[compiler-rt] [builtins] Implement the __chkstk function for ARM for MinGW

[mstorsjo]

Summary

On ARM (contrary to both x86 and ARM64), this function is available for linking in from kernel32.dll, but it's not allowed to link that function from there in Windows Store apps.

[smeenai]

Microsoft's implementation also checks the stack allocation against the thread's stack limit. Do you care about doing that?

LGTM otherwise, but I'd prefer to wait for @compnerd to take a look. (He's on vacation this week, but he'll be back next week.)

lib/builtins/arm/chkstk.S

10 ↗

(On Diff #154503)

It also clobbers the condition codes. We were bitten by this before; see https://reviews.llvm.org/rL311061.

[mstorsjo]

In D49055#1156485, @smeenai wrote:

Microsoft's implementation also checks the stack allocation against the thread's stack limit. Do you care about doing that?

That might be neat to do, but we don't do that in the existing implementations of __chkstk* for x86, nor in the aarch64 version I wrote a few months ago (where I justified it to myself with the same reason; it hasn't been done for x86 yet either).

lib/builtins/arm/chkstk.S

10 ↗

(On Diff #154503)

Good catch, I'll amend the comment for clarity.

[compnerd]

Seems fine with the adjusted comment

compiler-rt/trunk/lib/builtins/CMakeLists.txt

set(arm_Thumb1_SOURCES
   ${arm_Thumb1_icache_SOURCES})
 
 if(MINGW)
   set(arm_SOURCES
       arm/aeabi_idivmod.S
       arm/aeabi_ldivmod.S
       arm/aeabi_uidivmod.S
       arm/aeabi_uldivmod.S
+      arm/chkstk.S
       divmoddi4.c
       divmodsi4.c
       divdi3.c
       divsi3.c
       fixdfdi.c
       fixsfdi.c
       fixunsdfdi.c
       fixunssfdi.c

compiler-rt/trunk/lib/builtins/arm/chkstk.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"
+
+// __chkstk routine
+// This routine is windows specific.
+// http://msdn.microsoft.com/en-us/library/ms648426.aspx
+
+// This clobbers the register r12, and the condition codes, and uses r5 and r6
+// as temporaries by backing them up and restoring them afterwards.
+// Does not modify any memory or the stack pointer.
+
+//      movw    r4,  #256 // Number of bytes of stack, in units of 4 byte
+//      bl      __chkstk
+//      sub.w   sp, sp, r4
+
+#define PAGE_SIZE 4096
+
+        .p2align 2
+DEFINE_COMPILERRT_FUNCTION(__chkstk)
+        lsl    r4,  r4,  #2
+        mov    r12, sp
+        push   {r5, r6}
+        mov    r5,  r4
+1:
+        sub    r12, r12, #PAGE_SIZE
+        subs   r5,  r5,  #PAGE_SIZE
+        ldr    r6,  [r12]
+        bgt    1b
+
+        pop    {r5, r6}
+        bx     lr
+END_COMPILERRT_FUNCTION(__chkstk)