x86/intel_rdt/cqm: Improve limbo list processing

During a mkdir, the entire limbo list is synchronously checked on each
package for free RMIDs by sending IPIs. With a large number of RMIDs (SKL
has 192) this creates a intolerable amount of work in IPIs.

Replace the IPI based checking of the limbo list with asynchronous worker
threads on each package which periodically scan the limbo list and move the
RMIDs that have:

	llc_occupancy < threshold_occupancy

on all packages to the free list.

mkdir now returns -ENOSPC if the free list and the limbo list ere empty or
returns -EBUSY if there are RMIDs on the limbo list and the free list is
empty.

Getting rid of the IPIs also simplifies the data structures and the
serialization required for handling the lists.

[ tglx: Rewrote changelog ... ]

Signed-off-by: Vikas Shivappa <vikas.shivappa@linux.intel.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: ravi.v.shankar@intel.com
Cc: tony.luck@intel.com
Cc: fenghua.yu@intel.com
Cc: peterz@infradead.org
Cc: eranian@google.com
Cc: vikas.shivappa@intel.com
Cc: ak@linux.intel.com
Cc: davidcc@google.com
Link: http://lkml.kernel.org/r/1502845243-20454-3-git-send-email-vikas.shivappa@linux.intel.com

arch/x86/kernel/cpu/intel_rdt.c

@@ -426,6 +426,7 @@ static int domain_setup_mon_state(struct rdt_resource *r, struct rdt_domain *d)
 					   GFP_KERNEL);
 		if (!d->rmid_busy_llc)
 			return -ENOMEM;
+		INIT_DELAYED_WORK(&d->cqm_limbo, cqm_handle_limbo);
 	}
 	if (is_mbm_total_enabled()) {
 		tsize = sizeof(*d->mbm_total);
@@ -536,11 +537,33 @@ static void domain_remove_cpu(int cpu, struct rdt_resource *r)
 		list_del(&d->list);
 		if (is_mbm_enabled())
 			cancel_delayed_work(&d->mbm_over);
+		if (is_llc_occupancy_enabled() &&  has_busy_rmid(r, d)) {
+			/*
+			 * When a package is going down, forcefully
+			 * decrement rmid->ebusy. There is no way to know
+			 * that the L3 was flushed and hence may lead to
+			 * incorrect counts in rare scenarios, but leaving
+			 * the RMID as busy creates RMID leaks if the
+			 * package never comes back.
+			 */
+			__check_limbo(d, true);
+			cancel_delayed_work(&d->cqm_limbo);
+		}
+
 		kfree(d);
-	} else if (r == &rdt_resources_all[RDT_RESOURCE_L3] &&
-		   cpu == d->mbm_work_cpu && is_mbm_enabled()) {
-		cancel_delayed_work(&d->mbm_over);
-		mbm_setup_overflow_handler(d, 0);
+		return;
+	}
+
+	if (r == &rdt_resources_all[RDT_RESOURCE_L3]) {
+		if (is_mbm_enabled() && cpu == d->mbm_work_cpu) {
+			cancel_delayed_work(&d->mbm_over);
+			mbm_setup_overflow_handler(d, 0);
+		}
+		if (is_llc_occupancy_enabled() && cpu == d->cqm_work_cpu &&
+		    has_busy_rmid(r, d)) {
+			cancel_delayed_work(&d->cqm_limbo);
+			cqm_setup_limbo_handler(d, 0);
+		}
 	}
 }
 

arch/x86/kernel/cpu/intel_rdt.h

@@ -20,6 +20,8 @@
 #define QOS_L3_MBM_TOTAL_EVENT_ID	0x02
 #define QOS_L3_MBM_LOCAL_EVENT_ID	0x03
 
+#define CQM_LIMBOCHECK_INTERVAL	1000
+
 #define MBM_CNTR_WIDTH			24
 #define MBM_OVERFLOW_INTERVAL		1000
 
