In case the H/W decoder is not able to decode the stream (like too high
profile) try the S/W version.
This is done detecting the failure and trying to recreate the pipeline
in case:
- we are using a H/W pipeline;
- we didn't decode any frame (otherwise it means we lost the beginning
or it was not a problem of H/W decoder).
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Acked-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
This patch fixes a leak due to not freeing GstBus watch.
The watch is attached (as GSource) to the main loop and retains
a pointer to the bus so we need to remove it to release the bus
when we release the pipeline.
This was detected forcibly creating and destroying lot of streams.
After a while the client program consumed all file descriptors
and stopped working. This as GstBus retains a GPoll which,
under Unix, uses 2 file descriptors.
For some reasons using gst_pipeline_get_bus again in free_pipeline
do not fix entirely the leak so keep a pointer to the exact
bus we set our watch on.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Acked-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
We first try to detect if an Intel GPU is available (by looking into
udev's database) and then probe Gstreamer's registry cache to see
if there is h/w based decoder (element) available for the incoming
video codec format. If both these conditions are satisfied (i.e,
Intel Media SDK Gstreamer plugin (libgstmsdk.so) and associated
libraries are properly installed), we then create a simple decode
pipeline using appropriate h/w based decoder and post-processor
elements instead of relying on playbin -- which may not be able to
auto-select these elements.
For example, if the incoming codec format is h264, we then create
a pipeline using msdkh264dec and vaapipostproc elements instead of
avdec_h264 and videoconvert.
Cc: Frediano Ziglio <freddy77@gmail.com>
Cc: Gerd Hoffmann <kraxel@redhat.com>
Cc: Marc-André Lureau <marcandre.lureau@redhat.com>
Cc: Dongwon Kim <dongwon.kim@intel.com>
Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
Signed-off-by: Hazwan Arif Mazlan <hazwan.arif.mazlan@intel.com>
Signed-off-by: Jin Chung Teng <jin.chung.teng@intel.com>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
Although currently after free_pipeline we free the entire structure
the name and the function suggests that we only free the pipeline.
Also this is fixing a future possible problem with the series
from Vivek Kasireddy reusing the SpiceGstDecoder for another
pipeline if H/W encoder pipeline creation fails.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Acked-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
If playbin3 is available (checked in Gstreamer registry), it
makes sense to prefer it given the major improvements and
features it offers including the capability to handle higher
bitrates.
Cc: Frediano Ziglio <freddy77@gmail.com>
Cc: Dongwon Kim <dongwon.kim@intel.com>
Cc: Hazwan Arif Mazlan <hazwan.arif.mazlan@intel.com>
Signed-off-by: Jin Chung Teng <jin.chung.teng@intel.com>
Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
Adding the string "stream-format=byte-stream" to decoder capabilities
(dec_caps) for h264/h265 codecs stops Gstreamer from complaining
about missing stream format.
Cc: Frediano Ziglio <freddy77@gmail.com>
Cc: Dongwon Kim <dongwon.kim@intel.com>
Cc: Jin Chung Teng <jin.chung.teng@intel.com>
Cc: Hazwan Arif Mazlan <hazwan.arif.mazlan@intel.com>
Signed-off-by: Vivek Kasireddy <vivek.kasireddy@intel.com>
Acked-by: Frediano Ziglio <freddy77@gmail.com>
Solve double export issue of some functions between spice-gtk
and spice-glib libraries.
Some functions are exported by both spice-gtk and spice-glib.
As the state of these function can be duplicate this can create
some issues. Specifically you can enable debug but get only partly
enabled.
So to solve this and have a consistent behaviour across system
add new functions in spice-glib (where these functions where supposed
to be) and use them (we need to keep providing old functions for
ABI compatibility).
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
In case version script support is not available we can still
able to filter symbols exported.
This for instance will work with linkers not supporting
--version-script (like system MacOS one).
Use new version macros similar to Glib to detect also
deprecated function.
Tested that exported symbols remains the same from both Linux
and Windows shared libraries.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Fix some "Since" field in function documentation.
Nobody should nowadays use versions old like these but good to
have them fixed.
Function presence was checked looking at map-file and code
presence.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
One file for spice-glib and another for spice-gtk.
This avoid having duplicate exports in the future.
spice_g_signal_connect_object, spice_util_get_debug,
spice_util_get_version_string, spice_util_set_debug and
spice_uuid_to_string are exported by both libraries to keep ABI
compatibility.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
This option is intended for libtool, not for any linker.
Check the support of --version-script option using an empty
list of symbols to catch some faulty linker supporting that
option but not allowing not existing symbols (some buggy mold
versions).
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Win32 has #define for those.
Change variable names `near` and `far` to `_near` and `_far`.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Hardware support for Apple iOS is not available, check correctly
and not try to use it.
Signed-off-by: osy <osy@turing.llc>
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Tested-by: osy <osy@turing.llc>
This patch fixes bad filtering of USB devices.
The logic was introduced for windows clients in commit b838937 but was
moved in 2e19014 and executed also by non windows clients, resulting
in this bug.
Fixes: 2e19014 "usb-redir: isolate usage of libusb and usbredirhost"
Fixes: https://gitlab.freedesktop.org/spice/spice-gtk/-/issues/157
Signed-off-by: Victor Toso <victortoso@redhat.com>
The keycodemapdb Meson subproject provides a program and a source input.
Since it is a subproject, Meson wants to sandbox that and requires it to
be explicitly exported. But this never happened -- instead, we manually
poked at files using the actual string path "subprojects/......"
This was always a Meson sandbox violation, but Meson 0.63.0 started
noticing it and erroring out.
Instead, do the right thing. Update the subproject to a version that has
a meson.build with actually meaningful contents -- namely, a files
variable and a found program. Then use these in order to run the needed
custom_target.
In the process, it is also necessary to correct the argument ordering
when running keymap-gen.
Reviewed-by: Marc-André Lureau <marcandre.lureau@redhat.com>
libucontext is a lightweight implementation of ucontext for platforms
that do not have a built-in implementation. This allows us to use the
same code to support libucontext as ucontext.
by passing an open file descriptor in order to support the modern
Android USB device permissioning scheme.
During the process, a handle obtained by a call to
libusb_wrap_sys_device() is stored in the SpiceUsbDevice instance.
A call to libusb_open() is skipped if there is already a handle
present in the SpiceUsbDevice instance.
Finally, on Android, the option LIBUSB_OPTION_NO_DEVICE_DISCOVERY
must be set prior to the call to libusb_init() as per libusb
documentation.
Acked-by: Frediano Ziglio <freddy77@gmail.com>
This will pass hostname.
This fixes https://gitlab.freedesktop.org/spice/spice-gtk/-/issues/137.
You can easily verify the change with a network capture program running
something like
remote-viewer spice+tls://www.spice-space.org:443
you will see the hostname in the initial exchange.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Acked-by: Victor Toso <victortoso@redhat.com>
Use proper ioctls for MacOS.
Import new spice-common to solve an issue detecting Python3.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Acked-by: Victor Toso <victortoso@redhat.com>
Resolve some portability issue, like MacOS.
Macros from <endian.h> were used but not all systems have
this header.
Signed-off-by: Frediano Ziglio <freddy77@gmail.com>
Acked-by: Victor Toso <victortoso@redhat.com>
This fixes a regression from 032ca202f8
where a check on whether the usbids file contained anything but the
wrong variable was checked. This is visible on windows when trying to
share a USB device as devices won't get named based on the usbids file.
Same thing on linux if we can't read the name from the sysfs.
`usbids_vendor_info` is only initialized to something not NULL after it
was checked to be different from NULL which is obviously wrong. The
patch wanted to check that the `usbids_vendor_count` was different from
0 instead.
Acked-by: Frediano Ziglio <freddy77@gmail.com>
Note: that unusual function seems problematic for GI in general, we may want a simple pointer instead..
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
Internally, spice-widget has separate try_keyboard_ungrab() and
try_mouse_ungrab() functions to ungrab the keyboard and mouse
respectively. Both of these functions are called when the grab
sequence is pressed.
spice-widget exposes try_mouse_ungrab() as spice_display_mouse_ungrab()
to allow callers to explicitly ungrab. However, it does not currently
expose try_keyboard_ungrab(), so callers can only ungrab the mouse and
cannot ungrab the keyboard.
Correct this by exposing try_keyboard_ungrab() as
spice_display_keyboard_ungrab().
Fixes#73
gir fails to compute the size of the array, which results in invalid
class size. Deprecate usage of SPICE_RESERVED_PADDING, which isn't that
useful anyway.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
We are about to modify the struct fields, let's make sure we don't break
the ABI.
Once check-abi is added to our CI, this shouldn't be needed, and can
probably be reverted.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
The GSource added by gst_bus_add_watch() apparently holds
a reference to the GstBus. This can be seen by running
GST_DEBUG="GST_TRACER:7" GST_TRACERS="leaks" spicy:
GST_TRACER :0:: object-alive, type-name=(string)GstBus, address=(gpointer)0x1e76db0, description=(string)<bus6>, ref-count=(uint)1, trace=(string);
(note that gst_deinit() must be called for this output to be shown,
which doesn't happen in virt-viewer; see 0381e62)
To fix this, save the source's id returned by gst_bus_add_watch()
and remove the source when the pipe is unreferenced.
Signed-off-by: Jakub Janků <jjanku@redhat.com>
This avoids a critical warning. The function will eventually be called again on realize.
Signed-off-by: Marc-André Lureau <marcandre.lureau@redhat.com>
