The boards using CONFIG_SYS_DV_NOR_BOOT_CFG (i.e. calimain, da850evm_direct_nor and enbw_cmc) had the _start symbol defined after the CONFIG_SYS_DV_NOR_BOOT_CFG word rather than before it in arch/arm/lib/vectors.S. Because of that, if by lack of luck 'gd->mon_len = (ulong)&__bss_end - (ulong)_start' (see setup_mon_len()) was a multiple of 4 kiB (see reserve_uboot()), then the last BSS word overlapped the first word of the following reserved RAM area (or went beyond the top of RAM without such an area) after relocation because __image_copy_start did not match _start (see relocate_code()).

This was broken by commit 41623c9 'arm: move exception handling out of start.S files', which defined _start twice (before and after the CONFIG_SYS_DV_NOR_BOOT_CFG word), then by commit 0a26e1d 'arm: fix a double-definition error of _start symbol', which kept the definition of the _start symbol after the CONFIG_SYS_DV_NOR_BOOT_CFG word. This new commit fixes this issue by restoring the original behavior, i.e. by defining the _start symbol before the CONFIG_SYS_DV_NOR_BOOT_CFG word.

Signed-off-by: Benoît Thébaudeau benoit.thebaudeau.dev@gmail.com Cc: Albert Aribaud albert.u.boot@aribaud.net Cc: Manfred Rudigier manfred.rudigier@omicron.at Cc: Christian Riesch christian.riesch@omicron.at Cc: Sudhakar Rajashekhara sudhakar.raj@ti.com Cc: Heiko Schocher hs@denx.de --- arch/arm/lib/vectors.S | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/arch/arm/lib/vectors.S b/arch/arm/lib/vectors.S index 843b18f..0cb87ce 100644 --- a/arch/arm/lib/vectors.S +++ b/arch/arm/lib/vectors.S @@ -45,11 +45,12 @@ ************************************************************************* */

+_start: + #ifdef CONFIG_SYS_DV_NOR_BOOT_CFG .word CONFIG_SYS_DV_NOR_BOOT_CFG #endif

-_start: b reset ldr pc, _undefined_instruction ldr pc, _software_interrupt

On Wed, Sep 3, 2014 at 11:32 PM, Benoît Thébaudeau benoit.thebaudeau.dev@gmail.com wrote:

Some boards, like mx31pdk and tx25, require the beginning of the SPL code to be position-independent. For these two boards, this is because they use the i.MX external NAND boot, which starts by executing the first NAND Flash page from the NFC page buffer. The SPL then needs to copy itself to its actual link address in order to free the NFC page buffer and use it to load the non-SPL image from Flash before running it. This means that the SPL runtime address differs from its link address between the reset and the initial copy performed by board_init_f(), so this part of the SPL binary must be position-independent.

This requirement was broken by commit 41623c9 'arm: move exception handling out of start.S files', which used an absolute address to branch to the reset routine. This new commit restores the original behavior, which just performed a relative branch. This fixes the boot of mx31pdk and tx25.

Signed-off-by: Benoît Thébaudeau benoit.thebaudeau.dev@gmail.com Reported-by: Helmut Raiger helmut.raiger@hale.at Cc: Albert Aribaud albert.u.boot@aribaud.net Cc: Magnus Lilja lilja.magnus@gmail.com Cc: John Rigby jcrigby@gmail.com

---

arch/arm/lib/vectors.S | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/arch/arm/lib/vectors.S b/arch/arm/lib/vectors.S index 493f337..843b18f 100644 --- a/arch/arm/lib/vectors.S +++ b/arch/arm/lib/vectors.S @@ -50,7 +50,7 @@ #endif

_start:

•   ldr     pc, _reset
  

•   b       reset
    ldr     pc, _undefined_instruction
    ldr     pc, _software_interrupt
    ldr     pc, _prefetch_abort
  

@@ -77,7 +77,6 @@ _start: .globl _irq .globl _fiq

-_reset: .word reset _undefined_instruction: .word undefined_instruction _software_interrupt: .word software_interrupt _prefetch_abort: .word prefetch_abort -- 1.7.10.4

Magnus, can I have your 'Tested-by' for mx31pdk since you said you have tested this?

Tom, Albert, can you build-test all ARM boards with this patch (that would take eons on my ultra slow machine)? It would also be nice if someone could runtime-test an ARM board other than mx31pdk and tx25 with this patch.

Thanks in advance.

Best regards, Benoît

Hi Benoît,

On Wed, 3 Sep 2014 23:32:33 +0200, Benoît Thébaudeau benoit.thebaudeau.dev@gmail.com wrote:

Some boards, like mx31pdk and tx25, require the beginning of the SPL code to be position-independent. For these two boards, this is because they use the i.MX external NAND boot, which starts by executing the first NAND Flash page from the NFC page buffer. The SPL then needs to copy itself to its actual link address in order to free the NFC page buffer and use it to load the non-SPL image from Flash before running it. This means that the SPL runtime address differs from its link address between the reset and the initial copy performed by board_init_f(), so this part of the SPL binary must be position-independent.

This requirement was broken by commit 41623c9 'arm: move exception handling out of start.S files', which used an absolute address to branch to the reset routine. This new commit restores the original behavior, which just performed a relative branch. This fixes the boot of mx31pdk and tx25.

Signed-off-by: Benoît Thébaudeau benoit.thebaudeau.dev@gmail.com Reported-by: Helmut Raiger helmut.raiger@hale.at Cc: Albert Aribaud albert.u.boot@aribaud.net Cc: Magnus Lilja lilja.magnus@gmail.com Cc: John Rigby jcrigby@gmail.com

---

arch/arm/lib/vectors.S | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)

diff --git a/arch/arm/lib/vectors.S b/arch/arm/lib/vectors.S index 493f337..843b18f 100644 --- a/arch/arm/lib/vectors.S +++ b/arch/arm/lib/vectors.S @@ -50,7 +50,7 @@ #endif

_start:

• ldr pc, _reset

• b reset ldr pc, _undefined_instruction ldr pc, _software_interrupt ldr pc, _prefetch_abort

@@ -77,7 +77,6 @@ _start: .globl _irq .globl _fiq

-_reset: .word reset _undefined_instruction: .word undefined_instruction _software_interrupt: .word software_interrupt _prefetch_abort: .word prefetch_abort

Applied (as a bug fix) to u-boot-arm/master, thanks!

Amicalement,