anongit.freedesktop.org has unfortunately been very flaky in our CI
pipelines and causing builds to fail. I am running my own mirror which
syncs from the official FreeDesktop GitLab instance to GitHub, which
should allow CI pipelines to continue working on the last available
release should the GitLab sync fail.
It doesn't really help that much, but since we are defining it in
some headers, we might as well do it everywhere so it actually has
some effect in case the headers were included in a different order.
gitlab.freedesktop.org has become unreliable over the last year,
to prevent the ci pipeline from failing, fall back to the more stable
and trusted anongit.
This commit makes `CDebug` use Unicode internally instead of whatever
random code page is in use. It also gets rid of the horrible character
counting and replaces that with `vasprintf` and `vaswprintf` helpers
(partially inspired by Linux) which allocates a buffer.
For HRESULT logging, the error code in both hex and decimal are included.
The output is now guaranteed to be UTF-8.
An event, `m_signal`, is created and signalled when either `m_running` or
`m_connected` is changed by another thread, so that the pipe thread knows
to interrupt the read.
The pipe is now opened as async to allow interruption, and the I/O
operations now use overlapped I/O.
Other changes include:
* Changing `m_pipe` to `HandleT<HANDLETraits>` since `CreateFile` returns
`INVALID_HANDLE_VALUE` instead of `NULL` on error.
* Remove the call to `WaitNamedPipeA` because it's useless and returns
immediately without waiting if the pipe doesn't exist.
The driver runs under the account `NT AUTHORITY\USER MODE DRIVERS`
and as a result requires the key to be owned by this user so that
it is able to write to it.
Under windows there is no cursor enabled at all until it has been moved
for the fist time by the user. If our cursor information is invalid we
assume that this has occured and induce a wiggle to force windows to
create the cursor.
* Removes the vector in favour of a two element array
* Removes iteration on the vector
* Create and track a per-image fence
* bind the EGLImage to the texture at creation, avoiding re-binding
* sync on GL_SYNC_FLUSH_COMMANDS_BIT and remove glFlush
Custom modes can now be configured via the registry under
HKEY_LOCAL_MACHINE\SOFTWARE\LookingGlass\IDD
Create the value "Modes" as a REG_MULTI_SZ with the value as
a list of modes, for example:
1024x768@60
1920x1080@60
1920x1080@120*
The '*' denotes the preferred mode to default to if one has not
been selected by the user.
If the screen is blanked or removed through a user action
`IddCxMonitorQueryHardwareCursor` will return with a failure. This
is normal and we should not log this as an error.
It is invalid to call `IddCxMonitorSetupHardwareCursor` before
`IddCxSwapChainSetDevice`. This fixes this by moving the thread into
CSwapChainProcessor and starting it after `IddCxSwapChainSetDevice`
has succeeded.
GL_TEXTURE_EXTERNAL_OES is not a valid target for glTexImage2D and there
is no need to "create" the textures as this is done during import by
eglCreateImage
Pipewire now automatically moves non-rt clients into non-rt
threads so client-rt.conf is obsolete. Stop loading it during
context initialization for Pipewire 1.3.81 and newer.
See: 24bcacc619
Starting with version 258, "systemd-udevd ignores OWNER=/GROUP=
settings with a non-system user/group specified in udev rules files".
Change the example udev rule to remove the explicit user assignment and
instead use the "uaccess" tag to auto-assign the logged-in user
permissions. As "uaccess" tags are processed using the 73-seat-late
udev rule, the KVMFR rule file name must precede it (i.e. should use
a lower ordinal value), so change the file name in the example to reflect that.
Credit goes to Discord user "w1kl4s" for bringing this systemd change
to our attention and assisting in testing.
The callback runs in a random thread, we can't call directx methods
safely from it, so move reset so it's called automatically when a free
copy list is obtained.
`IddCxSwapChainFinishedProcessingFrame` must be called after every
frame, but we should do it as early as possible once all commands
are queued that use the frame.
As the resource size can be larger then the actual frame data, we
need to track this seperately so that we don't waste cycles copying
data the client will never use.
As the IDD itself runs in a WUMDF sandbox, it doesn't have enough
access to perform interactive operations such as moving the cursor.
This helper service communicates with the IDD over a named pipe,
so that we can perform these things, as well as in the future provide
a configuration GUI.
As the IDD requires being built in MSVC2022 with the WDK and then
signed to create a usable driver, this must be done on the LG build
server instead of github directly. This workflow will trigger this on
each commit
The new configuration option `input:evdev` accepts a comma separated
list of `/dev/input/` keyboard devices to use for input when in capture
mode. This makes it possible to capture only a specific keyboard instead
of all keyboards.
If an overlay is closed with overlayMsg_close, the message can be freed
while it is still being used by msg_render, resulting in a segfault. Lock
the message list for the duration of msg_render to fix this.
The renderer contains an optimization to skip ImGui if there is nothing to
show, but the status overlay always returns a damage rect, even if it
didn't draw anything, so the optimization can never take effect.
While resuming from hibernation lgmpTimer may transition app.state to
APP_STATE_REINIT_LGMP. Sometimes the transition is lost if the app_main
processing loop also changes app.state simultaneously. This seems to
occur frequently with the hibernation use case.
Separate lgmp timer state into its own enum. Move resulting app.state
transitions to the app_main loop so they are serialized.
When the heapTest fails for DMA copies to shared memory, fallback to
conventional CPU copy via mapped resources. While this is less optimal,
it is still faster then the older DXGI capture backend.
As reported by @Crispy-fried-chicken in issue #1133 there is a potential
XXE vulnerability here. This fixes this problem by verifying the value
of `vmf->pgff` does not exceed the bounds of the memory mapping.
Fixes: #1133
virt-manager (or libvirt itself) tends to preserve outer quotation marks but explode inner double-quotes into '"' which is annoying.
this looks nicer when pasted into the XML editor and applied
Also used double-quotes for the QEMU <6.2 example for consistency
PS: added myself to AUTHORS since this is my first contribution
AMD GPUs and older NVidia GPUs can initialize fine but fail when we
start to create resources in the shared memory heap, we must test it
early to detect this so we can fallback to a working capture method.
Moving this prevented the RX580 crash early in init, but later
presents during the capture. As we want to ensure fallback to DXGI
if this happens we need to catch this fault during init. This moves
the order back so that we fail back into a working state.
Often this log is provided instead of the actual host log, as this
log is largely useless for debugging this moves it to the temp
directory out of view of the user.
This is still incomplete as d3d12 doesn't have any proper sync with the
captured frame and as such is still not suggested for general usage. This
change though is monumental for this project as it removes a full memory
copy reducing bandwidth consumption enormously.
This patch performs several changes.
* Alters the fourcc codes to types that ignore the alpha channel where
possible to allow the gpu to internally use 24-bit formats.
* Attempts to use DRM_FORMAT_RGB888 first as some GPUs may support this
* If DMABUF is not in use the data is now imported directly as RGB24
without the post-processing shader
Compiled in resources often will not contain a null terminator, as such
we must not use functions that rely on it. This implements a memsearch
function that performs like strstr on a buffer instead of a null
terminated string.
sort_words.py outputs a diff that if applied will sort words.txt
It has a few options:
- '--add-word': Add a word before sorting (usable multiple times)
- '--save': Save changes to file instead of only outputting a diff
- '--quiet': Don't output a diff
This allows us to build with libdecor enabled as the selection to use it
is decided upon at runtime if the compositor `gnome-shell` is detected.
If the libdecor development headers are installed, by default it will
now be compiled in unless overridden by the user at compile time.
For the moment, this just increases texture memory usage, but does not
affect behavior.
In a future commit, I will modify the shaders to not pack data across
rows, in order to enable damage copies.
This new sphinx extension runs html.unescape
(from the Python Standard Library) on source files before they are
rendered, allowing escape sequences like ' ' for the no-break
space character.
I have also published this extension in my own name under a different
license (the same one Sphinx uses) for others to use:
https://github.com/JJRcop/sphinxcontrib-html_unescape
In 3625207801 I attempted to add
non-breaking spaces to a filepath so it would stay on one line.
Before this I had accidentally deleted my work but found it saved
in my sphinx build cache, so I copied my changes from that cache.
Unfortunately the cached version replaced non-breaking spaces with
real spaces and 3625207801 was made
reverted.
This commit re-adds the non-breaking spaces.
The handle is only needed so we can open the resource, once we have
it we can close the handle. We then cache the shared resource for
future reuse if possible.
DXGI doesn't take into account the SDRWhiteLevel that has already been
applied to the monitor when it converts to HDR10 which results in clipping.
This change set implements a HLSL shader to reverse this while at the same
time converting to HDR10.
This is still not perfect but far better then doing nothing.
Dr.Memory on Windows complains bitterly about invalid heap free as
it doesn't seem to be able to track this function's allocations. As
it's such a trivial function we can just implement it locally.
This change alters the small square dot cursor to a more visible 16x16
cursor for X11. A new option `spice:largeDotCursor` can be set to use an
alternative 32x32 cursor for the vision impaired.
This commit adds an interface to the X11 display server code to support
various window manage specific features, such as in this case, the i3
global full screen toggle.
This feature specifically uses the i3 IPC to cause looking glass to go
full screen across all monitors if the new option `i3:globalFullScreen`
is enabled.
There is no performance penalty to using ARGB10 as it's still a 32-bit copy per
pixel and the nvidia driver internally handles any conversions required to make
this work for both SDR and HDR.
According to Erik @ NVidia the open source NVidia driver will not
create a EGLImage from a DMABUF if the target is not
GL_TEXTURE_EXTERNAL_OES. This change set converts the dmabuf texture
from GL_TEXTURE_2D to GL_TEXTURE_EXTERNAL_OES and at runtime performs a
global search & replace on fragment shaders as needed to remain
compatible, replacing `sampler2D` with `samplerExternalOES`.
