Threaded file system benchmark

[Ryo ONODERA <ryo_on%yk.rim.or.jp@localhost>]

Hi,

I have added sched(3) support for fio, Flexible IO Tester.
(See: https://github.com/ryo-on/fio/tree/netbsd-cpu-affinity
and included patch.)
However my 'top -1' says only 1 CPU is consumed.

My sched(3) support code is wrong? Or is there another reason?
If there is another resason, could you explain me the reason?
If my code is wrong, could you give me some advices?

Thank you.

My machine:
NetBSD/amd64 current on 4 CPUs laptop.
My kernel is GENERIC with dtrace support.
$ uname -a
NetBSD angelcake.elements.tetera.org 7.99.19 NetBSD 7.99.19 (DTRACE7) #7: Thu Jul  9 21:58:33 JST 2015  ryo_on%angelcake.elements.tetera.org@localhost:/usr/world/7.99/amd64/obj/sys/arch/amd64/compile/DTRACE7 amd64

On one terminal:
$ cat random-write-test.fio
; random read of 128mb of data

[random-write]
rw=randwrite
size=1G
directory=/home/ryo_on/tmp2/fio-test

$ fio --numjobs 4 random-write-test.fio

Another terminal:
$ top -1

=== === patch === ===
diff --git a/Makefile b/Makefile
index 7fe7d4d..f795c74 100644
--- a/Makefile
+++ b/Makefile
@@ -284,7 +284,7 @@ override CFLAGS += -DFIO_VERSION='"$(FIO_VERSION)"'
 	@$(CC) -MM $(CFLAGS) $(CPPFLAGS) $(SRCDIR)/$*.c > $*.d
 	@mv -f $*.d $*.d.tmp
 	@sed -e 's|.*:|$*.o:|' < $*.d.tmp > $*.d
-	@sed -e 's/.*://' -e 's/\\$$//' < $*.d.tmp | fmt -1 | \
+	@sed -e 's/.*://' -e 's/\\$$//' < $*.d.tmp | fmt -g 1 | \
 		sed -e 's/^ *//' -e 's/$$/:/' >> $*.d
 	@rm -f $*.d.tmp
 
diff --git a/backend.c b/backend.c
index 3eafff6..384b2d3 100644
--- a/backend.c
+++ b/backend.c
@@ -1386,7 +1386,11 @@ static void *thread_main(void *data)
 	 * thread from this job
 	 */
 	if (o->gtod_cpu)
+#if defined(__NetBSD__)
+		fio_cpu_clear(o->cpumask, o->gtod_cpu);
+#else
 		fio_cpu_clear(&o->cpumask, o->gtod_cpu);
+#endif
 
 	/*
 	 * Set affinity first, in case it has an impact on the memory
@@ -1394,7 +1398,11 @@ static void *thread_main(void *data)
 	 */
 	if (fio_option_is_set(o, cpumask)) {
 		if (o->cpus_allowed_policy == FIO_CPUS_SPLIT) {
+#if defined(__NetBSD__)
+			ret = fio_cpus_split(o->cpumask, td->thread_number - 1);
+#else
 			ret = fio_cpus_split(&o->cpumask, td->thread_number - 1);
+#endif
 			if (!ret) {
 				log_err("fio: no CPUs set\n");
 				log_err("fio: Try increasing number of available CPUs\n");
@@ -1653,7 +1661,11 @@ err:
 	verify_free_state(td);
 
 	if (fio_option_is_set(o, cpumask)) {
+#if defined(__NetBSD__)
+		ret = fio_cpuset_exit(o->cpumask);
+#else
 		ret = fio_cpuset_exit(&o->cpumask);
+#endif
 		if (ret)
 			td_verror(td, ret, "fio_cpuset_exit");
 	}
diff --git a/gettime-thread.c b/gettime-thread.c
index 9bf85f0..6b02107 100644
--- a/gettime-thread.c
+++ b/gettime-thread.c
@@ -9,7 +9,11 @@
 struct timeval *fio_tv = NULL;
 int fio_gtod_offload = 0;
 static pthread_t gtod_thread;
+#if defined(__NetBSD__)
+static os_cpu_mask_t *fio_gtod_cpumask;
+#else
 static os_cpu_mask_t fio_gtod_cpumask;
+#endif
 
