configs: ast2600: Move SPL bss section to DRAM space
Subsystems (3)
PATCH 1/1 configs: ast2600: Move SPL bss section to DRAM space
The commit b583348ca8c8 ("image: fit: Align hash output buffers") places the hash output buffer at the .bss section. However, AST2600 by default executes SPL in the NOR flash XIP way. This results in the hash output cannot be written to the buffer as it is located at the R/X only region.
We need to move the .bss section out of the SPL body to the DRAM space, where hash output can be written to. This patch includes:
- Define the .bss section base and size
- A new SPL linker script is added with a separate .bss region specified
- Enable CONFIG_SPL_SEPARATE_BSS kconfig option
--- arch/arm/mach-aspeed/ast2600/u-boot-spl.lds | 94 +++++++++++++++++++++ configs/evb-ast2600_defconfig | 3 + include/configs/evb_ast2600.h | 3 + 3 files changed, 100 insertions(+) create mode 100644 arch/arm/mach-aspeed/ast2600/u-boot-spl.lds
arch/arm/mach-aspeed/ast2600/u-boot-spl.lds new file +94 -0
@@ -0,0 +1,94 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/* + * Copyright (c) 2004-2008 Texas Instruments + * + * (C) Copyright 2002 + * Gary Jennejohn, DENX Software Engineering, <[email protected]> + * + * (C) Copyright 2022 + * Chia-Wei Wang <[email protected]> + */ + +MEMORY { .nor : ORIGIN = CONFIG_SPL_TEXT_BASE, + LENGTH = CONFIG_SPL_SIZE_LIMIT } +MEMORY { .bss : ORIGIN = CONFIG_SPL_BSS_START_ADDR, + LENGTH = CONFIG_SPL_BSS_MAX_SIZE } + +OUTPUT_FORMAT("elf32-littlearm", "elf32-littlearm", "elf32-littlearm") +OUTPUT_ARCH(arm) +ENTRY(_start) +SECTIONS +{ + . = 0x00000000; + + . = ALIGN(4); + .text : + { + __image_copy_start = .; + *(.vectors) + CPUDIR/start.o (.text*) + *(.text*) + *(.glue*) + } > .nor + + . = ALIGN(4); + .rodata : { *(SORT_BY_ALIGNMENT(SORT_BY_NAME(.rodata*))) } > .nor + + . = ALIGN(4); + .data : { + *(.data*) + } > .nor + + . = ALIGN(4); + .u_boot_list : { + KEEP(*(SORT(.u_boot_list*))); + } > .nor + + . = ALIGN(4); + .binman_sym_table : { + __binman_sym_start = .; + KEEP(*(SORT(.binman_sym*))); + __binman_sym_end = .; + } > .nor + + . = ALIGN(4); + + __image_copy_end = .; + + .rel.dyn : { + __rel_dyn_start = .; + *(.rel*) + __rel_dyn_end = .; + } > .nor + + .end : + { + *(.__end) + } > .nor + + _image_binary_end = .; + + .bss : { + __bss_start = .; + *(.bss*) + . = ALIGN(4); + __bss_end = .; + } > .bss + + __bss_size = __bss_end - __bss_start; +} + +#if defined(IMAGE_MAX_SIZE) +ASSERT(__image_copy_end - __image_copy_start <= (IMAGE_MAX_SIZE), \ + "SPL image too big"); +#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
configs/evb-ast2600_defconfig +3 -0
@@ -1,6 +1,7 @@ CONFIG_ARM=y CONFIG_SYS_DCACHE_OFF=y CONFIG_SPL_SYS_THUMB_BUILD=y +CONFIG_SPL_LDSCRIPT="arch/arm/mach-aspeed/ast2600/u-boot-spl.lds" CONFIG_ARCH_ASPEED=y CONFIG_SYS_TEXT_BASE=0x80000000 CONFIG_SYS_MALLOC_LEN=0x2000000
@@ -35,6 +36,8 @@ CONFIG_SPL_SIZE_LIMIT_SUBTRACT_MALLOC=y CONFIG_SPL_SYS_MALLOC_SIMPLE=y CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x2000000 +CONFIG_SPL_SEPARATE_BSS=y +# CONFIG_TPL_SEPARATE_BSS is not set CONFIG_SPL_FIT_IMAGE_TINY=y CONFIG_SPL_DM_RESET=y CONFIG_SPL_RAM_SUPPORT=y
include/configs/evb_ast2600.h +3 -0
@@ -10,6 +10,9 @@#define CONFIG_SYS_UBOOT_BASE CONFIG_SYS_TEXT_BASE +#define CONFIG_SPL_BSS_START_ADDR 0x83000000 +#define CONFIG_SPL_BSS_MAX_SIZE 0x01000000 +/* Misc */ #define STR_HELPER (s) #s#define STR (s) STR_HELPER(s)
Signature · 1 line
2.25.1
Quoted 5 lines
The commit
b583348ca8c8("image: fit: Align hash output buffers") places the hash output buffer at the .bss section. However, AST2600 by default executes SPL in the NOR flash XIP way. This results in the hash output cannot be written to the buffer as it is located at the R/X only region.
