eliminate possible race in parallel TLB shootdown IPI
On the target side TLB shootdown IPI handler, prevent the compiler from performing a forward store optimization which may mask a subsequent update to the scoreboard by the initiator. Reported by: Max Laier, Anton Rang Discussed with: kib Sponsored by: Dell EMC Isilon
This commit is contained in:
parent
47e2650ea4
commit
42360f5c5b
@ -1094,7 +1094,7 @@ invlop_handler(void)
|
||||
for (;;) {
|
||||
for (initiator_cpu_id = 0; initiator_cpu_id <= mp_maxid;
|
||||
initiator_cpu_id++) {
|
||||
if (scoreboard[initiator_cpu_id] == 0)
|
||||
if (atomic_load_int(&scoreboard[initiator_cpu_id]) == 0)
|
||||
break;
|
||||
}
|
||||
if (initiator_cpu_id > mp_maxid)
|
||||
|
Loading…
x
Reference in New Issue
Block a user