[PATCH 1/2] arm64: Add missing GD_FLG_SKIP_RELOC handling

From: Marek Vasut marek.vasut+renesas@gmail.com
In case U-Boot enters relocation with GD_FLG_SKIP_RELOC, skip the relocation. The code still has to set up new_gd pointer and new stack pointer.
Signed-off-by: Marek Vasut marek.vasut+renesas@gmail.com Cc: Simon Glass sjg@chromium.org Cc: Tom Rini trini@konsulko.com --- arch/arm/lib/crt0_64.S | 4 ++++ lib/asm-offsets.c | 3 +++ 2 files changed, 7 insertions(+)
diff --git a/arch/arm/lib/crt0_64.S b/arch/arm/lib/crt0_64.S index 680e674fa3..28c8356aee 100644 --- a/arch/arm/lib/crt0_64.S +++ b/arch/arm/lib/crt0_64.S @@ -104,6 +104,10 @@ ENTRY(_main) bic sp, x0, #0xf /* 16-byte alignment for ABI compliance */ ldr x18, [x18, #GD_NEW_GD] /* x18 <- gd->new_gd */
+ /* Skip relocation in case gd->gd_flags & GD_FLG_SKIP_RELOC */ + ldr x0, [x18, #GD_FLAGS] /* x0 <- gd->flags */ + tbnz x0, 11, relocation_return /* GD_FLG_SKIP_RELOC is bit 11 */ + adr lr, relocation_return #if CONFIG_POSITION_INDEPENDENT /* Add in link-vs-runtime offset */ diff --git a/lib/asm-offsets.c b/lib/asm-offsets.c index c691066332..0808cd4b0c 100644 --- a/lib/asm-offsets.c +++ b/lib/asm-offsets.c @@ -29,6 +29,9 @@ int main(void) DEFINE(GD_SIZE, sizeof(struct global_data));
DEFINE(GD_BD, offsetof(struct global_data, bd)); + + DEFINE(GD_FLAGS, offsetof(struct global_data, flags)); + #if CONFIG_VAL(SYS_MALLOC_F_LEN) DEFINE(GD_MALLOC_BASE, offsetof(struct global_data, malloc_base)); #endif

From: Marek Vasut marek.vasut+renesas@gmail.com
In case U-Boot starts with GD_FLG_SKIP_RELOC, the U-Boot code is not relocated, however the stack and heap is at the end of DRAM after relocation. Reserve a LMB area for the non-relocated U-Boot code so it won't be overwritten.
Signed-off-by: Marek Vasut marek.vasut+renesas@gmail.com Cc: Simon Glass sjg@chromium.org Cc: Tom Rini trini@konsulko.com --- arch/arm/lib/stack.c | 9 +++++++++ 1 file changed, 9 insertions(+)
diff --git a/arch/arm/lib/stack.c b/arch/arm/lib/stack.c index 656084c7e5..d2e2715ecf 100644 --- a/arch/arm/lib/stack.c +++ b/arch/arm/lib/stack.c @@ -14,6 +14,7 @@ #include <init.h> #include <lmb.h> #include <asm/global_data.h> +#include <asm/sections.h>
DECLARE_GLOBAL_DATA_PTR;
@@ -46,4 +47,12 @@ static ulong get_sp(void) void arch_lmb_reserve(struct lmb *lmb) { arch_lmb_reserve_generic(lmb, get_sp(), gd->ram_top, 16384); + +#ifdef CONFIG_ARM + if (gd->flags & GD_FLG_SKIP_RELOC) { + lmb_reserve(lmb, (phys_addr_t)__image_copy_start, + (phys_addr_t)__image_copy_end - + (phys_addr_t)__image_copy_start); + } +#endif }

