2022-12-09 06:45:42

by Alexander Gordeev

Subject: [PATCH] jump_label: explicitly annotate inittext labels as init

inittext code may be out of [__init_begin, __init_end]
range on some architectures. Yet, the jump_label_init()
only calls init_section_contains() function to check if
a label needs to be annotated as init and inittext code
is left behind.

Fixes: 19483677684b ("jump_label: Annotate entries that operate on __init code earlier")
Signed-off-by: Alexander Gordeev <[email protected]>
---
kernel/jump_label.c | 5 ++++-
1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/kernel/jump_label.c b/kernel/jump_label.c
index 714ac4c3b556..77680665d374 100644
--- a/kernel/jump_label.c
+++ b/kernel/jump_label.c
@@ -481,13 +481,16 @@ void __init jump_label_init(void)

for (iter = iter_start; iter < iter_stop; iter++) {
struct static_key *iterk;
+ unsigned long addr;
bool in_init;

/* rewrite NOPs */
if (jump_label_type(iter) == JUMP_LABEL_NOP)
arch_jump_label_transform_static(iter, JUMP_LABEL_NOP);

- in_init = init_section_contains((void *)jump_entry_code(iter), 1);
+ addr = jump_entry_code(iter);
+ in_init = init_section_contains((void *)addr, 1) ||
+ is_kernel_inittext(addr);
jump_entry_set_init(iter, in_init);

iterk = jump_entry_key(iter);
--
2.34.1

2022-12-09 09:55:51

by Ard Biesheuvel

Subject: Re: [PATCH] jump_label: explicitly annotate inittext labels as init

On Fri, 9 Dec 2022 at 07:08, Alexander Gordeev <[email protected]> wrote:
>
> inittext code may be out of [__init_begin, __init_end]
> range on some architectures. Yet, the jump_label_init()
> only calls init_section_contains() function to check if
> a label needs to be annotated as init and inittext code
> is left behind.
>
> Fixes: 19483677684b ("jump_label: Annotate entries that operate on __init code earlier")
> Signed-off-by: Alexander Gordeev <[email protected]>
> ---
> kernel/jump_label.c | 5 ++++-
> 1 file changed, 4 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/jump_label.c b/kernel/jump_label.c
> index 714ac4c3b556..77680665d374 100644
> --- a/kernel/jump_label.c
> +++ b/kernel/jump_label.c
> @@ -481,13 +481,16 @@ void __init jump_label_init(void)
>
> for (iter = iter_start; iter < iter_stop; iter++) {
> struct static_key *iterk;
> + unsigned long addr;
> bool in_init;
>
> /* rewrite NOPs */
> if (jump_label_type(iter) == JUMP_LABEL_NOP)
> arch_jump_label_transform_static(iter, JUMP_LABEL_NOP);
>
> - in_init = init_section_contains((void *)jump_entry_code(iter), 1);
> + addr = jump_entry_code(iter);
> + in_init = init_section_contains((void *)addr, 1) ||
> + is_kernel_inittext(addr);

Isn't it sufficient to only call is_kenel_inittext here?

> jump_entry_set_init(iter, in_init);
>
> iterk = jump_entry_key(iter);
> --
> 2.34.1
>

2022-12-09 13:25:24

by Alexander Gordeev

Subject: Re: [PATCH] jump_label: explicitly annotate inittext labels as init

On Fri, Dec 09, 2022 at 10:41:55AM +0100, Ard Biesheuvel wrote:
> > diff --git a/kernel/jump_label.c b/kernel/jump_label.c
> > index 714ac4c3b556..77680665d374 100644
> > --- a/kernel/jump_label.c
> > +++ b/kernel/jump_label.c
> > @@ -481,13 +481,16 @@ void __init jump_label_init(void)
> >
> > for (iter = iter_start; iter < iter_stop; iter++) {
> > struct static_key *iterk;
> > + unsigned long addr;
> > bool in_init;
> >
> > /* rewrite NOPs */
> > if (jump_label_type(iter) == JUMP_LABEL_NOP)
> > arch_jump_label_transform_static(iter, JUMP_LABEL_NOP);
> >
> > - in_init = init_section_contains((void *)jump_entry_code(iter), 1);
> > + addr = jump_entry_code(iter);
> > + in_init = init_section_contains((void *)addr, 1) ||
> > + is_kernel_inittext(addr);
>
> Isn't it sufficient to only call is_kenel_inittext here?

I am not sure every arch out there would stick to putting
init code to inittext only.

However, should not is_kernel_exittext() (which does not
exist) also get checked here?

