Message-Id: <1539366951-8498-5-git-send-email-longman@redhat.com>
Date:   Fri, 12 Oct 2018 13:55:44 -0400
From:   Waiman Long <longman@...hat.com>
To:     Tejun Heo <tj@...nel.org>, Li Zefan <lizefan@...wei.com>,
        Johannes Weiner <hannes@...xchg.org>,
        Peter Zijlstra <peterz@...radead.org>,
        Ingo Molnar <mingo@...hat.com>
Cc:     cgroups@...r.kernel.org, linux-kernel@...r.kernel.org,
        linux-doc@...r.kernel.org, kernel-team@...com, pjt@...gle.com,
        luto@...capital.net, Mike Galbraith <efault@....de>,
        torvalds@...ux-foundation.org, Roman Gushchin <guro@...com>,
        Juri Lelli <juri.lelli@...hat.com>,
        Patrick Bellasi <patrick.bellasi@....com>,
        Waiman Long <longman@...hat.com>
Subject: [PATCH v13 04/11] cpuset: Add new v2 cpuset.sched.partition flag

A new cpuset.sched.partition boolean flag is added to cpuset v2.
This new flag, if set, indicates that the cgroup is the root of a
new scheduling domain or partition that includes itself and all its
descendants except those that are scheduling domain roots themselves
and their descendants.

With this new flag, one can directly create as many partitions as
necessary without ever using the v1 trick of turning off load balancing
in specific cpusets to create partitions as a side effect.

This new flag is owned by the parent and will cause the CPUs in the
cpuset to be removed from the effective CPUs of its parent.

This is implemented internally by adding a new subparts_cpus mask that
holds the CPUs belonging to child partitions so that:

        subparts_cpus | effective_cpus = cpus_allowed
        subparts_cpus & effective_cpus = 0

This new flag can only be turned on in a cpuset if its parent is a
partition root itself. The state of this flag cannot be changed if the
cpuset has children.

Once turned on, further changes to "cpuset.cpus" is allowed as long
as there is at least one CPU left that can be granted from the parent
and a child partition root cannot use up all the CPUs in the parent's
effective_cpus.

Signed-off-by: Waiman Long <longman@...hat.com>
---
 kernel/cgroup/cpuset.c | 362 +++++++++++++++++++++++++++++++++++++++++++++++--
 1 file changed, 349 insertions(+), 13 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 9ac5f94..9c2e73c 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -964,10 +964,191 @@ static void update_tasks_cpumask(struct cpuset *cs)
 	css_task_iter_end(&it);
 }
 
