Move to a shared static RWX section

Does away with the need to mark a region as executable on load.
Also allows the self-modified stuff and hook trampolines to fit in the
same page, so we take up ever-so-slightly less memory.

Unfortunately I had issues getting the test binary to build correctly
and eventually decided to give up as it wasn't testing much of value
anyway.

Also note that the build now allows for .c and .S files with the same
basename, which required renaming the .o files, so you might want to
clean out your .build/ directory before building this one.

1 files changed, 34 insertions, 59 deletions

diff --git a/src/con_.c b/src/con_.c
index ff71383..6d5ce03 100644
--- a/src/con_.c
+++ b/src/con_.c
@@ -75,26 +75,11 @@ DECL_VFUNC(struct ICvar, void *, FindVar_nonp2, 12, const char *)
 DECL_VFUNC(struct ICvar, void *, FindVar_OE, 7, const char *)
 #endif
 
-static struct ICvar *coniface;
-static void *colourmsgf;
+struct ICvar *_con_iface; // extern for con_.S
+void *_con_colourmsgf; // "
 
-#ifdef _WIN32
-#pragma section("selfmod", execute)
-__attribute((used, section("selfmod"), noinline))
-#endif
-asm_only void _con_colourmsg(void *dummy, const struct rgba *c,
-		const char *fmt, ...) {
-	// NE: ConsoleColorPrintf is virtual, so the dummy param is a carve-out for
-	// `this` (which is coniface).
-	__asm volatile (
-		"mov eax, %0\n"
-		"mov [esp + 4], eax\n" // put coniface in the empty stack slot
-		"jmp dword ptr %1\n" // jump to the real function
-		:
-		: "m" (coniface), "m" (colourmsgf)
-		: "eax", "memory"
-	);
-}
+// this is defined in con_.S, in order to set the proper section attributes.
+void _con_colourmsg(void *dummy, const struct rgba *c, const char *fmt, ...);
 
 #ifdef _WIN32
 // this function is defined as data because we'll be using it to self-modify the
@@ -114,23 +99,11 @@ asm_only static void _con_colourmsg_OE(void *dummy, const struct rgba *c,
 		"sub esp, 4\n" // pad the stack back out for the caller
 		"jmp ebx\n" // return to saved address
 		:
-		: "m" (coniface), "m" (colourmsgf)
+		: "m" (_con_iface), "m" (_con_colourmsgf)
 		: "eax", "ebx", "memory"
 	);
 }
 #define SELFMOD_LEN 15 // above instructions assemble to this many bytes!
-
-static bool selfmod() {
-	if (!os_mprot((void *)_con_colourmsg, SELFMOD_LEN, PAGE_EXECUTE_READWRITE)) {
-		errmsg_errorsys("couldn't make memory writable");
-		return false;
-	}
-	memcpy((void *)&_con_colourmsg, (void *)&_con_colourmsg_OE, SELFMOD_LEN);
-	if (!os_mprot((void *)_con_colourmsg, SELFMOD_LEN, PAGE_EXECUTE_READ)) {
-		errmsg_warnsys("couldn't restore self-modified page to read-only");
-	}
-	return true;
-}
 #endif
 
 static void VCALLCONV dtor(void *_) {} // we don't use constructors/destructors
@@ -244,13 +217,13 @@ static void VCALLCONV ChangeStringValue(struct con_var *this, const char *s,
 		float oldf) {
 	char *old = alloca(this->v2.strlen);
 	ChangeStringValue_common(this, &this->v2, old, s);
-	CallGlobalChangeCallbacks(coniface, this, old, oldf);
+	CallGlobalChangeCallbacks(_con_iface, this, old, oldf);
 }
 #ifdef _WIN32
 static void VCALLCONV ChangeStringValue_OE(struct con_var *this, const char *s) {
 	char *old = alloca(this->v1.strlen);
 	ChangeStringValue_common(this, &this->v1, old, s);
-	CallGlobalChangeCallbacks_OE(coniface, this, old);
+	CallGlobalChangeCallbacks_OE(_con_iface, this, old);
 }
 #endif
 
