Skip to contentSkip to conversation

rcuperf: Add ability to increase object allocation size

PATCH Superseded Joel Fernandes (Google) 3 participants 4 messages Mar 6, 2020 – Mar 9, 2020 RCU

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/2 rcuperf: Add ability to increase object allocation size

This allows us to increase memory pressure dynamically using a new rcuperf boot command line parameter called 'rcumult'.

---
 kernel/rcu/rcuperf.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
kernel/rcu/rcuperf.c +4 -1
index a4a8d097d84d9..16dd1e6b7c09f 100644
@@ -88,6 +88,7 @@ torture_param(bool, shutdown, RCUPERF_SHUTDOWN,
 torture_param(int, verbose, 1, "Enable verbose debugging printk()s");
 torture_param(int, writer_holdoff, 0, "Holdoff (us) between GPs, zero to disable");
 torture_param(int, kfree_rcu_test, 0, "Do we run a kfree_rcu() perf test?");
+torture_param(int, kfree_mult, 1, "Multiple of kfree_obj size to allocate.");
 
 static char *perf_type = "rcu";
 module_param(perf_type, charp, 0444);
@@ -635,7 +636,7 @@ kfree_perf_thread(void *arg)
 		}
 
 		for (i = 0; i < kfree_alloc_num; i++) {
-			alloc_ptr = kmalloc(sizeof(struct kfree_obj), GFP_KERNEL);
+			alloc_ptr = kmalloc(kfree_mult * sizeof(struct kfree_obj), GFP_KERNEL);
 			if (!alloc_ptr)
 				return -ENOMEM;
 
@@ -722,6 +723,8 @@ kfree_perf_init(void)
 		schedule_timeout_uninterruptible(1);
 	}
 
+	pr_alert("kfree object size=%lu\n", kfree_mult * sizeof(struct kfree_obj));
+
 	kfree_reader_tasks = kcalloc(kfree_nrealthreads, sizeof(kfree_reader_tasks[0]),
 			       GFP_KERNEL);
 	if (kfree_reader_tasks == NULL) {
Signature · 1 line
2.25.0.265.gbab2e86ba0-goog
#1 lore raw

PATCH 2/2 rcu/tree: Add a shrinker to prevent OOM due to kfree_rcu() batching

To reduce grace periods and improve kfree() performance, we have done batching recently dramatically bringing down the number of grace periods while giving us the ability to use kfree_bulk() for efficient kfree'ing.

However, this has increased the likelihood of OOM condition under heavy kfree_rcu() flood on small memory systems. This patch introduces a shrinker which starts grace periods right away if the system is under memory pressure due to existence of objects that have still not started a grace period.

With this patch, I do not observe an OOM anymore on a system with 512MB RAM and 8 CPUs, with the following rcuperf options:

rcuperf.kfree_loops=20000 rcuperf.kfree_alloc_num=8000 rcuperf.kfree_rcu_test=1 rcuperf.kfree_mult=2

Otherwise it easily OOMs with the above parameters.

NOTE:
1. On systems with no memory pressure, the patch has no effect as intended.
2. In the future, we can use this same mechanism to prevent grace periods from happening even more, by relying on shrinkers carefully.

---
 kernel/rcu/tree.c | 60 +++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 60 insertions(+)
kernel/rcu/tree.c +60 -0
index d3f52c30efb0c..2e0f66f04360e 100644
@@ -2823,6 +2823,8 @@ struct kfree_rcu_cpu {
 	struct delayed_work monitor_work;
 	bool monitor_todo;
 	bool initialized;
+	// Number of objects for which GP not started
+	int count;
 };
 
 static DEFINE_PER_CPU(struct kfree_rcu_cpu, krc);
@@ -2936,6 +2938,8 @@ static inline bool queue_kfree_rcu_work(struct kfree_rcu_cpu *krcp)
 				krcp->head = NULL;
 			}
 
+			krcp->count = 0;
+
 			/*
 			 * One work is per one batch, so there are two "free channels",
 			 * "bhead_free" and "head_free" the batch can handle. It can be
@@ -3072,6 +3076,8 @@ void kfree_call_rcu(struct rcu_head *head, rcu_callback_t func)
 		krcp->head = head;
 	}
 
