From: Waiman Long <longman@redhat.com> mainline inclusion from mainline-v6.15 commit f62a5d39368e34a966c8df63e1f05eed7fe9c5de category: bugfix bugzilla: https://atomgit.com/openeuler/kernel/issues/9899 Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i... -------------------------------- Currently, changes in exclusive CPUs are being handled in remote_partition_check() by disabling conflicting remote partitions. However, that may lead to results unexpected by the users. Fix this problem by removing remote_partition_check() and making update_cpumasks_hier() handle changes in descendant remote partitions properly. The compute_effective_exclusive_cpumask() function is enhanced to check the exclusive_cpus and effective_xcpus from siblings and excluded them in its effective exclusive CPUs computation and return a value to show if there is any sibling conflicts. This is somewhat like the cpu_exclusive flag check in validate_change(). This is the initial step to enable us to retire the use of cpu_exclusive flag in cgroup v2 in the future. One of the tests in the TEST_MATRIX of the test_cpuset_prs.sh script has to be updated due to changes in the way a child remote partition root is being handled (updated instead of invalidation) in update_cpumasks_hier(). Signed-off-by: Waiman Long <longman@redhat.com> Signed-off-by: Tejun Heo <tj@kernel.org> Conflicts: kernel/cgroup/cpuset.c tools/testing/selftests/cgroup/test_cpuset_prs.sh [ The partition_xcpus_add/del() helpers and isolcpus tracking are not backported; remote_partition_enable/disable() and remote_cpus_update() keep the direct cpumask operations. remote_cpus_update() takes the new 4-argument signature (cs, xcpus, excpus, tmp) and updates effective_xcpus/exclusive_cpus within the lock. The selftest is kept as-is (the OLK test matrix predates the upstream changes). ] Signed-off-by: Zhang Qiao <zhangqiao22@huawei.com> --- kernel/cgroup/cpuset.c | 270 +++++++++++++++++++++++------------------ 1 file changed, 154 insertions(+), 116 deletions(-) diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c index 7e572283100b..e0aa3a3c5523 100644 --- a/kernel/cgroup/cpuset.c +++ b/kernel/cgroup/cpuset.c @@ -237,6 +237,17 @@ static cpumask_var_t subpartitions_cpus; /* List of remote partition root children */ static struct list_head remote_children; +/* + * A flag to force sched domain rebuild at the end of an operation. + * It can be set in + * - update_partition_sd_lb() + * - update_cpumasks_hier() + * - cpuset_update_flag() + * - cpuset_hotplug_update_tasks() + * - cpuset_handle_hotplug() + * + * Protected by cpuset_mutex (with cpus_read_lock held) or cpus_write_lock. + */ /* * Partition root states: * @@ -1658,20 +1669,57 @@ static void reset_partition_data(struct cpuset *cs) * compute_effective_exclusive_cpumask - compute effective exclusive CPUs * @cs: cpuset * @xcpus: effective exclusive CPUs value to be set - * Return: true if xcpus is not empty, false otherwise. + * @real_cs: the real cpuset (can be NULL) + * Return: 0 if there is no sibling conflict, > 0 otherwise * - * Starting with exclusive_cpus (cpus_allowed if exclusive_cpus is not set), - * it must be a subset of parent's effective_xcpus. + * If exclusive_cpus isn't explicitly set or a real_cs is provided, we have to + * scan the sibling cpusets and exclude their exclusive_cpus or effective_xcpus + * as well. The provision of real_cs means that a cpumask is being changed and + * the given cs is a trial one. */ -static bool compute_effective_exclusive_cpumask(struct cpuset *cs, - struct cpumask *xcpus) +static int compute_effective_exclusive_cpumask(struct cpuset *cs, + struct cpumask *xcpus, + struct cpuset *real_cs) { + struct cgroup_subsys_state *css; struct cpuset *parent = parent_cs(cs); + struct