On Wed, 01 Aug 2018 18:26:48 PDT (-0700), zong...@gmail.com wrote:
Palmer Dabbelt <pal...@sifive.com> 於 2018年8月2日 週四 上午8:38寫道：
On Wed, 20 Jun 2018 18:40:07 PDT (-0700), z...@andestech.com wrote:
> The stvec's value must be 4 byte alignment by specification definition.
> This directive avoids to stvec be set the non-alignment value by the
> following code in head.S
>
> /* Point stvec to virtual address of intruction after satp write */
> la a0, 1f
> add a0, a0, a1
> csrw stvec, a0
>
> Signed-off-by: Zong Li <z...@andestech.com>
> ---
> arch/riscv/kernel/head.S | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/arch/riscv/kernel/head.S b/arch/riscv/kernel/head.S
> index 396ec7b349ce..ae7b204f531c 100644
> --- a/arch/riscv/kernel/head.S
> +++ b/arch/riscv/kernel/head.S
> @@ -94,6 +94,7 @@ relocate:
> or a0, a0, a1
> sfence.vma
> csrw sptbr, a0
> +.align 2
> 1:
> /* Set trap vector to spin forever to help debug */
> la a0, .Lsecondary_park
I don't think this is actually correct: you shouldn't be aligning the address
at which stvec is set, but the address that stvec is set to. If this is fixing
anything then it's probably just by coincidence as it's causing
.Lsecondary_park to be aligned.
I think this patch is the correct fix
diff --git a/arch/riscv/kernel/head.S b/arch/riscv/kernel/head.S
index 6e07ed37bbff..d1beecf1d060 100644
--- a/arch/riscv/kernel/head.S
+++ b/arch/riscv/kernel/head.S
@@ -143,6 +143,7 @@ relocate:
tail smp_callin
#endif
+.align 2
.Lsecondary_park:
/* We lack SMP support or have too many harts, so park this hart */
wfi
Thanks for pointing this out!
The position which I set in this patch is also be set to stvec for jumping to
correct VA at satp enabling. The label .Lsecondary_park is also be set to stvec,
I don't notice it because as you said, this label address is correct
just by coincidence.
I think there are two places need to be aligned.
The first setting as following:
relocate:
...
/* Point stvec to virtual address of intruction after satp write */
la a0, 1f
add a0, a0, a1
csrw stvec, a0 <----------- first set
...
sfence.vma
csrw sptbr, a0
1:
/* Set trap vector to spin forever to help debug */
la a0, .Lsecondary_park
csrw stvec, a0
I agree. I think this should do it:
commit a8c5f596bcd528c5468d621bc84cd901632a912a
gpg: Signature made Fri 03 Aug 2018 09:13:56 PM PDT
gpg: using RSA key 00CE76D1834960DFCE886DF8EF4CA1502CCBAB41
gpg: issuer "pal...@dabbelt.com"
gpg: Good signature from "Palmer Dabbelt <pal...@dabbelt.com>" [ultimate]
gpg: aka "Palmer Dabbelt <pal...@sifive.com>" [ultimate]
Author: Palmer Dabbelt <pal...@sifive.com>
Date: Wed Aug 1 17:38:30 2018 -0700
RISC-V: Align early stvec targets
The ISA mandantes that stvec is 4-byte aligned, but we weren't actually
respecting in the early boot code. These trap vector are never expected
to actually be used unless there's a bug in early in the boot process of
the kernel, which explains why the bug wasn't noticed until recently.
Thanks to Zong for finding the bug!
CC: z...@andestech.com
Signed-off-by: Palmer Dabbelt <pal...@sifive.com>
diff --git a/arch/riscv/kernel/head.S b/arch/riscv/kernel/head.S
index 6e07ed37bbff..c4d2c63f9a29 100644
--- a/arch/riscv/kernel/head.S
+++ b/arch/riscv/kernel/head.S
@@ -94,6 +94,7 @@ relocate:
or a0, a0, a1
sfence.vma
csrw sptbr, a0
+.align 2
1:
/* Set trap vector to spin forever to help debug */
la a0, .Lsecondary_park
@@ -143,6 +144,7 @@ relocate:
tail smp_callin
#endif
+.align 2
.Lsecondary_park:
/* We lack SMP support or have too many harts, so park this hart */
wfi
