NXP imx95 add basic pm support #91410

yongxu-wang15 · 2025-06-11T08:51:11Z

added basic pm support for imx95 evk m7 board, verified it in power_mgmt_soc test case, based on this case, M7 can enter/exit wait/suspend mode as we expected.

yongxu-wang15 · 2025-06-18T09:28:34Z

hi @ceolin @bjarki-andreasen @yangbolu1991, @JiafeiPan , can you help review this pr when you time free? its basic pm support for imx95 m7 core, thanks

JiafeiPan · 2025-06-18T09:48:52Z

drivers/firmware/scmi/core.c

+{
+	struct k_thread *current = k_current_get();
+
+	return strcmp(k_thread_name_get(current), "idle") == 0;


use z_is_idle_thread_object(current) ?

This is internal to kernel, but yep, this is better than check the name that depends on CONFIG_THREAD_NAME

thanks, have updated

JiafeiPan · 2025-06-18T09:49:20Z

drivers/firmware/scmi/core.c

@@ -158,6 +158,13 @@ static int scmi_send_message_post_kernel(struct scmi_protocol *proto,
 	return ret;
 }

+bool scmi_is_current_thread_idle(void)


JiafeiPan · 2025-06-18T09:50:50Z

drivers/firmware/scmi/core.c

+	 * the sm interrupt should be disabled in the soc low power phase.
+	 * In this case, the pre kernel is used instead of the post kernel.
+	 */
+	use_pre_kernel_path = use_pre_kernel_path || scmi_is_current_thread_idle();


use_pre_kernel_path |= scmi_is_current_thread_idle();

thanks, updated

JiafeiPan · 2025-06-18T09:55:37Z

soc/nxp/imx/imx9/imx95/m7/soc.c

+	irq_unlock(0);
+
+	/* disabled system manager irq */
+	irq_disable(DT_IRQN(DT_NODELABEL(mu5)));


Would you please help me to understand why still need to disable SM irq as you have already called __disable_irq(); which should have disable all interrupt?

the above code only mask irq to don't give reaction to interrupt, the capacity of wakeup Mcore from wfi is reserved, and the sm protocol use interrupt from scmi notification logic even though use scmi pre_kernel logic, so we need disabled mu5 now.

I agree @JiafeiPan i still don't get why mu5 irq needs to be disabled if IRQ's are disabled already.

the system manager will reply message in currently mechanism, so we need keep mu5 is be disabled to not wakeup m7 core from wfi after run scmi interface api.

the _disable_irq is to set PRIMASK register, it will mask all regular peripheral interrupt responce for cpu(such as m7 core), it mean that m7 can't enter nvic irq handler in this side, but this will not shield the triggering of interrupts, which will cause the CPU to be awakened from WFI. So in low power use case, although we have already set _disable_irq, but the cpu still have recognize this MU interrupt and thus leave WFI.

Isn't then the system design of system manager flawed? And needs reconsideration?

I think Linux uses the same mechanism using SCMI for PM does that work well then?

in linux side, the low power flow is in atf side, the atf use another mu, the atf mu use polling mode and set non-interrupt in there, but it have not this distinguish in zephyr now, and this way is hard to dela with in zephyr as it need another MU hardware resource, I think we can set mu hardware interface into non-interrupt mode in per kernel send scmi message side, in this it can fix this issue from my understand.

Yes putting SCMI into non-interrupt mode for PM makes sense. Then poll response after wake up, this doesn't do busy waiting. Then after wakeup put SCMI MU into interrupt mode.

sure, have updated it, now in pre kernel process(include idle task pm case), it will disable shem completion interrupt and mu side interrupt, and resume it after send message done. please help review it again

JiafeiPan · 2025-06-18T10:00:30Z

soc/nxp/imx/imx9/imx95/m7/soc.c

+		scmi_cpu_sleep_mode_set(&cpu_cfg);
+		__DSB();
+		__WFI();
+		__ISB();


whether need ISB after WFI

updated. thanks

bjarki-andreasen

Don't know anything about SCMI, but at least some irq_disable() may be superfluous :)

bjarki-andreasen · 2025-06-18T15:46:44Z

soc/nxp/imx/imx9/imx95/m7/soc.c

+	/* Set PRIMASK, Shield the CPU from responding to interrupts,
+	 * the CPU will not jump to the interrupt service routine (ISR).
+	 */
+	__disable_irq();
+	/* Set BASEPRI to 0 */
+	irq_unlock(0);


when we get here, irqs are already disabled by PM, see

zephyr/kernel/idle.c

Line 51 in a626d86

(void) arch_irq_lock();

Maybe because this arch using BASEPRI instead of masking interruptions globally ? This is still confusing for me too.