cpuset *sibling; + int retval = 0; if (!xcpus) xcpus = cs->effective_xcpus; - return cpumask_and(xcpus, user_xcpus(cs), parent->effective_xcpus); + cpumask_and(xcpus, user_xcpus(cs), parent->effective_xcpus); + + if (!real_cs) { + if (!cpumask_empty(cs->exclusive_cpus)) + return 0; + } else { + cs = real_cs; + } + + /* + * Exclude exclusive CPUs from siblings + */ + rcu_read_lock(); + cpuset_for_each_child(sibling, css, parent) { + if (sibling == cs) + continue; + + if (!cpumask_empty(sibling->exclusive_cpus) && + cpumask_intersects(xcpus, sibling->exclusive_cpus)) { + cpumask_andnot(xcpus, xcpus, sibling->exclusive_cpus); + retval++; + continue; + } + if (!cpumask_empty(sibling->effective_xcpus) && + cpumask_intersects(xcpus, sibling->effective_xcpus)) { + cpumask_andnot(xcpus, xcpus, sibling->effective_xcpus); + retval++; + } + } + rcu_read_unlock(); + return retval; } static inline bool is_remote_partition(struct cpuset *cs) @@ -1709,7 +1757,7 @@ static int remote_partition_enable(struct cpuset *cs, struct tmpmasks *tmp) * remote partition root underneath it, its exclusive_cpus must * have overlapped with subpartitions_cpus. */ - compute_effective_exclusive_cpumask(cs, tmp->new_cpus); + compute_effective_exclusive_cpumask(cs, tmp->new_cpus, NULL); if (cpumask_empty(tmp->new_cpus) || cpumask_intersects(tmp->new_cpus, subpartitions_cpus) || cpumask_subset(top_cpuset.effective_cpus, tmp->new_cpus)) @@ -1728,7 +1776,9 @@ static int remote_partition_enable(struct cpuset *cs, struct tmpmasks *tmp) parent->child_ecpus_count--; } list_add(&cs->remote_sibling, &remote_children); + cpumask_copy(cs->effective_xcpus, tmp->new_cpus); spin_unlock_irq(&callback_lock); + cpuset_force_rebuild(); cs->prs_err = 0; /* @@ -1751,9 +1801,8 @@ static int remote_partition_enable(struct cpuset *cs, struct tmpmasks *tmp) */ static void remote_partition_disable(struct cpuset *cs, struct tmpmasks *tmp) { - compute_effective_exclusive_cpumask(cs, tmp->new_cpus); WARN_ON_ONCE(!is_remote_partition(cs)); - WARN_ON_ONCE(!cpumask_subset(tmp->new_cpus, subpartitions_cpus)); + WARN_ON_ONCE(!cpumask_subset(cs->effective_xcpus, subpartitions_cpus)); spin_lock_irq(&callback_lock); cpumask_andnot(subpartitions_cpus, @@ -1767,8 +1816,12 @@ static void remote_partition_disable(struct cpuset *cs, struct tmpmasks *tmp) cs->partition_root_state = -cs->partition_root_state; else cs->partition_root_state = PRS_MEMBER; + + /* effective_xcpus may need to be changed */ + compute_effective_exclusive_cpumask(cs, NULL, NULL); reset_partition_data(cs); spin_unlock_irq(&callback_lock); + cpuset_force_rebuild(); /* * Proprogate changes in top_cpuset's effective_cpus down the hierarchy. @@ -1780,14 +1833,15 @@ static void remote_partition_disable(struct cpuset *cs, struct tmpmasks *tmp) /* * remote_cpus_update - cpus_exclusive change of remote partition * @cs: the cpuset to be updated - * @newmask: the new effective_xcpus mask - * @tmp: temparary masks + * @xcpus: the new exclusive_cpus mask, if non-NULL + * @excpus: the new effective_xcpus mask + * @tmp: temporary masks * * top_cpuset and subpartitions_cpus will be updated or partition can be * invalidated. */ -static void remote_cpus_update(struct cpuset *cs, struct cpumask *newmask, - struct tmpmasks *tmp) +static void remote_cpus_update(struct cpuset *cs, struct cpumask *xcpus, + struct cpumask *excpus, struct tmpmasks *tmp) { bool adding, deleting; @@ -1796,13 +1850,13 @@ static void remote_cpus_update(struct cpuset *cs, struct cpumask *newmask, WARN_ON_ONCE(!cpumask_subset(cs->effective_xcpus, subpartitions_cpus)); - if (cpumask_empty(newmask)) { + if (cpumask_empty(excpus)) { cs->prs_err = PERR_CPUSEMPTY; goto invalidate; } - adding = cpumask_andnot(tmp->addmask, newmask, cs->effective_xcpus); - deleting = cpumask_andnot(tmp->delmask, cs->effective_xcpus, newmask); + adding = cpumask_andnot(tmp->addmask, excpus, cs->effective_xcpus); + deleting = cpumask_andnot(tmp->delmask, cs->effective_xcpus, excpus); /* * Additions of remote CPUs is only allowed if those CPUs are @@ -1834,7 +1888,16 @@ static void remote_cpus_update(struct cpuset *cs, struct cpumask *newmask, cpumask_or(top_cpuset.effective_cpus, top_cpuset.effective_cpus, tmp->delmask); } + /* + * Need to update effective_xcpus and exclusive_cpus now as + * update_sibling_cpumasks() below may iterate back to the same cs. + */ + cpumask_copy(cs->effective_xcpus, excpus); + if (xcpus) + cpumask_copy(cs->exclusive_cpus, xcpus); spin_unlock_irq(&callback_lock); + if (adding || deleting) + cpuset_force_rebuild(); /* * Proprogate changes in top_cpuset's effective_cpus down the hierarchy. @@ -1847,47 +1910,6 @@ static void remote_cpus_update(struct cpuset *cs, struct cpumask *newmask, remote_partition_disable(cs, tmp); } -/* - * remote_partition_check - check if a child remote partition needs update - * @cs: the cpuset to be updated - * @newmask: the new effective_xcpus mask - * @delmask: temporary mask for deletion (not in tmp) - * @tmp: temparary masks - * - * This should be called before the given cs has updated its cpus_allowed - * and/or effective_xcpus. - */ -static void remote_partition_check(struct cpuset *cs, struct cpumask *newmask, - struct cpumask *delmask, struct tmpmasks *tmp) -{ - struct cpuset *child, *next; - int disable_cnt = 0; - - /* - * Compute the effective exclusive CPUs that will be deleted. - */ - if (!cpumask_andnot(delmask, cs->effective_xcpus, newmask) || - !cpumask_intersects(delmask, subpartitions_cpus)) - return; /* No deletion of exclusive CPUs in partitions */ - - /* - * Searching the remote children list to look for those that will - * be impacted by the deletion of exclusive CPUs. - * - * Since a cpuset must be removed from the remote children list - * before it can go offline and holding cpuset_mutex will prevent - * any change in cpuset status. RCU read lock isn't needed. - */ - lockdep_assert_held(&cpuset_mutex); - list_for_each_entry_safe(child, next, &remote_children, remote_sibling) - if (cpumask_intersects(child->effective_cpus, delmask)) { - remote_partition_disable(child, tmp); - disable_cnt++; - } - if (disable_cnt) - rebuild_sched_domains_locked(); -} - /* * prstate_housekeeping_conflict - check for partition & housekeeping conflicts * @prstate: partition root state to be checked @@ -1959,6 +1981,7 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd, bool nocpu; lockdep_assert_held(&cpuset_mutex); + WARN_ON_ONCE(is_remote_partition(cs)); /* * new_prs will only be changed for the partcmd_update and @@ -2000,13 +2023,20 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd, nocpu = tasks_nocpu_error(parent, cs, xcpus); if (cmd == partcmd_enable) { + /* + * Need to call compute_effective_exclusive_cpumask() in case + * exclusive_cpus not set. Sibling conflict should only happen + * if exclusive_cpus isn't set. + */ + xcpus = tmp->new_cpus; + if (compute_effective_exclusive_cpumask(cs, xcpus, NULL)) + WARN_ON_ONCE(!cpumask_empty(cs->exclusive_cpus)); + /* * Enabling partition root is not allowed if its - * effective_xcpus is empty or doesn't overlap with - * parent's effective_xcpus. + * effective_xcpus is empty. */ - if (cpumask_empty(xcpus) || - !cpumask_intersects(xcpus, parent->effective_xcpus)) + if (cpumask_empty(xcpus)) return PERR_INVCPUS; if (prstate_housekeeping_conflict(new_prs, xcpus)) @@ -2024,13 +2054,17 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd, subparts_delta++; } else if (cmd == partcmd_disable) { /* - * May need to add cpus to parent's effective_cpus