Enable deep suspend on SM8550 and drop the Pocket EVO fan tacho IRQ - #23
Closed
jesherman wants to merge 1 commit into
Closed
Enable deep suspend on SM8550 and drop the Pocket EVO fan tacho IRQ#23jesherman wants to merge 1 commit into
jesherman wants to merge 1 commit into
Conversation
The AYANEO Pocket EVO cannot hold suspend-to-RAM. Measured on device it leaves deep suspend after 2-92s, and with a Steam session running it fails to resume at all and needs a hard reset. It ships ARMADA_SUSPEND_MODE=fake as a result, which costs about 162mA of standby drain. Two independent causes. The first is SoC-wide and already solved upstream. ROCKNIX PR #2954 (and jaewun's #2952 before it) fixes the SM8550 wake sources and UFS resume races: TSENS passive threshold IRQs armed for wake, a geni UART IRQ left unmasked across suspend, and a set of UFS hibern8/link-recovery races. Six of those patches are picked up here. Four are left out: 0502 (qcom-ipcc IRQF_NO_SUSPEND) is already carried as 0504, 1004 (rsinput MCU resume) targets a gamepad driver the EVO does not use, and 1007/1015 do not apply to 7.1.5. Deep suspend was validated without 1007/1015. The second cause is specific to the AYANEO Pockets and is not covered upstream, whose SM8550 testing was on a Retroid Pocket 6. pocket-common.dtsi declares an edge-falling tacho interrupt on TLMM GPIO 64 for pwm-fan. It fires ~80 times a second whenever the fan spins and repeatedly pulls the SoC back out of deep suspend; /proc/interrupts showed it climbing past 230,000. The Odin 3 declares no tacho interrupt and suspends for hours. It is overridden in the EVO .dts rather than the shared DTSI, which six other boards include. Cost: hwmon fan1_input stops reporting RPM, as on the Odin 3. Also enable the PM and IRQ debug options. Diagnosing this without pm_wakeup_irq meant guessing at wake sources across a dozen hard resets; with them the offending IRQ names itself. That is how the fan tacho was found: /sys/kernel/debug/irq/irqs/<n> "wdepth" shows what is armed for wake. Measured on a Pocket EVO, on battery with Steam running: 1202s of uninterrupted deep suspend woken only by the RTC alarm, and standby drain of 27.8mA against 161.8mA for fake suspend - 5.8x lower, roughly 2 days of standby to 12. Those numbers are from the 7.0.11 build; 7.1.5 was then confirmed to hold deep suspend with the session running and resume cleanly. Note this set is version-fragile. Three of the eight patches broke moving from 7.0.11 to 7.1.5, and 0201 touches ufshcd_intr(), which 7.2 changes again by reverting the threaded IRQ handler. Plan to re-derive it rather than rebase on the next kernel bump. This only makes deep suspend work; it does not switch any device over to it. ARMADA_SUSPEND_MODE stays 'fake' everywhere until this has had a longer soak. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
jesherman
marked this pull request as draft
August 4, 2026 21:01
Contributor
Author
|
moving to draft as im getting inconsistent perf. |
Contributor
Author
|
no worries i completely agree. I'm still testing this as I found some intermittent performance issues/failure to wake and I do think there are some additional wrinkles here. Happy to pull a separate PR just for the fan DTS override |
Member
Yikes just realized Claude commented on this while I was having it inventory open PRs 🙃 But I (a human) agree with holding on this for now. |
xiaodoudou
pushed a commit
to xiaodoudou/distribution
that referenced
this pull request
Aug 6, 2026
qcs8550-ayaneo-pocket-common.dtsi gives pwm-fan an edge-falling TLMM 64 interrupt for the tachometer. Measured on a Pocket EVO it fires about 80 times a second whenever the fan spins and repeatedly pulls the SoC back out of deep suspend, with /proc/interrupts climbing past 230,000; with the fan stopped the same suspend holds. The AYN Odin 3 declares no tacho interrupt and suspends for hours. Overridden in the EVO .dts rather than in the shared DTSI, which six other boards include and whose fan wiring has not been checked. Cost: hwmon fan1_input stops reporting RPM, as on the Odin 3. Found and measured by jesherman in armada-os/armada-packages#23. The EVO stays outside the device gate; this removes one of the two reasons it is there.
