*: add optimization barrier to all "G trick" locations

archival/gzip.c

@@ -2042,8 +2042,8 @@ int gzip_main(int argc, char **argv)
 	//if (opt & 0x4) // -v
 	argv += optind;
 
-	PTR_TO_GLOBALS = xzalloc(sizeof(struct globals) + sizeof(struct globals2))
-			+ sizeof(struct globals);
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(struct globals) + sizeof(struct globals2))
+			+ sizeof(struct globals));
 	G2.l_desc.dyn_tree    = G2.dyn_ltree;
 	G2.l_desc.static_tree = G2.static_ltree;
 	G2.l_desc.extra_bits  = extra_lbits;

coreutils/test.c

@@ -180,6 +180,7 @@ static struct statics *const ptr_to_statics __attribute__ ((section (".data")));
 
 #define INIT_S() do { \
 	(*(struct statics**)&ptr_to_statics) = xzalloc(sizeof(S)); \
+	barrier(); \
 } while (0)
 #define DEINIT_S() do { \
 	free(ptr_to_statics); \

docs/keep_data_small.txt

@@ -99,9 +99,9 @@ and then declare that ptr_to_globals is a pointer to it:
 
 ptr_to_globals is declared as constant pointer.
 This helps gcc understand that it won't change, resulting in noticeably
-smaller code. In order to assign it, use PTR_TO_GLOBALS macro:
+smaller code. In order to assign it, use SET_PTR_TO_GLOBALS macro:
 
-	PTR_TO_GLOBALS = xzalloc(sizeof(G));
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G)));
 
 Typically it is done in <applet>_main().
 

editors/awk.c

@@ -394,7 +394,8 @@ static const uint16_t PRIMES[] ALIGN2 = { 251, 1021, 4093, 16381, 65521 };
 /* Globals. Split in two parts so that first one is addressed
  * with (mostly short) negative offsets */
 struct globals {
-	chain beginseq, mainseq, endseq, *seq;
+	chain beginseq, mainseq, endseq;
+	chain *seq;
 	node *break_ptr, *continue_ptr;
 	rstream *iF;
 	xhash *vhash, *ahash, *fdhash, *fnhash;
@@ -445,7 +446,7 @@ struct globals2 {
 	tsplitter fsplitter, rsplitter;
 };
 #define G1 (ptr_to_globals[-1])
-#define G (*(struct globals2 *const)ptr_to_globals)
+#define G (*(struct globals2 *)ptr_to_globals)
 /* For debug. nm --size-sort awk.o | grep -vi ' [tr] ' */
 /* char G1size[sizeof(G1)]; - 0x6c */
 /* char Gsize[sizeof(G)]; - 0x1cc */
@@ -485,7 +486,7 @@ struct globals2 {
 #define fsplitter    (G.fsplitter   )
 #define rsplitter    (G.rsplitter   )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G1) + sizeof(G)) + sizeof(G1); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G1) + sizeof(G)) + sizeof(G1)); \
 	G.next_token__ltclass = TC_OPTERM; \
 	G.evaluate__seed = 1; \
 } while (0)

editors/diff.c

@@ -151,7 +151,7 @@ struct globals {
 #define stb1               (G.stb1              )
 #define stb2               (G.stb2              )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	context = 3; \
 	max_context = 64; \
 } while (0)

editors/ed.c

@@ -51,7 +51,7 @@ struct globals {
 #define lines              (G.lines             )
 #define marks              (G.marks             )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 } while (0)
 
 

editors/vi.c

@@ -233,7 +233,7 @@ struct globals {
 #define readbuffer     (G.readbuffer    )
 #define scr_out_buf    (G.scr_out_buf   )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 } while (0)
 
 static int init_text_buffer(char *); // init from file or create new

include/libbb.h

@@ -1208,10 +1208,14 @@ extern char bb_common_bufsiz1[COMMON_BUFSIZE];
 struct globals;
 /* '*const' ptr makes gcc optimize code much better.
  * Magic prevents ptr_to_globals from going into rodata.
- * If you want to assign a value, use PTR_TO_GLOBALS = xxx */
+ * If you want to assign a value, use SET_PTR_TO_GLOBALS(x) */
 extern struct globals *const ptr_to_globals;
-#define PTR_TO_GLOBALS (*(struct globals**)&ptr_to_globals)
-
+/* At least gcc 3.4.6 on mipsel system needs optimization barrier */
+#define barrier() asm volatile("":::"memory")
+#define SET_PTR_TO_GLOBALS(x) do { \
+	(*(struct globals**)&ptr_to_globals) = (x); \
+	barrier(); \
+} while (0)
 
 /* You can change LIBBB_DEFAULT_LOGIN_SHELL, but don't use it,
  * use bb_default_login_shell and following defines.

