tree fa3c94f2b61915c9ad10281e0ad29d9840f7e931
parent d551bbb69e8c8394c98c1498a66c15d5d276f103
author Eli Lipsitz <elipsitz@pigweed.infra.roller.google.com> 1737739096 -0800
committer CQ Bot Account <pigweed-scoped@luci-project-accounts.iam.gserviceaccount.com> 1737739096 -0800

roll: pigweed pw_digital_io_mcuxpresso: Use hardware level interrupts instead of edge

This commit updates the internal implementation of
McuxpressoDigitalInOutInterrupt to use hardware level-sensitive
interrupts to implement all edge-sensitive interrupts. Previously, only
"kBothEdges" interrupts were implemented with level-sensitive
interrupts.

The reason for this change is to work around a hardware bug around
deep sleep and GPIO edge interrupts on the RT500 MCU.

If, for example, a falling-edge interrupt is set, and the chip enters
deep-sleep while the line is low, the chip will wake up immediately (the
wakeup signal from the GPIO block always acts as level-sensitive in
deep sleep). However, the edge detector in the block won't register an
edge (as none occurred), so no interrupt will be pended in the NVIC. The
chip will exit deep sleep, but there's nothing to do and no interrupts
pending, so the chip enters deep sleep again. All of this happens
without the CPU ever actually exiting WFI and executing instructions. If
the line is still low when the chip enters deep sleep again, it'll
repeat this over and over again.

This "edge-sensitive emulation" works, at the cost of 2x the interrupts
(handled internally in this class), because the system will wake up for
the opposing edge (for internal polarity swapping and bookkeeping), in
addition to the one it actually cares about.

Original-Bug: b/370770558
Tested: Tried both falling and rising interrupts, confirmed they fired
Tested: when expected.
Tested: Also enabled deep sleep and set up an falling-edge interrupt
Tested: attached to a button. Watched the PMIC_SEL0 pin in a logic
Tested: analyzer. Confirmed that before this change, pressing the
Tested: button causes rapid oscillations on PMIC_SEL0 (~5 us period).
Tested: After this change, each time the button is pressed or released
Tested: the system briefly wakes from deep sleep to service the
Tested: interrupt, but then goes back to sleep.
Original-Reviewed-on: https://pigweed-review.googlesource.com/c/pigweed/pigweed/+/261612
Presubmit-Verified: CQ Bot Account <pigweed-scoped@luci-project-accounts.iam.gserviceaccount.com>
Original-Revision: b2441ff1e55ad1fbd055099a1ff210f7f7e7c589

Rolled-Repo: https://pigweed.googlesource.com/pigweed/pigweed
Rolled-Commits: 3d1d7aec46d512..b2441ff1e55ad1
Roll-Count: 1
Roller-URL: https://cr-buildbucket.appspot.com/build/8724842441734211217
GitWatcher: ignore
CQ-Do-Not-Cancel-Tryjobs: true
Change-Id: I1ddd4c942f2406beb044bbde720570e186702876
Reviewed-on: https://pigweed-review.googlesource.com/c/pigweed/quickstart/bazel/+/262733
Lint: Lint 🤖 <android-build-ayeaye@system.gserviceaccount.com>
Bot-Commit: Pigweed Roller <pigweed-roller@pigweed-service-accounts.iam.gserviceaccount.com>
Commit-Queue: Pigweed Roller <pigweed-roller@pigweed-service-accounts.iam.gserviceaccount.com>