Contributor
Author
|
replaced by #28 |
xiaodoudou
pushed a commit
to xiaodoudou/distribution
that referenced
this pull request
Aug 7, 2026
qcs8550-ayaneo-pocket-common.dtsi gives pwm-fan an edge-falling TLMM 64 interrupt for the tachometer. Measured on a Pocket EVO it fires about 80 times a second whenever the fan spins and repeatedly pulls the SoC back out of deep suspend, with /proc/interrupts climbing past 230,000; with the fan stopped the same suspend holds. The AYN Odin 3 declares no tacho interrupt and suspends for hours. Overridden in the EVO .dts rather than in the shared DTSI, which six other boards include and whose fan wiring has not been checked. Cost: hwmon fan1_input stops reporting RPM, as on the Odin 3. Found and measured by jesherman in armada-os/armada-packages#23. The EVO stays outside the device gate; this removes one of the two reasons it is there.
xiaodoudou
pushed a commit
to xiaodoudou/distribution
that referenced
this pull request
Aug 8, 2026
qcs8550-ayaneo-pocket-common.dtsi gives pwm-fan an edge-falling TLMM 64 interrupt for the tachometer. Measured on a Pocket EVO it fires about 80 times a second whenever the fan spins and repeatedly pulls the SoC back out of deep suspend, with /proc/interrupts climbing past 230,000; with the fan stopped the same suspend holds. The AYN Odin 3 declares no tacho interrupt and suspends for hours. Overridden in the EVO .dts rather than in the shared DTSI, which six other boards include and whose fan wiring has not been checked. Cost: hwmon fan1_input stops reporting RPM, as on the Odin 3. Found and measured by jesherman in armada-os/armada-packages#23. The EVO stays outside the device gate; this removes one of the two reasons it is there.
xiaodoudou
pushed a commit
to xiaodoudou/distribution
that referenced
this pull request
Aug 10, 2026
qcs8550-ayaneo-pocket-common.dtsi gives pwm-fan an edge-falling TLMM 64 interrupt for the tachometer. Measured on a Pocket EVO it fires about 80 times a second whenever the fan spins and repeatedly pulls the SoC back out of deep suspend, with /proc/interrupts climbing past 230,000; with the fan stopped the same suspend holds. The AYN Odin 3 declares no tacho interrupt and suspends for hours. Overridden in the EVO .dts rather than in the shared DTSI, which six other boards include and whose fan wiring has not been checked. Cost: hwmon fan1_input stops reporting RPM, as on the Odin 3. Found and measured by jesherman in armada-os/armada-packages#23. The EVO stays outside the device gate; this removes one of the two reasons it is there.
xiaodoudou
pushed a commit
to xiaodoudou/distribution
that referenced
this pull request
Aug 11, 2026
qcs8550-ayaneo-pocket-common.dtsi gives pwm-fan an edge-falling TLMM 64 interrupt for the tachometer. Measured on a Pocket EVO it fires about 80 times a second whenever the fan spins and repeatedly pulls the SoC back out of deep suspend, with /proc/interrupts climbing past 230,000; with the fan stopped the same suspend holds. The AYN Odin 3 declares no tacho interrupt and suspends for hours. Overridden in the EVO .dts rather than in the shared DTSI, which six other boards include and whose fan wiring has not been checked. Cost: hwmon fan1_input stops reporting RPM, as on the Odin 3. Found and measured by jesherman in armada-os/armada-packages#23. The EVO stays outside the device gate; this removes one of the two reasons it is there.
xiaodoudou
pushed a commit
to xiaodoudou/distribution
that referenced
this pull request
Aug 22, 2026
qcs8550-ayaneo-pocket-common.dtsi gives pwm-fan an edge-falling TLMM 64 interrupt for the tachometer. Measured on a Pocket EVO it fires about 80 times a second whenever the fan spins and repeatedly pulls the SoC back out of deep suspend, with /proc/interrupts climbing past 230,000; with the fan stopped the same suspend holds. The AYN Odin 3 declares no tacho interrupt and suspends for hours. Overridden in the EVO .dts rather than in the shared DTSI, which six other boards include and whose fan wiring has not been checked. Cost: hwmon fan1_input stops reporting RPM, as on the Odin 3. Found and measured by jesherman in armada-os/armada-packages#23. The EVO stays outside the device gate; this removes one of the two reasons it is there.
