The latest development version of this page may be more current than this released 4.0.0 version.

Migration guide to Zephyr v3.5.0

This document describes the changes required or recommended when migrating your application from Zephyr v3.4.0 to Zephyr v3.5.0.

Any other changes (not directly related to migrating applications) can be found in the release notes.

Required changes

Kernel

  • The kernel k_mem_slab_free() function has changed its signature, now taking a void *mem pointer instead of a void **mem double-pointer. The new signature will not immediately trigger a compiler error or warning, instead likely causing a invalid memory access at runtime. A new _ASSERT statement, that you can enable with CONFIG_ASSERT, will detect if you pass the function memory not belonging to the memory blocks in the slab.

  • CONTAINER_OF now performs type checking, this was very commonly misused to obtain user structure from k_work pointers without passing from k_work_delayable. This would now result in a build error and have to be done correctly using k_work_delayable_from_work().

C Library

  • The default C library used on most targets has changed from the built-in minimal C library to Picolibc. While both provide standard C library interfaces and shouldn’t cause any behavioral regressions for applications, there are a few side effects to be aware of when migrating to Picolibc.

    • Picolibc enables thread local storage (CONFIG_THREAD_LOCAL_STORAGE) where supported. This changes some internal operations within the kernel that improve performance using some TLS variables. Zephyr places TLS variables in the memory reserved for the stack, so stack usage for every thread will increase by 8-16 bytes.

    • Picolibc uses the same malloc implementation as the minimal C library, but the default heap size depends on which C library is in use. When using the minimal C library, the default heap is zero bytes, which means that malloc will always fail. When using Picolibc, the default is 16kB with CONFIG_MMU or ARCH_POSIX, 2kB with CONFIG_USERSPACE and CONFIG_MPU_REQUIRES_POWER_OF_TWO_ALIGNMENT. For all other targets, the default heap uses all remaining memory on the system. You can change this by adjusting CONFIG_COMMON_LIBC_MALLOC_ARENA_SIZE.

    • Picolibc can either be built as part of the OS build or pulled from the toolchain. When building as part of the OS, the build will increase by approximately 1000 files.

    • When using the standard C++ library with Picolibc, both of those must come from the toolchain as the standard C++ library depends upon the C library ABI.

    • Picolibc removes the -ffreestanding compiler option. This allows significant compiler optimization improvements, but also means that the compiler will now warn about declarations of main which don’t conform to the Zephyr required type – int main(void).

    • Picolibc has four different printf/scanf variants supported in Zephyr, ‘double’, ‘long long’, ‘integer’, and ‘minimal. ‘double’ offers a complete printf implementation with exact floating point in decimal and hexadecimal formats, full integer support including long long, C99 integer size specifiers (j, z, t) and POSIX positional arguments. ‘long long’ mode removes float support, ‘integer’ removes long long support while ‘minimal’ mode also removes support for format modifiers and positional arguments. Building the library as a module allows finer control over the feature set provided at each level.

    • Picolibc’s default floating point input/output code is larger than the minimal C library version (this is necessary to conform with the C language “round trip” requirements for these operations). If you use CONFIG_CBPRINTF_FP_SUPPORT, you will see increased memory usage unless you also disable CONFIG_PICOLIBC_IO_FLOAT_EXACT, which switches Picolibc to a smaller, but inexact conversion algorithm. This requires building Picolibc as a module.

Optional Modules

The following modules have been made optional and are not downloaded with west update by default anymore:

  • chre

  • lz4

  • nanopb

  • psa-arch-tests

  • sof

  • tf-m-tests

  • tflite-micro

  • thrift

  • zscilib

To enable them again use the west config manifest.project-filter -- +<module name> command, or west config manifest.group-filter -- +optional to enable all optional modules, and then run west update again.

Device Drivers and Device Tree

  • zephyr,memory-region-mpu was renamed zephyr,memory-attr and its type moved from ‘enum’ to ‘int’. To have a seamless conversion this is the required change in the DT:

    - "RAM"         -> <( DT_MEM_ARM(ATTR_MPU_RAM) )>
    - "RAM_NOCACHE" -> <( DT_MEM_ARM(ATTR_MPU_RAM_NOCACHE) )>
    - "FLASH"       -> <( DT_MEM_ARM(ATTR_MPU_FLASH) )>
    - "PPB"         -> <( DT_MEM_ARM(ATTR_MPU_PPB) )>
    - "IO"          -> <( DT_MEM_ARM(ATTR_MPU_IO) )>
    - "EXTMEM"      -> <( DT_MEM_ARM(ATTR_MPU_EXTMEM) )>
    
  • Device dependencies (incorrectly referred as “device handles” in some areas) are now an optional feature enabled by CONFIG_DEVICE_DEPS. This means that an extra linker stage is no longer necessary if this option is not enabled.

