From: Waiman Long longman@redhat.com
mainline inclusion from mainline-v6.7-rc1 commit 46c521bac592251229acdd2cd67976a7b1f88bed category: bugfix bugzilla: https://gitee.com/openeuler/kernel/issues/I93ZFV
----------------------------------------------------------------------
When a local partition becomes invalid, it won't transition back to valid partition automatically if a proper "cpuset.cpus.exclusive" or "cpuset.cpus" change is made. Instead, system administrators have to explicitly echo "root" or "isolated" into the "cpuset.cpus.partition" file at the partition root.
This patch now enables the automatic transition of an invalid local partition back to valid when there is a proper "cpuset.cpus.exclusive" or "cpuset.cpus" change.
Automatic transition of an invalid remote partition to a valid one, however, is not covered by this patch. They still need an explicit write to "cpuset.cpus.partition" to become valid again.
The test_cpuset_prs.sh test script is updated to add new test cases to test this automatic state transition.
Reported-by: Pierre Gondois pierre.gondois@arm.com Link: https://lore.kernel.org/lkml/9777f0d2-2fdf-41cb-bd01-19c52939ef42@arm.com Signed-off-by: Waiman Long longman@redhat.com Signed-off-by: Tejun Heo tj@kernel.org
Conflicts: tools/testing/selftests/cgroup/test_cpuset_prs.sh
Signed-off-by: Chen Ridong chenridong@huawei.com --- kernel/cgroup/cpuset.c | 79 +++++++++++++++++++++++++----------------- 1 file changed, 48 insertions(+), 31 deletions(-)
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c index 51a398fc42ac..8ce22102b717 100644 --- a/kernel/cgroup/cpuset.c +++ b/kernel/cgroup/cpuset.c @@ -1913,17 +1913,28 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd, * * Compute add/delete mask to/from effective_cpus * - * addmask = effective_xcpus & ~newmask & parent->effective_xcpus - * delmask = newmask & ~cs->effective_xcpus - * & parent->effective_xcpus + * For valid partition: + * addmask = exclusive_cpus & ~newmask + * & parent->effective_xcpus + * delmask = newmask & ~exclusive_cpus + * & parent->effective_xcpus + * + * For invalid partition: + * delmask = newmask & parent->effective_xcpus */ - cpumask_andnot(tmp->addmask, xcpus, newmask); - adding = cpumask_and(tmp->addmask, tmp->addmask, - parent->effective_xcpus); + if (is_prs_invalid(old_prs)) { + adding = false; + deleting = cpumask_and(tmp->delmask, + newmask, parent->effective_xcpus); + } else { + cpumask_andnot(tmp->addmask, xcpus, newmask); + adding = cpumask_and(tmp->addmask, tmp->addmask, + parent->effective_xcpus);
- cpumask_andnot(tmp->delmask, newmask, xcpus); - deleting = cpumask_and(tmp->delmask, tmp->delmask, - parent->effective_xcpus); + cpumask_andnot(tmp->delmask, newmask, xcpus); + deleting = cpumask_and(tmp->delmask, tmp->delmask, + parent->effective_xcpus); + } /* * Make partition invalid if parent's effective_cpus could * become empty and there are tasks in the parent. @@ -2017,9 +2028,11 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd,
/* * Transitioning between invalid to valid or vice versa may require - * changing CS_CPU_EXCLUSIVE. + * changing CS_CPU_EXCLUSIVE. In the case of partcmd_update, + * validate_change() has already been successfully called and + * CPU lists in cs haven't been updated yet. So defer it to later. */ - if (old_prs != new_prs) { + if ((old_prs != new_prs) && (cmd != partcmd_update)) { int err = update_partition_exclusive(cs, new_prs);
if (err) @@ -2067,6 +2080,9 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd,
spin_unlock_irq(&callback_lock);
+ if ((old_prs != new_prs) && (cmd == partcmd_update)) + update_partition_exclusive(cs, new_prs); + if (adding || deleting) { update_tasks_cpumask(parent, tmp->addmask); update_sibling_cpumasks(parent, cs, tmp); @@ -2463,8 +2479,9 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, if (alloc_cpumasks(NULL, &tmp)) return -ENOMEM;
- if (is_partition_valid(cs)) { - if (cpumask_empty(trialcs->effective_xcpus)) { + if (old_prs) { + if (is_partition_valid(cs) && + cpumask_empty(trialcs->effective_xcpus)) { invalidate = true; cs->prs_err = PERR_INVCPUS; } else if (prstate_housekeeping_conflict(old_prs, trialcs->effective_xcpus)) { @@ -2498,13 +2515,16 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, */ invalidate = true; rcu_read_lock(); - cpuset_for_each_child(cp, css, parent) + cpuset_for_each_child(cp, css, parent) { + struct cpumask *xcpus = fetch_xcpus(trialcs); + if (is_partition_valid(cp) && - cpumask_intersects(trialcs->effective_xcpus, cp->effective_xcpus)) { + cpumask_intersects(xcpus, cp->effective_xcpus)) { rcu_read_unlock(); update_parent_effective_cpumask(cp, partcmd_invalidate, NULL, &tmp); rcu_read_lock(); } + } rcu_read_unlock(); retval = 0; } @@ -2512,18 +2532,24 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, if (retval < 0) goto out_free;
- if (is_partition_valid(cs)) { + if (is_partition_valid(cs) || + (is_partition_invalid(cs) && !invalidate)) { + struct cpumask *xcpus = trialcs->effective_xcpus; + + if (cpumask_empty(xcpus) && is_partition_invalid(cs)) + xcpus = trialcs->cpus_allowed; + /* * Call remote_cpus_update() to handle valid remote partition */ if (is_remote_partition(cs)) - remote_cpus_update(cs, trialcs->effective_xcpus, &tmp); + remote_cpus_update(cs, xcpus, &tmp); else if (invalidate) update_parent_effective_cpumask(cs, partcmd_invalidate, NULL, &tmp); else update_parent_effective_cpumask(cs, partcmd_update, - trialcs->effective_xcpus, &tmp); + xcpus, &tmp); } else if (!cpumask_empty(cs->exclusive_cpus)) { /* * Use trialcs->effective_cpus as a temp cpumask @@ -2600,7 +2626,7 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs, if (retval) return retval;
- if (is_partition_valid(cs)) { + if (old_prs) { if (cpumask_empty(trialcs->effective_xcpus)) { invalidate = true; cs->prs_err = PERR_INVCPUS; @@ -3034,19 +3060,10 @@ static int update_prstate(struct cpuset *cs, int new_prs) return 0;
/* - * For a previously invalid partition root with valid partition root - * parent, treat it as if it is a "member". Otherwise, reject it as - * remote partition cannot currently self-recover from an invalid - * state. + * Treat a previously invalid partition root as if it is a "member". */ - if (new_prs && is_prs_invalid(old_prs)) { - if (is_partition_valid(parent)) { - old_prs = PRS_MEMBER; - } else { - cs->partition_root_state = -new_prs; - return 0; - } - } + if (new_prs && is_prs_invalid(old_prs)) + old_prs = PRS_MEMBER;
if (alloc_cpumasks(NULL, &tmpmask)) return -ENOMEM;