Adrian Moreno was ruuning a kubernetes 1.19 + containerd/docker workload
using hugetlbfs. In this environment the issue is reproduced by:
1 - Start a simple pod that uses the recently added HugePages medium
feature (pod yaml attached)
2 - Start a DPDK app. It doesn't need to run successfully (as in transfer
packets) nor interact with real hardware. It seems just initializing
the EAL layer (which handles hugepage reservation and locking) is
enough to trigger the issue
3 - Delete the Pod (or let it "Complete").
This would result in a kworker thread going into a tight loop (top output):
1425 root 20 0 0 0 0 R 99.7 0.0 5:22.45
kworker/28:7+cgroup_destroy
'perf top -g' reports:
- 63.28% 0.01% [kernel] [k] worker_thread
- 49.97% worker_thread
- 52.64% process_one_work
- 62.08% css_killed_work_fn
- hugetlb_cgroup_css_offline
41.52% _raw_spin_lock
- 2.82% _cond_resched
rcu_all_qs
2.66% PageHuge
- 0.57% schedule
- 0.57% __schedule
We are spinning in the do-while loop in hugetlb_cgroup_css_offline.
Worse yet, we are holding the master cgroup lock (cgroup_mutex) while
infinitely spinning. Little else can be done on the system as the
cgroup_mutex can not be acquired.
Do note that the issue can be reproduced by simply offlining a hugetlb
cgroup containing pages with reservation counts.
The loop in hugetlb_cgroup_css_offline is moving page counts from the
cgroup being offlined to the parent cgroup. This is done for each hstate,
and is repeated until hugetlb_cgroup_have_usage returns false. The routine
moving counts (hugetlb_cgroup_move_parent) is only moving 'usage' counts.
The routine hugetlb_cgroup_have_usage is checking for both 'usage' and
'reservation' counts. Discussion about what to do with reservation
counts when reparenting was discussed here:
https://lore.kernel.org/linux-kselftest/CAHS8izMFAYTgxym-Hzb_JmkTK1N_S9tGN71uS6MFV+R7swYu5A@xxxxxxxxxxxxxx/
The decision was made to leave a zombie cgroup for with reservation
counts. Unfortunately, the code checking reservation counts was
incorrectly added to hugetlb_cgroup_have_usage.
To fix the issue, simply remove the check for reservation counts. While
fixing this issue, a related bug in hugetlb_cgroup_css_offline was noticed.
The hstate index is not reinitialized each time through the do-while loop.
Fix this as well.
Fixes: 1adc4d419aa2 ("hugetlb_cgroup: add interface for charge/uncharge hugetlb reservations")
Cc: <stable@xxxxxxxxxxxxxxx>
Reported-by: Adrian Moreno <amorenoz@xxxxxxxxxx>
Tested-by: Adrian Moreno <amorenoz@xxxxxxxxxx>
Signed-off-by: Mike Kravetz <mike.kravetz@xxxxxxxxxx>
---
mm/hugetlb_cgroup.c | 8 +++-----
1 file changed, 3 insertions(+), 5 deletions(-)
diff --git a/mm/hugetlb_cgroup.c b/mm/hugetlb_cgroup.c
index 1f87aec9ab5c..9182848dda3e 100644
--- a/mm/hugetlb_cgroup.c
+++ b/mm/hugetlb_cgroup.c
@@ -82,11 +82,8 @@ static inline bool hugetlb_cgroup_have_usage(struct hugetlb_cgroup *h_cg)
for (idx = 0; idx < hugetlb_max_hstate; idx++) {
if (page_counter_read(
- hugetlb_cgroup_counter_from_cgroup(h_cg, idx)) ||
- page_counter_read(hugetlb_cgroup_counter_from_cgroup_rsvd(
- h_cg, idx))) {
+ hugetlb_cgroup_counter_from_cgroup(h_cg, idx)))
return true;
- }
}
return false;
}
@@ -202,9 +199,10 @@ static void hugetlb_cgroup_css_offline(struct cgroup_subsys_state *css)
struct hugetlb_cgroup *h_cg = hugetlb_cgroup_from_css(css);
struct hstate *h;
struct page *page;
- int idx = 0;
+ int idx;
do {
+ idx = 0;
for_each_hstate(h) {
spin_lock(&hugetlb_lock);
list_for_each_entry(page, &h->hugepage_activelist, lru)
--
2.28.0