libbb/Kbuild

@@ -69,6 +69,7 @@ lib-y += pidfile.o
 lib-y += printable.o
 lib-y += process_escape_sequence.o
 lib-y += procps.o
+lib-y += ptr_to_globals.o
 lib-y += read.o
 lib-y += recursive_action.o
 lib-y += remove_file.o

libbb/appletlib.c

@@ -127,6 +127,7 @@ void lbb_prepare(const char *applet, char **argv)
 {
 #ifdef __GLIBC__
 	(*(int **)&bb_errno) = __errno_location();
+	barrier();
 #endif
 	applet_name = applet;
 

libbb/lineedit.c

@@ -146,6 +146,7 @@ static struct statics *const ptr_to_statics __attribute__ ((section (".data")));
 
 #define INIT_S() do { \
 	(*(struct statics**)&ptr_to_statics) = xzalloc(sizeof(S)); \
+	barrier(); \
 	cmdedit_termw = 80; \
 	USE_FEATURE_EDITING_FANCY_PROMPT(num_ok_lines = 1;) \
 	USE_FEATURE_GETUSERNAME_AND_HOMEDIR(home_pwd_buf = (char*)null_str;) \

libbb/messages.c

@@ -71,7 +71,3 @@ const char bb_path_wtmp_file[] ALIGN1 =
  * Since gcc insists on aligning struct global's members, it would be a pity
  * (and an alignment fault on some CPUs) to mess it up. */
 char bb_common_bufsiz1[COMMON_BUFSIZE] __attribute__(( aligned(sizeof(long long)) ));
-
-struct globals;
-/* Make it reside in R/W memory: */
-struct globals *const ptr_to_globals __attribute__ ((section (".data")));

libbb/ptr_to_globals.c

@@ -0,0 +1,11 @@
+/* vi: set sw=4 ts=4: */
+/*
+ * Copyright (C) 2008 by Denys Vlasenko <vda.linux@googlemail.com>
+ *
+ * Licensed under GPLv2, see file LICENSE in this tarball for details.
+ */
+
+/* We cheat here. It is declared as const ptr in libbb.h,
+ * but here we make it live in R/W memory */
+struct globals;
+struct globals *ptr_to_globals;

miscutils/less.c

@@ -153,14 +153,14 @@ struct globals {
 #define term_orig           (G.term_orig         )
 #define term_less           (G.term_less         )
 #define INIT_G() do { \
-		PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	less_gets_pos = -1; \
 	empty_line_marker = "~"; \
 	num_files = 1; \
 	current_file = 1; \
 	eof_error = 1; \
 	terminated = 1; \
-	} while (0)
+} while (0)
 
 /* Reset terminal input to normal */
 static void set_tty_cooked(void)

networking/httpd.c

@@ -322,7 +322,7 @@ struct globals {
 #define http_error_page   (G.http_error_page  )
 #define proxy             (G.proxy            )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	USE_FEATURE_HTTPD_BASIC_AUTH(g_realm = "Web Server Authentication";) \
 	bind_addr_or_port = "80"; \
 	file_size = -1; \

networking/ifenslave.c

@@ -129,7 +129,7 @@ struct globals {
 #define master     (G.master    )
 #define slave      (G.slave     )
 #define INIT_G() do { \
-        PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+        SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 } while (0)
 
 

networking/nc_bloaty.c

@@ -97,7 +97,6 @@ struct globals {
 };
 
 #define G (*ptr_to_globals)
-
 #define wrote_out  (G.wrote_out )
 #define wrote_net  (G.wrote_net )
 #define ouraddr    (G.ouraddr   )
@@ -115,6 +114,10 @@ struct globals {
 #else
 #define o_interval 0
 #endif
+#define INIT_G() do { \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
+} while (0)
+
 
 /* Must match getopt32 call! */
 enum {
@@ -678,9 +681,7 @@ int nc_main(int argc, char **argv)
 	int x;
 	unsigned o_lport = 0;
 
-	/* I was in this barbershop quartet in Skokie IL ... */
-	/* round up the usual suspects, i.e. malloc up all the stuff we need */
-	PTR_TO_GLOBALS = xzalloc(sizeof(G));
+	INIT_G();
 
