoted 48 lines · Michal Hocko
Quoted 47 lines · Oscar Salvador
Recently a customer of ours experienced a crash when booting the system while enabling memory-hotplug.
The problem is that Normal zones on different nodes don't get their private
zone->pagesetallocated, and keep sharing the initialboot_pageset. The sharing between zones is normally safe as explained by the comment forboot_pageset- it's a percpu structure, and manipulations are done with disabled interrupts, andboot_pagesetis set up in a way that any page placed on its pcplist is immediately flushed to shared zone's freelist, becausepcp->high== 1.
However, the hotplug operation updatespcp->highto a higher value as it expects to be operating on a private pageset.The problem is in
build_all_zonelists(), which is called when the first range of pages is onlined for the Normal zone of node X or Y:if (system_state == SYSTEM_BOOTING) { build_all_zonelists_init(); } else { #ifdef CONFIG_MEMORY_HOTPLUG if (zone) setup_zone_pageset(zone); #endif /* we have to stop all cpus to guarantee there is no user of zonelist */ stop_machine(__build_all_zonelists, pgdat, NULL); /* cpuset refresh routine should be here */ }When called during hotplug, it should execute the
setup_zone_pageset(zone)which allocates the private pageset.
However, with memhp_default_state=online, this happens early whilesystem_state==SYSTEM_BOOTINGis still true, hence this step is skipped. (andbuild_all_zonelists_init()is probably unsafe anyway at this point).Another hotplug operation on the same zone then leads to
zone_pcp_update(zone)called fromonline_pages(), which updates thepcp->highfor the sharedboot_pagesetto a value higher than 1.
At that point, pages freed from Node X and Y Normal zones can end up on the same pcplist and from there they can be freed to the wrong zone's freelist, leading to the corruption and crashes.Please, note that upstream has fixed that differently (and unintentionally) by adding another boot state (
SYSTEM_SCHEDULING), which is set beforesmp_init(). That should happen before memory hotplug events even with memhp_default_state=online. Backporting that would be too intrusive.
Quoted 8 lines
Quoted 2 lines
Signed-off-by Oscar Salvador <[email protected]>Debugged-by Vlastimil Babka <[email protected]>Yes, I believe this is the easiest and the least scary way to fix the issue for stable kernel users. Feel free to add
for that purpose.
Now queued up, thanks!
greg k-h
Quoted 42 lines · Oscar Salvador
Recently a customer of ours experienced a crash when booting the system while enabling memory-hotplug.
The problem is that Normal zones on different nodes don't get their private
zone->pagesetallocated, and keep sharing the initialboot_pageset. The sharing between zones is normally safe as explained by the comment forboot_pageset- it's a percpu structure, and manipulations are done with disabled interrupts, andboot_pagesetis set up in a way that any page placed on its pcplist is immediately flushed to shared zone's freelist, becausepcp->high== 1.
However, the hotplug operation updatespcp->highto a higher value as it expects to be operating on a private pageset.The problem is in
build_all_zonelists(), which is called when the first range of pages is onlined for the Normal zone of node X or Y:if (system_state == SYSTEM_BOOTING) { build_all_zonelists_init(); } else { #ifdef CONFIG_MEMORY_HOTPLUG if (zone) setup_zone_pageset(zone); #endif /* we have to stop all cpus to guarantee there is no user of zonelist */ stop_machine(__build_all_zonelists, pgdat, NULL); /* cpuset refresh routine should be here */ }When called during hotplug, it should execute the
setup_zone_pageset(zone)which allocates the private pageset.
However, with memhp_default_state=online, this happens early whilesystem_state==SYSTEM_BOOTINGis still true, hence this step is skipped. (andbuild_all_zonelists_init()is probably unsafe anyway at this point).Another hotplug operation on the same zone then leads to
zone_pcp_update(zone)called fromonline_pages(), which updates thepcp->highfor the sharedboot_pagesetto a value higher than 1.
At that point, pages freed from Node X and Y Normal zones can end up on the same pcplist and from there they can be freed to the wrong zone's freelist, leading to the corruption and crashes.
Quoted 7 lines
Please, note that upstream has fixed that differently (and unintentionally) by adding another boot state (
SYSTEM_SCHEDULING), which is set beforesmp_init(). That should happen before memory hotplug events even with memhp_default_state=online. Backporting that would be too intrusive.Signed-off-by Oscar Salvador <[email protected]>Debugged-by Vlastimil Babka <[email protected]>
So, we have ACPI running and already adding DIMMs while booting? Crazy.
Looks sane to me. Thanks!
Signature · 3 lines
Thanks, David / dhildenb
Patch "mm: Avoid calling build_all_zonelists_init under hotplug context" has been added to the 4.9-stable tree
In reply to Oscar SalvadorThis is a note to let you know that I've just added the patch titled
mm: Avoid calling build_all_zonelists_init under hotplug context
to the 4.9-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
mm-avoid-calling-build_all_zonelists_init-under-hotplug-context.patch
and it can be found in the queue-4.9 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree, please let <[email protected]> know about it.
From [email protected] Thu Aug 20 10:13:12 2020
From: Oscar Salvador <[email protected]>
Date: Tue, 18 Aug 2020 13:00:46 +0200
Subject: mm: Avoid calling build_all_zonelists_init under hotplug context
To: [email protected]
Cc: [email protected], [email protected], [email protected], [email protected], [email protected], Oscar Salvador <[email protected]>, Vlastimil Babka <[email protected]>
Message-ID: <[email protected]>
From: Oscar Salvador <[email protected]>
Recently a customer of ours experienced a crash when booting the system while enabling memory-hotplug.
