[SPAM] Re: [PATCH v2] xilinx: zynqmp: Do not use 0 as spl bss start address

[Michal Simek]

Hi,

On 6/13/22 11:02, Stefan Herbrechtsmeier wrote:
> [CAUTION: External Email]
> 
> Hi Michal,
> 
> Am 13.06.2022 um 10:21 schrieb Stefan Herbrechtsmeier:
>> Hi Michal,
>>
>> Am 13.06.2022 um 09:20 schrieb Michal Simek:
>>> Hi,
>>>
>>> On 6/10/22 18:48, Xavier Drudis Ferran wrote:
>>>> El Fri, Jun 10, 2022 at 04:42:55PM +0200, Stefan Herbrechtsmeier deia:
>>>>> Hi Michal,
>>>>>
>>>>> what is the default entry address for the aft / bl31.bin?
>>>>>
>>>>> I have a bl31.bin with an entry address of 0x1000 and this is inside
>>>>> the
>>>>> BSS.
>>>>>
>>>>
>>>> Me too, load address at 0x1000, but for me in SPL text, not BSS.
>>>>
>>>> I have a litle customized, a little old TF-A  for rk3399 / Rock pi 4
>>>> loading at address 0 with entry at 0x1000.
>>>>
>>>> But include/configs/rk3399_common.h sets my
>>>> CONFIG_SPL_BSS_START_ADDR=0x400000, away from harm.
>>>> I had problems booting anyway.
>>>>
>>>> Now I can load U-Boot from MMC with these patches
>>>> https://lists.denx.de/pipermail/u-boot/2022-June/485497.html
>>>>
>>>> In particular
>>>> CONFIG_ROCKCHIP_SPL_RESERVE_IRAM=0x15000
>>>>
>>>> This is defined in arch/arm/mach-rockchip/Kconfig and says it's
>>>> to avoid conflicts with SPL text area, not BSS
>>>>
>>>> But I found other boards with CONFIG_ROCKCHIP_SPL_RESERVE_IRAM=0x4000,
>>>> so I thought some low addresses where normal. I don't know.
>>>>
>>>> I had to modify the code loading from SPI because, unlike MMC code, it
>>>> thought address 0 meant no destination (I can send those patches when
>>>> I have them cleaner if anyone wants them).
>>>>
>>>> I just realised that I have CONFIG_SPL_TEXT_BASE=0x0.  I'm not finding
>>>> where that's defined, maybe it's simply because it's not defined
>>>> anywhere, so maybe the solution for me would be setting
>>>> CONFIG_SPL_TEXT_BASE
>>>> to 0x10000000 or something. Or maybe it needs to be at 0x0 because
>>>> it is bootrom who is loading it, and it won't look where I define it?
>>>> I can't remember whether I tried this.
>>>>
>>>> Maybe you can try to look at the size of a file bl31_0x00000000.bin
>>>> that is generated when you build U-boot with BL31 pointing at your
>>>> bl31.elf (check u-boot.its if that's not the name for you).
>>>>
>>>> Then set CONFIG_SPL_BSS_START_ADDR to that size + L (L= value of load
>>>> property
>>>> in entry atf_1 of u-boot.its). This should leave a hole at the beginning
>>>> of U-Boot to make room for your TF-A, and leave BSS elsewhere.
>>>>
>>>> The sources and build scripts for TF-A are public, so maybe one could
>>>> look at what's the criteria for putting images at different addresses?
>>>>
>>>
>>> I have no idea what rockchip has to do with memory layout on zynqmp.
>>> Anyway SPL must be placed to OCM that's why text base is at 0xfffc0000
>>> where OCM starts and which is also default address where a53s starts.
>>>
>>> TF-A can be placed to any address but address which I use in SPL flow
>>> is 0xfffe5000. PetaLinux which uses FSBL is using different address
>>> 0xfffea000 which is default in TF-A.
>>> In DDR I normally use end of low ddr locations. It means 0x7fxxxxxx.
>>>
>>> And of course you can't place TFA to location which is used by SPL or
>>> by u-boot or other SW.
>>
>> Thanks for the information. I will check why our TF-A is linked to the
>> lower DDR and move it back to the default address.
> 
> The TF-A was build with debug flag [1]:
> 
>  > By default, the Arm-trusted firmware builds for OCM space at address
>  > 0xFFFEA000. But, with DEBUG flag set to 1, it can't fit in OCM, so by
>  > default with DEBUG=1, it builds for DDR location 0x1000 with build
>  > flag DEBUG=1 mentioned while building.
> 
> Either the CONFIG_SPL_BSS_START_ADDR or the default location of the TF-A
> with DEBUG flag should be changed.
> 
> [1]
> https://xilinx-wiki.atlassian.net/wiki/spaces/A/pages/18842107/Arm+Trusted+Firmware

It is really question what should be default address with DEBUG=1.

I tried this

diff --git a/plat/xilinx/zynqmp/include/platform_def.h 
b/plat/xilinx/zynqmp/include/platform_def.h
index 9c1600a7b7e3..66bbf30a65a4 100644
--- a/plat/xilinx/zynqmp/include/platform_def.h
+++ b/plat/xilinx/zynqmp/include/platform_def.h
@@ -40,8 +40,8 @@
  # define BL31_BASE                     U(0xfffea000)
  # define BL31_LIMIT                    U(0x100000000)
  #else
-# define BL31_BASE                     U(0x1000)
-# define BL31_LIMIT                    U(0x7ffff)
+# define BL31_BASE                     U(0xfff5a000)
+# define BL31_LIMIT                    U(0x100000000)
  #endif
  #else
  # define BL31_BASE                     (ZYNQMP_ATF_MEM_BASE)


And with the latest TF-A this works fine.
No problem to change TF-A code to place it to different addresses. I think it 
should stay in OCM if possible or allocate space at the end of of low ddr 
address. It means any 0x7fxxxxxx address should be fine.
With any SPD on we won't be able to fit OCM anyway.

Thanks,
Michal