| Message ID | 20181031125840.23982-1-mhocko@kernel.org (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | memory_hotplug: cond_resched in __remove_pages | expand |
Acked-by: Johannes Thumshirn <jthumshirn@suse.de>
On Wed, 31 Oct 2018 13:58:40 +0100 Michal Hocko <mhocko@kernel.org> wrote: > From: Michal Hocko <mhocko@suse.com> > > We have received a bug report that unbinding a large pmem (>1TB) > can result in a soft lockup: > > ... > > It has been reported on an older (4.12) kernel but the current upstream > code doesn't cond_resched in the hot remove code at all and the given > range to remove might be really large. Fix the issue by calling cond_resched > once per memory section. > Worthy of a cc:stable, I suggest?
On Wed 31-10-18 12:15:50, Andrew Morton wrote: > On Wed, 31 Oct 2018 13:58:40 +0100 Michal Hocko <mhocko@kernel.org> wrote: > > > From: Michal Hocko <mhocko@suse.com> > > > > We have received a bug report that unbinding a large pmem (>1TB) > > can result in a soft lockup: > > > > ... > > > > It has been reported on an older (4.12) kernel but the current upstream > > code doesn't cond_resched in the hot remove code at all and the given > > range to remove might be really large. Fix the issue by calling cond_resched > > once per memory section. > > > > Worthy of a cc:stable, I suggest? It is simple enough and we will surely have it in 4.12 based SLES kernels. So I do not really mind cc: stable.
On Wed, Oct 31, 2018 at 01:58:40PM +0100, Michal Hocko wrote: > From: Michal Hocko <mhocko@suse.com> > > We have received a bug report that unbinding a large pmem (>1TB) > can result in a soft lockup: > [ 380.339203] NMI watchdog: BUG: soft lockup - CPU#9 stuck for 23s! [ndctl:4365] > [...] > [ 380.339316] Supported: Yes > [ 380.339318] CPU: 9 PID: 4365 Comm: ndctl Not tainted 4.12.14-94.40-default #1 SLE12-SP4 > [ 380.339318] Hardware name: Intel Corporation S2600WFD/S2600WFD, BIOS SE5C620.86B.01.00.0833.051120182255 05/11/2018 > [ 380.339319] task: ffff9cce7d4410c0 task.stack: ffffbe9eb1bc4000 > [ 380.339325] RIP: 0010:__put_page+0x62/0x80 > [ 380.339326] RSP: 0018:ffffbe9eb1bc7d30 EFLAGS: 00000282 ORIG_RAX: ffffffffffffff10 > [ 380.339327] RAX: 000040540081c0d3 RBX: ffffeb8f03557200 RCX: 000063af40000000 > [ 380.339328] RDX: 0000000000000002 RSI: ffff9cce75bff498 RDI: ffff9e4a76072ff8 > [ 380.339329] RBP: 0000000a43557200 R08: 0000000000000000 R09: ffffbe9eb1bc7bb0 > [ 380.339329] R10: ffffbe9eb1bc7d08 R11: 0000000000000000 R12: ffff9e194a22a0e0 > [ 380.339330] R13: ffff9cce7062fc10 R14: ffff9e194a22a0a0 R15: ffff9cce6559c0e0 > [ 380.339331] FS: 00007fd132368880(0000) GS:ffff9cce7ea40000(0000) knlGS:0000000000000000 > [ 380.339332] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > [ 380.339332] CR2: 00000000020820a0 CR3: 000000017ef7a003 CR4: 00000000007606e0 > [ 380.339333] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 > [ 380.339334] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 > [ 380.339334] PKRU: 55555554 > [ 380.339334] Call Trace: > [ 380.339338] devm_memremap_pages_release+0x152/0x260 > [ 380.339342] release_nodes+0x18d/0x1d0 > [ 380.339347] device_release_driver_internal+0x160/0x210 > [ 380.339350] unbind_store+0xb3/0xe0 > [ 380.339355] kernfs_fop_write+0x102/0x180 > [ 380.339358] __vfs_write+0x26/0x150 > [ 380.339363] ? security_file_permission+0x3c/0xc0 > [ 380.339364] vfs_write+0xad/0x1a0 > [ 380.339366] SyS_write+0x42/0x90 > [ 380.339370] do_syscall_64+0x74/0x150 > [ 380.339375] entry_SYSCALL_64_after_hwframe+0x3d/0xa2 > [ 380.339377] RIP: 0033:0x7fd13166b3d0 > > It has been reported on an older (4.12) kernel but the current upstream > code doesn't cond_resched in the hot remove code at all and the given > range to remove might be really large. Fix the issue by calling cond_resched > once per memory section. > > Signed-off-by: Michal Hocko <mhocko@suse.com> > --- > mm/memory_hotplug.c | 1 + > 1 file changed, 1 insertion(+) > > diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c > index 7e6509a53d79..1d87724fa558 100644 > --- a/mm/memory_hotplug.c > +++ b/mm/memory_hotplug.c > @@ -587,6 +587,7 @@ int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, > for (i = 0; i < sections_to_remove; i++) { > unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; > > + cond_resched(); > ret = __remove_section(zone, __pfn_to_section(pfn), map_offset, > altmap); > map_offset = 0; Quick math tells me we're doing less than 44GiB's per second of offlining then? Here is a quick untested patch that might help with the speed as well In hot remove, we try to clear poisoned pages, but a small optimization to check if num_poisoned_pages is 0 helps remove the iteration through nr_pages. NOTE: We can make num_poisoned_pages counter per section and speed this up even more in case we do have some poisoned pages Signed-off-by: Balbir Singh <bsingharora@gmail.com> --- mm/sparse.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/mm/sparse.c b/mm/sparse.c index 33307fc05c4d..c4280ef0f383 100644 --- a/mm/sparse.c +++ b/mm/sparse.c @@ -724,6 +724,9 @@ static void clear_hwpoisoned_pages(struct page *memmap, int nr_pages) if (!memmap) return; + if (atomic_long_read(&num_poisoned_pages) == 0) + return; + for (i = 0; i < nr_pages; i++) { if (PageHWPoison(&memmap[i])) { atomic_long_sub(1, &num_poisoned_pages); Anyway for this patch: Acked-by: Balbir Singh <bsingharora@gmail.com>