 	/* catch a signal or two for cleanup */
 	bb_signals(0

networking/sendmail.c

@@ -77,7 +77,7 @@ struct globals {
 #define xargs           (G.xargs     )
 #define fargs           (G.fargs     )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	xargs[0] = "openssl"; \
 	xargs[1] = "s_client"; \
 	xargs[2] = "-quiet"; \

networking/traceroute.c

@@ -378,7 +378,7 @@ struct globals {
 #define wherefrom (G.wherefrom)
 #define gwlist    (G.gwlist   )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	maxpacket = 32 * 1024; \
 	port = 32768 + 666; \
 	waittime = 5; \

procps/nmeter.c

@@ -75,11 +75,11 @@ struct globals {
 #define proc_diskstats     (G.proc_diskstats    )
 #define proc_sys_fs_filenr (G.proc_sys_fs_filenr)
 #define INIT_G() do { \
-		PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	cur_outbuf = outbuf; \
 	final_str = "\n"; \
 	deltanz = delta = 1000000; \
-	} while (0)
+} while (0)
 
 // We depend on this being a char[], not char* - we take sizeof() of it
 #define outbuf bb_common_bufsiz1

runit/svlogd.c

@@ -106,7 +106,7 @@ struct globals {
 #define fl_flag_0      (G.fl_flag_0     )
 #define dirn           (G.dirn          )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	linemax = 1000; \
 	/*buflen = 1024;*/ \
 	linecomplete = 1; \

shell/Kbuild

@@ -5,8 +5,7 @@
 # Licensed under the GPL v2, see the file LICENSE in this tarball.
 
 lib-y:=
-lib-$(CONFIG_ASH)  += ash.o
+lib-$(CONFIG_ASH)      += ash.o ash_ptr_hack.o
 lib-$(CONFIG_HUSH)     += hush.o
 lib-$(CONFIG_MSH)      += msh.o
-
 lib-$(CONFIG_CTTYHACK) += cttyhack.o

shell/ash.c

@@ -202,9 +202,8 @@ struct globals_misc {
 	/* indicates specified signal received */
 	char gotsig[NSIG - 1];
 };
-/* Make it reside in writable memory, yet make compiler understand that it is not going to change. */
-static struct globals_misc *const ptr_to_globals_misc __attribute__ ((section (".data")));
-#define G_misc (*ptr_to_globals_misc)
+extern struct globals_misc *const ash_ptr_to_globals_misc;
+#define G_misc (*ash_ptr_to_globals_misc)
 #define rootpid   (G_misc.rootpid  )
 #define shlvl     (G_misc.shlvl    )
 #define minusc    (G_misc.minusc   )
@@ -223,7 +222,8 @@ static struct globals_misc *const ptr_to_globals_misc __attribute__ ((section ("
 #define sigmode   (G_misc.sigmode  )
 #define gotsig    (G_misc.gotsig   )
 #define INIT_G_misc() do { \
-	(*(struct globals_misc**)&ptr_to_globals_misc) = xzalloc(sizeof(G_misc)); \
+	(*(struct globals_misc**)&ash_ptr_to_globals_misc) = xzalloc(sizeof(G_misc)); \
+	barrier(); \
 	curdir = nullstr; \
 	physdir = nullstr; \
 } while (0)
@@ -1147,9 +1147,8 @@ struct globals_memstack {
 	int    herefd; // = -1;
 	struct stack_block stackbase;
 };
-/* Make it reside in writable memory, yet make compiler understand that it is not going to change. */
-static struct globals_memstack *const ptr_to_globals_memstack __attribute__ ((section (".data")));
-#define G_memstack (*ptr_to_globals_memstack)
+extern struct globals_memstack *const ash_ptr_to_globals_memstack;
+#define G_memstack (*ash_ptr_to_globals_memstack)
 #define g_stackp     (G_memstack.g_stackp    )
 #define markp        (G_memstack.markp       )
 #define g_stacknxt   (G_memstack.g_stacknxt  )
@@ -1158,7 +1157,8 @@ static struct globals_memstack *const ptr_to_globals_memstack __attribute__ ((se
 #define herefd       (G_memstack.herefd      )
 #define stackbase    (G_memstack.stackbase   )
 #define INIT_G_memstack() do { \
-	(*(struct globals_memstack**)&ptr_to_globals_memstack) = xzalloc(sizeof(G_memstack)); \
+	(*(struct globals_memstack**)&ash_ptr_to_globals_memstack) = xzalloc(sizeof(G_memstack)); \
+	barrier(); \
 	g_stackp = &stackbase; \
 	g_stacknxt = stackbase.space; \
 	g_stacknleft = MINSIZE; \
@@ -1778,9 +1778,8 @@ struct globals_var {
 	struct var *vartab[VTABSIZE];
 	struct var varinit[ARRAY_SIZE(varinit_data)];
 };
-/* Make it reside in writable memory, yet make compiler understand that it is not going to change. */
-static struct globals_var *const ptr_to_globals_var __attribute__ ((section (".data")));
-#define G_var (*ptr_to_globals_var)
+extern struct globals_var *const ash_ptr_to_globals_var;
+#define G_var (*ash_ptr_to_globals_var)
 #define shellparam    (G_var.shellparam   )
 #define redirlist     (G_var.redirlist    )
 #define g_nullredirs  (G_var.g_nullredirs )
@@ -1789,7 +1788,8 @@ static struct globals_var *const ptr_to_globals_var __attribute__ ((section (".d
 #define varinit       (G_var.varinit      )
 #define INIT_G_var() do { \
 	int i; \
-	(*(struct globals_var**)&ptr_to_globals_var) = xzalloc(sizeof(G_var)); \
+	(*(struct globals_var**)&ash_ptr_to_globals_var) = xzalloc(sizeof(G_var)); \
+	barrier(); \
 	for (i = 0; i < ARRAY_SIZE(varinit_data); i++) { \
 		varinit[i].flags = varinit_data[i].flags; \
 		varinit[i].text  = varinit_data[i].text; \