+	krcp->count++;
+
 	// Set timer to drain after KFREE_DRAIN_JIFFIES.
 	if (rcu_scheduler_active == RCU_SCHEDULER_RUNNING &&
 	    !krcp->monitor_todo) {
@@ -3086,6 +3092,58 @@ void kfree_call_rcu(struct rcu_head *head, rcu_callback_t func)
 }
 EXPORT_SYMBOL_GPL(kfree_call_rcu);
 
+static unsigned long
+kfree_rcu_shrink_count(struct shrinker *shrink, struct shrink_control *sc)
+{
+	int cpu;
+	unsigned long flags, count = 0;
+
+	/* Snapshot count of all CPUs */
+	for_each_online_cpu(cpu) {
+		struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
+
+		spin_lock_irqsave(&krcp->lock, flags);
+		count += krcp->count;
+		spin_unlock_irqrestore(&krcp->lock, flags);
+	}
+
+	return count;
+}
+
+static unsigned long
+kfree_rcu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
+{
+	int cpu, freed = 0;
+	unsigned long flags;
+
+	for_each_online_cpu(cpu) {
+		int count;
+		struct kfree_rcu_cpu *krcp = per_cpu_ptr(&krc, cpu);
+
+		count = krcp->count;
+		spin_lock_irqsave(&krcp->lock, flags);
+		if (krcp->monitor_todo)
+			kfree_rcu_drain_unlock(krcp, flags);
+		else
+			spin_unlock_irqrestore(&krcp->lock, flags);
+
+		sc->nr_to_scan -= count;
+		freed += count;
+
+		if (sc->nr_to_scan <= 0)
+			break;
+	}
+
+	return freed;
+}
+
+static struct shrinker kfree_rcu_shrinker = {
+	.count_objects = kfree_rcu_shrink_count,
+	.scan_objects = kfree_rcu_shrink_scan,
+	.batch = 0,
+	.seeks = DEFAULT_SEEKS,
+};
+
 void __init kfree_rcu_scheduler_running(void)
 {
 	int cpu;
@@ -4007,6 +4065,8 @@ static void __init kfree_rcu_batch_init(void)
 		INIT_DELAYED_WORK(&krcp->monitor_work, kfree_rcu_monitor);
 		krcp->initialized = true;
 	}
+	if (register_shrinker(&kfree_rcu_shrinker))
+		pr_err("Failed to register kfree_rcu() shrinker!\n");
 }
 
 void __init rcu_init(void)
Signature · 1 line
2.25.0.265.gbab2e86ba0-goog
maintainer
#2 lore raw
In reply to Joel Fernandes (Google)
Quoted 4 lines · Joel Fernandes (Google)

This allows us to increase memory pressure dynamically using a new rcuperf boot command line parameter called 'rcumult'.

Looks plausible to me at a very quick first glance. Uladzislau, thoughts?

Thanx, Paul

Quoted 37 lines
---
 kernel/rcu/rcuperf.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
kernel/rcu/rcuperf.c +4 -1
index a4a8d097d84d9..16dd1e6b7c09f 100644
@@ -88,6 +88,7 @@ torture_param(bool, shutdown, RCUPERF_SHUTDOWN,
 torture_param(int, verbose, 1, "Enable verbose debugging printk()s");
 torture_param(int, writer_holdoff, 0, "Holdoff (us) between GPs, zero to disable");
 torture_param(int, kfree_rcu_test, 0, "Do we run a kfree_rcu() perf test?");
+torture_param(int, kfree_mult, 1, "Multiple of kfree_obj size to allocate.");
 
 static char *perf_type = "rcu";
 module_param(perf_type, charp, 0444);
@@ -635,7 +636,7 @@ kfree_perf_thread(void *arg)
 		}
 
 		for (i = 0; i < kfree_alloc_num; i++) {
-			alloc_ptr = kmalloc(sizeof(struct kfree_obj), GFP_KERNEL);
+			alloc_ptr = kmalloc(kfree_mult * sizeof(struct kfree_obj), GFP_KERNEL);
 			if (!alloc_ptr)
 				return -ENOMEM;
 
