diff --git a/libs/gui/BufferQueue.cpp b/libs/gui/BufferQueue.cpp
index f99bd29..cf81431 100644
--- a/libs/gui/BufferQueue.cpp
+++ b/libs/gui/BufferQueue.cpp
@@ -30,6 +30,7 @@
 
 #include <utils/Log.h>
 #include <utils/Trace.h>
+#include <utils/CallStack.h>
 
 // Macros for including the BufferQueue name in log messages
 #define ST_LOGV(x, ...) ALOGV("[%s] "x, mConsumerName.string(), ##__VA_ARGS__)
@@ -150,21 +151,21 @@
         }
 
         // Error out if the user has dequeued buffers
-        int maxBufferCount = getMaxBufferCountLocked();
-        for (int i=0 ; i<maxBufferCount; i++) {
+        for (int i=0 ; i<NUM_BUFFER_SLOTS; i++) {
             if (mSlots[i].mBufferState == BufferSlot::DEQUEUED) {
                 ST_LOGE("setBufferCount: client owns some buffers");
                 return -EINVAL;
             }
         }
 
-        const int minBufferSlots = getMinMaxBufferCountLocked();
         if (bufferCount == 0) {
             mOverrideMaxBufferCount = 0;
             mDequeueCondition.broadcast();
             return NO_ERROR;
         }
 
+        // fine to assume async to false before we're setting the buffer count
+        const int minBufferSlots = getMinMaxBufferCountLocked(false);
         if (bufferCount < minBufferSlots) {
             ST_LOGE("setBufferCount: requested buffer count (%d) is less than "
                     "minimum (%d)", bufferCount, minBufferSlots);
@@ -209,7 +210,7 @@
         value = mDefaultBufferFormat;
         break;
     case NATIVE_WINDOW_MIN_UNDEQUEUED_BUFFERS:
-        value = mMinUndequeuedBufferCount;
+        value = getMinUndequeuedBufferCount(false);
         break;
     case NATIVE_WINDOW_CONSUMER_RUNNING_BEHIND:
         value = (mQueue.size() >= 2);
@@ -229,15 +230,11 @@
         ST_LOGE("requestBuffer: BufferQueue has been abandoned!");
         return NO_INIT;
     }
-    int maxBufferCount = getMaxBufferCountLocked();
-    if (slot < 0 || maxBufferCount <= slot) {
+    if (slot < 0 || slot >= NUM_BUFFER_SLOTS) {
         ST_LOGE("requestBuffer: slot index out of range [0, %d]: %d",
-                maxBufferCount, slot);
+                NUM_BUFFER_SLOTS, slot);
         return BAD_VALUE;
     } else if (mSlots[slot].mBufferState != BufferSlot::DEQUEUED) {
-        // XXX: I vaguely recall there was some reason this can be valid, but
-        // for the life of me I can't recall under what circumstances that's
-        // the case.
         ST_LOGE("requestBuffer: slot %d is not owned by the client (state=%d)",
                 slot, mSlots[slot].mBufferState);
         return BAD_VALUE;
@@ -247,7 +244,7 @@
     return NO_ERROR;
 }
 
-status_t BufferQueue::dequeueBuffer(int *outBuf, sp<Fence>* outFence,
+status_t BufferQueue::dequeueBuffer(int *outBuf, sp<Fence>* outFence, bool async,
         uint32_t w, uint32_t h, uint32_t format, uint32_t usage) {
     ATRACE_CALL();
     ST_LOGV("dequeueBuffer: w=%d h=%d fmt=%#x usage=%#x", w, h, format, usage);
@@ -279,7 +276,16 @@
                 return NO_INIT;
             }
 
-            const int maxBufferCount = getMaxBufferCountLocked();
+            const int maxBufferCount = getMaxBufferCountLocked(async);
+            if (async && mOverrideMaxBufferCount) {
+                // FIXME: some drivers are manually setting the buffer-count (which they
+                // shouldn't), so we do this extra test here to handle that case.
+                // This is TEMPORARY, until we get this fixed.
+                if (mOverrideMaxBufferCount < maxBufferCount) {
+                    ST_LOGE("dequeueBuffer: async mode is invalid with buffercount override");
+                    return BAD_VALUE;
+                }
+            }
 
