diff mbox series

[v3,3/3] examples/l3fwd-power: enable PMD power monitor on Arm

Message ID 20221111072649.3304429-4-feifei.wang2@arm.com (mailing list archive)
State Superseded, archived
Delegated to: Thomas Monjalon
Headers
Series Enable PMD power management on Arm |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation success Compilation OK
ci/intel-Testing success Testing PASS
ci/github-robot: build success github build: passed

Commit Message

Feifei Wang Nov. 11, 2022, 7:26 a.m. UTC 
  For Arm aarch, power monitor uses WFE instruction to enable, which can
not exit automatically within the time limit. This means
'rte_power_monitor_wakeup' API needs to be called to wake up sleep cores
if there is no store operation to monitored address.

Furthermore, we disable power monitor feature on the main core so that
it can be used to wake up other sleeping cores after it exit from loop.

Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
Acked-by: David Hunt <david.hunt@intel.com>
---
 examples/l3fwd-power/main.c | 25 +++++++++++++++++++++++++
 1 file changed, 25 insertions(+)

Comments

Thomas Monjalon Nov. 11, 2022, 8:22 a.m. UTC | #1 
11/11/2022 08:26, Feifei Wang:
> For Arm aarch, power monitor uses WFE instruction to enable, which can
> not exit automatically within the time limit. This means
> 'rte_power_monitor_wakeup' API needs to be called to wake up sleep cores
> if there is no store operation to monitored address.
> 
> Furthermore, we disable power monitor feature on the main core so that
> it can be used to wake up other sleeping cores after it exit from loop.
> 
> Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
> Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> Acked-by: David Hunt <david.hunt@intel.com>
> ---
> +#if defined(RTE_ARCH_ARM64)
> +				/* Ensure the main lcore does not enter the power-monitor state,
> +				 * so that it can be used to wake up other lcores on ARM.
> +				 * This is due to WFE instruction has no timeout wake-up mechanism,
> +				 * and if users want to exit actively, the main lcore is needed
> +				 * to send SEV instruction to wake up other lcores.
> +				 */
> +				unsigned int main_lcore = rte_get_main_lcore();
> +				if (lcore_id == main_lcore &&
> +						pmgmt_type == RTE_POWER_MGMT_TYPE_MONITOR)
> +					continue;
> +#endif

We need to have this logic abstracted in the API for all architectures.
We cannot afford having such #ifdef per CPU in the application.
Feifei Wang Nov. 11, 2022, 10:21 a.m. UTC | #2 
Hi, Thomas
> -----邮件原件-----
> 发件人: Thomas Monjalon <thomas@monjalon.net>
> 发送时间: Friday, November 11, 2022 4:22 PM
> 收件人: Feifei Wang <Feifei.Wang2@arm.com>
> 抄送: David Hunt <david.hunt@intel.com>; dev@dpdk.org;
> david.marchand@redhat.com; stephen@networkplumber.org; nd
> <nd@arm.com>; Ruifeng Wang <Ruifeng.Wang@arm.com>; Honnappa
> Nagarahalli <Honnappa.Nagarahalli@arm.com>
> 主题: Re: [PATCH v3 3/3] examples/l3fwd-power: enable PMD power
> monitor on Arm
> 
> 11/11/2022 08:26, Feifei Wang:
> > For Arm aarch, power monitor uses WFE instruction to enable, which can
> > not exit automatically within the time limit. This means
> > 'rte_power_monitor_wakeup' API needs to be called to wake up sleep
> > cores if there is no store operation to monitored address.
> >
> > Furthermore, we disable power monitor feature on the main core so that
> > it can be used to wake up other sleeping cores after it exit from loop.
> >
> > Signed-off-by: Feifei Wang <feifei.wang2@arm.com>
> > Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> > Acked-by: David Hunt <david.hunt@intel.com>
> > ---
> > +#if defined(RTE_ARCH_ARM64)
> > +				/* Ensure the main lcore does not enter the
> power-monitor state,
> > +				 * so that it can be used to wake up other
> lcores on ARM.
> > +				 * This is due to WFE instruction has no
> timeout wake-up mechanism,
> > +				 * and if users want to exit actively, the main
> lcore is needed
> > +				 * to send SEV instruction to wake up other
> lcores.
> > +				 */
> > +				unsigned int main_lcore =
> rte_get_main_lcore();
> > +				if (lcore_id == main_lcore &&
> > +						pmgmt_type ==
> RTE_POWER_MGMT_TYPE_MONITOR)
> > +					continue;
> > +#endif
> 
> We need to have this logic abstracted in the API for all architectures.
> We cannot afford having such #ifdef per CPU in the application.

Thanks for the comment. I will move this into API.

Best Regards
Feifei
>
diff mbox series

Patch

diff --git a/examples/l3fwd-power/main.c b/examples/l3fwd-power/main.c
index fd3ade330f..7c41159c43 100644
--- a/examples/l3fwd-power/main.c
+++ b/examples/l3fwd-power/main.c
@@ -3006,6 +3006,18 @@  main(int argc, char **argv)
 						"Error setting scaling freq max: err=%d, lcore %d\n",
 							ret, lcore_id);
 
+#if defined(RTE_ARCH_ARM64)
+				/* Ensure the main lcore does not enter the power-monitor state,
+				 * so that it can be used to wake up other lcores on ARM.
+				 * This is due to WFE instruction has no timeout wake-up mechanism,
+				 * and if users want to exit actively, the main lcore is needed
+				 * to send SEV instruction to wake up other lcores.
+				 */
+				unsigned int main_lcore = rte_get_main_lcore();
+				if (lcore_id == main_lcore &&
+						pmgmt_type == RTE_POWER_MGMT_TYPE_MONITOR)
+					continue;
+#endif
 				ret = rte_power_ethdev_pmgmt_queue_enable(
 						lcore_id, portid, queueid,
 						pmgmt_type);
@@ -3113,6 +3125,19 @@  main(int argc, char **argv)
 	if (app_mode == APP_MODE_EMPTY_POLL || app_mode == APP_MODE_TELEMETRY)
 		launch_timer(rte_lcore_id());
 
+	/* wake up all worker cores from sleeping state */
+	if (pmgmt_type == RTE_POWER_MGMT_TYPE_MONITOR) {
+		for (lcore_id = 0; lcore_id < RTE_MAX_LCORE; lcore_id++) {
+			if (rte_lcore_is_enabled(lcore_id) == 0)
+				continue;
+
+			if (lcore_id == rte_get_main_lcore())
+				continue;
+
+			rte_power_monitor_wakeup(lcore_id);
+		}
+	}
+
 	RTE_LCORE_FOREACH_WORKER(lcore_id) {
 		if (rte_eal_wait_lcore(lcore_id) < 0)
 			return -1;