 void fio_gtod_init(void)
 {
@@ -96,6 +100,10 @@ err:
 void fio_gtod_set_cpu(unsigned int cpu)
 {
 #ifdef FIO_HAVE_CPU_AFFINITY
+#if defined(__NetBSD__)
+	fio_cpu_set(fio_gtod_cpumask, cpu);
+#else
 	fio_cpu_set(&fio_gtod_cpumask, cpu);
 #endif
+#endif
 }
diff --git a/gettime.c b/gettime.c
index ac54111..395342c 100644
--- a/gettime.c
+++ b/gettime.c
@@ -483,18 +483,30 @@ static void *clock_thread_fn(void *data)
 {
 	struct clock_thread *t = data;
 	struct clock_entry *c;
+#if defined(__NetBSD__)
+	os_cpu_mask_t *cpu_mask = cpuset_create();
+#else
 	os_cpu_mask_t cpu_mask;
+#endif
 	uint32_t last_seq;
 	int i;
 
-	if (fio_cpuset_init(&cpu_mask)) {
+#if defined(__NetBSD__)
+	if (fio_cpuset_init(cpu_mask)) {
+#else
+	if (fio_cpuset_init(cpu_mask)) {
+#endif
 		int __err = errno;
 
 		log_err("clock cpuset init failed: %s\n", strerror(__err));
 		goto err_out;
 	}
 
+#if defined(__NetBSD__)
+	fio_cpu_set(cpu_mask, t->cpu);
+#else
 	fio_cpu_set(&cpu_mask, t->cpu);
+#endif
 
 	if (fio_setaffinity(gettid(), cpu_mask) == -1) {
 		int __err = errno;
@@ -538,10 +550,18 @@ static void *clock_thread_fn(void *data)
 	if (!t->entries[i - 1].tsc && !t->entries[0].tsc)
 		goto err;
 
+#if defined(__NetBSD__)
+	fio_cpuset_exit(cpu_mask);
+#else
 	fio_cpuset_exit(&cpu_mask);
+#endif
 	return NULL;
 err:
+#if defined(__NetBSD__)
+	fio_cpuset_exit(cpu_mask);
+#else
 	fio_cpuset_exit(&cpu_mask);
+#endif
 err_out:
 	return (void *) 1;
 }
diff --git a/hash.h b/hash.h
index 02b0614..abafa3b 100644
--- a/hash.h
+++ b/hash.h
@@ -18,6 +18,10 @@
  * machines where multiplications are slow.
  */
 
+#if !defined(BITS_PER_LONG)
+#define	BITS_PER_LONG __SIZEOF_LONG__*8
+#endif
+
 #if BITS_PER_LONG == 32
 /* 2^31 + 2^29 - 2^25 + 2^22 - 2^19 - 2^16 + 1 */
 #define GOLDEN_RATIO_PRIME 0x9e370001UL
diff --git a/idletime.c b/idletime.c
index db272fe..370dcc2 100644
--- a/idletime.c
+++ b/idletime.c
@@ -46,23 +46,39 @@ static double calibrate_unit(unsigned char *data)
 static void free_cpu_affinity(struct idle_prof_thread *ipt)
 {
 #if defined(FIO_HAVE_CPU_AFFINITY)
+#if defined(__NetBSD__)
+	fio_cpuset_exit(ipt->cpu_mask);
+#else
 	fio_cpuset_exit(&ipt->cpu_mask);
 #endif
+#endif
 }
 
 static int set_cpu_affinity(struct idle_prof_thread *ipt)
 {
 #if defined(FIO_HAVE_CPU_AFFINITY)
+#if defined(__NetBSD__)
+	if (fio_cpuset_init(ipt->cpu_mask)) {
+#else
 	if (fio_cpuset_init(&ipt->cpu_mask)) {
+#endif
 		log_err("fio: cpuset init failed\n");
 		return -1;
 	}
 
