Skip to content

wayland demos hang in _egluNativeEventLoop loop on nVidia with __NV_PRIME_RENDER_OFFLOAD

I have a dual GPU setup. My primary GPU is an AMD RX 580 and my secondary is an nVidia GTX 970

Any attempt to run the mesa demos with __NV_PRIME_RENDER_OFFLOAD=1 cause these demos to block in __GI___poll with the following backtrace

(gdb) bt full
#0  0x00007ffff7b90f6f in __GI___poll (fds=0x7fffffffe300, nfds=3, timeout=-1) at ../sysdeps/unix/sysv/linux/poll.c:29
        sc_ret = -516
        sc_cancel_oldtype = 0
#1  0x000055555555b947 in event_loop () at ../src/egl/eglut/wsi/wayland.c:549
        poll_count = 3
        ret = 56
        pollfds = {{fd = 3, events = 1, revents = 0}, {fd = 3, events = 1, revents = 0}, {fd = 16, events = 1, revents = 0}}
#2  0x000055555555a615 in _eglutNativeEventLoop () at ../src/egl/eglut/wsi/wsi.c:70
#3  0x000055555555a226 in eglutMainLoop () at ../src/egl/eglut/eglut.c:269
        win = 0x555555629a70
#4  0x00005555555597fe in main (argc=1, argv=0x7fffffffe488) at ../src/egl/opengl/eglgears.c:319

I'm running: Gnome 45.3 Linux 6.6.10 Mesa 23.3.2 nVidia 545.29.06

This is admittedly a minor issue. Under normal circumstances (without __NV_PRIME_RENDER_OFFLOAD) these demos run just fine.

The upstream issue appears to be with one of nVidia/Wayland/Mutter, as it seems the window frame surface itself is not being created/presented

With WAYLAND_DEBUG enabled the last messages reported are:

wl_seat@23.capabilities(3)
 -> wl_seat@23.get_pointer(new id wl_pointer@18)
wl_seat@23.name("seat0")
xdg_toplevel@14.configure_bounds(3840, 2128)
xdg_toplevel@14.wm_capabilities(array[16])
xdg_toplevel@14.configure(0, 0, array[0])
xdg_surface@15.configure(35)
 -> xdg_toplevel@14.set_min_size(0, 0) 
 -> xdg_toplevel@14.set_max_size(0, 0) 
 -> wl_compositor@12.create_surface(new id wl_surface@25)
 -> wl_subcompositor@22.get_subsurface(new id wl_subsurface@26, wl_surface@25, wl_surface@10)
 -> wl_shm@17.create_pool(new id wl_shm_pool@27, fd 20, 484416)
 -> wl_shm_pool@27.create_buffer(new id wl_buffer@28, 0, 348, 348, 1392, 0)
 -> wl_shm_pool@27.destroy()
 -> wl_surface@25.attach(wl_buffer@28, 0, 0)
 -> wl_surface@25.set_buffer_scale(1)
 -> wl_surface@25.commit()
 -> wl_surface@25.damage_buffer(0, 0, 348, 348)
 -> wl_subsurface@26.set_position(-24, -24)
 -> wl_compositor@12.create_surface(new id wl_surface@29)
 -> wl_subcompositor@22.get_subsurface(new id wl_subsurface@30, wl_surface@29, wl_surface@10)
 -> wl_shm@17.create_pool(new id wl_shm_pool@31, fd 21, 44400)
 -> wl_shm_pool@31.create_buffer(new id wl_buffer@32, 0, 300, 37, 1200, 0)
 -> wl_shm_pool@31.destroy()
 -> wl_surface@29.attach(wl_buffer@32, 0, 0)
 -> wl_surface@29.set_buffer_scale(1)
 -> wl_surface@29.commit()
 -> wl_surface@29.damage_buffer(0, 0, 300, 37)
 -> wl_subsurface@30.set_position(0, -37)
 -> xdg_surface@15.set_window_geometry(0, -37, 300, 337)
 -> xdg_surface@15.ack_configure(35)
 -> wl_compositor@4.create_region(new id wl_region@33)
 -> wl_region@33.add(0, 0, 300, 300)
 -> wl_surface@10.set_opaque_region(wl_region@33)
 -> wl_region@33.destroy()

What seems to be missing from that xdg_surface.configure is everything from this point:

 -> wl_surface@10.frame(new id wl_callback@34)
 -> zwp_linux_dmabuf_v1@9.create_params(new id zwp_linux_buffer_params_v1@35)
 -> zwp_linux_buffer_params_v1@35.add(fd 20, 0, 0, 1216, 16777215, 4294967295)
 -> zwp_linux_buffer_params_v1@35.create_immed(new id wl_buffer@36, 300, 300, 808669784, 0)
 -> zwp_linux_buffer_params_v1@35.destroy()
 -> wl_surface@10.attach(wl_buffer@36, 0, 0)
 -> wl_surface@10.damage(0, 0, 2147483647, 2147483647)
 -> wl_surface@10.commit()

After which a normal execution would yield:

wl_display@1.delete_id(27)
wl_display@1.delete_id(31)
wl_display@1.delete_id(33)
wl_display@1.delete_id(35)
wl_buffer@28.release()
wl_buffer@32.release()
 -> wl_compositor@4.create_region(new id wl_region@35)
 -> wl_region@35.add(0, 0, 300, 300)
 -> wl_surface@10.set_opaque_region(wl_region@35)
 -> wl_region@35.destroy()
wl_display@1.delete_id(34)
wl_callback@34.done(672637)
 -> wl_surface@10.frame(new id wl_callback@34)
 -> zwp_linux_dmabuf_v1@9.create_params(new id zwp_linux_buffer_params_v1@33)
 -> zwp_linux_buffer_params_v1@33.add(fd 20, 0, 0, 1216, 16777215, 4294967295)
 -> zwp_linux_buffer_params_v1@33.create_immed(new id wl_buffer@31, 300, 300, 808669784, 0)
 -> zwp_linux_buffer_params_v1@33.destroy()
 -> wl_surface@10.attach(wl_buffer@31, 0, 0)
 -> wl_surface@10.damage(0, 0, 2147483647, 2147483647)
 -> wl_surface@10.commit()
xdg_toplevel@14.configure_bounds(3840, 2128)
xdg_toplevel@14.configure(300, 337, array[4])
xdg_surface@15.configure(37)
 -> xdg_toplevel@14.set_min_size(0, 0)
 -> xdg_toplevel@14.set_max_size(0, 0)
...
...

and so on

As for a solution, it's arguable that one is even necessary from these demos. Perhaps a timeout on the poll event?

My personal project does not appear to wait on any poll events, but without a window frame there isn't much left to do, since the resulting OpenGL context ends up with an undefined default framebuffer as a result.

I've attached the wayland debug logs in any case. wayland.noprime.txt wayland.prime.txt

I've confirmed this to affect the current release (mesa-demos-9) and also the current main (2e40dee9)

To upload designs, you'll need to enable LFS and have an admin enable hashed storage. More information