On Sat, Dec 07, 2024 at 07:02:55AM +0000, Wei Liu wrote:

On Wed, Nov 06, 2024 at 07:42:45AM -0800, mhkelley58@gmail.com wrote:

quoted

From: Michael Kelley <redacted>

Patch 1 fixes util_probe() to not force the error return value to
ENODEV when the util_init function fails -- just return the error
code from util_init so the real error code is displayed in messages.

Patch 2 fixes a more serious race condition between initialization
of the VMBus channel and initial operations of the user space
daemons for KVP and VSS. The fix reorders the initialization in
util_probe() so the race condition can't happen.

The two fixes are functionally independent, but Patch 2 introduces
the util_init_transport function that parallels the existing code
for the util_init function. Doing Patch 1 first avoids an
inconsistency in the error handling in similar code for these two
parts of util_probe().

This series is v2 of a single patch first posted by Dexuan Cui
to fix the race condition.[1] I've taken over the patch per
discussion with Dexuan.

[1] https://lore.kernel.org/linux-hyperv/20240909164719.41000-1-decui@microsoft.com/ (local)

Michael Kelley (2):
  Drivers: hv: util: Don't force error code to ENODEV in util_probe()
  Drivers: hv: util: Avoid accessing a ringbuffer not initialized yet

Acked-by: Wei Liu <wei.liu@kernel.org>

Applied to hyperv-fixes, thanks.