+#if defined(__NetBSD__)
+	fio_cpu_set(ipt->cpu_mask, ipt->cpu);
+#else
 	fio_cpu_set(&ipt->cpu_mask, ipt->cpu);
+#endif
 
 	if (fio_setaffinity(gettid(), ipt->cpu_mask)) {
 		log_err("fio: fio_setaffinity failed\n");
+#if defined(__NetBSD__)
+		fio_cpuset_exit(ipt->cpu_mask);
+#else
 		fio_cpuset_exit(&ipt->cpu_mask);
+#endif
 		return -1;
 	}
 
diff --git a/idletime.h b/idletime.h
index bd6dcef..be3a482 100644
--- a/idletime.h
+++ b/idletime.h
@@ -35,7 +35,11 @@ struct idle_prof_thread {
 	pthread_mutex_t init_lock;
 	pthread_mutex_t start_lock;
 
+#if defined(__NetBSD__)
+	os_cpu_mask_t *cpu_mask;
+#else
 	os_cpu_mask_t cpu_mask;
+#endif
 };
 
 struct idle_prof_common {
diff --git a/init.c b/init.c
index 5edd53e..2e13868 100644
--- a/init.c
+++ b/init.c
@@ -1702,7 +1702,11 @@ static int fill_def_thread(void)
 {
 	memset(&def_thread, 0, sizeof(def_thread));
 
+#if defined(__NetBSD__)
+	fio_getaffinity(getpid(), def_thread.o.cpumask);
+#else
 	fio_getaffinity(getpid(), &def_thread.o.cpumask);
+#endif
 	def_thread.o.error_dump = 1;
 
 	/*
diff --git a/options.c b/options.c
index ed5d37e..f1b44ef 100644
--- a/options.c
+++ b/options.c
@@ -435,7 +435,11 @@ static int str_cpumask_cb(void *data, unsigned long long *val)
 	if (parse_dryrun())
 		return 0;
 
+#if defined(__NetBSD__)
+	ret = fio_cpuset_init(td->o.cpumask);
+#else
 	ret = fio_cpuset_init(&td->o.cpumask);
+#endif
 	if (ret < 0) {
 		log_err("fio: cpuset_init failed\n");
 		td_verror(td, ret, "fio_cpuset_init");
@@ -452,7 +456,11 @@ static int str_cpumask_cb(void *data, unsigned long long *val)
 				return 1;
 			}
 			dprint(FD_PARSE, "set cpu allowed %d\n", i);
+#if defined(__NetBSD__)
+			fio_cpu_set(td->o.cpumask, i);
+#else
 			fio_cpu_set(&td->o.cpumask, i);
+#endif
 		}
 	}
 
@@ -532,14 +540,22 @@ static int str_cpus_allowed_cb(void *data, const char *input)
 	if (parse_dryrun())
 		return 0;
 
+#if defined(__NetBSD__)
+	return set_cpus_allowed(td, td->o.cpumask, input);
+#else
 	return set_cpus_allowed(td, &td->o.cpumask, input);
+#endif
 }
 
 static int str_verify_cpus_allowed_cb(void *data, const char *input)
 {
 	struct thread_data *td = data;
 
+#if defined(__NetBSD__)
+	return set_cpus_allowed(td, td->o.verify_cpumask, input);
+#else
 	return set_cpus_allowed(td, &td->o.verify_cpumask, input);
+#endif
 }
 #endif
 
diff --git a/os/os-netbsd.h b/os/os-netbsd.h
index 4b0269e..664edfd 100644
--- a/os/os-netbsd.h
+++ b/os/os-netbsd.h
@@ -12,6 +12,9 @@
 #undef rb_node
 #undef rb_left
 #undef rb_right
+#define _NetBSD_SOURCE
+#include <sched.h>
+#include <unistd.h>
 