             // Free up any buffers that are in slots beyond the max buffer
             // count.
@@ -328,7 +334,7 @@
                 // make sure the client is not trying to dequeue more buffers
                 // than allowed.
                 const int newUndequeuedCount = maxBufferCount - (dequeuedCount+1);
-                const int minUndequeuedCount = mMinUndequeuedBufferCount;
+                const int minUndequeuedCount = getMinUndequeuedBufferCount(async);
                 if (newUndequeuedCount < minUndequeuedCount) {
                     ST_LOGE("dequeueBuffer: min undequeued buffer count (%d) "
                             "exceeded (dequeued=%d undequeudCount=%d)",
@@ -448,9 +454,10 @@
     uint32_t transform;
     int scalingMode;
     int64_t timestamp;
+    bool async;
     sp<Fence> fence;
 
-    input.deflate(&timestamp, &crop, &scalingMode, &transform, &fence);
+    input.deflate(&timestamp, &crop, &scalingMode, &transform, &async, &fence);
 
     if (fence == NULL) {
         ST_LOGE("queueBuffer: fence is NULL");
@@ -477,7 +484,17 @@
             ST_LOGE("queueBuffer: BufferQueue has been abandoned!");
             return NO_INIT;
         }
-        int maxBufferCount = getMaxBufferCountLocked();
+
+        const int maxBufferCount = getMaxBufferCountLocked(async);
+        if (async && mOverrideMaxBufferCount) {
+            // FIXME: some drivers are manually setting the buffer-count (which they
+            // shouldn't), so we do this extra test here to handle that case.
+            // This is TEMPORARY, until we get this fixed.
+            if (mOverrideMaxBufferCount < maxBufferCount) {
+                ST_LOGE("queueBuffer: async mode is invalid with buffercount override");
+                return BAD_VALUE;
+            }
+        }
         if (buf < 0 || buf >= maxBufferCount) {
             ST_LOGE("queueBuffer: slot index out of range [0, %d]: %d",
                     maxBufferCount, buf);
@@ -523,7 +540,7 @@
         item.mFrameNumber = mFrameCounter;
         item.mBuf = buf;
         item.mFence = fence;
-        item.mDequeueBufferCannotBlock = mDequeueBufferCannotBlock;
+        item.mIsDroppable = mDequeueBufferCannotBlock || async;
 
         if (mQueue.empty()) {
             // when the queue is empty, we can ignore "mDequeueBufferCannotBlock", and
@@ -534,7 +551,7 @@
             // when the queue is not empty, we need to look at the front buffer
             // state and see if we need to replace it.
             Fifo::iterator front(mQueue.begin());
-            if (front->mDequeueBufferCannotBlock) {
+            if (front->mIsDroppable) {
                 // buffer slot currently queued is marked free if still tracked
                 if (stillTracking(front)) {
                     mSlots[front->mBuf].mBufferState = BufferSlot::FREE;
@@ -573,10 +590,9 @@
         return;
     }
 
-    int maxBufferCount = getMaxBufferCountLocked();
-    if (buf < 0 || buf >= maxBufferCount) {
+    if (buf < 0 || buf >= NUM_BUFFER_SLOTS) {
         ST_LOGE("cancelBuffer: slot index out of range [0, %d]: %d",
-                maxBufferCount, buf);
+                NUM_BUFFER_SLOTS, buf);
         return;
     } else if (mSlots[buf].mBufferState != BufferSlot::DEQUEUED) {
         ST_LOGE("cancelBuffer: slot %d is not owned by the client (state=%d)",
@@ -630,8 +646,6 @@
 
     mBufferHasBeenQueued = false;
     mDequeueBufferCannotBlock = mConsumerControlledByApp && producerControlledByApp;
-    mMinUndequeuedBufferCount = mDequeueBufferCannotBlock ?
-            mMaxAcquiredBufferCount+1 : mMaxAcquiredBufferCount;
 
     return err;
 }
@@ -704,12 +718,11 @@
         fifoSize++;
     }
 
-    int maxBufferCount = getMaxBufferCountLocked();
 
     result.appendFormat(
-            "%s-BufferQueue maxBufferCount=%d, mDequeueBufferCannotBlock=%d, default-size=[%dx%d], "
+            "%s-BufferQueue mMaxAcquiredBufferCount=%d, mDequeueBufferCannotBlock=%d, default-size=[%dx%d], "
             "default-format=%d, transform-hint=%02x, FIFO(%d)={%s}\n",
-            prefix, maxBufferCount, mDequeueBufferCannotBlock, mDefaultWidth,
+            prefix, mMaxAcquiredBufferCount, mDequeueBufferCannotBlock, mDefaultWidth,
             mDefaultHeight, mDefaultBufferFormat, mTransformHint,
             fifoSize, fifo.string());
 
