Fix binman_sym functionality on RISC-V port
cover Fix binman_sym functionality on RISC-V port
It's found that BINMAN_SYMBOLS_OK always evaluates to false on RISC-V, because our linkerscripts don't define symbol __image_copy_start, on which binman depends for determining the base address of an entry. Binman simply bails out in case of missing the symbol.
This series first defaults SPL_BINMAN_UBOOT_SYMBOLS to N on RISC-V to prevent binman from looking for a plain proper U-Boot image, which isn't desired since binman configuration on RISC-V wraps proper U-Boot in a FIT image. Then we define __image_copy_start and the paired __image_copy_end to really fix binman_sym's functionality.
I've tested building with
- ae350_rv64_defconfig
- ae350_rv64_spl_defconfig
- bananapi-f3_defconfig
- k230_canmv_defconfig
- microchip_mpfs_icicle_defconfig
and booting successfully on
- milkv_duo_defconfig
- qemu-riscv64_smode_defconfig
- qemu-riscv64_defconfig
- sifive_unleashed_defconfig
- starfive_visionfive2_defconfig
with either real hardware or QEMU.
Thanks for your time and review.
Yao Zi (2):
spl: riscv: Disable SPL_BINMAN_UBOOT_SYMBOLS by default
riscv: Provide __image_copy_{start_end} symbols in linkerscript
arch/riscv/cpu/u-boot-spl.lds | 2 ++ arch/riscv/cpu/u-boot.lds | 3 +++ common/spl/Kconfig | 2 ++ 3 files changed, 7 insertions(+)
Signature · 1 line
2.49.0
PATCH 1/2 spl: riscv: Disable SPL_BINMAN_UBOOT_SYMBOLS by default
The default binman configuration of RISC-V wraps proper U-Boot into a FIT image instead of shipping a plain image, thus there's no "u_boot_any" entry by default. Let's disable the option to prevent binman from looking for a plain proper U-Boot image, failing the build with message like
Section '/binman/spl-img': Symbol '_binman_u_boot_any_prop_size'
in entry '/binman/spl-img/mkimage/u-boot-spl/u-boot-spl-nodtb':
Entry 'u-boot-any' not found in list (u-boot-spl-nodtb, u-boot-spl-dtb,u-boot-spl,mkimage,spl-img)
--- common/spl/Kconfig | 2 ++ 1 file changed, 2 insertions(+)
common/spl/Kconfig +2 -0
@@ -214,6 +214,8 @@ config SPL_BINMAN_UBOOT_SYMBOLS bool "Declare binman symbols for U-Boot phases in SPL" depends on SPL_BINMAN_SYMBOLS default n if ARCH_IMX8M || ARCH_IMX8ULP || ARCH_IMX9 + # A FIT image is created with default binman configuration of RISC-V + default n if RISCV default y help This enables use of symbols in SPL which refer to U-Boot phases,
Signature · 1 line
2.49.0
Re: [PATCH 1/2] spl: riscv: Disable SPL_BINMAN_UBOOT_SYMBOLS by default
Hi Yao,
Quoted 11 lines · Yao Zi
The default binman configuration of RISC-V wraps proper U-Boot into a FIT image instead of shipping a plain image, thus there's no "
u_boot_any" entry by default. Let's disable the option to prevent binman from looking for a plain proper U-Boot image, failing the build with message likeSection '/binman/spl-img': Symbol '
_binman_u_boot_any_prop_size'
in entry '/binman/spl-img/mkimage/u-boot-spl/u-boot-spl-nodtb':
Entry 'u-boot-any' not found in list (u-boot-spl-nodtb, u-boot-spl-dtb,u-boot-spl,mkimage,spl-img)
Do you know where the symbol is being referenced? This sounds like a bug to me.
It is likely to be spl_get_image_pos(), perhaps called from spl_set_header_raw_uboot().
If you are using FIT that function should not be called.