hi @ceolin @bjarki-andreasen , this is as imx95 m7 is based on ARMv7-M architecture, For this architecture, the current implementation of arch_irq_lock of zephyr is based on BASEPRI, which will only retain abnormal interrupts such as NMI,
and all other interrupts from the CPU(including systemtick) will be masked, which makes the CORE unable to be woken up from WFI. so we used primask rather than defaultly basepri in there

is that a bug in the implementation of arch_irq_lock() for this arch? arch_irq_lock() is only supposed to block ISRs from being executed, not the CPU from waking up. Maybe a fix at arch level would be more fitting, or?

if we want keep arch_irq_lock keep wakeup source capacity for all arch, it have some issue from my understand, its best to fix in top arch level, https://github.com/zephyrproject-rtos/zephyr/blob/main/include/zephyr/arch/arm/cortex_m/exception.h#L53

ceolin · 2025-06-18T16:23:30Z

drivers/firmware/scmi/core.c

+{
+	struct k_thread *current = k_current_get();
+
+	return strcmp(k_thread_name_get(current), "idle") == 0;


This is internal to kernel, but yep, this is better than check the name that depends on CONFIG_THREAD_NAME

ceolin · 2025-06-18T16:33:56Z

dts/arm/nxp/nxp_imx95_m7.dtsi

+		};
+
+		power-states {
+			active: power-state-active {


What this state means ? Power states are used by the power management subsystem and are used only by the idle thread.

Since you don't have any special treatment for this state in pm_state_set you can remove it. This way the power subsystem won't find a suitable state and will just sit idle (wfi or wfe).

If you keep it the idle thread will continuously loop issuing pm_state_set and pm_state_exit_post_ops

thanks reminder, have deleted it

ceolin · 2025-06-18T16:50:14Z

soc/nxp/imx/imx9/imx95/m7/soc.c

+	/* Set PRIMASK, Shield the CPU from responding to interrupts,
+	 * the CPU will not jump to the interrupt service routine (ISR).
+	 */
+	__disable_irq();
+	/* Set BASEPRI to 0 */
+	irq_unlock(0);


Maybe because this arch using BASEPRI instead of masking interruptions globally ? This is still confusing for me too.

PetervdPerk-NXP

Could you verify if the systick is actually capable of waking up the core when doing wfi?
We've seen some issues with this, that other interrupts can wake the m7 but the systisck doesn't thus giving big delays.

yongxu-wang15 · 2025-06-20T03:33:29Z

hi @PetervdPerk-NXP , in my side, the system tick work fine on this patch

bjarki-andreasen

Regarding __disable_irq(), its soc dependent, so if it works, its fine. Regarding using internal kernel apis to check for idle thread context, that seems like a clear layering violation, and I have proposed a solution which seems to be more appropriate :)

drivers/firmware/scmi/CMakeLists.txt

drivers/firmware/scmi/core.c

added wait/stop/suspend pm node for imx95 m7 Signed-off-by: Yongxu Wang <[email protected]>

add basic pm_state_set and pm_state_exit_post_ops Signed-off-by: Yongxu Wang <[email protected]>

enable imx95_evk m7 for this power_mgmt_soc test case Signed-off-by: Yongxu Wang <[email protected]>

yongxu-wang15 · 2025-07-03T08:29:32Z

update this patch to re-name cpu into cpu0: cpu@0 in order to align pm test case, please help review this pr when time free, thanks a lot

SCMI now supports both polling and interrupt modes to determine message completion. By default, polling is used during the pre-kernel phase, while interrupts are used post-kernel. However, for certain SCMI APIs related to power management (PM), although they are invoked post-kernel, using interrupt mode may cause unintended CPU wakeups. To prevent this, polling mode should be used instead. This patch extends the scmi_send_message() interface by adding a 'pre_kernel' parameter, allowing callers to explicitly indicate the desired behavior based on context, rather than relying on internal kernel state detection. Signed-off-by: Yongxu Wang <[email protected]>

in pre kernel scmi send message process, it should use poll to receive system manager core notification, add disable and re-enable completion interrupt in this flow. Signed-off-by: Yongxu Wang <[email protected]>

add scmi_cpu_pd_lpm_set api for nxp imx scmi interface Signed-off-by: Yongxu Wang <[email protected]>

wakeupmix keep power on state is essential for system suspend mode, because of console uart locate in it. temporarily set the M7 mix to power on, further optimization will be carried out later Signed-off-by: Yongxu Wang <[email protected]>

yongxu-wang15 · 2025-07-22T08:39:01Z

newly update:

from @bjarki-andreasen suggestion, add pre_kernel variable into scmi_send_message api, and do kernel state judge in scmi higher level api
add scmi nxp cpu lpm interface, and set M7/wakeup mix to keep power on state when m7 cpu enter suspend mode

sonarqubecloud · 2025-07-22T08:40:55Z

Quality Gate passed

Issues
0 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

bjarki-andreasen

Nice work! looks clean :)

bjarki-andreasen · 2025-07-23T10:49:48Z

ping @ceolin

github-actions bot added platform: NXP MPU area: Power Management labels Jun 11, 2025

github-actions bot requested review from bjarki-andreasen, ceolin, dbaluta, dleach02, iuliana-prodan, JiafeiPan, JordanYates, nashif, teburd, tmleman and yangbolu1991 June 11, 2025 08:52

github-actions bot assigned ceolin and bjarki-andreasen Jun 11, 2025

yongxu-wang15 force-pushed the nxp-imx95-add-basic-pm-support branch 2 times, most recently from 9ab89c6 to 44fd99b Compare June 12, 2025 01:29

yongxu-wang15 changed the title ~~Nxp imx95 add basic pm support~~ NXP imx95 add basic pm support Jun 12, 2025