@@ -725,16 +738,25 @@
         }
     } stateName;
 
+    // just trim the free buffers to not spam the dump
+    int maxBufferCount = 0;
+    for (int i=NUM_BUFFER_SLOTS-1 ; i>=0 ; i--) {
+        const BufferSlot& slot(mSlots[i]);
+        if ((slot.mBufferState != BufferSlot::FREE) || (slot.mGraphicBuffer != NULL)) {
+            maxBufferCount = i+1;
+            break;
+        }
+    }
+
     for (int i=0 ; i<maxBufferCount ; i++) {
         const BufferSlot& slot(mSlots[i]);
+        const sp<GraphicBuffer>& buf(slot.mGraphicBuffer);
         result.appendFormat(
             "%s%s[%02d:%p] state=%-8s",
-                prefix, (slot.mBufferState == BufferSlot::ACQUIRED)?">":" ", i,
-                slot.mGraphicBuffer.get(),
+                prefix, (slot.mBufferState == BufferSlot::ACQUIRED)?">":" ", i, buf.get(),
                 stateName(slot.mBufferState)
         );
 
-        const sp<GraphicBuffer>& buf(slot.mGraphicBuffer);
         if (buf != NULL) {
             result.appendFormat(
                     ", %p [%4ux%4u:%4u,%3X]",
@@ -1018,12 +1040,17 @@
     return NO_ERROR;
 }
 
-int BufferQueue::getMinMaxBufferCountLocked() const {
-    return mMinUndequeuedBufferCount + 1;
+int BufferQueue::getMinUndequeuedBufferCount(bool async) const {
+    return (mDequeueBufferCannotBlock || async) ?
+                mMaxAcquiredBufferCount+1 : mMaxAcquiredBufferCount;
 }
 
-int BufferQueue::getMaxBufferCountLocked() const {
-    int minMaxBufferCount = getMinMaxBufferCountLocked();
+int BufferQueue::getMinMaxBufferCountLocked(bool async) const {
+    return getMinUndequeuedBufferCount(async) + 1;
+}
+
+int BufferQueue::getMaxBufferCountLocked(bool async) const {
+    int minMaxBufferCount = getMinMaxBufferCountLocked(async);
 
     int maxBufferCount = mDefaultMaxBufferCount;
     if (maxBufferCount < minMaxBufferCount) {
diff --git a/libs/gui/IGraphicBufferProducer.cpp b/libs/gui/IGraphicBufferProducer.cpp
index 9f65fc3..2e561df 100644
--- a/libs/gui/IGraphicBufferProducer.cpp
+++ b/libs/gui/IGraphicBufferProducer.cpp
@@ -80,10 +80,11 @@
         return result;
     }
 
-    virtual status_t dequeueBuffer(int *buf, sp<Fence>* fence,
+    virtual status_t dequeueBuffer(int *buf, sp<Fence>* fence, bool async,
             uint32_t w, uint32_t h, uint32_t format, uint32_t usage) {
         Parcel data, reply;
         data.writeInterfaceToken(IGraphicBufferProducer::getInterfaceDescriptor());
+        data.writeInt32(async);
         data.writeInt32(w);
         data.writeInt32(h);
         data.writeInt32(format);
@@ -197,13 +198,14 @@
         } break;
         case DEQUEUE_BUFFER: {
             CHECK_INTERFACE(IGraphicBufferProducer, data, reply);
+            bool async      = data.readInt32();
             uint32_t w      = data.readInt32();
             uint32_t h      = data.readInt32();
             uint32_t format = data.readInt32();
             uint32_t usage  = data.readInt32();
             int buf;
             sp<Fence> fence;
-            int result = dequeueBuffer(&buf, &fence, w, h, format, usage);
+            int result = dequeueBuffer(&buf, &fence, async, w, h, format, usage);
             reply->writeInt32(buf);
             reply->writeInt32(fence != NULL);
             if (fence != NULL) {
@@ -274,6 +276,7 @@
          + sizeof(crop)
          + sizeof(scalingMode)
          + sizeof(transform)
+         + sizeof(async)
          + fence->getFlattenedSize();
 }
 
