diff --git a/modules/gralloc/Android.mk b/modules/gralloc/Android.mk
index 1603f2d..8f044c5 100644
--- a/modules/gralloc/Android.mk
+++ b/modules/gralloc/Android.mk
@@ -21,7 +21,13 @@
 LOCAL_PRELINK_MODULE := false
 LOCAL_MODULE_PATH := $(TARGET_OUT_SHARED_LIBRARIES)/hw
 LOCAL_SHARED_LIBRARIES := liblog libcutils
-LOCAL_SRC_FILES := gralloc.cpp mapper.cpp framebuffer.cpp
+
+LOCAL_SRC_FILES := 	\
+	allocator.cpp 	\
+	gralloc.cpp 	\
+	framebuffer.cpp \
+	mapper.cpp
+	
 LOCAL_MODULE := gralloc.default
 LOCAL_CFLAGS:= -DLOG_TAG=\"gralloc\"
 include $(BUILD_SHARED_LIBRARY)
diff --git a/modules/gralloc/allocator.cpp b/modules/gralloc/allocator.cpp
new file mode 100644
index 0000000..f901741
--- /dev/null
+++ b/modules/gralloc/allocator.cpp
@@ -0,0 +1,154 @@
+/*
+ * Copyright (C) 2009 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+#include <cutils/log.h>
+
+#include "allocator.h"
+
+
+// align all the memory blocks on a cache-line boundary
+const int SimpleBestFitAllocator::kMemoryAlign = 32;
+
+SimpleBestFitAllocator::SimpleBestFitAllocator(size_t size)
+{
+    size_t pagesize = getpagesize();
+    mHeapSize = ((size + pagesize-1) & ~(pagesize-1));
+
+    chunk_t* node = new chunk_t(0, mHeapSize / kMemoryAlign);
+    mList.insertHead(node);
+}
+
+SimpleBestFitAllocator::~SimpleBestFitAllocator()
+{
+    while(!mList.isEmpty()) {
+        delete mList.remove(mList.head());
+    }
+}
+
+size_t SimpleBestFitAllocator::size() const
+{
+    return mHeapSize;
+}
+
+size_t SimpleBestFitAllocator::allocate(size_t size, uint32_t flags)
+{
+    Locker::Autolock _l(mLock);
+    ssize_t offset = alloc(size, flags);
+    return offset;
+}
+
+ssize_t SimpleBestFitAllocator::deallocate(size_t offset)
+{
+    Locker::Autolock _l(mLock);
+    chunk_t const * const freed = dealloc(offset);
+    if (freed) {
+        return 0;
+    }
+    return -ENOENT;
+}
+
+ssize_t SimpleBestFitAllocator::alloc(size_t size, uint32_t flags)
+{
+    if (size == 0) {
+        return 0;
+    }
+    size = (size + kMemoryAlign-1) / kMemoryAlign;
+    chunk_t* free_chunk = 0;
+    chunk_t* cur = mList.head();
+
+    size_t pagesize = getpagesize();
+    while (cur) {
+        int extra = ( -cur->start & ((pagesize/kMemoryAlign)-1) ) ;
+
+        // best fit
+        if (cur->free && (cur->size >= (size+extra))) {
+            if ((!free_chunk) || (cur->size < free_chunk->size)) {
+                free_chunk = cur;
+            }
+            if (cur->size == size) {
+                break;
+            }
+        }
+        cur = cur->next;
+    }
+
+    if (free_chunk) {
+        const size_t free_size = free_chunk->size;
+        free_chunk->free = 0;
+        free_chunk->size = size;
+        if (free_size > size) {
+            int extra = ( -free_chunk->start & ((pagesize/kMemoryAlign)-1) ) ;
+            if (extra) {
+                chunk_t* split = new chunk_t(free_chunk->start, extra);
+                free_chunk->start += extra;
+                mList.insertBefore(free_chunk, split);
+            }
+
+            LOGE_IF(((free_chunk->start*kMemoryAlign)&(pagesize-1)),
+                    "page is not aligned!!!");
+
+            const ssize_t tail_free = free_size - (size+extra);
+            if (tail_free > 0) {
+                chunk_t* split = new chunk_t(
+                        free_chunk->start + free_chunk->size, tail_free);
+                mList.insertAfter(free_chunk, split);
+            }
+        }
+        return (free_chunk->start)*kMemoryAlign;
+    }
+    return -ENOMEM;
+}
+
+SimpleBestFitAllocator::chunk_t* SimpleBestFitAllocator::dealloc(size_t start)
+{
+    start = start / kMemoryAlign;
+    chunk_t* cur = mList.head();
+    while (cur) {
+        if (cur->start == start) {
+            LOG_FATAL_IF(cur->free,
+                "block at offset 0x%08lX of size 0x%08lX already freed",
+                cur->start*kMemoryAlign, cur->size*kMemoryAlign);
+
+            // merge freed blocks together
+            chunk_t* freed = cur;
+            cur->free = 1;
+            do {
+                chunk_t* const p = cur->prev;
+                chunk_t* const n = cur->next;
+                if (p && (p->free || !cur->size)) {
