[PATCH v2 0/4] mkimage: sunxi_egon: add riscv support
This patchset tries to make mkimage -T sunxi_egon to be able to generate an eGON.BT0 image for Allwinner RISC-V SoCs (e.g. D1).
In addition, to keep the compatibility, it will still consider the architecture to be ARM when no architecture is specified.
Icenowy Zheng (4): mkimage: add a flag to describe whether -A is specified mkimage: sunxi_egon: refactor for multi-architecture support mkimage: sunxi_egon: add support for riscv sunxi: specify architecture when generating SPL boot image
scripts/Makefile.spl | 2 +- tools/imagetool.h | 1 + tools/mkimage.c | 1 + tools/sunxi_egon.c | 86 ++++++++++++++++++++++++++++++++++++++++---- 4 files changed, 82 insertions(+), 8 deletions(-)
The sunxi_egon type used to take no -A argument (because we assume sunxi targets are all ARM). However, as Allwinner D1 appears as the first RISC-V sunxi target, we need to support -A; in addition, as external projects rely on U-Boot mkimage to generate sunxi eGON.BT0 header, we need to keep compatibility with command line without -A.
As the default value of arch in mkimage is not proper (IH_ARCH_PPC instead of IH_ARCH_INVALID), to keep more compatibility, add an Aflag field to image parameters to describe whether an architecture is explicitly specified.
Reviewed-by: Tom Rini trini@konsulko.com Signed-off-by: Icenowy Zheng icenowy@aosc.io --- Changes in v2: - Added Tom Rini's Review tag.
tools/imagetool.h | 1 + tools/mkimage.c | 1 + 2 files changed, 2 insertions(+)
diff --git a/tools/imagetool.h b/tools/imagetool.h index e229a34ffc..5dc28312c2 100644 --- a/tools/imagetool.h +++ b/tools/imagetool.h @@ -51,6 +51,7 @@ struct image_tool_params { int pflag; int vflag; int xflag; + int Aflag; int skipcpy; int os; int arch; diff --git a/tools/mkimage.c b/tools/mkimage.c index cc7b242faf..54d8e3835a 100644 --- a/tools/mkimage.c +++ b/tools/mkimage.c @@ -168,6 +168,7 @@ static void process_args(int argc, char **argv) show_valid_options(IH_ARCH); usage("Invalid architecture"); } + params.Aflag = 1; break; case 'b': if (add_content(IH_TYPE_FLATDT, optarg)) {
Hi Icenowy,
On Sat, 19 Jun 2021 at 03:19, Icenowy Zheng icenowy@aosc.io wrote:
The sunxi_egon type used to take no -A argument (because we assume sunxi targets are all ARM). However, as Allwinner D1 appears as the first RISC-V sunxi target, we need to support -A; in addition, as external projects rely on U-Boot mkimage to generate sunxi eGON.BT0 header, we need to keep compatibility with command line without -A.
As the default value of arch in mkimage is not proper (IH_ARCH_PPC instead of IH_ARCH_INVALID), to keep more compatibility, add an Aflag field to image parameters to describe whether an architecture is explicitly specified.
Reviewed-by: Tom Rini trini@konsulko.com Signed-off-by: Icenowy Zheng icenowy@aosc.io
Changes in v2:
- Added Tom Rini's Review tag.
This is not actually a change so should not appear in the log.
Also note 'patman status' can be used to automatically collect these.
tools/imagetool.h | 1 + tools/mkimage.c | 1 + 2 files changed, 2 insertions(+)
diff --git a/tools/imagetool.h b/tools/imagetool.h index e229a34ffc..5dc28312c2 100644 --- a/tools/imagetool.h +++ b/tools/imagetool.h @@ -51,6 +51,7 @@ struct image_tool_params { int pflag; int vflag; int xflag;
int Aflag;
aflag
int skipcpy; int os; int arch;diff --git a/tools/mkimage.c b/tools/mkimage.c index cc7b242faf..54d8e3835a 100644 --- a/tools/mkimage.c +++ b/tools/mkimage.c @@ -168,6 +168,7 @@ static void process_args(int argc, char **argv) show_valid_options(IH_ARCH); usage("Invalid architecture"); }
params.Aflag = 1; break; case 'b': if (add_content(IH_TYPE_FLATDT, optarg)) {-- 2.30.2
Regards, Simon
在 2021-06-26星期六的 12:31 -0600，Simon Glass写道：
Hi Icenowy,
On Sat, 19 Jun 2021 at 03:19, Icenowy Zheng icenowy@aosc.io wrote:
The sunxi_egon type used to take no -A argument (because we assume sunxi targets are all ARM). However, as Allwinner D1 appears as the first RISC-V sunxi target, we need to support -A; in addition, as external projects rely on U-Boot mkimage to generate sunxi eGON.BT0 header, we need to keep compatibility with command line without -A.
