2020-04-30 09:55:32

by Zong Li

Subject: [PATCH] riscv: force __cpu_up_ variables to put in data section

Put __cpu_up_stack_pointer and __cpu_up_task_pointer in data section.
Currently, these two variables are put in bss section, there is a
potential risk that secondary harts get the uninitialized value before
main hart finishing the bss clearing. In this case, all secondary
harts would go through the waiting loop and enable the MMU before
main hart set up the page table.

Signed-off-by: Zong Li <[email protected]>
---
arch/riscv/kernel/cpu_ops.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/riscv/kernel/cpu_ops.c b/arch/riscv/kernel/cpu_ops.c
index c4c33bf02369..0ec22354018c 100644
--- a/arch/riscv/kernel/cpu_ops.c
+++ b/arch/riscv/kernel/cpu_ops.c
@@ -15,8 +15,8 @@

const struct cpu_operations *cpu_ops[NR_CPUS] __ro_after_init;

-void *__cpu_up_stack_pointer[NR_CPUS];
-void *__cpu_up_task_pointer[NR_CPUS];
+void *__cpu_up_stack_pointer[NR_CPUS] __section(.data);
+void *__cpu_up_task_pointer[NR_CPUS] __section(.data);

extern const struct cpu_operations cpu_ops_sbi;
extern const struct cpu_operations cpu_ops_spinwait;
--
2.26.1

2020-04-30 14:49:03

by Greentime Hu

Subject: Re: [PATCH] riscv: force __cpu_up_ variables to put in data section

Zong Li <[email protected]> 於 2020年4月30日 週四 下午5:53寫道：
>
> Put __cpu_up_stack_pointer and __cpu_up_task_pointer in data section.
> Currently, these two variables are put in bss section, there is a
> potential risk that secondary harts get the uninitialized value before
> main hart finishing the bss clearing. In this case, all secondary
> harts would go through the waiting loop and enable the MMU before
> main hart set up the page table.
>
> Signed-off-by: Zong Li <[email protected]>
> ---
> arch/riscv/kernel/cpu_ops.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/arch/riscv/kernel/cpu_ops.c b/arch/riscv/kernel/cpu_ops.c
> index c4c33bf02369..0ec22354018c 100644
> --- a/arch/riscv/kernel/cpu_ops.c
> +++ b/arch/riscv/kernel/cpu_ops.c
> @@ -15,8 +15,8 @@
>
> const struct cpu_operations *cpu_ops[NR_CPUS] __ro_after_init;
>
> -void *__cpu_up_stack_pointer[NR_CPUS];
> -void *__cpu_up_task_pointer[NR_CPUS];
> +void *__cpu_up_stack_pointer[NR_CPUS] __section(.data);
> +void *__cpu_up_task_pointer[NR_CPUS] __section(.data);
>
> extern const struct cpu_operations cpu_ops_sbi;
> extern const struct cpu_operations cpu_ops_spinwait;

Reviewed-by: Greentime Hu <[email protected]>

2020-04-30 18:25:32

by Atish Patra

Subject: Re: [PATCH] riscv: force __cpu_up_ variables to put in data section

On Thu, Apr 30, 2020 at 2:53 AM Zong Li <[email protected]> wrote:
>
> Put __cpu_up_stack_pointer and __cpu_up_task_pointer in data section.
> Currently, these two variables are put in bss section, there is a
> potential risk that secondary harts get the uninitialized value before
> main hart finishing the bss clearing. In this case, all secondary
> harts would go through the waiting loop and enable the MMU before
> main hart set up the page table.
>

That would be only true if you are using random booting protocol with
SBI v0.1 implementation.
With HSM extension in place, all the secondary cores are waiting in
firmware. The booting core
will bring them up one by one from cpu_up method.

The HSM extension is already available in OpenSBI v0.7 and Linux
kernel 5.7-rc1 onwards.

> Signed-off-by: Zong Li <[email protected]>
> ---
> arch/riscv/kernel/cpu_ops.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/arch/riscv/kernel/cpu_ops.c b/arch/riscv/kernel/cpu_ops.c
> index c4c33bf02369..0ec22354018c 100644
> --- a/arch/riscv/kernel/cpu_ops.c
> +++ b/arch/riscv/kernel/cpu_ops.c
> @@ -15,8 +15,8 @@
>
> const struct cpu_operations *cpu_ops[NR_CPUS] __ro_after_init;
>
> -void *__cpu_up_stack_pointer[NR_CPUS];
> -void *__cpu_up_task_pointer[NR_CPUS];
> +void *__cpu_up_stack_pointer[NR_CPUS] __section(.data);
> +void *__cpu_up_task_pointer[NR_CPUS] __section(.data);
>
> extern const struct cpu_operations cpu_ops_sbi;
> extern const struct cpu_operations cpu_ops_spinwait;
> --
> 2.26.1
>
>


--
Regards,
Atish

2020-05-02 06:02:40

by Zong Li

Subject: Re: [PATCH] riscv: force __cpu_up_ variables to put in data section

On Fri, May 1, 2020 at 2:23 AM Atish Patra <[email protected]> wrote:
>
> On Thu, Apr 30, 2020 at 2:53 AM Zong Li <[email protected]> wrote:
> >
> > Put __cpu_up_stack_pointer and __cpu_up_task_pointer in data section.
> > Currently, these two variables are put in bss section, there is a
> > potential risk that secondary harts get the uninitialized value before
> > main hart finishing the bss clearing. In this case, all secondary
> > harts would go through the waiting loop and enable the MMU before
> > main hart set up the page table.
> >
>
> That would be only true if you are using random booting protocol with
> SBI v0.1 implementation.
> With HSM extension in place, all the secondary cores are waiting in
> firmware. The booting core
> will bring them up one by one from cpu_up method.
>
> The HSM extension is already available in OpenSBI v0.7 and Linux
> kernel 5.7-rc1 onwards.