+                    freed = p;
+                    p->size += cur->size;
+                    mList.remove(cur);
+                    delete cur;
+                }
+                cur = n;
+            } while (cur && cur->free);
+
+            #ifndef NDEBUG
+                if (!freed->free) {
+                    dump_l("dealloc (!freed->free)");
+                }
+            #endif
+            LOG_FATAL_IF(!freed->free,
+                "freed block at offset 0x%08lX of size 0x%08lX is not free!",
+                freed->start * kMemoryAlign, freed->size * kMemoryAlign);
+
+            return freed;
+        }
+        cur = cur->next;
+    }
+    return 0;
+}
diff --git a/modules/gralloc/allocator.h b/modules/gralloc/allocator.h
new file mode 100644
index 0000000..dfb06f6
--- /dev/null
+++ b/modules/gralloc/allocator.h
@@ -0,0 +1,127 @@
+/*
+ * Copyright (C) 2009 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+
+#ifndef GRALLOC_ALLOCATOR_H_
+#define GRALLOC_ALLOCATOR_H_
+
+#include <stdint.h>
+#include <sys/types.h>
+
+#include "gralloc_priv.h"
+
+// ----------------------------------------------------------------------------
+
+/*
+ * A simple templatized doubly linked-list implementation
+ */
+
+template <typename NODE>
+class LinkedList
+{
+    NODE*  mFirst;
+    NODE*  mLast;
+
+public:
+                LinkedList() : mFirst(0), mLast(0) { }
+    bool        isEmpty() const { return mFirst == 0; }
+    NODE const* head() const { return mFirst; }
+    NODE*       head() { return mFirst; }
+    NODE const* tail() const { return mLast; }
+    NODE*       tail() { return mLast; }
+
+    void insertAfter(NODE* node, NODE* newNode) {
+        newNode->prev = node;
+        newNode->next = node->next;
+        if (node->next == 0) mLast = newNode;
+        else                 node->next->prev = newNode;
+        node->next = newNode;
+    }
+
+    void insertBefore(NODE* node, NODE* newNode) {
+         newNode->prev = node->prev;
+         newNode->next = node;
+         if (node->prev == 0)   mFirst = newNode;
+         else                   node->prev->next = newNode;
+         node->prev = newNode;
+    }
+
+    void insertHead(NODE* newNode) {
+        if (mFirst == 0) {
+            mFirst = mLast = newNode;
+            newNode->prev = newNode->next = 0;
+        } else {
+            newNode->prev = 0;
+            newNode->next = mFirst;
+            mFirst->prev = newNode;
+            mFirst = newNode;
+        }
+    }
+    
+    void insertTail(NODE* newNode) {
+        if (mLast == 0) {
+            insertHead(newNode);
+        } else {
+            newNode->prev = mLast;
+            newNode->next = 0;
+            mLast->next = newNode;
+            mLast = newNode;
+        }
+    }
+
+    NODE* remove(NODE* node) {
+        if (node->prev == 0)    mFirst = node->next;
+        else                    node->prev->next = node->next;
+        if (node->next == 0)    mLast = node->prev;
+        else                    node->next->prev = node->prev;
+        return node;
+    }
+};
+
+class SimpleBestFitAllocator
+{
+public:
+
+                        SimpleBestFitAllocator(size_t size);
+    virtual             ~SimpleBestFitAllocator();
+
+    virtual size_t      allocate(size_t size, uint32_t flags = 0);
+    virtual ssize_t     deallocate(size_t offset);
+    virtual size_t      size() const;
+
+private:
+    struct chunk_t {
+        chunk_t(size_t start, size_t size) 
+            : start(start), size(size), free(1), prev(0), next(0) {
+        }
+        size_t              start;
+        size_t              size : 28;
+        int                 free : 4;
+        mutable chunk_t*    prev;
+        mutable chunk_t*    next;
+    };
+
+    ssize_t  alloc(size_t size, uint32_t flags);
+    chunk_t* dealloc(size_t start);
+
+    static const int    kMemoryAlign;
+    mutable Locker      mLock;
+    LinkedList<chunk_t> mList;
+    size_t              mHeapSize;
+};
+
+
+#endif /* GRALLOC_ALLOCATOR_H_ */
diff --git a/modules/gralloc/gralloc.cpp b/modules/gralloc/gralloc.cpp
index 6ec65e9..b103daa 100644
--- a/modules/gralloc/gralloc.cpp
+++ b/modules/gralloc/gralloc.cpp
@@ -15,25 +15,29 @@
  */
 
