feat(linux): s2idle: Document the mode selection logic by DhruvaG2000 · Pull Request #641 · TexasInstruments/processor-sdk-doc

DhruvaG2000 · 2026-03-18T10:29:31Z

Document the mode selection logic using the s2idle flow

kwillis01

It might be good to see if there are other places in the previous parts of the doc that can be updated or that the new info can loop into and consolidate.

kwillis01

Mostly looks good, I think that the QoS Latency Constraints and Mode Selection section needs to be fixed up to flow better and be a little more condensed.

kwillis01 · 2026-04-15T15:38:06Z

Additionally, could you update the RTC + DDR section in pm_am62lx_low_power_modes.rst with the updated commands needed to enter RTC + DDR? Then you can point to this page from there if the user wants an indepth explanation on s2idle

github-actions · 2026-04-21T10:48:54Z

New warnings found with rstcheck:

source/linux/Foundational_Components/Power_Management/pm_psci_s2idle.rst:537: (ERROR/3) Content block expected for the "note" directive; none found.
source/linux/Foundational_Components/Power_Management/pm_psci_s2idle.rst:538: (WARNING/2) Explicit markup ends without a blank line; unexpected unindent.

github-actions · 2026-04-22T06:14:28Z

New warnings found with rstcheck:

source/linux/Foundational_Components/Power_Management/pm_psci_s2idle.rst:505: (ERROR/3) Content block expected for the "note" directive; none found.
source/linux/Foundational_Components/Power_Management/pm_psci_s2idle.rst:506: (WARNING/2) Explicit markup ends without a blank line; unexpected unindent.

kwillis01

It might be good to put explicit instructions on how to enter RTC + DDR via s2idle (besides the constraint setting) in pm_am62lx_low_power_modes. It would give users a quick way to grab the instructions and go.

kwillis01 · 2026-04-24T14:49:16Z

+   If regular [mem] interface is selected in `/sys/power/mem_sleep`, the RTC Only + DDR low power mode requires special steps
   to enter. The steps are documented in :ref:`am62l_suspend_workarounds`.

+RTC Only Plus DDR


Why was this moved? The note added should be in the RTC + DDR section not before it.

Why should it be only in the RTC +DDR section? The note applies to low power mode selection in general. I feel it should belong at the top and hence this was done intentionally.

Yes, but now it is in the "RTC Only" section so it looks like it only applies to that mode. If anything, it should be put in the top of this page during the "Overview" section. If it is moved to the overview section, it should be noted s2idle only applies to DeepSleep and RTC + DDR for now.

However, in this release, DeepSleep is unstable using s2idle, so it might be good to move the note to inside RTC + DDR for now, then we can move it later when the bugs are fixed.

makes sense.

kwillis01 · 2026-04-24T15:45:59Z

+3. Only idle states with ``exit-latency-us + entry-latency-us`` ≤ constraint are considered
+4. The deepest eligible state is selected
+
+**Selecting Specific Low-Power Modes:**


Could you make an explicit mention of /dev/cpu_wakeup_latency here? I think it would be confusing to someone reading this section where QoS constraint is set, but which one?

okay will add that

kwillis01 · 2026-04-24T15:48:20Z

+
+Execute this in a subshell to avoid accidentally keeping it open indefinitely.
+
+**How It Sets QoS Constraints:**


The explanation in this section would make more sense under the How QoS Constraints Work section. Explaining this constraint AFTER giving an example with it doesn't make much sense. This explanation would help set the stage for the examples given in this section.

ti-scaria · 2026-04-27T05:58:32Z

+     - ``syscore_suspend/resume()`` **called** (saves critical platform state)
+
+   * - **Driver Suspend**
+     - Same device PM callbacks as deep sleep


Since we can have both deep sleep and RTC+DDR using s2idle, mentioning specifically deep sleep might create confusion

removing this entirely, since it's not a difference

ti-scaria · 2026-04-27T06:00:58Z

-     - **cpuidle**: Uses the standard CPUidle framework. Additionally, each driver is made idle by calling respective runtime suspend hooks.
-     - **suspend_ops**: Uses driver specific suspend operations before ``PSCI_SYSTEM_SUSPEND`` is called.
-       No governors exist to make any decisions.
+.. list-table:: S2Idle vs Deep Sleep Comparison


is this table for the deep(mem) path or deep sleep mode specifically

ti-scaria · 2026-04-27T06:07:57Z

+
+.. code-block:: bash
+
+   exec 4<>/dev/cpu_wakeup_latency; echo 0x3e8 >&4


Might want to keep the value same as in the above example

cshilwant · 2026-04-30T21:07:38Z

@DhruvaG2000 can you re-spin after checking comments from @kwillis01 @ti-scaria

kwillis01 · 2026-05-12T14:59:09Z

+   If regular [mem] interface is selected in `/sys/power/mem_sleep`, the RTC Only + DDR low power mode requires special steps
   to enter. The steps are documented in :ref:`am62l_suspend_workarounds`.

+RTC Only Plus DDR


Yes, but now it is in the "RTC Only" section so it looks like it only applies to that mode. If anything, it should be put in the top of this page during the "Overview" section. If it is moved to the overview section, it should be noted s2idle only applies to DeepSleep and RTC + DDR for now.