@@ -291,6 +294,7 @@
     memcpy(p, &crop,        sizeof(crop));        p += sizeof(crop);
     memcpy(p, &scalingMode, sizeof(scalingMode)); p += sizeof(scalingMode);
     memcpy(p, &transform,   sizeof(transform));   p += sizeof(transform);
+    memcpy(p, &async,       sizeof(async));       p += sizeof(async);
     err = fence->flatten(p, size - (p - (char*)buffer), fds, count);
     return err;
 }
@@ -304,6 +308,7 @@
     memcpy(&crop,        p, sizeof(crop));        p += sizeof(crop);
     memcpy(&scalingMode, p, sizeof(scalingMode)); p += sizeof(scalingMode);
     memcpy(&transform,   p, sizeof(transform));   p += sizeof(transform);
+    memcpy(&async,       p, sizeof(async));       p += sizeof(async);
     fence = new Fence();
     err = fence->unflatten(p, size - (p - (const char*)buffer), fds, count);
     return err;
diff --git a/libs/gui/Surface.cpp b/libs/gui/Surface.cpp
index 0d4449a..998ea8a 100644
--- a/libs/gui/Surface.cpp
+++ b/libs/gui/Surface.cpp
@@ -73,6 +73,7 @@
     mConsumerRunningBehind = false;
     mConnectedToCpu = false;
     mProducerControlledByApp = true;
+    mSwapIntervalZero = false;
 }
 
 Surface::~Surface() {
@@ -162,7 +163,6 @@
     // EGL specification states:
     //  interval is silently clamped to minimum and maximum implementation
     //  dependent values before being stored.
-    // Although we don't have to, we apply the same logic here.
 
     if (interval < minSwapInterval)
         interval = minSwapInterval;
@@ -170,11 +170,9 @@
     if (interval > maxSwapInterval)
         interval = maxSwapInterval;
 
-    // FIXME: re-implement swap-interval
-    //status_t res = mGraphicBufferProducer->setSynchronousMode(interval ? true : false);
-    status_t res = NO_ERROR;
+    mSwapIntervalZero = (interval == 0);
 
-    return res;
+    return NO_ERROR;
 }
 
 int Surface::dequeueBuffer(android_native_buffer_t** buffer,
@@ -186,7 +184,7 @@
     int reqW = mReqWidth ? mReqWidth : mUserWidth;
     int reqH = mReqHeight ? mReqHeight : mUserHeight;
     sp<Fence> fence;
-    status_t result = mGraphicBufferProducer->dequeueBuffer(&buf, &fence,
+    status_t result = mGraphicBufferProducer->dequeueBuffer(&buf, &fence, mSwapIntervalZero,
             reqW, reqH, mReqFormat, mReqUsage);
     if (result < 0) {
         ALOGV("dequeueBuffer: IGraphicBufferProducer::dequeueBuffer(%d, %d, %d, %d)"
@@ -282,7 +280,7 @@
     sp<Fence> fence(fenceFd >= 0 ? new Fence(fenceFd) : Fence::NO_FENCE);
     IGraphicBufferProducer::QueueBufferOutput output;
     IGraphicBufferProducer::QueueBufferInput input(timestamp, crop, mScalingMode,
-            mTransform, fence);
+            mTransform, mSwapIntervalZero, fence);
     status_t err = mGraphicBufferProducer->queueBuffer(i, input, &output);
     if (err != OK)  {
         ALOGE("queueBuffer: error queuing buffer to SurfaceTexture, %d", err);
diff --git a/libs/gui/tests/BufferQueue_test.cpp b/libs/gui/tests/BufferQueue_test.cpp
index 1f8e7fa..b691fc1 100644
--- a/libs/gui/tests/BufferQueue_test.cpp
+++ b/libs/gui/tests/BufferQueue_test.cpp
@@ -71,12 +71,12 @@
     sp<Fence> fence;
     sp<GraphicBuffer> buf;
     IGraphicBufferProducer::QueueBufferInput qbi(0, Rect(0, 0, 1, 1),
-            NATIVE_WINDOW_SCALING_MODE_FREEZE, 0, Fence::NO_FENCE);
+            NATIVE_WINDOW_SCALING_MODE_FREEZE, 0, false, Fence::NO_FENCE);
     BufferQueue::BufferItem item;
 
     for (int i = 0; i < 2; i++) {
         ASSERT_EQ(IGraphicBufferProducer::BUFFER_NEEDS_REALLOCATION,
-                mBQ->dequeueBuffer(&slot, &fence, 1, 1, 0,
+                mBQ->dequeueBuffer(&slot, &fence, false, 1, 1, 0,
                     GRALLOC_USAGE_SW_READ_OFTEN));
         ASSERT_EQ(OK, mBQ->requestBuffer(slot, &buf));
         ASSERT_EQ(OK, mBQ->queueBuffer(slot, qbi, &qbo));
@@ -84,7 +84,7 @@
     }
 