+/**
+ * compute_effective_cpumask - Compute the effective cpumask of the cpuset
+ * @new_cpus: the temp variable for the new effective_cpus mask
+ * @cs: the cpuset the need to recompute the new effective_cpus mask
+ * @parent: the parent cpuset
+ *
+ * If the parent has subpartition CPUs, include them in the list of
+ * allowable CPUs in computing the new effective_cpus mask.
+ */
+static void compute_effective_cpumask(struct cpumask *new_cpus,
+				      struct cpuset *cs, struct cpuset *parent)
+{
+	if (parent->nr_subparts_cpus) {
+		cpumask_or(new_cpus, parent->effective_cpus,
+			   parent->subparts_cpus);
+		cpumask_and(new_cpus, new_cpus, cs->cpus_allowed);
+	} else {
+		cpumask_and(new_cpus, cs->cpus_allowed, parent->effective_cpus);
+	}
+}
+
+/*
+ * Commands for update_parent_subparts_cpumask
+ */
+enum subparts_cmd {
+	partcmd_enable,		/* Enable partition root	 */
+	partcmd_disable,	/* Disable partition root	 */
+	partcmd_update,		/* Update parent's subparts_cpus */
+};
+
+/**
+ * update_parent_subparts_cpumask - update subparts_cpus mask of parent cpuset
+ * @cpuset:  The cpuset that requests change in partition root state
+ * @cmd:     Partition root state change command
+ * @newmask: Optional new cpumask for partcmd_update
+ * @tmp:     Temporary addmask and delmask
+ * Return:   0, 1 or an error code
+ *
+ * For partcmd_enable, the cpuset is being transformed from a non-partition
+ * root to a partition root. The cpus_allowed mask of the given cpuset will
+ * be put into parent's subparts_cpus and taken away from parent's
+ * effective_cpus. The function will return 0 if all the CPUs listed in
+ * cpus_allowed can be granted or an error code will be returned.
+ *
+ * For partcmd_disable, the cpuset is being transofrmed from a partition
+ * root back to a non-partition root. any CPUs in cpus_allowed that are in
+ * parent's subparts_cpus will be taken away from that cpumask and put back
+ * into parent's effective_cpus. 0 should always be returned.
+ *
+ * For partcmd_update, if the optional newmask is specified, the cpu
+ * list is to be changed from cpus_allowed to newmask. Otherwise,
+ * cpus_allowed is assumed to remain the same.  The function will return
+ * 1 if changes to parent's subparts_cpus and effective_cpus happen or 0
+ * otherwise. In case of error, an error code will be returned.
+ *
+ * The partcmd_enable and partcmd_disable commands are used by
+ * update_prstate(). The partcmd_update command is used by
+ * update_cpumasks_hier() with newmask NULL and update_cpumask() with
+ * newmask set.
+ *
+ * The checking is more strict when enabling partition root than the
+ * other two commands.
+ *
+ * Because of the implicit cpu exclusive nature of a partition root,
+ * cpumask changes that violates the cpu exclusivity rule will not be
+ * permitted when checked by validate_change(). The validate_change()
+ * function will also prevent any changes to the cpu list if it is not
+ * a superset of children's cpu lists.
+ *
+ * Called with cpuset_mutex held.
+ */
+static int update_parent_subparts_cpumask(struct cpuset *cpuset, int cmd,
+					  struct cpumask *newmask,
+					  struct tmpmasks *tmp)
+{
+	struct cpuset *parent = parent_cs(cpuset);
+	int adding;	/* Moving cpus from effective_cpus to subparts_cpus */
+	int deleting;	/* Moving cpus from subparts_cpus to effective_cpus */
+
+	/*
+	 * The parent must be a partition root.
+	 * The new cpumask, if present, or the current cpus_allowed must
+	 * not be empty.
+	 */
+	if (!is_partition_root(parent) ||
+	   (newmask && cpumask_empty(newmask)) ||
+	   (!newmask && cpumask_empty(cpuset->cpus_allowed)))
+		return -EINVAL;
+
+	/*
+	 * Enabling/disabling partition root is not allowed if there are
+	 * online children.
+	 */
+	if ((cmd != partcmd_update) && css_has_online_children(&cpuset->css))
+		return -EBUSY;
+
+	/*
+	 * Enabling partition root is not allowed if not all the CPUs
+	 * can be granted from parent's effective_cpus or at least one
+	 * CPU will be left after that.
+	 */
+	if ((cmd == partcmd_enable) &&
+	   (!cpumask_subset(cpuset->cpus_allowed, parent->effective_cpus) ||
+	     cpumask_equal(cpuset->cpus_allowed, parent->effective_cpus)))
+		return -EINVAL;
+
+	/*
+	 * A cpumask update cannot make parent's effective_cpus become empty.
+	 */
+	adding = deleting = false;
+	if (cmd == partcmd_enable) {
+		cpumask_copy(tmp->addmask, cpuset->cpus_allowed);
+		adding = true;
+	} else if (cmd == partcmd_disable) {
+		deleting = cpumask_and(tmp->delmask, cpuset->cpus_allowed,
+				       parent->subparts_cpus);
+	} else if (newmask) {
+		/*
+		 * partcmd_update with newmask:
+		 *
+		 * delmask = cpus_allowed & ~newmask & parent->subparts_cpus
+		 * addmask = newmask & parent->effective_cpus
+		 *		     & ~parent->subparts_cpus
+		 */
+		cpumask_andnot(tmp->delmask, cpuset->cpus_allowed, newmask);
+		deleting = cpumask_and(tmp->delmask, tmp->delmask,
+				       parent->subparts_cpus);
+
+		cpumask_and(tmp->addmask, newmask, parent->effective_cpus);
+		adding = cpumask_andnot(tmp->addmask, tmp->addmask,
+					parent->subparts_cpus);
+		/*
+		 * Return error if the new effective_cpus could become empty.
+		 */
+		if (adding && !deleting &&
+		    cpumask_equal(parent->effective_cpus, tmp->addmask))
+			return -EINVAL;
+	} else {
+		/*
+		 * partcmd_update w/o newmask:
+		 *
+		 * addmask = cpus_allowed & parent->effectiveb_cpus
+		 *
+		 * Note that parent's subparts_cpus may have been
+		 * pre-shrunk in case the CPUs granted to the parent
+		 * by the grandparent changes. So no deletion is needed.
+		 */
+		adding = cpumask_and(tmp->addmask, cpuset->cpus_allowed,
+				     parent->effective_cpus);
+		if (cpumask_equal(tmp->addmask, parent->effective_cpus))
+			return -EINVAL;
+	}
+
+	if (!adding && !deleting)
+		return 0;
+
+	/*
+	 * Change the parent's subparts_cpus.
+	 * Newly added CPUs will be removed from effective_cpus and
+	 * newly deleted ones will be added back to effective_cpus.
+	 */
+	spin_lock_irq(&callback_lock);
+	if (adding) {
+		cpumask_or(parent->subparts_cpus,
+			   parent->subparts_cpus, tmp->addmask);
+		cpumask_andnot(parent->effective_cpus,
+			       parent->effective_cpus, tmp->addmask);
+	}
+	if (deleting) {
+		cpumask_andnot(parent->subparts_cpus,
+			       parent->subparts_cpus, tmp->delmask);
+		cpumask_or(parent->effective_cpus,
+			   parent->effective_cpus, tmp->delmask);
+	}
+
+	parent->nr_subparts_cpus = cpumask_weight(parent->subparts_cpus);
+	spin_unlock_irq(&callback_lock);
+
+	return cmd == partcmd_update;
+}
+
 /*
  * update_cpumasks_hier - Update effective cpumasks and tasks in the subtree
- * @cs: the cpuset to consider
- * @new_cpus: temp variable for calculating new effective_cpus
+ * @cs:  the cpuset to consider
+ * @tmp: temp variables for calculating effective_cpus & partition setup
  *
  * When congifured cpumask is changed, the effective cpumasks of this cpuset
  * and all its descendants need to be updated.
@@ -976,7 +1157,7 @@ static void update_tasks_cpumask(struct cpuset *cs)
  *
  * Called with cpuset_mutex held
  */
-static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus)
+static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp)
 {
 	struct cpuset *cp;
 	struct cgroup_subsys_state *pos_css;
@@ -985,28 +1166,61 @@ static void update_cpumasks_hier(struct cpuset *cs, struct cpumask *new_cpus)
 	rcu_read_lock();
 	cpuset_for_each_descendant_pre(cp, pos_css, cs) {
 		struct cpuset *parent = parent_cs(cp);
+		bool cs_empty;
 