However, in this release, DeepSleep is unstable using s2idle, so it might be good to move the note to inside RTC + DDR for now, then we can move it later when the bugs are fixed.

kwillis01 · 2026-05-12T15:07:11Z

+This is a generic, pure software, light-weight variant of system suspend (also referred to as S2I or S2Idle).
+It allows more energy to be saved relative to runtime idle by freezing user space, suspending the timekeeping
+and putting all I/O devices into low-power states (possibly lower-power than available in the working state),
+such that the processors can spend time in their deepest idle states while the system is suspended.


This repeats a lot of the overview plus I don't think it gets what you want it to get across. Maybe focus on how the difference between the design principles is that CPUIdle takes the reigns in deciding the system state for s2idle while for DeepSleep it is pretty much all based on the firmware choosing the LPM based off of QoS constraints.

kwillis01 · 2026-05-12T15:19:57Z

+
+.. code-block:: console
+
+   root@am62lxx-evm:~# gcc testqos.c -o testqos


Note above the testqos.c sets the /dev/cpu_wakeup_latency QoS constraint and an example program is shown later.

Document the mode selection logic using the s2idle flow Signed-off-by: Dhruva Gole <d-gole@ti.com>

Add mention of the s2idle based mode selection being part of low power mode support as of this release Signed-off-by: Dhruva Gole <d-gole@ti.com>

Also removes myself from AM62X device owner. Add Scaria as another PM component owner Signed-off-by: Dhruva Gole <d-gole@ti.com>

github-actions Bot assigned devarsht, DhruvaG2000, gehariprasath, praneethbajjuri, r-vignesh, uditkumarti and vishalmti Mar 18, 2026

kwillis01 suggested changes Mar 18, 2026

View reviewed changes

ti-scaria reviewed Mar 20, 2026

View reviewed changes

DhruvaG2000 force-pushed the s2idle_mode_sel_v1 branch from 995b85e to 6209735 Compare March 31, 2026 17:26

github-actions Bot requested review from devarsht, gehariprasath, praneethbajjuri, r-vignesh, uditkumarti and vishalmti March 31, 2026 17:27

kwillis01 suggested changes Apr 1, 2026

View reviewed changes

DhruvaG2000 force-pushed the s2idle_mode_sel_v1 branch from 6209735 to 8532d2e Compare April 21, 2026 10:48

github-actions Bot assigned kwillis01 Apr 21, 2026

StaticRocket requested changes Apr 21, 2026

View reviewed changes

Comment thread source/linux/Foundational_Components/Power_Management/pm_psci_s2idle.rst Outdated

Comment thread source/linux/Foundational_Components/Power_Management/pm_psci_s2idle.rst Outdated

Comment thread source/linux/Foundational_Components/Power_Management/pm_psci_s2idle.rst

DhruvaG2000 force-pushed the s2idle_mode_sel_v1 branch from 8532d2e to 27dc856 Compare April 22, 2026 06:13

DhruvaG2000 force-pushed the s2idle_mode_sel_v1 branch from 3de7535 to f4ad757 Compare April 24, 2026 12:23

DhruvaG2000 changed the title ~~[RFC] feat(linux): s2idle: Document the mode selection logic~~ feat(linux): s2idle: Document the mode selection logic Apr 24, 2026

DhruvaG2000 marked this pull request as ready for review April 24, 2026 12:39

DhruvaG2000 requested review from VeeruPrudhvi, cshilwant and jeevantelukula as code owners April 24, 2026 12:39

kwillis01 suggested changes Apr 24, 2026

View reviewed changes

ti-scaria reviewed Apr 27, 2026

View reviewed changes

DhruvaG2000 force-pushed the s2idle_mode_sel_v1 branch from f4ad757 to 2c5e7bd Compare May 12, 2026 10:08

kwillis01 suggested changes May 12, 2026

View reviewed changes

DhruvaG2000 force-pushed the s2idle_mode_sel_v1 branch from 2c5e7bd to 19bd7c0 Compare May 13, 2026 08:36

DhruvaG2000 requested review from kwillis01 and ti-scaria May 13, 2026 08:36

DhruvaG2000 added 3 commits May 13, 2026 17:17

feat(linux): s2idle: Document the mode selection logic

f7dfaeb

Document the mode selection logic using the s2idle flow Signed-off-by: Dhruva Gole <d-gole@ti.com>

feat(pm_am62lx_low_power_modes): mention s2idle approach

4b77e16

Add mention of the s2idle based mode selection being part of low power mode support as of this release Signed-off-by: Dhruva Gole <d-gole@ti.com>

chore(owners): Update Power Management component owners

34ae59d

Also removes myself from AM62X device owner. Add Scaria as another PM component owner Signed-off-by: Dhruva Gole <d-gole@ti.com>

DhruvaG2000 force-pushed the s2idle_mode_sel_v1 branch from 19bd7c0 to 34ae59d Compare May 13, 2026 11:47

StaticRocket self-requested a review May 13, 2026 22:15


		Execute this in a subshell to avoid accidentally keeping it open indefinitely.

		How It Sets QoS Constraints:


		.. code-block:: bash

		exec 4<>/dev/cpu_wakeup_latency; echo 0x3e8 >&4


		.. code-block:: console

		root@am62lxx-evm:~# gcc testqos.c -o testqos

Conversation

DhruvaG2000 commented Mar 18, 2026

Uh oh!

kwillis01 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

kwillis01 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

kwillis01 commented Apr 15, 2026

Uh oh!

github-actions Bot commented Apr 21, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

github-actions Bot commented Apr 22, 2026

Uh oh!

kwillis01 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!

Choose a reason for hiding this comment

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!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

cshilwant commented Apr 30, 2026

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!

Uh oh!

Uh oh!

Uh oh!