[REGRESSION][BISECTED] vmwgfx crashes with command buffer error after update

[Zack Rusin]

On Thu, Aug 15, 2024 at 4:30 PM Andreas Piesk <a.piesk at mailbox.org> wrote:
>
> Hello,
>
> the bug was first reported on VMware Workstation by rdkehn.
>
> On my setup (archlinux text mode only VM on ESXi 8.0U3 latest) the kernel does NOT crash, the screen just goes dark after switching the console from
>
> [    2.745694] Console: switching to colour dummy device 80x25
>
> to
>
> [    2.771998] Console: switching to colour frame buffer device 160x50
>
> You see the VMware remote console resizing, then going black and from this point no more output.
>
> I have attached boot_journal and vmware.log from my setup. VM uses EFI boot and SVGA with defaults as display adapter, I attached the vmx file too.

Thanks! I see. I have a patch out that fixes it, but in general I
think those vm's with 16mb for graphics are very risky and I'd suggest
bumping them to at least 32mb. The vram portion can stay at 16mb, but
the graphicsMemoryKB can be safely set to fourth or even half of
memsize (in your config 256mb or even 512mb), which will make the vm's
a lot safer and allow actual ui usage because with console being
pinned we just don't have a lot of wiggle room otherwise and we just
can't migrate pinned framebuffers.
The patch that "regressed" this makes dumb buffers surface that
actually respect pinning, but as long as you don't have gpu host side
things will be ok. Otherwise we can't make a config with 16mb of
available graphics memory and graphics acceleration work.

z