Ref: https://github.com/NVIDIA/open-gpu-kernel-modules/discussions/243#discussioncomment-3283415
If the cursor was grabbed the window the cursor moves over when it is
ungrabbed will recieve an EnterNotify event with the mode of
NotifyUngrab, unfortunatly some window manages such as i3 will ignore
this message and as such focus follows mouse will not function
correctly. This patch injects a normal EnterNotify to work around this
issue.
This fixes a rounding issue on certain hardware (NVidia) which actually
implement mediump as half precision (16-bit) float. It's safe to assume
`highp` is available as if the GPU does not support it, then the shader
compiler will try to find a lower precision that is supported by the GPU
@quantum has observed nvfbc under rare circumstances fail to initialize,
this adds a retry to the init with a short delay to hopefully recover
from this situation.
This is rare and I'm not sure what causes it, but PipeWire sometimes uses a
larger period size than requested for no obvious reason (e.g., we could
request a period size of 512, but PipeWire uses 2048 anyway). This causes
us to stay in a permanent state of underrunning because the target latency
is too low.
With this change, we use the actual device period in the target latency
calculation if it is larger than the expected maximum. We may still get
some glitches at the beginning of playback (because the startup latency is
based upon the expected maximum period size), but it will recover after a
few seconds as it adjusts to the new target latency.
This fixes the following crash when attempt to toggle capture at the
right after closing the overlay:
zwp_pointer_constraints_v1@12: error 1: a pointer constraint with a wl_pointer of the same wl_seat is already on this surface
There is no action to be taken by the Looking Glass client, but a
handler needs to exist as certain other Wayland clients chooses to
send this message for copy-paste operations despite the fact it's
supposed to be used for drag-and-drop negotiation.
Previously, if the client's subscription to the frame buffer became
invalid for any reason, the video feed in OBS would freeze until the
user goes in and changes any of the settings. This commit allows the
plugin to automatically attempt to recover.
This commit makes it show a prettier error message when no display
server is available, including the display servers compiled. This
replaces the old assert.
Example output:
[E] 572277145932 main.c:1167 | lg_run | No display servers available, tried:
[E] 572277145934 main.c:1169 | lg_run | * Wayland
[E] 572277145935 main.c:1169 | lg_run | * X11
Certain window managers give us a scale before it gives us a size.
This commit makes the Wayland backend avoid passing a zero size to
wp_viewport_set_source, which is a protocol error.
With checkout v1, it checks out all the repository history, which takes
~20 seconds to complete with all the submodules that we have. With v2,
takes <10 seconds to complete the checkout by virtue of checking out
only the latest commit.
For some cursed reason, spell check starts failing if docs uses v2
checkout, but doesn't fail if it uses v1.
The Wayland display server has a graph for presentation times, but the
backend shuts down after the overlays, so calling
overlayGraph_unregister causes memory corruption. We can avoid this
problem by making unregister a noop after all graphs have been freed.
This is safe because updating the graph doesn't use the graph handle:
instead you update the ringbuffer which is owned by the user.
A failure to connect to spice would cause LG to exit late, this adds a
startup condition that prevents the LG initialization to complete until
the spice connection has been established.
This brings nanosvg into the project for SVG loading and rendering.
Unfortunatly we can not at this time use a submodule for this project
until https://github.com/memononen/nanosvg/pull/214 is merged.
Profiling shows that a considerable amount of time is spent in
glBindBuffer and glBufferSubData when the damage rects are updated.
Since the amount of data here is quite small it's far faster to check if
it's different then to just blindly overwrite the buffer on each call.
Profiled on an Intel CPU with UHD P630 Graphics using magic-trace
The prior commit to expose the FrameBuffer internals makes use of an
atomic from `stdatomic.h`. Unfortunatly C++ has no notion of _Atomic and
as such `stdatomic.h` is not compatible. To work around this we instead
just forward declare the type here.
This fixes an issue where the window position would be ignored if the
application was launched in full screen mode from the command line
causing the client to enter full screen on the wrong monitor in
multi-monitor configurations.
* Moved the LG license and version onto a seperate tab.
* Added general donation section and link to the website donation page
* Removed donation details under gnif's section
This makes it possible to define the escape key by name rather then just
it's integer code, while still allowing fallback to using an integer
value for codes that may not be defined.
Example: `input:escapeKey=KEY_F1`
An invalid string value will also print a list of all valid string
values.
This is an experimental & incomplete feature for those using
supersampling. Anything > 1200p will be downsampled by 50% before
copying out of the GPU to save on memory bandwidth.
Unfinished! Has issues with damage tracking and currently can not
be configured. Only dx11 has been tested at this point, everything
else will likely have problems/crash.
Pipewire documents the mute parameter as a bool, however `pw_stream_set_control` expects a float value and converts it to a bool.
6ad6300ec6/src/pipewire/stream.c (L2063)
The state is never updated when a message box is dismissed, so the
cursor is never displayed when a second message box shows up.
The only other caller, app_setOverlay, has state tracking already.
The nsleep() call lets d3d12 sleep for a more precise amount of
time while maintaining the current millisecond-scale sleep
interface in the configuration file.
Under Wayland, if the mouse pointer is disconnected whilst captured
(like say via KVM switch), the waylandWarpPointer code will be called
but the pointer will be NULL. This results in the cryptic message:
error marshalling arguments for confine_pointer (signature noo?ou): null value passed for arg 2
Error marshalling request: Invalid argument
This patch adds a check on the wlWm.pointer pointer before attempting
to warp the pointer, and avoids the crash.
The current minimum target latency is partially based upon the default qemu
behaviour whereby audio packets are delivered in a sawtooth pattern, with
packet timestamps drifting between 5ms above and below the measured clock.
This 5ms error is baked into the minimum target latency to avoid
underrunning.
This sawtooth pattern can be reduced by specifying a lower timer period in
the qemu configuration, so remove it from the hardcoded minimum latency and
add it to the default configurable buffer latency instead. This allows
users that have configured their VM appropriately to reduce the overall
latency.
The best quality resampler has an intrinsic latency of about 3ms, and the
processing itself takes another 1-2ms per 10ms block. The faster setting
has an intrinsic latency of about 0.4ms, with about 0.04ms processing time.
This makes for an overall saving of about 4ms, with negligible loss in
quality.
This adds a new `audio:bufferLatency` option which allows the user to
adjust the amount of buffering LG does over the absolute bare minimum. By
default, this is set large enough to absorb typical timing jitter from
Spice. Users may reduce this if they care more about latency than audio
quality.
This adds a new `audio:periodSize` option which defaults to 2048 frames.
For PipeWire, this controls the `PIPEWIRE_LATENCY` value. For PulseAudio,
the controls the target buffer length (`tlength`) value.
If the audio device starts earlier than required, we slew the read pointer
backwards to avoid underrunning. We need to apply this same offset to the
recorded device position, otherwise the Spice thread will think playback is
further ahead than it really is and inject unnecessary latency to
compensate.
When the 'keep alive' playback times out, playback is stopped from the
audio callback, resulting in an assertion failure inside PulseAudio as we
try to lock the main loop thread while already inside it.
The actual time between opening the device and the device starting to pull
data can range anywhere between nearly instant and hundreds of
milliseconds. To minimise startup latency, open the device as soon as the
first playback data is received from Spice. If the device starts earlier
than required, insert a period of silence at the beginning of playback to
avoid underrunning. If it starts later, just accept the higher latency and
let the adaptive resampling deal with it.
We can set the startup latency for the next playback far more precisely if
we have the device open already.
Only keep the device open with no playback for 30 seconds to avoid keeping
the device open unnecessarily forever.
Underruns can still happen quite easily at the beginning of playback,
particularly at very low latency settings. Further increase the startup
latency to avoid this.
Many X11 window managers will present an application on their
taskbar as a combination of the application name and an icon
imagery pulled from the X-Property _NET_WM_ICON. Applications
built under frameworks such as Qt or GTK have this property
populated by the framework. This commit adds the Atom _NET_WM_ICON
and populates it with a 64x64 icon of Looking Glass.
The desktop doesn't need its own sampler, there is already an identically
configured one in the `desktop->texture`.
For some reason, using the texture sampler fixes a black screen issue
with my GTX 660 using the 470.86 driver. Maybe hitting some limit
for how many samplers can be allocated?
PipeWire startup latency varies wildly depending on what else is, or was
last using the audio device. In the worst case, PipeWire can request two
full buffers within a very short period of time immediately at the start of
playback, so make sure we've got enough data in the buffer to support this.
The target latency is now based upon the device maximum period size
(which may be configured by setting the `PIPEWIRE_LATENCY` environment
variable if using PipeWire), with some allowance for timing jitter from
Spice and the audio device.
PipeWire can change the period size dynamically at any time which must be
taken into account when selecting the target latency to avoid underruns
when the period size is increased. This is explained in detail within the
commit body.
The recent `pwnkit` exploit brought this to my attention, not that we
are a setuid process we should still do this properly... who knows where
this code might get used in the future.
Previously this was hardcoded to 100ms which is far too high in most
instances, instead we get the initial period size and use whichever is
greater out of 50ms or the period size.
The idea is to reduce the amount of time it takes for the latency to
come down after initial stream start.
This removes the need for locking while also giving a better result in
the graph output. Also when the graph is disabled via the overlay
options it will no longer cause redraws.
This change allows the audiodevs to return the minimum period frames
needed to start playback instead of having to rely on a pull to obtain
these details.
Additionally we are using this information to select an initial start
latency as well as to train the desired latency in order to keep it as
low as possible.
This change is based on the techniques described in [1] and [2].
The input audio stream from Spice is not synchronised to the audio playback
device. While the input and output may be both nominally running at 48 kHz,
when compared against each other, they will differ by a tiny fraction of a
percent. Given enough time (typically on the order of a few hours), this
will result in the ring buffer becoming completely full or completely
empty. It will stay in this state permanently, periodically resulting in
glitches as the buffer repeatedly underruns or overruns.
