Re: [PATCH RFC v2 07/21] kasan, arm64: move initialization message

[Andrey Konovalov]

On Wed, Oct 28, 2020 at 11:56 AM Dmitry Vyukov <dvyukov@xxxxxxxxxx> wrote:
>
> On Thu, Oct 22, 2020 at 3:19 PM Andrey Konovalov <andreyknvl@xxxxxxxxxx> wrote:
> >
> > Tag-based KASAN modes are fully initialized with kasan_init_tags(),
> > while the generic mode only requireds kasan_init(). Move the
> > initialization message for tag-based modes into kasan_init_tags().
> >
> > Also fix pr_fmt() usage for KASAN code: generic mode doesn't need it,
>
> Why doesn't it need it? What's the difference with tag modes?

I need to reword the patch descriptions: it's not the mode that
doesn't need it, it's the generic.c file, as it doesn't use any pr_*()
functions.

>
> > tag-based modes should use "kasan:" instead of KBUILD_MODNAME.
>
> With generic KASAN I currently see:
>
> [    0.571473][    T0] kasan: KernelAddressSanitizer initialized
>
> So KBUILD_MODNAME somehow works. Is there some difference between files?

That code is printed from arch/xxx/mm/kasan_init*.c, which has its own
pr_fmt defined.

>
> > Signed-off-by: Andrey Konovalov <andreyknvl@xxxxxxxxxx>
> > Link: https://linux-review.googlesource.com/id/Idfd1e50625ffdf42dfc3dbf7455b11bd200a0a49
> > ---
> >  arch/arm64/mm/kasan_init.c | 3 +++
> >  mm/kasan/generic.c         | 2 --
> >  mm/kasan/hw_tags.c         | 4 ++++
> >  mm/kasan/sw_tags.c         | 4 +++-
> >  4 files changed, 10 insertions(+), 3 deletions(-)
> >
> > diff --git a/arch/arm64/mm/kasan_init.c b/arch/arm64/mm/kasan_init.c
> > index b6b9d55bb72e..8f17fa834b62 100644
> > --- a/arch/arm64/mm/kasan_init.c
> > +++ b/arch/arm64/mm/kasan_init.c
> > @@ -290,5 +290,8 @@ void __init kasan_init(void)
> >  {
> >         kasan_init_shadow();
> >         kasan_init_depth();
> > +#if defined(CONFIG_KASAN_GENERIC)
> > +       /* CONFIG_KASAN_SW/HW_TAGS also requires kasan_init_tags(). */
>
> A bit cleaner way may be to introduce kasan_init_early() and
> kasan_init_late(). Late() will do tag init and always print the
> message.

It appears we'll also need kasan_init_even_later() for some
MTE-related stuff. I'll try to figure out some sane naming scheme here
and include it into the next version.