UF2 is a file format, developed by Microsoft for PXT (also known as Microsoft MakeCode), that is particularly suitable for flashing microcontrollers over MSC (Mass Storage Class; aka removable flash drive).
For a more friendly explanation, check out this blog post. Also, take a look at the list of implementations at the bottom of this document.
The UF2 file consists of 512 byte blocks, each of which is self-contained and independent of others. Each 512 byte block consists of (see below for details):
- magic numbers at the beginning and at the end
- address where the data should be flashed
- up to 476 bytes of data
The data transfers over MSC always arrive in multiples of 512 bytes. Together with the FAT file system structure, this means that blocks of the UF2 file are always aligned with the MSC writes - the microcontroller never gets a partial file.
The magic numbers let the microcontroller distinguish an UF2 file block from other data (eg., FAT table entry, or various book-keeping files stored by some operating systems). When a UF2 block is recognized, it can be flashed immediately (unless flash page size is more than 256 bytes; in that case a buffer is needed). The actual handling of file format during writing is very simple (~10 lines of C code in simplest version).
A UF2 file consists of 512 byte blocks. Each block starts with a 32 byte header, followed by data, and a final magic number. All fields, except for data, are 32 bit unsigned little endian integers.
| Offset | Size | Value |
|---|---|---|
| 0 | 4 | First magic number, 0x0A324655 ("UF2\n") |
| 4 | 4 | Second magic number, 0x9E5D5157 |
| 8 | 4 | Flags |
| 12 | 4 | Address in flash where the data should be written |
| 16 | 4 | Number of bytes used in data (often 256) |
| 20 | 4 | Sequential block number; starts at 0 |
| 24 | 4 | Total number of blocks in file |
| 28 | 4 | File size or board family ID or zero |
| 32 | 476 | Data, padded with zeros |
| 508 | 4 | Final magic number, 0x0AB16F30 |
The following C struct can be used:
struct UF2_Block {
// 32 byte header
uint32_t magicStart0;
uint32_t magicStart1;
uint32_t flags;
uint32_t targetAddr;
uint32_t payloadSize;
uint32_t blockNo;
uint32_t numBlocks;
uint32_t fileSize; // or familyID;
uint8_t data[476];
uint32_t magicEnd;
} UF2_Block;Currently, there are two flags is defined:
0x00000001- not main flash - this block should be skipped when writing the device flash; it can be used to store "comments" in the file, typically embedded source code or debug info that does not fit on the device flash0x00001000- file container - see below0x00002000- familyID present - when set, thefileSize/familyIDholds a value identifying the board family (usually corresponds to an MCU)
This field is optional, and should be set only when the corresponding flag is set. It is recommended that new bootloaders require the field to be set appropriately, and refuse to flash UF2 files without it. If you're developing your own bootloader, and your board family isn't listed here, pick a new family ID at random. It's good to also send a PR here, so your family can be listed.
If the familyID doesn't match, the bootloader should disregard the
entire block, including blockNo and numBlocks fields.
In particular, writing a full UF2 file with non-matching familyID
should not reset the board.
This also allows for several files with different familyID to be
simply concatenated together, and the whole resulting file to be copied
to the device with only one actually being written to flash.
The reason to pick numbers at random is to minimize risk of collisions
in the wild. Do not pick random numbers by banging on keyboard, or by using
0xdeadf00d, 0x42424242 etc. A good way is to use the following
shell command: printf "0x%04x%04x\n" $RANDOM $RANDOM
This procedure was unfortunately not used for the SAMD51 below.
- Microchip (Atmel) SAMD21 - 0x68ed2b88
- Microchip (Atmel) SAMD51 - 0x55114460
- Nordic NRF52840 - 0x1b57745f
- ST STM32F103 - 0x5ee21072
- ST STM32F401 - 0x57755a57
- Microchip (Atmel) ATmega32 - 0x16573617
- Cypress FX2 - 0x5a18069b
The magic number at the end is meant to mitigate partial block writes.
Second and final magic numbers were randomly selected, except for the last byte
of final magic number, which was forced to be '\n' (0xA). Together with the
first magic number being "UF2\n" this makes it easy to identify UF2 blocks in
a text editor.
The header is padded to 32 bytes, as hex editors commonly use 16 or 32 bytes as line length. This way, the data payload is aligned to line start.
32 bit integers are used for all fields so that large flash sizes can be supported in future, as well as for simplicity. Little endian is used, as most microcontrollers are little endian. 8 bit microcontrollers can choose to just use the first 16 bits of various header fields.
The total number of blocks in the file and the sequential block number make it easy for the bootloader to detect that all blocks have been transferred. It requires one bit of memory per block (eg., on SAMD21G18A it's 128 bytes). Alternatively, the bootloader might ignore that and just implement a reset after say 1 second break in incoming UF2 blocks.
