[rp2040] Add interrupt guard to millis() accumulator

Wiegand and ZyAura call millis() from ISR handlers on all platforms including RP2040 (no platform restrictions on either component). The accumulator's mutable statics need protection from concurrent ISR access. Uses save_and_disable_interrupts()/restore_interrupts() — same pattern as InterruptLock in rp2040/helpers.cpp.
[rp2040] Bound the while loop in millis() accumulator
2026-07-05 06:29:41 +00:00 · 2026-04-12 09:11:41 -10:00 · 2026-04-11 23:22:12 -10:00 · 2026-04-11 23:17:12 -10:00 · 2026-04-11 22:38:29 -10:00 · 2026-04-11 22:22:32 -10:00
28 changed files with 263 additions and 89 deletions
@@ -47,7 +47,7 @@ runs:

    - name: Build and push to ghcr by digest
      id: build-ghcr
-      uses: docker/build-push-action@d08e5c354a6adb9ed34480a06d141179aa583294 # v7.0.0
+      uses: docker/build-push-action@bcafcacb16a39f128d818304e6c9c0c18556b85f # v7.1.0
      env:
        DOCKER_BUILD_SUMMARY: false
        DOCKER_BUILD_RECORD_UPLOAD: false
@@ -73,7 +73,7 @@ runs:

    - name: Build and push to dockerhub by digest
      id: build-dockerhub
-      uses: docker/build-push-action@d08e5c354a6adb9ed34480a06d141179aa583294 # v7.0.0
+      uses: docker/build-push-action@bcafcacb16a39f128d818304e6c9c0c18556b85f # v7.1.0
      env:
        DOCKER_BUILD_SUMMARY: false
        DOCKER_BUILD_RECORD_UPLOAD: false
@@ -33,7 +33,7 @@ jobs:
          private-key: ${{ secrets.ESPHOME_GITHUB_APP_PRIVATE_KEY }}

      - name: Auto Label PR
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          github-token: ${{ steps.generate-token.outputs.token }}
          script: |
@@ -47,7 +47,7 @@ jobs:
          fi
      - if: failure()
        name: Review PR
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          script: |
            await github.rest.pulls.createReview({
@@ -62,7 +62,7 @@ jobs:
        run: git diff
      - if: failure()
        name: Archive artifacts
-        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7.0.0
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: generated-proto-files
          path: |
@@ -70,7 +70,7 @@ jobs:
            esphome/components/api/api_pb2_service.*
      - if: success()
        name: Dismiss review
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          script: |
            let reviews = await github.rest.pulls.listReviews({
@@ -42,7 +42,7 @@ jobs:

      - if: failure() && github.event.pull_request.head.repo.full_name == github.repository
        name: Request changes
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          script: |
            await github.rest.pulls.createReview({
@@ -55,7 +55,7 @@ jobs:

      - if: success() && github.event.pull_request.head.repo.full_name == github.repository
        name: Dismiss review
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          script: |
            let reviews = await github.rest.pulls.listReviews({
@@ -904,7 +904,7 @@ jobs:
          fi

      - name: Upload memory analysis JSON
-        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7.0.0
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: memory-analysis-target
          path: memory-analysis-target.json
@@ -969,7 +969,7 @@ jobs:
            --platform "$platform"

      - name: Upload memory analysis JSON
-        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7.0.0
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: memory-analysis-pr
          path: memory-analysis-pr.json
@@ -34,7 +34,7 @@ jobs:
            CODEOWNERS

      - name: Check codeowner approval and update label
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        env:
          PR_NUMBER: ${{ github.event.pull_request.number }}
        with:
@@ -33,7 +33,7 @@ jobs:
          ref: ${{ github.event.pull_request.base.sha }}

      - name: Request reviews from component codeowners
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          script: |
            const { loadCodeowners, getEffectiveOwners } = require('./.github/scripts/codeowners.js');
@@ -15,7 +15,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Add external component comment
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          github-token: ${{ secrets.GITHUB_TOKEN }}
          script: |
@@ -19,7 +19,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Notify codeowners for component issues
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          script: |
            const owner = context.repo.owner;
@@ -18,7 +18,7 @@ jobs:
    steps:
      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2

-      - uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+      - uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          script: |
            const {
@@ -138,7 +138,7 @@ jobs:
      #     version: ${{ needs.init.outputs.tag }}

      - name: Upload digests
-        uses: actions/upload-artifact@bbbca2ddaa5d8feaa63e36b76fdaad77386f024f # v7.0.0
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
        with:
          name: digests-${{ matrix.platform.arch }}
          path: /tmp/digests
@@ -229,7 +229,7 @@ jobs:
          repositories: home-assistant-addon

      - name: Trigger Workflow
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          github-token: ${{ steps.generate-token.outputs.token }}
          script: |
@@ -264,7 +264,7 @@ jobs:
          repositories: esphome-schema

      - name: Trigger Workflow
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          github-token: ${{ steps.generate-token.outputs.token }}
          script: |
@@ -295,7 +295,7 @@ jobs:
          repositories: version-notifier

