Use percpu_counter for obj_alloc counter of Linux-backed caches

A previous commit enabled the tracking of object allocations
in Linux-backed caches from the SPL layer for debuggability.
The commit is: 9a170fc6fe54f1e852b6c39630fe5ef2bbd97c16

Unfortunately, it also introduced minor performance regressions
that were highlighted by the ZFS perf test-suite. Within Delphix
we found that the regression would be from -1%, all the way up
to -8% for some workloads.

This commit brings performance back up to par by creating a
separate counter for those caches and making it a percpu in
order to avoid lock-contention.

The initial performance testing was done by myself, and the
final round was conducted by @tonynguien who was also the one
that discovered the regression and highlighted the culprit.

Reviewed-by: Matt Ahrens <matt@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes #10397

config/kernel-percpu.m4

@@ -0,0 +1,34 @@
+dnl #
+dnl # 3.18 API change,
+dnl # The function percpu_counter_init now must be passed a GFP mask.
+dnl #
+AC_DEFUN([ZFS_AC_KERNEL_SRC_PERCPU_COUNTER_INIT], [
+	ZFS_LINUX_TEST_SRC([percpu_counter_init_with_gfp], [
+		#include <linux/gfp.h>
+		#include <linux/percpu_counter.h>
+	],[
+		struct percpu_counter counter;
+		int error;
+
+		error = percpu_counter_init(&counter, 0, GFP_KERNEL);
+	])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_PERCPU_COUNTER_INIT], [
+	AC_MSG_CHECKING([whether percpu_counter_init() wants gfp_t])
+	ZFS_LINUX_TEST_RESULT([percpu_counter_init_with_gfp], [
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_PERCPU_COUNTER_INIT_WITH_GFP, 1,
+		    [percpu_counter_init() wants gfp_t])
+	],[
+		AC_MSG_RESULT(no)
+	])
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_SRC_PERCPU], [
+	ZFS_AC_KERNEL_SRC_PERCPU_COUNTER_INIT
+])
+
+AC_DEFUN([ZFS_AC_KERNEL_PERCPU], [
+	ZFS_AC_KERNEL_PERCPU_COUNTER_INIT
+])

config/kernel.m4

@@ -121,6 +121,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_SRC], [
 	ZFS_AC_KERNEL_SRC_TOTALRAM_PAGES_FUNC
 	ZFS_AC_KERNEL_SRC_TOTALHIGH_PAGES
 	ZFS_AC_KERNEL_SRC_KSTRTOUL
+	ZFS_AC_KERNEL_SRC_PERCPU
 
 	AC_MSG_CHECKING([for available kernel interfaces])
 	ZFS_LINUX_TEST_COMPILE_ALL([kabi])
@@ -216,6 +217,7 @@ AC_DEFUN([ZFS_AC_KERNEL_TEST_RESULT], [
 	ZFS_AC_KERNEL_TOTALRAM_PAGES_FUNC
 	ZFS_AC_KERNEL_TOTALHIGH_PAGES
 	ZFS_AC_KERNEL_KSTRTOUL
+	ZFS_AC_KERNEL_PERCPU
 ])
 
 dnl #

include/os/linux/kernel/linux/Makefile.am

@@ -5,6 +5,7 @@ KERNEL_H = \
 	blkdev_compat.h \
 	utsname_compat.h \
 	kmap_compat.h \
+	percpu_compat.h \
 	simd.h \
 	simd_x86.h \
 	simd_aarch64.h \

include/os/linux/kernel/linux/percpu_compat.h

@@ -0,0 +1,44 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+
+/*
+ * Copyright (c) 2020 by Delphix. All rights reserved.
+ */
+
+#ifndef _ZFS_PERCPU_H
+#define	_ZFS_PERCPU_H
+
+#include <linux/percpu_counter.h>
+
+/*
+ * 3.18 API change,
+ * percpu_counter_init() now must be passed a gfp mask which will be
+ * used for the dynamic allocation of the actual counter.
+ */
+#ifdef HAVE_PERCPU_COUNTER_INIT_WITH_GFP
+#define	percpu_counter_init_common(counter, n, gfp) \
+	percpu_counter_init(counter, n, gfp)
+#else
+#define	percpu_counter_init_common(counter, n, gfp) \
+	percpu_counter_init(counter, n)
+#endif
+
+#endif /* _ZFS_PERCPU_H */

