On Tue, Jan 17, 2017 at 9:55 AM, Masahiro Yamada yamada.masahiro@socionext.com wrote:

2016-12-26 23:20 GMT+09:00 Oded Gabbay oded.gabbay@gmail.com:

...

This patch copies from arm u-boot-spl.lds some asserts that check that the size of the SPL image and BSS doesn't violate their max size.

Signed-off-by: Oded Gabbay oded.gabbay@gmail.com Cc: Albert Aribaud albert.u.boot@aribaud.net

---

arch/arm/cpu/armv8/u-boot-spl.lds | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+)

diff --git a/arch/arm/cpu/armv8/u-boot-spl.lds b/arch/arm/cpu/armv8/u-boot-spl.lds index e7799cc..0876a94 100644 --- a/arch/arm/cpu/armv8/u-boot-spl.lds +++ b/arch/arm/cpu/armv8/u-boot-spl.lds @@ -78,3 +78,22 @@ SECTIONS /DISCARD/ : { *(.interp*) } /DISCARD/ : { *(.gnu*) } }

+#if defined(CONFIG_SPL_MAX_SIZE) +ASSERT(__image_copy_end - __image_copy_start < (CONFIG_SPL_MAX_SIZE), \

•   "SPL image too big");
  

+#endif

+#if defined(CONFIG_SPL_BSS_MAX_SIZE) && defined(CONFIG_SPL_MAX_FOOTPRINT) +ASSERT(0, "CONFIG_SPL_BSS_MAX_SIZE and CONFIG_SPL_MAX_FOOTPRINT must not be defined together"); +#endif

+#if defined(CONFIG_SPL_BSS_MAX_SIZE) +ASSERT(__bss_end - __bss_start < (CONFIG_SPL_BSS_MAX_SIZE), \

•   "SPL image BSS too big");
  

+#endif

+#if defined(CONFIG_SPL_MAX_FOOTPRINT) +ASSERT(__bss_end - _start < (CONFIG_SPL_MAX_FOOTPRINT), \

•   "SPL image plus BSS too big");
  

+#endif

2.7.4

This patch is wrong in multiple ways.

See the top lines of this file.

MEMORY { .sram : ORIGIN = CONFIG_SPL_TEXT_BASE, LENGTH = CONFIG_SPL_MAX_SIZE } MEMORY { .sdram : ORIGIN = CONFIG_SPL_BSS_START_ADDR, LENGTH = CONFIG_SPL_BSS_MAX_SIZE }

This means CONFIG_SPL_BSS_MAX_SIZE must always be defined.

Your code

...

+#if defined(CONFIG_SPL_BSS_MAX_SIZE) && defined(CONFIG_SPL_MAX_FOOTPRINT) +ASSERT(0, "CONFIG_SPL_BSS_MAX_SIZE and CONFIG_SPL_MAX_FOOTPRINT must not be defined together"); +#endif

... means CONFIG_SPL_BSS_MAX_FOOTPRINT can never be defined.

Correct. According to the README:

"CONFIG_SPL_MAX_FOOTPRINT and CONFIG_SPL_BSS_MAX_SIZE must not be both defined at the same time."

So this assert checks this. If this is wrong, then we should fix the README instead.

As a result,

...

+#if defined(CONFIG_SPL_BSS_MAX_SIZE) +ASSERT(__bss_end - __bss_start < (CONFIG_SPL_BSS_MAX_SIZE), \

•   "SPL image BSS too big");
  

+#endif

the ifdef is always true, so meaningless.

Correct, we can remove this #ifdef

...

+ASSERT(__bss_end - __bss_start < (CONFIG_SPL_BSS_MAX_SIZE), \

•   "SPL image BSS too big");
  

is already size-checked by the following:

MEMORY { .sdram : ORIGIN = CONFIG_SPL_BSS_START_ADDR, LENGTH = CONFIG_SPL_BSS_MAX_SIZE }

so, meaningless.

...

+#if defined(CONFIG_SPL_MAX_FOOTPRINT) +ASSERT(__bss_end - _start < (CONFIG_SPL_MAX_FOOTPRINT), \

•   "SPL image plus BSS too big");
  

+#endif

This code is never enabled.