Quoted 20 lines
Signed-off-by Yao Zi <[email protected]>--- common/spl/Kconfig | 2 ++ 1 file changed, 2 insertions(+)common/spl/Kconfig +2
-1@@ -214,6 +214,8 @@ config SPL_BINMAN_UBOOT_SYMBOLS bool "Declare binman symbols for U-Boot phases in SPL" depends on SPL_BINMAN_SYMBOLS default n if ARCH_IMX8M || ARCH_IMX8ULP || ARCH_IMX9 + # A FIT image is created with default binman configuration of RISC-V + default n if RISCV default y help This enables use of symbols in SPL which refer to U-Boot phases, --
Quoted 2 lines
2.49.0
Regards,
Simon
Re: [PATCH 1/2] spl: riscv: Disable SPL_BINMAN_UBOOT_SYMBOLS by default
Quoted 15 lines · Simon Glass
Hi Yao,
Quoted 11 lines · Yao Zi
The default binman configuration of RISC-V wraps proper U-Boot into a FIT image instead of shipping a plain image, thus there's no "
u_boot_any" entry by default. Let's disable the option to prevent binman from looking for a plain proper U-Boot image, failing the build with message likeSection '/binman/spl-img': Symbol '
_binman_u_boot_any_prop_size'
in entry '/binman/spl-img/mkimage/u-boot-spl/u-boot-spl-nodtb':
Entry 'u-boot-any' not found in list (u-boot-spl-nodtb, u-boot-spl-dtb,u-boot-spl,mkimage,spl-img)
Quoted 4 lines
Do you know where the symbol is being referenced? This sounds like a bug to me.
It is likely to be
spl_get_image_pos(), perhaps called fromspl_set_header_raw_uboot().
The symbol is referenced in common/spl/spl.c:54,
#if CONFIG_IS_ENABLED(BINMAN_UBOOT_SYMBOLS) /* See spl.h for information about this */ #if defined(CONFIG_SPL_BUILD) binman_sym_declare(ulong, u_boot_any, image_pos); binman_sym_declare(ulong, u_boot_any, size); #endif
Quoted 1 line
If you are using FIT that function should not be called.
The place where binman_sym() is used isn't where an entry in .binman_sym section is introduced. Instead, binman_sym_declare() actually defines the entry.
So whether spl_get_image_pos() is invoked doesn't affect the failure, since in both cases binman_sym_declare() puts an entry into .binman_sym, confusing binman.
imho SPL_BINMAN_UBOOT_SYMBOLS should depend on !SPL_LOAD_FIT in an ideal world, but I'm not sure the impact on current codebase thus limited the change to RISC-V only.
Best regards,
Yao Zi
Quoted 25 lines
Quoted 20 lines
Signed-off-by Yao Zi <[email protected]>--- common/spl/Kconfig | 2 ++ 1 file changed, 2 insertions(+)common/spl/Kconfig +2
-1@@ -214,6 +214,8 @@ config SPL_BINMAN_UBOOT_SYMBOLS bool "Declare binman symbols for U-Boot phases in SPL" depends on SPL_BINMAN_SYMBOLS default n if ARCH_IMX8M || ARCH_IMX8ULP || ARCH_IMX9 + # A FIT image is created with default binman configuration of RISC-V + default n if RISCV default y help This enables use of symbols in SPL which refer to U-Boot phases, --Quoted 2 lines
2.49.0
Regards,
Simon
Re: [PATCH 1/2] spl: riscv: Disable SPL_BINMAN_UBOOT_SYMBOLS by default
In reply to Yao ZiHi,
Quoted 21 lines · Yao Zi
The default binman configuration of RISC-V wraps proper U-Boot into a FIT image instead of shipping a plain image, thus there's no "
u_boot_any" entry by default. Let's disable the option to prevent binman from looking for a plain proper U-Boot image, failing the build with message likeSection '/binman/spl-img': Symbol '
_binman_u_boot_any_prop_size'
in entry '/binman/spl-img/mkimage/u-boot-spl/u-boot-spl-nodtb':
Entry 'u-boot-any' not found in list (u-boot-spl-nodtb, u-boot-spl-dtb,u-boot-spl,mkimage,spl-img)Signed-off-by Yao Zi <[email protected]>--- common/spl/Kconfig | 2 ++ 1 file changed, 2 insertions(+)common/spl/Kconfig +0
-0
common/spl/Kconfig config SPL_BINMAN_UBOOT_SYMBOLS · line 217
Quoted 4 lines
common/spl/Kconfig +1
-0@@ -214,6 +214,8 @@ config SPL_BINMAN_UBOOT_SYMBOLS bool "Declare binman symbols for U-Boot phases in SPL" depends on SPL_BINMAN_SYMBOLS default n if ARCH_IMX8M || ARCH_IMX8ULP || ARCH_IMX9 + # A FIT image is created with default binman configuration of RISC-V
Use of a FIT does not really matter, Rockchip use a FIT for U-Boot proper and symbols can be located.
