For most connectors associating monitors with our virtual
'heads', we merely track their ids. This means we don't have
to bother freeing any client side state at deallocation time.
There's one exception, though. The main connector, connector0,
we keep an open reference to. We do this, because it owns the
mode object we use in SetCrtc calls.
This commit ensures that connector0 for each head is properly freed
when that head is deallocated.
Some underpowered video cards will have multiple connectors
tied to one controller. In this case all connectors get
the same "cloned" output automatically.
This commit detects this situation and prevents plymouth from
wastefully trying to allocate several frame buffers. This
commit also prevents plymouth from constantly trying to switch
between those allocated frame buffers every frame of the animation.
We only really support full color layouts at the moment
for kernel modesetting drivers, but this commit adds a
"color_depth" member to the buffer structs so we can at
least recognize non-conforming buffers.
Right now if a user erroneously puts, e.g., console=ttyS0 twice
on their kernel command line, plymouth will open that tty twice
and write to it twice, and read from it twice, etc.
This commit filters those duplicates out.
There are times when plymouth is running that the filesystem isn't
accessible. For instance, if a user has /usr as a separate partition,
then when first leaving the initrd, plymouth won't have access to its
plugins.
In those cases we really need to survive if the user hits escape.
This commit compiles details into the binary. In this way, if the
plugins aren't available, we still have something to fall back to.
In the event should_ignore_show_splash_calls () returns
true, we won't ever have a keyboard set. This commit
make sure that we don't try to use the keyboard if its
unavailable.
Right now, we put plymouth into a sort of degraded mode when we
find init=/anything on the kernel command line. This is so if
the user does init=/bin/sh to get fix their system, we don't
get in the way.
This breaks plymouth for things like init=/sbin/bootchartd and even
init=/sbin/init. We've previously had a plymouth.override-splash
kernel command line option to force plymouth on in those cases.
This commit flips things around a bit. Now init=/sbin/bootchartd
will work by default and there's a new option
plymouth.ignore-show-splash which forces things back into degraded
mode. We also will implicitly do the degraded mode for e.g.
init=/foo/barsh
Plymouth has historically used "plymouth:" to identify arguments destined
for it on the kernel command line. Most other things seem to the period
instead of colon, so this commit just makes plymouth conform.
For compatibility, we still support the old way.
Because of an apparent think-o in the script, if a custom theme
has a broken symlink in it, plymouth-populate-initrd would silently
stop processing files and ship an incomplete set of data files.
This commit changes "break" to "continue" so that broken symlinks
are ignored, which was probably the original intent.
It's important to make sure the theme name is properly
quoted when passed to the basename command. This
is because, if the theme name is empty we want the empty
string returned, not the suffix that would otherwise be
stripped off.
Some discussion here:
https://bugzilla.redhat.com/show_bug.cgi?id=606634
If the user has the details plugin configured, and the
system does plymouth quit --retain-splash, then their
may be a crash.
This is because the plugin event loop exit handler will called
after the plugin is destroyed. This commit makes sure
the exit handler is removed before the plugin is freed.
Before we defaulted to tty1, but some systems just
don't have tty1. Normally, those systems specify
an alternate console on the kernel command line,
but not always.
This commit tries to make things work in those cases
as well.
I'd like to find a more generic way to make this all
work.
epoll_wait was getting passed a pointer to 64 events,
but was only being told to use up to 8 of those events
at a time because it was using sizeof (pointer) for the
argument specifying the number of events instead of the
actual number of events.
plymouth has naïvely been treating the first available mode in
the connector object as the active mode. While this is true
most of the time, it doesn't hold true if the user overrides the
mode on the kernel command line.
This commit changes things to look up the actual active mode, and go
with that, which should help prevent flicker at start up in some cases.
This patch is based heavily on a patch from Forest Bond
<forest@alittletooquiet.net> here:
http://lists.freedesktop.org/archives/plymouth/2010-June/000369.html
and first mentioned in commit abfda7550a
Previously, we would store the mode of each head as pointer in
the array of modes on the connector object. Now we just store
the index into that array.
This is to make it clearer that we don't own the memory
associated with it directly, and that the memory associated with
it is automatically cleaned up when the connector is.
This also helps to highlight a problem mentioned by
Forest Bond <forest@alittletooquiet.net> here:
http://lists.freedesktop.org/archives/plymouth/2010-June/000369.html
Namely, we've been naïvely treating the first available mode in
the connector object as the active mode. While this is true
most of the time, it doesn't hold true if the user overrides the
mode on the kernel command line.
Adds an ignore-keystroke command. This does the same thing as the
ignore-keystroke option, but is migrated to a command. The option still exists
for compatibility.
Right now, we always assume the tty can be reopened when it
gets disconnected. This isn't always true. While plymouth
clearly won't function 100% properly without a tty, we shouldn't
blow an assertion.
Mandriva has cases where they need to drop to details mode
before /usr is mounted. /usr has the details splash plugin,
so this causes plymouth to explicitly exit.
This commit makes plymouth cope with that scenario slightly
better. It takes out the exit(1) call which was clearly bogus,
and also makes parts of the code handle a NULL splash better.
This is not a real fix though.
Ultimately, we should make "details" be built-in as a nice failsafe.
There's no reason we should need to load it off the disk as a plugin.