On Sun, Oct 10, 2021 at 11:52:09PM +0200, marek.vasut@gmail.com wrote:
From: Marek Vasut marek.vasut+renesas@gmail.com
In case U-Boot starts with GD_FLG_SKIP_RELOC, the U-Boot code is not relocated, however the stack and heap is at the end of DRAM after relocation. Reserve a LMB area for the non-relocated U-Boot code so it won't be overwritten.
Signed-off-by: Marek Vasut marek.vasut+renesas@gmail.com Cc: Simon Glass sjg@chromium.org Cc: Tom Rini trini@konsulko.com
arch/arm/lib/stack.c | 9 +++++++++ 1 file changed, 9 insertions(+)
diff --git a/arch/arm/lib/stack.c b/arch/arm/lib/stack.c index 656084c7e5..d2e2715ecf 100644 --- a/arch/arm/lib/stack.c +++ b/arch/arm/lib/stack.c @@ -14,6 +14,7 @@ #include <init.h> #include <lmb.h> #include <asm/global_data.h> +#include <asm/sections.h>
DECLARE_GLOBAL_DATA_PTR;
@@ -46,4 +47,12 @@ static ulong get_sp(void) void arch_lmb_reserve(struct lmb *lmb) { arch_lmb_reserve_generic(lmb, get_sp(), gd->ram_top, 16384);
+#ifdef CONFIG_ARM
if (gd->flags & GD_FLG_SKIP_RELOC) {
lmb_reserve(lmb, (phys_addr_t)__image_copy_start,
(phys_addr_t)__image_copy_end -
(phys_addr_t)__image_copy_start);
}
+#endif }
Erm, this is in the arm code, the patch subject says arm64 and you're testing for CONFIG_ARM (which will always be true here). Can you please respin? Also, the whitespace of the new code is too indented. Thanks.

On 2021/10/11 5:52, marek.vasut@gmail.com wrote:
From: Marek Vasut marek.vasut+renesas@gmail.com
In case U-Boot enters relocation with GD_FLG_SKIP_RELOC, skip the relocation. The code still has to set up new_gd pointer and new stack pointer.
Signed-off-by: Marek Vasut marek.vasut+renesas@gmail.com Cc: Simon Glass sjg@chromium.org Cc: Tom Rini trini@konsulko.com
arch/arm/lib/crt0_64.S | 4 ++++ lib/asm-offsets.c | 3 +++ 2 files changed, 7 insertions(+)
diff --git a/arch/arm/lib/crt0_64.S b/arch/arm/lib/crt0_64.S index 680e674fa3..28c8356aee 100644 --- a/arch/arm/lib/crt0_64.S +++ b/arch/arm/lib/crt0_64.S @@ -104,6 +104,10 @@ ENTRY(_main) bic sp, x0, #0xf /* 16-byte alignment for ABI compliance */ ldr x18, [x18, #GD_NEW_GD] /* x18 <- gd->new_gd */
- /* Skip relocation in case gd->gd_flags & GD_FLG_SKIP_RELOC */
- ldr x0, [x18, #GD_FLAGS] /* x0 <- gd->flags */
You are using new_gd, that means bit 11 needs to be set after new_gd has been filled with gd.
I would prefer use gd, not new_gd.
Thanks, Peng.
- tbnz x0, 11, relocation_return /* GD_FLG_SKIP_RELOC is bit 11 */
- adr lr, relocation_return #if CONFIG_POSITION_INDEPENDENT /* Add in link-vs-runtime offset */
diff --git a/lib/asm-offsets.c b/lib/asm-offsets.c index c691066332..0808cd4b0c 100644 --- a/lib/asm-offsets.c +++ b/lib/asm-offsets.c @@ -29,6 +29,9 @@ int main(void) DEFINE(GD_SIZE, sizeof(struct global_data));
DEFINE(GD_BD, offsetof(struct global_data, bd));
- DEFINE(GD_FLAGS, offsetof(struct global_data, flags));
- #if CONFIG_VAL(SYS_MALLOC_F_LEN) DEFINE(GD_MALLOC_BASE, offsetof(struct global_data, malloc_base)); #endif

On 10/14/21 3:55 AM, Peng Fan (OSS) wrote: [...]
diff --git a/arch/arm/lib/crt0_64.S b/arch/arm/lib/crt0_64.S index 680e674fa3..28c8356aee 100644 --- a/arch/arm/lib/crt0_64.S +++ b/arch/arm/lib/crt0_64.S @@ -104,6 +104,10 @@ ENTRY(_main) bic sp, x0, #0xf /* 16-byte alignment for ABI compliance */ ldr x18, [x18, #GD_NEW_GD] /* x18 <- gd->new_gd */ + /* Skip relocation in case gd->gd_flags & GD_FLG_SKIP_RELOC */ + ldr x0, [x18, #GD_FLAGS] /* x0 <- gd->flags */
You are using new_gd, that means bit 11 needs to be set after new_gd has been filled with gd.
I would prefer use gd, not new_gd.
Both gd and newgd have GD_FLG_SKIP_RELOC set very early on, in u-boot that's currently done by one board in mach_cpu_init, so that should be no problem. Moreover, both gd and newgd flags must be identical as far as I can tell.
participants (4)
-
Marek Vasut
-
marek.vasut@gmail.com
-
Peng Fan (OSS)
-
Tom Rini