To address this, adjust the speed of the received data to match the rate at
which it is being consumed by the audio device. This will result in a
slight pitch shift, but the changes should be small and smooth enough that
this is unnoticeable to the user.
The process works roughly as follows:
1. Every time audio data is received from Spice, or consumed by the audio
device, sample the current time. These are fed into a pair of delay
locked loops to produce smoothed approximations of the two clocks.
2. Compute the difference between the two clocks and compare this against
the target latency to produce an error value. This error value will be
quite stable during normal operation, but can change quite rapidly due
to external factors, particularly at the start of playback. To smooth
out any sudden changes in playback speed, which would be noticeable to
the user, this value is also filtered through another delay locked loop.
3. Feed this error value into a PI controller to produce a ratio value.
This is the target playback speed in order to bring the error value
towards zero.
4. Resample the input audio using the computed ratio to apply the speed
change. The output of the resampler is what is ultimately inserted into
the ring buffer for consumption by the audio device.
Since this process targets a specific latency value, rather than simply
trying to rate match the input and output, it also has the effect of
'correcting' latency issues. If a high latency application (such as a media
player) is already running, the time between requesting the start of
playback and the audio device actually starting to consume samples can be
very high, easily in the hundreds of milliseconds. The changes here will
automatically adjust the playback speed over the course of a few minutes to
bring the latency back down to the target value.
[1] https://kokkinizita.linuxaudio.org/papers/adapt-resamp.pdf
[2] https://kokkinizita.linuxaudio.org/papers/usingdll.pdf
In unbounded mode, the read and write pointers are free to move
independently of one another. This is useful where the input and output
streams are progressing at the same rate on average, and we want to keep
the latency stable in the event than an underrun or overrun occurs.
If an underrun occurs (i.e., there is not enough data in the buffer to
satisfy a read request), the missing values with be filled with zeros. When
the writer catches up, the same number of values will be skipped from the
input.
If an overrun occurs (i.e., there is not enough free space in the buffer to
satisfy a write request), excess values will be discarded. When the reader
catches up, the same number of values will be zeroed in the output.
Unbounded mode is currently unused since our audio input and output
streams are not synchronised. This will be implemented in a later commit.
Also reimplemented as a lock-free queue which is safer for use in audio
device callbacks.
If a new playback is started while the previous playback is still flushing,
we simply allow the stream to continue playing and effectively cancel the
flush. In general this is not safe because there may not be enough data in
the buffer to avoid underrunning. We could handle this better later by
trying to insert the right number of silent samples into the buffer, but
for now just completely stop the previous stream before starting the new
one.
Automatically restarting playback once draining has completed could result
in playback starting too early (i.e., before there is enough data in the
ring buffer to avoid underrunning). `audio_playbackData` will keep invoking
`start` until it returns true anyway, so we can just allow draining to
complete normally and wait for `start` to be called again.
To avoid client showing "Using : NVFBC (NVidia Frame Buffer Capt".
This happens because the string is truncated to 31 characters to fit
in the char capture[32]; member of KVMFRRecord_VMInfo.
For Windows 10, it so happens that the major.minor is 10.0. This is not
usually a given, e.g. on Windows 7 where it would read 6.1, on
Windows 8 it would read 6.2, and on Windows 8.1 it would read 6.3.
This is obviously undesirable, so we should just read the ProductName
from registry if possible. This results in something like:
OS Name: Windows 10 Pro for Workstations (Build: 19043)
This will cache up to 10 handles, in practice I have never seen DXGI
return anything but the same resource each time but we allow for more
anyway should MS change something in the future.
Should the cache get over filled it is disabled entirely and we revert
to the original behaviour.
This is an MIT-licensed header from Microsoft, which contains the
Direct3D 12 debug layer.
This header is slightly modified to be able to compile on older
MinGW versions.
These are implemented as ScrollLock+Up/Down for volume up and down, and
ScrollLock+M to toggle audio mute. These should prove useful especially
when Looking Glass now supports streaming audio, and the volume is
defined in the guest and set on the output stream.
This prevents LGMP_ERR_QUEUE_FULL from happening with high polling rate
mice, which is caused by receiving many more mouse events while the
guest cursor warps, triggering more warps.
On my machine (Intel UHD Graphics 770), texture processing occasionally
(about 5% of the time) takes more than 20ms (the highest I have seen is
around 32ms) when the host resolution is 2560x1440. This results in the
frame being discarded and the client displays a stale image. Increase the
timeout to 40ms.
The buffer input sizes to the `IDXGIOutputDuplication` methods are measured
in bytes. This dramatically increases the number of dirty/move rects that
can be handled.
Now LG uses a 25Hz tick timer it is an issue that `create_timer` spawns
a new thread for every single timer event, so instead multiplex all the
timers into a single thread with a 1ms resolution.
This prevents severe buffer underruns if the PipeWire quantum is bigger
than the ring buffer size. This could happen if a media player is running
at the same time as Looking Glass if it requests a very large quantum size,
for example.
This stops the end of the playback from being truncated. It also prevents
an audible glitch when playback next starts due to the truncated data being
left behind in the ring buffer.
When using jitRender, or on the first frame of an alert the window
doesn't get resized immediately causing it to cut off the end of the
text.
ImGui needs two passes to calulate the bounding box for automatically
sized windows, this is per it's design and not a bug, see:
https://github.com/ocornut/imgui/issues/2158#issuecomment-434223618
This new feature while helps on some systems, others using freesync or
higher refresh rates where the capture can't keep up will limit to
fractions of the refresh rate. Better to disable and allow users to
opt-in.
This change reduces the host GPU and CPU load by a large margin
improving guest system performance along with removing latency spikes
when moving the mouse. This is default enabled but can be disabled with
the new option `dxgi:dwmFlush=no` as it limits the capture rate to the
refresh rate of the guests output which may not be desireable.
If the guest supports sending us it's UUID and PureSpice has also
reported the guest's UUID, check them to see if the user has
accidentially connected to the wrong spice socket.
This change allows the host to provide information to the client about
how the VM is configured, information such as the UUID, CPU
configuration and capture method both for informational display in the
client as well as debugging in the client's logs.
The format of the records allows this to be extended later with new
record types without needing to bump the KVMFR version.
If there is LGMP corruption the LGMP thread will set the state to
REINIT which if this happens early enough will get overwritten if the
inital app state is set too late. Instead set the application initial
state early to avoid this.
As the ringbuffer is now in use for audio it makes sense to provide bulk
append and consume functions that are thread safe instead of adding
locking over all of the functions. This partially reverts the prior
commit that added the extra locking.
g_state.posInfoValid could become valid after the guest reports the
cursor position, in which case we did not show the cursor until another
update occurs.
This commit eliminates the race by performing the update when
g_state.posInfoValid becomes true.
Latest versions of QEMU are running out of PCIe mapping
through libvirt as the commandline device is mapping
ivshmem before other root devices.
Use JSON style configuration in the commandline block.
JSON arguments are loaded after all regular ones
and will guarantee to get loaded after all other libvirt mapped
PCIe devices.
For an unknwon reason when LG is on another desktop (hidden) and the
user switches to that desktop, the first attempt to grab the pointer
results in a GrabFrozen result. This adds some simple retry logic to
attempt again after a short (100ms) delay which seems to resolve the
issue.
After many months of hard work from our team, I am pleased to announce Beta 5 Release Candidate 1 which brings with it a huge number of improvements.
If you like this project and find it useful and would like to help out you can support this project directly by using the following platforms.
* [GitHub](https://github.com/sponsors/gnif)
* [Ko-Fi](https://ko-fi.com/lookingglass)
* [Patreon](https://www.patreon.com/gnif)
* [Paypal](https://www.paypal.com/cgi-bin/webscr?cmd=_s-xclick&hosted_button_id=ESQ72XUPGKXRY)
* BTC - 14ZFcYjsKPiVreHqcaekvHGL846u3ZuT13
* ETH - 0x6f8aEe454384122bF9ed28f025FBCe2Bce98db85
Changelog:
**B4-rc1**
* removed SDL displayserver
* reduced amount of GPU buffer allocations in DMABUF import path
* implemented frame damage tracking, used to reduce amount of data copies needed
* added frame damage display mode in EGL backend
* added cimgui/imgui to the project
* added frame timings collection and display
* improved LGMP cursor performance by releasing the memory faster
* implemented eglSwapBuffersWithDamage for X11
* don't allow the mouse to trigger redraws if the video feed is disabled
* New "Overlay Mode" which allows interaction with imgui windows/widgets
* Allow the FPS display to be moved
* Observe XDG_CONFIG_DIR when looking for the configuration
* Added EGL texture import timing graph
* Alerts are now rendered using ImGui
* Added high DPI support
* add JIT rendering mode for wayland
* improve fractional scaling for wayland
* made numbering of IVSHMEM devices consistent
* Added guest cursor warp support to the KVMFR protocol
* Added the ability to have post-processing filters
* Added AMD FidelityFX CAS post-processor
* Added AMD FidelityFX FSR upscaling post-processor
* add DMABUF import support in OBS plugin
* add keyboard LED synchronization
* fix EGL backend not being detected on some systems (e.g. mesa without libglvnd) (edited)
* fix cursor blinking issue on certain hardware (e.g. old Intel integrated graphics)
* Guest cursor is now aligned to the local cursor on window entry using guest cursor warp (Fixes mouse acceleration issues)
* add ability to reorder post-processing filters
* add ability to load and save post-processing filter presets.
* add ability to capture via xdg-desktop-portal+PipeWire on Linux
* windows host installers are now distributed as 64-bit executables
If a new client connects but there has not been a capture timeout the
new client count wont be zeroed on a valid capture. If there is a
timeout later the host will still think there is a new client and
re-send a frame when it should not. This fixes this by reading the value
on a valid frame which zeros the new subs count.
