Index: lib/builtins/CMakeLists.txt
===================================================================
--- lib/builtins/CMakeLists.txt
+++ lib/builtins/CMakeLists.txt
@@ -152,6 +152,7 @@
   x86_64/floatundidf.S
   x86_64/floatundisf.S
   x86_64/floatundixf.S
+  x86_64/probestack.S
   ${GENERIC_SOURCES})
 
 if(WIN32)
@@ -173,6 +174,7 @@
   i386/lshrdi3.S
   i386/moddi3.S
   i386/muldi3.S
+  i386/probestack.S
   i386/udivdi3.S
   i386/umoddi3.S
   ${GENERIC_SOURCES})
Index: lib/builtins/README.txt
===================================================================
--- lib/builtins/README.txt
+++ lib/builtins/README.txt
@@ -196,6 +196,14 @@
 
 //         Runtime support
 
+// __probestack() is used to touch all pages of `size` bytes which will
+// later be allocated relative to the call site's stack pointer. It assumes
+// that the first and the last byte after the allocation will be touched by
+// something else. It has a custom platform specific calling convention.
+// This function is not available on Windows platforms as those have their
+// own builtins.
+void __probestack(du_int size);
+
 // __clear_cache() is used to tell process that new instructions have been
 // written to an address range.  Necessary on processors that do not have
 // a unified instruction and data cache.
Index: lib/builtins/i386/probestack.S
===================================================================
--- /dev/null
+++ lib/builtins/i386/probestack.S
@@ -0,0 +1,58 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// du_int __probestack(du_int size, du_int guard_size);
+
+// `size` is passed in eax, `guard_size` is passed in ebx,
+// and this does not clobber any registers but eax.
+// `size` is returned in eax aligned up to 4 bytes.
+
+#ifndef __WIN32__
+#ifdef __i386__
+
+#if defined(__APPLE__)
+	.const
+#elif defined(__ELF__)
+	.section .rodata
+#else
+	.section .rdata,"rd"
+#endif
+
+.text
+.balign 4
+DEFINE_COMPILERRT_FUNCTION(__probestack)
+.cfi_startproc
+	// Align size to 4 bytes
+	addl $3, %eax
+	andl $-4, %eax
+
+	pushl %eax
+.cfi_adjust_cfa_offset 4
+.cfi_rel_offset %eax, 0
+	pushl %ecx
+.cfi_adjust_cfa_offset 4
+.cfi_rel_offset %ecx, 0
+
+	// Load the pre-call ESP into ECX
+	leal 12(%esp), %ecx
+
+1:
+	subl %ebx, %ecx
+	orb $0, (%ecx)
+	subl %ebx, %eax
+	ja 1b
+
+	popl %ecx
+.cfi_adjust_cfa_offset -4
+.cfi_restore %ecx
+	popl %eax
+.cfi_adjust_cfa_offset -4
+.cfi_restore %eax
+	ret
+.cfi_endproc
+END_COMPILERRT_FUNCTION(__probestack)
+
+#endif // __i386__
+#endif // __WIN32__
Index: lib/builtins/x86_64/probestack.S
===================================================================
--- /dev/null
+++ lib/builtins/x86_64/probestack.S
@@ -0,0 +1,58 @@
+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+// du_int __probestack(du_int size, du_int guard_size);
+
+// `size` is passed in rax, `guard_size` is passed in rbx,
+// and this does not clobber any registers but rax.
+// `size` is returned in rax aligned up to 16 bytes.
+
+#ifndef __WIN32__
+#ifdef __x86_64__
+
+#if defined(__APPLE__)
+	.const
+#elif defined(__ELF__)
+	.section .rodata
+#else
+	.section .rdata,"rd"
+#endif
+
+.text
+.balign 4
+DEFINE_COMPILERRT_FUNCTION(__probestack)
+.cfi_startproc
+	// Align size to 16 bytes
+	addq $15, %rax
+	andq $-16, %rax
+
+	pushq %rax
+.cfi_adjust_cfa_offset 8
+.cfi_rel_offset %rax, 0
+	pushq %r11
+.cfi_adjust_cfa_offset 8
+.cfi_rel_offset %r11, 0
+
+	// Load the pre-call RSP into R11
+	leaq 24(%rsp), %r11
+
+1:
+	subq %rbx, %r11
+	orb $0, (%r11)
+	subq %rbx, %rax
+	ja 1b
+
+	popq %r11
+.cfi_adjust_cfa_offset -8
+.cfi_restore %r11
+	popq %rax
+.cfi_adjust_cfa_offset -8
+.cfi_restore %rax
+	ret
+.cfi_endproc
+END_COMPILERRT_FUNCTION(__probestack)
+
+#endif // __x86_64__
+#endif // __WIN32__