common/spl/Kconfig config SPL_BINMAN_UBOOT_SYMBOLS · line 221
Quoted 4 lines
common/spl/Kconfig +1
-0+ default n if RISCV default y help This enables use of symbols in SPL which refer to U-Boot phases,
Maybe something like this is what you instead want to use:
arch/riscv/dts/binman.dtsi +3 -4
@@ -30,20 +30,19 @@ uboot { description = "U-Boot"; type = "standalone"; - os = "U-Boot"; + os = "u-boot"; arch = "riscv"; compression = "none"; load = /bits/ 64 <CONFIG_TEXT_BASE>; - uboot_blob: blob-ext { - filename = "u-boot-nodtb.bin"; + uboot_blob: u-boot-nodtb { }; }; #else linux { description = "Linux"; type = "standalone"; - os = "Linux"; + os = "linux"; arch = "riscv"; compression = "none"; load = /bits/ 64 <CONFIG_TEXT_BASE>;
arch/riscv/dts/starfive-visionfive2-binman.dtsi +1 -0
@@ -20,6 +20,7 @@ args = "-T sfspl"; u-boot-spl { + no-write-symbols; }; }; };
Regards,
Jonas
Re: [PATCH 1/2] spl: riscv: Disable SPL_BINMAN_UBOOT_SYMBOLS by default
Quoted 19 lines · Jonas Karlman
Hi,
Quoted 16 lines · Yao Zi
The default binman configuration of RISC-V wraps proper U-Boot into a FIT image instead of shipping a plain image, thus there's no "
u_boot_any" entry by default. Let's disable the option to prevent binman from looking for a plain proper U-Boot image, failing the build with message likeSection '/binman/spl-img': Symbol '
_binman_u_boot_any_prop_size'
in entry '/binman/spl-img/mkimage/u-boot-spl/u-boot-spl-nodtb':
Entry 'u-boot-any' not found in list (u-boot-spl-nodtb, u-boot-spl-dtb,u-boot-spl,mkimage,spl-img)Signed-off-by Yao Zi <[email protected]>--- common/spl/Kconfig | 2 ++ 1 file changed, 2 insertions(+)
Quoted 12 lines
common/spl/Kconfig config SPL_BINMAN_UBOOT_SYMBOLS · line 217
Quoted 9 lines
common/spl/Kconfig +1
-0@@ -214,6 +214,8 @@ config SPL_BINMAN_UBOOT_SYMBOLS bool "Declare binman symbols for U-Boot phases in SPL" depends on SPL_BINMAN_SYMBOLS default n if ARCH_IMX8M || ARCH_IMX8ULP || ARCH_IMX9 + # A FIT image is created with default binman configuration of RISC-VUse of a FIT does not really matter, Rockchip use a FIT for U-Boot proper and symbols can be located.
You're right. binman is able to match a U-Boot blob in FIT image as long as its entry name is correct. We should replace blob-ext with u-boot-nodtb here.