This allows the client to work when the OpenGL implementation fails if
EGL_RENDER_BUFFER is passed, printing a warning. This should fix issues
with Nvidia proprietary drivers on Wayland.
NVIDIA still do not implement a complete/working DMABUF implementation
yet advertise support. Best to tell the public to complain to NVIDIA
instead of assuming it's a LG bug or an issue with their system.
filters are now not run if `egl_filterSetup` returns false, as such we
do not need additional `enable` checks in `prepare` or `run` and we can
bypass the filters even earlier if they are not enabled reducing load.
Since this->prepared will never be set to true unless the filter is
enabled, this results in the framebuffer setup being done every frame
for no reason, causing a lot of texture reallocations.
The cursorThread prevents the host from going to sleep when the
video feed is disabled as it's subscribed to the cursor queue. Stopping
the cursorThread will unsubscribe from the queue and allow the host
application to disable capture.
This prevents us from exceeding the minimum GL/EGL versions that we are
targetting or forced to use indirection for due to lack of guarantee in
the ABI.
This commit adds check for the extensions that we need and then calls
the functions indirectly through gl_dynprocs.
This should improve compatibility with older versions of OpenGL, as we
now fallback to the ARB extensions if possible, and in the case of
glGenerateMipmap, we can handle the function not existing at all.
The Linux OpenGL ABI does not guarantee that glXSwapIntervalEXT will be
exported statically from any library, and indeed on some systems this
function does not link at load time, e.g. with amdgpu-pro. All other
GLX functions that we use are from GLX 1.0, which is guaranteed to be
exported statically.
This commit solves this issue by using glXGetProcAddressARB to load the
function. Note that only the ARB version of glXGetProcAddress is
guaranteed to exist by the Linux OpenGL ABI, which is why we must use
it.
If the wait times out, we used to simply restart the loop, which causes
it to check this->stop and exit if set to true. However, nvfbc_stop
already calls lgSignalEvent, which would wake up the pointer thread to
perform the check, so there is no need to set a timeout on the wait.
Sometimes, e.g. when xdg-open has to start the browser, the xdg-open
process can stay around until the browser exits, which freezes the
client. Instead, we should not wait for xdg-open to exit.
However, we can't simply not call wait, as that would leave the
xdg-open process around as a zombie. We could turn off the SIGCHLD
handler, but that's a global solution to a local problem. Instead, we
call setsid and fork again to detach the xdg-open process as if it's a
daemon, and let init take care of the reaping process.
Co-Authored-By: Tudor Brindus <me@tbrindus.ca>
The host is a 64-bit application, so requiring 64-bit Windows isn't an
issue. However, not using 32-bit installers has an advantage: we don't
need to require WoW64.
This prevents attempts to grab the pointer after the guest side warp
finishes if the pointer has left the window in the meantime. On Wayland,
this would result in the pointer moving to the middle of the window when
the confine is created.
Previously, all progress made during sleep is reset, so if the thread keeps
getting interrupted before the sleep finishes, the sleep will never complete.
This saves a lot of GPU power for partial updates. Running testufo with
lanczos downscaling and FSR upscaling consumed over 90 W, but with this
commit, consumed only 75 W.
The old code will not correctly report the number of threads on CPUs with
more than one processor group, i.e. when there are more than 64 logical
processors (threads).
The translucent white modal background sort of cancels out the dark
background we apply to the overlay, which is undesirable. It should
instead further darken the background.
For consistency, we now use igGetColorU32Col(ImGuiCol_ModalWindowDimBg)
to draw the overlay background, to avoid hardcoding the same colour in
multiple places.
Currently, this is visible through how fast the cursor blinks, with it
blinking faster at higher refresh rates. This commit makes the timing
consistent.
This new function dumps all options marked as preset instead of dumping
individual sections. This should allow filter options to not be all grouped
into the [eglFilter] section.
This makes vector_push return a pointer to the pushed element.
It also allows the user to push a NULL pointer, which means allocating the
memory for the element but do not copy anything into it.
It works, with the following limitations:
1. user is forced to select the monitor through platform-specific mechanisms
every time the client starts.
2. cursor is composed onto the screen, and no position can be reported.
imgui really hates it when we update the modifier key state after igNewFrame.
The result is:
void ImGui::ErrorCheckEndFrameSanityChecks(): Assertion
`(key_mod_flags == 0 || g.IO.KeyMods == key_mod_flags) &&
"Mismatching io.KeyCtrl/io.KeyShift/io.KeyAlt/io.KeySuper vs io.KeyMods"'
failed.
Therefore, we buffer the modifier state information and update it in the IO
object right before we call igNewFrame.
This avoids warping the host cursor when the guest-side warp has not finished,
which will result in the host cursor exiting at the wrong position if it exits
at that moment.
Using 4x4 means that some pixels will be outside of the lanczos window. The
ideal lanczos function should in fact be zero in those areas, so we shouldn't
waste time processing those pixels.
I can't notice any difference in the results.
The signature for calloc is void *calloc(size_t num, size_t size), where num
is the number of elements to allocate, and size is the size. Therefore, to
allocate a single struct, we should pass 1 for num and the size of the struct
as size.
In some places, we use the opposite order, and we should flip it.
According to the documentation for eglQueryString:
> EGL_BAD_DISPLAY is generated if display is not an EGL display connection,
> unless display is EGL_NO_DISPLAY and name is EGL_EXTENSIONS.
Therefore, we should check EGL by doing:
eglQueryString(EGL_NO_DISPLAY, EGL_EXTENSIONS)
Indeed, the old way of eglQueryString(EGL_NO_DISPLAY, EGL_VERSION) works on
libglvnd but not using mesa's libEGL.so directly.
Also added a warning to make it more obvious that EGL is not available.
With the new keymap feature, we are now able to properly support letting
the user enter exact values into the sliders. This commit adds a tooltip
to help the user discover this feature.
Note that this currently only works on Wayland. The X11 backend will need
to call app_handleKeyboardModifiers.
Due to the way assert is defined in standard C, compilers in release mode
will not treat it as unreachable. This explains a lot about those pesky
uninitialized variable bugs, actually.
This lets us mark code as unreachable and signals the compiler that this
is the case with __builtin_unreachable().
We also mark DEBUG_FATAL as unreachable.
This, unlike the standard assert macro, is guaranteed to print the failed
assertion to our log file, and tests the assertion even with NDEBUG defined
so we can more easily catch failures in production binaries without crashing
the program.
The motivation of this is how MinGW handles assertion failures: it creates a
dialog window that the headless user will not be able to see, and blocks the
program from being restarted by the service. Since the failed assertion is
displayed in the dialog, it doesn't print anything to the log, making it
impossible to diagnose issues.
Before, if the size is exactly the multiple of the page size, an extra padding
page is added for no reason. This commit fixes the logic and also uses the
page size obtained dynamically.
This commit makes the crash handler show relative paths instead of absolute
ones, which makes the stack traces generated easier to read.
On the other hand, absolute paths makes sense on Linux, since the user is
expected to build the binaries themselves, and gdb will be able to find the
source code.
This also displays a tooltip to explain that quality can be changed by
altering guest resolution and also show the resolutions needed to achieve
each quality mode.
This commit adds damage tracking to the DXGI textures, and only copies the
damaged areas to the textures with ID3D11DeviceContext::CopySubresourceRegion.
The sleep logic in waitFrame makes it difficult for this to reduce the
latency, but removing it shows significant improvements (6-7 ms to ~3 ms)
when a tiny portion of the screen is damaged, while showing no difference on
full screen damage.
This prevents issues like obscure characters getting transformed in symbol
names, resulting in an endless game of whack-a-mole finding symbols that are
replaced, such as 58964ce317.
gl_Position is expected to be using homogeneous coordinates, which requires
w to be a coordinate scale factor, usually 1.0. z should also be set in order
for depth to be well-defined. Therefore, we should set gl_Position.zw to
vec2(0.0, 1.0).
This avoids race conditions in GL drivers when attempting to render and
call glEGLImageTargetTexture2DOES on the same texture.
Also, when using glEGLImageTargetTexture2DOES, we do not need to allocate
storage for textures.
External events like launching other applications can cause latency
spikes while X11 initializes the application, we should only start
adjusting our delay if we see excessive skips over a 1s period.
`process` and `bind` are called from the same thread in order, there is
no need for atomic usage here.
This reverts commit 3d7dbd6371.
This reverts commit b3db1ba10b.
It used to be the case that we overwrite this->sync even if it was non-zero
when updating the texture, without deleting the sync object. If we update
faster than we render, the result would be leaking sync objects.
This commit ensures that sync objects are deleted when they are replaced.
Invalidating the entire window on an Expose event causes poor WM
performance when dragging the window around. Instead flag to redraw and
wait for the expose events to stop for 100ms before doing it.
This uses the same line sweep algorithm originally created to copy DXGI
textures to IVSHMEM to implement the copy from IVSHMEM to memory-mapped
pixel buffer objects.
This implementation uses a line sweep algorithm to copy the precisely the
intersection of all accumulated damage rectangles, ensuring that every
pixel is copied exactly once, and no pixel is ever copied multiple times.
Furthermore, once a row has been swept, we update the framebuffer write
pointer immediately.
It looks really weird having a separator right after a sentence ending in :.
A separator makes the list look detached from the paragraph that introduces
it, which looks awkward. Instead, this commit moves the separator before the
introducing paragraph.
Also added logic to properly pluralize the sentence.
Certain drivers do not support pitches that are not multiples of 128 bytes,
and instead just does some kind of rounding internally. On DXGI, this is not
a problem because the API rounds pixel pitch, but NvFBC does not. This causes
certain resolutions to simply not work with dmabuf, most notably 3440x1440,
which is 1440p ultrawide.
Since we are copying pixels with the CPU anyways, we might as well round the
pitch up to 128 bytes (32 pixels).
This is currently only implemented for Linux.
On Windows, ShellExecute should be used, but that should be done when it's
actually needed so it could be tested.
