ARM: cleanup gd init
Subsystems (3)
cover ARM: cleanup gd init
Hello,
As I noticed at [1] gd is setup multiple times. As pointed Tom pointed out out me this actually leads to problems as well, see [2]. These patches attempt to cleanup gd usage (a bit, board_f.c is not included since that one is rather trivial), some questions I have about it:
1) The major one, do these init changes brick any board?
2) Is there any board which needs gdata as a global?
3) What to do with lowlevel_init on armv7. Hide it under a CONFIG_*, make it weak.. or just remove it completely?
4) Keep the s_init in crt0.S or move it to the board_init_f? The disadvantage of the later is that all the different board_init_f's need to call system_init.
5) Where to put the __weak s_init (or hide this under a define)
Regards,
Jeroen
[1] http://lists.denx.de/pipermail/u-boot/2013-August/160933.html
[2] http://lists.denx.de/pipermail/u-boot/2013-July/158144.html
Jeroen Hofstee (3):
ARM,crt0.S: call s_init instead from ctr0.S ARM,crt0.S: optional init gd to gdata for spl
ARM: do not assign gd outside of crt0.S
arch/arm/cpu/arm926ejs/davinci/spl.c | 3 +-- arch/arm/cpu/armv7/exynos/spl_boot.c | 6 ++---- arch/arm/cpu/armv7/lowlevel_init.S | 23 +---------------------- arch/arm/cpu/armv7/omap-common/hwinit-common.c | 2 -- arch/arm/cpu/armv7/omap3/board.c | 2 -- arch/arm/cpu/armv7/omap3/lowlevel_init.S | 8 ++++---- arch/arm/cpu/armv7/rmobile/lowlevel_init.S | 6 ------ arch/arm/lib/crt0.S | 8 ++++++++ arch/arm/lib/reset.c | 5 +++++ arch/arm/lib/spl.c | 5 ++--- board/isee/igep0033/board.c | 1 - board/phytec/pcm051/board.c | 2 -- board/ti/am335x/board.c | 2 -- board/ti/omap5912osk/lowlevel_init.S | 11 ----------- board/ti/ti814x/evm.c | 2 -- board/woodburn/woodburn.c | 3 --- 16 files changed, 23 insertions(+), 66 deletions(-)
Signature · 1 line
1.8.1.2
RFC 1/3 ARM,crt0.S: call s_init instead from ctr0.S
The only reason gd is setup before _main seems to be the call to s_init. Therefore call s_init in crt0.S after gd has been setup. (It might become system_init one day, but that has the disadvantage that all versions of board_init_f must call system_init).
--- arch/arm/cpu/armv7/lowlevel_init.S | 23 +---------------------- arch/arm/cpu/armv7/omap3/lowlevel_init.S | 8 ++++---- arch/arm/cpu/armv7/rmobile/lowlevel_init.S | 6 ------ arch/arm/lib/crt0.S | 3 +++ arch/arm/lib/reset.c | 5 +++++ board/ti/omap5912osk/lowlevel_init.S | 11 ----------- 6 files changed, 13 insertions(+), 43 deletions(-)
arch/arm/cpu/armv7/lowlevel_init.S +1 -22
@@ -16,26 +16,5 @@ #include <linux/linkage.h> ENTRY(lowlevel_init) - /* - * Setup a temporary stack - */ - ldr sp, =CONFIG_SYS_INIT_SP_ADDR - bic sp, sp, #7 /* 8-byte alignment for ABI compliance */ -#ifdef CONFIG_SPL_BUILD - ldr r8, =gdata -#else - sub sp, #GD_SIZE - bic sp, sp, #7 - mov r8, sp -#endif - /* - * Save the old lr(passed in ip) and the current lr to stack - */ - push {ip, lr} - - /* - * go setup pll, mux, memory - */ - bl s_init - pop {ip, pc} + mov pc, lr ENDPROC(lowlevel_init)
arch/arm/cpu/armv7/omap3/lowlevel_init.S +4 -4
@@ -197,22 +197,22 @@ pll_div_val5: #endif ENTRY(lowlevel_init) +#if !defined(CONFIG_SYS_NAND_BOOT) && !defined(CONFIG_SYS_ONENAND_BOOT) ldr sp, SRAM_STACK str ip, [sp] /* stash ip register */ mov ip, lr /* save link reg across call */ -#if !defined(CONFIG_SYS_NAND_BOOT) && !defined(CONFIG_SYS_ONENAND_BOOT) /* * No need to copy/exec the clock code - DPLL adjust already done * in NAND/oneNAND Boot. */ ldr r1, =SRAM_CLK_CODE bl cpy_clk_code -#endif /* NAND Boot */ + mov lr, ip /* restore link reg */ ldr ip, [sp] /* restore save ip */ - /* tail-call s_init to setup pll, mux, memory */ - b s_init +#endif /* NAND Boot */ + mov pc, lr ENDPROC(lowlevel_init) /* the literal pools origin */
arch/arm/cpu/armv7/rmobile/lowlevel_init.S +0 -6