2022-12-09 13:54:07

by Ard Biesheuvel

Subject: Re: [PATCH] jump_label: explicitly annotate inittext labels as init

On Fri, 9 Dec 2022 at 13:49, Alexander Gordeev <[email protected]> wrote:
>
> On Fri, Dec 09, 2022 at 10:41:55AM +0100, Ard Biesheuvel wrote:
> > > diff --git a/kernel/jump_label.c b/kernel/jump_label.c
> > > index 714ac4c3b556..77680665d374 100644
> > > --- a/kernel/jump_label.c
> > > +++ b/kernel/jump_label.c
> > > @@ -481,13 +481,16 @@ void __init jump_label_init(void)
> > >
> > > for (iter = iter_start; iter < iter_stop; iter++) {
> > > struct static_key *iterk;
> > > + unsigned long addr;
> > > bool in_init;
> > >
> > > /* rewrite NOPs */
> > > if (jump_label_type(iter) == JUMP_LABEL_NOP)
> > > arch_jump_label_transform_static(iter, JUMP_LABEL_NOP);
> > >
> > > - in_init = init_section_contains((void *)jump_entry_code(iter), 1);
> > > + addr = jump_entry_code(iter);
> > > + in_init = init_section_contains((void *)addr, 1) ||
> > > + is_kernel_inittext(addr);
> >
> > Isn't it sufficient to only call is_kenel_inittext here?
>
> I am not sure every arch out there would stick to putting
> init code to inittext only.
>
> However, should not is_kernel_exittext() (which does not
> exist) also get checked here?

No, exittext either exists or it doesn't, rather than disappearing at
runtime like inittext.

2022-12-09 15:10:13

by Ard Biesheuvel

Subject: Re: [PATCH] jump_label: explicitly annotate inittext labels as init

On Fri, 9 Dec 2022 at 15:48, Alexander Gordeev <[email protected]> wrote:
>
> On Fri, Dec 09, 2022 at 02:48:49PM +0100, Ard Biesheuvel wrote:
> > On Fri, 9 Dec 2022 at 13:49, Alexander Gordeev <[email protected]> wrote:
> > >
> > > On Fri, Dec 09, 2022 at 10:41:55AM +0100, Ard Biesheuvel wrote:
> > > > > diff --git a/kernel/jump_label.c b/kernel/jump_label.c
> > > > > index 714ac4c3b556..77680665d374 100644
> > > > > --- a/kernel/jump_label.c
> > > > > +++ b/kernel/jump_label.c
> > > > > @@ -481,13 +481,16 @@ void __init jump_label_init(void)
> > > > >
> > > > > for (iter = iter_start; iter < iter_stop; iter++) {
> > > > > struct static_key *iterk;
> > > > > + unsigned long addr;
> > > > > bool in_init;
> > > > >
> > > > > /* rewrite NOPs */
> > > > > if (jump_label_type(iter) == JUMP_LABEL_NOP)
> > > > > arch_jump_label_transform_static(iter, JUMP_LABEL_NOP);
> > > > >
> > > > > - in_init = init_section_contains((void *)jump_entry_code(iter), 1);
> > > > > + addr = jump_entry_code(iter);
> > > > > + in_init = init_section_contains((void *)addr, 1) ||
> > > > > + is_kernel_inittext(addr);
> > > >
> > > > Isn't it sufficient to only call is_kenel_inittext here?
> > >
> > > I am not sure every arch out there would stick to putting
> > > init code to inittext only.
> > >
> > > However, should not is_kernel_exittext() (which does not
> > > exist) also get checked here?
> >
> > No, exittext either exists or it doesn't, rather than disappearing at
> > runtime like inittext.
>
> May be just swap the order of init_section_contains() and
> is_kernel_inittext() checks then?

No I don't think we need both.

2022-12-09 15:47:46

by Alexander Gordeev

Subject: Re: [PATCH] jump_label: explicitly annotate inittext labels as init

On Fri, Dec 09, 2022 at 02:48:49PM +0100, Ard Biesheuvel wrote:
> On Fri, 9 Dec 2022 at 13:49, Alexander Gordeev <[email protected]> wrote:
> >
> > On Fri, Dec 09, 2022 at 10:41:55AM +0100, Ard Biesheuvel wrote:
> > > > diff --git a/kernel/jump_label.c b/kernel/jump_label.c
> > > > index 714ac4c3b556..77680665d374 100644
> > > > --- a/kernel/jump_label.c
> > > > +++ b/kernel/jump_label.c
> > > > @@ -481,13 +481,16 @@ void __init jump_label_init(void)
> > > >
> > > > for (iter = iter_start; iter < iter_stop; iter++) {
> > > > struct static_key *iterk;
> > > > + unsigned long addr;
> > > > bool in_init;
> > > >
> > > > /* rewrite NOPs */
> > > > if (jump_label_type(iter) == JUMP_LABEL_NOP)
> > > > arch_jump_label_transform_static(iter, JUMP_LABEL_NOP);
> > > >
> > > > - in_init = init_section_contains((void *)jump_entry_code(iter), 1);
> > > > + addr = jump_entry_code(iter);
> > > > + in_init = init_section_contains((void *)addr, 1) ||
> > > > + is_kernel_inittext(addr);
> > >
> > > Isn't it sufficient to only call is_kenel_inittext here?
> >
> > I am not sure every arch out there would stick to putting
> > init code to inittext only.
> >
> > However, should not is_kernel_exittext() (which does not
> > exist) also get checked here?
>
> No, exittext either exists or it doesn't, rather than disappearing at
> runtime like inittext.

May be just swap the order of init_section_contains() and
is_kernel_inittext() checks then?