This commit adds a new host configuration option, nvfbc:diffRes, which
specifies the dimensions of every block in the diff map. This defaults to
128, meaning the default 128x128 block size.
Since block sizes other than 128x128 is not guaranteed to be supported by
NvFBC, the function NvFBCGetDiffMapBlockSize was introduced to query the
support and output the actual block size used.
The display servers and renderers may want to register their own
overlays in the future, as such we need g_state.overlays to be
initialized to allow for this.
Since we only update imgui's cursor location when the overlay is
enabled, if the last cursor position was showing a shape that is
incorrect when we re-enter the overlay the cursor will be wrong. This
corrects this by updating the location as we enter overlay mode.
This adds a new `earlyInit` call which allows the overlay to register
options before actually being intialized. Also the keybind handling and
state tracking for each overlay has been moved internal to the overlay
itself.
When entering overlay mode if the cursor was previously grabbed we
should restore the state when exiting overlay mode. This will also
correct the pointer setting it to NONE or SQUARE depending on the prior
grab state.
X11 needs to calibrate to get the best possible latency, as such it
needs the scene to render so that the render time of the scene can be
accounted for in the delay calculation.
This replaces the scaled `destRect` with a version that uses doubles
correcting the rounding error that is causing a failure to properly
clear the black bar areas.
This mesh will later be used to render only damaged portions of the desktop.
We also moved the coordinate transformation for damage overlay into a matrix
and computed by the shader.
XPresent doesn't give us the time before presentation, but the time just
after. This code calculates and calibrates a delay to sleep for before
signaling the wait event for render when using jitRender
After the damage queue PR, EGL damage count 0 means no change, and -1 means
invalidate the entire window. However, several other places have different
semantics, and we are not handling them correctly:
1. KVMFR uses 0 to signal invalidating the entire frame, so if we receive 0
rectangles in egl_on_frame, we should set damage count to -1.
2. The damage overlay treated 0 as full damage, which is now incorrect. This
is fixed, and now it treats 0 as no update, and -1 as full damage.
The way things were handled in EGLTexture is not only very hard to
follow, but broken. This change set breaks up EGLTexture into a modular
design making it easier to implement the various versions.
Note that DMABUF is currently broken and needs to be re-implemented.
There used to be a possible race when a bunch of rectangle is appended, but
the total count is not updated before it's read. Using a lock eliminates
all such races.
Without configuring Wayland compositors to send frame callbacks as late as
possible, JIT rendering can increase latency by more than one frame.
For example, by default, sway asks applications to render right after a
vblank, and does its own composition right after a vblank, resulting in
~2 frame's worth of latency. If max_render_time is set on the output,
it composes that many milliseconds before the vblank, losing ~1 frame's
worth of latency. If max_render_time is set on the window also, the frame
callback is sent that many milliseconds before composition, and we achieve
perfectly low latency.
Therefore, out of the box, JIT rendering should not be enabled, as manual
compositor configuration is required for optimal results.
For reference, the following sway settings results in the best latency:
output <insert output name> max_render_time 1
for_window [app_id="looking-glass-client"] max_render_time 1
This reverts commit 3baed05728.
If we invalidate the window, we used to not update this->cursorLast, and
this causes us to lose track of the cursor. Now we update this->cursorLast
unconditionally, and this fixes the issue.
Version 3 does not send xdg_output.done events, instead guaranteeing that
all xdg_output.* events are sent before wl_output.done. This saves us from
doing the work twice.
The method used is not guaranteed to work on all Wayland compositors,
so offer a way out. We need to support it anyways in case xdg_output
or wp_viewporter protocols are not available.
Currently, we scale the desktop up to the next largest integer, and rely on
the wayland compositor to scale it back down to the correct size.
This is obviously undesirable.
In this commit, we attempt to detect the actual fractional scaling by finding
the current active mode in wl_output, and dividing it by the logical screen
size reported by xdg_output, taking into consideration screen rotation.
We then use wp_viewporter to set the exact buffer and viewport sizes if
fractional scaling is needed.
When requested, JIT render mode will be used if the display server supports it.
Otherwise, a warning is generated instead.
This essentially uses the signalNextFrame logic for imgui, but for everything.
We automatically enable this mode when overlay is on.
Currently, this exposes some damage tracking bugs in the EGL renderer.
This prevents damage from being overwritten when frames are received
faster than could be rendered.
This implementation cycles between two queues, removing all need for
memory allocation.
This method takes an LGEvent and signals it when the next frame should be
rendered in time for the next vblank.
We will be using this to render imgui at screen refresh rate, but this could
potentially be used later to implement a better form of vsync for supported
display servers.
This must be invoked before swapping buffers.
The imgui overlay requires input even if the display is not captured and
operating in raw mode. XInput2 correctly only sends
XI_Press/ReleaseButton events if the device has not been captured, as
such it's safe to handle both raw and non raw buttons events at the same
time.
We look for the client config in $XDG_CONFIG_HOME/looking-glass/client.ini.
This is done because it's more conventional, and also allows us to add
additional configuration files, e.g. for the host.
We fallback to $HOME/.config as is standard, and then as a last resort use
getpwuid(getuid())->pw_dir. This is also recommended by the getpwuid manpage:
> An application that wants to determine its user's home directory should
> inspect the value of HOME (rather than the value getpwuid(getuid())->pw_dir)
> since this allows the user to modify their notion of "the home directory"
> during a login session.
Currently, we load /etc/looking-glass-client.ini and/or
~/.config/looking-glass-client.ini as long as they exist, even if they are
not files. We should only load them if they are files.
We don't want to encourage craziness of people making the client suid to
bypass permission issues on the shm file.
Note: I see no evidence of this happening in the wild, but let's be
proactive.
The refresh-copyright script now automatically updates the copyright string
embedded in config.c. In order to achieve this, refresh-copyright gained the
ability to reflow text as the situation needs.
We now give ImGui the true logical size of the window and tell it to scale
the framebuffer. To fix the blurry fonts, we continue to load fonts at the
scale necessary for the DPI and use FontGlobalScale to shrink the fonts back
to the logical size. The font rectangle is then expanded by the framebuffer
scaling, resulting in good text rendering.
This method has the advantage of not messing up the sizes of resizable
overlays when moving across monitors.
The default of [0, 50] makes sense for FPS/UPS graphs, but does not for
things like the import graph. The latter should not take more than 5 ms
for sure.
This commit allows the min/max y-axis value to be specified when registering
the graph.
We used to increment the source buffer index by width * bpp, not by pitch.
This is incorrect and the root cause behind #670.
This is a regression that appeared in 196050bd23.
We now enumerate all IVSHMEM devices, sort them based on PCI bus, slot,
and function numbers, then index from the resulting order. This should
be consistent across boots.
To help the user identify the correct IVSHMEM device, we also print the
list of IVSHMEM devices on startup.
We implement nanotime by converting QueryPerformanceTimer output with
floating point arithmetic. This is necessary to preserve precision on
platforms where each tick is not an integer number of nanoseconds.
Furthermore, struct timespec is included C11 and appears to be supported
on Windows, so we no longer need to #ifdef it out.
Now that we are drawing with damage rects, when the window is hidden and
then exposed the window may not get fully redrawn. This provides
`app_invalidateWindow` for the display server backend to call when the
screen needs a full redraw.
When a new client connects to our session the host will repeat the last
valid frame for the new client. This change will detect this and skip
the duplicated frame.
1. Correctly detect ON state so that host is not built as -march=native by
default.
2. Merge OFF and GENERIC options as we need at least -march=nehalem to build
properly.
This is necessary in case overlays change size. When this happens, we must
damage the larger of the overlays' rectangles this frame and last frame.
This erases the overlay from where it is no longer appears.
In order to do this, we must keep track of the rectangles for every overlay
with no exception. We cannot short-circuit the generation of rectangles if
we run out of buffer space, and we must allocate space for MAX_OVERLAY_RECTS
rectangles for every frame. Otherwise, we will not know where to erase the
overlay if it disappears.
When our window is destroyed, our timers are also destroyed. This causes our
attempt at destruction to fail. Instead, set MessageHWND to NULL in the
WM_DESTROY handler and don't try destroying the timers if the window is gone.
This is desirable because certain uncommon configurations like libdecor or
clang may break, and this shouldn't stop us from seeing if unrelated changes
pass.
DestroyWindow can only be invoked on the thread that created the window.
All other threads must use WM_CLOSE or another message to signal tell the
window to destroy itself.
Currently, we dispatch the events on the wayland display server ourselves.
This is fine when using the cairo backend of libdecor, as it does the same
thign we do, but other backends may require other things to be dispatched.
This commit lets libdecor dispatch events instead through libdecor_get_fd
and libdecor_dispatch, which should hopefully makes things less sketchy.
Our code to generate object files from shaders with the linker seems to
depend on GNU ld. More accurately, it works with lld, but we must specify
the correct object format via -m, which is very difficult to detect in a
satisfactory manner. Therefore we simply force use of GNU ld.
While the renderer can internally track this it would be better to
simply provide this information to the renderer directly so it can make
better decisions on how best to update the screen.
MinGW seems to decide at random whether it wants to use memcpy from
mscvrt.dll or ntdll.dll. Currently, on Debian buster, ntdll.dll is chosen,
while on sid, mscvrt.dll is chosen.
This commit declares a new .def file for ntdll containing only the
functions we want to link from ntdll.dll, and generates ntdll.a from it
with dlltool. This way, MinGW will never be tempted to link functions
like memcpy from ntdll.dll.
This function is sometimes flaky and may fail for no apparent reason,
see https://stackoverflow.com/q/3945003. This has also been experienced
during the development of #610.
This commit adds logging so we may see if it ever fails for no reason
and work out some way to fix it.
We were using an auto-reset event to signal the mousehook exit. This was
fine when there was only one thread, but with the addition of the update
thread, only one thread is signaled, causing the wait to last forever.