     ASSERT_EQ(IGraphicBufferProducer::BUFFER_NEEDS_REALLOCATION,
-            mBQ->dequeueBuffer(&slot, &fence, 1, 1, 0,
+            mBQ->dequeueBuffer(&slot, &fence, false, 1, 1, 0,
                 GRALLOC_USAGE_SW_READ_OFTEN));
     ASSERT_EQ(OK, mBQ->requestBuffer(slot, &buf));
     ASSERT_EQ(OK, mBQ->queueBuffer(slot, qbi, &qbo));
diff --git a/libs/gui/tests/SurfaceTextureClient_test.cpp b/libs/gui/tests/SurfaceTextureClient_test.cpp
index 9908cc9..158c94b 100644
--- a/libs/gui/tests/SurfaceTextureClient_test.cpp
+++ b/libs/gui/tests/SurfaceTextureClient_test.cpp
@@ -361,7 +361,6 @@
 
 TEST_F(SurfaceTextureClientTest, SurfaceTextureSyncModeSlowRetire) {
     android_native_buffer_t* buf[3];
-    ASSERT_EQ(OK, mANW->setSwapInterval(mANW.get(), 1));
     ASSERT_EQ(OK, native_window_set_buffer_count(mANW.get(), 4));
     ASSERT_EQ(OK, native_window_dequeue_buffer_and_wait(mANW.get(), &buf[0]));
     ASSERT_EQ(OK, native_window_dequeue_buffer_and_wait(mANW.get(), &buf[1]));
@@ -382,7 +381,6 @@
 
 TEST_F(SurfaceTextureClientTest, SurfaceTextureSyncModeFastRetire) {
     android_native_buffer_t* buf[3];
-    ASSERT_EQ(OK, mANW->setSwapInterval(mANW.get(), 1));
     ASSERT_EQ(OK, native_window_set_buffer_count(mANW.get(), 4));
     ASSERT_EQ(OK, native_window_dequeue_buffer_and_wait(mANW.get(), &buf[0]));
     ASSERT_EQ(OK, native_window_dequeue_buffer_and_wait(mANW.get(), &buf[1]));
@@ -403,7 +401,6 @@
 
 TEST_F(SurfaceTextureClientTest, SurfaceTextureSyncModeDQQR) {
     android_native_buffer_t* buf[3];
-    ASSERT_EQ(OK, mANW->setSwapInterval(mANW.get(), 1));
     ASSERT_EQ(OK, native_window_set_buffer_count(mANW.get(), 3));
 
     ASSERT_EQ(OK, native_window_dequeue_buffer_and_wait(mANW.get(), &buf[0]));
@@ -429,7 +426,6 @@
 TEST_F(SurfaceTextureClientTest, DISABLED_SurfaceTextureSyncModeDequeueCurrent) {
     android_native_buffer_t* buf[3];
     android_native_buffer_t* firstBuf;
-    ASSERT_EQ(OK, mANW->setSwapInterval(mANW.get(), 1));
     ASSERT_EQ(OK, native_window_set_buffer_count(mANW.get(), 3));
     ASSERT_EQ(OK, native_window_dequeue_buffer_and_wait(mANW.get(), &firstBuf));
     ASSERT_EQ(OK, mANW->queueBuffer(mANW.get(), firstBuf, -1));
@@ -449,7 +445,6 @@
 
 TEST_F(SurfaceTextureClientTest, SurfaceTextureSyncModeMinUndequeued) {
     android_native_buffer_t* buf[3];
-    ASSERT_EQ(OK, mANW->setSwapInterval(mANW.get(), 1));
     ASSERT_EQ(OK, native_window_set_buffer_count(mANW.get(), 3));
 
     // We should be able to dequeue all the buffers before we've queued mANWy.
@@ -528,7 +523,6 @@
     };
 
     android_native_buffer_t* buf[3];
-    ASSERT_EQ(OK, mANW->setSwapInterval(mANW.get(), 1));
     ASSERT_EQ(OK, native_window_set_buffer_count(mANW.get(), 3));
     // dequeue/queue/update so we have a current buffer
     ASSERT_EQ(OK, native_window_dequeue_buffer_and_wait(mANW.get(), &buf[0]));