 #include "../file.h"
 
@@ -20,8 +23,61 @@
 #define FIO_USE_GENERIC_RAND
 #define FIO_USE_GENERIC_INIT_RANDOM_STATE
 #define FIO_HAVE_GETTID
+#define	FIO_HAVE_CPU_AFFINITY
+#define	FIO_MAX_CPUS 1024
 
-#undef	FIO_HAVE_CPU_AFFINITY	/* XXX notyet */
+typedef	cpuset_t os_cpu_mask_t;
+
+static inline int fio_cpu_isset(os_cpu_mask_t *mask, const cpuid_t cpu)
+{
+	return cpuset_isset(cpu, mask);
+}
+
+static inline int fio_setaffinity(int tid, os_cpu_mask_t *cpumask)
+{
+	return _sched_getaffinity(getpid(), tid, cpuset_size(cpumask), cpumask);
+}
+
+static inline int fio_getaffinity(int tid, os_cpu_mask_t *cpumask)
+{
+	return _sched_getaffinity(getpid(), tid, cpuset_size(cpumask), cpumask);
+}
+
+static inline int fio_cpuset_init(os_cpu_mask_t *mask)
+{
+	cpuset_zero(mask);
+	return 0;
+}
+
+static inline void fio_cpu_set(os_cpu_mask_t *mask, int cpu)
+{
+	cpuset_set(cpu, mask);
+	return;
+}
+
+static inline int fio_cpuset_exit(os_cpu_mask_t *mask)
+{
+	return 0;
+}
+
+static inline int fio_cpu_count(os_cpu_mask_t *mask)
+{
+	int cpu = sysconf(_SC_NPROCESSORS_CONF);
+	int count = 0;
+
+	for (int i = 0; i < cpu; i++) {
+		if(cpuset_isset(i, mask)) {
+			count++;
+		}
+	}
+	return count;
+}
+
+static inline void fio_cpu_clear(os_cpu_mask_t *mask, int cpu)
+{
+	cpuset_clr(cpu, mask);
+	return;
+}
 
 #define OS_MAP_ANON		MAP_ANON
 
diff --git a/parse.c b/parse.c
index 745056b..f7f164e 100644
--- a/parse.c
+++ b/parse.c
@@ -152,7 +152,7 @@ static unsigned long long get_mult_time(const char *str, int len,
 
 	c = strdup(p);
 	for (int i = 0; i < strlen(c); i++)
-		c[i] = tolower(c[i]);
+		c[i] = tolower((int)c[i]);
 
 	if (!strncmp("us", c, 2) || !strncmp("usec", c, 4))
 		mult = 1;
@@ -198,7 +198,7 @@ static unsigned long long __get_mult_bytes(const char *p, void *data,
 	c = strdup(p);
 
 	for (i = 0; i < strlen(c); i++) {
-		c[i] = tolower(c[i]);
+		c[i] = tolower((int)c[i]);
 		if (is_separator(c[i])) {
 			c[i] = '\0';
 			break;
diff --git a/thread_options.h b/thread_options.h
index 6604a37..8a66824 100644
--- a/thread_options.h
+++ b/thread_options.h
@@ -163,8 +163,13 @@ struct thread_options {
 	unsigned int stonewall;
 	unsigned int new_group;
 	unsigned int numjobs;
+#if defined(__NetBSD__)
+	os_cpu_mask_t *cpumask;
+	os_cpu_mask_t *verify_cpumask;
+#else
 	os_cpu_mask_t cpumask;
 	os_cpu_mask_t verify_cpumask;
+#endif
 	unsigned int cpus_allowed_policy;
 	char *numa_cpunodes;
 	unsigned short numa_mem_mode;


--
Ryo ONODERA // ryo_on%yk.rim.or.jp@localhost
PGP fingerprint = 82A2 DC91 76E0 A10A 8ABB  FD1B F404 27FA C7D1 15F3