* [PATCH] MIPS: crazy spinlock speed test.
@ 2010-02-16 23:26 David Daney
2010-02-24 15:55 ` Ralf Baechle
0 siblings, 1 reply; 2+ messages in thread
From: David Daney @ 2010-02-16 23:26 UTC (permalink / raw)
To: linux-mips, ralf; +Cc: David Daney
This is just a test program for raw_spinlocks. The main reason I
wrote it is to validate my spinlock changes that I sent in a previous
patch.
To use it enable CONFIG_DEBUG_FS and CONFIG_SPINLOCK_TEST then at run
time do:
# mount -t debugfs none /sys/kernel/debug/
# cat /sys/kernel/debug/mips/spin_single
# cat /sys/kernel/debug/mips/spin_multi
On my 600MHz octeon cn5860 (16 CPUs) I get
spin_single spin_multi
base 106885 247941
spinlock_patch 75194 219465
This shows that for uncontended locks the spinlock patch gives 41%
improvement and for contended locks 12% improvement (1/time).
Signed-off-by: David Daney <ddaney@caviumnetworks.com>
---
arch/mips/Kconfig.debug | 7 ++
arch/mips/kernel/Makefile | 1 +
arch/mips/kernel/spinlock_test.c | 141 ++++++++++++++++++++++++++++++++++++++
3 files changed, 149 insertions(+), 0 deletions(-)
create mode 100644 arch/mips/kernel/spinlock_test.c
diff --git a/arch/mips/Kconfig.debug b/arch/mips/Kconfig.debug
index d8c0fd7..8354a18 100644
--- a/arch/mips/Kconfig.debug
+++ b/arch/mips/Kconfig.debug
@@ -131,4 +131,11 @@ config DEBUG_ZBOOT
to reduce the kernel image size and speed up the booting procedure a
little.
+config SPINLOCK_TEST
+ bool "Enable spinlock timing tests in debugfs"
+ depends on DEBUG_FS
+ default n
+ help
+ Add several files to the debugfs to test spinlock speed.
+
endmenu
diff --git a/arch/mips/kernel/Makefile b/arch/mips/kernel/Makefile
index 642ae95..aac78e3 100644
--- a/arch/mips/kernel/Makefile
+++ b/arch/mips/kernel/Makefile
@@ -93,6 +93,7 @@ obj-$(CONFIG_GPIO_TXX9) += gpio_txx9.o
obj-$(CONFIG_KEXEC) += machine_kexec.o relocate_kernel.o
obj-$(CONFIG_EARLY_PRINTK) += early_printk.o
+obj-$(CONFIG_SPINLOCK_TEST) += spinlock_test.o
CFLAGS_cpu-bugs64.o = $(shell if $(CC) $(KBUILD_CFLAGS) -Wa,-mdaddi -c -o /dev/null -xc /dev/null >/dev/null 2>&1; then echo "-DHAVE_AS_SET_DADDI"; fi)
diff --git a/arch/mips/kernel/spinlock_test.c b/arch/mips/kernel/spinlock_test.c
new file mode 100644
index 0000000..da61134
--- /dev/null
+++ b/arch/mips/kernel/spinlock_test.c
@@ -0,0 +1,141 @@
+#include <linux/init.h>
+#include <linux/kthread.h>
+#include <linux/hrtimer.h>
+#include <linux/fs.h>
+#include <linux/debugfs.h>
+#include <linux/module.h>
+#include <linux/spinlock.h>
+
+
+static int ss_get(void *data, u64 *val)
+{
+ ktime_t start, finish;
+ int loops;
+ int cont;
+ DEFINE_RAW_SPINLOCK(ss_spin);
+
+ loops = 1000000;
+ cont = 1;
+
+ start = ktime_get();
+
+ while (cont) {
+ raw_spin_lock(&ss_spin);
+ loops--;
+ if (loops == 0)
+ cont = 0;
+ raw_spin_unlock(&ss_spin);
+ }
+
+ finish = ktime_get();
+
+ *val = ktime_us_delta(finish, start);
+
+ return 0;
+}
+
+DEFINE_SIMPLE_ATTRIBUTE(fops_ss, ss_get, NULL, "%llu\n");
+
+
+
+struct spin_multi_state {
+ raw_spinlock_t lock;
+ atomic_t start_wait;
+ atomic_t enter_wait;
+ atomic_t exit_wait;
+ int loops;
+};
+
+struct spin_multi_per_thread {
