Hi Dmitry, Answering the three questions, and flagging a claim in the commit message that your question made us re-read.
1) Is it an interrupt controller? Yes. MDSS is one in DT: sm8750.dtsi's mdss node has "interrupt-controller;" and "#interrupt-cells = <1>", and the DPU child takes its IRQ through "interrupts-extended = <&mdss 0>". INT2 is the second interrupt block inside MDSS; dispcc-sm8750 exposes it as MDSS_INT2_GDSC plus DISP_CC_MDSS_CORE_INT2_BCR (0xa000), next to DISP_CC_MDSS_CORE_BCR (0x8000). It is not a separate device. 2) What is the side effect? With the probe ordering taken into account, none: msm_mdss_reset() runs early in msm_mdss_probe() (msm_mdss.c:372), while the irq domain is created later (:417) and of_platform_populate() (:537) brings up the DPU after that. The group is asserted for 20 ms (the existing "one frame" value) and then released; dispcc PLLs and clocks are not touched, and RSCC is gone from the series as you asked. 3) How was it tested? At two levels, and we should have said which is which: - tested: the reset at this position (mdss probe, after dispcc PLL programming) with core+INT2+RSCC as one bulk group -- 5/5 cold POR and 2/2 warm resets with no bad boot (card0 present, DSI-1 connected, storm signature 0), against 5 STORM in 9 cold boots on the base arm that reset only the core; - not tested: this exact v3 shape (array API, two resets), and we never isolated INT2 as a single variable -- the failure is intermittent (~50%), so a per-reset A/B would need a large number of boots before it means anything. So 3/3's commit message overclaims: "command DMA timeouts ... because INT2 was left out of reset" is a causal statement we cannot support. We will send a v4 that keeps the DT change (it already has both R-bs) and drops the causal wording, leaving only what holds: the bootloader can hand over with both MDSS blocks running, so DT describes both resets and the driver resets them as a group. 2/3 only exists to make the driver honour a DT that lists more than one MDSS reset. If you would prefer 3/3 alone after the v4 rewrite, say so. AI-assistance disclosure: this reply was drafted with an LLM coding assistant; the author verified every statement above against the tree and against his own logs and takes responsibility for it. Qianhui