JiafeiPan reviewed Jun 18, 2025

View reviewed changes

bjarki-andreasen requested changes Jun 18, 2025

View reviewed changes

ceolin requested changes Jun 18, 2025

View reviewed changes

PetervdPerk-NXP reviewed Jun 19, 2025

View reviewed changes

yongxu-wang15 force-pushed the nxp-imx95-add-basic-pm-support branch from 44fd99b to d9fc857 Compare June 20, 2025 03:20

yongxu-wang15 force-pushed the nxp-imx95-add-basic-pm-support branch from d9fc857 to 4708bc7 Compare June 20, 2025 05:15

yongxu-wang15 requested review from JiafeiPan, PetervdPerk-NXP, ceolin and bjarki-andreasen June 23, 2025 01:26

bjarki-andreasen requested changes Jun 23, 2025

View reviewed changes

drivers/firmware/scmi/CMakeLists.txt Outdated Show resolved Hide resolved

drivers/firmware/scmi/core.c Outdated Show resolved Hide resolved

yongxu-wang15 force-pushed the nxp-imx95-add-basic-pm-support branch from 4708bc7 to c966036 Compare June 26, 2025 09:40

yongxu-wang15 added 3 commits July 3, 2025 13:54

dts: arm: nxp_imx95_m7: add pm node

e4251f2

added wait/stop/suspend pm node for imx95 m7 Signed-off-by: Yongxu Wang <[email protected]>

soc: nxp: imx9: add basic pm process for i.MX95 M7

da58d7c

add basic pm_state_set and pm_state_exit_post_ops Signed-off-by: Yongxu Wang <[email protected]>

pm: power_mgmt_soc: add imx95_evk m7 into test case yml

0c59458

enable imx95_evk m7 for this power_mgmt_soc test case Signed-off-by: Yongxu Wang <[email protected]>

yongxu-wang15 force-pushed the nxp-imx95-add-basic-pm-support branch from c966036 to fe2807b Compare July 3, 2025 08:26

yongxu-wang15 added 2 commits July 22, 2025 14:25

drivers: scmi: disable peer core interrupt when use pre kernel process

94f3ca1

in pre kernel scmi send message process, it should use poll to receive system manager core notification, add disable and re-enable completion interrupt in this flow. Signed-off-by: Yongxu Wang <[email protected]>

yongxu-wang15 force-pushed the nxp-imx95-add-basic-pm-support branch from fe2807b to 866dd58 Compare July 22, 2025 07:10

zephyrbot added the platform: NXP Drivers NXP Semiconductors, drivers label Jul 22, 2025

zephyrbot requested review from mmahadevan108 and Zhiqiang-Hou July 22, 2025 07:11

yongxu-wang15 added 2 commits July 22, 2025 16:03

drivers: scmi: add cpu lpm interface api

3c10388

add scmi_cpu_pd_lpm_set api for nxp imx scmi interface Signed-off-by: Yongxu Wang <[email protected]>

yongxu-wang15 force-pushed the nxp-imx95-add-basic-pm-support branch from 866dd58 to 138eef8 Compare July 22, 2025 08:07

bjarki-andreasen approved these changes Jul 23, 2025

View reviewed changes

NXP imx95 add basic pm support #91410

Are you sure you want to change the base?

NXP imx95 add basic pm support #91410

Conversation

yongxu-wang15 commented Jun 11, 2025

Uh oh!

yongxu-wang15 commented Jun 18, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yongxu-wang15 Jun 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bjarki-andreasen left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

bjarki-andreasen Jun 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

yongxu-wang15 Jun 20, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

PetervdPerk-NXP left a comment

Choose a reason for hiding this comment

Uh oh!

yongxu-wang15 commented Jun 20, 2025

Uh oh!

bjarki-andreasen left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

yongxu-wang15 commented Jul 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

yongxu-wang15 commented Jun 18, 2025 •

edited

Loading

yongxu-wang15 Jun 25, 2025 •

edited

Loading

bjarki-andreasen Jun 20, 2025 •

edited

Loading

yongxu-wang15 Jun 20, 2025 •

edited

Loading

yongxu-wang15 commented Jul 3, 2025 •

edited

Loading

yongxu-wang15 commented Jul 22, 2025 •

edited

Loading