[AArch64][Darwin] Add new ARM64 stack probing function for Darwin

[aemerson]

Summary

[AArch64][Darwin] Add new ARM64 stack probing function for Darwin.

This is the ARM64 equivalent of the Darwin x86 stack probing function and likewise it doesn't modify the stack pointer. However it does have a custom calling convention (uses registers X9-X11).

rdar://25859140

[mstorsjo]

lib/builtins/CMakeLists.txt

447

If this file is darwin specific, it would need to be added conditionally to aarch64_SOURCES. In particular, the .weak_definition directive won't assemble when targeting windows.

lib/builtins/aarch64/chkstk_darwin.S

23

You could avoid clobbering x11 altogether by doing ldr xzr, [x10].

Is the initial cmp w9, #PAGE_SZ strictly necessary? The compiler shouldn't emit these calls if the requested amount is less than one page right?

I implemented something very similar to this in D41134 - there, the main loop is 1 instruction shorter. There I also skip touching the final page, assuming that the caller will touch that page before allocating even more stack space, but perhaps that's not an assumption one can make.

lib/builtins/CMakeLists.txt

 endif()
 
 set(aarch64_SOURCES
+  aarch64/chkstk_darwin.S
   ${GENERIC_TF_SOURCES}
   ${GENERIC_SOURCES})
 

lib/builtins/aarch64/chkstk_darwin.S

+// This file is dual licensed under the MIT and the University of Illinois Open
+// Source Licenses. See LICENSE.TXT for details.
+
+#include "../assembly.h"
+
+#ifdef __aarch64__
+
+#define PAGE_SZ 4096
+
+// ___chkstk_darwin - a generic stack probing function.
+// The number of bytes to probe is passed in w9.
+// This clobbers registers x9, x10 and x11.
+// Does not modify any memory or the stack pointer
+
+.balign 4
+DEFINE_COMPILERRT_FUNCTION(___chkstk_darwin)
+.weak_definition ___chkstk_darwin
+        cmp     w9, #PAGE_SZ
+        mov     x10, sp
+        b.lo    Lend
+Lloop:
+        sub     x10, x10, #PAGE_SZ
+        ldr     x11, [x10]
+        sub     w9, w9, #PAGE_SZ
+        cmp     w9, #PAGE_SZ
+        b.hi    Lloop
+Lend:
+        sub     x10, x10, x9
+        ldr     x11, [x10]
+        ret
+END_COMPILERRT_FUNCTION(___chkstk_darwin)
+
+#endif // __aarch64__

lib/builtins/assembly.h

 
 #if defined(__POWERPC__) || defined(__powerpc__) || defined(__ppc__)
 #define SEPARATOR @
+#elif defined(__aarch64__)
+#define SEPARATOR %%
 #else
 #define SEPARATOR ;
 #endif