[RFC,v2,5/6] ftrace: Add ftrace_partial_regs() for converting ftrace_regs to pt_regs

  From: Masami Hiramatsu (Google) <mhiramat@kernel.org>

Add ftrace_partial_regs() which converts the ftrace_regas to pt_regs.
If the architecture defines its own ftrace_regs, this copies partial
registers to pt_regs and returns it. If not, ftrace_regs is the same as
pt_regs and ftrace_partial_regs() will return ftrace_regs::regs.

Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
---
 arch/arm64/include/asm/ftrace.h |   11 +++++++++++
 include/linux/ftrace.h          |   11 +++++++++++
 2 files changed, 22 insertions(+)
  

On Mon, Aug 7, 2023 at 8:49 AM Masami Hiramatsu (Google)
<mhiramat@kernel.org> wrote:
>
> From: Masami Hiramatsu (Google) <mhiramat@kernel.org>
>
> Add ftrace_partial_regs() which converts the ftrace_regas to pt_regs.

ftrace_regs*

> If the architecture defines its own ftrace_regs, this copies partial
> registers to pt_regs and returns it. If not, ftrace_regs is the same as
> pt_regs and ftrace_partial_regs() will return ftrace_regs::regs.
>
> Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
> ---
>  arch/arm64/include/asm/ftrace.h |   11 +++++++++++
>  include/linux/ftrace.h          |   11 +++++++++++
>  2 files changed, 22 insertions(+)
>
> diff --git a/arch/arm64/include/asm/ftrace.h b/arch/arm64/include/asm/ftrace.h
> index ab158196480c..b108cd6718cf 100644
> --- a/arch/arm64/include/asm/ftrace.h
> +++ b/arch/arm64/include/asm/ftrace.h
> @@ -137,6 +137,17 @@ ftrace_override_function_with_return(struct ftrace_regs *fregs)
>         fregs->pc = fregs->lr;
>  }
>
> +static __always_inline struct pt_regs *
> +ftrace_partial_regs(const struct ftrace_regs *fregs, struct pt_regs *regs)
> +{
> +       memcpy(regs->regs, fregs->regs, sizeof(u64) * 10);

Are you intentionally copying that tenth value (fregs.direct_tramp)
into pt_regs.regs[9] ? This seems wrong and it looks like it will bite
us back one day. Isn't it one of these cases where we can simply use
sizeof(fregs->regs) ?

> +       regs->sp = fregs->sp;
> +       regs->pc = fregs->pc;
> +       regs->x[29] = fregs->fp;
> +       regs->x[30] = fregs->lr;
> +       return regs;
> +}
> +
>  int ftrace_regs_query_register_offset(const char *name);
>
>  int ftrace_init_nop(struct module *mod, struct dyn_ftrace *rec);
> diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h
> index 3fb94a1a2461..7f45654441b7 100644
> --- a/include/linux/ftrace.h
> +++ b/include/linux/ftrace.h
> @@ -155,6 +155,17 @@ static __always_inline struct pt_regs *ftrace_get_regs(struct ftrace_regs *fregs
>         return arch_ftrace_get_regs(fregs);
>  }
>
> +#if !defined(CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS) || \
> +       defined(CONFIG_HAVE_PT_REGS_COMPAT_FTRACE_REGS)
> +
> +static __always_inline struct pt_regs *
> +ftrace_partial_regs(const struct ftrace_regs *fregs, struct pt_regs *regs)
> +{
> +       return arch_ftrace_get_regs((struct ftrace_regs *)fregs);
> +}

I don't think this works. Suppose you are on x86, WITH_ARGS, and with
HAVE_PT_REGS_COMPAT_FTRACE_REGS. If you register to ftrace without
FTRACE_OPS_FL_SAVE_REGS you will receive a ftrace_regs from the light
ftrace pre-trampoline that has a CS register equal to 0 and
arch_ftrace_get_regs will return NULL here, which should never happen.

Have you tested your series without registering as FTRACE_OPS_FL_SAVE_REGS ?
  