      - name: Trigger Workflow
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          github-token: ${{ steps.generate-token.outputs.token }}
          script: |
@@ -14,7 +14,7 @@ jobs:
    runs-on: ubuntu-latest
    steps:
      - name: Check for blocking labels
-        uses: actions/github-script@ed597411d8f924073f98dfc5c65a23a2325f34cd # v8.0.0
+        uses: actions/github-script@3a2844b7e9c422d3c10d287c895573f7108da1b3 # v9.0.0
        with:
          script: |
            const blockingLabels = ['needs-docs', 'merge-after-release', 'chained-pr'];
@@ -41,7 +41,7 @@ jobs:
          python script/run-in-env.py pre-commit run --all-files

      - name: Commit changes
-        uses: peter-evans/create-pull-request@c0f553fe549906ede9cf27b5156039d195d2ece0 # v8.1.0
+        uses: peter-evans/create-pull-request@5f6978faf089d4d20b00c7766989d076bb2fc7f1 # v8.1.1
        with:
          commit-message: "Synchronise Device Classes from Home Assistant"
          committer: esphomebot <esphome@openhomefoundation.org>
@@ -298,33 +298,15 @@ constexpr uint32_t VARINT_MAX_2_BYTE = 1 << 14;  // 16384
 class ProtoEncode {
 public:
  /// Write a multi-byte varint directly through a pos pointer.
-  /// Unrolled based on the compile-time max varint length for T
-  /// (5 bytes for uint32_t, 10 bytes for uint64_t). The explicit unroll gives
-  /// the compiler straight-line code with early exits instead of a
-  /// data-dependent back-edge branch, which measurably speeds up BLE raw
-  /// advertisement MAC encoding (always 7-byte varints) among other hot paths.
-  ///
-  /// Takes/returns pos by value so this function can remain out-of-line without
-  /// forcing the caller to spill pos to memory on every byte write. Callers
-  /// (which are ALWAYS_INLINE) should assign the returned pos back to their
-  /// local. Code size win vs. inlining the whole unroll at every call site.
  template<typename T>
-  static uint8_t *encode_varint_raw_loop(uint8_t *__restrict__ pos PROTO_ENCODE_DEBUG_PARAM, T value) {
-    constexpr int MAX_VARINT_BYTES = (sizeof(T) * 8 + 6) / 7;  // 5 for u32, 10 for u64
-#pragma GCC unroll 10
-    for (int i = 0; i < MAX_VARINT_BYTES - 1; i++) {
+  static inline void encode_varint_raw_loop(uint8_t *__restrict__ &pos PROTO_ENCODE_DEBUG_PARAM, T value) {
+    do {
      PROTO_ENCODE_CHECK_BOUNDS(pos, 1);
-      *pos++ = static_cast<uint8_t>(value) | 0x80;
+      *pos++ = static_cast<uint8_t>(value | 0x80);
      value >>= 7;
-      if (value <= 0x7F) {
-        PROTO_ENCODE_CHECK_BOUNDS(pos, 1);
-        *pos++ = static_cast<uint8_t>(value);
-        return pos;
-      }
-    }
+    } while (value > 0x7F);
    PROTO_ENCODE_CHECK_BOUNDS(pos, 1);
    *pos++ = static_cast<uint8_t>(value);
-    return pos;
  }
  static inline void ESPHOME_ALWAYS_INLINE encode_varint_raw(uint8_t *__restrict__ &pos PROTO_ENCODE_DEBUG_PARAM,
                                                             uint32_t value) {
@@ -333,7 +315,7 @@ class ProtoEncode {
      *pos++ = static_cast<uint8_t>(value);
      return;
    }
-    pos = encode_varint_raw_loop(pos PROTO_ENCODE_DEBUG_ARG, value);
+    encode_varint_raw_loop(pos PROTO_ENCODE_DEBUG_ARG, value);
  }
  /// Encode a varint that is expected to be 1-2 bytes (e.g. zigzag RSSI, small lengths).
  static inline void ESPHOME_ALWAYS_INLINE encode_varint_raw_short(uint8_t *__restrict__ &pos PROTO_ENCODE_DEBUG_PARAM,
@@ -349,7 +331,7 @@ class ProtoEncode {
      *pos++ = static_cast<uint8_t>(value >> 7);
      return;
    }
-    pos = encode_varint_raw_loop(pos PROTO_ENCODE_DEBUG_ARG, value);
+    encode_varint_raw_loop(pos PROTO_ENCODE_DEBUG_ARG, value);
  }
  static inline void ESPHOME_ALWAYS_INLINE encode_varint_raw_64(uint8_t *__restrict__ &pos PROTO_ENCODE_DEBUG_PARAM,
                                                                uint64_t value) {
@@ -358,7 +340,7 @@ class ProtoEncode {
      *pos++ = static_cast<uint8_t>(value);
      return;
    }
-    pos = encode_varint_raw_loop(pos PROTO_ENCODE_DEBUG_ARG, value);
+    encode_varint_raw_loop(pos PROTO_ENCODE_DEBUG_ARG, value);
  }
  static inline void ESPHOME_ALWAYS_INLINE encode_field_raw(uint8_t *__restrict__ &pos PROTO_ENCODE_DEBUG_PARAM,
                                                            uint32_t field_id, uint32_t type) {
@@ -420,7 +402,7 @@ class ProtoEncode {
      PROTO_ENCODE_CHECK_BOUNDS(pos, 1 + len);
      *pos++ = static_cast<uint8_t>(len);
    } else {
-      pos = encode_varint_raw_loop(pos PROTO_ENCODE_DEBUG_ARG, len);
+      encode_varint_raw_loop(pos PROTO_ENCODE_DEBUG_ARG, len);
      PROTO_ENCODE_CHECK_BOUNDS(pos, len);
    }
    std::memcpy(pos, string, len);
@@ -102,8 +102,34 @@ CC1101Component::CC1101Component() {
  memset(this->pa_table_, 0, sizeof(this->pa_table_));
 }

