Fix several issues in the ESP32-S3 CAM driver:
- stop_capture: reset DMA channel, CAM module and AFIFO under
spinlock to fully quiesce hardware before returning. Clear
pending VSYNC interrupt to prevent stale ISR firing.
- uninit: reset CAM/AFIFO before releasing DMA to prevent
in-flight transfers after channel detach. Use esp_teardown_irq
with correct peripheral ID (ESP32S3_PERIPH_LCD_CAM) instead of
irq_detach which corrupts the shared IRQ mapping table. Mask
interrupts and clear pending flags under spinlock before
detaching handler.
- uninit: preserve XCLK output so the sensor remains accessible
via I2C for subsequent re-initialization.
- set_buf/uninit: track driver-allocated vs user-provided frame
buffers with fb_allocated flag to prevent double-free when
using V4L2 USERPTR mode.
Signed-off-by: wangjianyu3 <wangjianyu3@xiaomi.com>
Protect the common Espressif Wi-Fi netdev RX queue with a spinlock. The Wi-Fi RX callback enqueues packets while the netdev upper-half RX thread dequeues them, and the IOB queue helpers are not internally serialized.
Under sustained receive traffic that race can corrupt the queue state, strand IOBs, and degrade throughput over time as reported in issue #16915.
Initialize the lock during device setup and use it when clearing the queue on ifup, enqueueing received packets, and dequeuing them for the upper half.
Signed-off-by: aviralgarg05 <gargaviral99@gmail.com>
debug.h is a NuttX-specific, non-POSIX header. Placing it in the
top-level include/ directory creates naming conflicts with external
projects that define their own debug.h.
This commit moves the canonical header to include/nuttx/debug.h,
following the NuttX convention for non-POSIX/non-standard headers,
and updates all in-tree references.
A backward-compatibility shim is left at include/debug.h that
emits a deprecation #warning and re-includes <nuttx/debug.h>,
allowing out-of-tree code to continue building while migrating.
Signed-off-by: Piyush Patle <piyushpatle228@gmail.com>
esp_ws2812.h exposes esp_ws2812_setup() with a struct rmt_dev_s * argument
but did not include esp_rmt.h directly.
After guarding the LIRC board registration path with CONFIG_DRIVERS_RC,
non-RC ESP configs no longer pulled in the RMT type indirectly and CI
started failing with an incomplete struct rmt_dev_s declaration.
Include esp_rmt.h in the Xtensa and RISC-V ws2812 headers.
Signed-off-by: Piyush Patle <piyushpatle228@gmail.com>
Add documentation for the RC/LIRC character driver subsystem covering
device registration, the LIRC interface, and usage from user space.
Remove placeholder empty files under drivers/rmt that were left over
from the rmtchar era and are no longer referenced.
Signed-off-by: Piyush Patle <piyushpatle228@gmail.com>
Replace the ESP-specific rmtchar upper-half with arch-local esp_lirc
adapters for Xtensa and RISC-V.
This moves the RMT upper-half out of drivers/rmt, registers LIRC
devices from the ESP board bring-up paths, and removes the old common
rmtchar driver and headers.
Also update the ESP Kconfig and build wiring to build esp_lirc when
ESP_RMT and DRIVERS_RC are enabled.
Fixes discovered during hardware validation:
- register TX as /dev/lirc1 so RX and TX do not collide
- parse the RX worker thread argument from the correct argv slot
- keep RX devices from advertising TX capability
Signed-off-by: Piyush Patle <piyushpatle228@gmail.com>
The I2S BCLK frequency is calculated as:
bclk = sample_rate * total_slot * data_width
Previously, total_slot was always taken from the static config
(esp_i2s0_config / esp_i2s1_config), which is hardcoded to 2.
This is correct for mono and stereo, but wrong for TDM modes
with more than 2 channels (e.g. 4-channel recording with ES7210).
