Add dumb allocator

[UjinT34]

• Adds dumb allocator and buffer implementation. Mostly copied from gbm and kwin.
• Adds fallbackAllocator API

fallbackAllocator defaults to nullptr, returns dumb allocator for drm backend.

hyprwm/Hyprland#7040

[moetayuko]

I tested this PR and hyprwm/Hyprland#7040, they broke my hw cursor
hyprland.log

[hc] loadThemeStyle: loading for size 48
[hc] getShapesC: found 1 images for left_ptr
[TRACE] [AQ] atomic moveCursor
[TRACE] [AQ] CDRMOutput::scheduleFrame: reason 4, needsFrame true, isPageFlipPending true, frameEventScheduled true
[TRACE] [AQ] GBM: Allocating a buffer: size [Vector2D: x: 128, y: 128], format INVALID, cursor: true, multigpu: true, scanout: true
[LOG] [AQ] GBM: Automatically selected format AR24 for new GBM buffer
[LOG] [AQ] GBM: Buffer is marked as multigpu, forcing linear
[TRACE] [AQ] GBM: Using modifier-based allocation, modifiers: 1
[TRACE] [AQ] GBM: | mod 0x0
[ERR] [AQ] GBM: Allocating with modifiers and flags failed for cursor plane, falling back to dumb
[ERR] [AQ] Couldn't allocate a gbm buffer with size [Vector2D: x: 128, y: 128] and format INVALID
[TRACE] [AQ] GBM: Allocating a dumb buffer: size [Vector2D: x: 128, y: 128], format INVALID
[LOG] [AQ] GBM: Allocated a new dumb buffer with size [Vector2D: x: 128, y: 128] and format AR24
[TRACE] [AQ] GBM: Allocating a buffer: size [Vector2D: x: 128, y: 128], format INVALID, cursor: true, multigpu: true, scanout: true
[LOG] [AQ] GBM: Automatically selected format AR24 for new GBM buffer
[LOG] [AQ] GBM: Buffer is marked as multigpu, forcing linear
[TRACE] [AQ] GBM: Using modifier-based allocation, modifiers: 1
[TRACE] [AQ] GBM: | mod 0x0
[ERR] [AQ] GBM: Allocating with modifiers and flags failed for cursor plane, falling back to dumb
[ERR] [AQ] Couldn't allocate a gbm buffer with size [Vector2D: x: 128, y: 128] and format INVALID
[TRACE] [AQ] GBM: Allocating a dumb buffer: size [Vector2D: x: 128, y: 128], format INVALID
[LOG] [AQ] GBM: Allocated a new dumb buffer with size [Vector2D: x: 128, y: 128] and format AR24
[LOG] [AQ] Swapchain: Reconfigured a swapchain to [Vector2D: x: 128, y: 128] AR24 of length 2
[TRACE] Failed to create cursor RB with format 875713089, mod 0
[TRACE] [pointer] hw cursor failed rendering
[TRACE] [pointer] hw transformed hotspot for DP-1: [Vector2D: x: 10, y: 3]
[TRACE] [AQ] CDRMOutput::scheduleFrame: reason 2, needsFrame true, isPageFlipPending true, frameEventScheduled true
[TRACE] [AQ] CDRMOutput::scheduleFrame: reason 3, needsFrame true, isPageFlipPending true, frameEventScheduled true
[TRACE] [AQ] CDRMOutput::scheduleFrame: reason 3, needsFrame true, isPageFlipPending true, frameEventScheduled true
[TRACE] Output DP-1 rejected hardware cursors, falling back to sw

my cursor buffer is expected to be allocated by the modifier-less allocation path:

aquamarine/src/allocator/GBM.cpp

Lines 147 to 154 in 0ab8ffa

	if (!bo) {
	if (explicitModifiers.size() == 1 && explicitModifiers[0] == DRM_FORMAT_MOD_LINEAR) {
	flags |= GBM_BO_USE_LINEAR;
	allocator->backend->log(AQ_LOG_ERROR, "GBM: Allocating with modifiers failed, falling back to modifier-less allocation");
	} else
	allocator->backend->log(AQ_LOG_ERROR, "GBM: Allocating with modifiers failed, falling back to implicit");
	bo = gbm_bo_create(allocator->gbmDevice, attrs.size.x, attrs.size.y, attrs.format, flags);
	}

it's short-circuited by the dumb buffer path w/ this PR

[UjinT34]

You need either cursor:allow_dumb_copy = true or cursor:allow_slow_copy for this to work.
Logs [pointer] performing slow copy or [pointer] performing dumb copy.
If you have working hw cursors on main without cursor:allow_dumb_copy = true then you don't need this PR. And after required fixes gbm allocator will work as before.

[moetayuko]

You need either cursor:allow_dumb_copy = true or cursor:allow_slow_copy for this to work. Logs [pointer] performing slow copy or [pointer] performing dumb copy. If you have working hw cursors on main without cursor:allow_dumb_copy = true then you don't need this PR. And after required fixes gbm allocator will work as before.

I have working hw cursor on main. To avoid regression I guess the dumb buffer allocation should be moved after gbm_bo_create to avoid short-circuiting it?

[UjinT34]

No regression here. GBM allocator logic restored.
There is no point to move dumb allocator after implicit allocation. Implicit works on nvidia but it'll return an unsuitable buffer for cursor plane. That's why it allows non-renderable linear buffer just before the implicit. Might not be needed with GBM_BO_USE_LINEAR flag though. Didn't test it.

[vaxerski]

sorry for the delay, on a "break"

[vaxerski]

no, no, no. The dumb allocator doesn't really hold any big allocs, so it should be per-drm-backend (can be always present, no big deal)

[UjinT34]

Don't really understand what needs to be done here. How HL should get access to dumb allocator?

[vaxerski]

in CDRMBackend SP<CDumbAllocator>

[vaxerski]

then it can ->output->backend, cast that to drm and ->dumballocator