  • On all STM32 ADC, it is no longer possible to read sensor channels (Vref, Vbat or temperature) using the ADC driver. The dedicated sensor driver should be used instead. This change is due to a limitation on STM32F4 where the channels for temperature and Vbat are identical, and the impossibility of determining what we want to measure using solely the ADC API.

  • The RAM disk driver has been changed to support multiple instances and instantiation using devicetree. As a result, Kconfig option CONFIG_DISK_RAM_VOLUME_SIZE and Kconfig option CONFIG_DISK_RAM_VOLUME_NAME are removed, and the application using the RAM disk must instantiate it using devicetree, as in the following example:

    / {
        ramdisk0 {
            compatible = "zephyr,ram-disk";
            disk-name = "RAM";
            sector-size = <512>;
            sector-count = <192>;
        };
    };
    
  • The goodix,gt911, xptek,xpt2046 and hynitron,cst816s drivers have been converted from Kscan to Input, they can still be used with Kscan applications by adding a zephyr,kscan-input node.

  • The zephyr,gpio-keys binding has been merged into gpio-keys and the callback definition has been renamed from INPUT_LISTENER_CB_DEFINE to INPUT_CALLBACK_DEFINE.

  • The ti,bq274xx driver was using incorrect units for capacity and power channels, these have been fixed and scaled by x1000 factor from the previous implementation, any application using them has to be changed accordingly.

  • The configuration options for the SSD1306 display driver can now be provided via the Devicetree binding solomon,ssd1306fb. The following Kconfig options: CONFIG_SSD1306_DEFAULT, CONFIG_SSD1306_SH1106_COMPATIBLE, and CONFIG_SSD1306_REVERSE_MODE have been removed.

    • You can remove CONFIG_SSD1306_DEFAULT without any other modification.

    • CONFIG_SSD1306_SH1106_COMPATIBLE was used to assert that the device is (compatible with) SH1106. This has been replaced by a dedicated dts compatible declaration. You may update an existing sh1106 node to change the compatible designation from solomon,ssd1306fb to sinowealth,sh1106.

    • CONFIG_SSD1306_REVERSE_MODE is now set using the inversion-on property of the devicetree node.

  • GPIO drivers not implementing IRQ related operations must now provide NULL to the relevant operations: pin_interrupt_configure, manage_callback, get_pending_int. The public API will return -ENOSYS when these are not available, instead of -ENOTSUP.

  • STM32 Ethernet driver was misusing hwinfo_get_device_id() to generate last 3 bytes of mac address, resulting in a high risk of collision when using SoCs from the same lot. This is now fixed to use the whole range of entropy available from the unique ID (96 bits). Devices using unique ID based mac address will see last 3 bytes of their MAC address modified by this change.

  • On all STM32 (except F1x and F37x series), two new required properties have been added to ADC to configure the source clock and the prescaler. st,adc-clock-source allows choosing either synchronous or asynchronous clock source. st,adc-prescaler allows setting the value of the prescaler for the chosen clock source. Not all combinations are allowed. Refer to the appropriate RefMan for more information. When choosing asynchronous clock, the choice of the kernel source clock is made in the clocks node as it is done for other peripherals, for example, to select HSI16 as clock source for STM32G0:

    &adc {
        clocks = <&rcc STM32_CLOCK_BUS_APB1_2 0x00100000>,
                 <&rcc STM32_SRC_HSI ADC_SEL(2)>;
      };
    
  • On NXP boards with LPC DMA, the DMA controller node used to have its dma-channels property set in the board DTS as a way to configure the amount of structures the driver will allocate. This did not match the zephyr dma-controller binding, so this property is now fixed and set in the SOC devicetree definition. Downstream boards should not override this property and instead use the new driver Kconfig CONFIG_DMA_MCUX_LPC_NUMBER_OF_CHANNELS_ALLOCATED.

  • The LPC55XXX series SOC (except LPC55S06) default main clock has been updated to PLL1 source from XTAL32K running at 144MHZ. If the new kconfig option CONFIG_INIT_PLL1 is disabled then the main clock is muxed to FRO_HR as before.

  • The Kconfig option CONFIG_GPIO_NCT38XX_INTERRUPT has been renamed to CONFIG_GPIO_NCT38XX_ALERT.

