Bug 1561088 - emit unwind information for libffi aarch64/win assembly; r=dmajor,gsvelto

The hand-written assembly for libffi on aarch64/windows doesn't emit
unwind information.  If we ever tried to unwind through these functions,
they'd look like leaf functions, which is decidedly not true and would
cause great pain.

For whatever reason, the original aarch64 libffi functions used
x21/x22/x23/x24 as their (callee-saved) scratch registers.  This
convention works on windows as well, but the unwind information on
windows mandates that we start saving callee-saved registers starting
from x19, rather than x21.  Rather than rewriting the assembly to use
x19/x20 instead of x21/x22, which would be a large change, we chose
instead to simply save/restore extra registers in the prolog/epilog.
This change does make the stack frame sizes slightly bigger, but an
extra 16 bytes in libffi stack frames should not matter.

The `-TC` change is necessary to make the compiler play nicely with .asm
file suffixes.

Differential Revision: https://phabricator.services.mozilla.com/D35714

--HG--
extra : moz-landing-system : lando

config/external/ffi/moz.build

@@ -78,7 +78,18 @@ else:
     elif CONFIG['FFI_TARGET'] == 'AARCH64':
         ffi_srcs = ('sysv.S', 'ffi.c')
     elif CONFIG['FFI_TARGET'] == 'ARM64_WIN64':
-        ffi_srcs = ('win64.asm', 'ffi.c')
+        ffi_srcs = ['ffi.c']
+
+        GENERATED_FILES += ['win64_aarch.asm']
+        asm = GENERATED_FILES['win64_aarch.asm']
+        asm.inputs = [
+            '/js/src/ctypes/libffi/src/aarch64/win64.asm',
+            '!../../../js/src/ctypes/libffi/fficonfig.h',
+            '!../../../js/src/ctypes/libffi/include/ffi.h',
+        ]
+        asm.script = 'preprocess_libffi_asm.py'
+        asm.flags = ['$(DEFINES)', '$(LOCAL_INCLUDES)']
+        SOURCES += ['!win64_aarch.asm']
     elif CONFIG['FFI_TARGET'] == 'X86':
         ffi_srcs = ('ffi.c', 'sysv.S', 'win32.S')
     elif CONFIG['FFI_TARGET'] == 'X86_64':

config/external/ffi/preprocess_libffi_asm.py

@@ -15,7 +15,8 @@ def main(output, input_asm, ffi_h, ffi_config_h, defines, includes):
     defines = shlex.split(defines)
     includes = shlex.split(includes)
     # CPP uses -E which generates #line directives. -EP suppresses them.
-    cpp = buildconfig.substs['CPP'] + ['-EP']
+    # -TC forces the compiler to treat the input as C.
+    cpp = buildconfig.substs['CPP'] + ['-EP'] + ['-TC']
     input_asm = mozpath.relpath(input_asm, os.getcwd())
     args = cpp + defines + includes + [input_asm]
     print(' '.join(args))

js/src/ctypes/libffi/src/aarch64/win64.asm

@@ -19,9 +19,11 @@
 ;; TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
 ;; SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 
+#include "ksarm64.h"
+
 ;; Hand-converted from the sysv.S file in this directory.
 
-        AREA |.text|, CODE, ARM64
+        TEXTAREA
 
    ;; ffi_call_SYSV()
 
@@ -58,30 +60,26 @@
    ;; This function uses the following stack frame layout:
 
    ;; ==
-   ;;              saved x30(lr)
-   ;; x29(fp)->    saved x29(fp)
    ;;              saved x24
    ;;              saved x23
    ;;              saved x22
-   ;; sp'    ->    saved x21
+   ;;              saved x21
+   ;;              saved x20
+   ;;              saved x19
+   ;;              saved x30(lr)
+   ;; x29(fp)->    saved x29(fp)
    ;;              ...
    ;; sp     ->    (constructed callee stack arguments)
    ;; ==
 
    ;; Voila!
 