Quoted 15 lines
common/spl/Kconfig config SPL_BINMAN_UBOOT_SYMBOLS · line 221
Quoted 5 lines
common/spl/Kconfig +1
-0+ default n if RISCV default y help This enables use of symbols in SPL which refer to U-Boot phases,Maybe something like this is what you instead want to use:
arch/riscv/dts/binman.dtsi +0
-0@@ -30,20 +30,19 @@ uboot { description = "U-Boot";
arch/riscv/dts/binman.dtsi line 38
Quoted 10 lines
arch/riscv/dts/binman.dtsi +2
-3type = "standalone"; - os = "U-Boot"; + os = "u-boot"; arch = "riscv"; compression = "none"; load = /bits/ 64 <CONFIG_TEXT_BASE>; - uboot_blob: blob-ext { - filename = "u-boot-nodtb.bin"; + uboot_blob: u-boot-nodtb {
I think the change to uboot_blob node is enough to fix the binman issue, right?
The value of property "os" is probably wrong since spl_fit.c compares it to "u-boot" case-sensitively to look for a U-Boot image. But this should go in another patch IMHO.
Quoted 18 lines
arch/riscv/dts/binman.dtsi +1
-1}; }; #else linux { description = "Linux"; type = "standalone"; - os = "Linux"; + os = "linux"; arch = "riscv"; compression = "none"; load = /bits/ 64 <CONFIG_TEXT_BASE>;arch/riscv/dts/starfive-visionfive2-binman.dtsi +0
-0
arch/riscv/dts/starfive-visionfive2-binman.dtsi line 26
Quoted 7 lines
arch/riscv/dts/starfive-visionfive2-binman.dtsi +1
-0@@ -20,6 +20,7 @@ args = "-T sfspl"; u-boot-spl { + no-write-symbols; }; }; };
At the very first sight I wrongly thought "no-write-symbols" would prevent binman to function and should be avoided. But visionfive2's case seems actually special as its spl and proper U-Boot are split into different images, where binman cannot help much.
I've applied the fix of uboot_blob node to my WIP branch of TH1520 SPL support and verified it does fix the binman error. Thanks for your hint and review! I'll include the fix and drop this patch in v2.
Quoted 2 lines
Regards,
Jonas
Thanks,
Yao Zi
PATCH 2/2 riscv: Provide __image_copy_{start_end} symbols in linkerscript
In reply to Yao ZiBinman looks for __image_copy_start to determine the base address of an entry if elf-base-sym isn't specified, which is missing in RISC-V port. This causes binman skips RISC-V SPL entries without filling addresses into its .binman_sym_table section.
This patch defines __image_copy_start in linkerscript of both SPL and proper U-Boot to ensure binman_sym functions correctly with the default binman.dtsi. The paired symbol, __image_copy_end, is introduced as well for completeness.
--- arch/riscv/cpu/u-boot-spl.lds | 2 ++ arch/riscv/cpu/u-boot.lds | 3 +++ 2 files changed, 5 insertions(+)
arch/riscv/cpu/u-boot-spl.lds +2 -0
@@ -16,6 +16,7 @@ ENTRY(_start) SECTIONS { . = ALIGN(4); + __image_copy_start = ADDR(.text); .text : { arch/riscv/cpu/start.o (.text) *(.text*)
@@ -46,6 +47,7 @@ SECTIONS _end = .; _image_binary_end = .; + _image_copy_end = .; .bss : { __bss_start = .;
arch/riscv/cpu/u-boot.lds +3 -0
@@ -10,6 +10,7 @@ ENTRY(_start) SECTIONS { . = ALIGN(4); + __image_copy_start = ADDR(.text); .text : { arch/riscv/cpu/start.o (.text) }
@@ -57,6 +58,8 @@ SECTIONS __efi_runtime_rel_stop = .; } + __image_copy_end = .; + /DISCARD/ : { *(.rela.plt*) } .rela.dyn : { __rel_dyn_start = .;
Signature · 1 line
2.49.0
Re: [PATCH 2/2] riscv: Provide __image_copy_{start_end} symbols in linkerscript
+Heinrich Schuchardt
Quoted 11 lines · Yao Zi
Binman looks for
__image_copy_startto determine the base address of an entry if elf-base-sym isn't specified, which is missing in RISC-V port. This causes binman skips RISC-V SPL entries without filling addresses into its .binman_sym_table section.This patch defines
__image_copy_startin linkerscript of both SPL and proper U-Boot to ensurebinman_symfunctions correctly with the defaultbinman.dtsi. The paired symbol,__image_copy_end, is introduced as well for completeness.