+void IRAM_ATTR CC1101Component::gpio_intr(CC1101Component *arg) { arg->enable_loop_soon_any_context(); }
+
 void CC1101Component::setup() {
  this->spi_setup();
+
+  if (this->gdo0_pin_ != nullptr) {
+    this->gdo0_pin_->setup();
+  }
+
+  this->configure();
+  if (this->is_failed()) {
+    return;
+  }
+
+  // Defer pin mode setup until after all components have completed setup()
+  // This handles the case where remote_transmitter runs after CC1101 and changes pin mode
+  if (this->gdo0_pin_ != nullptr) {
+    this->defer([this]() {
+      this->gdo0_pin_->pin_mode(gpio::FLAG_INPUT);
+      if (this->state_.PKT_FORMAT == static_cast<uint8_t>(PacketFormat::PACKET_FORMAT_FIFO)) {
+        this->gdo0_pin_->attach_interrupt(&CC1101Component::gpio_intr, this, gpio::INTERRUPT_RISING_EDGE);
+      }
+    });
+  }
+}
+
+void CC1101Component::configure() {
+  // Manual reset sequence per CC1101 datasheet section 19.1.2
  this->cs_->digital_write(true);
  delayMicroseconds(1);
  this->cs_->digital_write(false);
@@ -126,11 +152,6 @@ void CC1101Component::setup() {
    return;
  }

-  // Setup GDO0 pin if configured
-  if (this->gdo0_pin_ != nullptr) {
-    this->gdo0_pin_->setup();
-  }
-
  this->initialized_ = true;

  for (uint8_t i = 0; i <= static_cast<uint8_t>(Register::TEST0); i++) {
@@ -140,20 +161,11 @@ void CC1101Component::setup() {
    this->write_(static_cast<Register>(i));
  }
  this->set_output_power(this->output_power_requested_);
+
  if (!this->enter_rx_()) {
    this->mark_failed();
    return;
  }
-
-  // Defer pin mode setup until after all components have completed setup()
-  // This handles the case where remote_transmitter runs after CC1101 and changes pin mode
-  if (this->gdo0_pin_ != nullptr) {
-    this->defer([this]() { this->gdo0_pin_->pin_mode(gpio::FLAG_INPUT); });
-  }
-
-  if (this->state_.PKT_FORMAT != static_cast<uint8_t>(PacketFormat::PACKET_FORMAT_FIFO)) {
-    this->disable_loop();
-  }
 }

 void CC1101Component::call_listeners_(const std::vector<uint8_t> &packet, float freq_offset, float rssi, uint8_t lqi) {
@@ -164,6 +176,7 @@ void CC1101Component::call_listeners_(const std::vector<uint8_t> &packet, float
 }

 void CC1101Component::loop() {
+  this->disable_loop();
  if (this->state_.PKT_FORMAT != static_cast<uint8_t>(PacketFormat::PACKET_FORMAT_FIFO) || this->gdo0_pin_ == nullptr ||
      !this->gdo0_pin_->digital_read()) {
    return;
@@ -244,6 +257,7 @@ void CC1101Component::begin_tx() {
  this->write_(Register::PKTCTRL0, 0x32);
  ESP_LOGV(TAG, "Beginning TX sequence");
  if (this->gdo0_pin_ != nullptr) {
+    this->gdo0_pin_->detach_interrupt();
    this->gdo0_pin_->pin_mode(gpio::FLAG_OUTPUT);
  }
  // Transition through IDLE to bypass CCA (Clear Channel Assessment) which can
@@ -268,7 +282,7 @@ void CC1101Component::begin_rx() {

 void CC1101Component::reset() {
  this->strobe_(Command::RES);
-  this->setup();
+  this->configure();
 }

 void CC1101Component::set_idle() {
@@ -673,10 +687,12 @@ void CC1101Component::set_packet_mode(bool value) {
    this->state_.GDO0_CFG = 0x0D;
  }
  if (this->initialized_) {
-    if (value) {
-      this->enable_loop();
-    } else {
-      this->disable_loop();
+    if (this->gdo0_pin_ != nullptr) {
+      if (value) {
+        this->gdo0_pin_->attach_interrupt(&CC1101Component::gpio_intr, this, gpio::INTERRUPT_RISING_EDGE);
+      } else {
+        this->gdo0_pin_->detach_interrupt();
+      }
    }
    this->write_(Register::PKTCTRL0);
    this->write_(Register::PKTCTRL1);
@@ -25,6 +25,7 @@ class CC1101Component : public Component,
  void setup() override;
  void loop() override;
  void dump_config() override;
+  void configure();

