spi_flash_generic_timeout is currently defined in the header file,
which makes it a common symbol between translation units where this
header file is included. It is also defined in
spi_flash_chip_generic.c.
Add an explicit `extern` to prevent multiple definition errors.
Clang warns that the original code wouldn't have any effect:
warning: indirection of non-volatile null pointer will be deleted,
not trap [-Wnull-dereference]
note: consider using __builtin_trap() or qualifying pointer
with 'volatile'
__builtin_trap translates to 'break 1, 15' instruction on Xtensa,
which might be okay in this case. However to absolutely certainly not
break anything for GCC builds, add 'volatile' instead.
Toolchain CMake files for clang currently pass -ffreestanding option
to prevent clang from picking the wrong copy of stdint.h. This is a
temporary hack until we fix clang distributions to not include the
GCC version of stdint.h. This hack, however, results in setting
__STDC_HOSTED__=0, which is being checked by cbor header files,
making some required functions unavailable as a result.
Undefine __STDC_HOSTED__ as a workaround. This flag is only passed
when compiling cbor itself.
We will likely need to check clang version as well, eventually.
However at the moment clang builds change too often for such check
to be maintainable. Disable it for clang.
This feature was added when we thought that target components might be
added out of IDF tree. That never became possible, and with recent
changes in IDF architecture it is unlikely that this feature will be
necessary any time soon. Remove it, simplifying the code.
This is an experimental feature intended at the moment for ESP-IDF
developers only.
If IDF_TOOLCHAIN=clang environment variable or CMake variable is set,
use toolchain-clang-esp32xx.cmake instead of toolchain-esp32xx.cmake.
These changes aren't sufficient to actually build any IDF project with
clang; subsequent commits add a few workarounds required to do this.
Toolchain files are added for esp32 and esp32s2, which are the targets
supported in our llvm-project fork at the moment.
Since these functions receive the pointer to reent structure, they
should set errno in it rather than using thread-local errno.
This didn't cause practical issues because console functions in IDF
are only called from threads, and in that case 'r' is a pointer to
the thread-local structure, so &errno is the same thing as
&__errno_r(r). Still, fixing this for consistency.
When CONFIG_VFS_SUPPORT_IO is disabled, _read_r and _write_r
implementations in syscalls.c are used to provide console I/O via
esp_rom_uart_tx_one_char/esp_rom_uart_rx_one_char.
When newlib opens a (FILE*) stream, it calls fstat to check if the
underlying file is character-oriented. In this case, it configures the
stream to use line buffering. Otherwise (or if fstat fails) the stream
is opened as block buffered.
Since fstat wasn't provided, stdin/stdout/stderr streams got opened in
block buffered mode. For console, we need line buffered output so that
the stream buffer is flushed each time a complete line (ending with
'\n') is sent to stdout or stderr.
Fix by implementing _fstat_r stub, setting st->st_mdoe=S_IFCHR.
If a partition offset is not aligned to 4 kB, erase operations will
fail, even if they are aligned to 4 kB relative to the partition
start. This is because the underlying esp_flash_erase_range function
only works when the address is aligned to 4 kB.
Not making this an error for now, since applications might be using
read-only non-4kB aligned partitions, which still work fine.
Will change this behavior in IDF 5.0, requiring 4 kB alignment for all
partitions.
Closes https://github.com/espressif/esp-idf/issues/7295
Closes https://github.com/espressif/esp-idf/issues/7350
...only. Previously this check was done for both app and data
partitions. While checking that data partition size is aligned is
a _good thing_, the check would only occur when secure boot was on.
These files were added into the component directory back when we
didn't have the documentation directory, and weren't moved once the
documentation directory was created.
The code checked CONFIG_ETH_USE_SPI_ETHERNET (which is usually set),
but CONFIG_EXAMPLE_ETH_SPI_xxx_GPIO options are only defined if
CONFIG_EXAMPLE_USE_SPI_ETHERNET is set. Fix the ifdef accordingly.
Regression from aea901f0.
* fix failing tests: expect s2 binutils to be installed on s3,
expect risc-v toolchain to be installed on s2 and s3.