As the default value of arch in mkimage is not proper (IH_ARCH_PPC instead of IH_ARCH_INVALID), to keep more compatibility, add an Aflag field to image parameters to describe whether an architecture is explicitly specified.
Reviewed-by: Tom Rini trini@konsulko.com Signed-off-by: Icenowy Zheng icenowy@aosc.io
Changes in v2:
- Added Tom Rini's Review tag.
This is not actually a change so should not appear in the log.
Also note 'patman status' can be used to automatically collect these.
tools/imagetool.h | 1 + tools/mkimage.c | 1 + 2 files changed, 2 insertions(+)
diff --git a/tools/imagetool.h b/tools/imagetool.h index e229a34ffc..5dc28312c2 100644 --- a/tools/imagetool.h +++ b/tools/imagetool.h @@ -51,6 +51,7 @@ struct image_tool_params { int pflag; int vflag; int xflag; + int Aflag;
aflag
I think a problem is that -a is a different flag than -A.
-a is addr, -A is architecture.
int skipcpy; int os; int arch; diff --git a/tools/mkimage.c b/tools/mkimage.c index cc7b242faf..54d8e3835a 100644 --- a/tools/mkimage.c +++ b/tools/mkimage.c @@ -168,6 +168,7 @@ static void process_args(int argc, char **argv) show_valid_options(IH_ARCH); usage("Invalid architecture"); } + params.Aflag = 1; break; case 'b': if (add_content(IH_TYPE_FLATDT, optarg)) { -- 2.30.2
Regards, Simon
Hi Icenowy,
On Sat, 26 Jun 2021 at 17:57, Icenowy Zheng icenowy@aosc.io wrote:
在 2021-06-26星期六的 12:31 -0600，Simon Glass写道：
Hi Icenowy,
On Sat, 19 Jun 2021 at 03:19, Icenowy Zheng icenowy@aosc.io wrote:
The sunxi_egon type used to take no -A argument (because we assume sunxi targets are all ARM). However, as Allwinner D1 appears as the first RISC-V sunxi target, we need to support -A; in addition, as external projects rely on U-Boot mkimage to generate sunxi eGON.BT0 header, we need to keep compatibility with command line without -A.
As the default value of arch in mkimage is not proper (IH_ARCH_PPC instead of IH_ARCH_INVALID), to keep more compatibility, add an Aflag field to image parameters to describe whether an architecture is explicitly specified.
Reviewed-by: Tom Rini trini@konsulko.com Signed-off-by: Icenowy Zheng icenowy@aosc.io
Changes in v2:
- Added Tom Rini's Review tag.
This is not actually a change so should not appear in the log.
Also note 'patman status' can be used to automatically collect these.
tools/imagetool.h | 1 + tools/mkimage.c | 1 + 2 files changed, 2 insertions(+)
diff --git a/tools/imagetool.h b/tools/imagetool.h index e229a34ffc..5dc28312c2 100644 --- a/tools/imagetool.h +++ b/tools/imagetool.h @@ -51,6 +51,7 @@ struct image_tool_params { int pflag; int vflag; int xflag;
int Aflag;aflag
I think a problem is that -a is a different flag than -A.
-a is addr, -A is architecture.
Yes OK, but the code style doesn't allow capital letters, as I understand it.
