Received-SPF: pass (google.com: domain of linux-kernel+bounces-50682-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:4601:e00::3 as permitted sender) client-ip=2604:1380:4601:e00::3;
Precedence: bulk
MIME-Version: 1.0
References: <20240121182040.GBZa1geI5NxWSslvt0@fat_crate.local>
 <20240130220845.1978329-2-kevinloughlin@google.com> <4ljfvg7c23g2wx4hcqz6x6tx7uvp6fffayqvu4ptsmfqajlujr@zngokbd2awma>
 <CAGdbjm+dt2LMqsWkzME-Gya9F=pZMubTuW3Ad3dStj=L+MW+ng@mail.gmail.com>
In-Reply-To: <CAGdbjm+dt2LMqsWkzME-Gya9F=pZMubTuW3Ad3dStj=L+MW+ng@mail.gmail.com>
From: Ard Biesheuvel <ardb@kernel.org>
Date: Fri, 2 Feb 2024 23:47:24 +0100
Message-ID: <CAMj1kXGSZkJ-1opx9mrpfhzyreCM+P616xM+MhPOaR+EgP5NNg@mail.gmail.com>
Subject: Re: [PATCH v3 1/2] x86/sev: enforce RIP-relative accesses in early
 SEV/SME code
To: Kevin Loughlin <kevinloughlin@google.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>, Thomas Gleixner <tglx@linutronix.de>, 
	Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>, 
	Dave Hansen <dave.hansen@linux.intel.com>, x86@kernel.org, 
	Nick Desaulniers <ndesaulniers@google.com>, Justin Stitt <justinstitt@google.com>, 
	Tom Lendacky <thomas.lendacky@amd.com>, Pankaj Gupta <pankaj.gupta@amd.com>, 
	Hou Wenlong <houwenlong.hwl@antgroup.com>, Dionna Glaze <dionnaglaze@google.com>, 
	Brijesh Singh <brijesh.singh@amd.com>, Michael Roth <michael.roth@amd.com>, 
	linux-kernel@vger.kernel.org, llvm@lists.linux.dev, 
	linux-coco@lists.linux.dev, Ashish Kalra <ashish.kalra@amd.com>, 
	Andi Kleen <ak@linux.intel.com>, Adam Dunlap <acdunlap@google.com>, 
	Peter Gonda <pgonda@google.com>, Jacob Xu <jacobhxu@google.com>, 
	Sidharth Telang <sidtelang@google.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Fri, 2 Feb 2024 at 23:00, Kevin Loughlin <kevinloughlin@google.com> wrot=
e:
>
> On Wed, Jan 31, 2024 at 12:20=E2=80=AFAM Kirill A. Shutemov
> <kirill.shutemov@linux.intel.com> wrote:
> >
> > On Tue, Jan 30, 2024 at 10:08:44PM +0000, Kevin Loughlin wrote:
> > > diff --git a/arch/x86/include/asm/mem_encrypt.h b/arch/x86/include/as=
m/mem_encrypt.h
> > > index 359ada486fa9..b65e66ee79c4 100644
> > > --- a/arch/x86/include/asm/mem_encrypt.h
> > > +++ b/arch/x86/include/asm/mem_encrypt.h
> > > @@ -17,6 +17,20 @@
> > >
> > >  #include <asm/bootparam.h>
> > >
> > > +/*
> > > + * Like the address operator "&", evaluates to the address of a LHS =
variable
> > > + * "var", but also enforces the use of RIP-relative logic. This macr=
o can be
> > > + * used to safely access global data variables prior to kernel reloc=
ation.
> > > + */
> > > +#define RIP_RELATIVE_ADDR(var)               \
> > > +({                                   \
> > > +     void *rip_rel_ptr;              \
> > > +     asm ("lea "#var"(%%rip), %0"    \
> > > +             : "=3Dr" (rip_rel_ptr)    \
> > > +             : "p" (&var));          \
> > > +     rip_rel_ptr;                    \
> > > +})
> > > +
> >
> > I don't think it is the right place for the macro. The next patch uses =
for
> > things unrelated to memory encryption.
>
> You're right; with the cleanup, I agree it becomes more general. We
> can move it to arch/x86/include/asm/asm.h.
>
> >
> > > @@ -239,14 +244,14 @@ unsigned long __head __startup_64(unsigned long=
 physaddr,
> > >        */
> > >
> > >       next_pgt_ptr =3D fixup_pointer(&next_early_pgt, physaddr);
> > > -     pud =3D fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++], ph=
ysaddr);
> > > -     pmd =3D fixup_pointer(early_dynamic_pgts[(*next_pgt_ptr)++], ph=
ysaddr);
> > > +     early_dynamic_pgts_ptr =3D fixup_pointer(early_dynamic_pgts, ph=
ysaddr);
> > > +     pud =3D (pudval_t *) early_dynamic_pgts_ptr[(*next_pgt_ptr)++];
> > > +     pmd =3D (pmdval_t *) early_dynamic_pgts_ptr[(*next_pgt_ptr)++];
> > >
> >
> > This change doesn't belong to this patch. Maybe move it into the next
> > patch and combine with removing fixup_pointer().
>
> I'll put it in a separate commit even preceding this one, as it's
> actually a bug in the existing fixup pointer logic that I noticed when
> transitioning to the use of the RIP-relative macro. Specifically,
> early_dynamic_pgts is a global variable just like next_early_pgt and
> thus also needs to be fixed up to guarantee the correct address is
> used across toolchains.

No, this is not a bug.

early_dynamic_pgts is a two dimensional array

extern pmd_t early_dynamic_pgts[EARLY_DYNAMIC_PAGE_TABLES][PTRS_PER_PMD];

and so this expression

early_dynamic_pgts[(*next_pgt_ptr)++]

produces the kernel virtual address of one of the top level elements,
which itself is an array, and so this can be fixed up as usual.
IOW, the [] operation does not result in pointer dereference here,
only in pointer arithmetic.