  // Actions
  void begin_tx();
@@ -93,6 +94,7 @@ class CC1101Component : public Component,

  // GDO pin for packet reception
  InternalGPIOPin *gdo0_pin_{nullptr};
+  static void IRAM_ATTR gpio_intr(CC1101Component *arg);

  // Packet handling
  void call_listeners_(const std::vector<uint8_t> &packet, float freq_offset, float rssi, uint8_t lqi);
@@ -43,3 +43,11 @@ wave_4_26.extend(
        },
    },
 )
+
+
+ssd1677.extend(
+    "waveshare-3.97in",
+    width=800,
+    height=480,
+    mirror_x=True,
+)
@@ -451,6 +451,8 @@ async def to_code(config):
    ota.request_ota_state_listeners()

    esp32.add_idf_component(name="espressif/esp-tflite-micro", ref="1.3.3~1")
+    # Pin esp-nn for stable future builds (esp-tflite-micro depends on esp-nn)
+    esp32.add_idf_component(name="espressif/esp-nn", ref="1.2.1")

    cg.add_build_flag("-DTF_LITE_STATIC_MEMORY")
    cg.add_build_flag("-DTF_LITE_DISABLE_X86_NEON")
@@ -29,14 +29,6 @@ void VADModel::log_model_config() {
 bool StreamingModel::load_model_() {
  RAMAllocator<uint8_t> arena_allocator;

-  if (this->tensor_arena_ == nullptr) {
-    this->tensor_arena_ = arena_allocator.allocate(this->tensor_arena_size_);
-    if (this->tensor_arena_ == nullptr) {
-      ESP_LOGE(TAG, "Could not allocate the streaming model's tensor arena.");
-      return false;
-    }
-  }
-
  if (this->var_arena_ == nullptr) {
    this->var_arena_ = arena_allocator.allocate(STREAMING_MODEL_VARIABLE_ARENA_SIZE);
    if (this->var_arena_ == nullptr) {
@@ -53,6 +45,26 @@ bool StreamingModel::load_model_() {
    return false;
  }

+  // Probe for the actual required tensor arena size if not yet determined
+  if (!this->tensor_arena_size_probed_) {
+    size_t probed_size = this->probe_arena_size_();
+    if (probed_size > 0) {
+      ESP_LOGD(TAG, "Probed tensor arena size: %zu bytes", probed_size);
+      this->tensor_arena_size_ = probed_size;
+    } else {
+      ESP_LOGW(TAG, "Arena size probe failed, using manifest size: %zu bytes", this->tensor_arena_size_);
+    }
+    this->tensor_arena_size_probed_ = true;
+  }
+
+  if (this->tensor_arena_ == nullptr) {
+    this->tensor_arena_ = arena_allocator.allocate(this->tensor_arena_size_);
+    if (this->tensor_arena_ == nullptr) {
+      ESP_LOGE(TAG, "Could not allocate the streaming model's tensor arena.");
+      return false;
+    }
+  }
+
  if (this->interpreter_ == nullptr) {
    this->interpreter_ =
        make_unique<tflite::MicroInterpreter>(tflite::GetModel(this->model_start_), this->streaming_op_resolver_,
@@ -94,6 +106,70 @@ bool StreamingModel::load_model_() {
  return true;
 }