The number of data bytes is configurable and depends on the size of the flash page (that is the smallest size that can be erased) on the microcontroller.
- if the page size is more than
476bytes, the bootloader should support any payload size, as it needs to buffer the entire page in memory anyway - if the page size is less than
476bytes, the payload should be a multiple of page size, so it can be written without buffering; the target address should also be a multiple of page size
In any event, payload size and target address should always be 4-byte aligned.
Note that payload size of 256 is always correct, and makes it easy to convert
between flash addresses and UF2 file offsets.
For example, on Atmel's SAMD21 chips the page size is 256 bytes, and this
also is the payload size. If the page size was 128 bytes, one could use
payload of 128*3. Nordic nRF51 has page size of 1024 bytes, and thus
any payload size should be allowed.
Some IDEs will embed program sources in the UF2 file. This allows a UF2 files to be loaded by the IDE and serve as a natural backup and transfer format. This can be done in two ways:
- using the "not main flash" flag
- using normal blocks that are flashed to the device
If the bootloader can expose CURRENT.UF2 file (see below) and there is enough
flash available, than the second option is more desirable, as it allows sharing
programs directly from the board.
See https://makecode.com/source-embedding for more info.
The file format is designed specifically to deal with the following problems:
- operating system (OS) writing blocks in different order than occurs in a file
- OS writing blocks multiple times
- OS writing data that is not UF2 blocks
- OS writing first/final part of a block, possibly for metadata detection or search indexing
The only file system assumption we make is that blocks of file are aligned with blocks on the hard drive. It's likely true of many file systems besides FAT.
We also assume that USB MSC device reports its block size to be a multiple of 512
bytes. In the wild these devices always almost report exactly 512, and some
operating systems do not support other values.
Bootloaders may expose virtual files in their MSC devices. These are standardized here, so that flashing tools can automatically detect the bootloaders.
INFO_UF2.TXT- contains information about the bootloader build and the board on which it is runningINDEX.HTM- redirects to a page that contains an IDE or other informationCURRENT.UF2- the contents of the entire flash of the device, starting at0x00000000, with256payload size; thus, the size of this file will report as twice the size of flash
Flashing tools can use the presence of INFO_UF2.TXT (in upper or lower case,
as FAT is case-insensitive) file as an indication that a given directory is
actually a connected UF2 board. The other files should not be used for
detection.
Typical INFO_UF2.TXT file looks like this:
UF2 Bootloader v1.1.3 SFA
Model: Arduino Zero
Board-ID: SAMD21G18A-Zero-v0
The Board-ID field is machine-readable and consists of a number of dash-separated tokens.
The first token is the CPU type, second is the board type, and third is the board revision.
More tokens can be also added.
The bootloader should contain its info file as a static string somewhere in its code.
If possible, the last word of the bootloader code should point to this string.
This way, the info file can be found in the initial section of the CURRENT.UF2
file as well. Thus, a board type can be determined from the contents of CURRENT.UF2.
This is particularly useful with the source embedding (see above).
It is also possible to use the UF2 format as a container for one or more regular files (akin to a TAR file, or ZIP archive without compression). This is useful when the embedded device being flashed sports a file system.
The program to run may reside in one of the files, or in the main flash memory.
In such a usage the file container flag is set on blocks, the field fileSize
holds the file size of the current file, and the field targetAddr holds the
offset in current file.
The not main flash flag on blocks should be ignored when the file container is set.
The file name is stored at &data[payloadSize] (ie., right after the actual payload) and
terminated with a 0x00 byte. The format of filename is dependent on the
bootloader (usually implemented as some sort of file system daemon).
The bootloader will usually allow any size of the payload.
The current files on device might be exposed as multiple UF2 files, instead of
a single CURRENT.UF2. They may reside in directories, however, due to UF2 general
design, it doesn't matter which directory the UF2 file is written to.
Typical writing procedure is as follows:
- validate UF2 magic numbers
- make sure that
targetAddr < fileSizeand thatfileSizeisn't out of reasonable range - write
0x00atdata[475]to ensure NUL termination of file name - read file name from
&data[payloadSize]; perform any mapping on the file name - create a directory where the file is to be written if doesn't exists
- open the file for writing
- truncate the file to
fileSize - seek
targetAddr - write the payload (ie.,
data[0 ... payloadSize - 1]) - close the file
The fields blockNo and numBlocks refer to the entire UF2 file, not the current
file.
- Microchip ATSAMD21 and ATSAMD51
- Arduino UNO
- STM32F103
- STM32F4
- Nordic NRF52840
- Linux (RPi Zero)
- Cypress FX2
There's an ongoing effort to implement UF2 in Codal.
MIT
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.