Re: [PATCH v2 1/4] arm64: kasan: don't populate vmalloc area for CONFIG_KASAN_VMALLOC

[Will Deacon <will@kernel.org>]

On Fri, Feb 05, 2021 at 06:30:44PM +0100, Andrey Konovalov wrote:
> On Fri, Feb 5, 2021 at 6:19 PM Will Deacon <will@kernel.org> wrote:
> >
> > On Fri, Feb 05, 2021 at 12:37:21AM +0800, Lecopzer Chen wrote:
> > >
> > > > On Thu, Feb 04, 2021 at 10:46:12PM +0800, Lecopzer Chen wrote:
> > > > > > On Sat, Jan 09, 2021 at 06:32:49PM +0800, Lecopzer Chen wrote:
> > > > > > > Linux support KAsan for VMALLOC since commit 3c5c3cfb9ef4da9
> > > > > > > ("kasan: support backing vmalloc space with real shadow memory")
> > > > > > >
> > > > > > > Like how the MODULES_VADDR does now, just not to early populate
> > > > > > > the VMALLOC_START between VMALLOC_END.
> > > > > > > similarly, the kernel code mapping is now in the VMALLOC area and
> > > > > > > should keep these area populated.
> > > > > > >
> > > > > > > Signed-off-by: Lecopzer Chen <lecopzer.chen@mediatek.com>
> > > > > > > ---
> > > > > > >  arch/arm64/mm/kasan_init.c | 23 ++++++++++++++++++-----
> > > > > > >  1 file changed, 18 insertions(+), 5 deletions(-)
> > > > > > >
> > > > > > > diff --git a/arch/arm64/mm/kasan_init.c b/arch/arm64/mm/kasan_init.c
> > > > > > > index d8e66c78440e..39b218a64279 100644
> > > > > > > --- a/arch/arm64/mm/kasan_init.c
> > > > > > > +++ b/arch/arm64/mm/kasan_init.c
> > > > > > > @@ -214,6 +214,7 @@ static void __init kasan_init_shadow(void)
> > > > > > >  {
> > > > > > >   u64 kimg_shadow_start, kimg_shadow_end;
> > > > > > >   u64 mod_shadow_start, mod_shadow_end;
> > > > > > > + u64 vmalloc_shadow_start, vmalloc_shadow_end;
> > > > > > >   phys_addr_t pa_start, pa_end;
> > > > > > >   u64 i;
> > > > > > >
> > > > > > > @@ -223,6 +224,9 @@ static void __init kasan_init_shadow(void)
> > > > > > >   mod_shadow_start = (u64)kasan_mem_to_shadow((void *)MODULES_VADDR);
> > > > > > >   mod_shadow_end = (u64)kasan_mem_to_shadow((void *)MODULES_END);
> > > > > > >
> > > > > > > + vmalloc_shadow_start = (u64)kasan_mem_to_shadow((void *)VMALLOC_START);
> > > > > > > + vmalloc_shadow_end = (u64)kasan_mem_to_shadow((void *)VMALLOC_END);
> > > > > > > +
> > > > > > >   /*
> > > > > > >    * We are going to perform proper setup of shadow memory.
> > > > > > >    * At first we should unmap early shadow (clear_pgds() call below).
> > > > > > > @@ -241,12 +245,21 @@ static void __init kasan_init_shadow(void)
> > > > > > >
> > > > > > >   kasan_populate_early_shadow(kasan_mem_to_shadow((void *)PAGE_END),
> > > > > > >                              (void *)mod_shadow_start);
> > > > > > > - kasan_populate_early_shadow((void *)kimg_shadow_end,
> > > > > > > -                            (void *)KASAN_SHADOW_END);
> > > > > > > + if (IS_ENABLED(CONFIG_KASAN_VMALLOC)) {
> > > > > >
> > > > > > Do we really need yet another CONFIG option for KASAN? What's the use-case
> > > > > > for *not* enabling this if you're already enabling one of the KASAN
> > > > > > backends?
> > > > >
> > > > > As I know, KASAN_VMALLOC now only supports KASAN_GENERIC and also
> > > > > KASAN_VMALLOC uses more memory to map real shadow memory (1/8 of vmalloc va).
> > > >
> > > > The shadow is allocated dynamically though, isn't it?
> > >
> > > Yes, but It's still a cost.
> > >
> > > > > There should be someone can enable KASAN_GENERIC but can't use VMALLOC
> > > > > due to memory issue.
> > > >
> > > > That doesn't sound particularly realistic to me. The reason I'm pushing here
> > > > is because I would _really_ like to move to VMAP stack unconditionally, and
> > > > that would effectively force KASAN_VMALLOC to be set if KASAN is in use.
> > > >
> > > > So unless there's a really good reason not to do that, please can we make
> > > > this unconditional for arm64? Pretty please?
> > >
> > > I think it's fine since we have a good reason.
> > > Also if someone have memory issue in KASAN_VMALLOC,
> > > they can use SW_TAG, right?
> > >
> > > However the SW_TAG/HW_TAG is not supported VMALLOC yet.
> > > So the code would be like
> > >
> > >       if (IS_ENABLED(CONFIG_KASAN_GENERIC))
> >
> > Just make this CONFIG_KASAN_VMALLOC, since that depends on KASAN_GENERIC.
> >
> > >               /* explain the relationship between
> > >                * KASAN_GENERIC and KASAN_VMALLOC in arm64
> > >                * XXX: because we want VMAP stack....
> > >                */
> >
> > I don't understand the relation with SW_TAGS. The VMAP_STACK dependency is:
> >
> >         depends on !KASAN || KASAN_HW_TAGS || KASAN_VMALLOC
> 
> This means that VMAP_STACK can be only enabled if KASAN_HW_TAGS=y or
> if KASAN_VMALLOC=y for other modes.
> 
> >
> > which doesn't mention SW_TAGS at all. So that seems to imply that SW_TAGS
> > and VMAP_STACK are mutually exclusive :(
> 
> SW_TAGS doesn't yet have vmalloc support, so it's not compatible with
> VMAP_STACK. Once vmalloc support is added to SW_TAGS, KASAN_VMALLOC
> should be allowed to be enabled with SW_TAGS. This series is a step
> towards having that support, but doesn't implement it. That will be a
> separate effort.

Ok, thanks. Then I think we should try to invert the dependency here, if
possible, so that the KASAN backends depend on !VMAP_STACK if they don't
support it, rather than silently disabling VMAP_STACK when they are
selected.

Will