Re: does drive_get_next(IF_NONE) make sense?

[Philippe Mathieu-Daudé]

On 11/2/21 16:14, Peter Maydell wrote:
> Does it make sense for a device/board to do drive_get_next(IF_NONE) ?
> 
> At the moment we have exactly one user of this, which is
> hw/misc/sifive_u_otp.c. This is a model of a one-time-programmable
> fuse, and the drive is providing the backing store for the fuse
> contents. Borrowing an IF_NONE for this seems a bit odd, but
> it's not clear any of the other IF_ types is better.
> 
> We also just (this release cycle) added models of the Xilinx
> efuse OTP fuses. Those have been implemented to use IF_PFLASH.
> (This is a somewhat unfortunate inconsistency I guess.)
> 
> We also have a patchseries currently in the code review stage
> which uses IF_NONE:
> https://patchew.org/QEMU/20211101232346.1070813-1-wuhaotsh@google.com/20211101232346.1070813-6-wuhaotsh@google.com/
> Here we are trying to provide a drive as backing store for some
> EEPROMs that hang off the i2c buses on some npcm7xx boards.
> 
> Are these uses of IF_NONE OK, or should we be doing something
> else (using IF_PFLASH, defining a new IF_*, ???)

IIUC '-drive if=xxx' is deprecated, replaced by '-blockdev'.

Personally I expect a BlockInterfaceType to be a bus interface
and see IF_NONE as a buggy case.

See also:
https://lore.kernel.org/qemu-devel/YDY7KKI1Xme29UlQ@stefanha-x1.localdomain/

I am not sure about the amount of work required to fully leverage
-blockdev and remove -drive. Can it be sugar-expanded from the CLI?

What else is missing to finish the blockdev conversion?