+size_t StreamingModel::probe_arena_size_() {
+  RAMAllocator<uint8_t> arena_allocator;
+
+  // Try with the manifest size first, then escalates to 1.5, then 2x if it fails. Different platforms and different
+  // versions of the esp-nn library require different amounts of memory, so the manifest size may not always be correct,
+  // and probing allows us to find the actual required size for the current build and platform. Aligns test sizes to 16
+  // bytes.
+  size_t attempt_sizes[] = {(this->tensor_arena_size_ + 15) & ~15, (this->tensor_arena_size_ * 3 / 2 + 15) & ~15,
+                            (this->tensor_arena_size_ * 2 + 15) & ~15};
+
+  for (size_t attempt_size : attempt_sizes) {
+    uint8_t *probe_arena = arena_allocator.allocate(attempt_size);
+    if (probe_arena == nullptr) {
+      continue;
+    }
+
+    // Verify the model works at all with this arena size
+    auto probe_interpreter = make_unique<tflite::MicroInterpreter>(
+        tflite::GetModel(this->model_start_), this->streaming_op_resolver_, probe_arena, attempt_size, this->mrv_);
+
+    if (probe_interpreter->AllocateTensors() != kTfLiteOk) {
+      probe_interpreter.reset();
+      arena_allocator.deallocate(probe_arena, attempt_size);
+      this->ma_ = tflite::MicroAllocator::Create(this->var_arena_, STREAMING_MODEL_VARIABLE_ARENA_SIZE);
+      this->mrv_ = tflite::MicroResourceVariables::Create(this->ma_, 20);
+      continue;
+    }
+
+    // Try to shrink the arena. Start with arena_used_bytes() + 16 (rounded to 16-byte alignment).
+    // If that works, use it. Otherwise, try midpoints between that and the full size until one succeeds.
+    size_t lower = (probe_interpreter->arena_used_bytes() + 16 + 15) & ~15;
+    probe_interpreter.reset();
+    this->ma_ = tflite::MicroAllocator::Create(this->var_arena_, STREAMING_MODEL_VARIABLE_ARENA_SIZE);
+    this->mrv_ = tflite::MicroResourceVariables::Create(this->ma_, 20);
+
+    size_t upper = attempt_size;
+
+    while (lower < upper) {
+      auto test_interpreter = make_unique<tflite::MicroInterpreter>(
+          tflite::GetModel(this->model_start_), this->streaming_op_resolver_, probe_arena, lower, this->mrv_);
+
+      bool ok = test_interpreter->AllocateTensors() == kTfLiteOk;
+
+      test_interpreter.reset();
+      this->ma_ = tflite::MicroAllocator::Create(this->var_arena_, STREAMING_MODEL_VARIABLE_ARENA_SIZE);
+      this->mrv_ = tflite::MicroResourceVariables::Create(this->ma_, 20);
+
+      if (ok) {
+        // Found a working size smaller than the full arena
+        upper = lower + 16;  // Pad by 16 bytes to be safe for future allocations
+        break;
+      }
+
+      // Try the midpoint between current attempt and full size
+      lower = ((lower + upper) / 2 + 15) & ~15;
+    }
+
+    arena_allocator.deallocate(probe_arena, attempt_size);
+    return upper;
+  }
+
+  return 0;
+}
+
 void StreamingModel::unload_model() {
  this->interpreter_.reset();

@@ -63,6 +63,10 @@ class StreamingModel {
  /// @brief Allocates tensor and variable arenas and sets up the model interpreter
  /// @return True if successful, false otherwise
  bool load_model_();
+  /// @brief Probes the actual required tensor arena size by trial allocation.
+  /// Tries the manifest size first, then 2x if that fails.
+  /// @return The required arena size rounded up to 16-byte alignment, or 0 on failure.
+  size_t probe_arena_size_();
  /// @brief Returns true if successfully registered the streaming model's TensorFlow operations
  bool register_streaming_ops_(tflite::MicroMutableOpResolver<20> &op_resolver);

@@ -70,6 +74,7 @@ class StreamingModel {

  bool loaded_{false};
  bool enabled_{true};
+  bool tensor_arena_size_probed_{false};
  bool unprocessed_probability_status_{false};
  uint8_t current_stride_step_{0};
  int16_t ignore_windows_{-MIN_SLICES_BEFORE_DETECTION};
@@ -223,6 +223,10 @@ async def to_code(config):
        for symbol in ("vprintf", "printf", "fprintf"):
            cg.add_build_flag(f"-Wl,--wrap={symbol}")

+    # Wrap Arduino's millis() so all callers use our fast accumulator instead of
+    # the expensive time_us_64() + micros_to_millis() 64-bit conversion path.
+    cg.add_build_flag("-Wl,--wrap=millis")
+
    cg.add_platformio_option("board_build.core", "earlephilhower")
    # In testing mode, use all flash for sketch to allow linking grouped component tests.
    # Real RP2040 hardware uses 1MB filesystem + 1MB sketch, but CI tests may combine
@@ -14,8 +14,51 @@
 namespace esphome {