  • The CAN controller timing API functions can_set_timing() and can_set_timing_data() no longer fallback to the (Re-)Synchronization Jump Width (SJW) value set in the devicetree properties for the given CAN controller upon encountering an SJW value corresponding to CAN_SJW_NO_CHANGE (which is no longer available). The caller will therefore need to fill in the sjw field in can_timing. To aid in this, the can_calc_timing() and can_calc_timing_data() functions now automatically calculate a suitable SJW. The calculated SJW can be overwritten by the caller if needed. The CAN controller API functions can_set_bitrate() and can_set_bitrate_data() now also automatically calculate a suitable SJW, but their SJW cannot be overwritten by the caller.

  • The CAN ISO-TP message configuration in isotp_msg_id is changed to use the following flags instead of bit fields:

    The two new flags ISOTP_MSG_FDF and ISOTP_MSG_BRS were added for CAN FD mode.

  • NXP i.MX RT based boards should now enable CONFIG_DEVICE_CONFIGURATION_DATA at the board level when using a DCD with the RT bootrom, and enable CONFIG_NXP_IMX_EXTERNAL_SDRAM when using external SDRAM via the SEMC

  • NXP i.MX RT11xx series SNVS pin control name identifiers have been updated to match with the source data for these SOCs. The pin names have had the suffix dig added. For example, iomuxc_snvs_wakeup_gpio13_io00 has been renamed to iomuxc_snvs_wakeup_dig_gpio13_io00

Power Management

  • Platforms that implement power management hooks must explicitly select CONFIG_HAS_PM in Kconfig. This is now a dependency of CONFIG_PM. Before this change all platforms could enable CONFIG_PM because empty weak stubs were provided, however, this is no longer supported. As a result of this change, power management hooks are no longer defined as weaks.

  • Multiple platforms no longer support powering the system off using pm_state_force(). The new sys_poweroff() API must be used. Migrated platforms include Nordic nRF, STM32, ESP32 and TI CC13XX/26XX. The new API is independent from CONFIG_PM. It requires CONFIG_POWEROFF to be enabled, which depends on CONFIG_HAS_POWEROFF, an option selected by platforms implementing the required new hooks.

Bootloader

Bluetooth

  • The accept() callback’s signature in bt_l2cap_server has changed to int (*accept)(struct bt_conn *conn, struct bt_l2cap_server *server, struct bt_l2cap_chan **chan), adding a new server parameter pointing to the bt_l2cap_server structure instance the callback relates to. GitHub #60536

Networking

  • A new networking Kconfig option CONFIG_NET_INTERFACE_NAME defaults to y. The option allows user to set a name to a network interface. During system startup a default name is assigned to the network interface like eth0 to the first Ethernet network interface. The option affects the behavior of SO_BINDTODEVICE BSD socket option. If the Kconfig option is set to n, which is how the system worked earlier, then the name of the device assigned to the network interface is used by the SO_BINDTODEVICE socket option. If the Kconfig option is set to y (current default), then the network interface name is used by the SO_BINDTODEVICE socket option.

  • Ethernet PHY devicetree bindings were updated to use the standard reg property for the PHY address instead of a custom address property. As a result, MDIO controller nodes now require #address-cells and #size-cells properties. Similarly, Ethernet PHY devicetree nodes and corresponding driver were updated to consistently use the node name ethernet-phy instead of phy. Devicetrees and overlays must be updated accordingly:

    mdio {
        compatible = "mdio-controller";
        #address-cells = <1>;
        #size-cells = <0>;
    
        ethernet-phy@0 {
            compatible = "ethernet-phy";
            reg = <0>;
        };
    };
    

Other Subsystems

  • ZBus runtime observers implementation now relies on the HEAP memory instead of a memory slab. Thus, zbus’ configuration (kconfig) related to runtime observers has changed. To keep your runtime observers code working correctly, you need to:

  • The zbus VDED delivery sequence has changed. Check the documentation to verify if it will affect your code.

  • MCUmgr SMP version 2 error codes entry has changed due to a collision with an existing response in shell_mgmt. Previously, these errors had the entry ret but now have the entry err. smp_add_cmd_ret() is now deprecated and smp_add_cmd_err() should be used instead, MGMT_CB_ERROR_RET is now deprecated and MGMT_CB_ERROR_ERR should be used instead. SMP version 2 error code defines for in-tree modules have been updated to replace the *_RET_RC_* parts with *_ERR_*.

  • MCUmgr SMP version 2 error translation (to legacy MCUmgr error code) is now handled in function handlers by setting the mg_translate_error function pointer of mgmt_group when registering a group. See smp_translate_error_fn for function details. Any SMP version 2 handlers made for Zephyr 3.4 need to be updated to include these translation functions when the groups are registered.

ARM

  • ARM SoC initialization routines no longer need to call NMI_INIT(). The macro call has been removed as it was not doing anything useful.

RISC V