 #include <limits.h>
-
-#include <sys/mman.h>
-#include <sys/stat.h>
-#include <sys/types.h>
-
-#include <cutils/ashmem.h>
-#include <cutils/log.h>
-
-#include <hardware/hardware.h>
-#include <hardware/gralloc.h>
-
+#include <unistd.h>
 #include <fcntl.h>
 #include <errno.h>
 #include <pthread.h>
 
+#include <sys/mman.h>
+#include <sys/stat.h>
+#include <sys/types.h>
+#include <sys/ioctl.h>
+
+#include <cutils/ashmem.h>
 #include <cutils/log.h>
 #include <cutils/atomic.h>
 
+#include <hardware/hardware.h>
+#include <hardware/gralloc.h>
+
 #include "gralloc_priv.h"
+#include "allocator.h"
+
+#if HAVE_ANDROID_OS
+#include <linux/android_pmem.h>
+#endif
 
 /*****************************************************************************/
 
@@ -94,7 +98,9 @@
     numBuffers: 0,
     bufferMask: 0,
     lock: PTHREAD_MUTEX_INITIALIZER,
-    currentBuffer: 0
+    currentBuffer: 0,
+    pmem_master: -1,
+    pmem_master_base: 0
 };
 
 /*****************************************************************************/
@@ -164,32 +170,104 @@
     return err;
 }
 
+static SimpleBestFitAllocator sAllocator(8*1024*1024);
+
+static int init_pmem_area(private_module_t* m)
+{
+    int err = 0;
+    int master_fd = open("/dev/pmem", O_RDWR, 0);
+    if (master_fd >= 0) {
+        void* base = mmap(0, sAllocator.size(),
+                PROT_READ|PROT_WRITE, MAP_SHARED, master_fd, 0);
+        if (base == MAP_FAILED) {
+            err = -errno;
+            base = 0;
+            close(master_fd);
+            master_fd = -1;
+        }
+        m->pmem_master = master_fd;
+        m->pmem_master_base = base;
+    } else {
+        err = -errno;
+    }
+    return err;
+}
 