Quoted 5 lines
Signed-off-by Yao Zi <[email protected]>--- arch/riscv/cpu/u-boot-spl.lds | 2 ++ arch/riscv/cpu/u-boot.lds | 3 +++ 2 files changed, 5 insertions(+)
Quoted 45 lines
arch/riscv/cpu/u-boot-spl.lds +2
-0@@ -16,6 +16,7 @@ ENTRY(_start) SECTIONS { . = ALIGN(4); + __image_copy_start = ADDR(.text); .text : { arch/riscv/cpu/start.o (.text) *(.text*)@@ -46,6 +47,7 @@ SECTIONS _end = .; _image_binary_end = .; + _image_copy_end = .; .bss : { __bss_start = .;arch/riscv/cpu/u-boot.lds +3
-1@@ -10,6 +10,7 @@ ENTRY(_start) SECTIONS { . = ALIGN(4); + __image_copy_start = ADDR(.text); .text : { arch/riscv/cpu/start.o (.text) }@@ -57,6 +58,8 @@ SECTIONS __efi_runtime_rel_stop = .; } + __image_copy_end = .; + /DISCARD/ : { *(.rela.plt*) } .rela.dyn : { __rel_dyn_start = .; --2.49.0
Re: [PATCH 2/2] riscv: Provide __image_copy_{start_end} symbols in linkerscript
In reply to Yao ZiHi,
Quoted 29 lines · Yao Zi
Binman looks for
__image_copy_startto determine the base address of an entry if elf-base-sym isn't specified, which is missing in RISC-V port. This causes binman skips RISC-V SPL entries without filling addresses into its .binman_sym_table section.This patch defines
__image_copy_startin linkerscript of both SPL and proper U-Boot to ensurebinman_symfunctions correctly with the defaultbinman.dtsi. The paired symbol,__image_copy_end, is introduced as well for completeness.Signed-off-by Yao Zi <[email protected]>--- arch/riscv/cpu/u-boot-spl.lds | 2 ++ arch/riscv/cpu/u-boot.lds | 3 +++ 2 files changed, 5 insertions(+)arch/riscv/cpu/u-boot-spl.lds +1
-0@@ -16,6 +16,7 @@ ENTRY(_start) SECTIONS { . = ALIGN(4); + __image_copy_start = ADDR(.text); .text : { arch/riscv/cpu/start.o (.text) *(.text*)
arch/riscv/cpu/u-boot-spl.lds SECTIONS · line 50
Quoted 4 lines
arch/riscv/cpu/u-boot-spl.lds +1
-0@@ -46,6 +47,7 @@ SECTIONS _end = .; _image_binary_end = .; + _image_copy_end = .;
This looks like a typo, should probably be with two _ ?