The fix is switching to a manual reset event and call ResetEvent after
the threads have exited.
The type of the QuadPart member of the LARGE_INTEGER union is actually
LONGLONG, so we should cast to LONGLONG instead of int.
This avoids truncation should (ms * 10000.0f) exceed 2^31-1.
If the guest is not sending frames at a constant rate, the minimum FPS
timeout may expire drawing an additional frame. This change calculates
the average ups frame time over the past 100ms and adds this to the
timeout value allowing this value to be dynamic.
The accumulated time is not the best way to do this as the timer
function callback may not be exactly every 1000ms, by using the
monotonic clock we will get more accurate results.
It used to be the case that when updating app.manifest, the resource file
it not automatically rebuilt. This made it a headache to update the manifest.
We set OBJECT_DEPENDS so that cmake knows to make the res file depend on
app.manifest and icon.ico.
This commit is based on PR #579 and should be rebased on it after it's merged.
GCC 11 will support x86_64 micro-architecture feature levels.
What we really want to support is nehalem or newer, which is x86-64-v2,
and specifying this instead of nehalem means that we are not tuning for
nehalem specifically.
This function is available since Windows Vista and can therefore be used
directly without going through GetProcAddress. Unfortunately, MinGW does
not have d3dkmthk.h, but we can declare the prototype ourselves and link
against gdi32.dll.
There is no need to LoadLibrary and GetProcAddress to get pointers to
NtDelayExecution or NtSetTimerResolution. These functions don't have
prototypes in any SDK header, but they are exported in ntdll.dll and
we can simply declare the prototype and link ntdll.
There is also no chance that the functions do not exist: I checked an
old install of Windows NT 4.0 and both of these functions exist.
Also used NtSetTimerResolution instead of ZeSetTimerResolution for
consistency (they are the same).
Also changed system timer resolution log message units to μs with
one decimal digit for readability. This is the actual amount of
precision available to us.
1. Use atomics and return exact cursor positions from egl_cursor_render
to avoid race conditions between cursor render and update.
2. Instead of messing with lastCursorValid in various overlays, simply use
the hasOverlay/hadOverlay logic for cursor damage. This simplifies the
logic greatly.
As a result, I believe all cursor-related artifacts are fixed.
Note to reviewer: as atomic_init and atomic_store are implemented as macros,
it is currently not possible to pass structs as compound literals due to the
comma being interpreted as an argument separator by the preprocessor.
According to MSDN documentation for CreateEnvironmentBlock, "[i]f the
environment block is passed to CreateProcessAsUser, you must also
specify the CREATE_UNICODE_ENVIRONMENT flag."
Also pass DETACHED_PROCESS because the host is a GUI application and
doesn't use the console.
Since with the service, we are already running as SYSTEM, we don't need
to use dupeSystemProcessToken to get the token for SYSTEM. This removes
the need for having SeDebugPrivilege, SeTcbPrivilege, and
SeAssignPrimaryTokenPrivilege, or otherwise doing sketchy things.
Furthermore, we now only open the token with the privileges we actually
need.
This allows the process to be terminated without resorting to
TerminateProcess. With some fixes, this allows the notification icon to be
removed when the service is restarted.
Furthermore, instead of sending WM_DESTROY to fool the window into believing
it's being destroyed, we actually call DestroyWindow now.
This commit creates a new utility library, eglutil.h, which contains code
to detect and use EGL_KHR_swap_buffers_with_damage or its EXT equivalent.
This logic used to be duplicated between the X11 and Wayland display servers,
which is not ideal.
Instead of using the desktop <GL/gl.h>, we properly use the OpenGL ES 3.x
headers. Also, we now use GL_EXT_buffer_storage for MAP_PERSISTENT_BIT_EXT
and MAP_COHERENT_BIT_EXT as the core versions are only available in desktop
OpenGL 4.4. Similarly, we need GL_EXT_texture_format_BGRA8888 for GL_BGRA_EXT
as GL_BGRA is desktop-only.
This allows us to set a callback to read a value out before it's about
to be overwritten which can be useful for things like calculating a
running average.
For adjacent changed regions, we actually use the bounding box for the
entire polygon. This may result in more area being damaged than strictly
necessary, but is nevertheless desirable since it reduces the number of
rectangles.
The renderer may take time to process the cursor update due to various
internal factors, as such it's best we copy the data and mark the
message as done ASAP. This prevents the host from filling up the queue
as easily when a high dpi mouse is in use.
The windows hook WH_MOUSE_LL is called in such a way that any delay in
processing causes a system wide stall. This change spawns an extra
thread which waits on an event set by the hook which is then used to
call the callback with an artifical limit of 1000Hz.
Before we try and perhaps fail to init DXGI, we should print out what
the device is so that when there is an error report we can immediately
see if the user has the QXL device attached still.
While it's correct for DXGI to use a asyncronous waitFrame model, other
capture interfaces such as NvFBC it is not correct. This change allows
the capture interface to specify which is more correct for it and moves
the waitFrame/post into the main thread if async is not desired.
This changes the host to use a seperate pool of LGMP memory for cursor
positionl updates without shape information helping to prevent
corruption of the shape entries if they are still pending. While this is
not a perfect solution it resolves the issue without making major
changes to LGMP during the RC phase we are currently in.
Before, we only break out of the current row when a change is detected,
and all subsequent rows are still scanned. Now we break out of the entire
loop. This should make change detection ever so slightly faster.
Testing shows that `D3DKMTSetProcessSchedulingPriorityClass` has a
positive performance impact for NvFBC as well as DXGI, as such always
try to boost the priority for the windows host.
This so called "enhanced" event logic is completely flawed and can never
work correctly, better to strip it out and put our faith in windows to
handle the events for us.
And yes, I am fully aware I wrote the utter trash in the first place :)
Due to a failure to understand atomics when this code was originally
written it has a critical flaw with the fast path where an event could
be signalled when it should not be. This change set corrects this issue
by using atomic operations.
People often miss the warnings about invalid arguments in their command
line, this last minute patch attempts to address this by making
warnings, errors, fixme's and fatal errors stand out if stdout is a TTY.
If the guest VM is not showing a cursor when it starts such as on the
Windows login screen, the client never gets the current position of the
cursor, which prevents the client from attempting to send mouse
movements. This change ensures the client gets the mouse location on
startup.
Conversion from the float values srcW/srcH to the int values for the client window dimensions would sometimes round down, causing the client to scale instead of matching the host's resolution.
In order to make it easier for package maintainers that manage LG inside
their own git repository we should use the version file if it exists
over calling git describe.
A resolution switch could cause the renderer state to become invalid as
the texture format may change while it's being rendered. This fixes this
by adding a lock around the format change and render calls to the
renderer.
We should only advance the pointerIndex if the buffer was not swapped
out for storage. This is to ensure that we do not overwrite cursor
memory that the client(s) may still be using.
This reverts commit d82f2e510d.
While the proposed change is more correct, it breaks the generation of
the file due to failure to locate the resource files, such as
`resources/icon.ico`.
When a new cursor shape is provided by the capture interface we need to
retain a copy of it incase a new client connects which will not yet have
the cursor shape. The logic here was flawed causing the wrong shape to
be sent to a new client in some instances.
Drawing to the front buffer directly requires special handling to
prevent seeing the draw progress (avoiding glClear, etc) and as a result
the output is quite bad unless a compositor is running. Also vsync if
enabled will not function without double buffering enabled.
As OpenGL is the legacy fallback, there are no plans to implement clean
front buffer draw support, so just enable double buffering.
We receive values as int64_t, but when we compute the sum, we store it as
int. This doesn't make sense as we eventually cast it to double when
computing the average. We should instead store the sum as int64_t.
If a compositor has not yet started or is not running the atom
`_NET_WM_BYPASS_COMPOSITOR` may not have yet been created. As such we
need to create it so that if a compositor is started it will see this
propery and honour it.
This should fix the occasional Wayland protocol errors that arise when
the UI thread and the cursor thread race.
Example of error that is fixed:
zwp_pointer_constraints_v1@11: error 1: a pointer constraint with a wl_pointer of the same wl_seat is already on this surface
The existing code would overwrite the texture's data even if the texture
is currently being used to render to screen. This changeset generates a
texture for each buffer preventing this invalid usage.
Extensive profiling reveals that the glibc memcpy performs up to 2x
faster then the existing SIMD implementation that was in use here. This
patch also will copy large 1MB chunks if the pitch of the source and
destination match further increasing throughput.
Under windows text editors like to default to saving in UTF8 with a byte
order mask prefix preventing proper parsing of any option that is on the
first line of the file. Since the configuration file is only intended to
be plain ascii this change ignores all non-ascii characters solving this
issue.
The incorrect spelling, NETWM_BYPASS_COMPOSITOR, somehow worked in the
past, but it appears to not work right now. Corrgan on Discord reported
the issue and confirmed that changing the spelling allowed the compositor
to be bypassed and the client to update faster than 60 Hz on his mixed
refresh rate setup.
This change adds an average function to time how long it takes the GPU
to copy and map the texture, and then uses this average to sleep for 80%
of this average lowering CPU usage and potentially decreasing lock
contention.
This commit restructures the Wayland clipboard handling for host->VM.
Before, we select one clipboard format and buffers the data for it, to
be presented to spice when needed.
Now, we simply offer all clipboard formats supported, and only when spice
asks for the data do we actually read the wl_data_offer. The wl_data_offer
is kept around until a new offer is presented, the offer invalidated, or
when we lose keyboard focus. This is in accordance with the specification
for wl_data_device::selection, which states that:
> The data_offer is valid until a new data_offer or NULL is received or
> until the client loses keyboard focus. The client must destroy the
> previous selection data_offer, if any, upon receiving this event.
We still buffer the entire clipboard data into memory because we have no
knowledge of the clipboard data size in advance and cannot do incremental
transfers.