+ struct spin_multi_state *state;
+ ktime_t start;
+};
+
+static int multi_other(void *data)
+{
+ int loops;
+ int cont;
+ struct spin_multi_per_thread *pt = data;
+ struct spin_multi_state *s = pt->state;
+
+ loops = s->loops;
+ cont = 1;
+
+ atomic_dec(&s->enter_wait);
+
+ while (atomic_read(&s->enter_wait))
+ ; /* spin */
+
+ pt->start = ktime_get();
+
+ atomic_dec(&s->start_wait);
+
+ while (atomic_read(&s->start_wait))
+ ; /* spin */
+
+ while (cont) {
+ raw_spin_lock(&s->lock);
+ loops--;
+ if (loops == 0)
+ cont = 0;
+ raw_spin_unlock(&s->lock);
+ }
+
+ atomic_dec(&s->exit_wait);
+ while (atomic_read(&s->exit_wait))
+ ; /* spin */
+ return 0;
+}
+
+static int multi_get(void *data, u64 *val)
+{
+ ktime_t finish;
+ struct spin_multi_state ms;
+ struct spin_multi_per_thread t1, t2;
+
+ ms.lock = __RAW_SPIN_LOCK_UNLOCKED("multi_get");
+ ms.loops = 1000000;
+
+ atomic_set(&ms.start_wait, 2);
+ atomic_set(&ms.enter_wait, 2);
+ atomic_set(&ms.exit_wait, 2);
+ t1.state = &ms;
+ t2.state = &ms;
+
+ kthread_run(multi_other, &t2, "multi_get");
+
+ multi_other(&t1);
+
+ finish = ktime_get();
+
+ *val = ktime_us_delta(finish, t1.start);
+
+ return 0;
+}
+
+DEFINE_SIMPLE_ATTRIBUTE(fops_multi, multi_get, NULL, "%llu\n");
+
+
+extern struct dentry *mips_debugfs_dir;
+static int __init spinlock_test(void)
+{
+ struct dentry *d;
+
+ if (!mips_debugfs_dir)
+ return -ENODEV;
+
+ d = debugfs_create_file("spin_single", S_IRUGO,
+ mips_debugfs_dir, NULL,
+ &fops_ss);
+ if (!d)
+ return -ENOMEM;
+
+ d = debugfs_create_file("spin_multi", S_IRUGO,
+ mips_debugfs_dir, NULL,
+ &fops_multi);
+ if (!d)
+ return -ENOMEM;
+
+ return 0;
+}
+device_initcall(spinlock_test);
--
1.6.6
^ permalink raw reply related [flat|nested] 2+ messages in thread* Re: [PATCH] MIPS: crazy spinlock speed test.
2010-02-16 23:26 [PATCH] MIPS: crazy spinlock speed test David Daney
@ 2010-02-24 15:55 ` Ralf Baechle
0 siblings, 0 replies; 2+ messages in thread
From: Ralf Baechle @ 2010-02-24 15:55 UTC (permalink / raw)
To: David Daney; +Cc: linux-mips
On Tue, Feb 16, 2010 at 03:26:35PM -0800, David Daney wrote:
> This is just a test program for raw_spinlocks. The main reason I
> wrote it is to validate my spinlock changes that I sent in a previous
> patch.
>
> To use it enable CONFIG_DEBUG_FS and CONFIG_SPINLOCK_TEST then at run
> time do:
>
> # mount -t debugfs none /sys/kernel/debug/
> # cat /sys/kernel/debug/mips/spin_single
> # cat /sys/kernel/debug/mips/spin_multi
>
> On my 600MHz octeon cn5860 (16 CPUs) I get
>
> spin_single spin_multi
> base 106885 247941
> spinlock_patch 75194 219465
>
> This shows that for uncontended locks the spinlock patch gives 41%
> improvement and for contended locks 12% improvement (1/time).
>
> Signed-off-by: David Daney <ddaney@caviumnetworks.com>
Also queued for 2.6.34.
Thanks!
Ralf
^ permalink raw reply [flat|nested] 2+ messages in thread
end of thread, other threads:[~2010-02-24 15:55 UTC | newest]
Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2010-02-16 23:26 [PATCH] MIPS: crazy spinlock speed test David Daney
2010-02-24 15:55 ` Ralf Baechle
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox