Previously we call Thread::setThreadAffinityInternal in
Thread::startThread. The purpose was to avoid the main workload being
run on incorrect CPUs. This leads to a race condition of setting
`Thread::thread_` in `Thread::start()` and accessing
`Thread::setThreadAffinityInternal` though.
This patch moves the call after the construction of std::thread to avoid
the race condition. The downside is that the first tasks, if any, upon
starting a thread might be run on incorrect CPUs.
Fixes: 4d9db06d66 ("libcamera: add method to set thread affinity")
Signed-off-by: Harvey Yang <chenghaoyang@chromium.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Move the section of the Thread support page dealing with thread safety
to a dedicated .dox file at Documentation/. This is done to support the
splitting of the Documentation into a public and internal version. With
a separate page, references can be made to thread safety without having
to include the Thread class in the doxygen run. Some sections of the new
page are still specific to internal implementations and so are hidden
with the \internal flag and an internal section which is conditionally
included. For now, hardcode it to be included in the Doxyfile.
Signed-off-by: Daniel Scally <dan.scally@ideasonboard.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Source files in libcamera start by a comment block header, which
includes the file name and a one-line description of the file contents.
While the latter is useful to get a quick overview of the file contents
at a glance, the former is mostly a source of inconvenience. The name in
the comments can easily get out of sync with the file name when files
are renamed, and copy & paste during development have often lead to
incorrect names being used to start with.
Readers of the source code are expected to know which file they're
looking it. Drop the file name from the header comment block.
The change was generated with the following script:
----------------------------------------
dirs="include/libcamera src test utils"
declare -rA patterns=(
['c']=' \* '
['cpp']=' \* '
['h']=' \* '
['py']='# '
['sh']='# '
)
for ext in ${!patterns[@]} ; do
files=$(for dir in $dirs ; do find $dir -name "*.${ext}" ; done)
pattern=${patterns[${ext}]}
for file in $files ; do
name=$(basename ${file})
sed -i "s/^\(${pattern}\)${name} - /\1/" "$file"
done
done
----------------------------------------
This misses several files that are out of sync with the comment block
header. Those will be addressed separately and manually.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Objects can be scheduled for deletion with Object::deleteLater(), which
queues a deferred deletion to the thread's event loop. As the
deleteLater() function is meant to be called from a different thread,
this may race with thread termination, and deferred deletions queued
just before calling Thread::exit() may not be processed by the event
loop. Make sure they get processed when finishing the thread, before
stopping.
This eliminates the race condition that occurs when calling
Object::deleteLater() followed by Thread::exit() from the same thread.
Calling deleteLater() from neither the thread the object is bound to or
the thread calling Thread::exit() is still inherently racy.
The change fixes a failure in the object-delete unit test.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Milan Zamazal <mzamazal@redhat.com>
The signal.h header doesn't need to include object.h. Replace it with a
forward declaration, and instead include object.h in source files that
require it. It can speed up compilation a little bit, but more
importantly avoids unintended dependencies from the Signal class to the
Object class to be added later as the compiler will catch them.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Milan Zamazal <mzamazal@redhat.com>
This replaces Mutex and MutexLocker with our own defined classes.
The classes are annotated by clang thread safety annotations.
So we can add annotation to code where the classes are used.
v4l2 code needs to be annotated, which violates Mutex capability.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
ConditionVariable is alias to std::condition_variable. This replaces
std::condition_variable with the ConditionVariable. It enables
replacing ConditionVariable implementation easily in the following
patches.
Signed-off-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Many signals used in internal and public APIs carry the emitter pointer
as a signal argument. This was done to allow slots connected to multiple
signal instances to differentiate between emitters. While starting from
a good intention of facilitating the implementation of slots, it turned
out to be a bad API design as the signal isn't meant to know what it
will be connected to, and thus shouldn't carry parameters that are
solely meant to support a use case specific to the connected slot.
These pointers turn out to be unused in all slots but one. In the only
case where it is needed, it can be obtained by wrapping the slot in a
lambda function when connecting the signal. Do so, and drop the emitter
pointer from all signals.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Umang Jain <umang.jain@ideasonboard.com>
Usage of 'method' to refer to member functions comes from Java. The C++
standard uses the term 'function' only. Replace 'method' with 'function'
or 'member function' through the whole code base and documentation.
While at it, fix two typos (s/backeng/backend/).
The BoundMethod and Object::invokeMethod() are left as-is here, and will
be addressed separately.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Acked-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
There are use cases for calling the dispatchMessages() function
recursively, from within a message handler. This can be used, for
instance, to force delivery of messages posted to a thread concurrently
to stopping the thread. This currently causes access, in the outer
dispatchMessages() call, to iterators that have been invalidated by
erasing list elements in the recursive call, leading to undefined
behaviour (most likely double-free or other crashes).
Fix it by only erasing messages from the list at the end of the outer
call, identified using a recursion counter.
Bug: https://bugs.libcamera.org/show_bug.cgi?id=26
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Jacopo Mondi <jacopo@jmondi.org>
Tested-by: David Plowman <david.plowman@raspberrypi.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Move the functionality for the following components to the new
base support library:
- BoundMethod
- EventDispatcher
- EventDispatcherPoll
- Log
- Message
- Object
- Signal
- Semaphore
- Thread
- Timer
While it would be preferable to see these split to move one component
per commit, these components are all interdependent upon each other,
which leaves us with one big change performing the move for all of them.
Reviewed-by: Hirokazu Honda <hiroh@chromium.org>
Reviewed-by: Paul Elder <paul.elder@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