How about addr_flag ?
int skipcpy; int os; int arch;diff --git a/tools/mkimage.c b/tools/mkimage.c index cc7b242faf..54d8e3835a 100644 --- a/tools/mkimage.c +++ b/tools/mkimage.c @@ -168,6 +168,7 @@ static void process_args(int argc, char **argv) show_valid_options(IH_ARCH); usage("Invalid architecture"); }
params.Aflag = 1; break; case 'b': if (add_content(IH_TYPE_FLATDT, optarg)) {-- 2.30.2
Regards, Simon
在 2021-06-27星期日的 13:32 -0600，Simon Glass写道：
Hi Icenowy,
On Sat, 26 Jun 2021 at 17:57, Icenowy Zheng icenowy@aosc.io wrote:
在 2021-06-26星期六的 12:31 -0600，Simon Glass写道：
Hi Icenowy,
On Sat, 19 Jun 2021 at 03:19, Icenowy Zheng icenowy@aosc.io wrote:
The sunxi_egon type used to take no -A argument (because we assume sunxi targets are all ARM). However, as Allwinner D1 appears as the first RISC-V sunxi target, we need to support -A; in addition, as external projects rely on U-Boot mkimage to generate sunxi eGON.BT0 header, we need to keep compatibility with command line without -A.
As the default value of arch in mkimage is not proper (IH_ARCH_PPC instead of IH_ARCH_INVALID), to keep more compatibility, add an Aflag field to image parameters to describe whether an architecture is explicitly specified.
Reviewed-by: Tom Rini trini@konsulko.com Signed-off-by: Icenowy Zheng icenowy@aosc.io
Changes in v2:
- Added Tom Rini's Review tag.
This is not actually a change so should not appear in the log.
Also note 'patman status' can be used to automatically collect these.
tools/imagetool.h | 1 + tools/mkimage.c | 1 + 2 files changed, 2 insertions(+)
diff --git a/tools/imagetool.h b/tools/imagetool.h index e229a34ffc..5dc28312c2 100644 --- a/tools/imagetool.h +++ b/tools/imagetool.h @@ -51,6 +51,7 @@ struct image_tool_params { int pflag; int vflag; int xflag; + int Aflag;
aflag
I think a problem is that -a is a different flag than -A.
-a is addr, -A is architecture.
Yes OK, but the code style doesn't allow capital letters, as I understand it.
How about addr_flag ?
maybe we just call -a aflag, and -A arch_flag?
int skipcpy; int os; int arch; diff --git a/tools/mkimage.c b/tools/mkimage.c index cc7b242faf..54d8e3835a 100644 --- a/tools/mkimage.c +++ b/tools/mkimage.c @@ -168,6 +168,7 @@ static void process_args(int argc, char **argv) show_valid_options(IH_ARCH); usage("Invalid architecture"); } + params.Aflag = 1; break; case 'b': if (add_content(IH_TYPE_FLATDT, optarg)) { -- 2.30.2
Regards, Simon
On Mon, Jun 28, 2021 at 06:17:27AM +0800, Icenowy Zheng wrote:
在 2021-06-27星期日的 13:32 -0600，Simon Glass写道：
Hi Icenowy,
On Sat, 26 Jun 2021 at 17:57, Icenowy Zheng icenowy@aosc.io wrote:
在 2021-06-26星期六的 12:31 -0600，Simon Glass写道：
Hi Icenowy,
On Sat, 19 Jun 2021 at 03:19, Icenowy Zheng icenowy@aosc.io wrote:
The sunxi_egon type used to take no -A argument (because we assume sunxi targets are all ARM). However, as Allwinner D1 appears as the first RISC-V sunxi target, we need to support -A; in addition, as external projects rely on U-Boot mkimage to generate sunxi eGON.BT0 header, we need to keep compatibility with command line without -A.
As the default value of arch in mkimage is not proper (IH_ARCH_PPC instead of IH_ARCH_INVALID), to keep more compatibility, add an Aflag field to image parameters to describe whether an architecture is explicitly specified.
Reviewed-by: Tom Rini trini@konsulko.com Signed-off-by: Icenowy Zheng icenowy@aosc.io
Changes in v2:
- Added Tom Rini's Review tag.
This is not actually a change so should not appear in the log.
Also note 'patman status' can be used to automatically collect these.
tools/imagetool.h | 1 + tools/mkimage.c | 1 + 2 files changed, 2 insertions(+)
diff --git a/tools/imagetool.h b/tools/imagetool.h index e229a34ffc..5dc28312c2 100644 --- a/tools/imagetool.h +++ b/tools/imagetool.h @@ -51,6 +51,7 @@ struct image_tool_params { int pflag; int vflag; int xflag; + int Aflag;
aflag
I think a problem is that -a is a different flag than -A.