shell/ash_ptr_hack.c

@@ -0,0 +1,16 @@
+/* vi: set sw=4 ts=4: */
+/*
+ * Copyright (C) 2008 by Denys Vlasenko <vda.linux@googlemail.com>
+ *
+ * Licensed under GPLv2, see file LICENSE in this tarball for details.
+ */
+
+/* We cheat here. They are declared as const ptr in ash.c,
+ * but here we make them live in R/W memory */
+struct globals_misc;
+struct globals_memstack;
+struct globals_var;
+
+struct globals_misc     *ash_ptr_to_globals_misc;
+struct globals_memstack *ash_ptr_to_globals_memstack;
+struct globals_var      *ash_ptr_to_globals_var;

shell/hush.c

@@ -460,6 +460,9 @@ enum { run_list_level = 0 };
 #endif
 #define charmap          (G.charmap         )
 #define user_input_buf   (G.user_input_buf  )
+#define INIT_G() do { \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
+} while (0)
 
 
 #define B_CHUNK  100
@@ -3778,7 +3781,7 @@ int hush_main(int argc, char **argv)
 	char **e;
 	struct variable *cur_var;
 
-	PTR_TO_GLOBALS = xzalloc(sizeof(G));
+	INIT_G();
 
 	/* Deal with HUSH_VERSION */
 	shell_ver = const_shell_ver; /* copying struct here */

shell/msh.c

@@ -705,7 +705,7 @@ struct globals {
 #define child_cmd       (G.child_cmd      )
 #define iostack         (G.iostack        )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	global_env.linep = line; \
 	global_env.iobase = iostack; \
 	global_env.iop = iostack - 1; \

sysklogd/syslogd.c

@@ -136,7 +136,7 @@ static const struct init_globals init_data = {
 
 #define G (*ptr_to_globals)
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = memcpy(xzalloc(sizeof(G)), &init_data, sizeof(init_data)); \
+	SET_PTR_TO_GLOBALS(memcpy(xzalloc(sizeof(G)), &init_data, sizeof(init_data))); \
 } while (0)
 
 

util-linux/fdisk.c

@@ -309,7 +309,7 @@ struct globals {
 #define MBRbuffer       (G.MBRbuffer)
 #define ptes            (G.ptes)
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	sector_size = DEFAULT_SECTOR_SIZE; \
 	sector_offset = 1; \
 	g_partitions = 4; \

util-linux/fsck_minix.c

@@ -190,7 +190,7 @@ struct globals {
 #define check_file_blk     (G.check_file_blk     )
 #define current_name       (G.current_name       )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	dirsize = 16; \
 	namelen = 14; \
 	current_name[0] = '/'; \

util-linux/mkfs_minix.c

@@ -122,7 +122,7 @@ struct globals {
 };
 #define G (*ptr_to_globals)
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 } while (0)
 
 static ALWAYS_INLINE unsigned div_roundup(unsigned size, unsigned n)

util-linux/more.c

@@ -28,9 +28,7 @@ struct globals {
 	struct termios new_settings;
 };
 #define G (*(struct globals*)bb_common_bufsiz1)
-//#define G (*ptr_to_globals)
 #define INIT_G() ((void)0)
-//#define INIT_G() PTR_TO_GLOBALS = xzalloc(sizeof(G))
 #define initial_settings (G.initial_settings)
 #define new_settings     (G.new_settings    )
 #define cin_fileno       (G.cin_fileno      )

util-linux/script.c

@@ -25,7 +25,7 @@ struct globals {
 #define tt        (G.tt       )
 #define fname     (G.fname    )
 #define INIT_G() do { \
-	PTR_TO_GLOBALS = xzalloc(sizeof(G)); \
+	SET_PTR_TO_GLOBALS(xzalloc(sizeof(G))); \
 	fname = "typescript"; \
 } while (0)