Add support for UFS devices

[runningnak3d]

LG devices with UFS NAND use a sector size of 4096 not 512. Also, they have 7 block devices (/dev/block/sda to /dev/block/sdg) instead of the single block device (mmcblk0) that an eMMC device has.

I propose adding two switches:

--ufs to set the blocksize to 4096 --block to pass the block device that you want to work with

Alternatively could parse the GPT of all 7 block devices, and then issue the OPEN based on the partition that gets passed.

I am happy to open a PR, but some feedback on the best method of dealing with UFS would be good first :)

-- Brian

[tuxuser]

Hey @runningnak3d! Does an UFS device advertise somehow that it's using UFS? Maybe in the LAF Properties.. or in another way?

[runningnak3d]

Hmmm, well memory_id:'H28U74303AMR \n' but then we would have to have a list of all the NAND chips.

Another possibility is to parse dmesg:

[LAF] lun_sizes[0] = 124370944

[LAF] lun_sizes[1] = 32768

[LAF] lun_sizes[2] = 16384

[LAF] lun_sizes[3] = 16384

[LAF] lun_sizes[4] = 475136

[LAF] lun_sizes[5] = 16384

[LAF] lun_sizes[6] = 16384

Lastly, could try to open /dev/block/sda if it fails, it is an eMMC device, if it opens, it is UFS.

Those are the only auto detection methods that I can think of.

[Lekensteyn]

Parsing dmesg is probably fragile, maybe there is some /sys/block//queue/ file that could list the actual size (/sys/block/sda/queue/physical_block_size or logical_block_size)? Maybe it is even safe to use 4K in both UFS/eMMC cases if the offsets are aligned at that number (I am just thinking aloud).

eMMC has a single block device with a partition table, how does UFS work? What are the different block devices sda..sdg for? Do they have partitions like sda1, sda2, etc.?)

[runningnak3d]

Yes the /sys file system has the size of the devices, but we can't read those with protocol version 1.0.0.4. No way -- no cat command :)

The partitions that are on the single mmcblk0 block device are spread across the 7 block devices. For example laf on the v20 is /dev/block/sda1 recovery is /dev/block/sde2.

Yea, I agree that parsing dmesg is messy. IMHO trying to open sda is the cleanest. It will fail on eMMC devices and succeed on UFS devices. Can't check for mmcblk0 because if someone has a UFS device and has an SD card in, it will exist.

[runningnak3d]

I don't really like this solution because it relies on ls and that could be pulled in the future....

With that said, we could ls /dev/block/sda -- if it exists, it is a UFS device, if not eMMC. If it is a UFS, then we need to find out how many block devices there are as some devices use less than the 7 that are available on UFS. Iterate through sda to sdg till nothing is returned, then read each one to get the partition info into an array. At that point based on the partition name that is passed, the appropriate block device can be opened.

Thoughts?

[shadabmozaffar]

To know if memory device is UFS or eMMC, you can rely on block size, which is 4K for UFS. To detect, you can just read physical_part0 GPT, and if EFI PART sign in at 4K address, you have UFS device. EFI PART signature is always at Sector 1.

Btw, I am still thinking how to open physical part 1 and others with OPEN command?