Regression: desk.cpl hangs while trying to apply properties in Live CD

Noticed this problem while trying to investigate CORE-19049. Tested build 0.4.15-dev-6359-g0e8cf6f.

This is not related to switching screen modes, as I thought earlier. ReactOS LiveCD, when the explorer shell has been started, enters some kind of stasis for some period of time, in which it hangs when trying to apply desktop properties.

Steps to reproduce:

• Start Live CD
• Go to Desktop properties
• Select (but do not apply) a different screensaver
• Revert the selection
• Click Apply
• Expected result: Settings are applied immediately
• Observed result: Desktop properties window hangs (system is still responsive) for some time
• Then after a minute it unhangs and applies the changes

These debug prints are being spammed while in stasis:

(ntoskrnl/se/accesschk.c:588) SepAccessCheck(): Failed to grant access rights. RemainingAccess = 0x00000001  DesiredAccess = 0x00000001

(ntoskrnl/se/accesschk.c:588) SepAccessCheck(): Failed to grant access rights. RemainingAccess = 0x00000001  DesiredAccess = 0x00000001

(ntoskrnl/se/accesschk.c:588) SepAccessCheck(): Failed to grant access rights. RemainingAccess = 0x00000001  DesiredAccess = 0x00000001

Here is debug log when booting LiveCD, entering the desktop, and waiting some time before it becomes usable: 0.4.15-dev-6402-g6299340-idle.txt

It exits the stasis when these debug prints are no more printed, and then you can normally apply the desktop properties.

--------------

Previous steps to reproduce:

• Use QEMU with Bochs video adapter (default)
• Start Live CD
• Go to Desktop properties, try setting video mode to 640x480x32
• Expected result: Video mode is being changed immediately
• Observed result: Desktop properties window hangs (system is still responsive) for some time
• Then after a minute it unhangs and switches the mode

Screenshots:

Before regression	After regression

In order to find guilty commit I had to perform deep bisection test:

ReactOS Build	Debug log	Clean	With a389f8a applied
0.4.15-dev-6049-ge54b32b	0.4.15-dev-6049-ge54b32b.txt
0.4.15-dev-6050-g13e6cbc	–
0.4.15-dev-6051-g024daea	–		–
0.4.15-dev-6052-g2626c72	–		–
0.4.15-dev-6053-gaeffa57	–		–
0.4.15-dev-6054-gbddd8b0	–		–
0.4.15-dev-6055-g36cdd34	–		–
0.4.15-dev-6056-gb29b268	–		–
0.4.15-dev-6057-gd708c79	–		–
0.4.15-dev-6058-ga389f8a	0.4.15-dev-6058-ga389f8a.txt		–

• - no regression / good
• - BSOD happens
• - regressed / bad

As you can see, 0.4.15-dev-6058-ga389f8a is the first "bad" build, however when applying the fix from this commit on top of other commits, it reveals the actual first bad commit is 0.4.15-dev-6050-g13e6cbc, related to CORE-18986.