by Borislav Petkov

[permalink] [raw]

Subject: Re: [PATCH v3 23/75] x86/boot/compressed/64: Setup GHCB Based VC Exception handler

On Tue, May 12, 2020 at 11:08:12PM +0200, Joerg Roedel wrote:
> No, this is a recent addition, otherwise this breaks out-of-tree builds
> (make O=/some/path ...) because inat-tables.c (included from inat.c) is
> generated during build and ends up in the $(objtree).

Please add a blurb about this above it otherwise no one would have a
clue why it is there.

> Because the immediate is the last part of the instruction which is
> decoded (even if there is no immediate). The .got field is set when
> either the immediate was decoded successfully or, in case the
> instruction has no immediate, when the rest of the instruction was
> decoded successfully. So testing immediate.got is the indicator whether
> decoding was successful.

Hm, whether the immediate was parsed correctly or it wasn't there and
using that as the sign whether the instruction was decoded successfully
sounds kinda arbitrary.

@Masami: shouldn't that insn_get_length() thing return a value to denote
whether the decode was successful or not instead of testing arbitrary
fields?

Pasting for you the code sequence again:

...
insn_get_length(&ctxt->insn);

ret = ctxt->insn.immediate.got ? ES_OK : ES_DECODE_FAILED;
...

I wonder if one would be able to do instead:

if (insn_get_length(&ctxt->insn))
return ES_OK;

return ES_DECODE_FAILED;

to have it straight-forward...

> Probably better rename the function, it also does the sev-es protocol
> version negotiation and all other related setup tasks. Maybe
> sev_es_setup() is a better name?

Sure.

> If the above call fails its the end of the systems lifetime, because we
> can't continue to boot an SEV-ES guest when we have no GHCB.
>
> BUG_ON() and friends are also not available in the pre-decompression
> boot stage.

Oh ok, error() does hlt the system.

Thx.

--
Regards/Gruss,
Boris.

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

2020-06-03 10:11:30

by Joerg Roedel

[permalink] [raw]

Subject: Re: [PATCH v3 23/75] x86/boot/compressed/64: Setup GHCB Based VC Exception handler

On Mon, May 11, 2020 at 10:07:09PM +0200, Borislav Petkov wrote:
> On Tue, Apr 28, 2020 at 05:16:33PM +0200, Joerg Roedel wrote:
> > @@ -63,3 +175,45 @@ void __init do_vc_no_ghcb(struct pt_regs *regs, unsigned long exit_code)
> > while (true)
> > asm volatile("hlt\n");
> > }
> > +
> > +static enum es_result vc_insn_string_read(struct es_em_ctxt *ctxt,
> > + void *src, char *buf,
> > + unsigned int data_size,
> > + unsigned int count,
> > + bool backwards)
> > +{
> > + int i, b = backwards ? -1 : 1;
> > + enum es_result ret = ES_OK;
> > +
> > + for (i = 0; i < count; i++) {
> > + void *s = src + (i * data_size * b);
> > + char *d = buf + (i * data_size);
>
> >From a previous review:
>
> Where are we checking whether that count is not exceeding @buf or
> similar discrepancies?

These two functions are only called from vc_handle_ioio() and buf always
points to ghcb->shared_buffer.

In general, the caller has to make sure that sizeof(*buf) is at least
data_size*count, and handle_ioio() calculates count based on the size of
*buf.

Joerg