If in usb_host_interface_release() the underlying pipe was halted while having an URB in-flight,
usb_host_client_handle_events() may not have had a chance to process the URB yet.
Check if there are any pending URBs on the pipe before attempting to free it.
Closes https://github.com/espressif/esp-idf/issues/17707
Starting with ESP32-P4 we can have targets that have more than 1 USB-OTG peripheral.
This commit adds an option to choose which peripherals will be used by USB Host lib.
Internally, we will still have only 1 Root HUB but with multiple Root ports.
# Conflicts:
# components/usb/host_test/usb_host_layer_test/main/usb_host_install_unit_test.cpp
# components/usb/hub.c
# components/usb/test_apps/common/phy_common.c
# components/usb/test_apps/common/phy_common.h
# components/usb/test_apps/hcd/main/test_hcd_common.c
# components/usb/test_apps/usb_host/main/test_app_main.c
# components/usb/usb_host.c
754d357f refactor(ext_hub): Fixed device release, optimized the order of closing usbh device
fc61875a refactor(ext_hub): Pospone the device release, if device is not IDLE
3fd17b8b refactor(hub): Applied new ext_hub api, refactor func names
3003362b refactor(usb_host): Cancel hub porpagation to the user, rename non-critical func
19ce9ed6 refactor(test_usb_host_async): Added host_lib_task finish notification
f238d75b refactor(ext_port): Remove the error verification, as error will be handled in ext hub driver
Co-authored-by: Roman Leonov <roman.leonov@espressif.com>
Introduce a new HAL API `usb_dwc_hal_set_fifo_config()` that allows advanced users
to manually configure RX, Non-Periodic TX, and Periodic TX FIFO sizes. This offers
fine-grained control beyond the previous bias-based sizing approach.
The HAL function no longer returns `esp_err_t`, and internal validations are enforced
via `HAL_ASSERT()`. Responsibility for input validation has been moved to the HCD layer.
FIFO configuration must be applied before any USB pipes are created or activated.
This feature is intended for use during `usb_host_install()`.
If no custom FIFO configuration is provided (i.e., all values are zero),
the driver falls back to a bias-based default layout based on Kconfig settings
(`CONFIG_USB_HOST_HW_BUFFER_BIAS_*`). Bias resolution is done inside `hcd_port_init()`.
The `port_obj_t` structure has been extended with a `fifo_config` field, which stores
the configuration to allow re-application after a USB port reset.
Obsolete FIFO bias enums (`usb_hal_fifo_bias_t`, `hcd_port_fifo_bias_t`) and related
APIs (`hcd_port_set_fifo_bias()`) have been removed in favor of the new structure-based mechanism.
The HCD initialization and port reset flow has been updated to use the explicit
FIFO configuration.
USB Host maintainer documentation (`maintainers.md`) has been updated accordingly.
Test cases were updated to remove the usage of removed bias API and now rely on default
or custom FIFO configuration.
- DWC-OTG internal DMA can access psram on esp32p4
- Move DMA memory buffs to psram, to save internal ram
- HCD tests and MSC example runs in CI with psram enabled
Changed error code from ESP_ERR_INVALID_STATE to ESP_ERR_NOT_FOUND
when the client never opened the device.
Updated function documentation to correctly reflect return values.
If you call *usb_host_device_close()* for a device that isn't open, the function exits early,
without giving back the semaphore it took, which causes any other call that tries to take
that semaphore to hang indefinitely.
Strangely, there's redundant handling of this condition, with two checks in a row that both handle
the case where `_check_client_opened_device(client_obj, dev_addr)` returns `false`:
```c
HOST_CHECK_FROM_CRIT(_check_client_opened_device(client_obj, dev_addr), ESP_ERR_NOT_FOUND);
if (!_check_client_opened_device(client_obj, dev_addr)) {
// Client never opened this device
ret = ESP_ERR_INVALID_STATE;
HOST_EXIT_CRITICAL();
goto exit;
}
…
exit:
xSemaphoreGive(p_host_lib_obj->constant.mux_lock);
return ret;
```
The first line is the one that exits early, as HOST_CHECK_FROM_CRIT returns its second parameter
if its first parameter is false, without giving back the semaphore (although it does exit
the critical section).
The subsequent block handles the exact same case, except that it ensures the semaphore is given
back before returning. Currently, this block is never reached.
Perhaps the first check was added, then someone noticed the issue and added the second check,
but they forgot to remove the first one.
In any case, this PR removes the first check, so the second check can properly handle this case
by giving back the semaphore before returning.
This bug appears to have been present in the initial commit of the USB Host library to the ESP-IDF
repo: accbaee57c
Of course, if you never try to close a non-opened device, then you won't encounter it!
Unfortunately, I have some code that tried to do that, which is how I found the issue.
Previously, we included symbols from soc/usb_dwc_cfg.h and configured
the HAL and LL according to it. Now we get the configuration in runtime
from USB-DWC registers.
Added missing definition for USB FS peripheral on ESP32-P4.
Fixes issue where ESP_ERR_NO_MEM was being silently discarded after
cleaning up after a failed malloc in usb_host_client_register.
Signed-off-by: Daniel Mangum <georgedanielmangum@gmail.com>
