diff --git a/arch/x86/kernel/include/arch-memory.h b/arch/x86/kernel/include/arch-memory.h
index a1c9196c..65ca4710 100644
--- a/arch/x86/kernel/include/arch-memory.h
+++ b/arch/x86/kernel/include/arch-memory.h
@@ -16,10 +16,12 @@
 #define PAGE_SHIFT         12
 #define PAGE_SIZE          (1UL << PAGE_SHIFT)
 #define PAGE_MASK          (~((unsigned long)PAGE_SIZE - 1))
+#define PAGE_P2ALIGN       0
 
 #define LARGE_PAGE_SHIFT   21
 #define LARGE_PAGE_SIZE    (1UL << LARGE_PAGE_SHIFT)
 #define LARGE_PAGE_MASK    (~((unsigned long)LARGE_PAGE_SIZE - 1))
+#define LARGE_PAGE_P2ALIGN (LARGE_PAGE_SHIFT - PAGE_SHIFT)
 
 #define USER_END           0x0000800000000000UL
 #define MAP_ST_START       0xffff800000000000UL
diff --git a/arch/x86/kernel/memory.c b/arch/x86/kernel/memory.c
index 50419478..6c71effd 100644
--- a/arch/x86/kernel/memory.c
+++ b/arch/x86/kernel/memory.c
@@ -35,7 +35,7 @@ void *early_alloc_page(void)
 void *arch_alloc_page(enum ihk_mc_ap_flag flag)
 {
 	if (pa_ops)
-		return pa_ops->alloc_page(1, flag);
+		return pa_ops->alloc_page(1, PAGE_P2ALIGN, flag);
 	else
 		return early_alloc_page();
 }
@@ -45,14 +45,19 @@ void arch_free_page(void *ptr)
 		pa_ops->free_page(ptr, 1);
 }
 
-void *ihk_mc_alloc_pages(int npages, enum ihk_mc_ap_flag flag)
+void *ihk_mc_alloc_aligned_pages(int npages, int p2align, enum ihk_mc_ap_flag flag)
 {
 	if (pa_ops)
-		return pa_ops->alloc_page(npages, flag);
+		return pa_ops->alloc_page(npages, p2align, flag);
 	else
 		return NULL;
 }
 
+void *ihk_mc_alloc_pages(int npages, enum ihk_mc_ap_flag flag)
+{
+	return ihk_mc_alloc_aligned_pages(npages, PAGE_P2ALIGN, flag);
+}
+
 void ihk_mc_free_pages(void *p, int npages)
 {
 	if (pa_ops)
diff --git a/kernel/mem.c b/kernel/mem.c
index 5dc7bf06..8a7949ff 100644
--- a/kernel/mem.c
+++ b/kernel/mem.c
@@ -42,9 +42,9 @@ static void reserve_pages(unsigned long start, unsigned long end, int type)
 	ihk_pagealloc_reserve(pa_allocator, start, end);
 }
 
-void *allocate_pages(int npages, enum ihk_mc_ap_flag flag)
+void *allocate_aligned_pages(int npages, int p2align, enum ihk_mc_ap_flag flag)
 {
-	unsigned long pa = ihk_pagealloc_alloc(pa_allocator, npages);
+	unsigned long pa = ihk_pagealloc_alloc(pa_allocator, npages, p2align);
 	/* all_pagealloc_alloc returns zero when error occured, 
 	   and callee (in mcos/kernel/process.c) so propagate it */
 	if(pa)
@@ -54,13 +54,18 @@ void *allocate_pages(int npages, enum ihk_mc_ap_flag flag)
 	return NULL;
 }
 
+void *allocate_pages(int npages, enum ihk_mc_ap_flag flag)
+{
+	return allocate_aligned_pages(npages, PAGE_P2ALIGN, flag);
+}
+
 void free_pages(void *va, int npages)
 {
 	ihk_pagealloc_free(pa_allocator, virt_to_phys(va), npages);
 }
 
 static struct ihk_mc_pa_ops allocator = {
-	.alloc_page = allocate_pages,
+	.alloc_page = allocate_aligned_pages,
 	.free_page = free_pages,
 };
 
@@ -176,12 +181,15 @@ static void page_allocator_init(void)
 	unsigned long page_map_pa, pages;
 	void *page_map;
 	unsigned int i;
+	uint64_t start;
+	uint64_t end;
 
-	pa_start = ihk_mc_get_memory_address(IHK_MC_GMA_AVAIL_START, 0);
-	pa_end = ihk_mc_get_memory_address(IHK_MC_GMA_AVAIL_END, 0);
+	start = ihk_mc_get_memory_address(IHK_MC_GMA_AVAIL_START, 0);
+	end = ihk_mc_get_memory_address(IHK_MC_GMA_AVAIL_END, 0);
 
-	pa_start &= PAGE_MASK;
-	pa_end = (pa_end + PAGE_SIZE - 1) & PAGE_MASK;
+	start &= PAGE_MASK;
+	pa_start = start & LARGE_PAGE_MASK;
+	pa_end = (end + PAGE_SIZE - 1) & PAGE_MASK;
 
 #ifndef ATTACHED_MIC
 	page_map_pa = ihk_mc_get_memory_address(IHK_MC_GMA_HEAP_START, 0);
@@ -198,6 +206,9 @@ static void page_allocator_init(void)
 	                                    PAGE_SIZE, page_map, &pages);
 
 	reserve_pages(page_map_pa, page_map_pa + pages * PAGE_SIZE, 0);
+	if (pa_start < start) {
+		reserve_pages(pa_start, start, 0);
+	}
 