Regards,
Jonas
Quoted 24 lines
arch/riscv/cpu/u-boot-spl.lds +0
-0.bss : { __bss_start = .;arch/riscv/cpu/u-boot.lds +3
-0@@ -10,6 +10,7 @@ ENTRY(_start) SECTIONS { . = ALIGN(4); + __image_copy_start = ADDR(.text); .text : { arch/riscv/cpu/start.o (.text) }@@ -57,6 +58,8 @@ SECTIONS __efi_runtime_rel_stop = .; } + __image_copy_end = .; + /DISCARD/ : { *(.rela.plt*) } .rela.dyn : { __rel_dyn_start = .;
Re: [PATCH 2/2] riscv: Provide __image_copy_{start_end} symbols in linkerscript
Quoted 33 lines · Jonas Karlman
Hi,
Quoted 30 lines · Yao Zi
Binman looks for
__image_copy_startto determine the base address of an entry if elf-base-sym isn't specified, which is missing in RISC-V port. This causes binman skips RISC-V SPL entries without filling addresses into its .binman_sym_table section.This patch defines
__image_copy_startin linkerscript of both SPL and proper U-Boot to ensurebinman_symfunctions correctly with the defaultbinman.dtsi. The paired symbol,__image_copy_end, is introduced as well for completeness.Signed-off-by Yao Zi <[email protected]>--- arch/riscv/cpu/u-boot-spl.lds | 2 ++ arch/riscv/cpu/u-boot.lds | 3 +++ 2 files changed, 5 insertions(+)arch/riscv/cpu/u-boot-spl.lds +1
-0@@ -16,6 +16,7 @@ ENTRY(_start) SECTIONS { . = ALIGN(4); + __image_copy_start = ADDR(.text); .text : { arch/riscv/cpu/start.o (.text) *(.text*)@@ -46,6 +47,7 @@ SECTIONS
Quoted 5 lines
arch/riscv/cpu/u-boot-spl.lds SECTIONS · line 50
Quoted 3 lines
arch/riscv/cpu/u-boot-spl.lds +1
-0_end = .; _image_binary_end = .; + _image_copy_end = .;This looks like a typo, should probably be with two _ ?
Yes, thanks for catching this. I'll correct it and apply Simon's tag in v2. Simon, is it okay for you?
Thanks,
Yao Zi
Quoted 28 lines
Regards,
JonasQuoted 24 lines
arch/riscv/cpu/u-boot-spl.lds +0
-0.bss : { __bss_start = .;arch/riscv/cpu/u-boot.lds +3
-0@@ -10,6 +10,7 @@ ENTRY(_start) SECTIONS { . = ALIGN(4); + __image_copy_start = ADDR(.text); .text : { arch/riscv/cpu/start.o (.text) }@@ -57,6 +58,8 @@ SECTIONS __efi_runtime_rel_stop = .; } + __image_copy_end = .; + /DISCARD/ : { *(.rela.plt*) } .rela.dyn : { __rel_dyn_start = .;
Re: [PATCH 2/2] riscv: Provide __image_copy_{start_end} symbols in linkerscript
Hi Yao,
Quoted 35 lines · Yao Zi
Quoted 33 lines · Jonas Karlman
Hi,
Quoted 30 lines · Yao Zi
Binman looks for
__image_copy_startto determine the base address of an entry if elf-base-sym isn't specified, which is missing in RISC-V port. This causes binman skips RISC-V SPL entries without filling addresses into its .binman_sym_table section.This patch defines
__image_copy_startin linkerscript of both SPL and proper U-Boot to ensurebinman_symfunctions correctly with the defaultbinman.dtsi. The paired symbol,__image_copy_end, is introduced as well for completeness.Signed-off-by Yao Zi <[email protected]>--- arch/riscv/cpu/u-boot-spl.lds | 2 ++ arch/riscv/cpu/u-boot.lds | 3 +++ 2 files changed, 5 insertions(+)arch/riscv/cpu/u-boot-spl.lds +1
-0@@ -16,6 +16,7 @@ ENTRY(_start) SECTIONS { . = ALIGN(4); + __image_copy_start = ADDR(.text); .text : { arch/riscv/cpu/start.o (.text) *(.text*)@@ -46,6 +47,7 @@ SECTIONS
Quoted 8 lines
Quoted 5 lines
arch/riscv/cpu/u-boot-spl.lds SECTIONS · line 50
Quoted 3 lines
arch/riscv/cpu/u-boot-spl.lds +1
-0_end = .; _image_binary_end = .; + _image_copy_end = .;This looks like a typo, should probably be with two _ ?
Yes, thanks for catching this. I'll correct it and apply Simon's tag in v2. Simon, is it okay for you?
Yes.
Regards,
Simon