x11/nvidia-driver versions 358.16 and above - lockup when X server starts

[Bengt Ahlgren]

There seems to be a locking problem in the newer nvidia drivers (358.16
and above that have the new nvidia-modeset.ko module).  These are not in
the ports tree, but a port patch is being worked on to update it:

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=201340

With a WITNESS kernel, my system always locks up when the X server
starts after printing this:

Apr  8 19:44:11 ivy kernel: acquiring duplicate lock of same type: "os.lock_sx"
Apr  8 19:44:11 ivy kernel: 1st os.lock_sx @ nvidia_os.c:593
Apr  8 19:44:11 ivy kernel: 2nd os.lock_sx @ nvidia_os.c:593
Apr  8 19:44:11 ivy kernel: KDB: stack backtrace:
Apr  8 19:44:11 ivy kernel: db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame 0xfffffe06651c7410
Apr  8 19:44:11 ivy kernel: kdb_backtrace() at kdb_backtrace+0x39/frame 0xfffffe06651c74c0
Apr  8 19:44:11 ivy kernel: witness_checkorder() at witness_checkorder+0xe24/frame 0xfffffe06651c7550
Apr  8 19:44:11 ivy kernel: _sx_xlock() at _sx_xlock+0x75/frame 0xfffffe06651c7590
Apr  8 19:44:11 ivy kernel: os_acquire_mutex() at os_acquire_mutex+0x32/frame 0xfffffe06651c75b0
Apr  8 19:44:11 ivy kernel: _nv014117rm() at _nv014117rm+0x18/frame 0xfffffe000a188e60
Apr  8 19:44:11 ivy kernel: dmapbase() at 0xfffff80011527800

The X server gets at least this far:

[   164.014] (==) NVIDIA(0): Depth 24, (==) framebuffer bpp 32
[   164.014] (==) NVIDIA(0): RGB weight 888
[   164.014] (==) NVIDIA(0): Default visual is TrueColor
[   164.014] (==) NVIDIA(0): Using gamma correction (1.0, 1.0, 1.0)
[   164.015] (**) NVIDIA(0): Enabling 2D acceleration

Without WITNESS, it locks up perhaps 1/3 of the time, and works fine
afterways if it didn't.  I run FreeBSD 10.3-PRERELEASE amd64 r296673 on
a desktop with a Z77 motherboard and a GeForce GTX 550 Ti.

Is there anything that the FreeBSD community can do with this info?  I
have submitted a problem report to NVIDIA.

Bengt