 void HOT yield() { ::yield(); }
+// Arduino-pico's millis() uses time_us_64() (64-bit hardware timer read) then
+// micros_to_millis() which does 64-bit multiply-shift conversion on an ARM
+// Cortex-M0+ (~789 ns/call measured). Replace with a simple accumulator that
+// tracks a running millis counter from 32-bit micros() deltas using pure
+// 32-bit ops. ::micros() on RP2040 is a fast 32-bit hardware read (~220 ns).
+//
+// Overflow safety: ::micros() wraps every ~71.6 minutes. Unsigned subtraction
+// handles one wrap correctly. ESPHome calls millis() thousands of times per
+// second, so missing a full wrap is not a realistic concern. At boot, both
+// s_last_us and ::micros() start at 0, so no special initialization needed.
+//
+// Also installed as __wrap_millis (via -Wl,--wrap=millis) so Arduino library
+// code calling ::millis() directly gets the fast version. Interrupts are
+// briefly disabled to protect the static state — Wiegand and ZyAura call
+// millis() from ISR handlers on all platforms including RP2040.
+static constexpr uint32_t MILLIS_RARE_PATH_THRESHOLD_US = 10000;
+static constexpr uint32_t US_PER_MS = 1000;
+
+uint32_t HOT millis() {
+  static struct {
+    uint32_t cache;
+    uint32_t remainder;
+    uint32_t last_us;
+  } state = {0, 0, 0};
+  uint32_t ps = save_and_disable_interrupts();
+  uint32_t now_us = ::micros();
+  uint32_t delta = now_us - state.last_us;
+  state.last_us = now_us;
+  state.remainder += delta;
+  if (state.remainder >= MILLIS_RARE_PATH_THRESHOLD_US) {
+    uint32_t ms = state.remainder / US_PER_MS;
+    state.cache += ms;
+    state.remainder -= ms * US_PER_MS;
+  } else {
+    while (state.remainder >= US_PER_MS) {
+      state.cache++;
+      state.remainder -= US_PER_MS;
+    }
+  }
+  uint32_t result = state.cache;
+  restore_interrupts(ps);
+  return result;
+}
+// millis_64() keeps the full 64-bit timer for precision — called once per loop by Scheduler.
 uint64_t millis_64() { return micros_to_millis<uint64_t>(time_us_64()); }
-uint32_t HOT millis() { return micros_to_millis(time_us_64()); }
 void HOT delay(uint32_t ms) { ::delay(ms); }
 uint32_t HOT micros() { return ::micros(); }
 void HOT delayMicroseconds(uint32_t us) { delay_microseconds_safe(us); }
@@ -42,4 +85,9 @@ uint32_t arch_get_cpu_freq_hz() { return RP2040::f_cpu(); }

 }  // namespace esphome

+// Linker wrap: redirect all ::millis() calls (Arduino libs) to our accumulator.
+// Requires -Wl,--wrap=millis in build flags (added by __init__.py).
+// NOLINTNEXTLINE(bugprone-reserved-identifier,cert-dcl37-c,cert-dcl51-cpp,readability-identifier-naming)
+extern "C" uint32_t __wrap_millis() { return esphome::millis(); }
+
 #endif  // USE_RP2040
@@ -9,7 +9,7 @@
 #include <WiFi.h>
 #include <pico/cyw43_arch.h>  // For cyw43_arch_lwip_begin/end (LwIPLock)
 #elif defined(USE_ETHERNET)
-#include <LwipEthernet.h>  // For ethernet_arch_lwip_begin/end (LwIPLock)
+#include <lwip_wrap.h>  // For LWIPMutex — LwIPLock mirrors its semantics (see below)
 #include "esphome/components/ethernet/ethernet_component.h"
 #endif
 #include <hardware/structs/rosc.h>
@@ -43,9 +43,18 @@ IRAM_ATTR InterruptLock::~InterruptLock() { restore_interrupts(state_); }
 // main loop, corrupting the shared rx_buf_ pbuf chain (use-after-free, pbuf_cat
 // assertion failures). See esphome#10681.
 //
-// WiFi uses cyw43_arch_lwip_begin/end; Ethernet uses ethernet_arch_lwip_begin/end.
-// Both acquire the async_context recursive mutex to prevent IRQ callbacks from
-// firing during critical sections.
+// WiFi uses cyw43_arch_lwip_begin/end.
+//
+// For wired Ethernet, taking only the async_context lock is NOT enough. The
+// W5500 GPIO IRQ path (LwipIntfDev::_irq) checks arduino-pico's `__inLWIP`
+// counter to decide whether to defer packet processing. If we hold the
+// async_context lock without bumping `__inLWIP`, an interrupt-driven packet
+// arrival re-enters lwIP from IRQ context and corrupts pbufs (the `pbuf_cat`
+// assertion crash on wiznet-w5500-evb-pico). We mirror arduino-pico's
+// LWIPMutex (cores/rp2040/lwip_wrap.h) exactly: bump `__inLWIP`, take the
+// lock, and on release re-unmask any GPIO IRQs that were deferred while we
+// held it. We can't `using LwIPLock = LWIPMutex;` in helpers.h because
+// pulling lwip_wrap.h there poisons many TUs with lwIP types.
 //
 // When neither WiFi nor Ethernet is configured, this is a no-op since
 // there's no network stack and no lwip callbacks to race with.
@@ -53,8 +62,18 @@ IRAM_ATTR InterruptLock::~InterruptLock() { restore_interrupts(state_); }
 LwIPLock::LwIPLock() { cyw43_arch_lwip_begin(); }
 LwIPLock::~LwIPLock() { cyw43_arch_lwip_end(); }
 #elif defined(USE_ETHERNET)
-LwIPLock::LwIPLock() { ethernet_arch_lwip_begin(); }
-LwIPLock::~LwIPLock() { ethernet_arch_lwip_end(); }
+LwIPLock::LwIPLock() {
+  __inLWIP++;
+  ethernet_arch_lwip_begin();
+}
+LwIPLock::~LwIPLock() {
+  ethernet_arch_lwip_end();
+  __inLWIP--;
+  if (__needsIRQEN && !__inLWIP) {
+    __needsIRQEN = false;
+    ethernet_arch_lwip_gpio_unmask();
+  }
+}
 #else
 LwIPLock::LwIPLock() {}
 LwIPLock::~LwIPLock() {}
@@ -104,11 +104,17 @@ void SX126x::write_register_(uint16_t reg, uint8_t *data, uint8_t size) {
  delayMicroseconds(SWITCHING_DELAY_US);
 }