-a is addr, -A is architecture.
Yes OK, but the code style doesn't allow capital letters, as I understand it.
How about addr_flag ?
maybe we just call -a aflag, and -A arch_flag?
Frankly I think Aflag is the most readable option here. They're all literally ${FLAG_USE}flag variables, yes? CamelCase isn't great, but there's times where the most readable thing isn't always what we use everywhere else.
Hi Tom,
On Sun, 27 Jun 2021 at 17:43, Tom Rini trini@konsulko.com wrote:
On Mon, Jun 28, 2021 at 06:17:27AM +0800, Icenowy Zheng wrote:
在 2021-06-27星期日的 13:32 -0600，Simon Glass写道：
Hi Icenowy,
On Sat, 26 Jun 2021 at 17:57, Icenowy Zheng icenowy@aosc.io wrote:
在 2021-06-26星期六的 12:31 -0600，Simon Glass写道：
Hi Icenowy,
On Sat, 19 Jun 2021 at 03:19, Icenowy Zheng icenowy@aosc.io wrote:
The sunxi_egon type used to take no -A argument (because we assume sunxi targets are all ARM). However, as Allwinner D1 appears as the first RISC-V sunxi target, we need to support -A; in addition, as external projects rely on U-Boot mkimage to generate sunxi eGON.BT0 header, we need to keep compatibility with command line without -A.
As the default value of arch in mkimage is not proper (IH_ARCH_PPC instead of IH_ARCH_INVALID), to keep more compatibility, add an Aflag field to image parameters to describe whether an architecture is explicitly specified.
Reviewed-by: Tom Rini trini@konsulko.com Signed-off-by: Icenowy Zheng icenowy@aosc.io
Changes in v2:
- Added Tom Rini's Review tag.
This is not actually a change so should not appear in the log.
Also note 'patman status' can be used to automatically collect these.
tools/imagetool.h | 1 + tools/mkimage.c | 1 + 2 files changed, 2 insertions(+)
diff --git a/tools/imagetool.h b/tools/imagetool.h index e229a34ffc..5dc28312c2 100644 --- a/tools/imagetool.h +++ b/tools/imagetool.h @@ -51,6 +51,7 @@ struct image_tool_params { int pflag; int vflag; int xflag;
int Aflag;aflag
I think a problem is that -a is a different flag than -A.
-a is addr, -A is architecture.
Yes OK, but the code style doesn't allow capital letters, as I understand it.
How about addr_flag ?
maybe we just call -a aflag, and -A arch_flag?
Frankly I think Aflag is the most readable option here. They're all literally ${FLAG_USE}flag variables, yes? CamelCase isn't great, but there's times where the most readable thing isn't always what we use everywhere else.
I mean, sort of, but 'a' and 'A' are not that descriptive so you could make the case that they should be named for the full flag meaning/purpose.
Anyway, I see it has your review tag, so I'll leave it to you.
Regards, Simon
Hi Icenowy,
On Sun, 27 Jun 2021 at 16:17, Icenowy Zheng icenowy@aosc.io wrote:
在 2021-06-27星期日的 13:32 -0600，Simon Glass写道：
Hi Icenowy,
On Sat, 26 Jun 2021 at 17:57, Icenowy Zheng icenowy@aosc.io wrote:
在 2021-06-26星期六的 12:31 -0600，Simon Glass写道：
Hi Icenowy,
On Sat, 19 Jun 2021 at 03:19, Icenowy Zheng icenowy@aosc.io wrote:
The sunxi_egon type used to take no -A argument (because we assume sunxi targets are all ARM). However, as Allwinner D1 appears as the first RISC-V sunxi target, we need to support -A; in addition, as external projects rely on U-Boot mkimage to generate sunxi eGON.BT0 header, we need to keep compatibility with command line without -A.
As the default value of arch in mkimage is not proper (IH_ARCH_PPC instead of IH_ARCH_INVALID), to keep more compatibility, add an Aflag field to image parameters to describe whether an architecture is explicitly specified.
Reviewed-by: Tom Rini trini@konsulko.com Signed-off-by: Icenowy Zheng icenowy@aosc.io
Changes in v2:
- Added Tom Rini's Review tag.