Correct, assuming we leave the above "#if defined(CONFIG_SPL_BSS_MAX_SIZE) && defined(CONFIG_SPL_MAX_FOOTPRINT)" in place.

Thanks, Oded

-- Best Regards Masahiro Yamada

Hi.

(CCing Simon because he wrote the code addressed in this discussion.)

2017-01-17 18:27 GMT+09:00 Oded Gabbay oded.gabbay@gmail.com:

Hi Mashiro, Could you please also take a look at the patch I sent to add the missing __bss_size ?

Thanks, Oded

On Mon, Dec 26, 2016 at 4:20 PM, Oded Gabbay oded.gabbay@gmail.com wrote:

...

This patch adds a missing __bss_size symbol to the default armv8 u-boot-spl.lds file. Makefile.spl relies on __bss_size to be present when it creates the SPL image. It uses that symbol to create a pad file that will be used to place the dtb after the bss section. In ARMv8 default u-boot-spl.lds, __bss_size was missing and therefore, the pad file was always 0. As a result, the dtb was placed after _image_binary_end, which caused a failure when loading it inside the SPL.

Signed-off-by: Oded Gabbay oded.gabbay@gmail.com Cc: Albert Aribaud albert.u.boot@aribaud.net

---

arch/arm/cpu/armv8/u-boot-spl.lds | 1 + 1 file changed, 1 insertion(+)

diff --git a/arch/arm/cpu/armv8/u-boot-spl.lds b/arch/arm/cpu/armv8/u-boot-spl.lds index cc427c3..e7799cc 100644 --- a/arch/arm/cpu/armv8/u-boot-spl.lds +++ b/arch/arm/cpu/armv8/u-boot-spl.lds @@ -70,6 +70,7 @@ SECTIONS KEEP(*(.__bss_end)); } >.sdram

•   __bss_size = __bss_end - _image_binary_end;
    /DISCARD/ : { *(.dynsym) }
    /DISCARD/ : { *(.dynstr*) }
    /DISCARD/ : { *(.dynamic*) }
  

This patch seems wrong.

This linker script seems to support separate two memory regions.

MEMORY { .sram : ORIGIN = CONFIG_SPL_TEXT_BASE, LENGTH = CONFIG_SPL_MAX_SIZE } MEMORY { .sdram : ORIGIN = CONFIG_SPL_BSS_START_ADDR, LENGTH = CONFIG_SPL_BSS_MAX_SIZE }

_image_binary_end is the end of the first region (.sram), __bss_end is the end of the second region (.sdram).

So, with your patch,

__bss_size = (gap between .sram and .sdram) + (size of .bss)

This will append really big padding there is a gap between the two regions. (at least, it will break my boards.)

If you take a look at fdtdec_setup(),

# ifdef CONFIG_SPL_BUILD /* FDT is at end of BSS unless it is in a different memory region */ if (IS_ENABLED(CONFIG_SPL_SEPARATE_BSS)) gd->fdt_blob = (ulong *)&_image_binary_end; else gd->fdt_blob = (ulong *)&__bss_end;

I think you are supposed to enable CONFIG_SPL_SEPARATE_BSS if you want to use SPL_OF_CONTROL on your arm64 boards.

Checking the following commit will help you.

commit 10172962479ddd6609101fdb83bde66c0719852c Author: Simon Glass sjg@chromium.org Date: Sat Oct 17 19:41:19 2015 -0600

dm: spl: Support device tree when BSS is in a different section

At present in SPL we place the device tree immediately after BSS. This avoids needing to copy it out of the way before BSS can be used. However on some boards BSS is not placed with the image - e.g. it can be in RAM if available.

Add an option to tell U-Boot that the device tree should be placed at the end of the image binary (_image_binary_end) instead of at the end of BSS.

Note: A common reason to place BSS in RAM is to support the FAT filesystem. We should update the code so that it does not use so much BSS.

Signed-off-by: Simon Glass sjg@chromium.org Signed-off-by: Michal Simek michal.simek@xilinx.com