summaryrefslogtreecommitdiff
path: root/libs/gui/BufferQueueThreadState.cpp
Commit message (Collapse)AuthorAgeFilesLines
* Remove libbinder usage from media.swcodec apexChong Zhang2020-02-071-0/+10
| | | | | | | | | | | | libgui_bufferqueue_static is used by com.android.media.swcodec apex which doesn't do any IPC through binder. Remove binder dependency to eliminate any possibility of IPC via any unstable interfaces. b/147759770 b/139201422 test: atest CtsMediaTestCases -- --module-arg CtsMediaTestCases:size:small Change-Id: I98dbe751527bf4725c7764ba21f46b6af60968b4
* Remove need for libbinderthreadstate.Steven Moreland2020-02-051-2/+3
| | | | | | | | | | | | | | Instead of having this library, libbinder/libhwbinder can keep track of stack pointers so that when they recurse, we know which one was visited most recently. As with the original implementation of libbinderthreadstate, this is somewhat of a hack. An explanation of why this is and what to do instead is added in CallerUtils.h. Bug: 148692216 Test: libbinderthreadstateutils_test Change-Id: Ief28663728fb8786b06bf9e72238052b9af81d87
* BufferQueueProducer: use the correct IPCThreadState.Jayant Chowdhary2019-03-081-0/+38
The system variant of libgui maybe be double loaded. Also IGraphicBufferProducer functions may be called from hwbinder threads due to the presence of TWGraphicBufferProducer wrappers (hybrid interfaces). Therefore, we should use the correct IPCThreadState/ hardware::IPCThreadState to query callingPids. This also avoids access to /dev/binder in vendor processes, in case the system variant of the library is loaded, for eg: in libmediandk. Bug: 124128212 Test: Selinux denials realted to/dev/binder acccess are not present when AImageReader from libmediandk is used in a vendor process. Test: Play Youtube movies on Chrome, use camera to take pictures/ record videos (sanity). Change-Id: I27d78e30e16b7df5e3dfbb130121f3d7078671a3 Signed-off-by: Jayant Chowdhary <jchowdhary@google.com>