-		cpumask_and(new_cpus, cp->cpus_allowed, parent->effective_cpus);
+		compute_effective_cpumask(tmp->new_cpus, cp, parent);
+		cs_empty = cpumask_empty(tmp->new_cpus);
+
+		/*
+		 * A partition root cannot have empty effective_cpus
+		 */
+		WARN_ON_ONCE(cs_empty && is_partition_root(cp));
 
 		/*
 		 * If it becomes empty, inherit the effective mask of the
 		 * parent, which is guaranteed to have some CPUs.
 		 */
-		if (is_in_v2_mode() && cpumask_empty(new_cpus))
-			cpumask_copy(new_cpus, parent->effective_cpus);
+		if (is_in_v2_mode() && cs_empty)
+			cpumask_copy(tmp->new_cpus, parent->effective_cpus);
 
-		/* Skip the whole subtree if the cpumask remains the same. */
-		if (cpumask_equal(new_cpus, cp->effective_cpus)) {
+		/*
+		 * Skip the whole subtree if the cpumask remains the same
+		 * and has no partition root state.
+		 */
+		if (!is_partition_root(cp) &&
+		    cpumask_equal(tmp->new_cpus, cp->effective_cpus)) {
 			pos_css = css_rightmost_descendant(pos_css);
 			continue;
 		}
 