@@ -59,14 +59,8 @@ loop0: subs r0, r0, #1 bne loop0 - ldr sp, MERAM_STACK - b s_init - .pool .align 4 ENDPROC(lowlevel_init) .ltorg - -MERAM_STACK: - .word LOW_LEVEL_MERAM_STACK
arch/arm/lib/crt0.S +3 -0
@@ -70,6 +70,9 @@ ENTRY(_main) sub sp, #GD_SIZE /* allocate one GD above SP */ bic sp, sp, #7 /* 8-byte alignment for ABI compliance */ mov r8, sp /* GD is above SP */ + + bl s_init + mov r0, #0 bl board_init_f
arch/arm/lib/reset.c +5 -0
@@ -35,3 +35,8 @@ int do_reset(cmd_tbl_t *cmdtp, int flag, int argc, char * const argv[])/*NOTREACHED*/ return 0 ; } + +/* HACK */ +__weak void s_init (void ) +{ +}
board/ti/omap5912osk/lowlevel_init.S +0 -11
@@ -296,17 +296,6 @@ common_tc: ldr sp, SRAM_STACK bic sp, sp, #7 /* 8-byte alignment for ABI compliance */ - /* - * Save the old lr(passed in ip) and the current lr to stack - */ - push {ip, lr} - - /* - * go setup pll, mux, memory - */ - bl s_init - pop {ip, pc} - /* back to arch calling code */ mov pc, lr
Signature · 1 line
1.8.1.2
RFC 1/3 ARM,crt0.S: call s_init instead from ctr0.S
arch/arm/cpu/armv7/rmobile/lowlevel_init.S loop0: · line 64
Quoted 12 lines · Jeroen Hofstee
/* the literal pools origin */
arch/arm/cpu/armv7/rmobile/lowlevel_init.S +0
-3@@ -59,14 +59,8 @@ loop0: subs r0, r0, #1 bne loop0 - ldr sp, MERAM_STACK - b s_init -
right, this ^^^ will brick things nicely, how about returning to the caller...
Quoted 8 lines
arch/arm/cpu/armv7/rmobile/lowlevel_init.S +0
-3.pool .align 4 ENDPROC(lowlevel_init) .ltorg - -MERAM_STACK: - .word LOW_LEVEL_MERAM_STACK
RFC 2/3 ARM,crt0.S: optional init gd to gdata for spl
In reply to Jeroen HofsteeThe common/spl changes gd to gdata in board_init_f. I fail to see why. The only reason I can think of to use the gdata is the case where is won't fit into the region where the stack lives.
Move the init to crt0.S and make it a CONFIG.
--- arch/arm/lib/crt0.S | 5 +++++ arch/arm/lib/spl.c | 5 ++--- 2 files changed, 7 insertions(+), 3 deletions(-)
arch/arm/lib/crt0.S +5 -0
@@ -67,9 +67,14 @@ ENTRY(_main) ldr sp, =(CONFIG_SYS_INIT_SP_ADDR) #endif bic sp, sp, #7 /* 8-byte alignment for ABI compliance */ + +#if defined(CONFIG_SPL_BUILD) && defined(CONFIG_SPL_GD_GLOBAL) + ldr r8, =gdata /* SPL assigns GD directly to &gdata */ +#else sub sp, #GD_SIZE /* allocate one GD above SP */ bic sp, sp, #7 /* 8-byte alignment for ABI compliance */ mov r8, sp /* GD is above SP */ +#endif bl s_init
arch/arm/lib/spl.c +2 -3
@@ -15,7 +15,9 @@/* Pointer to as well as the global data structure for SPL */ DECLARE_GLOBAL_DATA_PTR ; +#ifdef CONFIG_SPL_GD_GLOBAL gd_t gdata __attribute__ ((section(".data"))); +#endif /* * In the context of SPL, board_init_f must ensure that any clocks/etc for
@@ -31,9 +33,6 @@ void __weak board_init_f(ulong dummy)/* Clear the BSS. */ memset (__bss_start,0 , __bss_end - __bss_start); -/* Set global data pointer. */ - gd = &gdata; -board_init_r (NULL ,0 ); }
Signature · 1 line
1.8.1.2
RFC 3/3 ARM: do not assign gd outside of crt0.S
In reply to Jeroen Hofstee--- arch/arm/cpu/arm926ejs/davinci/spl.c | 3 +-- arch/arm/cpu/armv7/exynos/spl_boot.c | 6 ++---- arch/arm/cpu/armv7/omap-common/hwinit-common.c | 2 -- arch/arm/cpu/armv7/omap3/board.c | 2 -- board/isee/igep0033/board.c | 1 - board/phytec/pcm051/board.c | 2 -- board/ti/am335x/board.c | 2 -- board/ti/ti814x/evm.c | 2 -- board/woodburn/woodburn.c | 3 --- 9 files changed, 3 insertions(+), 20 deletions(-)
arch/arm/cpu/arm926ejs/davinci/spl.c +1 -2
@@ -50,8 +50,7 @@ void board_init_f(ulong dummy)/* Third, we clear the BSS. */ memset (__bss_start,0 , __bss_end - __bss_start); -/* Finally, setup gd and move to the next step. */ - gd = &gdata; +/* Finally move to the next step. */ board_init_r (NULL ,0 ); }
arch/arm/cpu/armv7/exynos/spl_boot.c +2 -4
@@ -149,9 +149,8 @@ void memzero(void *s, size_t n)* * @param gdp Value to give to gd */ -static void setup_global_data (gd_t *gdp) +static void setup_global_data (void ) { - gd = gdp;memzero ((void *)gd,sizeof (gd_t)); gd->flags |=GD_FLG_RELOC ; gd->baudrate =CONFIG_BAUDRATE ;
@@ -160,10 +159,9 @@ static void setup_global_data(gd_t *gdp)void board_init_f (unsigned long bootflag) { -__aligned(8) gd_t local_gd; __attribute__((noreturn))void (*uboot)(void ); -setup_global_data (&local_gd); +setup_global_data ();if (do_lowlevel_init ())power_exit_wakeup ();
arch/arm/cpu/armv7/omap-common/hwinit-common.c +0 -2
@@ -143,8 +143,6 @@ void s_init(void)srcomp_enable ();setup_clocks_for_console (); - gd = &gdata; -preloader_console_init ();do_io_settings ();#endif
arch/arm/cpu/armv7/omap3/board.c +0 -2
@@ -240,8 +240,6 @@ void s_init(void)#endif #ifdef CONFIG_SPL_BUILD - gd = &gdata; -preloader_console_init ();timer_init ();
board/isee/igep0033/board.c +0 -1
@@ -102,7 +102,6 @@ void s_init(void)enable_uart0_pin_mux ();uart_soft_reset (); - gd = &gdata;preloader_console_init ();
board/phytec/pcm051/board.c +0 -2
@@ -113,8 +113,6 @@ void s_init(void)enable_uart0_pin_mux ();uart_soft_reset (); - gd = &gdata; -preloader_console_init ();/* Initalize the board header */
board/ti/am335x/board.c +0 -2
@@ -328,8 +328,6 @@ void s_init(void)uart_soft_reset (); - gd = &gdata; -preloader_console_init ();/* Initalize the board header */
board/ti/ti814x/evm.c +0 -2
@@ -143,8 +143,6 @@ void s_init(void)/* Enable UART */ uart_enable (); - gd = &gdata; -preloader_console_init ();config_dmm (&evm_lisa_map_regs);
board/woodburn/woodburn.c +0 -3
@@ -137,9 +137,6 @@ void board_init_f(ulong dummy)/* Clear the BSS. */ memset (__bss_start,0 , __bss_end - __bss_start); -/* Set global data pointer. */ - gd = &gdata; -preloader_console_init ();timer_init ();
Signature · 1 line
1.8.1.2
cover ARM: cleanup gd init
In reply to Jeroen HofsteeHello,
Quoted 3 lines · Jeroen Hofstee
4) Keep the
s_initincrt0.Sor move it to theboard_init_f? The disadvantage of the later is that all the different board_init_f's need to callsystem_init.
Since this has been on the mailing-list for a month without a reply, let's push this a bit.
Moving s_init to board_init_f is not smart since there are many board_init_f already and likely there will be more in the future since SPL uses it as well. It also provides a chance to save bootrom registers to gd. After boad_init_f r1 is clobbered at least..
Regards,
Jeroen
cover ARM: cleanup gd init
Hi Jeroen,
Quoted 6 lines · Jeroen Hofstee
Hello,
Quoted 3 lines · Jeroen Hofstee
4) Keep the
s_initincrt0.Sor move it to theboard_init_f? The disadvantage of the later is that all the different board_init_f's need to callsystem_init.
Quoted 7 lines
Since this has been on the mailing-list for a month without a reply, let's push this a bit.
Moving
s_inittoboard_init_fis not smart since there are manyboard_init_falready and likely there will be more in the future since SPL uses it as well. It also provides a chance to save bootrom registers to gd. Afterboad_init_fr1 is clobbered at least..
Not sure what you mean by there being "many" board_init_f()s -- I count five occurrences, one of which is the standard board_init_f(), one is a weak default in SPL, and three are custom versions; the last four may or may not be merged into 'the' board_init_f() one.
Regarding saving bootrom (or other) registers, that is a feature for start.S, more precisely from _start, which is the only place in the whole of U-Boot that has not clobbered any register yet.
The right process is thus to split s_init(): any part that deals with saving registers at boot should go in start.S; any part that deals with anything else and is not absolutely required before entering crt0.S should move in a function called from board_init_f().
Quoted 2 lines
Regards,
Jeroen
Amicalement,
Signature · 1 line
Albert.