Quoted 7 lines
We need to move the .bss section out of the SPL body to the DRAM space, where hash output can be written to. This patch includes:
- Define the .bss section base and size
- A new SPL linker script is added with a separate .bss region specified
- EnableCONFIG_SPL_SEPARATE_BSSkconfig optionSigned-off-by Chia-Wei Wang <[email protected]>
LGTM.
Best Regards,
-Neal
Quoted 7 lines
--- arch/arm/mach-aspeed/ast2600/u-boot-spl.lds | 94 +++++++++++++++++++++ configs/evb-ast2600_defconfig | 3 + include/configs/evb_ast2600.h | 3 + 3 files changed, 100 insertions(+) create mode 100644 arch/arm/mach-aspeed/ast2600/u-boot-spl.lds
Quoted 5 lines · Chia-Wei Wang
The commit
b583348ca8c8("image: fit: Align hash output buffers") places the hash output buffer at the .bss section. However, AST2600 by default executes SPL in the NOR flash XIP way. This results in the hash output cannot be written to the buffer as it is located at the R/X only region.
Quoted 8 lines
We need to move the .bss section out of the SPL body to the DRAM space, where hash output can be written to. This patch includes:
- Define the .bss section base and size
- A new SPL linker script is added with a separate .bss region specified
- EnableCONFIG_SPL_SEPARATE_BSSkconfig optionSigned-off-by Chia-Wei Wang <[email protected]>Reviewed-by Neal Liu <[email protected]>
Applied to u-boot/next, thanks!
Signature · 1 line
Tom
Joel Stanley
maintainerHi Chai Wei,
Quoted 6 lines · Chia-Wei Wang
The commit
b583348ca8c8("image: fit: Align hash output buffers") places the hash output buffer at the .bss section. However, AST2600 by default executes SPL in the NOR flash XIP way. This results in the hash output cannot be written to the buffer as it is located at the R/X only region.
Quoted 7 lines
We need to move the .bss section out of the SPL body to the DRAM space, where hash output can be written to. This patch includes:
- Define the .bss section base and size
- A new SPL linker script is added with a separate .bss region specified
- EnableCONFIG_SPL_SEPARATE_BSSkconfig optionSigned-off-by Chia-Wei Wang <[email protected]>
This patch breaks booting for me.
My concern with the approach is it creates extra maintenance work. When changes are made to the main linker script they need to be mirrored here, or else the aspeed port will miss out. (Having the machine tested in CI will help this somewhat, but only for the code paths we can test under emulation).
I know the patch has been merged, but I have a few questions:
I imagine the ast2600 is not the only board that runs XIP. How do other boards solve the problem?
What happens when a symbol that is used before DRAM training has completed is placed in bss?
How do you plan to support systems that don't have NOR?
