The EDL Utility is a Win32 utility for accessing the Qualcomm Emergency Download interface on Qualcomm processors.
Note: The usage has changed and /e is only for erasing NAND memories.
To zero out sections of eMMC or UFS use the new /f fill command.
Qualcomm processors support two different protocols, "Sahara" and "Firehose". Sahara is supported in ROM and is always present. Firehose is implemented in downloadable loaders in ELF format.
The usual procedure is to first get your device in EDL mode, i.e. where it is presenting USB VID/PID 05c6/9008. This can be achieved by:
Everything under Windows needs some kind of driver. Zadig is a simple generic driver generator. Do NOT use any Windows drivers from Qualcomm. They will try to present your device as a serial port. Now you can do a simple check if you like.
This shows you that the device is connected and has the right driver.
Next, you must use the Sahara protocol to load a loader for the Firehose protocol.
Loaders are specific to processor, device manufacturer, possibly flash memory type and hash.
To decide which one you need you need to collect some basic info.
There are reports that Sahara protocol version 3.0 does not support querying the HWID or Hash.
If this happens to you, use the /qbc quirk (see below).
Loaders may be found (thank you!) at https://github.com/bkerler/Loaders/. These files often use .bin or .mbn as the extension despite it actually being a normal ELF file. The file names are based on the 16 hexit HWID and the first 16 hexits of the Hash. By one Russian web site they are listed under the last 8 hexits of the Hash. As the filenames tend to be cumbersome, you might rename them something short and mnemonic.
From this point on the processor is using the Firehose protocol and you need not (can not) reload the loader unless you reboot.
A device might be using eMMC storage (older devices), NAND storage or UFS storage (newer devices).
The /u flag must be used for all operation in Firehose on devices with UFS.
The flags /u (LUN), /p (partition), /s (start block), /c (count of blocks) and /b (block size) are used to specify the range of operation.
If the partition is specified then the start block is relative to the start of the partition.
If partition is not specified, then the start block is absolute.
Zero is the default for both start block and count of blocks.
Partition operations often do not specify either start block or count of blocks.
Operations on raw devices (i.e. not a partition) require an explicit /s and /c to prevent accidents like edl /e.
| Start | Count | Partition | |
|---|---|---|---|
| Unspecified | Specified | ||
| 0 | 0 | Whole device | Whole partition |
| 0 | + | Start of device | Start of partition |
| + | + | Middle of device | Middle of partition |
| + | 0 | End of device | End of partition |
| − | 0 | End of device | End of partition |
| − | + | Part of end of device | Part of end of partition |
The major operations are /r (read), /e (erase), /w (write).
The erase and write operations can be combined which yields the non-optimized operations of full erase and (possibly) partial write (depending on the size of the input file).
Be very careful when you specify /e (erase), /w (write) as not specifying a partition means the whole device!
Partitions are sized for the maximum anticipated size of the contents.
Often the fraction of a partition that is actively being used is as low as 20%.
(There are often many partitons with all zeroes in them also.)
There is no particular need to transfer a whole partition when 20% will do.
Of course, if you still want to transfer another 50MB of zeroes, just don't use the /t flag.
Also note that some images have signing or other (sometimes) necessary things after the end of the normal image.
Currently the EDL utility has the capability to recognize the actual size of:
Android images are naturally aligned to pagesize (normally 4096 bytes) but ELF files can be any size. Therefore, when they are read, even when truncated, they are rounded up to the current device blocksize (normally 512 or 4096 bytes). This simplifies matters when/if they are written back to the device.
NAND memory has two peculiarities that require special handling. The first peculiarity is that they have "bad blocks" (an erase block is sometimes 64 x 4096 bytes). During a read the output file will be filled with 0xff wherever bad blocks are to maintain alignment. During a write the input file will be skipped over wherever bad blocks are to maintain alignment. The second peculiarity is that because of the hidden CRC32 and ECC on each page, a freshly erased page must never be written with all 0xff values. The EDL utility will do explicit multiple writes (in the hundreds) around the bad blocks and the empty pages. The EDL utility will do explicit multiple reads (a few) around the bad blocks. You must explicitly erase whichever region of the NAND memory before writing but this may be combined in the same command. The EDL utility now supports NAND volume tables analogously to GPT partition tables.
Quirks are idiosyncracies, anomalies or incorrect implementations of Firehose loaders.
By specifying the /q flag you can bypass problematic parts.
/qbc, for example, will not query HWID and hash.
| A | Do not query serial number |
| B | Do not query HWID |
| C | Do not query hash |
| D | Allow CSD read to fail (Sony Vivo) |
Show usage:
Query basic info:
Load a loader (needs to be done only once after a fresh start):
List the partitions:
Download the MBR of a UFS LUN:
Download the boot partition (and truncate to its actual size):
Erase the the last 4096 bytes of /vendor (removes FEC correction):
Flash the recovery partition:
Erase and write to NAND memory blocks:
Try some random XML:
Reboot to normal system:
Reboot to fastboot (probably only works on Motorola):
Download edl.exe, the EDL utiliy.
Download ubi.exe, a simple utility for examining full dumps of NAND/UBI.
See also: QcomView – a utility for analyzing Qualcomm xbl/abl/Firehose loaders