linux-kernel - Re: [PATCH] x86/sev-es: Fix SEV-ES OUT/IN immediate opcode vc handling

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20201228223650.GD20321@zn.tnic>
Date:   Mon, 28 Dec 2020 23:36:50 +0100
From:   Borislav Petkov <bp@...en8.de>
To:     Sean Christopherson <seanjc@...gle.com>,
        Peter Gonda <pgonda@...gle.com>
Cc:     Thomas Gleixner <tglx@...utronix.de>,
        Ingo Molnar <mingo@...hat.com>, Joerg Roedel <jroedel@...e.de>,
        Tom Lendacky <thomas.lendacky@....com>,
        linux-kernel@...r.kernel.org, x86@...nel.org,
        David Rientjes <rientjes@...gle.com>
Subject: Re: [PATCH] x86/sev-es: Fix SEV-ES OUT/IN immediate opcode vc
 handling

On Thu, Dec 17, 2020 at 09:19:13AM -0800, Sean Christopherson wrote:
> On Wed, Dec 16, 2020, Peter Gonda wrote:
> > 
> > The IN and OUT immediate instructions only use an 8-bit immediate. The
> > current VC handler uses the entire 32-bit immediate value. These
> > instructions only set the first bytes.
> > 
> > Tested with a loop back port with "outb %0,$0xe0". Before the port seen
> > by KVM was 0xffffffffffffffe0 instead of 0xe0. After the correct port
> > was seen by KVM and the guests loop back OUT then IN were equal.
> > 
> > 
> > Signed-off-by: Peter Gonda <pgonda@...gle.com>
> > Acked-by: David Rientjes <rientjes@...gle.com>
> > 
> > 
> > ---
> >  arch/x86/kernel/sev-es-shared.c | 8 ++++++--
> >  drivers/Makefile                | 1 +
> >  2 files changed, 7 insertions(+), 2 deletions(-)
> > 
> > diff --git a/arch/x86/kernel/sev-es-shared.c b/arch/x86/kernel/sev-es-shared.c
> > index 7d04b356d44d..6c790377c55c 100644
> > --- a/arch/x86/kernel/sev-es-shared.c
> > +++ b/arch/x86/kernel/sev-es-shared.c
> > @@ -305,14 +305,14 @@ static enum es_result vc_ioio_exitinfo(struct es_em_ctxt *ctxt, u64 *exitinfo)
> >  	case 0xe4:
> >  	case 0xe5:
> >  		*exitinfo |= IOIO_TYPE_IN;
> > -		*exitinfo |= (u64)insn->immediate.value << 16;
> > +		*exitinfo |= insn->immediate.bytes[0] << 16;
> 
> Can't we just drop the explicit cast to u64?  Or explicitly cast to u8?  Doesn't
> really matter, but poking into the backing bytes feels a bit backwards.

GHCB spec says for IOIO_PROT (exit code 0x7b):

"SW_EXITINFO1 will be set as documented in AMD64 Architecture
Programmer’s Manual Volume 2: System Programming, Section 15.10.2"

That section has "Figure 15-2. EXITINFO1 for IOIO Intercept" where
[31:16] is the intercepted I/O port.

Now, insn->immediate.value is a union between a signed int and a
unsigned char bytes[4] array so in order to be absolutely correct, that
should use:

	insn->immediate.bytes[0]

instead, to get the port byte only, before shifting it into [31:16]. The
(u64) cast doesn't matter because *exitinfo = 0 at function entry and
that is part of the first assignment afterwards.

I still don't see how that sign extension 0xffffffffffffffe0 can happen,
though. Maybe something gets added during instruction decoding...

Hmmm.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette