Hi Albert,
On Wed, Nov 11, 2015 at 6:33 PM, Albert ARIBAUD albert.u.boot@aribaud.net wrote:
+#if defined(CONFIG_SYS_MALLOC_F_LEN)
sub sp, sp, #CONFIG_SYS_MALLOC_F_LENstr sp, [r9, #GD_MALLOC_BASE]+#endif
NAK, as this only papers over the actual issue. Board_init_f_mem should have set the malloc base in GD. Therefore, rather than doing it again later, we must determine why it was not properly done earlier.11111
Can you give me the toolchain version, board name and commit ID that I could use to reproduce the *faulty* build and check the generated code?
Sure, I am testing top of head U-boot (cad049907). Target is mx6sabresd_spl_defconfig.
Toolchain is: arm-linux-gnueabi-gcc (Ubuntu/Linaro 4.7.3-12ubuntu1) 4.7.3
In order to reproduce the malloc failure, please apply this patch against mainline:
--- a/board/freescale/mx6sabresd/mx6sabresd.c +++ b/board/freescale/mx6sabresd/mx6sabresd.c @@ -30,6 +30,7 @@ #include "../common/pfuze.h" #include <asm/arch/mx6-ddr.h> #include <usb.h> +#include <malloc.h>
DECLARE_GLOBAL_DATA_PTR;
@@ -833,6 +834,8 @@ static void spl_dram_init(void)
void board_init_f(ulong dummy) { + void __iomem *ptr; + /* setup AIPS and disable watchdog */ arch_cpu_init();
@@ -848,6 +851,12 @@ void board_init_f(ulong dummy) /* UART clocks enabled and gd valid - init serial console */ preloader_console_init();
+ spl_init(); + + ptr = malloc(64); + if (!ptr) + puts("******* malloc returned NULL\n"); + /* DDR initialization */ spl_dram_init();
Also, as I just explained to Simon if I remove the ifdefery like this:
--- a/common/init/board_init.c +++ b/common/init/board_init.c @@ -50,11 +50,8 @@ ulong board_init_f_mem(ulong top) #endif arch_setup_gd(gd_ptr);
-#if defined(CONFIG_SYS_MALLOC_F) && \ - (!defined(CONFIG_SPL_BUILD) || !defined(CONFIG_SYS_SPL_MALLOC_START)) top -= CONFIG_SYS_MALLOC_F_LEN; gd->malloc_base = top; -#endif
return top; }
Then malloc() works fine in SPL.
So it seems I need to find a way to make CONFIG_SPL_BUILD=n or CONFIG_SYS_SPL_MALLOC_START=n.
Thanks,
Fabio Estevam