This is not actually a change so should not appear in the log.
Also note 'patman status' can be used to automatically collect these.
tools/imagetool.h | 1 + tools/mkimage.c | 1 + 2 files changed, 2 insertions(+)
diff --git a/tools/imagetool.h b/tools/imagetool.h index e229a34ffc..5dc28312c2 100644 --- a/tools/imagetool.h +++ b/tools/imagetool.h @@ -51,6 +51,7 @@ struct image_tool_params { int pflag; int vflag; int xflag;
int Aflag;aflag
I think a problem is that -a is a different flag than -A.
-a is addr, -A is architecture.
Yes OK, but the code style doesn't allow capital letters, as I understand it.
How about addr_flag ?
maybe we just call -a aflag, and -A arch_flag?
Sure if you prefer.
Regards, Simon
Refactor some functions in mkimage sunxi_egon type, in order to prepare for adding support for more CPU architectures (e.g. RISC-V). In addition, compatibility for operation w/o specified architecture is kept, in this case the architecture is assumed as ARM.
Signed-off-by: Icenowy Zheng icenowy@aosc.io --- Changes in v2: - Merged fixes in the next patch in v1 (patch rebase issue).
tools/sunxi_egon.c | 63 ++++++++++++++++++++++++++++++++++++++++------ 1 file changed, 56 insertions(+), 7 deletions(-)
diff --git a/tools/sunxi_egon.c b/tools/sunxi_egon.c index a5299eb6a1..062c9bc151 100644 --- a/tools/sunxi_egon.c +++ b/tools/sunxi_egon.c @@ -16,7 +16,25 @@
static int egon_check_params(struct image_tool_params *params) { - /* We just need a binary image file. */ + int arch; + + /* Assume ARM when no architecture specified for compatibility */ + if (params->Aflag) + arch = params->arch; + else + arch = IH_ARCH_ARM; + + /* + * Check whether the architecture is supported. + */ + switch (arch) { + case IH_ARCH_ARM: + break; + default: + return EXIT_FAILURE; + } + + /* We need a binary image file. */ return !params->dflag; }
@@ -25,10 +43,26 @@ static int egon_verify_header(unsigned char *ptr, int image_size, { const struct boot_file_head *header = (void *)ptr; uint32_t length; + int arch;
- /* First 4 bytes must be an ARM branch instruction. */ - if ((le32_to_cpu(header->b_instruction) & 0xff000000) != 0xea000000) - return EXIT_FAILURE; + /* Assume ARM when no architecture specified for compatibility */ + if (params->Aflag) + arch = params->arch; + else + arch = IH_ARCH_ARM; + + /* + * First 4 bytes must be a branch instruction of the corresponding + * architecture. + */ + switch (arch) { + case IH_ARCH_ARM: + if ((le32_to_cpu(header->b_instruction) & 0xff000000) != 0xea000000) + return EXIT_FAILURE; + break; + default: + return EXIT_FAILURE; /* Unknown architecture */ + }
if (memcmp(header->magic, BOOT0_MAGIC, sizeof(header->magic))) return EXIT_FAILURE; @@ -76,10 +110,25 @@ static void egon_set_header(void *buf, struct stat *sbuf, int infd, uint32_t *buf32 = buf; uint32_t checksum = 0, value; int i; + int arch;
- /* Generate an ARM branch instruction to jump over the header. */ - value = 0xea000000 | (sizeof(struct boot_file_head) / 4 - 2); - header->b_instruction = cpu_to_le32(value); + /* Assume ARM when no architecture specified for compatibility */ + if (params->Aflag) + arch = params->arch; + else + arch = IH_ARCH_ARM; + + /* + * Different architectures need different first instruction to + * branch to the body. + */ + switch (arch) { + case IH_ARCH_ARM: + /* Generate an ARM branch instruction to jump over the header. */ + value = 0xea000000 | (sizeof(struct boot_file_head) / 4 - 2); + header->b_instruction = cpu_to_le32(value); + break; + }
memcpy(header->magic, BOOT0_MAGIC, sizeof(header->magic)); header->check_sum = cpu_to_le32(BROM_STAMP_VALUE);
On Sat, 19 Jun 2021 17:20:04 +0800 Icenowy Zheng icenowy@aosc.io wrote:
Refactor some functions in mkimage sunxi_egon type, in order to prepare for adding support for more CPU architectures (e.g. RISC-V). In addition, compatibility for operation w/o specified architecture is kept, in this case the architecture is assumed as ARM.