for - * valid partition root. + * May need to add cpus back to parent's effective_cpus + * (and maybe removed from subpartitions_cpus/isolated_cpus) + * for valid partition root. xcpus may contain CPUs that + * shouldn't be removed from the two global cpumasks. */ - adding = !is_prs_invalid(old_prs) && - cpumask_and(tmp->addmask, xcpus, parent->effective_xcpus); - if (adding) + if (is_partition_valid(cs)) { + cpumask_copy(tmp->addmask, cs->effective_xcpus); + adding = true; subparts_delta--; + } + new_prs = PRS_MEMBER; } else if (newmask) { /* * Empty cpumask is not allowed @@ -2039,6 +2073,7 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd, part_error = PERR_CPUSEMPTY; goto write_error; } + /* Check newmask again, whether cpus are available for parent/cs */ nocpu |= tasks_nocpu_error(parent, cs, newmask); @@ -2265,7 +2300,7 @@ static void compute_partition_effective_cpumask(struct cpuset *cs, * 2) All the effective_cpus will be used up and cp * has tasks */ - compute_effective_exclusive_cpumask(cs, new_ecpus); + compute_effective_exclusive_cpumask(cs, new_ecpus, NULL); cpumask_and(new_ecpus, new_ecpus, cpu_active_mask); rcu_read_lock(); @@ -2273,6 +2308,11 @@ static void compute_partition_effective_cpumask(struct cpuset *cs, if (!is_partition_valid(child)) continue; + /* + * There shouldn't be a remote partition underneath another + * partition root. + */ + WARN_ON_ONCE(is_remote_partition(child)); child->prs_err = 0; if (!cpumask_subset(child->effective_xcpus, cs->effective_xcpus)) @@ -2334,32 +2374,39 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp, bool remote = is_remote_partition(cp); bool update_parent = false; + old_prs = new_prs = cp->partition_root_state; + /* - * Skip descendent remote partition that acquires CPUs - * directly from top cpuset unless it is cs. + * For child remote partition root (!= cs), we need to call + * remote_cpus_update() if effective_xcpus will be changed. + * Otherwise, we can skip the whole subtree. + * + * remote_cpus_update() will reuse tmp->new_cpus only after + * its value is being processed. */ if (remote && (cp != cs)) { - pos_css = css_rightmost_descendant(pos_css); - continue; - } + compute_effective_exclusive_cpumask(cp, tmp->new_cpus, NULL); + if (cpumask_equal(cp->effective_xcpus, tmp->new_cpus)) { + pos_css = css_rightmost_descendant(pos_css); + continue; + } + rcu_read_unlock(); + remote_cpus_update(cp, NULL, tmp->new_cpus, tmp); + rcu_read_lock(); - /* - * Update effective_xcpus if exclusive_cpus set. - * The case when exclusive_cpus isn't set is handled later. - */ - if (!cpumask_empty(cp->exclusive_cpus) && (cp != cs)) { - spin_lock_irq(&callback_lock); - compute_effective_exclusive_cpumask(cp, NULL); - spin_unlock_irq(&callback_lock); + /* Remote partition may be invalidated */ + new_prs = cp->partition_root_state; + remote = (new_prs == old_prs); } - old_prs = new_prs = cp->partition_root_state; - if (remote || (is_partition_valid(parent) && - is_partition_valid(cp))) + if (remote || (is_partition_valid(parent) && is_partition_valid(cp))) compute_partition_effective_cpumask(cp, tmp->new_cpus); else compute_effective_cpumask(tmp->new_cpus, cp, parent); + if (remote) + goto get_css; /* Ready to update cpuset data */ + /* * A partition with no effective_cpus is allowed as long as * there is no task associated with it. Call @@ -2388,9 +2435,6 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp, parent->child_ecpus_count--; } - if (remote) - goto get_css; - /* * Skip the whole subtree if * 1) the cpumask remains the same, @@ -2451,6 +2495,9 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp, spin_lock_irq(&callback_lock); cpumask_copy(cp->effective_cpus, tmp->new_cpus); cp->partition_root_state = new_prs; + if (!cpumask_empty(cp->exclusive_cpus) && (cp != cs)) + compute_effective_exclusive_cpumask(cp, NULL, NULL); + /* * Make sure effective_xcpus is properly set for a valid * partition root. @@ -2543,7 +2590,14 @@ static void update_sibling_cpumasks(struct cpuset *parent, struct cpuset *cs, parent); if (cpumask_equal(tmp->new_cpus, sibling->effective_cpus)) continue; + } else if (is_remote_partition(sibling)) { + /* + * Change in a sibling cpuset won't affect a remote + * partition root. + */ + continue; } + if (!css_tryget_online(&sibling->css)) continue; @@ -2600,8 +2654,9 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, * trialcs->effective_xcpus is used as a temporary cpumask * for checking validity of the partition root. */ + trialcs->partition_root_state = PRS_MEMBER; if (!cpumask_empty(trialcs->exclusive_cpus) || is_partition_valid(cs)) - compute_effective_exclusive_cpumask(trialcs, NULL); + compute_effective_exclusive_cpumask(trialcs, NULL, cs); } /* Nothing to do if the cpus didn't change */ @@ -2675,19 +2730,13 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs, * Call remote_cpus_update() to handle valid remote partition */ if (is_remote_partition(cs)) - remote_cpus_update(cs, xcpus, &tmp); + remote_cpus_update(cs, NULL, xcpus, &tmp); else if (invalidate) update_parent_effective_cpumask(cs, partcmd_invalidate, NULL, &tmp); else update_parent_effective_cpumask(cs, partcmd_update, xcpus, &tmp); - } else if (!cpumask_empty(cs->exclusive_cpus)) { - /* - * Use trialcs->effective_cpus as a temp cpumask - */ - remote_partition_check(cs, trialcs->effective_xcpus, - trialcs->effective_cpus, &tmp); } spin_lock_irq(&callback_lock); @@ -2739,8 +2788,15 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs, if (cpumask_equal(cs->exclusive_cpus, trialcs->exclusive_cpus)) return 0; - if (*buf) - compute_effective_exclusive_cpumask(trialcs, NULL); + if (*buf) { + trialcs->partition_root_state = PRS_MEMBER; + /* + * Reject the change if there is exclusive CPUs conflict with + * the siblings. + */ + if (compute_effective_exclusive_cpumask(trialcs, NULL, cs)) + return -EINVAL; + } /* * Check all the descendants in update_cpumasks_hier() if @@ -2772,8 +2828,8 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs, if (invalidate) remote_partition_disable(cs, &tmp); else - remote_cpus_update(cs, trialcs->effective_xcpus, - &tmp); + remote_cpus_update(cs, trialcs->exclusive_cpus, + trialcs->effective_xcpus, &tmp); } else if (invalidate) { update_parent_effective_cpumask(cs, partcmd_invalidate, NULL, &tmp); @@ -2781,12 +2837,6 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs, update_parent_effective_cpumask(cs, partcmd_update, trialcs->effective_xcpus, &tmp); } - } else if (!cpumask_empty(trialcs->exclusive_cpus)) { - /* - * Use trialcs->effective_cpus as a temp cpumask - */ - remote_partition_check(cs, trialcs->effective_xcpus, - trialcs->effective_cpus, &tmp); } spin_lock_irq(&callback_lock); cpumask_copy(cs->exclusive_cpus, trialcs->exclusive_cpus); @@ -3203,17 +3253,6 @@ static int update_prstate(struct cpuset *cs, int new_prs) if (alloc_cpumasks(NULL, &tmpmask)) return -ENOMEM; - /* - * Setup effective_xcpus if not properly set yet, it will be cleared - * later if partition becomes invalid. - */ - if ((new_prs > 0) && cpumask_empty(cs->exclusive_cpus)) { - spin_lock_irq(&callback_lock); - cpumask_and(cs->effective_xcpus, - cs->cpus_allowed, parent->effective_xcpus); - spin_unlock_irq(&callback_lock); - } - err = update_partition_exclusive(cs, new_prs); if (err) goto out; @@ -4665,7 +4704,6 @@ static void cpuset_hotplug_update_tasks(struct cpuset *cs, struct tmpmasks *tmp) remote_partition_disable(cs, tmp); compute_effective_cpumask(&new_cpus, cs, parent); remote = false; - cpuset_force_rebuild(); } /* -- 2.18.0