@@ -446,7 +419,7 @@ void con_regvar(struct con_var *v) {
 	struct con_var_common *c = con_getvarcommon(v);
 	c->strval = extmalloc(c->strlen); // note: _DEF_CVAR() sets strlen member
 	memcpy(c->strval, c->defaultval, c->strlen);
-	RegisterConCommand(coniface, v);
+	RegisterConCommand(_con_iface, v);
 }
 
 void con_regcmd(struct con_cmd *c) {
@@ -454,7 +427,7 @@ void con_regcmd(struct con_cmd *c) {
 	if_hot (!GAMETYPE_MATCHES(OE)) if (c->base.flags & CON_INIT_HIDDEN) {
 		c->base.flags = (c->base.flags & ~CON_INIT_HIDDEN) | _CON_NE_HIDDEN;
 	}
-	RegisterConCommand(coniface, c);
+	RegisterConCommand(_con_iface, c);
 }
 
 void con_hide(struct con_cmdbase *b) {
@@ -504,7 +477,7 @@ static bool find_Con_ColorPrintf() {
 	for (uchar *p = insns; p - insns < 320;) {
 		if (p[0] == X86_PUSHECX && p[1] == X86_PUSHIW && p[6] == X86_CALL &&
 				p[11] == X86_ALUMI8S && p[12] == X86_MODRM(3, 0, 4)) {
-			colourmsgf = p + 11 + mem_loads32(p + 7);
+			_con_colourmsgf = p + 11 + mem_loads32(p + 7);
 			return true;
 		}
 		NEXT_INSN(p, "Con_ColorPrintf function");
@@ -529,39 +502,39 @@ static void badver() {
 }
 
 bool con_detect(int pluginver) {
-	if (coniface = factory_engine("VEngineCvar007", 0)) {
+	if (_con_iface = factory_engine("VEngineCvar007", 0)) {
 		// GENIUS HACK (BUT STILL BAD): Portal 2 has everything in ICvar shifted
 		// down 3 places due to the extra stuff in IAppSystem. This means that
 		// if we look up the Portal 2-specific cvar using FindCommandBase, it
 		// *actually* calls the const-overloaded FindVar on other branches,
 		// which just happens to still work fine. From there, we can figure out
 		// the actual ABI to use to avoid spectacular crashes.
-		if (FindCommandBase_p2(coniface, "portal2_square_portals")) {
+		if (FindCommandBase_p2(_con_iface, "portal2_square_portals")) {
 			_gametype_tag |= _gametype_tag_Portal2;
 			return true;
 		}
-		if (FindCommand_nonp2(coniface, "l4d2_snd_adrenaline")) {
+		if (FindCommand_nonp2(_con_iface, "l4d2_snd_adrenaline")) {
 			// while we're here, also distinguish Survivors, the stupid Japanese
 			// arcade game a few people seem to care about for some reason
 			// (which for some other reason also has some vtable changes)
-			if (FindVar_nonp2(coniface, "avatarbasemodel")) {
+			if (FindVar_nonp2(_con_iface, "avatarbasemodel")) {
 				_gametype_tag |= _gametype_tag_L4DS;
 			}
 			else {
 				_gametype_tag |= _gametype_tag_L4D2;
 			}
-			if (FindVar_nonp2(coniface, "sv_zombie_touch_trigger_delay")) {
+			if (FindVar_nonp2(_con_iface, "sv_zombie_touch_trigger_delay")) {
 				_gametype_tag |= _gametype_tag_L4D2_2125plus;
 			}
-			if (FindVar_nonp2(coniface, "director_cs_weapon_spawn_chance")) {
+			if (FindVar_nonp2(_con_iface, "director_cs_weapon_spawn_chance")) {
 				_gametype_tag |= _gametype_tag_TheLastStand;
 			}
 			return true;
 		}
-		if (FindVar_nonp2(coniface, "z_difficulty")) {
+		if (FindVar_nonp2(_con_iface, "z_difficulty")) {
 			_gametype_tag |= _gametype_tag_L4D1;
 			// Crash Course update
-			if (FindCommand_nonp2(coniface, "director_log_scavenge_items")) {
+			if (FindCommand_nonp2(_con_iface, "director_log_scavenge_items")) {
 				_gametype_tag |= _gametype_tag_L4D1_1015plus;
 				// seems there was some code shuffling in the Mac update (1022).
 				// this update came out like 2-3 weeks after The Sacrifice
@@ -575,43 +548,45 @@ bool con_detect(int pluginver) {
 		helpuserhelpus(pluginver, '7');
 		return false;
 	}
-	if (coniface = factory_engine("VEngineCvar004", 0)) {
+	if (_con_iface = factory_engine("VEngineCvar004", 0)) {
 		// TODO(compat): are there any cases where 004 is incompatible? could
 		// this crash? find out!
 		if (pluginver == 3) _gametype_tag |= _gametype_tag_2013;
 		else _gametype_tag |= _gametype_tag_OrangeBox;
 		// detect Portal 1 versions while we're here...
-		if (FindCommand_nonp2(coniface, "upgrade_portalgun")) {
+		if (FindCommand_nonp2(_con_iface, "upgrade_portalgun")) {
 			_gametype_tag |= _gametype_tag_Portal1;
-			if (!FindVar_nonp2(coniface, "tf_escort_score_rate")) {
+			if (!FindVar_nonp2(_con_iface, "tf_escort_score_rate")) {
 				_gametype_tag |= _gametype_tag_Portal1_3420;
 			}
 		}
-		else if (FindCommand_nonp2(coniface, "phys_swap")) {
+		else if (FindCommand_nonp2(_con_iface, "phys_swap")) {
 			_gametype_tag |= _gametype_tag_HL2series;
 		}
 		return true;
 	}
-	if (coniface = factory_engine("VEngineCvar003", 0)) {
+	if (_con_iface = factory_engine("VEngineCvar003", 0)) {
 #ifdef _WIN32 // there's no OE on linux!
 		_gametype_tag |= _gametype_tag_OE;
 		// for deletion/unlinking on unload, we need an indirect linked list
 		// pointer. calling GetCommands gives us a direct pointer. so we have to
 		// actually pull out the indirect pointer from the actual asm lol.
-		if (!find_linkedlist((uchar *)VFUNC(coniface, GetCommands_OE))) {
+		if (!find_linkedlist((uchar *)VFUNC(_con_iface, GetCommands_OE))) {
 			errmsg_errorx("couldn't find command list pointer");
 			return false;
 		}
 		if (!find_argcargv()) return false;
 		if (!find_Con_ColorPrintf()) return false;
-		if (!selfmod()) return false;
+		// note: _con_colourmsg is already RWX, just overwrite it.
+		memcpy((void *)&_con_colourmsg, (void *)&_con_colourmsg_OE,
+				SELFMOD_LEN);
 		// NOTE: the default static struct layout is for NE; immediately after
 		// engineapi init finishes, the generated glue code will shunt
 		// everything along for OE if required, in shuntvars(). since all the
 		// gluegen code is currently hooked up in sst.c this is a little bit
 		// annoyingly removed from here. not sure how to do it better, sorry.
 		off_cvar_common = offsetof(struct con_var, v1);
-		if (FindVar_OE(coniface, "mm_ai_facehugger_enablehugeattack")) {
+		if (FindVar_OE(_con_iface, "mm_ai_facehugger_enablehugeattack")) {
 			_gametype_tag |= _gametype_tag_DMoMM;
 		}
 		return true;
@@ -632,7 +607,7 @@ bool con_detect(int pluginver) {
 }
 
