We rather confusingly still call a secondary display card that is
GPU-less (DisplayLink or other basic KMS device) a "secondary GPU",
so just because secondary_gpu_state is non-NULL doesn't mean we
can use it for rendering. The clearest indication of this is when
there is no EGL surface.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2341>
At first glance the `goto` looks like a loop, or potentially an infinite
loop. It's not a loop because the mode has changed at that point to
`META_SHARED_FRAMEBUFFER_COPY_MODE_PRIMARY`. But we can make it more
obvious and avoid the need for a goto.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2240>
There was a sanity check that complained if there was still a "next
framebuffer" when disposing an onscreen. This is correct to complain
about under normal operation, as we always wait until receiving the page
flip callback before cleaning up the onscreen and their state.
However, when there are many hotplugs occurring, we might end up with
race conditions when the above sanity check is not valid: when we have
more than one monitor active, paint 1 one of them, but receive a hotplug
event before we paint the other(s), we will discard the already painted
onscreen before really issuing a page flip.
In this situation, we will have the "next framebuffer", but having that
is not a bug, it's a race condition, thus to not leak in this situation,
make sure to clean up the next framebuffer here too.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/2081
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2225>
This significantly increases the chance of a fullscreen surface buffer
being scanned out instead of being painted via composition. This is
assuming the client supports the DMA buffer feedback Wayland protocol.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2146>
An incorrect assumption that after mode set there would be no pending
page flips was made. This meant that if there was a mode set, followed
by a page flip, if that page flip was for a CRTC on a now unused GPU,
we'd crash due to the renderer GPU data having already been freed. This
commit avoids that by keeping it alive as long as the page flips are
still in the air. It fixes crashes with backtraces such as
0) meta_render_device_get_egl_display (render_device=0x0)
at ../src/backends/native/meta-render-device.c:320
1) secondary_gpu_state_free (secondary_gpu_state=0x1c8cc30)
at ../src/backends/native/meta-onscreen-native.c:560
2) meta_onscreen_native_dispose (object=0x1cb65e0)
at ../src/backends/native/meta-onscreen-native.c:2168
3) g_object_unref (_object=<optimized out>)
at ../gobject/gobject.c:3540
4) g_object_unref (_object=0x1cb65e0)
at ../gobject/gobject.c:3470
5) clutter_stage_view_finalize (object=0x1cbb450)
at ../clutter/clutter/clutter-stage-view.c:1412
6) g_object_unref (_object=<optimized out>)
at ../gobject/gobject.c:3578
7) g_object_unref (_object=0x1cbb450)
at ../gobject/gobject.c:3470
8) meta_kms_page_flip_closure_free (closure=0x1d47e60)
at ../src/backends/native/meta-kms-page-flip.c:76
9) g_list_foreach (list=<optimized out>, func=0x7fb3ada67111 <meta_kms_page_flip_closure_free>, user_data=0x0)
at ../glib/glist.c:1090
10) g_list_free_full (list=0x1cb4d20 = {...}, free_func=<optimized out>)
at ../glib/glist.c:244
11) meta_kms_page_flip_data_unref (page_flip_data=0x1c65510)
at ../src/backends/native/meta-kms-page-flip.c:109
12) meta_kms_callback_data_free (callback_data=0x227ebf0)
at ../src/backends/native/meta-kms.c:372
13) flush_callbacks (kms=0x18e2630)
at ../src/backends/native/meta-kms.c:391
14) callback_idle (user_data=0x18e2630)
at ../src/backends/native/meta-kms.c
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2147>
It was dropping to zero after the first frame because it hadn't been
incremented high enough. So the second frame would crash with:
```
#0 g_type_check_instance_cast
#1 META_DRM_BUFFER
#2 copy_shared_framebuffer_cpu
```
That's the CPU-copy path (fallback-fallback) that probably no one is using
but it does work after this fix. Exactly the same issue as was fixed
in `copy_shared_framebuffer_primary_gpu` by 36352f44f9.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2104>
Systems with AMD GPUs do not take advantage of Mutter's zero-copy path
when driving DisplayLink screens. This is due to a very slow CPU access
to the zero-copy texture. Instead they fall back on primary GPU doing a
copy of the texture for fast CPU access. This commit accelerates texture
copy by working through damage regions only.