+void IRAM_ATTR SX126x::gpio_intr(SX126x *arg) { arg->enable_loop_soon_any_context(); }
+
 void SX126x::setup() {
  // setup pins
  this->busy_pin_->setup();
  this->rst_pin_->setup();
  this->dio1_pin_->setup();
+  if (this->dio1_pin_->is_internal()) {
+    static_cast<InternalGPIOPin *>(this->dio1_pin_)
+        ->attach_interrupt(&SX126x::gpio_intr, this, gpio::INTERRUPT_RISING_EDGE);
+  }

  // start spi
  this->spi_setup();
@@ -348,6 +354,9 @@ void SX126x::call_listeners_(const std::vector<uint8_t> &packet, float rssi, flo
 }

 void SX126x::loop() {
+  if (this->dio1_pin_->is_internal()) {
+    this->disable_loop();
+  }
  if (!this->dio1_pin_->digital_read()) {
    return;
  }
@@ -3,6 +3,7 @@
 #include "esphome/components/spi/spi.h"
 #include "esphome/core/automation.h"
 #include "esphome/core/component.h"
+#include "esphome/core/hal.h"
 #include "sx126x_reg.h"
 #include <utility>
 #include <vector>
@@ -100,6 +101,7 @@ class SX126x : public Component,
  Trigger<std::vector<uint8_t>, float, float> *get_packet_trigger() { return &this->packet_trigger_; }

 protected:
+  static void IRAM_ATTR gpio_intr(SX126x *arg);
  void configure_fsk_ook_();
  void configure_lora_();
  void set_packet_params_(uint8_t payload_length);
@@ -53,6 +53,8 @@ void SX127x::write_fifo_(const std::vector<uint8_t> &packet) {
  this->disable();
 }

+void IRAM_ATTR SX127x::gpio_intr(SX127x *arg) { arg->enable_loop_soon_any_context(); }
+
 void SX127x::setup() {
  // setup reset
  this->rst_pin_->setup();
@@ -60,6 +62,7 @@ void SX127x::setup() {
  // setup dio0
  if (this->dio0_pin_) {
    this->dio0_pin_->setup();
+    this->dio0_pin_->attach_interrupt(&SX127x::gpio_intr, this, gpio::INTERRUPT_RISING_EDGE);
  }

  // start spi
@@ -313,6 +316,7 @@ void SX127x::call_listeners_(const std::vector<uint8_t> &packet, float rssi, flo
 }

 void SX127x::loop() {
+  this->disable_loop();
  if (this->dio0_pin_ == nullptr || !this->dio0_pin_->digital_read()) {
    return;
  }
@@ -386,11 +390,6 @@ void SX127x::set_mode_(uint8_t modulation, uint8_t mode) {
      return;
    }
  }
-  if (mode == MODE_RX && (modulation == MOD_LORA || this->packet_mode_)) {
-    this->enable_loop();
-  } else {
-    this->disable_loop();
-  }
 }

 void SX127x::set_mode_rx() {
@@ -4,6 +4,7 @@
 #include "esphome/components/spi/spi.h"
 #include "esphome/core/automation.h"
 #include "esphome/core/component.h"
+#include "esphome/core/hal.h"
 #include <vector>

 namespace esphome {
@@ -86,6 +87,7 @@ class SX127x : public Component,
  Trigger<std::vector<uint8_t>, float, float> *get_packet_trigger() { return &this->packet_trigger_; }