* improve readability by adding assert_tool_installed and
assert_tool_not_installed functions, instead of using an optional
assertIn argument
* rename "RISC" to "RISCV_ELF" as ULP is arguably very much "R" as in
"RISC"!
CMake sorts result of file(GLOB) command since version 3.6.0:
https://gitlab.kitware.com/cmake/cmake/-/commit/edcccde7d
Since ESP-IDF sets cmake_minimum_required version to 3.5, and version
3.5.1 is used in CI, sort file lists obtained from file(GLOB)
manually.
This helps obtain reproducible order of libraries passed to the linker
and to ldgen.
In practice, calling memcpy with NULL buffer and 0 size works on the
ESP32, but the C standard considers this an undefined behavior.
When building with UBSAN checks enabled, compiler will insert a check
that memcpy argument is non-NULL, regardless of the size argument.
This caused a failure in tools/test_apps/system/panic, which is built
with USBAN enabled for several components.
Before newlib 3.3.0, <dirent.h> bundled in newlib did not include any
function declarations. Instead, the file included the platform-
specific <sys/dirent.h>. This inclusion was inside a C++ guard block.
ESP-IDF provided sys/dirent.h inside newlib component, and this file
contained all the necessary function and structure declarations.
Since https://github.com/espressif/newlib-esp32/commit/da418955f5dca2f8570561c5dd6e79e25ebaeba2,
common function declarations have been added to <dirent.h> in newlib.
However, the inclusion of sys/dirent.h has been moved out of the C++
guard block. However we didn't notice this change and did not update
sys/dirent.h in ESP-IDF newlib component to and the now-required
C++ guards there.
This commit adds the missing C++ guards to the platform-specific
sys/dirent.h.
The declarations of common dirent.h functions are now present both in
<dirent.h> (provided by newlib) and in sys/dirent.h (provided by IDF).
We keep the declarations in sys/dirent.h for compatibility, since some
ESP-IDF files and applications may include <sys/dirent.h> directly,
rather than <dirent.h>.
Closes https://github.com/espressif/esp-idf/issues/7204
FreeRTOS timer tests use quite a lot of stack, close to the limit on
the ESP32. ESP32-S3 has larger coprocessor save area, so there is less
space left for the actual stack, and the tests fail with a stack
overflow. Increase the stack size to make the tests work reliably.
From LX7_ESP32_S3_MP_linux_redist.tgz with SHA256:
63f305a982b2ee94cc78e5c20e3e3eb8bf0edeeaf703af0227a418bc34f7b848
copied from xtensa-elf/arch/include/xtensa/config/ with changes:
1) pre-commit fixes applied
2) re-added CONFIGID0 and CONFIGID1 in specreg.h
esp_partition: add ESP_PARTITION_TYPE_ANY for iterating over all partition types (Github PR)
Closes IDFGH-4783
See merge request espressif/esp-idf!13434
esp_hid: allow setting stack size for the event task, increase stack size in the example (Github PR)
Closes IDFGH-4568 and IDFGH-4563
See merge request espressif/esp-idf!13438
* Extend sdmmc_slot_config_t with GPIO pin numbers for all SD bus
signals. These new fields are available if SOC_SDMMC_USE_GPIO_MATRIX
is set.
* Add shorter "sd" and "wp" aliases for "gpio_sd" and "gpio_wp" field
names in sdmmc_slot_config_t.
* Deprecate sdmmc_host_pullup_en, prefer to enable this feature using
SDMMC_SLOT_FLAG_INTERNAL_PULLUP instead.
* sync the latest struct header file from ESP32
* add soc_caps.h macros to distinguish between IO MUX and GPIO Matrix
support in SDMMC on different chips.
* store GPIO matrix signal numbers in sdmmc_slot_info_t
This option is not enabled by default because many existing tests
use integer assertions to check the pointers:
TEST_ASSERT_EQUAL(NULL, pointer)
This causes a "cast from pointer to integer of different size"
(-Wpointer-to-int-cast) warning to be generated, as Unity converts
every argument to UNITY_UINT first, and with 64-bit support enabled,
UNITY_UINT becomes a 64-bit unsigned type.
Ivan Grokhotkov