 static int gralloc_alloc_buffer(alloc_device_t* dev,
         size_t size, int usage, buffer_handle_t* pHandle)
 {
-    size = roundUpToPageSize(size);
+    int err = 0;
     int flags = 0;
+
+    int fd = -1;
+    void* base = 0;
+    int offset = 0;
+    int lockState = 0;
+
+    size = roundUpToPageSize(size);
+    
+    if (usage & GRALLOC_USAGE_HW_TEXTURE) {
+        // enable pmem in that case, so our software GL can fallback to
+        // the copybit module.
+        flags |= private_handle_t::PRIV_FLAGS_USES_PMEM;
+    }
+    
     if (usage & GRALLOC_USAGE_HW_2D) {
         flags |= private_handle_t::PRIV_FLAGS_USES_PMEM;
     }
-    int fd;
+
     if ((flags & private_handle_t::PRIV_FLAGS_USES_PMEM) == 0) {
         fd = ashmem_create_region("Buffer", size);
+        if (fd < 0) {
+            err = -errno;
+        }
     } else {
-        fd = open("/dev/pmem", O_RDWR, 0);
-        // Note: Currently pmem get sized when it is mmaped.
-        // This means that just doing the open doesn't actually allocate
-        // anything. We basically need to do an implicit "mmap"
-        // (under the hood) for pmem regions. However, the current
-        // code will work okay for now thanks to the reference-counting.
+        private_module_t* m = reinterpret_cast<private_module_t*>(
+                dev->common.module);
+        
+        pthread_mutex_lock(&m->lock);
+        if (m->pmem_master == -1)
+            err = init_pmem_area(m);
+        pthread_mutex_unlock(&m->lock);
+        
+        if (m->pmem_master >= 0) {
+            // PMEM buffers are always mmapped
+            base = m->pmem_master_base;
+            lockState |= private_handle_t::LOCK_STATE_MAPPED;
+
+            offset = sAllocator.allocate(size);
+            if (offset < 0) {
+                err = -ENOMEM;
+            } else {
+                fd = open("/dev/pmem", O_RDWR, 0);
+                err = ioctl(fd, PMEM_CONNECT, m->pmem_master);
+                if (err < 0) {
+                    err = -errno;
+                } else {
+                    struct pmem_region sub = { offset, size };
+                    err = ioctl(fd, PMEM_MAP, &sub);
+                }
+                if (err < 0) {
+                    close(fd);
+                    sAllocator.deallocate(offset);
+                    fd = -1;
+                }
+                LOGD_IF(!err, "allocating pmem size=%d, offset=%d", size, offset);
+            }
+        }
     }
-    if (fd < 0) {
-        return -errno;
+
+    if (err == 0) {
+        private_handle_t* hnd = new private_handle_t(fd, size, flags);
+        hnd->offset = offset;
+        hnd->base = int(base)+offset;
+        hnd->lockState = lockState;
+        *pHandle = hnd;
     }
-    private_handle_t* hnd = new private_handle_t(fd, size, flags);
-    *pHandle = hnd;
-    return 0;
+    
+    LOGE_IF(err, "gralloc failed err=%s", strerror(-err));
+    
+    return err;
 }
 
 /*****************************************************************************/
@@ -249,6 +327,10 @@
         const size_t bufferSize = m->finfo.line_length * m->info.yres;
         int index = (hnd->base - m->framebuffer->base) / bufferSize;
         m->bufferMask &= ~(1<<index); 
+    } else if (true || hnd->flags & private_handle_t::PRIV_FLAGS_USES_PMEM) {
+        if (hnd->fd >= 0) {
+            sAllocator.deallocate(hnd->offset);
+       }
     }
 
     gralloc_module_t* m = reinterpret_cast<gralloc_module_t*>(
diff --git a/modules/gralloc/gralloc_priv.h b/modules/gralloc/gralloc_priv.h
index 4d5afdc..8adb4f4 100644
--- a/modules/gralloc/gralloc_priv.h
+++ b/modules/gralloc/gralloc_priv.h
@@ -23,6 +23,8 @@
 #include <sys/cdefs.h>
 #include <hardware/gralloc.h>
 #include <pthread.h>
+#include <errno.h>
+#include <unistd.h>
 
 #include <cutils/native_handle.h>
 
@@ -40,6 +42,23 @@
 
 /*****************************************************************************/
 
+class Locker {
+    pthread_mutex_t mutex;
+public:
+    class Autolock {
+        Locker& locker;
+    public:
+        inline Autolock(Locker& locker) : locker(locker) {  locker.lock(); }
+        inline ~Autolock() { locker.unlock(); }
+    };
+    inline Locker()        { pthread_mutex_init(&mutex, 0); }
+    inline ~Locker()       { pthread_mutex_destroy(&mutex); }
+    inline void lock()     { pthread_mutex_lock(&mutex); }
+    inline void unlock()   { pthread_mutex_unlock(&mutex); }
+};
+
+/*****************************************************************************/
+
 struct private_handle_t;
 
 struct private_module_t {
@@ -51,6 +70,9 @@
     uint32_t bufferMask;
     pthread_mutex_t lock;
     buffer_handle_t currentBuffer;
+    int pmem_master;
+    void* pmem_master_base;
+
     struct fb_var_screeninfo info;
     struct fb_fix_screeninfo finfo;
     float xdpi;
@@ -82,25 +104,25 @@
     int     magic;
     int     flags;
     int     size;
-    int     offset; // used with copybit
+    int     offset;
     // FIXME: the attributes below should be out-of-line
     int     base;
     int     lockState;
     int     writeOwner;
+    int     pid;
 
