Re: [PATCH 8/8] riscv: add two hook functions of ftrace

[Zong Li]

Palmer Dabbelt <palmer@xxxxxxxxxxx> æ 2020å3æ5æ éå äå9:44åéï
>
> On Mon, 17 Feb 2020 00:32:23 PST (-0800), zong.li@xxxxxxxxxx wrote:
> > After the text section be mask as non-writable, the ftrace have to
> > change the permission of text for dynamic patching the intructions.
> > Add ftrace_arch_code_modify_prepare and
> > ftrace_arch_code_modify_post_process to change permission.
> >
> > Signed-off-by: Zong Li <zong.li@xxxxxxxxxx>
> > ---
> >  arch/riscv/kernel/ftrace.c | 18 ++++++++++++++++++
> >  1 file changed, 18 insertions(+)
> >
> > diff --git a/arch/riscv/kernel/ftrace.c b/arch/riscv/kernel/ftrace.c
> > index c40fdcdeb950..576df0807200 100644
> > --- a/arch/riscv/kernel/ftrace.c
> > +++ b/arch/riscv/kernel/ftrace.c
> > @@ -7,9 +7,27 @@
> >
> >  #include <linux/ftrace.h>
> >  #include <linux/uaccess.h>
> > +#include <linux/memory.h>
> > +#include <asm/set_memory.h>
> >  #include <asm/cacheflush.h>
> >
> >  #ifdef CONFIG_DYNAMIC_FTRACE
> > +int ftrace_arch_code_modify_prepare(void) __acquires(&text_mutex)
> > +{
> > +     mutex_lock(&text_mutex);
> > +     set_kernel_text_rw();
> > +     set_all_modules_text_rw();
> > +     return 0;
> > +}
>
> None of the other architectures are doing anything remotely like this in these
> functions, despite many supporting STRICT_KERNEL_RWX.  Having a function that
> maps all text as RW seems super dangerous, as one stack attack means NX is
> gone.
>
> Looks like FIX_TEXT_POKE0 is the magic that makes the other ports work.
>

Your concern is right. Let me change the way.

> > +
> > +int ftrace_arch_code_modify_post_process(void) __releases(&text_mutex)
> > +{
> > +     set_all_modules_text_ro();
> > +     set_kernel_text_ro();
> > +     mutex_unlock(&text_mutex);
> > +     return 0;
> > +}
>
> Presumably this needs a icache flush as well?  Probably better to do so when
> installing the instructions, though.
>

Yes, I think I lost it. Thanks.

> > +
> >  static int ftrace_check_current_call(unsigned long hook_pos,
> >                                    unsigned int *expected)
> >  {
>