On 03/11/2016 02:43 AM, Martin K. Petersen wrote:

quoted

quoted

quoted

quoted

quoted

quoted

"Arnd" == Arnd Bergmann [off-list ref] writes:

Arnd> Looking through what other drivers do, I've found a couple of
Arnd> patterns now. n particular, most use the SG_IO ioctl to pass down
Arnd> commands from user space into a device specific command
Arnd> queue. Have you looked at that interface in the past to see if it
Arnd> would fit your use case?

Arnd> There is also a 'bsg' API that some drivers implement, which I
Arnd> think would be another alternative.

Arnd> Could any of the SCSI experts comment on what they expect a driver
Arnd> to use out of those three alternatives (if any):

My preference is that if you want to expose any information that a user
or admin would care about, put it in sysfs where it is easily accessible
and can be scripted.

For things that need to directly send commands to the hardware (to
configure vendor specific settings or inspect parameters for development
purposes) use the sg or bsg interfaces like we do for both ATA and
SCSI. bsg is newer and supports more features like bidirectional
commands. But for this particular use case I don't think it offers any
particular advantages over SG_IO. Both interfaces take the same
descriptors so it really doesn't matter much.

See:

	http://sg.danny.cz/sg/

And indeed I would expose this 'special' UFS RPMB as a 'generic'
SCSI device and add the required fields in sysfs, much like Greg KH
suggested.
Plus adding a 'bsg' interface for sending 'raw' UFS commands sounds
much more appealing; that's what we do for SAS and FC, too.

But then, none of the above can be done if there is no hardware
available. At the same time, you wouldn't be able to test any
proposed fixes to the original ioctl patch.
So I guess you'll have to retract this patch or get hold of the
hardware to actually _test_ something there.

Or send me some, I'll be happy to help out here :-)

I will submit proposal or new RPMB layer for both UFS and EMMC 
It also sports IOCTL interface, similar and replacing to mmc_ioc_cmd which is also used to access RPMB partition on emmc.
Blocks  of 512 bytes has to be transferred back and force so sysfs API is not good enough.  RPMB partition is not suitable to be accessed via block layer.  I've tried going via SICSI generic layers, it is a bit awkward as the protocol is very UFS specific.  
Currently I have working solution for both EMMC and UFS2.0.

Thanks
Tomas