-    static const int sNumInts = 7;
+    static const int sNumInts = 8;
     static const int sNumFds = 1;
     static const int sMagic = 0x3141592;
 
     private_handle_t(int fd, int size, int flags) :
         fd(fd), magic(sMagic), flags(flags), size(size), offset(0),
-        base(0), lockState(0), writeOwner(0) 
+        base(0), lockState(0), writeOwner(0), pid(getpid())
     {
         version = sizeof(native_handle);
         numInts = sNumInts;
         numFds = sNumFds;
     }
-
     ~private_handle_t() {
         magic = 0;
     }
diff --git a/modules/gralloc/mapper.cpp b/modules/gralloc/mapper.cpp
index f13c315..1ab70ec 100644
--- a/modules/gralloc/mapper.cpp
+++ b/modules/gralloc/mapper.cpp
@@ -17,6 +17,7 @@
 #include <limits.h>
 #include <errno.h>
 #include <pthread.h>
+#include <unistd.h>
 
 #include <sys/mman.h>
 #include <sys/stat.h>
@@ -30,6 +31,10 @@
 
 #include "gralloc_priv.h"
 
+
+// we need this for now because pmem cannot mmap at an offset
+#define PMEM_HACK   1
+
 /*****************************************************************************/
 
 static int gralloc_map(gralloc_module_t const* module,
@@ -38,12 +43,19 @@
 {
     private_handle_t* hnd = (private_handle_t*)handle;
     if (!(hnd->flags & private_handle_t::PRIV_FLAGS_FRAMEBUFFER)) {
-        void* mappedAddress = mmap(0, hnd->size, PROT_READ|PROT_WRITE,
-                MAP_SHARED, hnd->fd, 0);
+        size_t size = hnd->size;
+#if PMEM_HACK
+        size += hnd->offset;
+#endif
+        void* mappedAddress = mmap(0, size,
+                PROT_READ|PROT_WRITE, MAP_SHARED, hnd->fd, 0);
         if (mappedAddress == MAP_FAILED) {
+            LOGE("Could not mmap %s", strerror(errno));
             return -errno;
         }
-        hnd->base = intptr_t(mappedAddress);
+        hnd->base = intptr_t(mappedAddress) + hnd->offset;
+        //LOGD("gralloc_map() succeeded fd=%d, off=%d, size=%d, vaddr=%p", 
+        //        hnd->fd, hnd->offset, hnd->size, mappedAddress);
     }
     *vaddr = (void*)hnd->base;
     return 0;
@@ -55,7 +67,7 @@
     private_handle_t* hnd = (private_handle_t*)handle;
     if (!(hnd->flags & private_handle_t::PRIV_FLAGS_FRAMEBUFFER)) {
         if (munmap((void*)hnd->base, hnd->size) < 0) {
-            LOGE("Could not unmap %d", errno);
+            LOGE("Could not unmap %s", strerror(errno));
         }
     }
     hnd->base = 0;
@@ -82,10 +94,14 @@
      * handle, but instead maintained in the kernel or at least 
      * out-of-line
      */ 
+
+    // if this handle was created in this process, then we keep it as is.
     private_handle_t* hnd = (private_handle_t*)handle;
-    hnd->base = 0;
-    hnd->lockState  = 0;
-    hnd->writeOwner = 0;
+    if (hnd->pid != getpid()) {
+        hnd->base = 0;
+        hnd->lockState  = 0;
+        hnd->writeOwner = 0;
+    }
     return 0;
 }
 
@@ -102,20 +118,20 @@
      */
 
     private_handle_t* hnd = (private_handle_t*)handle;
-    int32_t current_value, new_value;
-    int retry;
     
     LOGE_IF(hnd->lockState & private_handle_t::LOCK_STATE_READ_MASK,
             "handle %p still locked (state=%08x)",
             hnd, hnd->lockState);
 
-    if (hnd->lockState & private_handle_t::LOCK_STATE_MAPPED) {
-        gralloc_unmap(module, handle);
+    // never unmap buffers that were created in this process
+    if (hnd->pid != getpid()) {
+        if (hnd->lockState & private_handle_t::LOCK_STATE_MAPPED) {
+            gralloc_unmap(module, handle);
+        }
+        hnd->base = 0;
+        hnd->lockState  = 0;
+        hnd->writeOwner = 0;
     }
-
-    hnd->base = 0;
-    hnd->lockState  = 0;
-    hnd->writeOwner = 0;
     return 0;
 }