@@ -722,6 +723,8 @@ kfree_perf_init(void)
 		schedule_timeout_uninterruptible(1);
 	}
 
+	pr_alert("kfree object size=%lu\n", kfree_mult * sizeof(struct kfree_obj));
+
 	kfree_reader_tasks = kcalloc(kfree_nrealthreads, sizeof(kfree_reader_tasks[0]),
 			       GFP_KERNEL);
 	if (kfree_reader_tasks == NULL) {
Signature · 1 line
2.25.0.265.gbab2e86ba0-goog
#3 lore raw
In reply to Joel Fernandes (Google)
Thank you for the patch! Perhaps something to improve:

Hi "Joel,

Thank you for the patch! Perhaps something to improve:

[auto build test WARNING on rcu/dev]
[also build test WARNING on rcu/rcu/next linus/master v5.6-rc5 next-20200306] [cannot apply to linux/master]
[if your patch is applied to the wrong git tree, please drop us a note to help improve the system. BTW, we also suggest to use '--base' option to specify the base tree in git format-patch, please see https://stackoverflow.com/a/37406982]

url:    https://github.com/0day-ci/linux/commits/Joel-Fernandes-Google/rcuperf-Add-ability-to-increase-object-allocation-size/20200306-110336
base:   https://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git dev
config: i386-randconfig-a001-20200308 (attached as .config)
compiler: gcc-4.9 (Debian 4.9.2-10+deb8u1) 4.9.2
reproduce:
        # save the attached .config to linux build tree
        make ARCH=i386 

If you fix the issue, kindly add following tag

All warnings (new ones prefixed by >>):

In file included from include/linux/kernel.h:15:0,
from kernel/rcu/rcuperf.c:13:
kernel/rcu/rcuperf.c: In function 'kfree_perf_init':

>> kernel/rcu/rcuperf.c:726:65: warning: format '%lu' expects argument of type 'long unsigned int', but argument 2 has type 'unsigned int' [-Wformat=]
     pr_alert("kfree object size=%lu\n", kfree_mult * sizeof(struct kfree_obj));
                                                                    ^
   include/linux/printk.h:300:35: note: in definition of macro 'pr_alert'
     printk(KERN_ALERT pr_fmt(fmt), ##__VA_ARGS__)
                                      ^

vim +726 kernel/rcu/rcuperf.c

708
709 static int __init
710 kfree_perf_init(void)
711 {
712 long i;
713 int firsterr = 0;
714
715 kfree_nrealthreads = compute_real(kfree_nthreads);
716 /* Start up the kthreads. */
717 if (shutdown) {
718 init_waitqueue_head(&shutdown_wq);
719 firsterr = torture_create_kthread(kfree_perf_shutdown, NULL,
720 shutdown_task);
721 if (firsterr)
722 goto unwind;
723 schedule_timeout_uninterruptible(1);
724 }
725

 > 726		pr_alert("kfree object size=%lu\n", kfree_mult * sizeof(struct kfree_obj));

727
728 kfree_reader_tasks = kcalloc(kfree_nrealthreads, sizeof(kfree_reader_tasks[0]),
729 GFP_KERNEL);
730 if (kfree_reader_tasks == NULL) {
731 firsterr = -ENOMEM;
732 goto unwind;
733 }
734
735 for (i = 0; i < kfree_nrealthreads; i++) {
736 firsterr = torture_create_kthread(kfree_perf_thread, (void *)i,
737 kfree_reader_tasks[i]);
738 if (firsterr)
739 goto unwind;
740 }
741
742 while (atomic_read(&n_kfree_perf_thread_started) < kfree_nrealthreads)
743 schedule_timeout_uninterruptible(1);
744
745 torture_init_end();
746 return 0;
747
748 unwind:
749 torture_init_end();
750 kfree_perf_cleanup();
751 return firsterr;
752 }
753

---
0-DAY CI Kernel Test Service, Intel Corporation
https://lists.01.org/hyperkitty/list/[email protected]