Re: [PATCH v4 3/3] scsi: ufs: rpmb: Drop the unregistered ufs_rpmb bus

[Stanley Jhu <stanleyjhu@google.com>]

On Mon, 2026-09-14 at 12:05 +0200, Bean Huo wrote:
> "has always failed" is not correct, bus_add_device() only started rejecting
> devices on an unregistered bus after this commit:
> 36f35b8df697 ("driver core: reject devices with unregistered buses").
> Before that, device_register() succeeded and the device was just not added to
> the bus, so UFS RPMB did register until that commit 36f35b8df697.

You are right, and the consequence is larger than the wording. I built
two kernels from the same base with the same config, the only difference
being a revert of 36f35b8df697, with none of this series applied:

  base:                  /sys/class/rpmb empty, one "cannot add device"
  36f35b8df697 reverted: rpmb0..rpmb3, four regions registered

After unbinding the host, the reverted kernel still has all four devices
in /sys/class/rpmb while /sys/class/scsi_device is empty.

So the leak that patch 2 fixes is real on v6.19 through v7.1, and this
patch re-enables what v7.2-rc1 disabled.

v5 restores Cc: stable on both UFS patches. The cover letter cites
36f35b8df697 with the measurement above, and notes that a backport
needs all three patches in order.