Signed-off-by: Icenowy Zheng icenowy@aosc.io
Looks alright: Reviewed-by: Andre Przywara andre.przywara@arm.com
One thing you might want to improve (just when you respin anyway):
Changes in v2:
- Merged fixes in the next patch in v1 (patch rebase issue).
tools/sunxi_egon.c | 63 ++++++++++++++++++++++++++++++++++++++++------ 1 file changed, 56 insertions(+), 7 deletions(-)
diff --git a/tools/sunxi_egon.c b/tools/sunxi_egon.c index a5299eb6a1..062c9bc151 100644 --- a/tools/sunxi_egon.c +++ b/tools/sunxi_egon.c @@ -16,7 +16,25 @@
static int egon_check_params(struct image_tool_params *params) {
- /* We just need a binary image file. */
- int arch;
- /* Assume ARM when no architecture specified for compatibility */
- if (params->Aflag)
arch = params->arch;- else
arch = IH_ARCH_ARM;
This code snippet will be used three times in this file, so you might want to factor this out.
Cheers, Andre
- /*
* Check whether the architecture is supported.*/- switch (arch) {
- case IH_ARCH_ARM:
break;- default:
return EXIT_FAILURE;- }
- /* We need a binary image file. */ return !params->dflag;
}
@@ -25,10 +43,26 @@ static int egon_verify_header(unsigned char *ptr, int image_size, { const struct boot_file_head *header = (void *)ptr; uint32_t length;
- int arch;
- /* First 4 bytes must be an ARM branch instruction. */
- if ((le32_to_cpu(header->b_instruction) & 0xff000000) != 0xea000000)
return EXIT_FAILURE;
/* Assume ARM when no architecture specified for compatibility */
if (params->Aflag)
arch = params->arch;else
arch = IH_ARCH_ARM;/*
* First 4 bytes must be a branch instruction of the corresponding* architecture.*/switch (arch) {
case IH_ARCH_ARM:
if ((le32_to_cpu(header->b_instruction) & 0xff000000) != 0xea000000)return EXIT_FAILURE;break;default:
return EXIT_FAILURE; /* Unknown architecture */}
if (memcmp(header->magic, BOOT0_MAGIC, sizeof(header->magic))) return EXIT_FAILURE;
@@ -76,10 +110,25 @@ static void egon_set_header(void *buf, struct stat *sbuf, int infd, uint32_t *buf32 = buf; uint32_t checksum = 0, value; int i;
- int arch;
- /* Generate an ARM branch instruction to jump over the header. */
- value = 0xea000000 | (sizeof(struct boot_file_head) / 4 - 2);
- header->b_instruction = cpu_to_le32(value);
/* Assume ARM when no architecture specified for compatibility */
if (params->Aflag)
arch = params->arch;else
arch = IH_ARCH_ARM;/*
* Different architectures need different first instruction to* branch to the body.*/switch (arch) {
case IH_ARCH_ARM:
/* Generate an ARM branch instruction to jump over the header. */value = 0xea000000 | (sizeof(struct boot_file_head) / 4 - 2);header->b_instruction = cpu_to_le32(value);break;}
memcpy(header->magic, BOOT0_MAGIC, sizeof(header->magic)); header->check_sum = cpu_to_le32(BROM_STAMP_VALUE);
There's now a sun20i family in sunxi, which uses RISC-V CPU.
Add support for making eGON.BT0 image for RISC-V.