Furthermore, if the user performs drag-and-drop on our window, we may have
need to handle multiple wl_data_offer objects at the same time. Therefore,
instead of storing state on the global wlCb object, we instead allocate
memory and store it as user_data on the wl_data_offer. As a result, we also
handle drag-and-drop so that we can free the memory.
It has been detemined that a failure to init NvFBC causes a 20-30%
performance penalty on non NvFBC supported hardware (GeForce) when using
DXGI, as such reverse the order and default to using DXGI as our first
option.
If NvFBC is still desired, pr #500 added the option `app:capture` which
can be used to force NvFBC.
One of the most common issues reported in the support channels is the
IVSHMEM size being too small. This change adds a calculation to
determine an optimal size and uses the new `os_showMessage` platform
method to display a message box to the user with the error.
Support for non-PNG types is optional in the spice agent, so we should
avoid sending those if PNG is available.
Currently, the spice VDAgent supports only PNG and BMP formats.
This fixes a regression caused by the move from SDL2 which handled this
itself. We should only minimize when focus is lost if the application
was in full screen mode.
util_guestCurToLocal may not be able to provide the local position if
we do not yet know where the guest cursor is, or the destination render
rect dimensions. Acting on this when this information is unknown causes
undefined behaivour.
To start a clipboard incr transfer the client has to delete the INCR
window property as the reply to the selection. This deletion generates a
property change event with the type delete, errornously triggering the
incr processing of the data. This patch corrects this by ignoring
property deletions.
Imported Installation, FAQ, Troubleshooting, OBS, and Technical FAQ pages
from wiki. Edits done to content as this is no longer a wiki, but versioned
in-repo documentation
Instead of damaging the entire surface when rendering a cursor move,
we can use the EGL_KHR_swap_buffers_with_damage extension to only
damage the part of the window covered by the cursor. This should
reduce the cursor movement latency on Wayland.
We previously used strstr, which can be prone to false positives when
the name of one extension is a substring of another extension.
This commit creates the helper function util_hasGLExt, which asserts
that the substring found in extension list is bounded by either spaces
or the beginning/end of the string.
Corners of table have '+' added, and adds new command line flag --rst-help,
which adds some extra formatting to the make it an rST compliant table for the
in-line docs.
Using util_cursorToInt messes with the error tracking for normal movements,
and is not necessary since we are computing an absolute position on the
client window.
Instead, we should pass doubles directly to display servers and let them
decide how to best handle them. For example, XIWarpPointer accepts doubles
directly.
This prevents the host cursor from moving into another window in capture
mode, solving the problem of input going to an overlapping window in
capture mode, and also preventing loss of focus with focus_follows_mouse.
Currently, (un)grabPointer is used both for tracking/confining the mouse
in normal mode, as well as entering/exiting capture mode. This makes it
impossible to use separate cursor logic for capture mode, which is needed
to deal with overlapping windows for the Wayland backend.
This commit creates separate (un)capturePointer for entering/exiting
capture mode. There should be no behaviour changes.
This adds a new method to the display server interface to allow the
application to notify the ds when there is a guest cursor position
update along with the translated local guest cursor position. This makes
it possible for the display server to keep the local cursor position in
sync with the guest cursor so that window leave events can be detected
when the cursor would move into an overlapping window.
Wayland currently just has a stub for this, and the X11 implementation
still needs some minor tweaking.
This option controls the time period (in ms) after which the help menu
appears when holding down the escape key. After this time period,
capture mode is no longer toggled.
This fixes#527.
Due to the logic in the event loop property events may get filtered out
that were clipboard related. This changes ensures the clipboard event
handler code gets to run first avoiding this issue.
The clipboard atoms may not exist yet and as such we must create them if
this is the case. Failure to do so results in `SEL_DATA` being zero
breaking the clipboard paste mechanics
Since we now let the mouse hook linger until the process is killed, the
cursor event that the hook signals may now be null, as the capture could
have stopped. If the hook fires during this time, a crash occurs.
Instead of converting every SID to string with ConvertSidToStringSidA
and compare it with the magical SID string for local system with strcmp,
we could instead create the local system SID and compare directly with
EqualSid.
We don't actually have any handles that should be inherited, so specifying
TRUE for bInheritHandles to CreateProcessAsUserA is pointless.
Furthermore, according to MSDN, "[y]ou cannot inherit handles across
sessions," and we are spawning the host in a different session, so this
is even more pointless.
This commit fixes the -Wmissing-field-initializers warning, which can only
be disabled with a pragma. GCC wants us to Initialize libdecor reserved
fields, which requires knowing how many reserved fields there are.
This is an implementation detail, and so we can only disable the warning.
This also fixes -Wincompatible-pointer-types, which is an actual bug.
If the window manager does not support the motif hints then fallback to
creating a utility window, do not do both. A utility window is a
sub-optimal fallback as it may prevent the application being shown in
the taskbar or as a running application as has been reported on KDE.
Using the first two valuators present in the event is incorrect. Events
with only one valuator set, such as those sent by the Xorg evdev driver
when the mouse moved along one axis only, were being discarded. On the
other hand, mice with multiple scroll wheels may be able to emit events
with two scroll wheel valuators set.
The XInput2 specification is light on details, but "Rel X" and "Rel Y"
appear to be the de facto standard names for the motion valuators. If
valuators with those labels are not found, fall back to using valuators
with numbers 0 and 1.
Due to the confusing nature of the x11 protocol, bit_gravity and
win_gravity are not what they appear to be. These do not describe the
window position but rather the pixels/subwindows when the window is
resized. Instead set the gravity via the WM_SIZE_HINTS property which
all modern window managers should respect.
This fixes the fullscreen and likely borderless issue too that people
have been reporting on X11 under gnome. Thanks to tdb in discord for
spotting the error.
Instead of doing ShellExecute from the service, we instead get the token
of the currently logged in user, and do CreateProcessAsUserA to run
notepad with that token. This should be safe.
Also for failure to parse command line. For these errors, restarting
with exponential backoff will not help: no amount of restarting the
service could possibly make the ivshmem device exist or larger, so
we shouldn't try.
Certain users of Radeon cards have observed that the host fails to start
at boot, with D3D11CreateDevice failing with HSTATUS 0x887a0004, which
translates to "The specified device interface or feature level is not
supported on this system."
This failure results in a LG_HOST_EXIT_FAILED exit code, which the service
does not attempt to restart. The user has to manually restart the service
for the host application to work.
These users reported that the host application started fine on
B2. This strongly suggests that the fix to enable capturing the login
screen made the host application start too early during the boot process,
and the graphics driver did not have time to initialize fully.
This PR allows the service to retry a few times on LG_HOST_EXIT_FAILED,
with exponential backoff, before giving up. This should cover this bug
and other similar bugs related to the early initialization which I do not
have logs for.
When linking against libbfd.so, just passing libbfd.so to the compiler is
sufficient. When linking against the static version libbfd.a, however,
we must additionally link against libiberty.a and libz.a.
This commit adds a CMake helper to find the correct libraries that need
to be passed to link against libbfd correctly.
Essentially, debug.h defines printTrace as an empty macro when the macro
ENABLE_BACKTRACE is not defined, breaking the compilation of crash.c.
Fixed by defining a private macro for debug.h only to avoid clashing.
This commit introduces a new option, app:capture, which can be set to
either DXGI or NvFBC to force the host application to use that backend.
This is very useful for testing DXGI on Quadro cards, which would default
to running with NvFBC.
This is needed for proper cleanup.
Freeing the capture interface also avoids a crash when using the NvFBC
backend. This is because we moved the mouse hook removal to nvfbc_free.
If nvfbc_free is not called before we start freeing LGMP memory, the
mouse hook would end up writing the cursor position into an invalid
memory address, causing an access violation.
The only addition to v4 was `wl_surface_damage_buffer`, which we do not
use.
This change should allow running on more compositors (even though v4 is
already old -- 5 years now).
Ref
3384f69ecf.
We don't necessarily need `wl_output.release`, which is the only
addition in v3.
This allows Looking Glass to run on Ubuntu 20.04 without having to go
difficult lengths to acquire newer Wayland packages. Since 20.04 is an
LTS release, this seems worthwhile for the small amount of complexity
this introduces.
Fixes
https://forum.level1techs.com/t/latest-build-allow-inhibiting-shortcuts-dialog-ubuntu/168684/6.
Calling abort() instead of exit() will generate a core dump, allowing gdb
to be used when the client crashes. This is desirable for the following
reasons:
1. gdb can be used to inspect the call stack with far more detail than the
our quick stack trace code, and also allows the access to the heap.
2. Our SIGSEGV handler is unable to use debug symbols for shared libraries,
making it impossible to debug bugs involving drivers and similar.
When users press escapeKey for a long time, they probably want to
see the help text instead of actually toggling capture. Therefore,
if the key is held down for more than 500 ms, we assume the user
wants to look at the help text and do not toggle capture mode.
500 ms seems to be a decent compromise, allowing slow presses, but
is not enough time for the user to have looked at the help text.
This was missed when splitting up wayland.c into multiple modules.
This commit also drops the useless #include <SDL2/SDL.h>, bringing
SDL removal one step closer.
If the scale factor of an wl_output changes while the client is running,
the maximum scale factor is not updated. This may result in incorrect
scaling.
Therefore, when the scale factor is changed, we should generate a
resize event.
During the refactor/rebase period with B3-next the conditional was
accidentally reversed. This would cause the cursor to be ungrabbed
simply when toggling capture mode instead of waiting for the cursor to
exit the window.
Mouse move deltas greater then 10 are rare, let alone the 20 this code
now uses. Any movements that exceed 20 pixels will disable the exit
detection code path preventing rapid movements in FPV games from causing
the cursor to exit the window if autoCapture is enabled.
As we now are using our own backends instead of SDL, there is no longer
any need to warp back to the center of the window when in autoCapture
mode. This breaks the SDL ds backend behaviour, however as SDL is
planned to be removed this is not an issue.