include/os/linux/spl/sys/kmem_cache.h

@@ -202,6 +202,7 @@ typedef struct spl_kmem_cache {
 	uint64_t		skc_slab_max;	/* Slab max historic  */
 	uint64_t		skc_obj_total;	/* Obj total current */
 	uint64_t		skc_obj_alloc;	/* Obj alloc current */
+	struct percpu_counter	skc_linux_alloc;   /* Linux-backed Obj alloc  */
 	uint64_t		skc_obj_max;	/* Obj max historic */
 	uint64_t		skc_obj_deadlock;  /* Obj emergency deadlocks */
 	uint64_t		skc_obj_emergency; /* Obj emergency current */

module/os/linux/spl/spl-kmem-cache.c

@@ -31,6 +31,7 @@
 #include <sys/wait.h>
 #include <linux/slab.h>
 #include <linux/swap.h>
+#include <linux/percpu_compat.h>
 #include <linux/prefetch.h>
 
 /*
@@ -948,6 +949,13 @@ spl_kmem_cache_create(char *name, size_t size, size_t align,
 	skc->skc_obj_emergency = 0;
 	skc->skc_obj_emergency_max = 0;
 
+	rc = percpu_counter_init_common(&skc->skc_linux_alloc, 0,
+	    GFP_KERNEL);
+	if (rc != 0) {
+		kfree(skc);
+		return (NULL);
+	}
+
 	/*
 	 * Verify the requested alignment restriction is sane.
 	 */
@@ -1047,6 +1055,7 @@ spl_kmem_cache_create(char *name, size_t size, size_t align,
 	return (skc);
 out:
 	kfree(skc->skc_name);
+	percpu_counter_destroy(&skc->skc_linux_alloc);
 	kfree(skc);
 	return (NULL);
 }
@@ -1117,6 +1126,9 @@ spl_kmem_cache_destroy(spl_kmem_cache_t *skc)
 	ASSERT3U(skc->skc_obj_emergency, ==, 0);
 	ASSERT(list_empty(&skc->skc_complete_list));
 
+	ASSERT3U(percpu_counter_sum(&skc->skc_linux_alloc), ==, 0);
+	percpu_counter_destroy(&skc->skc_linux_alloc);
+
 	spin_unlock(&skc->skc_lock);
 
 	kfree(skc->skc_name);
@@ -1473,9 +1485,7 @@ spl_kmem_cache_alloc(spl_kmem_cache_t *skc, int flags)
 			 * how many objects we've allocated in it for
 			 * better debuggability.
 			 */
-			spin_lock(&skc->skc_lock);
+			percpu_counter_inc(&skc->skc_linux_alloc);
-			skc->skc_obj_alloc++;
-			spin_unlock(&skc->skc_lock);
 		}
 		goto ret;
 	}
@@ -1550,9 +1560,7 @@ spl_kmem_cache_free(spl_kmem_cache_t *skc, void *obj)
 	 */
 	if (skc->skc_flags & KMC_SLAB) {
 		kmem_cache_free(skc->skc_linux_cache, obj);
-		spin_lock(&skc->skc_lock);
+		percpu_counter_dec(&skc->skc_linux_alloc);
-		skc->skc_obj_alloc--;
-		spin_unlock(&skc->skc_lock);
 		return;
 	}
 

module/os/linux/spl/spl-proc.c

@@ -446,16 +446,18 @@ slab_seq_show(struct seq_file *f, void *p)
 		 * the underlying Linux cache please refer to /proc/slabinfo.
 		 */
 		spin_lock(&skc->skc_lock);
+		uint64_t objs_allocated =
+		    percpu_counter_sum(&skc->skc_linux_alloc);
 		seq_printf(f, "%-36s  ", skc->skc_name);
 		seq_printf(f, "0x%05lx %9s %9lu %8s %8u  "
 		    "%5s %5s %5s  %5s %5lu %5s  %5s %5s %5s\n",
 		    (long unsigned)skc->skc_flags,
 		    "-",
-		    (long unsigned)(skc->skc_obj_size * skc->skc_obj_alloc),
+		    (long unsigned)(skc->skc_obj_size * objs_allocated),
 		    "-",
 		    (unsigned)skc->skc_obj_size,
 		    "-", "-", "-", "-",
-		    (long unsigned)skc->skc_obj_alloc,
+		    (long unsigned)objs_allocated,
 		    "-", "-", "-", "-");
 		spin_unlock(&skc->skc_lock);
 		return (0);