Skip to contentSkip to conversation

powerpc/opal: Wake up kopald polling thread before waiting for events

PATCH Applied Benjamin Herrenschmidt 3 participants 5 messages Jul 4, 2016 – Jul 11, 2016 PowerPC

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 powerpc/opal: Wake up kopald polling thread before waiting for events

On some environments (prototype machines, some simulators, etc...) there is no functional interrupt source to signal completion, so we rely on the fairly slow OPAL heartbeat.

In a number of cases, the calls complete very quickly or even immediately. We've observed that it helps a lot to wakeup the OPAL heartbeat thread before waiting for event in those cases, it will call OPAL immediately to collect completions for anything that finished fast enough.

---
 arch/powerpc/include/asm/opal.h             |  2 ++
 arch/powerpc/platforms/powernv/opal-async.c |  5 +++++
 arch/powerpc/platforms/powernv/opal.c       | 12 ++++++++++--
 3 files changed, 17 insertions(+), 2 deletions(-)
arch/powerpc/include/asm/opal.h +2 -0
index 6135816..0c76bc0 100644
@@ -277,6 +277,8 @@ extern int opal_error_code(int rc);
 
 ssize_t opal_msglog_copy(char *to, loff_t pos, size_t count);
 
+void opal_wake_poller(void);
+
 #endif /* __ASSEMBLY__ */
 
 #endif /* _ASM_POWERPC_OPAL_H */
arch/powerpc/platforms/powernv/opal-async.c +5 -0
index bdc8c0c..83bebee 100644
@@ -117,6 +117,11 @@ int opal_async_wait_response(uint64_t token, struct opal_msg *msg)
 		return -EINVAL;
 	}
 
+	/* Wakeup the poller before we wait for events to speed things
+	 * up on platforms or simulators where the interrupts aren't
+	 * functional.
+	 */
+	opal_wake_poller();
 	wait_event(opal_async_wait, test_bit(token, opal_async_complete_map));
 	memcpy(msg, &opal_async_responses[token], sizeof(*msg));
 
arch/powerpc/platforms/powernv/opal.c +10 -2
index 802f3b7..7f13302 100644
@@ -55,6 +55,7 @@ struct device_node *opal_node;
 static DEFINE_SPINLOCK(opal_write_lock);
 static struct atomic_notifier_head opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
 static uint32_t opal_heartbeat;
+static struct task_struct *kopald_tsk;
 
 void opal_configure_cores(void)
 {
@@ -650,6 +651,7 @@ static void opal_i2c_create_devs(void)
 
 static int kopald(void *unused)
 {
+	unsigned long timeout = msecs_to_jiffies(opal_heartbeat) + 1;
 	__be64 events;
 
 	set_freezable();
@@ -657,12 +659,18 @@ static int kopald(void *unused)
 		try_to_freeze();
 		opal_poll_events(&events);
 		opal_handle_events(be64_to_cpu(events));
-		msleep_interruptible(opal_heartbeat);
+		schedule_timeout_interruptible(timeout);
 	} while (!kthread_should_stop());
 
 	return 0;
 }
 
+void opal_wake_poller(void)
+{
+	if (kopald_tsk)
+		wake_up_process(kopald_tsk);
+}
+
 static void opal_init_heartbeat(void)
 {
 	/* Old firwmware, we assume the HVC heartbeat is sufficient */
@@ -671,7 +679,7 @@ static void opal_init_heartbeat(void)
 		opal_heartbeat = 0;
 
 	if (opal_heartbeat)
-		kthread_run(kopald, NULL, "kopald");
+		kopald_tsk = kthread_run(kopald, NULL, "kopald");
 }
 
 static int __init opal_init(void)
#1 lore raw
Quoted 9 lines · Benjamin Herrenschmidt

On some environments (prototype machines, some simulators, etc...) there is no functional interrupt source to signal completion, so we rely on the fairly slow OPAL heartbeat.

In a number of cases, the calls complete very quickly or even immediately. We've observed that it helps a lot to wakeup the OPAL heartbeat thread before waiting for event in those cases, it will call OPAL immediately to collect completions for anything that finished fast enough.

Any performance/jitter impact on normal systems?

cheers

#2 lore raw
Quoted 5 lines · Michael Ellerman
Quoted 4 lines · Benjamin Herrenschmidt

On some environments (prototype machines, some simulators, etc...) there is no functional interrupt source to signal completion, so we rely on the fairly slow OPAL heartbeat.

Quoted 7 lines
Quoted 5 lines

In a number of cases, the calls complete very quickly or even immediately. We've observed that it helps a lot to wakeup the OPAL heartbeat thread before waiting for event in those cases, it will call OPAL immediately to collect completions for anything that finished fast enough.

Any performance/jitter impact on normal systems?

Not that I can think of.

Cheers,
Ben.

Acked-by on v1 1/1
#3 lore raw
In reply to Benjamin Herrenschmidt
Quoted 11 lines · Benjamin Herrenschmidt

On some environments (prototype machines, some simulators, etc...) there is no functional interrupt source to signal completion, so we rely on the fairly slow OPAL heartbeat.
=20
In a number of cases, the calls complete very quickly or even immediately. We've observed that it helps a lot to wakeup the OPAL heartbeat thread before waiting for event in those cases, it will call OPAL immediately to collect completions for anything that finished fast enough.
=20

Without this using drivers/mtd/devices/powernv_flash.c on mambo is impossible. =C2=A0Feels like it's at least 1000x faster.