The $escape+S keybinding now cycles through the available scale algorithms.
This allows the user to switch between algorithms if the automatic detection
turns out to be problematic.
The algorithms are renumbered so that 0 can be LG_SCALE_AUTO.
Clang doesn't know the warning option `format-truncation`, which GCC
falsely triggers on when calling vsnprintf. Further to this, GCC doesn't
understand the warning group `unknown-warning-option`, as such we must
disable these warnings in the cmake.
Reusing cached EGLImages while the frame format has changed will result
in visual artifacts. We should instead destroy the EGLImages and let
them be recreated.
Instead of duplicating the #defines from the shader into the C code,
this commit adds a custom CMake rule that exports all the #defines
from a shader into a C header.
This commit fixes the issues with the meaning of useNearest being flipped
by removing the variable and use enumerations.
We define an enumeration EGL_DesktopScaleType to express the type of scaling
needed to be performed: no scaling, upscaling, or downscaling. This is
updated when either the frame size or the viewport size changes.
Previously, we only updated the useNearest when the frame size changes.
The desktop shader can now support an enumeration of scaling algorithms,
although only nearest and linear are currently implemented.
Like before, nearest is used when not scaling or upscaling, and linear is
used when downscaling.
It appears that the PCI BAR memory is slow to access with remap_pfn_range
and that it should instead be faulted in one page at a time.
The commit 5774e21965 implemented the former
behaviour and caused a performance regression in the VM->VM case.
This commit retores the old behaviour, but extends it to support mmaping
the kvmfr device directly, without going through a dmabuf.
The Wayland display server is getting unwieldy due to the sheer size.
To make it easier to edit in the future, I split it into many components
based on logical boundaries.
This is enabled on default. Specify wayland:warpSupport=no to disable it,
which may be useful on certain compositors that do not warp when the
pointer is confined.
This commit implements support for LG_DS_WARP_SURFACE, as well as a warp
routine based on cursor confines.
This may not necessarily work for all compositors. As such, the old cursor
routines are still kept, and used when wm.warpSupport is set to false.
This commit converts the output of ds->getProp(LG_DS_WARP_SUPPORT) to
an enum containing three items:
* LG_DS_WARP_NONE: warp is not supported at all
* LG_DS_WARP_SURFACE: warp is possible, but only inside the window
* LG_DS_WARP_SCREEN: warp is possible anywhere on the screen
LG_DS_WARP_NONE corresponds to the old false return value, and
LG_DS_WARP_SCREEN corresponds to the old true return value.
LG_DS_WARP_SURFACE is designed for Wayland, where warping is possible,
but only in our window. In this case, since we cannot warp outside
the window, we can warp the cursor to the edge when we attempt to exit.
If the cursor leaves, the normal leave routine gets called, and the
cursor disappears. If the cursor does not end up leaving, we grab it
again.
This makes dealing with window manager shortcuts that overlap with guest
keys more pleasant, while retaining the previous functionality for users
who prefer it.
For instance, previously, using Alt+Tab (or $mod as Alt in i3/sway
movement commands) would result in the guest retaining Alt as pressed.
When the guest regained focus, it would continue thinking Alt is
pressed, leading to accidentally triggering obscure shortcuts. One had
to remember to press Alt again to "unstick" things, which was
suboptimal.
This allows buffers to be shared between different asynchronous operations.
Once all users no longer need the buffer, it will be freed.
The motivation for this is being able to stream Wayland clipboard data
asynchronously to multiple clients. The buffer should only be freed after
the clipboard has changed and all ongoing transfer completes.
This commit forces the DMA'd memory to be copied into the texture in
the EGL on_frame handler. This avoids tearing when the LG host inevitably
updates the underlying memory. We need an additional copy inside the GPU,
but this is cheap compared to copying from system memory.
We could have used logic to lock the memory buffer, but that would require
performing DMA on every frame, which wastes memory bandwidth. This
manifests as reduced frame rate when moving the mouse compared to the
non-DMA implementation.
We also keep multiple EGLImages, one for each DMA fd, to avoid issues
with the OpenGL driver.
It used to be the case that you need to create dmabuf for kvmfr devices
to be able to mmap them. But after #457, this is no longer needed.
Directly mmaping the kvmfr device has the advantage of avoiding the
creation of a dmabuf, which has cost (e.g. the list of pages, the
scatterlist, etc.).
This commit makes the test program try the following cases:
* mmaping 0-offset dmabuf with 0 offset
* mmaping 0-offset dmabuf with 1 page offset
* mmaping page-offset dmabuf with 0 offset
* mmaping page-offset dmabuf with 1 page offset
* mmaping device with 0 offset
* mmaping device with 1 page offset
This allows PCI kvmfr devices to be directly mmap'd just like in-memory
ones. Also, the more efficient mmap implementation is used for mapping
the dmabuf, avoiding the faulting code entirely.
Added an array option static_size_mb to the kvmfr module to create a
list of in-memory kvmfr devices. These devices support dmabuf just like
normal kvmfr devices. Additionally, they can be mmap'd, which allows
them to be passed to qemu as ivshmem devices.
This new implementation uses a special mimetype to tag data copied from
the guest, instead of using flags. This should make it easier to
implement asynchronous transfers in the future. Also, it's simpler to
understand and less error-prone.
The pid is included in the mimetype in order to distinguish between
different instances of looking glass: you might want to copy between
two different VMs, for example.
This commit adds a new option, win:autoScreensaver, which when set to yes,
automatically disables the screensaver when requested by an application
running in the guest, and enables it when the application no longer wants
it disabled.
This is useful when doing media playback in the guest.
This will allow us to add an option to disable the screensaver on the client
when an application in the guest requests it. This behaviour may be useful
when the guest is doing media playback.
It appears that the keyboard should only be grabbed if the client is
focused and the cursor is in the view. However, the relevant logic was
missing from core_setCursorInView, and the keyboard was never actually
grabbed.
This commit adds the call to g_state.ds->grabKeyboard(), allowing grabbing
to work.
Before, if you want to see the FPS, you need to close the client and
restart it with the -k switch to see the FPS. This is annoying.
This PR introduces a new keybind, ScrollLock+D, which, when pressed,
toggles the display of the FPS.
This is implemented for both EGL and OpenGL backends.
egl_help_set_text and egl_help_render were both accessing bmp->help from
different threads. This creates a race condition in which if the help text
is quickly toggled on and off, it stays on.
This has been fixed with an atomic exchange.
This should prevent the looking-glass-client window from having an alpha
channel. On Wayland, the alpha channel is used to compose the window onto
the desktop, so the wallpaper would bleed through unless set to complete
opaque.
We worked around this by using constant alpha for rendering, but it was
not sustainable. Instead, we should just ask for 24-bit context.
One of the major issues with the old tracking code is a data race
between the cursor thread updating g_cursor.guest and the
app_handleMouseBasic function. Specifically, the latter may have
sent mouse input via spice that has not been processed by the guest
and updated g_cursor.guest, but the guest may overwrite g_cursor.guest
to a previous state before the input is processed. This causes some
movements to be doubled. Eventually, the cursor positions will
synchronize, but this nevertheless causes a lot of jitter.
In this commit, we introduce a new field g_cursor.projected, which
is unambiguously the position of the cursor after taking into account
all the input already sent via spice. This is synced up to the guest
cursor upon entering the window and when the host restarts. Afterwards,
all mouse movements will be based on this position. This eliminates
all cursor jitter as far as I could tell.
Also, the cursor is now synced to the host position when exiting
capture mode.
A downside of this commit is that if the 1:1 movement patch is not
correctly applied, the cursor position would be wildly off instead
of simply jittering, but that is an unsupported configuration and
should not matter.
Also unsupported is when an application in guest moves the cursor
programmatically and bypassing spice. When using those applications,
capture mode must be on. Before this commit, we try to move the guest
cursor back to where it should be, but it's inherently fragile and
may lead to scenarios such as wild movements in first-person shooters.
We used to test for the EGL_KHR_platform_base and EGL_EXT_platform_base,
but those only really signal the availability of eglGetPlatformDisplay(EXT)
functions, not whether the constant EGL_PLATFORM_WAYLAND_KHR or
EGL_PLATFORM_WAYLAND_EXT is accepted by their respective functions.
Instead, we switch to test for the extensions that tells us whether the
Wayland platform is supported.
Using a macro ENABLE_OPENGL just like ENABLE_EGL to optionally remove
OpenGL implementation code. This is mostly because on Wayland it's just
a rehash of the EGL code (as EGL is the only way to create OpenGL
contexts on Wayland).
`$DISPLAY` will be set even in a Wayland session, which causes LG to
initialize itself under Xwayland unless it is explicitly compiled with
`-DENABLE_X11=OFF`.
We could add a Wayland check within the X11 backend, but reordering the
code-generated array seems like a better solution.
We ask for 32-bit colour buffer when creating the EGL context. On Wayland,
this sometimes give contexts with alpha channels, resulting in unwanted
transparency. So we clear the alpha channel in the desktop shader.
We also switch to using constant alpha for blending the splash, which
avoids more alpha issues.
As the window manager may change our mode to full screen without our
request we must ask the ds backend for the current state when we want to
toggle the mode.
Should this all go well you should be left with the file `looking-glass-client`
### Removing Wayland or X11 support
Wayland and/or X11 support can be disabled with the compile options
`ENABLE_WAYLAND` and `ENABLE_X11`, if both are specified only `SDL2` will remain
and the client will fallback to using it.
cmake ../ -DENABLE_WAYLAND=OFF
At this time, X11 is the perferred and best supported interface. Wayland is not
far behind, however it lacks some of the seamless interaction features that X11
has due to the lack of cursor warp (programmatic movement of the local cusror) on
Wayland.
---
## Usage Tips
### Key Bindings
By default Looking Glass uses the `Scroll Lock` key as the escape key for commands as well as the input capture mode toggle, this can be changed using the `-m` switch if you desire a different key.
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