+		/*
+		 * update_parent_subparts_cpumask() should have been called
+		 * for cs already in update_cpumask(). We should also call
+		 * update_tasks_cpumask() again for tasks in the parent
+		 * cpuset if the parent's subparts_cpus changes.
+		 */
+		if ((cp != cs) && cp->partition_root_state &&
+		    update_parent_subparts_cpumask(cp, partcmd_update,
+						   NULL, tmp)) {
+			if (parent != &top_cpuset)
+				update_tasks_cpumask(parent);
+		}
+
 		if (!css_tryget_online(&cp->css))
 			continue;
 		rcu_read_unlock();
 
 		spin_lock_irq(&callback_lock);
-		cpumask_copy(cp->effective_cpus, new_cpus);
+
+		cpumask_copy(cp->effective_cpus, tmp->new_cpus);
+		if (cp->nr_subparts_cpus) {
+			/*
+			 * Make sure that effective_cpus & subparts_cpus
+			 * are mutually exclusive.
+			 */
+			cpumask_andnot(cp->effective_cpus, cp->effective_cpus,
+				       cp->subparts_cpus);
+		}
 		spin_unlock_irq(&callback_lock);
 
 		WARN_ON(!is_in_v2_mode() &&
@@ -1041,6 +1255,7 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 			  const char *buf)
 {
 	int retval;
+	struct tmpmasks tmp;
 
 	/* top_cpuset.cpus_allowed tracks cpu_online_mask; it's read-only */
 	if (cs == &top_cpuset)
@@ -1072,12 +1287,36 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	if (retval < 0)
 		return retval;
 
+	/*
+	 * Use the cpumasks in trialcs for tmpmasks.
+	 */
+	tmp.addmask  = trialcs->subparts_cpus;
+	tmp.delmask  = trialcs->effective_cpus;
+	tmp.new_cpus = trialcs->cpus_allowed;
+
+	if (cs->partition_root_state) {
+		/* Cpumask of a partition root cannot be empty */
+		if (cpumask_empty(trialcs->cpus_allowed))
+			return -EINVAL;
+		if (update_parent_subparts_cpumask(cs, partcmd_update,
+					trialcs->cpus_allowed, &tmp) < 0)
+			return -EINVAL;
+	}
+
 	spin_lock_irq(&callback_lock);
 	cpumask_copy(cs->cpus_allowed, trialcs->cpus_allowed);
+
+	/*
+	 * Make sure that subparts_cpus is a subset of cpus_allowed.
+	 */
+	if (cs->nr_subparts_cpus) {
+		cpumask_andnot(cs->subparts_cpus, cs->subparts_cpus,
+			       cs->cpus_allowed);
+		cs->nr_subparts_cpus = cpumask_weight(cs->subparts_cpus);
+	}
 	spin_unlock_irq(&callback_lock);
 
-	/* use trialcs->cpus_allowed as a temp variable */
-	update_cpumasks_hier(cs, trialcs->cpus_allowed);
+	update_cpumasks_hier(cs, &tmp);
 	return 0;
 }
 
@@ -1436,6 +1675,80 @@ static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
 }
 