 static int *find_host_initialized() {
-	const uchar *insns = colourmsgf;
+	const uchar *insns = _con_colourmsgf;
 	for (const uchar *p = insns; p - insns < 32;) {
 		// cmp byte ptr [<pointer>], <value>
 		if (p[0] == X86_ALUMI8 && p[1] == X86_MODRM(0, 7, 5)) {
@@ -653,8 +628,8 @@ void con_init() {
 		if (host_initialized && *host_initialized == 0) *host_initialized = 1;
 	}
 	else {
-		colourmsgf = coniface->vtable[vtidx_ConsoleColorPrintf];
-		dllid = AllocateDLLIdentifier(coniface);
+		_con_colourmsgf = _con_iface->vtable[vtidx_ConsoleColorPrintf];
+		dllid = AllocateDLLIdentifier(_con_iface);
 	}
 
 	void **pc = _con_vtab_cmd + 3 + NVDTOR, **pv = _con_vtab_var + 3 + NVDTOR,
@@ -770,11 +745,11 @@ void con_disconnect() {
 		return;
 	}
 #endif
-	UnregisterConCommands(coniface, dllid);
+	UnregisterConCommands(_con_iface, dllid);
 }
 
 struct con_var *con_findvar(const char *name) {
-	return FindVar(coniface, name);
+	return FindVar(_con_iface, name);
 }
 
 struct con_cmd *con_findcmd(const char *name) {
@@ -790,7 +765,7 @@ struct con_cmd *con_findcmd(const char *name) {
 		return 0;
 	}
 #endif
-	return FindCommand(coniface, name);
+	return FindCommand(_con_iface, name);
 }
 
 // NOTE: getters here still go through the parent pointer although we stopped