-	EXPORT |ffi_call_SYSV|
+        NESTED_ENTRY |ffi_call_SYSV|
 
-|ffi_call_SYSV| PROC
-;#define ffi_call_SYSV_FS (8 * 4)
-
-        stp     x29, x30, [sp, #-16]!
-
-        mov     x29, sp
-        sub     sp, sp, #32 	; ffi_call_SYSV_FS
-
-        stp     x21, x22, [sp, #0]
-        stp     x23, x24, [sp, #16]
+        PROLOG_SAVE_REG_PAIR x29, x30, #-64!
+        PROLOG_SAVE_REG_PAIR x19, x20, #16
+        PROLOG_SAVE_REG_PAIR x21, x22, #32
+        PROLOG_SAVE_REG_PAIR x23, x24, #48
 
         mov     x21, x1
         mov     x22, x2
@@ -145,19 +143,14 @@ noload_call
 
 nosave_call
         ; All done, unwind our stack frame.
-        ldp     x21, x22, [x29,  # - 32] ; ffi_call_SYSV_FS
+        EPILOG_STACK_RESTORE
+        EPILOG_RESTORE_REG_PAIR x19, x20, #16
+        EPILOG_RESTORE_REG_PAIR x21, x22, #32
+        EPILOG_RESTORE_REG_PAIR x23, x24, #48
+        EPILOG_RESTORE_REG_PAIR x29, x30, #64!
+        EPILOG_RETURN
 
-        ldp     x23, x24, [x29,  # - 32 + 16] ; ffi_call_SYSV_FS
-
-        mov     sp, x29
-
-        ldp     x29, x30, [sp], #16
-
-        ret
-
-	ENDP
-
-; #define ffi_closure_SYSV_FS (8 * 2 + AARCH64_CALL_CONTEXT_SIZE)
+        NESTED_END |ffi_call_SYSV|
 
    ;; ffi_closure_SYSV
 
@@ -186,10 +179,12 @@ nosave_call
    ;; This function uses the following stack frame layout:
 
    ;; ==
-   ;;              saved x30(lr)
-   ;; x29(fp)->    saved x29(fp)
    ;;              saved x22
    ;;              saved x21
+   ;;              saved x20
+   ;;              saved x19
+   ;;              saved x30(lr)
+   ;; x29(fp)->    saved x29(fp)
    ;;              ...
    ;; sp     ->    call_context
    ;; ==
@@ -197,16 +192,14 @@ nosave_call
    ;; Voila!
 
 	IMPORT |ffi_closure_SYSV_inner|
-	EXPORT |ffi_closure_SYSV|
 
-|ffi_closure_SYSV| PROC
-        stp     x29, x30, [sp, #-16]!
+        NESTED_ENTRY |ffi_closure_SYSV|
 
-        mov     x29, sp
+        PROLOG_SAVE_REG_PAIR fp, lr, #-48!
+        PROLOG_SAVE_REG_PAIR x19, x20, #16
+        PROLOG_SAVE_REG_PAIR x21, x22, #32
 
-        sub     sp, sp, #256+512+16
-
-        stp     x21, x22, [x29, #-16]
+        sub     sp, sp, #256+512
 
         ; Load x21 with &call_context.
         mov     x21, sp
@@ -260,13 +253,12 @@ noload_closure
         ; Note nothing useful is returned in x8.
 
         ; We are done, unwind our frame.
-        ldp     x21, x22, [x29,  #-16]
+        EPILOG_STACK_RESTORE
+        EPILOG_RESTORE_REG_PAIR x19, x20, #16
+        EPILOG_RESTORE_REG_PAIR x21, x22, #32
+        EPILOG_RESTORE_REG_PAIR x29, x30, #48!
+        EPILOG_RETURN
 
-        mov     sp, x29
+        NESTED_END |ffi_closure_SYSV|
 
-        ldp     x29, x30, [sp], #16
-
-        ret
-
-	ENDP
-	END
+        END