 	/* BIOS reserved ranges */
 	for (i = 1; i <= ihk_mc_get_memory_address(IHK_MC_NR_RESERVED_AREAS, 0); 
@@ -249,7 +260,7 @@ void *ihk_mc_map_virtual(unsigned long phys, int npages,
 	offset = (phys & (PAGE_SIZE - 1));
 	phys = phys & PAGE_MASK;
 
-	p = (void *)ihk_pagealloc_alloc(vmap_allocator, npages);
+	p = (void *)ihk_pagealloc_alloc(vmap_allocator, npages, PAGE_P2ALIGN);
 	if (!p) {
 		return NULL;
 	}
diff --git a/lib/include/ihk/mm.h b/lib/include/ihk/mm.h
index 76de6e68..14e547d6 100644
--- a/lib/include/ihk/mm.h
+++ b/lib/include/ihk/mm.h
@@ -50,7 +50,7 @@ void ihk_mc_reserve_arch_pages(unsigned long start, unsigned long end,
                                void (*cb)(unsigned long, unsigned long, int));
 
 struct ihk_mc_pa_ops {
-	void *(*alloc_page)(int, enum ihk_mc_ap_flag);
+	void *(*alloc_page)(int, int, enum ihk_mc_ap_flag);
 	void (*free_page)(void *, int);
 
 	void *(*alloc)(int, enum ihk_mc_ap_flag);
@@ -74,6 +74,7 @@ void ihk_mc_map_micpa(unsigned long host_pa, unsigned long* mic_pa);
 int ihk_mc_free_micpa(unsigned long mic_pa);
 void ihk_mc_clean_micpa(void);
 
+void *ihk_mc_alloc_aligned_pages(int npages, int p2align, enum ihk_mc_ap_flag flag);
 void *ihk_mc_alloc_pages(int npages, enum ihk_mc_ap_flag flag);
 void ihk_mc_free_pages(void *p, int npages);
 void *ihk_mc_allocate(int size, enum ihk_mc_ap_flag flag);
diff --git a/lib/include/ihk/page_alloc.h b/lib/include/ihk/page_alloc.h
index 0f17aa8a..4cc2d67c 100644
--- a/lib/include/ihk/page_alloc.h
+++ b/lib/include/ihk/page_alloc.h
@@ -20,7 +20,7 @@ void *__ihk_pagealloc_init(unsigned long start, unsigned long size,
 void *ihk_pagealloc_init(unsigned long start, unsigned long size,
                          unsigned long unit);
 void ihk_pagealloc_destroy(void *__desc);
-unsigned long ihk_pagealloc_alloc(void *__desc, int npages);
+unsigned long ihk_pagealloc_alloc(void *__desc, int npages, int p2align);
 void ihk_pagealloc_reserve(void *desc, unsigned long start, unsigned long end);
 void ihk_pagealloc_free(void *__desc, unsigned long address, int npages);
 unsigned long ihk_pagealloc_count(void *__desc);
diff --git a/lib/page_alloc.c b/lib/page_alloc.c
index 9c2c2109..e1fdffd0 100644
--- a/lib/page_alloc.c
+++ b/lib/page_alloc.c
@@ -88,13 +88,14 @@ void ihk_pagealloc_destroy(void *__desc)
 }
 
 static unsigned long __ihk_pagealloc_large(struct ihk_page_allocator_desc *desc,
-                                           int npages)
+                                           int npages, int p2align)
 {
 	unsigned long flags;
 	unsigned int i, j, mi;
 	int nblocks;
 	int nfrags;
 	unsigned long mask;
+	int mialign;
 
 	nblocks = (npages / 64);
 	mask = -1;
@@ -103,13 +104,14 @@ static unsigned long __ihk_pagealloc_large(struct ihk_page_allocator_desc *desc,
 		++nblocks;
 		mask = (1UL << nfrags) - 1;
 	}
+	mialign = (p2align <= 6)? 1: (1 << (p2align - 6));
 
 	flags = ihk_mc_spinlock_lock(&desc->lock);
 	for (i = 0, mi = desc->last; i < desc->count; i++, mi++) {
 		if (mi >= desc->count) {
 			mi = 0;
 		}
-		if (mi + nblocks >= desc->count) {
+		if ((mi + nblocks >= desc->count) || (mi % mialign)) {
 			continue;
 		}
 		for (j = mi; j < mi + nblocks - 1; j++) {
@@ -131,18 +133,20 @@ static unsigned long __ihk_pagealloc_large(struct ihk_page_allocator_desc *desc,
 	return 0;
 }
 
-unsigned long ihk_pagealloc_alloc(void *__desc, int npages)
+unsigned long ihk_pagealloc_alloc(void *__desc, int npages, int p2align)
 {
 	struct ihk_page_allocator_desc *desc = __desc;
 	unsigned int i, mi;
 	int j;
 	unsigned long v, mask, flags;
+	int jalign;
 
-	if (npages >= 32) {
-		return __ihk_pagealloc_large(desc, npages);
+	if ((npages >= 32) || (p2align >= 5)) {
+		return __ihk_pagealloc_large(desc, npages, p2align);
 	}
 
 	mask = (1UL << npages) - 1;
+	jalign = (p2align <= 0)? 1: (1 << p2align);
 
 	flags = ihk_mc_spinlock_lock(&desc->lock);
 	for (i = 0, mi = desc->last; i < desc->count; i++, mi++) {
@@ -155,6 +159,9 @@ unsigned long ihk_pagealloc_alloc(void *__desc, int npages)
 			continue;
 		
 		for (j = 0; j <= 64 - npages; j++) {
+			if (j % jalign) {
+				continue;
+			}
 			if (!(v & (mask << j))) { /* free */
 				desc->map[mi] |= (mask << j);