diff --git a/proxyclient/m1n1/agx/context.py b/proxyclient/m1n1/agx/context.py
index ac8f95c2..126f88b5 100644
--- a/proxyclient/m1n1/agx/context.py
+++ b/proxyclient/m1n1/agx/context.py
@@ -32,6 +32,7 @@ class GPUContext:
 
         # 32K VA pages since buffer manager needs that
         self.uobj = GPUAllocator(agx, "Userspace", 0x1600000000, 0x100000000, ctx=None,
+                                 guard_pages=1,
                                  va_block=32768, nG=1, AP=0, PXN=1, UXN=1)
 
     def bind(self, ctx_id):
diff --git a/proxyclient/m1n1/agx/object.py b/proxyclient/m1n1/agx/object.py
index 4a6383be..b2763693 100644
--- a/proxyclient/m1n1/agx/object.py
+++ b/proxyclient/m1n1/agx/object.py
@@ -102,7 +102,7 @@ class GPUObject:
 
 class GPUAllocator:
     def __init__(self, agx, name, start, size,
-                 ctx=0, page_size=16384, va_block=None, **kwargs):
+                 ctx=0, page_size=16384, va_block=None, guard_pages=1, **kwargs):
         self.page_size = page_size
         if va_block is None:
             va_block = page_size
@@ -111,6 +111,7 @@ class GPUAllocator:
         self.name = name
         self.va = Heap(start, start + size, block=va_block)
         self.verbose = 1
+        self.guard_pages = guard_pages
         self.objects = {}
         self.flags = kwargs
 
@@ -123,8 +124,10 @@ class GPUAllocator:
         if name is not None:
             obj._name = name
 
+        guard_size = self.page_size * self.guard_pages
+
         size_align = align_up(obj._size, self.page_size)
-        addr = self.va.malloc(size_align + self.page_size)
+        addr = self.va.malloc(size_align + guard_size)
         paddr = self.agx.u.memalign(self.page_size, size_align)
         off = size_align - obj._size