I have a linux guest running Debian Buster (though this problem goes much further back) under a Windows 10 host. I have "Accelerate 3D graphics" enabled, with 2GB of graphics memory allocated. This VM has most of the resources of my workstation dedicated to it.
I'm getting, fairly frequently, errors in dmesg that look like this:
[255847.503765] [drm:vmw_cmdbuf_work_func [vmwgfx]] *ERROR* Command "SVGA_3D_CMD_DX_SET_QUERY_OFFSET" causing device error.
[255847.503779] [drm:vmw_cmdbuf_work_func [vmwgfx]] *ERROR* Command buffer offset is 0
[255847.503786] [drm:vmw_cmdbuf_work_func [vmwgfx]] *ERROR* Command size is 16
[255847.503935] [drm:vmw_cmdbuf_work_func [vmwgfx]] *ERROR* Command "SVGA_3D_CMD_DX_BEGIN_QUERY" causing device error.
[255847.503942] [drm:vmw_cmdbuf_work_func [vmwgfx]] *ERROR* Command buffer offset is 0
[255847.503950] [drm:vmw_cmdbuf_work_func [vmwgfx]] *ERROR* Command size is 12
[255847.504071] [drm:vmw_cmdbuf_work_func [vmwgfx]] *ERROR* Command "SVGA_3D_CMD_DX_END_QUERY" causing device error.
[255847.504079] [drm:vmw_cmdbuf_work_func [vmwgfx]] *ERROR* Command buffer offset is 0
[255847.504086] [drm:vmw_cmdbuf_work_func [vmwgfx]] *ERROR* Command size is 12
[255847.511801] code[77548]: segfault at 7f045f489fff ip 00007f036952f627 sp 00007ffddd154cd0 error 6 in vmwgfx_dri.so[7f0369176000+82e000]
There are usually a few hundred of these former errors before the segfault. Once I get a few segfaults, the entire guest locks up. Sometimes the segfaults look more directly like they came from the driver: those are the ones that fully lock the machine.
The easiest (though not only) way of recreating this is to open vscode and use it for a while. Many other electron apps will eventually cause the same issue. That said, it's probably a bad sign that
Turning off the accelerate 3D option isn't really a workable solution, as the applications become unbearably and unusably slow.
Does anyone know of a workaround or fix? I'm getting awfully close to getting rid of the workflow I've been using for many, many years.