Cheers,
Joel
Quoted 145 lines
--- arch/arm/mach-aspeed/ast2600/u-boot-spl.lds | 94 +++++++++++++++++++++ configs/evb-ast2600_defconfig | 3 + include/configs/evb_ast2600.h | 3 + 3 files changed, 100 insertions(+) create mode 100644 arch/arm/mach-aspeed/ast2600/u-boot-spl.ldsarch/arm/mach-aspeed/ast2600/u-boot-spl.lds new file +94
-0@@ -0,0 +1,94 @@ +/* SPDX-License-Identifier: GPL-2.0+ */ +/* + * Copyright (c) 2004-2008 Texas Instruments + * + * (C) Copyright 2002 + * Gary Jennejohn, DENX Software Engineering, <[email protected]> + * + * (C) Copyright 2022 + * Chia-Wei Wang <[email protected]> + */ + +MEMORY { .nor : ORIGIN = CONFIG_SPL_TEXT_BASE, + LENGTH = CONFIG_SPL_SIZE_LIMIT } +MEMORY { .bss : ORIGIN = CONFIG_SPL_BSS_START_ADDR, + LENGTH = CONFIG_SPL_BSS_MAX_SIZE } + +OUTPUT_FORMAT("elf32-littlearm", "elf32-littlearm", "elf32-littlearm") +OUTPUT_ARCH(arm) +ENTRY(_start) +SECTIONS +{ + . = 0x00000000; + + . = ALIGN(4); + .text : + { + __image_copy_start = .; + *(.vectors) + CPUDIR/start.o (.text*) + *(.text*) + *(.glue*) + } > .nor + + . = ALIGN(4); + .rodata : { *(SORT_BY_ALIGNMENT(SORT_BY_NAME(.rodata*))) } > .nor + + . = ALIGN(4); + .data : { + *(.data*) + } > .nor + + . = ALIGN(4); + .u_boot_list : { + KEEP(*(SORT(.u_boot_list*))); + } > .nor + + . = ALIGN(4); + .binman_sym_table : { + __binman_sym_start = .; + KEEP(*(SORT(.binman_sym*))); + __binman_sym_end = .; + } > .nor + + . = ALIGN(4); + + __image_copy_end = .; + + .rel.dyn : { + __rel_dyn_start = .; + *(.rel*) + __rel_dyn_end = .; + } > .nor + + .end : + { + *(.__end) + } > .nor + + _image_binary_end = .; + + .bss : { + __bss_start = .; + *(.bss*) + . = ALIGN(4); + __bss_end = .; + } > .bss + + __bss_size = __bss_end - __bss_start; +} + +#if defined(IMAGE_MAX_SIZE) +ASSERT(__image_copy_end - __image_copy_start <= (IMAGE_MAX_SIZE), \ + "SPL image too big"); +#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"); +#endifconfigs/evb-ast2600_defconfig +3
-0@@ -1,6 +1,7 @@ CONFIG_ARM=y CONFIG_SYS_DCACHE_OFF=y CONFIG_SPL_SYS_THUMB_BUILD=y +CONFIG_SPL_LDSCRIPT="arch/arm/mach-aspeed/ast2600/u-boot-spl.lds" CONFIG_ARCH_ASPEED=y CONFIG_SYS_TEXT_BASE=0x80000000 CONFIG_SYS_MALLOC_LEN=0x2000000@@ -35,6 +36,8 @@ CONFIG_SPL_SIZE_LIMIT_SUBTRACT_MALLOC=y CONFIG_SPL_SYS_MALLOC_SIMPLE=y CONFIG_SPL_STACK_R=y CONFIG_SPL_STACK_R_MALLOC_SIMPLE_LEN=0x2000000 +CONFIG_SPL_SEPARATE_BSS=y +# CONFIG_TPL_SEPARATE_BSS is not set CONFIG_SPL_FIT_IMAGE_TINY=y CONFIG_SPL_DM_RESET=y CONFIG_SPL_RAM_SUPPORT=yinclude/configs/evb_ast2600.h +3
-1@@ -10,6 +10,9 @@#define CONFIG_SYS_UBOOT_BASE CONFIG_SYS_TEXT_BASE +#define CONFIG_SPL_BSS_START_ADDR 0x83000000 +#define CONFIG_SPL_BSS_MAX_SIZE 0x01000000 +/* Misc */ #define STR_HELPER (s) #s#define STR (s) STR_HELPER(s) --2.25.1
Hi Chai,
Quoted 15 lines · Joel Stanley
Hi Chai Wei,
Quoted 12 lines · Chia-Wei Wang
The commit
b583348ca8c8("image: fit: Align hash output buffers") places the hash output buffer at the .bss section. However, AST2600 by default executes SPL in the NOR flash XIP way. This results in the hash output cannot be written to the buffer as it is located at the R/X only region.We need to move the .bss section out of the SPL body to the DRAM space, where hash output can be written to. This patch includes:
- Define the .bss section base and size
- A new SPL linker script is added with a separate .bss region specified
- EnableCONFIG_SPL_SEPARATE_BSSkconfig option
Quoted 3 lines
Quoted 1 line
Signed-off-by Chia-Wei Wang <[email protected]>This patch breaks booting for me.