On Fri 02-11-18 14:52:05, Balbir Singh wrote: > On Wed, Oct 31, 2018 at 01:58:40PM +0100, Michal Hocko wrote: > > From: Michal Hocko <mhocko@suse.com> > > > > We have received a bug report that unbinding a large pmem (>1TB) > > can result in a soft lockup: > > [ 380.339203] NMI watchdog: BUG: soft lockup - CPU#9 stuck for 23s! [ndctl:4365] > > [...] > > [ 380.339316] Supported: Yes > > [ 380.339318] CPU: 9 PID: 4365 Comm: ndctl Not tainted 4.12.14-94.40-default #1 SLE12-SP4 > > [ 380.339318] Hardware name: Intel Corporation S2600WFD/S2600WFD, BIOS SE5C620.86B.01.00.0833.051120182255 05/11/2018 > > [ 380.339319] task: ffff9cce7d4410c0 task.stack: ffffbe9eb1bc4000 > > [ 380.339325] RIP: 0010:__put_page+0x62/0x80 > > [ 380.339326] RSP: 0018:ffffbe9eb1bc7d30 EFLAGS: 00000282 ORIG_RAX: ffffffffffffff10 > > [ 380.339327] RAX: 000040540081c0d3 RBX: ffffeb8f03557200 RCX: 000063af40000000 > > [ 380.339328] RDX: 0000000000000002 RSI: ffff9cce75bff498 RDI: ffff9e4a76072ff8 > > [ 380.339329] RBP: 0000000a43557200 R08: 0000000000000000 R09: ffffbe9eb1bc7bb0 > > [ 380.339329] R10: ffffbe9eb1bc7d08 R11: 0000000000000000 R12: ffff9e194a22a0e0 > > [ 380.339330] R13: ffff9cce7062fc10 R14: ffff9e194a22a0a0 R15: ffff9cce6559c0e0 > > [ 380.339331] FS: 00007fd132368880(0000) GS:ffff9cce7ea40000(0000) knlGS:0000000000000000 > > [ 380.339332] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 > > [ 380.339332] CR2: 00000000020820a0 CR3: 000000017ef7a003 CR4: 00000000007606e0 > > [ 380.339333] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 > > [ 380.339334] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 > > [ 380.339334] PKRU: 55555554 > > [ 380.339334] Call Trace: > > [ 380.339338] devm_memremap_pages_release+0x152/0x260 > > [ 380.339342] release_nodes+0x18d/0x1d0 > > [ 380.339347] device_release_driver_internal+0x160/0x210 > > [ 380.339350] unbind_store+0xb3/0xe0 > > [ 380.339355] kernfs_fop_write+0x102/0x180 > > [ 380.339358] __vfs_write+0x26/0x150 > > [ 380.339363] ? security_file_permission+0x3c/0xc0 > > [ 380.339364] vfs_write+0xad/0x1a0 > > [ 380.339366] SyS_write+0x42/0x90 > > [ 380.339370] do_syscall_64+0x74/0x150 > > [ 380.339375] entry_SYSCALL_64_after_hwframe+0x3d/0xa2 > > [ 380.339377] RIP: 0033:0x7fd13166b3d0 > > > > It has been reported on an older (4.12) kernel but the current upstream > > code doesn't cond_resched in the hot remove code at all and the given > > range to remove might be really large. Fix the issue by calling cond_resched > > once per memory section. > > > > Signed-off-by: Michal Hocko <mhocko@suse.com> > > --- > > mm/memory_hotplug.c | 1 + > > 1 file changed, 1 insertion(+) > > > > diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c > > index 7e6509a53d79..1d87724fa558 100644 > > --- a/mm/memory_hotplug.c > > +++ b/mm/memory_hotplug.c > > @@ -587,6 +587,7 @@ int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, > > for (i = 0; i < sections_to_remove; i++) { > > unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; > > > > + cond_resched(); > > ret = __remove_section(zone, __pfn_to_section(pfn), map_offset, > > altmap); > > map_offset = 0; > > Quick math tells me we're doing less than 44GiB's per second of offlining then? > > Here is a quick untested patch that might help with the speed as well > > In hot remove, we try to clear poisoned pages, but > a small optimization to check if num_poisoned_pages > is 0 helps remove the iteration through nr_pages. > > NOTE: We can make num_poisoned_pages counter per > section and speed this up even more in case we > do have some poisoned pages yes this makes sense. Could you post a proper patch so that this doesn't get lost in this thread?
diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c index 7e6509a53d79..1d87724fa558 100644 --- a/mm/memory_hotplug.c +++ b/mm/memory_hotplug.c @@ -587,6 +587,7 @@ int __remove_pages(struct zone *zone, unsigned long phys_start_pfn, for (i = 0; i < sections_to_remove; i++) { unsigned long pfn = phys_start_pfn + i*PAGES_PER_SECTION; + cond_resched(); ret = __remove_section(zone, __pfn_to_section(pfn), map_offset, altmap); map_offset = 0;