With 4 channels but total_slot=2, the BCLK is only half of what
it should be, causing the actual sample rate to be half of the
requested rate (e.g. requesting 16 kHz actually samples at 8 kHz).
Replace priv->config->total_slot with MAX(priv->channels, 2) in
all three BCLK calculation sites (i2s_check_mclkfrequency and
i2s_set_clock for both master and slave paths). The MIN of 2
preserves the I2S protocol requirement that at least 2 slots
exist even for mono, while correctly scaling for multi-channel
TDM configurations.
Signed-off-by: wangjianyu3 <wangjianyu3@xiaomi.com>
Fix multiple bugs in the I2S RX receive path on ESP32-S3 with GDMA:
- Set streaming=true in i2s_rxdma_start so completed buffers are not
incorrectly marked AUDIO_APB_FINAL
- Restore up_enable_irq calls for DMA interrupt after esp_dma_load
resets the channel
- Skip i2s_hal_rx_reset/rx_reset_fifo on ESP32-S3 which clears the
TDM slot configuration set by i2s_configure
- Accept 1-16 RX channels on ESP32-S3 (was limited to 2)
- Add getreg32 read-back after DMA start to flush CPU write buffer
- Auto-stop RX channel when no pending buffers remain, preventing
hardware from running without DMA descriptors after stop
- Remove duplicate return OK in i2s_rxdma_start
- Fix i2s_ll_tx_enable_mono_mode to i2s_ll_rx_enable_mono_mode in
i2s_rxchannels
- Fix tx_slot_cfg.std to rx_slot_cfg.std in RX PCM slot config
- Fix DMA direction, peripheral ID, data_width, rx_start sequencing,
and rx_eof_num word count in RX path
- Fix esp32s3_board_i2s.c I2S port index
Signed-off-by: wangjianyu3 <wangjianyu3@xiaomi.com>
- Remove debug polling loop from esp32s3_cam_start_capture() that
busy-waited on DMA status register.
- Fix DMA ISR race: stop DMA before invoking capture callback to
prevent ISR re-triggering while callback processes the buffer.
Check priv->capturing before restarting DMA in ISR.
- Move DMA descriptors from struct to heap allocation, avoiding
stack/BSS alignment issues with cache-line-aligned descriptors.
Signed-off-by: wangjianyu3 <wangjianyu3@xiaomi.com>
This commit updates the common-source integration for Xtensa-based
Espressif devices (ESP32, ESP32-S2, and ESP32-S3). This is part of a larger
common-source update split by architecture for better maintainability.
Major components updated:
- IRQ allocator refactoring with intr_alloc integration
- Common-source drivers (GPIO, RMT, I2C, SPI, UART, etc.)
- Espressif components upgrade to release/master.b-test
- Peripheral drivers (ADC, PWM, LEDC, MCPWM, PCNT, Temperature Sensor, etc.)
- Wireless adapters (Wi-Fi and BLE)
- esp_timer migration to the common-source path for Xtensa devices
- Common-source power management implementation (auto-sleep and wakeup paths)
- Board defconfigs for all Xtensa Espressif boards
- SMP support improvements for ESP32-S3
- Critical section handling improvements
Key architectural changes:
- IRQ Allocator: The new interrupt allocator enables multiple mapping
options from interrupt sources to CPU interrupts, providing flexibility
required by modern peripherals. Although this introduces breaking changes
to the interrupt handling API, the required ARCH_MINIMAL_VECTORTABLE
Kconfig option is explicitly checked during startup to ensure proper
configuration. This validation prevents runtime issues from configuration
mismatches.
- Xtensa-specific interrupt handling via esp_xtensa_intr.c providing
NuttX-native implementations of xt_ints_on/off and interrupt handlers,
avoiding conflicts with NuttX's core Xtensa macros.
- Timer/RTC unification: ESP32/ESP32-S2/ESP32-S3 move from chip-specific
RTC/RT-timer code to common-source Espressif integration, including
esp_timer_adapter/esp_rtc paths and the required bringup/defconfig updates.