Tests on a 4K screen with windowed applications show significant
reduction of GPU utilisation.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/2033>
This replaces functionality that MetaRenderDevice and friends has
learned, e.g. buffer allocation, EGLDisplay creation, with the usage of
those helper objects. The main objective is to shrink
meta-renderer-native.c and by extension meta-onscreen-native.c, moving
its functionality into more isolated objects.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1854>
Scanout doesn't go through the usual path of compositing and doing
eglSwapBuffers, therefore it doesn't hit the timestamp query placed in
that path. Instead, get the timings by binding the scanout buffer to an
FBO and doing a timestamp query on the FBO.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1762>
The DRM buffers aren't really tied to mode setting, so they shouldn't
need to have an associated mode setting device. Now that we have a
device file level object that can fill this role, port over
MetaDrmBuffer and friends away from MetaKmsDevice to MetaDeviceFile.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1828>
When drmModePageFlip() or drmModeAtomicCommit() unexpectedly failed (e.g.
ENOSPC, which has been seen in the wild), this failure was not handled
very gracefully. The page flip listener for the scanout was left in the
MetaKmsUpdate, meaning when the primary plane composition was later page
flipped, two page flip listeners were added, one for the primary plane,
and one for the scanout. This caused the 'page-flipped' event to be
handled twice, the second time being fatal.
Handle this by making 'no-discard' listener flag be somewhat reversed,
and say 'drop-on-error', and then drop all 'drop-on-error' listeners
when a MetaKmsUpdate failed to be processed.
Also for a "preserve" flagged update, don't ever trigger "discard"
callbacks just yet, as preserved updates are used again for the primary
plane composition, in order to not miss e.g. CRTC gamma updates, or
cursor plane updates, which were added separately.
Closes: https://gitlab.gnome.org/GNOME/mutter/-/issues/1809
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1910>
We already swapped the front buffer, and even if it didn't get
presented, we should still swap our representation of the state, to not
get into a confused buffer tracking state.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1822>
On hybrid graphics system, the primary path used to transfer the stage
framebuffer onto the dedicated GPU's video memory preparing for scanout,
is using the dedicated GPU to glBlitFramebuffer() the content from the
iGPU texture onto the scanout buffer.
After we have done this, we reset the current EGL context back to the
one managed by cogl. What we failed to do, however, was to reset the
current EGL context when we inhibited the actual page flip due to having
entered power save mode.
When we later started to paint again, Cogl thought the current EGL
context was still the correct one, but in fact it was the one used for
the iGPU -> dGPU blit, causing various EGL surface errors, and as a side
effect, eventually hitting an assert.
Fix this by making sure we reset to the Cogl managed EGL context also
for this case.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1803>
Destroying the EGLSurface frees the underlying container structs. When
we call gbm_surface_release_buffer() with a gbm_surface the EGLSurface
was created from, doing that after the EGLSurface was destroyed results
in attempts to access freed memory. Fix this by releasing any buffer
first, followed by destroying the EGLSurface, and lastly, the
gbm_surface.
This was not a problem prior to CoglOnscreen turning into a GObject, as
in that case, the dispose-chain was not setup correctly, and the
EGLSurface destruction was done in the native backend implementation.
This also changes a g_return_if_fail() to a g_warn_if_fail(), as if we
hit the unexpected case, we still need to call up to the parent dispose
vfunc to not cause critical issues.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1803>
It's handled by CoglOnscreenEgl's dispose() implementation. It was
failed to be invoked in the past because the old non-GObject web of
vtables were not setup correctly, meaning the old generic EGL layer of
the CoglOnscreen de-init was never invoked.
When the type inheritence was cleaned up, this mistake was not cleaned
up, so do that now.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1803>
This eliminates the need for any render node or device nodes, thus can
be used without any graphics hardware available at all, or with a
graphics driver without any render node available.
The surfaceless mode currently requires EGL_KHR_no_config_context to
configure the initial EGL display.
This also means we can enable the native backend tests in CI, as it
should work without any additional privileges.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1698>
This concerns only the cases when the presentation timestamp is received
directly from the device (from KMS or from GLX). In the majority of
cases this timestamp is already MONOTONIC. When it isn't, after this
commit, the current value of the MONOTONIC clock is sampled instead.
The alternative is to store the clock id alongside the timestamp, with
possible values of MONOTONIC, REALTIME (from KMS) and GETTIMEOFDAY (from
GLX; this might be the same as REALTIME, I'm not sure), and then
"convert" the timestamp to MONOTONIC when needed. An example of such a
conversion was done in compositor.c (removed in this commit). It would
also be needed for the presentation-time Wayland protocol. However, it
seems that the vast majority of up-to-date systems are using MONOTONIC
anyway, making this effort not justified.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1484>
KMS and GLX device timestamps have microsecond precision, and whenever
we sample the time ourselves it's not the real presentation time anyway,
so nanosecond precision for that case is unnecessary.
The presentation timestamp in ClutterFrameInfo is in microseconds, too,
so this commit makes them have the same precision.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1484>
A flag indicating whether the presentation timestamp was provided by the
display hardware (rather than sampled in user space).
It will be used for the presentation-time Wayland protocol.
This is definitely the case for page_flip_handler(), and I'm assuming
this is also the case for the two instances in the GLX code.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1484>
Each next and current scanout buffer has a reference on them making sure
they stay alive. When dumb buffers were used on the secondary GPU state,
this didn't happen, leading to crashes due to unref:ing one time too
many, with backtraces such as
0) g_type_check_instance_is_fundamentally_a ()
1) g_object_unref ()
2) secondary_gpu_release_dumb ()
3) import_shared_framebuffer ()
4) update_secondary_gpu_state_post_swap_buffers ()
5) meta_onscreen_native_swap_buffers_with_damage ()
6) cogl_onscreen_swap_buffers_with_damage ()
7) swap_framebuffer ()
8) clutter_stage_cogl_redraw_view_primary ()
9) clutter_stage_cogl_redraw_view ()
10) _clutter_stage_window_redraw_view ()
11) handle_frame_clock_frame ()
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1746>
g_set_error_literal() asserts that the provided message is not NULL.
If it is NULL, the function is entirely no-op.
This resulted in a NULL dereference of the GError, which remained
NULL in this case, when trying to print a warning in
clutter_stage_cogl_redraw_view().
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1715>
Instead of calling "init_onscreen()" on two different separate vtables
from the allocate() funtion, just have the CoglOnscreen sub types
themself implement allocate() and initialize in there.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1514>
Thins means that e.g. MetaOnscreenNative now inherits CoglOnscreenEgl,
which inherits CoglOnscreen which inherits CoglFramebuffer, all being
the same GObject instance.
This makes it necessary to the one creating the onscreen to know what it
wants to create. For the X11 backend, the type of renderer (Xlib EGL or
GLX) determines the type, and for the native backend, it's currently
always MetaOnscreenNative.
The "winsys" vfunc entries related to onscreens hasn't been moved yet,
that will come later.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1514>
To get meta-renderer-native.c down to a bit more managable size, and to
isolate "onscreen" functionality from other (at least partly), move out
the things related to CoglOnscreen to meta-onscreen-native.[ch].
A couple of structs are moved to a new shared header file, as
abstracting those types (e.g. (primary, secondary) render devices) will
be dealt with later.
Part-of: <https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1514>