@@ -187,8 +189,11 @@ struct mbm_state {
  * @mbm_total:	saved state for MBM total bandwidth
  * @mbm_local:	saved state for MBM local bandwidth
  * @mbm_over:	worker to periodically read MBM h/w counters
+ * @cqm_limbo:	worker to periodically read CQM h/w counters
  * @mbm_work_cpu:
  *		worker cpu for MBM h/w counters
+ * @cqm_work_cpu:
+ *		worker cpu for CQM h/w counters
  * @ctrl_val:	array of cache or mem ctrl values (indexed by CLOSID)
  * @new_ctrl:	new ctrl value to be loaded
  * @have_new_ctrl: did user provide new_ctrl for this domain
@@ -201,7 +206,9 @@ struct rdt_domain {
 	struct mbm_state	*mbm_total;
 	struct mbm_state	*mbm_local;
 	struct delayed_work	mbm_over;
+	struct delayed_work	cqm_limbo;
 	int			mbm_work_cpu;
+	int			cqm_work_cpu;
 	u32			*ctrl_val;
 	u32			new_ctrl;
 	bool			have_new_ctrl;
@@ -422,7 +429,12 @@ void mkdir_mondata_subdir_allrdtgrp(struct rdt_resource *r,
 				    struct rdt_domain *d);
 void mon_event_read(struct rmid_read *rr, struct rdt_domain *d,
 		    struct rdtgroup *rdtgrp, int evtid, int first);
-void mbm_setup_overflow_handler(struct rdt_domain *dom, unsigned long delay_ms);
+void mbm_setup_overflow_handler(struct rdt_domain *dom,
+				unsigned long delay_ms);
 void mbm_handle_overflow(struct work_struct *work);
+void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms);
+void cqm_handle_limbo(struct work_struct *work);
+bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d);
+void __check_limbo(struct rdt_domain *d, bool force_free);
 
 #endif /* _ASM_X86_INTEL_RDT_H */

arch/x86/kernel/cpu/intel_rdt_monitor.c

@@ -33,7 +33,7 @@
 
 struct rmid_entry {
 	u32				rmid;
-	atomic_t			busy;
+	int				busy;
 	struct list_head		list;
 };
 
@@ -45,13 +45,13 @@ struct rmid_entry {
 static LIST_HEAD(rmid_free_lru);
 
 /**
- * @rmid_limbo_lru       list of currently unused but (potentially)
+ * @rmid_limbo_count     count of currently unused but (potentially)
  *     dirty RMIDs.
- *     This list contains RMIDs that no one is currently using but that
+ *     This counts RMIDs that no one is currently using but that
  *     may have a occupancy value > intel_cqm_threshold. User can change
  *     the threshold occupancy value.
  */
-static LIST_HEAD(rmid_limbo_lru);
+unsigned int rmid_limbo_count;
 
 /**
  * @rmid_entry - The entry in the limbo and free lists.
@@ -103,124 +103,53 @@ static u64 __rmid_read(u32 rmid, u32 eventid)
 	return val;
 }
 
-/*
- * Walk the limbo list looking at any RMIDs that are flagged in the
- * domain rmid_busy_llc bitmap as busy. If the reported LLC occupancy
- * is below the threshold clear the busy bit and decrement the count.
- * If the busy count gets to zero on an RMID we stop looking.
- * This can be called from an IPI.
- * We need an atomic for the busy count because multiple CPUs may check
- * the same RMID at the same time.
- */
-static bool __check_limbo(struct rdt_domain *d)
-{
-	struct rmid_entry *entry;
-	u64 val;
-
-	list_for_each_entry(entry, &rmid_limbo_lru, list) {
-		if (!test_bit(entry->rmid, d->rmid_busy_llc))
-			continue;
-		val = __rmid_read(entry->rmid, QOS_L3_OCCUP_EVENT_ID);
-		if (val <= intel_cqm_threshold) {
-			clear_bit(entry->rmid, d->rmid_busy_llc);
-			if (atomic_dec_and_test(&entry->busy))
-				return true;
-		}
-	}
-	return false;
-}
-
-static void check_limbo(void *arg)
+static bool rmid_dirty(struct rmid_entry *entry)
 {
-	struct rdt_domain *d;
-
-	d = get_domain_from_cpu(smp_processor_id(),
-				&rdt_resources_all[RDT_RESOURCE_L3]);
-
-	if (d)
-		__check_limbo(d);
-}
+	u64 val = __rmid_read(entry->rmid, QOS_L3_OCCUP_EVENT_ID);
 