The problem is that Normal zones on different nodes don't get their private zone->pageset allocated, and keep sharing the initial boot_pageset. The sharing between zones is normally safe as explained by the comment for boot_pageset - it's a percpu structure, and manipulations are done with disabled interrupts, and boot_pageset is set up in a way that any page placed on its pcplist is immediately flushed to shared zone's freelist, because pcp->high == 1.
However, the hotplug operation updates pcp->high to a higher value as it expects to be operating on a private pageset.
The problem is in build_all_zonelists(), which is called when the first range of pages is onlined for the Normal zone of node X or Y:
if (system_state == SYSTEM_BOOTING) {
build_all_zonelists_init();
} else {
#ifdef CONFIG_MEMORY_HOTPLUG
if (zone)
setup_zone_pageset(zone);
#endif
/* we have to stop all cpus to guarantee there is no user
of zonelist */
stop_machine(__build_all_zonelists, pgdat, NULL);
/* cpuset refresh routine should be here */
}When called during hotplug, it should execute the setup_zone_pageset(zone) which allocates the private pageset.
However, with memhp_default_state=online, this happens early while system_state == SYSTEM_BOOTING is still true, hence this step is skipped. (and build_all_zonelists_init() is probably unsafe anyway at this point).
Another hotplug operation on the same zone then leads to zone_pcp_update(zone) called from online_pages(), which updates the pcp->high for the shared boot_pageset to a value higher than 1.
At that point, pages freed from Node X and Y Normal zones can end up on the same pcplist and from there they can be freed to the wrong zone's freelist, leading to the corruption and crashes.
Please, note that upstream has fixed that differently (and unintentionally) by adding another boot state (SYSTEM_SCHEDULING), which is set before smp_init(). That should happen before memory hotplug events even with memhp_default_state=online. Backporting that would be too intrusive.
--- include/linux/mmzone.h | 3 ++- init/main.c | 2 +- mm/memory_hotplug.c | 10 +++++----- mm/page_alloc.c | 7 ++++--- 4 files changed, 12 insertions(+), 10 deletions(-)
include/linux/mmzone.h +2 -1
@@ -756,7 +756,8 @@ static inline bool is_dev_zone(const str#include <linux/memory_hotplug.h> extern struct mutex zonelists_mutex; -void build_all_zonelists (pg_data_t *pgdat,struct zone *zone); +void build_all_zonelists (pg_data_t *pgdat,struct zone *zone, +bool hotplug_context);void wakeup_kswapd (struct zone *zone,int order,enum zone_type classzone_idx); bool __zone_watermark_ok(struct zone *z, unsigned int order, unsigned long mark, int classzone_idx, unsigned int alloc_flags,
init/main.c +1 -1
@@ -512,7 +512,7 @@ asmlinkage __visible void __init start_ksmp_prepare_boot_cpu ();/* arch-specific boot-cpu hooks */ boot_cpu_hotplug_init (); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , false);page_alloc_init ();pr_notice ("Kernel command line: %s\n" , boot_command_line);
mm/memory_hotplug.c +5 -5
@@ -1125,7 +1125,7 @@ int __ref online_pages(unsigned long pfnmutex_lock (&zonelists_mutex);if (!populated_zone (zone)) { need_zonelists_rebuild =1 ; -build_all_zonelists (NULL , zone); +build_all_zonelists (NULL , zone, true); } ret = walk_system_ram_range(pfn, nr_pages, &onlined_pages,
@@ -1146,7 +1146,7 @@ int __ref online_pages(unsigned long pfnif (onlined_pages) {node_states_set_node (nid, &arg);if (need_zonelists_rebuild) -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , true);else zone_pcp_update (zone); }
@@ -1220,7 +1220,7 @@ static pg_data_t __ref *hotadd_new_pgdat* to access not-initialized zonelist, build here. */ mutex_lock (&zonelists_mutex); -build_all_zonelists (pgdat,NULL ); +build_all_zonelists (pgdat,NULL , true);mutex_unlock (&zonelists_mutex);/*
@@ -1276,7 +1276,7 @@ int try_online_node(int nid)if (pgdat->node_zonelists->_zonerefs->zone ==NULL ) {mutex_lock (&zonelists_mutex); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , true);mutex_unlock (&zonelists_mutex); }
@@ -2016,7 +2016,7 @@ repeat:if (!populated_zone (zone)) {zone_pcp_reset (zone);mutex_lock (&zonelists_mutex); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , true);mutex_unlock (&zonelists_mutex); }else zone_pcp_update (zone);
mm/page_alloc.c +4 -3
@@ -4608,7 +4608,7 @@ int numa_zonelist_order_handler(struct c user_zonelist_order = oldval; }else if (oldval != user_zonelist_order) {mutex_lock (&zonelists_mutex); -build_all_zonelists (NULL ,NULL ); +build_all_zonelists (NULL ,NULL , false);mutex_unlock (&zonelists_mutex); } }
@@ -4988,11 +4988,12 @@ build_all_zonelists_init(void)* (2) call of __init annotated helper build_all_zonelists_init * [protected by SYSTEM_BOOTING]. */ -void __ref build_all_zonelists (pg_data_t *pgdat,struct zone *zone) +void __ref build_all_zonelists (pg_data_t *pgdat,struct zone *zone, +bool hotplug_context) {set_zonelist_order (); -if (system_state ==SYSTEM_BOOTING ) { +if (system_state ==SYSTEM_BOOTING && !hotplug_context) {build_all_zonelists_init (); }else {#ifdef CONFIG_MEMORY_HOTPLUG
Patches currently in stable-queue which might be from [email protected] are
queue-4.9/mm-avoid-calling-build_all_zonelists_init-under-hotplug-context.patch