SF: Fix a couple of Layer ref count issues
This is an attempt at fixing two reference counting issues for Layers.
The first issue is that since we were holding an sp<IBinder> (really a
reference to a LayerCleaner) inside the layer state for deferred
transactions, there was a possibility that it could end up being the
last strong reference to the LayerCleaner such that when it was
destroyed while applying a non-deferred transaction, it would attempt
to grab the SurfaceFlinger main lock to destroy its Layer. Since this
occurred in the main SurfaceFlinger loop, which was already holding
the lock to process transactions, this would cause a deadlock.
To fix this, the sp<IBinder> inside the layer state was changed to a
wp<IBinder>, only being promoted when it actually needs to be accessed
(i.e., when the deferred transaction is created).
The second issue is that we were promoting and holding a strong
reference to a Layer before calling into SurfaceFlinger to destroy it
on the onLayerDestroyed path (triggered when a LayerCleaner is
destroyed). After returning from the attempt to grab the SurfaceFlinger
main lock, it was possible that this strong reference was the last one
keeping the Layer alive, and destroying it at this point could cause
the HWC2 version of the layer to be destroyed at effectively any point,
even between validate/present.
To fix this, the promotion of the weak Layer reference was moved inside
the critical section where the SurfaceFlinger main lock is held.
Test: Cherry-pick from internal branch
Bug: 30503916
Bug: 30281222
Change-Id: I1c6a271f9a7b5d6eea9a9db61d971f262d0cfe84
diff --git a/services/surfaceflinger/Client.cpp b/services/surfaceflinger/Client.cpp
index 2a025b8..3e48cd2 100644
--- a/services/surfaceflinger/Client.cpp
+++ b/services/surfaceflinger/Client.cpp
@@ -43,10 +43,7 @@
{
const size_t count = mLayers.size();
for (size_t i=0 ; i<count ; i++) {
- sp<Layer> layer(mLayers.valueAt(i).promote());
- if (layer != 0) {
- mFlinger->removeLayer(layer);
- }
+ mFlinger->removeLayer(mLayers.valueAt(i));
}
}
diff --git a/services/surfaceflinger/Layer.cpp b/services/surfaceflinger/Layer.cpp
index a512acc..9173165 100644
--- a/services/surfaceflinger/Layer.cpp
+++ b/services/surfaceflinger/Layer.cpp
@@ -1289,9 +1289,14 @@
// If this transaction is waiting on the receipt of a frame, generate a sync
// point and send it to the remote layer.
if (mCurrentState.handle != nullptr) {
- sp<Handle> handle = static_cast<Handle*>(mCurrentState.handle.get());
- sp<Layer> handleLayer = handle->owner.promote();
- if (handleLayer == nullptr) {
+ sp<IBinder> strongBinder = mCurrentState.handle.promote();
+ sp<Handle> handle = nullptr;
+ sp<Layer> handleLayer = nullptr;
+ if (strongBinder != nullptr) {
+ handle = static_cast<Handle*>(strongBinder.get());
+ handleLayer = handle->owner.promote();
+ }
+ if (strongBinder == nullptr || handleLayer == nullptr) {
ALOGE("[%s] Unable to promote Layer handle", mName.string());
// If we can't promote the layer we are intended to wait on,
// then it is expired or otherwise invalid. Allow this transaction
diff --git a/services/surfaceflinger/Layer.h b/services/surfaceflinger/Layer.h
index 55e3b54..c96e7d5 100644
--- a/services/surfaceflinger/Layer.h
+++ b/services/surfaceflinger/Layer.h
@@ -124,7 +124,7 @@
// If set, defers this state update until the Layer identified by handle
// receives a frame with the given frameNumber
- sp<IBinder> handle;
+ wp<IBinder> handle;
uint64_t frameNumber;
// the transparentRegion hint is a bit special, it's latched only
diff --git a/services/surfaceflinger/SurfaceFlinger.cpp b/services/surfaceflinger/SurfaceFlinger.cpp
index a4e3568..47f5401 100644
--- a/services/surfaceflinger/SurfaceFlinger.cpp
+++ b/services/surfaceflinger/SurfaceFlinger.cpp
@@ -2106,8 +2106,14 @@
return NO_ERROR;
}
-status_t SurfaceFlinger::removeLayer(const sp<Layer>& layer) {
+status_t SurfaceFlinger::removeLayer(const wp<Layer>& weakLayer) {
Mutex::Autolock _l(mStateLock);
+ sp<Layer> layer = weakLayer.promote();
+ if (layer == nullptr) {
+ // The layer has already been removed, carry on
+ return NO_ERROR;
+ }
+
ssize_t index = mCurrentState.layersSortedByZ.remove(layer);
if (index >= 0) {
mLayersPendingRemoval.push(layer);
@@ -2448,14 +2454,7 @@
{
// called by ~LayerCleaner() when all references to the IBinder (handle)
// are gone
- status_t err = NO_ERROR;
- sp<Layer> l(layer.promote());
- if (l != NULL) {
- err = removeLayer(l);
- ALOGE_IF(err<0 && err != NAME_NOT_FOUND,
- "error removing layer=%p (%s)", l.get(), strerror(-err));
- }
- return err;
+ return removeLayer(layer);
}
// ---------------------------------------------------------------------------
diff --git a/services/surfaceflinger/SurfaceFlinger.h b/services/surfaceflinger/SurfaceFlinger.h
index 633e956..8279a85 100644
--- a/services/surfaceflinger/SurfaceFlinger.h
+++ b/services/surfaceflinger/SurfaceFlinger.h
@@ -311,7 +311,7 @@
status_t onLayerDestroyed(const wp<Layer>& layer);
// remove a layer from SurfaceFlinger immediately
- status_t removeLayer(const sp<Layer>& layer);
+ status_t removeLayer(const wp<Layer>& layer);
// add a layer to SurfaceFlinger
status_t addClientLayer(const sp<Client>& client,
diff --git a/services/surfaceflinger/SurfaceFlinger_hwc1.cpp b/services/surfaceflinger/SurfaceFlinger_hwc1.cpp
index 7f3b269..072de81 100644
--- a/services/surfaceflinger/SurfaceFlinger_hwc1.cpp
+++ b/services/surfaceflinger/SurfaceFlinger_hwc1.cpp
@@ -2096,8 +2096,14 @@
return NO_ERROR;
}
-status_t SurfaceFlinger::removeLayer(const sp<Layer>& layer) {
+status_t SurfaceFlinger::removeLayer(const wp<Layer>& weakLayer) {
Mutex::Autolock _l(mStateLock);
+ sp<Layer> layer = weakLayer.promote();
+ if (layer == nullptr) {
+ // The layer has already been removed, carry on
+ return NO_ERROR;
+ }
+
ssize_t index = mCurrentState.layersSortedByZ.remove(layer);
if (index >= 0) {
mLayersPendingRemoval.push(layer);
@@ -2438,14 +2444,7 @@
{
// called by ~LayerCleaner() when all references to the IBinder (handle)
// are gone
- status_t err = NO_ERROR;
- sp<Layer> l(layer.promote());
- if (l != NULL) {
- err = removeLayer(l);
- ALOGE_IF(err<0 && err != NAME_NOT_FOUND,
- "error removing layer=%p (%s)", l.get(), strerror(-err));
- }
- return err;
+ return removeLayer(layer);
}
// ---------------------------------------------------------------------------