-static bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d)
-{
-	return find_first_bit(d->rmid_busy_llc, r->num_rmid) != r->num_rmid;
+	return val >= intel_cqm_threshold;
 }
 
 /*
- * Scan the limbo list and move all entries that are below the
- * intel_cqm_threshold to the free list.
- * Return "true" if the limbo list is empty, "false" if there are
- * still some RMIDs there.
+ * Check the RMIDs that are marked as busy for this domain. If the
+ * reported LLC occupancy is below the threshold clear the busy bit and
+ * decrement the count. If the busy count gets to zero on an RMID, we
+ * free the RMID
  */
-static bool try_freeing_limbo_rmid(void)
+void __check_limbo(struct rdt_domain *d, bool force_free)
 {
-	struct rmid_entry *entry, *tmp;
+	struct rmid_entry *entry;
 	struct rdt_resource *r;
-	cpumask_var_t cpu_mask;
-	struct rdt_domain *d;
-	bool ret = true;
-	int cpu;
-
-	if (list_empty(&rmid_limbo_lru))
-		return ret;
+	u32 crmid = 1, nrmid;
 
 	r = &rdt_resources_all[RDT_RESOURCE_L3];
 
-	cpu = get_cpu();
-
 	/*
-	 * First see if we can free up an RMID by checking busy values
-	 * on the local package.
+	 * Skip RMID 0 and start from RMID 1 and check all the RMIDs that
+	 * are marked as busy for occupancy < threshold. If the occupancy
+	 * is less than the threshold decrement the busy counter of the
+	 * RMID and move it to the free list when the counter reaches 0.
 	 */
-	d = get_domain_from_cpu(cpu, r);
-	if (d && has_busy_rmid(r, d) && __check_limbo(d)) {
-		list_for_each_entry_safe(entry, tmp, &rmid_limbo_lru, list) {
-			if (atomic_read(&entry->busy) == 0) {
-				list_del(&entry->list);
+	for (;;) {
+		nrmid = find_next_bit(d->rmid_busy_llc, r->num_rmid, crmid);
+		if (nrmid >= r->num_rmid)
+			break;
+
+		entry = __rmid_entry(nrmid);
+		if (force_free || !rmid_dirty(entry)) {
+			clear_bit(entry->rmid, d->rmid_busy_llc);
+			if (!--entry->busy) {
+				rmid_limbo_count--;
 				list_add_tail(&entry->list, &rmid_free_lru);
-				goto done;
 			}
 		}
+		crmid = nrmid + 1;
 	}
+}
 
-	if (!zalloc_cpumask_var(&cpu_mask, GFP_KERNEL)) {
-		ret = false;
-		goto done;
-	}
-
-	/*
-	 * Build a mask of other domains that have busy RMIDs
-	 */
-	list_for_each_entry(d, &r->domains, list) {
-		if (!cpumask_test_cpu(cpu, &d->cpu_mask) &&
-		    has_busy_rmid(r, d))
-			cpumask_set_cpu(cpumask_any(&d->cpu_mask), cpu_mask);
-	}
-	if (cpumask_empty(cpu_mask)) {
-		ret = false;
-		goto free_mask;
-	}
-
-	/*
-	 * Scan domains with busy RMIDs to check if they still are busy
-	 */
-	on_each_cpu_mask(cpu_mask, check_limbo, NULL, true);
-
-	/* Walk limbo list moving all free RMIDs to the &rmid_free_lru list */
-	list_for_each_entry_safe(entry, tmp, &rmid_limbo_lru, list) {
-		if (atomic_read(&entry->busy) != 0) {
-			ret = false;
-			continue;
-		}
-		list_del(&entry->list);
-		list_add_tail(&entry->list, &rmid_free_lru);
-	}
-
-free_mask:
-	free_cpumask_var(cpu_mask);
-done:
-	put_cpu();
-	return ret;
+bool has_busy_rmid(struct rdt_resource *r, struct rdt_domain *d)
+{
+	return find_first_bit(d->rmid_busy_llc, r->num_rmid) != r->num_rmid;
 }
 
 /*
@@ -231,15 +160,11 @@ done:
 int alloc_rmid(void)
 {
 	struct rmid_entry *entry;
-	bool ret;
 
 	lockdep_assert_held(&rdtgroup_mutex);
 