xiaodoudou
pushed a commit
to xiaodoudou/distribution
that referenced
this pull request
Aug 23, 2026
qcs8550-ayaneo-pocket-common.dtsi gives pwm-fan an edge-falling TLMM 64 interrupt for the tachometer. Measured on a Pocket EVO it fires about 80 times a second whenever the fan spins and repeatedly pulls the SoC back out of deep suspend, with /proc/interrupts climbing past 230,000; with the fan stopped the same suspend holds. The AYN Odin 3 declares no tacho interrupt and suspends for hours. Overridden in the EVO .dts rather than in the shared DTSI, which six other boards include and whose fan wiring has not been checked. Cost: hwmon fan1_input stops reporting RPM, as on the Odin 3. Found and measured by jesherman in armada-os/armada-packages#23. The EVO stays outside the device gate; this removes one of the two reasons it is there.
xiaodoudou
pushed a commit
to xiaodoudou/distribution
that referenced
this pull request
Aug 27, 2026
qcs8550-ayaneo-pocket-common.dtsi gives pwm-fan an edge-falling TLMM 64 interrupt for the tachometer. Measured on a Pocket EVO it fires about 80 times a second whenever the fan spins and repeatedly pulls the SoC back out of deep suspend, with /proc/interrupts climbing past 230,000; with the fan stopped the same suspend holds. The AYN Odin 3 declares no tacho interrupt and suspends for hours. Overridden in the EVO .dts rather than in the shared DTSI, which six other boards include and whose fan wiring has not been checked. Cost: hwmon fan1_input stops reporting RPM, as on the Odin 3. Found and measured by jesherman in armada-os/armada-packages#23. The EVO stays outside the device gate; this removes one of the two reasons it is there.
xiaodoudou
pushed a commit
to xiaodoudou/distribution
that referenced
this pull request
Aug 27, 2026
qcs8550-ayaneo-pocket-common.dtsi gives pwm-fan an edge-falling TLMM 64 interrupt for the tachometer. Measured on a Pocket EVO it fires about 80 times a second whenever the fan spins and repeatedly pulls the SoC back out of deep suspend, with /proc/interrupts climbing past 230,000; with the fan stopped the same suspend holds. The AYN Odin 3 declares no tacho interrupt and suspends for hours. Overridden in the EVO .dts rather than in the shared DTSI, which six other boards include and whose fan wiring has not been checked. Cost: hwmon fan1_input stops reporting RPM, as on the Odin 3. Found and measured by jesherman in armada-os/armada-packages#23. The EVO stays outside the device gate; this removes one of the two reasons it is there.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Makes suspend-to-RAM hold on the AYANEO Pocket EVO. Doesn't switch any device to it — that's armada#215.
Changes
Six patches from ROCKNIX #2954 — TSENS wake IRQs, geni UART, UFS resume races. Four of that PR's patches omitted:
0502already carried as0504;1004is for a gamepad driver the EVO doesn't use;1007/1015don't apply to 7.1.5 (kept as files, commented out inseries).Fan tacho DTS override — not covered upstream, whose SM8550 testing was on a Retroid Pocket 6.
pocket-common.dtsigivespwm-fanan edge-falling IRQ on TLMM GPIO 64; it fires ~80×/sec while the fan spins and repeatedly pulls the SoC out of suspend (/proc/interruptspassed 230,000). The Odin 3 declares none and sleeps for hours. Scoped to the EVO.dts, not the DTSI six other boards include. Costsfan1_inputRPM reporting, as on the Odin 3.PM_DEBUG/PM_SLEEP_DEBUG/GENERIC_IRQ_DEBUGFS—pm_wakeup_irqis what made this diagnosable. Easy to drop if you'd rather not ship debug options.Results
Pocket EVO, battery, Steam running: 1202s deep suspend woken only by the RTC alarm,
success=3 fail=0. Standby 27.8 mA vs 161.8 mA — 5.8× lower, ~2 days to ~12.From the 7.0.11 build. 7.1.5 confirmed working (session running, clean resume) but only a ~4 min run.
Caveat: version-fragile
Three of eight patches broke moving 7.0.11 → 7.1.5.
0201is the risk — only UFS-core patch, 424 lines inufshcd.c, trackingufshcd_intr(), which 7.2 changes again by reverting the threaded IRQ handler. Re-derive rather than rebase next bump. Dropping it hangs resume with Steam running, so it's load-bearing here.Testing
Pocket EVO, 7.1.5, 125 patches applied / 0 failed. Untested on other AYANEO Pockets; DTS change is EVO-scoped, kernel patches are
qcom,sm8550-guarded or Qualcomm-generic.🤖 Generated with Claude Code