builtins: Add __probestack support functions for x86

[Zoxc]

Summary

This adds the __probestack function which is used by the probe-stack attribute in http://reviews.llvm.org/D9654

[Zoxc]

This aligns the size argument to 4 and 16 bytes for x86 and x64. The guard region size is passed as a parameter in EBX/RBX.

lib/builtins/CMakeLists.txt

 
 set(x86_64_SOURCES
   x86_64/floatdidf.c
   x86_64/floatdisf.c
   x86_64/floatdixf.c
   x86_64/floatundidf.S
   x86_64/floatundisf.S
   x86_64/floatundixf.S
+  x86_64/probestack.S
   ${GENERIC_SOURCES})
 
 if(WIN32)
   set(x86_64_SOURCES
       ${x86_64_SOURCES}
       x86_64/chkstk.S)
 endif()
 
 set(i386_SOURCES
   i386/ashldi3.S
   i386/ashrdi3.S
   i386/divdi3.S
   i386/floatdidf.S
   i386/floatdisf.S
   i386/floatdixf.S
   i386/floatundidf.S
   i386/floatundisf.S
   i386/floatundixf.S
   i386/lshrdi3.S
   i386/moddi3.S
   i386/muldi3.S
+  i386/probestack.S
   i386/udivdi3.S
   i386/umoddi3.S
   ${GENERIC_SOURCES})
 
 if(WIN32)
   set(i386_SOURCES
       ${i386_SOURCES}
       i386/chkstk.S)

lib/builtins/README.txt

 long double _Complex __divxc3(long double a, long double b,
                               long double c, long double d);
 long double _Complex __divtc3(long double a, long double b,
                               long double c, long double d);  // ppc only
 
 
 //         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.
 void __clear_cache(void* start, void* end);
 
 // __enable_execute_stack() is used with nested functions when a trampoline
 // function is written onto the stack and that page range needs to be made
 // executable.

lib/builtins/i386/probestack.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"
+
+// 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__

lib/builtins/x86_64/probestack.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"
+
+// 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__