On Wed, 9 Aug 2023 12:31:27 +0200
Florent Revest <revest@chromium.org> wrote:

> On Mon, Aug 7, 2023 at 8:49 AM Masami Hiramatsu (Google)
> <mhiramat@kernel.org> wrote:
> >
> > From: Masami Hiramatsu (Google) <mhiramat@kernel.org>
> >
> > Add ftrace_partial_regs() which converts the ftrace_regas to pt_regs.
> 
> ftrace_regs*

Oops, thanks.

> 
> > If the architecture defines its own ftrace_regs, this copies partial
> > registers to pt_regs and returns it. If not, ftrace_regs is the same as
> > pt_regs and ftrace_partial_regs() will return ftrace_regs::regs.
> >
> > Signed-off-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
> > ---
> >  arch/arm64/include/asm/ftrace.h |   11 +++++++++++
> >  include/linux/ftrace.h          |   11 +++++++++++
> >  2 files changed, 22 insertions(+)
> >
> > diff --git a/arch/arm64/include/asm/ftrace.h b/arch/arm64/include/asm/ftrace.h
> > index ab158196480c..b108cd6718cf 100644
> > --- a/arch/arm64/include/asm/ftrace.h
> > +++ b/arch/arm64/include/asm/ftrace.h
> > @@ -137,6 +137,17 @@ ftrace_override_function_with_return(struct ftrace_regs *fregs)
> >         fregs->pc = fregs->lr;
> >  }
> >
> > +static __always_inline struct pt_regs *
> > +ftrace_partial_regs(const struct ftrace_regs *fregs, struct pt_regs *regs)
> > +{
> > +       memcpy(regs->regs, fregs->regs, sizeof(u64) * 10);
> 
> Are you intentionally copying that tenth value (fregs.direct_tramp)
> into pt_regs.regs[9] ? This seems wrong and it looks like it will bite
> us back one day. Isn't it one of these cases where we can simply use
> sizeof(fregs->regs) ?

Ah, sorry, it was my mistake. It should be "sizeof(u64) * 9".
I would like to know how can I handle the 'direct_tramp' thing?
Can I just ignore it?

> 
> > +       regs->sp = fregs->sp;
> > +       regs->pc = fregs->pc;
> > +       regs->x[29] = fregs->fp;
> > +       regs->x[30] = fregs->lr;
> > +       return regs;
> > +}
> > +
> >  int ftrace_regs_query_register_offset(const char *name);
> >
> >  int ftrace_init_nop(struct module *mod, struct dyn_ftrace *rec);
> > diff --git a/include/linux/ftrace.h b/include/linux/ftrace.h
> > index 3fb94a1a2461..7f45654441b7 100644
> > --- a/include/linux/ftrace.h
> > +++ b/include/linux/ftrace.h
> > @@ -155,6 +155,17 @@ static __always_inline struct pt_regs *ftrace_get_regs(struct ftrace_regs *fregs
> >         return arch_ftrace_get_regs(fregs);
> >  }
> >
> > +#if !defined(CONFIG_HAVE_DYNAMIC_FTRACE_WITH_ARGS) || \
> > +       defined(CONFIG_HAVE_PT_REGS_COMPAT_FTRACE_REGS)
> > +
> > +static __always_inline struct pt_regs *
> > +ftrace_partial_regs(const struct ftrace_regs *fregs, struct pt_regs *regs)
> > +{
> > +       return arch_ftrace_get_regs((struct ftrace_regs *)fregs);
> > +}
> 
> I don't think this works. Suppose you are on x86, WITH_ARGS, and with
> HAVE_PT_REGS_COMPAT_FTRACE_REGS. If you register to ftrace without
> FTRACE_OPS_FL_SAVE_REGS you will receive a ftrace_regs from the light
> ftrace pre-trampoline that has a CS register equal to 0 and
> arch_ftrace_get_regs will return NULL here, which should never happen.

Yes, Jiri also pointed it. So I simply made it (also remove 'const' from fregs)

return &fregs->regs;

Thank you,

> 
> Have you tested your series without registering as FTRACE_OPS_FL_SAVE_REGS ?