Does the patch Joel posted [1] fix your issue? It seems like I used the wrong macro in the first place, so hopefully this patch shouldn't be necessary.
--Sean
Hi Sean,
Quoted 25 lines
From: Sean Anderson <[email protected]>
Sent: Tuesday, June 28, 2022 12:57 PMHi Chai,
Quoted 3 lines · Joel Stanley
Hi Chai Wei,
On Wed, 1 Jun 2022 at 08:21, Chia-Wei Wang
Quoted 5 lines · <[email protected]> wrote:
Quoted 5 lines
The commit
b583348ca8c8("image: fit: Align hash output buffers") places the hash output buffer at the .bss section. However, AST2600 by default executes SPL in the NOR flash XIP way. This results in the hash output cannot be written to the buffer as it is located at the R/X onlyregion.
Quoted 9 lines
Quoted 8 lines
We need to move the .bss section out of the SPL body to the DRAM space, where hash output can be written to. This patch includes:
- Define the .bss section base and size
- A new SPL linker script is added with a separate .bss region specified
- EnableCONFIG_SPL_SEPARATE_BSSkconfig optionSigned-off-by Chia-Wei Wang <[email protected]>
Quoted 8 lines
Quoted 1 line
This patch breaks booting for me.
Does the patch Joel posted [1] fix your issue? It seems like I used the wrong macro in the first place, so hopefully this patch shouldn't be necessary.
--Sean
Yes. Joel's patch also solved that issue.
Relocating .bss to DRAM space can avoid similar issues. But it do create additional maintenance work.
Chiawei
Hi Joel,
Quoted 23 lines
From: Joel Stanley <[email protected]>
Sent: Tuesday, June 28, 2022 12:24 PMHi Chai Wei,
Quoted 5 lines · Chia-Wei Wang
The commit
b583348ca8c8("image: fit: Align hash output buffers") places the hash output buffer at the .bss section. However, AST2600 by default executes SPL in the NOR flash XIP way. This results in the hash output cannot be written to the buffer as it is located at the R/X onlyregion.
Quoted 9 lines
We need to move the .bss section out of the SPL body to the DRAM space, where hash output can be written to. This patch includes:
- Define the .bss section base and size
- A new SPL linker script is added with a separate .bss region
specified
- EnableCONFIG_SPL_SEPARATE_BSSkconfig optionSigned-off-by Chia-Wei Wang <[email protected]>
Quoted 7 lines
This patch breaks booting for me.
My concern with the approach is it creates extra maintenance work. When changes are made to the main linker script they need to be mirrored here, or else the aspeed port will miss out. (Having the machine tested in CI will help this somewhat, but only for the code paths we can test under emulation).
The patch was trying to solve the hash buffer allocation change to common code and to avoid similar issues. But I agree there is additional maintenance work on the customized linker script.
Quoted 8 lines
I know the patch has been merged, but I have a few questions:
I imagine the ast2600 is not the only board that runs XIP. How do other boards solve the problem?
What happens when a symbol that is used before DRAM training has completed is placed in bss?
Honestly, I am not sure how other platforms/boards dealing with this.
But like U-Boot REAME stated, the initial global data is read-only.
I guess we have to be careful about the global variable use before DRAM initialization or even variable relocation.
Quoted 2 lines
How do you plan to support systems that don't have NOR?
We would like to have another defconfig for eMMC booting.
If the NOR-based linker script is not applicable, the default one can be used.
Chiawei