Signed-off-by: Icenowy Zheng icenowy@aosc.io --- Changes in v2: - Removed changes that should belong to the previous patch in v1.
tools/sunxi_egon.c | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+)
diff --git a/tools/sunxi_egon.c b/tools/sunxi_egon.c index 062c9bc151..836e99a6e6 100644 --- a/tools/sunxi_egon.c +++ b/tools/sunxi_egon.c @@ -29,6 +29,7 @@ static int egon_check_params(struct image_tool_params *params) */ switch (arch) { case IH_ARCH_ARM: + case IH_ARCH_RISCV: break; default: return EXIT_FAILURE; @@ -60,6 +61,10 @@ static int egon_verify_header(unsigned char *ptr, int image_size, if ((le32_to_cpu(header->b_instruction) & 0xff000000) != 0xea000000) return EXIT_FAILURE; break; + case IH_ARCH_RISCV: + if ((le32_to_cpu(header->b_instruction) & 0x00000fff) != 0x0000006f) + return EXIT_FAILURE; + break; default: return EXIT_FAILURE; /* Unknown architecture */ } @@ -128,6 +133,24 @@ static void egon_set_header(void *buf, struct stat *sbuf, int infd, value = 0xea000000 | (sizeof(struct boot_file_head) / 4 - 2); header->b_instruction = cpu_to_le32(value); break; + case IH_ARCH_RISCV: + /* + * Generate a RISC-V JAL instruction with rd=x0 + * (pseudo instruction J, jump without side effects). + * + * The following weird bit operation maps imm[20] + * to inst[31], imm[10:1] to inst[30:21], + * imm[11] to inst[20], imm[19:12] to inst[19:12], + * and imm[0] is dropped (because 1-byte RISC-V instruction + * is not allowed). + */ + value = 0x0000006f | + ((sizeof(struct boot_file_head) & 0x00100000) << 11) | + ((sizeof(struct boot_file_head) & 0x000007fe) << 20) | + ((sizeof(struct boot_file_head) & 0x00000800) << 9) | + ((sizeof(struct boot_file_head) & 0x000ff000) << 0); + header->b_instruction = cpu_to_le32(value); + break; }
memcpy(header->magic, BOOT0_MAGIC, sizeof(header->magic));
On Sat, 19 Jun 2021 17:20:05 +0800 Icenowy Zheng icenowy@aosc.io wrote:
There's now a sun20i family in sunxi, which uses RISC-V CPU.
Add support for making eGON.BT0 image for RISC-V.
Signed-off-by: Icenowy Zheng icenowy@aosc.io
Compared against the RISC-V manual.
Reviewed-by: Andre Przywara andre.przywara@arm.com
Cheers, Andre
Changes in v2:
- Removed changes that should belong to the previous patch in v1.
tools/sunxi_egon.c | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+)
diff --git a/tools/sunxi_egon.c b/tools/sunxi_egon.c index 062c9bc151..836e99a6e6 100644 --- a/tools/sunxi_egon.c +++ b/tools/sunxi_egon.c @@ -29,6 +29,7 @@ static int egon_check_params(struct image_tool_params *params) */ switch (arch) { case IH_ARCH_ARM:
- case IH_ARCH_RISCV: break; default: return EXIT_FAILURE;
@@ -60,6 +61,10 @@ static int egon_verify_header(unsigned char *ptr, int image_size, if ((le32_to_cpu(header->b_instruction) & 0xff000000) != 0xea000000) return EXIT_FAILURE; break;
- case IH_ARCH_RISCV:
if ((le32_to_cpu(header->b_instruction) & 0x00000fff) != 0x0000006f)return EXIT_FAILURE;break;@@ -128,6 +133,24 @@ static void egon_set_header(void *buf, struct stat *sbuf, int infd, value = 0xea000000 | (sizeof(struct boot_file_head) / 4 - 2); header->b_instruction = cpu_to_le32(value); break;
case IH_ARCH_RISCV:
/** Generate a RISC-V JAL instruction with rd=x0* (pseudo instruction J, jump without side effects).** The following weird bit operation maps imm[20]* to inst[31], imm[10:1] to inst[30:21],* imm[11] to inst[20], imm[19:12] to inst[19:12],* and imm[0] is dropped (because 1-byte RISC-V instruction* is not allowed).*/value = 0x0000006f |((sizeof(struct boot_file_head) & 0x00100000) << 11) |((sizeof(struct boot_file_head) & 0x000007fe) << 20) |((sizeof(struct boot_file_head) & 0x00000800) << 9) |((sizeof(struct boot_file_head) & 0x000ff000) << 0);header->b_instruction = cpu_to_le32(value);break;}
memcpy(header->magic, BOOT0_MAGIC, sizeof(header->magic));
As mkimage -T sunxi_egon now gains support for -A parameter, specify the architecture when generating SPL boot image for sunxi.