Quoted 2 lines

---
=C2=A0arch/powerpc/include/asm/opal.h=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=

=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0|=C2=A0=C2=A02 ++

Quoted 2 lines

=C2=A0arch/powerpc/platforms/powernv/opal-async.c |=C2=A0=C2=A05 +++++ =C2=A0arch/powerpc/platforms/powernv/opal.c=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=

=C2=A0=C2=A0| 12 ++++++++++--

Quoted 88 lines

=C2=A03 files changed, 17 insertions(+), 2 deletions(-) =20

arch/powerpc/include/asm/opal.h +0 -0
index 6135816..0c76bc0 100644
@@ -277,6 +277,8 @@ extern int opal_error_code(int rc);

=C2=A0
=C2=A0ssize_t opal_msglog_copy(char *to, loff_t pos, size_t count); =C2=A0

arch/powerpc/include/asm/opal.h +2 -0
+void opal_wake_poller(void);
+

=C2=A0#endif /* __ASSEMBLY__ */
=C2=A0
=C2=A0#endif /* _ASM_POWERPC_OPAL_H */

arch/powerpc/platforms/powernv/opal-async.c +0 -0
index bdc8c0c..83bebee 100644
@@ -117,6 +117,11 @@ int opal_async_wait_response(uint64_t token, struct
 opal_msg *msg)
 =C2=A0		return -EINVAL;
 =C2=A0	}

=C2=A0

arch/powerpc/platforms/powernv/opal-async.c +5 -0
+	/* Wakeup the poller before we wait for events to speed things
+	=C2=A0* up on platforms or simulators where the interrupts aren't
+	=C2=A0* functional.
+	=C2=A0*/
+	opal_wake_poller();

=C2=A0 wait_event(opal_async_wait, test_bit(token,
opal_async_complete_map));
=C2=A0 memcpy(msg, &opal_async_responses[token], sizeof(*msg));
=C2=A0

arch/powerpc/platforms/powernv/opal.c +0 -0
index 802f3b7..7f13302 100644
@@ -55,6 +55,7 @@ struct device_node *opal_node;
 =C2=A0static DEFINE_SPINLOCK(opal_write_lock);

=C2=A0static struct atomic_notifier_head
opal_msg_notifier_head[OPAL_MSG_TYPE_MAX];
=C2=A0static uint32_t opal_heartbeat;
+static struct task_struct *kopald_tsk;
=C2=A0
=C2=A0void opal_configure_cores(void)
=C2=A0{

arch/powerpc/platforms/powernv/opal.c +0 -0
@@ -650,6 +651,7 @@ static void opal_i2c_create_devs(void)

=C2=A0
=C2=A0static int kopald(void *unused)
=C2=A0{

arch/powerpc/platforms/powernv/opal.c +1 -0
+	unsigned long timeout =3D msecs_to_jiffies(opal_heartbeat) + 1;

=C2=A0 __be64 events;
=C2=A0
=C2=A0 set_freezable();

arch/powerpc/platforms/powernv/opal.c +1 -1
@@ -657,12 +659,18 @@ static int kopald(void *unused)
 =C2=A0		try_to_freeze();
 =C2=A0		opal_poll_events(&events);
 =C2=A0		opal_handle_events(be64_to_cpu(events));
-		msleep_interruptible(opal_heartbeat);
+		schedule_timeout_interruptible(timeout);
 =C2=A0	} while (!kthread_should_stop());

=C2=A0
=C2=A0 return 0;
=C2=A0}
=C2=A0

arch/powerpc/platforms/powernv/opal.c +6 -0
+void opal_wake_poller(void)
+{
+	if (kopald_tsk)
+		wake_up_process(kopald_tsk);
+}
+

=C2=A0static void opal_init_heartbeat(void)
=C2=A0{
=C2=A0 /* Old firwmware, we assume the HVC heartbeat is sufficient */

arch/powerpc/platforms/powernv/opal.c +0 -0
@@ -671,7 +679,7 @@ static void opal_init_heartbeat(void)
 =C2=A0		opal_heartbeat =3D 0;

=C2=A0
=C2=A0 if (opal_heartbeat)

arch/powerpc/platforms/powernv/opal.c +1 -1
-		kthread_run(kopald, NULL, "kopald");
+		kopald_tsk =3D kthread_run(kopald, NULL, "kopald");

=C2=A0}
=C2=A0
=C2=A0static int __init opal_init(void)
=20
=20

reviewer Applied
#4 lore raw
In reply to Benjamin Herrenschmidt
Quoted 4 lines · Benjamin Herrenschmidt

On some environments (prototype machines, some simulators, etc...) there is no functional interrupt source to signal completion, so we rely on the fairly slow OPAL heartbeat.

Quoted 8 lines

In a number of cases, the calls complete very quickly or even immediately. We've observed that it helps a lot to wakeup the OPAL heartbeat thread before waiting for event in those cases, it will call OPAL immediately to collect completions for anything that finished fast enough.

Applied to powerpc next, thanks.

https://git.kernel.org/powerpc/c/a203658b5ed37c11e5016d3fbb

cheers