| Révision | 87d725793ec559e1f1ad67108a2c4806313e8a26 (tree) |
|---|---|
| l'heure | 2020-05-21 11:16:53 |
| Auteur | Matt Moeller <matthew.moeller@kinz...> |
| Commiter | Chih-Wei Huang |
ISurfaceComposer: Workaround for screenshots on intel graphics
Prevent the GraphicBuffer destructor until after the screencap client
has had a chance to consume it.
libs/gui/ISurfaceComposer.cpp (diff)| @@ -19,6 +19,7 @@ | ||
| 19 | 19 | |
| 20 | 20 | #include <stdint.h> |
| 21 | 21 | #include <sys/types.h> |
| 22 | +#include <thread> | |
| 22 | 23 | |
| 23 | 24 | #include <binder/Parcel.h> |
| 24 | 25 | #include <binder/IPCThreadState.h> |
| @@ -631,6 +632,20 @@ status_t BnSurfaceComposer::onTransact( | ||
| 631 | 632 | reply->write(*outBuffer); |
| 632 | 633 | reply->writeBool(capturedSecureLayers); |
| 633 | 634 | } |
| 635 | + | |
| 636 | + // XXX | |
| 637 | + // When outBuffer goes out of scope, the destructor gets called and on some | |
| 638 | + // graphics platforms (intel i915) this will actually free the underlying buffer | |
| 639 | + // containing the screen capture. The guy on the other end of this binder transaction | |
| 640 | + // will try to open the buffer that has been freed and fail. | |
| 641 | + // | |
| 642 | + // The super-hacky workaround here is to delay the destructor call in a thread so | |
| 643 | + // the binder client can obtain the buffer in question. When the thread timeout expires | |
| 644 | + // the destructor will get called and clean up the buffer so there is no memory leak. | |
| 645 | + std::thread([outBuffer]() { | |
| 646 | + std::this_thread::sleep_for(std::chrono::seconds(1)); | |
| 647 | + }).detach(); | |
| 648 | + | |
| 634 | 649 | return NO_ERROR; |
| 635 | 650 | } |
| 636 | 651 | case CAPTURE_LAYERS: { |
Android-x86
Fork