-	if (list_empty(&rmid_free_lru)) {
-		ret = try_freeing_limbo_rmid();
-		if (list_empty(&rmid_free_lru))
-			return ret ? -ENOSPC : -EBUSY;
-	}
+	if (list_empty(&rmid_free_lru))
+		return rmid_limbo_count ? -EBUSY : -ENOSPC;
 
 	entry = list_first_entry(&rmid_free_lru,
 				 struct rmid_entry, list);
@@ -252,11 +177,12 @@ static void add_rmid_to_limbo(struct rmid_entry *entry)
 {
 	struct rdt_resource *r;
 	struct rdt_domain *d;
-	int cpu, nbusy = 0;
+	int cpu;
 	u64 val;
 
 	r = &rdt_resources_all[RDT_RESOURCE_L3];
 
+	entry->busy = 0;
 	cpu = get_cpu();
 	list_for_each_entry(d, &r->domains, list) {
 		if (cpumask_test_cpu(cpu, &d->cpu_mask)) {
@@ -264,17 +190,22 @@ static void add_rmid_to_limbo(struct rmid_entry *entry)
 			if (val <= intel_cqm_threshold)
 				continue;
 		}
+
+		/*
+		 * For the first limbo RMID in the domain,
+		 * setup up the limbo worker.
+		 */
+		if (!has_busy_rmid(r, d))
+			cqm_setup_limbo_handler(d, CQM_LIMBOCHECK_INTERVAL);
 		set_bit(entry->rmid, d->rmid_busy_llc);
-		nbusy++;
+		entry->busy++;
 	}
 	put_cpu();
 
-	if (nbusy) {
-		atomic_set(&entry->busy, nbusy);
-		list_add_tail(&entry->list, &rmid_limbo_lru);
-	} else {
+	if (entry->busy)
+		rmid_limbo_count++;
+	else
 		list_add_tail(&entry->list, &rmid_free_lru);
-	}
 }
 
 void free_rmid(u32 rmid)
@@ -387,6 +318,50 @@ static void mbm_update(struct rdt_domain *d, int rmid)
 	}
 }
 
+/*
+ * Handler to scan the limbo list and move the RMIDs
+ * to free list whose occupancy < threshold_occupancy.
+ */
+void cqm_handle_limbo(struct work_struct *work)
+{
+	unsigned long delay = msecs_to_jiffies(CQM_LIMBOCHECK_INTERVAL);
+	int cpu = smp_processor_id();
+	struct rdt_resource *r;
+	struct rdt_domain *d;
+
+	mutex_lock(&rdtgroup_mutex);
+
+	r = &rdt_resources_all[RDT_RESOURCE_L3];
+	d = get_domain_from_cpu(cpu, r);
+
+	if (!d) {
+		pr_warn_once("Failure to get domain for limbo worker\n");
+		goto out_unlock;
+	}
+
+	__check_limbo(d, false);
+
+	if (has_busy_rmid(r, d))
+		schedule_delayed_work_on(cpu, &d->cqm_limbo, delay);
+
+out_unlock:
+	mutex_unlock(&rdtgroup_mutex);
+}
+
+void cqm_setup_limbo_handler(struct rdt_domain *dom, unsigned long delay_ms)
+{
+	unsigned long delay = msecs_to_jiffies(delay_ms);
+	struct rdt_resource *r;
+	int cpu;
+
+	r = &rdt_resources_all[RDT_RESOURCE_L3];
+
+	cpu = cpumask_any(&dom->cpu_mask);
+	dom->cqm_work_cpu = cpu;
+
+	schedule_delayed_work_on(cpu, &dom->cqm_limbo, delay);
+}
+
 void mbm_handle_overflow(struct work_struct *work)
 {
 	unsigned long delay = msecs_to_jiffies(MBM_OVERFLOW_INTERVAL);
@@ -413,6 +388,7 @@ void mbm_handle_overflow(struct work_struct *work)
 	}
 
 	schedule_delayed_work_on(cpu, &d->mbm_over, delay);
+
 out_unlock:
 	mutex_unlock(&rdtgroup_mutex);
 }