 /*
+ * update_prstate - update partititon_root_state
+ * cs:	the cpuset to update
+ * val: 0 - disabled, 1 - enabled
+ *
+ * Call with cpuset_mutex held.
+ */
+static int update_prstate(struct cpuset *cs, int val)
+{
+	int err;
+	struct cpuset *parent = parent_cs(cs);
+	struct tmpmasks tmp;
+
+	if ((val != 0) && (val != 1))
+		return -EINVAL;
+	if (val == cs->partition_root_state)
+		return 0;
+
+	/*
+	 * Cannot force a partial or erroneous partition root to a full
+	 * partition root.
+	 */
+	if (val && cs->partition_root_state)
+		return -EINVAL;
+
+	if (alloc_cpumasks(NULL, &tmp))
+		return -ENOMEM;
+
+	err = -EINVAL;
+	if (!cs->partition_root_state) {
+		/*
+		 * Turning on partition root requires setting the
+		 * CS_CPU_EXCLUSIVE bit implicitly as well and cpus_allowed
+		 * cannot be NULL.
+		 */
+		if (cpumask_empty(cs->cpus_allowed))
+			goto out;
+
+		err = update_flag(CS_CPU_EXCLUSIVE, cs, 1);
+		if (err)
+			goto out;
+
+		err = update_parent_subparts_cpumask(cs, partcmd_enable,
+						     NULL, &tmp);
+		if (err) {
+			update_flag(CS_CPU_EXCLUSIVE, cs, 0);
+			goto out;
+		}
+		cs->partition_root_state = PRS_ENABLED;
+	} else {
+		err = update_parent_subparts_cpumask(cs, partcmd_disable,
+						     NULL, &tmp);
+		if (err)
+			goto out;
+
+		cs->partition_root_state = 0;
+
+		/* Turning off CS_CPU_EXCLUSIVE will not return error */
+		update_flag(CS_CPU_EXCLUSIVE, cs, 0);
+	}
+
+	/*
+	 * Update cpumask of parent's tasks except when it is the top
+	 * cpuset as some system daemons cannot be mapped to other CPUs.
+	 */
+	if (parent != &top_cpuset)
+		update_tasks_cpumask(parent);
+
+	rebuild_sched_domains_locked();
+out:
+	free_cpumasks(&tmp);
+	return err;
+}
+
+/*
  * Frequency meter - How fast is some event occurring?
  *
  * These routines manage a digitally filtered, constant time based,
@@ -1680,6 +1993,7 @@ static void cpuset_attach(struct cgroup_taskset *tset)
 	FILE_MEM_EXCLUSIVE,
 	FILE_MEM_HARDWALL,
 	FILE_SCHED_LOAD_BALANCE,
+	FILE_PARTITION_ROOT,
 	FILE_SCHED_RELAX_DOMAIN_LEVEL,
 	FILE_MEMORY_PRESSURE_ENABLED,
 	FILE_MEMORY_PRESSURE,
@@ -1749,6 +2063,9 @@ static int cpuset_write_s64(struct cgroup_subsys_state *css, struct cftype *cft,
 	case FILE_SCHED_RELAX_DOMAIN_LEVEL:
 		retval = update_relax_domain_level(cs, val);
 		break;
+	case FILE_PARTITION_ROOT:
+		retval = update_prstate(cs, val);
+		break;
 	default:
 		retval = -EINVAL;
 		break;
@@ -1899,6 +2216,8 @@ static s64 cpuset_read_s64(struct cgroup_subsys_state *css, struct cftype *cft)
 	switch (type) {
 	case FILE_SCHED_RELAX_DOMAIN_LEVEL:
 		return cs->relax_domain_level;
+	case FILE_PARTITION_ROOT:
+		return cs->partition_root_state;
 	default:
 		BUG();
 	}
@@ -2050,6 +2369,14 @@ static s64 cpuset_read_s64(struct cgroup_subsys_state *css, struct cftype *cft)
 		.flags = CFTYPE_NOT_ON_ROOT,
 	},
 
+	{
+		.name = "sched.partition",
+		.read_s64 = cpuset_read_s64,
+		.write_s64 = cpuset_write_s64,
+		.private = FILE_PARTITION_ROOT,
+		.flags = CFTYPE_NOT_ON_ROOT,
+	},
+
 	{ }	/* terminate */
 };
 
@@ -2151,7 +2478,12 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
 /*
  * If the cpuset being removed has its flag 'sched_load_balance'
  * enabled, then simulate turning sched_load_balance off, which
- * will call rebuild_sched_domains_locked().
+ * will call rebuild_sched_domains_locked(). That is not needed
+ * in the default hierarchy where only changes in partition
+ * will cause repartitioning.
+ *
+ * If the cpuset has the 'sched.partition' flag enabled, simulate
+ * turning 'sched.partition" off.
  */
 
 static void cpuset_css_offline(struct cgroup_subsys_state *css)
@@ -2160,7 +2492,11 @@ static void cpuset_css_offline(struct cgroup_subsys_state *css)
 
 	mutex_lock(&cpuset_mutex);
 
-	if (is_sched_load_balance(cs))
+	if (is_partition_root(cs))
+		update_prstate(cs, 0);
+
+	if (!cgroup_subsys_on_dfl(cpuset_cgrp_subsys) &&
+	    is_sched_load_balance(cs))
 		update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
 
 	cpuset_dec();
-- 
1.8.3.1

Powered by blists - more mailing lists