If I understand correctly, the newest kernel still works with earlier
OpenSBI than v0.7 or BBL. It seems to me that we need to consider the
use cases if we don't limit it to up to OpenSBI v0.7.

>
> > Signed-off-by: Zong Li <[email protected]>
> > ---
> > arch/riscv/kernel/cpu_ops.c | 4 ++--
> > 1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/riscv/kernel/cpu_ops.c b/arch/riscv/kernel/cpu_ops.c
> > index c4c33bf02369..0ec22354018c 100644
> > --- a/arch/riscv/kernel/cpu_ops.c
> > +++ b/arch/riscv/kernel/cpu_ops.c
> > @@ -15,8 +15,8 @@
> >
> > const struct cpu_operations *cpu_ops[NR_CPUS] __ro_after_init;
> >
> > -void *__cpu_up_stack_pointer[NR_CPUS];
> > -void *__cpu_up_task_pointer[NR_CPUS];
> > +void *__cpu_up_stack_pointer[NR_CPUS] __section(.data);
> > +void *__cpu_up_task_pointer[NR_CPUS] __section(.data);
> >
> > extern const struct cpu_operations cpu_ops_sbi;
> > extern const struct cpu_operations cpu_ops_spinwait;
> > --
> > 2.26.1
> >
> >
>
>
> --
> Regards,
> Atish

2020-05-02 06:15:52

by Anup Patel

Subject: Re: [PATCH] riscv: force __cpu_up_ variables to put in data section

On Sat, May 2, 2020 at 11:30 AM Zong Li <[email protected]> wrote:
>
> On Fri, May 1, 2020 at 2:23 AM Atish Patra <[email protected]> wrote:
> >
> > On Thu, Apr 30, 2020 at 2:53 AM Zong Li <[email protected]> wrote:
> > >
> > > Put __cpu_up_stack_pointer and __cpu_up_task_pointer in data section.
> > > Currently, these two variables are put in bss section, there is a
> > > potential risk that secondary harts get the uninitialized value before
> > > main hart finishing the bss clearing. In this case, all secondary
> > > harts would go through the waiting loop and enable the MMU before
> > > main hart set up the page table.
> > >
> >
> > That would be only true if you are using random booting protocol with
> > SBI v0.1 implementation.
> > With HSM extension in place, all the secondary cores are waiting in
> > firmware. The booting core
> > will bring them up one by one from cpu_up method.
> >
> > The HSM extension is already available in OpenSBI v0.7 and Linux
> > kernel 5.7-rc1 onwards.
>
> If I understand correctly, the newest kernel still works with earlier
> OpenSBI than v0.7 or BBL. It seems to me that we need to consider the
> use cases if we don't limit it to up to OpenSBI v0.7.

I think the commit description should clearly state that the issue will
manifest only for random booting of multiple HARTs which means it
will manifest only for BBL and OpenSBI v0.6 (or older).

Regards,
Anup

2020-05-02 09:55:31

by Zong Li

Subject: Re: [PATCH] riscv: force __cpu_up_ variables to put in data section

On Sat, May 2, 2020 at 2:13 PM Anup Patel <[email protected]> wrote:
>
> On Sat, May 2, 2020 at 11:30 AM Zong Li <[email protected]> wrote:
> >
> > On Fri, May 1, 2020 at 2:23 AM Atish Patra <[email protected]> wrote:
> > >
> > > On Thu, Apr 30, 2020 at 2:53 AM Zong Li <[email protected]> wrote:
> > > >
> > > > Put __cpu_up_stack_pointer and __cpu_up_task_pointer in data section.
> > > > Currently, these two variables are put in bss section, there is a
> > > > potential risk that secondary harts get the uninitialized value before
> > > > main hart finishing the bss clearing. In this case, all secondary
> > > > harts would go through the waiting loop and enable the MMU before
> > > > main hart set up the page table.
> > > >
> > >
> > > That would be only true if you are using random booting protocol with
> > > SBI v0.1 implementation.
> > > With HSM extension in place, all the secondary cores are waiting in
> > > firmware. The booting core
> > > will bring them up one by one from cpu_up method.
> > >
> > > The HSM extension is already available in OpenSBI v0.7 and Linux
> > > kernel 5.7-rc1 onwards.
> >
> > If I understand correctly, the newest kernel still works with earlier
> > OpenSBI than v0.7 or BBL. It seems to me that we need to consider the
> > use cases if we don't limit it to up to OpenSBI v0.7.
>
> I think the commit description should clearly state that the issue will
> manifest only for random booting of multiple HARTs which means it
> will manifest only for BBL and OpenSBI v0.6 (or older).
>

OK. Let me modify the description. Thanks.

> Regards,
> Anup