- Power management consolidation: Xtensa PM follows the common-source
implementation, including common-source auto-sleep behavior, UART/Wi-Fi
wakeup coordination, and tickless-safe sleep flow compatibility.
Note: This is a large commit to maintain bisectability. Breaking the
changes into smaller commits would result in non-building intermediate
states across the common-source infrastructure update.
Tested configurations:
- All defconfigs were tested, including `ostest`.
Signed-off-by: Tiago Medicci Serrano <tiago.medicci@espressif.com>
Fix 269 occurrences of duplicate "the" word typo found in 209 files
across source code, header files, and configuration.
Signed-off-by: Huang Qi <huangqi3@xiaomi.com>
Add ESP32-S3 CAM controller driver implementing the NuttX imgdata
interface. Supports 8-bit DVP camera input with DMA-based frame
capture, VSYNC interrupt handling, and PSRAM frame buffer allocation.
The driver is sensor-agnostic: resolution and pixel format are
negotiated at runtime through the V4L2 pipeline. XCLK output
(24 MHz) is started during initialization for sensor communication.
Signed-off-by: wangjianyu3 <wangjianyu3@xiaomi.com>
This commits adds a function that can be implemented by any vendor
to allow SoC-specific functions to be called by `up_initialize`.
Please note that `up_initialize` is provided by the arch level, but
it doesn't allow SoC-specific initialization. Although it could be
possible to run such initialization on board-level, semantically
it isn't related to the board, but with the SoC. As an example of
such implementation, some SoCs require RTC subsystem initialization
to occur before the OS is completely initialized.
Signed-off-by: Tiago Medicci Serrano <tiago.medicci@espressif.com>
The esp32_spi_poll_exchange() function previously used a hardcoded
length of sizeof(uint32_t) (4 bytes) for memcpy() operations and
pointer increments, disregarding the actual remaining transfer_size.
When an SPI transaction involved a data length not aligned to 4 bytes
(such as a 1-byte payload commonly used in sub-GHz radio drivers),
the driver forced a 4-byte memory transfer. This caused out-of-bounds
reads and writes, leading to stack smashing and fatal CPU exceptions
(e.g., EXCCAUSE=0000).
This patch resolves the memory corruption vulnerability by:
1. Calculating a precise `chunk` size for each iteration, strictly
bounded by the remaining `transfer_size`.
2. Limiting the memcpy() operations and the subsequent pointer
increments (tp and rp) to this exact chunk size.
3. Zero-initializing the temporary TX register (w_wd) before copying
to clear padding and prevent reading uninitialized memory.
4. Adding a NULL check for the receive pointer (rp) prior to the
memcpy operation to prevent unintended memory writes.
Signed-off-by: Chip L. <chplee@gmail.com>
When using a key that is longer than the block size of the hashing
algorithm used, the key must be hashed before it is used.
Signed-off-by: Vlad Pruteanu <pruteanuvlad1611@yahoo.com>
Intermediate files of make depend like .ddc and .dds may remain
when make is interrupted. Remove them using make distclean.
Signed-off-by: SPRESENSE <41312067+SPRESENSE@users.noreply.github.com>
The benefits of doing this are:
1. It makes the code logic clearer, with different resources protected by different locks.
2. It improves system responsiveness and avoids contention issues caused by acquiring the same large lock.
Signed-off-by: wangzhi16 <wangzhi16@xiaomi.com>
since the judgment for network card selection was changed from IS_UP to
IS_RUNNING, drivers that lack carrier_on need to add the carrier_on
operation; otherwise, network access issues will occur.
Signed-off-by: zhanghongyu <zhanghongyu@xiaomi.com>
INT32_C(x) is currently defined as x ## ll on Xtensa, which
produces a long long int constant. However, int32_t is
defined as long int on this architecture.