Signed-off-by: Icenowy Zheng icenowy@aosc.io --- No changes in v2.
scripts/Makefile.spl | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/scripts/Makefile.spl b/scripts/Makefile.spl index 5be1a9ba1b..c13a41a243 100644 --- a/scripts/Makefile.spl +++ b/scripts/Makefile.spl @@ -414,7 +414,7 @@ endif $(obj)/$(SPL_BIN).sfp: $(obj)/$(SPL_BIN).bin FORCE $(call if_changed,mkimage)
-MKIMAGEFLAGS_sunxi-spl.bin = -T sunxi_egon \ +MKIMAGEFLAGS_sunxi-spl.bin = -A $(ARCH) -T sunxi_egon \ -n $(CONFIG_DEFAULT_DEVICE_TREE)
OBJCOPYFLAGS_u-boot-spl-dtb.hex := -I binary -O ihex --change-address=$(CONFIG_SPL_TEXT_BASE)
On Sat, 19 Jun 2021 17:21:11 +0800 Icenowy Zheng icenowy@aosc.io wrote:
As mkimage -T sunxi_egon now gains support for -A parameter, specify the architecture when generating SPL boot image for sunxi.
Signed-off-by: Icenowy Zheng icenowy@aosc.io
Reviewed-by: Andre Przywara andre.przywara@arm.com
Thanks! Andre
No changes in v2.
scripts/Makefile.spl | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/scripts/Makefile.spl b/scripts/Makefile.spl index 5be1a9ba1b..c13a41a243 100644 --- a/scripts/Makefile.spl +++ b/scripts/Makefile.spl @@ -414,7 +414,7 @@ endif $(obj)/$(SPL_BIN).sfp: $(obj)/$(SPL_BIN).bin FORCE $(call if_changed,mkimage)
-MKIMAGEFLAGS_sunxi-spl.bin = -T sunxi_egon \ +MKIMAGEFLAGS_sunxi-spl.bin = -A $(ARCH) -T sunxi_egon \ -n $(CONFIG_DEFAULT_DEVICE_TREE)
OBJCOPYFLAGS_u-boot-spl-dtb.hex := -I binary -O ihex --change-address=$(CONFIG_SPL_TEXT_BASE)
On 6/19/21 4:18 AM, Icenowy Zheng wrote:
This patchset tries to make mkimage -T sunxi_egon to be able to generate an eGON.BT0 image for Allwinner RISC-V SoCs (e.g. D1).
In addition, to keep the compatibility, it will still consider the architecture to be ARM when no architecture is specified.
Icenowy Zheng (4): mkimage: add a flag to describe whether -A is specified mkimage: sunxi_egon: refactor for multi-architecture support mkimage: sunxi_egon: add support for riscv sunxi: specify architecture when generating SPL boot image
scripts/Makefile.spl | 2 +- tools/imagetool.h | 1 + tools/mkimage.c | 1 + tools/sunxi_egon.c | 86 ++++++++++++++++++++++++++++++++++++++++---- 4 files changed, 82 insertions(+), 8 deletions(-)
This works great on D1. For the whole series:
Tested-by: Samuel Holland samuel@sholland.org
On Sat, 19 Jun 2021 17:18:34 +0800 Icenowy Zheng icenowy@aosc.io wrote:
This patchset tries to make mkimage -T sunxi_egon to be able to generate an eGON.BT0 image for Allwinner RISC-V SoCs (e.g. D1).
In addition, to keep the compatibility, it will still consider the architecture to be ARM when no architecture is specified.
All merged into sunxi/master (which already landed in mainline).
Thanks! Andre
Icenowy Zheng (4): mkimage: add a flag to describe whether -A is specified mkimage: sunxi_egon: refactor for multi-architecture support mkimage: sunxi_egon: add support for riscv sunxi: specify architecture when generating SPL boot image
scripts/Makefile.spl | 2 +- tools/imagetool.h | 1 + tools/mkimage.c | 1 + tools/sunxi_egon.c | 86 ++++++++++++++++++++++++++++++++++++++++---- 4 files changed, 82 insertions(+), 8 deletions(-)
participants (5)
-
Andre Przywara -
Icenowy Zheng -
Samuel Holland -
Simon Glass -
Tom Rini