 protected:
+  static void IRAM_ATTR gpio_intr(SX127x *arg);
  void configure_fsk_ook_();
  void configure_lora_();
  void set_mode_(uint8_t modulation, uint8_t mode);
@@ -12,14 +12,14 @@ platformio==6.1.19
 esptool==5.2.0
 click==8.3.2
 esphome-dashboard==20260408.1
-aioesphomeapi==44.13.1
+aioesphomeapi==44.13.3
 zeroconf==0.148.0
 puremagic==1.30
 ruamel.yaml==0.19.1 # dashboard_import
 ruamel.yaml.clib==0.2.15 # dashboard_import
 esphome-glyphsets==0.2.0
 pillow==12.2.0
-resvg-py==0.2.6
+resvg-py==0.3.1
 freetype-py==2.5.1
 jinja2==3.1.6
 bleak==2.1.1
Author	SHA1	Message	Date
J. Nick Koston	feb734c8b7	[rp2040] Add interrupt guard to millis() accumulator Wiegand and ZyAura call millis() from ISR handlers on all platforms including RP2040 (no platform restrictions on either component). The accumulator's mutable statics need protection from concurrent ISR access. Uses save_and_disable_interrupts()/restore_interrupts() — same pattern as InterruptLock in rp2040/helpers.cpp.	2026-04-12 09:11:41 -10:00
J. Nick Koston	12d227dee8	[rp2040] Bound the while loop in millis() accumulator Same approach as ESP8266: split into common path (while loop, ≤10 iterations) and rare path (constant-time /1000 multiply for gaps >10 ms). RP2040 has no ISR callers so no interrupt guard needed, but the bounded loop avoids worst-case latency if millis() is called from a context that was blocked for a long time.	2026-04-11 23:22:12 -10:00
J. Nick Koston	85c51967f5	[rp2040] Merge accumulator into millis(), struct statics, cleanup Merge millis_accumulator() into millis() and pack statics into a struct for smaller code. Remove unused USE_FAST_MILLIS_ACCUMULATOR define. Add NOLINTNEXTLINE for __wrap_millis reserved identifier.	2026-04-11 23:17:12 -10:00
J. Nick Koston	31aa00f1d5	[rp2040] Add USE_FAST_MILLIS_ACCUMULATOR define for benchmark guard	2026-04-11 22:38:29 -10:00
J. Nick Koston	a50a24aed3	[rp2040] Fix clang-format: drop trailing underscore from static function	2026-04-11 22:22:32 -10:00
J. Nick Koston	357835cc20	[rp2040] Replace millis() with fast accumulator, wrap Arduino callers Arduino-pico's millis() uses time_us_64() (64-bit hardware timer read) then micros_to_millis() for 64-bit multiply-shift conversion on ARM Cortex-M0+. Benchmarked at ~789 ns/call. Replace with a simple accumulator that tracks a running millis counter from 32-bit ::micros() deltas (220 ns on RP2040) using pure 32-bit integer ops. No 64-bit math needed. Use -Wl,--wrap=millis to intercept all ::millis() calls so Arduino libraries also get the fast version. millis_64() is left on time_us_64() for full 64-bit precision — it is only called once per loop by the Scheduler. Overflow safety: ::micros() wraps every ~71.6 minutes. Unsigned 32-bit delta arithmetic handles one wrap correctly. ESPHome calls millis() thousands of times per second, so missing a full wrap is not a realistic concern. Benchmarked on real RP2040 hardware: Before: 789 ns/call (time_us_64 + micros_to_millis) After: ~270 ns/call (accumulator, estimated)	2026-04-11 22:20:36 -10:00
Jonathan Swoboda	bef4c8a86c	[cc1101] Extract chip configuration into configure() method (#15635 )	2026-04-11 17:36:27 -04:00
Farmer-shin	6e67864510	[epaper_spi] Add Waveshare 3.97inch E-Paper Display (#15466 )	2026-04-11 21:27:25 +10:00
dependabot[bot]	c2af4874f9	Bump aioesphomeapi from 44.13.2 to 44.13.3 (#15641 ) Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>	2026-04-11 08:58:20 +00:00
dependabot[bot]	2001b91280	Bump resvg-py from 0.3.0 to 0.3.1 (#15640 ) Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>	2026-04-11 08:57:39 +00:00
dependabot[bot]	5460ee7edd	Bump aioesphomeapi from 44.13.1 to 44.13.2 (#15637 ) Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>	2026-04-10 15:55:15 -10:00
J. Nick Koston	40081e5ae7	[rp2040] Fix W5500 Ethernet pbuf corruption by mirroring LWIPMutex semantics (#15624 ) Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>	2026-04-10 13:13:05 -10:00
Jonathan Swoboda	a7c5b0ab46	[sx127x][cc1101][sx126x] Use GPIO interrupt to wake loop (#15627 )	2026-04-10 16:26:09 -04:00
dependabot[bot]	e1a813e11f	Bump peter-evans/create-pull-request from 8.1.0 to 8.1.1 (#15630 ) Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>	2026-04-10 10:21:01 -10:00
dependabot[bot]	1dfeef0265	Bump actions/github-script from 8.0.0 to 9.0.0 (#15632 ) Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>	2026-04-10 10:20:43 -10:00
dependabot[bot]	395610c117	Bump docker/build-push-action from 7.0.0 to 7.1.0 in /.github/actions/build-image (#15633 ) Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>	2026-04-10 10:20:17 -10:00
dependabot[bot]	ae96f82b82	Bump actions/upload-artifact from 7.0.0 to 7.0.1 (#15631 ) Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>	2026-04-10 10:20:04 -10:00
dependabot[bot]	2c610abcd0	Bump resvg-py from 0.2.6 to 0.3.0 (#15629 ) Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>	2026-04-10 10:19:52 -10:00
Kevin Ahrendt	d3591c8d9e	[micro_wake_word] Pin esp-nn version (#15628 )	2026-04-10 15:21:26 -04:00