diff --git a/libgralloc-qsd8k/gpu.cpp b/libgralloc-qsd8k/gpu.cpp
index 9006528..b3a83ce 100755
--- a/libgralloc-qsd8k/gpu.cpp
+++ b/libgralloc-qsd8k/gpu.cpp
@@ -218,7 +218,7 @@ int gpu_context_t::gralloc_alloc_buffer(size_t size, int usage, buffer_handle_t*
         // PMEM buffers are always mmapped
         lockState |= private_handle_t::LOCK_STATE_MAPPED;
 
-        err = pma->alloc_pmem_buffer(size, usage, &base, &offset, &fd, &masterFd);
+        err = pma->alloc_pmem_buffer(size, usage, &base, &offset, &fd, &masterFd, format);
         if (err < 0) {
             if (((usage & GRALLOC_USAGE_HW_MASK) == 0) &&
                 ((usage & GRALLOC_USAGE_PRIVATE_PMEM_ADSP) == 0)) {
diff --git a/libgralloc-qsd8k/gralloc.cpp b/libgralloc-qsd8k/gralloc.cpp
index 019a8cd..16f026e 100755
--- a/libgralloc-qsd8k/gralloc.cpp
+++ b/libgralloc-qsd8k/gralloc.cpp
@@ -96,6 +96,13 @@ class PmemAllocatorDepsDeviceImpl : public PmemUserspaceAllocator::Deps,
         return ioctl(fd, PMEM_UNMAP, &sub);
     }
 
+    virtual int alignPmem(int fd, size_t size, int align) {
+        struct pmem_allocation allocation;
+        allocation.size = size;
+        allocation.align = align;
+        return ioctl(fd, PMEM_ALLOCATE_ALIGNED, &allocation);
+    }
+
     virtual int getErrno() {
         return errno;
     }
diff --git a/libgralloc-qsd8k/pmemalloc.cpp b/libgralloc-qsd8k/pmemalloc.cpp
index dc756be..ea30b2b 100755
--- a/libgralloc-qsd8k/pmemalloc.cpp
+++ b/libgralloc-qsd8k/pmemalloc.cpp
@@ -143,7 +143,7 @@ int PmemUserspaceAllocator::init_pmem_area()
 
 
 int PmemUserspaceAllocator::alloc_pmem_buffer(size_t size, int usage,
-        void** pBase, int* pOffset, int* pFd, int* masterFd)
+        void** pBase, int* pOffset, int* pFd, int* masterFd, int format)
 {
     BEGIN_FUNC;
     int err = init_pmem_area();
@@ -260,7 +260,7 @@ static unsigned clp2(unsigned x) {
 
 
 int PmemKernelAllocator::alloc_pmem_buffer(size_t size, int usage,
-        void** pBase,int* pOffset, int* pFd, int* masterFd)
+        void** pBase,int* pOffset, int* pFd, int* masterFd, int format)
 {
     BEGIN_FUNC;
 
@@ -292,6 +292,14 @@ int PmemKernelAllocator::alloc_pmem_buffer(size_t size, int usage,
     // The size should already be page aligned, now round it up to a power of 2.
     //size = clp2(size);
 
+    if (format == HAL_PIXEL_FORMAT_YCbCr_420_SP_TILED) {
+        // Tile format buffers need physical alignment to 8K
+        err = deps.alignPmem(master_fd, size, 8192);
+        if (err < 0) {
+            LOGE("alignPmem failed");
+        }
+    }
+
     void* base = deps.mmap(0, size, PROT_READ|PROT_WRITE, MAP_SHARED, master_fd, 0);
     if (base == MAP_FAILED) {
         LOGE("%s: failed to map pmem fd: %s", device,
diff --git a/libgralloc-qsd8k/pmemalloc.h b/libgralloc-qsd8k/pmemalloc.h
index dfe114e..a6d98b8 100755
--- a/libgralloc-qsd8k/pmemalloc.h
+++ b/libgralloc-qsd8k/pmemalloc.h
@@ -36,7 +36,7 @@ class PmemAllocator {
     virtual void* get_base_address() = 0;
 
     virtual int alloc_pmem_buffer(size_t size, int usage, void** pBase,
-            int* pOffset, int* pFd, int* masterFd) = 0;
+            int* pOffset, int* pFd, int* masterFd, int format) = 0;
     virtual int free_pmem_buffer(size_t size, void* base, int offset, int fd) = 0;
 };
 
@@ -89,7 +89,7 @@ class PmemUserspaceAllocator: public PmemAllocator {
     virtual int init_pmem_area_locked();
     virtual int init_pmem_area();
     virtual int alloc_pmem_buffer(size_t size, int usage, void** pBase,
-            int* pOffset, int* pFd, int* masterFd);
+            int* pOffset, int* pFd, int* masterFd, int format);
     virtual int free_pmem_buffer(size_t size, void* base, int offset, int fd);
 
 #ifndef ANDROID_OS
@@ -142,6 +142,7 @@ class PmemKernelAllocator: public PmemAllocator {
         virtual int connectPmem(int fd, int master_fd) = 0;
         virtual int mapPmem(int fd, int offset, size_t size) = 0;
         virtual int unmapPmem(int fd, int offset, size_t size) = 0;
+        virtual int alignPmem(int fd, size_t size, int align) = 0;
     };
 
     PmemKernelAllocator(Deps& deps);
@@ -151,7 +152,7 @@ class PmemKernelAllocator: public PmemAllocator {
     virtual void* get_base_address();
 
     virtual int alloc_pmem_buffer(size_t size, int usage, void** pBase,
-            int* pOffset, int* pFd, int* masterFd);
+            int* pOffset, int* pFd, int* masterFd, int format);
     virtual int free_pmem_buffer(size_t size, void* base, int offset, int fd);
 
  private: