[cli/trampolines] Clean up definitions, fix win32 exporting issue (Ju…

…liaLang#39543)

cli/Makefile

@@ -53,7 +53,7 @@ $(BUILDDIR)/loader_exe.o : $(SRCDIR)/loader_exe.c $(HEADERS) $(JULIAHOME)/VERSIO
 	@$(call PRINT_CC, $(CC) $(SHIPFLAGS) $(LOADER_CFLAGS) -c $< -o $@)
 $(BUILDDIR)/loader_exe.dbg.obj : $(SRCDIR)/loader_exe.c $(HEADERS) $(JULIAHOME)/VERSION
 	@$(call PRINT_CC, $(CC) $(DEBUGFLAGS) $(LOADER_CFLAGS) -c $< -o $@)
-$(BUILDDIR)/loader_trampolines.o : $(SRCDIR)/trampolines/trampolines_$(ARCH).S
+$(BUILDDIR)/loader_trampolines.o : $(SRCDIR)/trampolines/trampolines_$(ARCH).S $(HEADERS) $(SRCDIR)/trampolines/common.h
 	@$(call PRINT_CC, $(CC) $(SHIPFLAGS) $(LOADER_CFLAGS) $< -c -o $@)
 
 # Debugging target to help us see what kind of code is being generated for our trampolines

cli/trampolines/common.h

@@ -0,0 +1,70 @@
+// Preprocessor annoyances
+#define CONCAT_(x,y)    x##y
+#define CONCAT(x,y)     CONCAT_(x, y)
+#define CNAMEADDR(name) CONCAT(CNAME(name),_addr)
+#define STR_(x)         #x
+#define STR(x)          STR_(x)
+#define I(x)            x
+
+// On macOS and 32-bit windows, we need to prepend underscores on symbols to match the C ABI
+#if defined(__APPLE__) || (defined(_WIN32) && !defined(_WIN64))
+#define UNDERSCORE(x) _##x
+#else
+#define UNDERSCORE(x)    x
+#endif
+
+// Windows requires some help with the linker when it comes to debuginfo/exporting
+#if defined(_WIN32) || defined(_WIN64)
+#define DEBUGINFO(name)     .def name; \
+                            .scl 2; \
+                            .type 32; \
+                            .endef
+#define EXPORT(name)        .section .drectve,"r"; \
+                            .ascii STR(-export:##I(name)); \
+                            .ascii " "; \
+                            .section .text
+#else
+#define DEBUGINFO(name)
+#define EXPORT(name)
+#endif
+
+// Windows 64-bit uses SEH
+#if defined(_WIN64)
+#define SEH_START1(name)    .seh_proc CNAME(name)
+#define SEH_START2()        .seh_endprologue
+#define SEH_END()           .seh_endproc
+#else
+#define SEH_START1(name)
+#define SEH_START2()
+#define SEH_END()
+#endif
+
+// If we're compiling with control-flow branch protection, mark the trampoline entry
+// points with `endbr{32,64}`, as appropriate on this arch
+#if defined(__CET__) && __CET__ & 1 != 0
+#if defined(__x86_64__)
+#define CET_START()     endbr64
+#else
+#define CET_START()     endbr32
+#endif
+#else
+#define CET_START()
+#endif
+
+// aarch64 on mac requires some special assembler syntax for both calculating memory
+// offsets and even just the assembler statement separator token
+#if defined(__aarch64__)
+#if defined(__APPLE__)
+#define PAGE(x)     x##@PAGE
+#define PAGEOFF(x)  x##@PAGEOFF
+#define SEP         %%
+#else
+#define PAGE(x)     x
+#define PAGEOFF(x)  :lo12:##x
+#define SEP         ;
+#endif
+#endif
+
+// If someday we need to mangle everything, we do so by defining this `CNAME()`
+// to do something more complex than just `UNDERSCORE(x)`.
+#define CNAME(x)    UNDERSCORE(x)

cli/trampolines/trampolines_aarch64.S

@@ -1,18 +1,6 @@
+#include "common.h"
 #include "../../src/jl_exported_funcs.inc"
 
-// On macOS, we need to prepend underscores on symbols
-#if defined(__APPLE__) && defined(__MACH__)
-#define CNAME(x)    _##x
-#define PAGE(x)     x##@PAGE
-#define PAGEOFF(x)  x##@PAGEOFF
-#define SEP         %%
-#else
-#define CNAME(x)    x
-#define PAGE(x)     x
-#define PAGEOFF(x)  #:lo12:##x
-#define SEP         ;
-#endif
-
 #define XX(name) \
 .global CNAME(name) SEP \
 .cfi_startproc SEP \

cli/trampolines/trampolines_arm.S

@@ -1,16 +1,17 @@
+#include "common.h"
 #include "../../src/jl_exported_funcs.inc"
 
 #define XX(name) \
-.global name; \
+.global CNAME(name); \
 .cfi_startproc; \
-name##:; \
-    ldr ip, .L##name; \
-.L##name: ;\
+CNAME(name)##:; \
+    ldr ip, CONCAT(.L,CNAME(name)); \
+CONCAT(.L,CNAME(name)): ;\
     add ip, pc, ip; \
     ldr pc, [ip]; \
     .align 2; \
-.L##name##_addr: ; \
-    .word name##_addr-(.L##name + 8); \
+CONCAT(.L,CNAMEADDR(name))##: ; \
+    .word CNAMEADDR(name)##-(CONCAT(.L,CNAME(name)) + 8); \
 .cfi_endproc; \
 
 JL_EXPORTED_FUNCS(XX)

cli/trampolines/trampolines_i686.S

@@ -1,32 +1,16 @@
+#include "common.h"
 #include "../../src/jl_exported_funcs.inc"
 
-// On Windows, we need to prepend underscores on symbols
-#if defined(_WIN32)
-#define CNAME(x) _##x
-#define DEBUGINFO(name)     .def CNAME(name); \
-                            .scl 2; \
-                            .type 32; \
-                            .endef
-#else
-#define CNAME(x) x
-#define DEBUGINFO(name)
-#endif
-
-#if defined(__CET__) && __CET__ & 1 != 0
-#define CET_START() endbr32
-#else
-#define CET_START()
-#endif
-
 #define XX(name) \
-DEBUGINFO(name); \
+DEBUGINFO(CNAME(name)); \
 .global CNAME(name); \
 .cfi_startproc; \
 CNAME(name)##:; \
     CET_START(); \
-    jmpl *(CNAME(name##_addr)); \
+    jmpl *(CNAMEADDR(name)); \
     ud2; \
 .cfi_endproc; \
+EXPORT(name); \
 
 JL_EXPORTED_FUNCS(XX)
 #undef XX

cli/trampolines/trampolines_powerpc64le.S

@@ -1,3 +1,4 @@
+#include "common.h"
 #include "../../src/jl_exported_funcs.inc"
 
 // Notes:
@@ -7,19 +8,19 @@
 // See 64-Bit ELF V2 ABI Specification: Power Architecture v1.4
 
 #define XX(name) \
-.global name; \
-.type name##, @function; \
+.global CNAME(name); \
+.type CNAME(name)##, @function; \
 .cfi_startproc; \
-name##: ; \
-    addis 2, 12, .TOC.-name##@ha; \
-    addi 2, 2, .TOC.-name##@l; \
-    .localentry name##,.-name##; \
-    addis 9,2,name##_addr@toc@ha; \
-    ld 12,name##_addr@toc@l(9); \
+CNAME(name)##: ; \
+    addis 2, 12, .TOC.-CNAME(name)##@ha; \
+    addi 2, 2, .TOC.-CNAME(name)##@l; \
+    .localentry CNAME(name)##,.-CNAME(name)##; \
+    addis 12,2,CNAMEADDR(name)##@toc@ha; \
+    ld 12,CNAMEADDR(name)##@toc@l(12); \
     mtctr 12; \
     bctr; \
 .cfi_endproc; \
-.size name##,.-name##; \
+.size CNAME(name)##,.-CNAME(name)##; \
 
 JL_EXPORTED_FUNCS(XX)
 #undef XX

cli/trampolines/trampolines_x86_64.S

@@ -1,37 +1,6 @@
+#include "common.h"
 #include "../../src/jl_exported_funcs.inc"
 
-// On macOS, we need to prepend underscores on symbols
-#if defined(__APPLE__) && defined(__MACH__)
-#define CNAME(x) _##x
-#else
-#define CNAME(x) x
-#endif
-
-#if defined(_WIN64)
-#define DEBUGINFO(name)     .def CNAME(name); \
-                            .scl 2; \
-                            .type 32; \
-                            .endef
-#define EXPORT(name)        .section .drectve,"r"; \
-                            .ascii " -export:"#name""; \
-                            .section .text
-#define SEH_START1(name)    .seh_proc name
-#define SEH_START2()        .seh_endprologue
-#define SEH_END()           .seh_endproc
-#else
-#define DEBUGINFO(name)
-#define EXPORT(name)
-#define SEH_START1(name)
-#define SEH_START2()
-#define SEH_END()
-#endif
-
-#if defined(__CET__) && __CET__ & 1 != 0
-#define CET_START() endbr64
-#else
-#define CET_START()
-#endif
-
 #define XX(name) \
 DEBUGINFO(name); \
 .global CNAME(name); \
@@ -40,7 +9,7 @@ SEH_START1(name); \
 CNAME(name)##:; \
 SEH_START2(); \
     CET_START(); \
-    mov CNAME(name##_addr)(%rip),%r11; \
+    mov CNAMEADDR(name)(%rip),%r11; \
     jmpq *%r11; \
     ud2; \
 SEH_END(); \