This mismatch can break C++ template overload resolution
and causes build failures in downstream projects such as PX4.
Fix the macro by using x ## l so INT32_C expands to the same
underlying type as int32_t.
Signed-off-by: Adwait Godbole <adwaitngodbole@gmail.com>
SPI transmits incorrect bytes when DMA is enabled and `txbuffer` of
`esp32s3_spi_dma_exchange()` is NULL, causing mounted FAT32 SD cards
to fail with -ENODEV during read operations.
Signed-off-by: liu <liu334275@gmail.com>
boot mode is configured
1. Explicitly define the .rtc_reserved section to ensure RTC data
that requires fixed address is correctly placed and does not
conflict with other sections
2. When legacy boot is configured, vecbase must also be set and
clear the BSS section
3. Cache must be disabled during Flash operations, so all
Flash-operation-related functions must be placed in IRAM
(internal RAM). Update the linker script for legacy boot mode
Signed-off-by: nuttxs <zhaoqing.zhang@sony.com>
Extend CONFIG_LIBCXXTOOLCHAIN support to all remaining platforms
in CMake builds by explicitly locating and linking libstdc++.a
from the toolchain. This prevents build failures caused by STL
header/library mismatches where toolchain headers are included
without the corresponding library.
* Adds nuttx_find_toolchain_lib() calls for all platforms.
* Aligns CMake behavior with Make build system.
Signed-off-by: trns1997 <trns1997@gmail.com>
Add validation to ensure allocated stack size does not exceed TLS_MAXSTACK when
CONFIG_TLS_ALIGNED is enabled, and verify proper stack alignment using STACK_ALIGN_MASK
across all architectures. This improves stack safety and prevents potential TLS overflow conditions.
Signed-off-by: hujun5 <hujun5@xiaomi.com>
Rename STACK_ALIGNMENT macro to STACKFRAME_ALIGN throughout the codebase
to provide clearer naming semantics. The new name better reflects the macro's
purpose of frame alignment rather than general stack alignment.
Signed-off-by: hujun5 <hujun5@xiaomi.com>
Move stack alignment and kernel stack macros from architecture-specific internal
headers to public include/nuttx/irq.h. Consolidates duplicate definitions across
17 architecture families, reducing code duplication while enabling common code
to access these core alignment utilities without architecture dependencies.
Signed-off-by: hujun5 <hujun5@xiaomi.com>
Add support to disable signals actions related struct
Co-authored-by: guoshichao <guoshichao@xiaomi.com>
Signed-off-by: Chengdong Wang <wangchengdong@lixiang.com>
After addrenv_switch(), the current running task (this_task) may change due to
deferred work execution. Update all architecture interrupt, syscall, and exit
handlers to re-fetch tcb = this_task() after addrenv_switch(). Ensures scheduler
and context operations use the correct TCB, preventing context corruption and
exceptions across SMP and memory-protected builds.
Signed-off-by: hujun5 <hujun5@xiaomi.com>
The items is not used in up_flush_dcache, and we should use it in
function cache_invalidate_dcache_items.
Signed-off-by: rongyichang <rongyichang@xiaommi.com>
This commit add the option that was missing in the esp32_start.c
to ignore the chip version for boards with old ESP32 silicon tape out.
Signed-off-by: Alan C. Assis <acassis@gmail.com>
Upperhalf supports multiple working modes at the same time,
which is specified by NIC when register
Signed-off-by: zhanghongyu <zhanghongyu@xiaomi.com>
Modifies Kconfig and build tools to support flash encryption
and E-Fuse burning when flash encryption is enabled.
Signed-off-by: Filipe Cavalcanti <filipe.cavalcanti@espressif.com>
The efuse initialization requires critical section to avoid a very
hidden corner case when flash encryption is enabled.
Signed-off-by: Filipe Cavalcanti <filipe.cavalcanti@espressif.com>
