When loading code segment descriptor in iret instruction emulation, the
checks are same as far return instruction emulation, so transfer type
should be X86_TRANSFER_RET in __load_segment_descriptor(). Although,
only iret in real mode is implemented now, and no checks are actually
needed for real mode, it would still be better to mark transfer type as
X86_TRANSFER_RET.
No functional change intended.
Signed-off-by: Hou Wenlong <[email protected]>
---
arch/x86/kvm/emulate.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
index 3b27622d4642..5052eb480068 100644
--- a/arch/x86/kvm/emulate.c
+++ b/arch/x86/kvm/emulate.c
@@ -2100,6 +2100,7 @@ static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
X86_EFLAGS_FIXED;
unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
X86_EFLAGS_VIP;
+ u8 cpl = ctxt->ops->cpl(ctxt);
/* TODO: Add stack limit check */
@@ -2121,7 +2122,8 @@ static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
if (rc != X86EMUL_CONTINUE)
return rc;
- rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
+ rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
+ X86_TRANSFER_RET, NULL);
if (rc != X86EMUL_CONTINUE)
return rc;
--
2.31.1
On Wed, Oct 12, 2022, Hou Wenlong wrote:
> When loading code segment descriptor in iret instruction emulation, the
> checks are same as far return instruction emulation, so transfer type
> should be X86_TRANSFER_RET in __load_segment_descriptor(). Although,
> only iret in real mode is implemented now, and no checks are actually
> needed for real mode, it would still be better to mark transfer type as
> X86_TRANSFER_RET.
It's not strictly a RET though. The RPL vs. DPL checks in __load_segment_descriptor()
might do the right thing, but there's a rather large pile of stuff IRET can do that
RET can't (ignoring the fact that KVM doesn't even emulate FAR RET to outer privilege
levels).
And __emulate_int_real() also loads CS with X86_TRANSFER_NONE, i.e. KVM still has
a weird path to worry about.
Rather than make the IRET case slightly less wrong, what about adding a sanity
check in __load_segment_descriptor() that KVM doesn't attempt to load CS in Protected
Mode with X86_TRANSFER_NONE?
diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
index 3b27622d4642..fe735e18c419 100644
--- a/arch/x86/kvm/emulate.c
+++ b/arch/x86/kvm/emulate.c
@@ -1641,6 +1641,14 @@ static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
goto exception;
break;
case VCPU_SREG_CS:
+ /*
+ * KVM uses "none" when loading CS as part of emulating Real
+ * Mode exceptions and IRET (handled above). In all other
+ * cases, loading CS without a control transfer is a KVM bug.
+ */
+ if (WARN_ON_ONCE(transfer == X86_TRANSFER_NONE))
+ goto exception;
+
if (!(seg_desc.type & 8))
goto exception;
>
> No functional change intended.
>
> Signed-off-by: Hou Wenlong <[email protected]>
> ---
> arch/x86/kvm/emulate.c | 4 +++-
> 1 file changed, 3 insertions(+), 1 deletion(-)
>
> diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
> index 3b27622d4642..5052eb480068 100644
> --- a/arch/x86/kvm/emulate.c
> +++ b/arch/x86/kvm/emulate.c
> @@ -2100,6 +2100,7 @@ static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
> X86_EFLAGS_FIXED;
> unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
> X86_EFLAGS_VIP;
> + u8 cpl = ctxt->ops->cpl(ctxt);
>
> /* TODO: Add stack limit check */
>
> @@ -2121,7 +2122,8 @@ static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
> if (rc != X86EMUL_CONTINUE)
> return rc;
>
> - rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
> + rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
> + X86_TRANSFER_RET, NULL);
>
> if (rc != X86EMUL_CONTINUE)
> return rc;
> --
> 2.31.1
>
On Thu, Oct 13, 2022 at 12:34:57AM +0800, Sean Christopherson wrote:
> On Wed, Oct 12, 2022, Hou Wenlong wrote:
> > When loading code segment descriptor in iret instruction emulation, the
> > checks are same as far return instruction emulation, so transfer type
> > should be X86_TRANSFER_RET in __load_segment_descriptor(). Although,
> > only iret in real mode is implemented now, and no checks are actually
> > needed for real mode, it would still be better to mark transfer type as
> > X86_TRANSFER_RET.
>
> It's not strictly a RET though. The RPL vs. DPL checks in __load_segment_descriptor()
> might do the right thing, but there's a rather large pile of stuff IRET can do that
> RET can't (ignoring the fact that KVM doesn't even emulate FAR RET to outer privilege
> levels).
>
Yes, if EFLAGS.NT is set, IRET performs a task switch instead of far return in Protected
Mode.
> And __emulate_int_real() also loads CS with X86_TRANSFER_NONE, i.e. KVM still has
> a weird path to worry about.
>
> Rather than make the IRET case slightly less wrong, what about adding a sanity
> check in __load_segment_descriptor() that KVM doesn't attempt to load CS in Protected
> Mode with X86_TRANSFER_NONE?
>
OK, sanity check is needed if someone wants to implement IRET in Protected Mode but
use load_segment_descriptor().
> diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
> index 3b27622d4642..fe735e18c419 100644
> --- a/arch/x86/kvm/emulate.c
> +++ b/arch/x86/kvm/emulate.c
> @@ -1641,6 +1641,14 @@ static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
> goto exception;
> break;
> case VCPU_SREG_CS:
> + /*
> + * KVM uses "none" when loading CS as part of emulating Real
> + * Mode exceptions and IRET (handled above). In all other
> + * cases, loading CS without a control transfer is a KVM bug.
> + */
> + if (WARN_ON_ONCE(transfer == X86_TRANSFER_NONE))
> + goto exception;
> +
> if (!(seg_desc.type & 8))
> goto exception;
>
Do I need to prepare this patch or you will add this directly?
> >
> > No functional change intended.
> >
> > Signed-off-by: Hou Wenlong <[email protected]>
> > ---
> > arch/x86/kvm/emulate.c | 4 +++-
> > 1 file changed, 3 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
> > index 3b27622d4642..5052eb480068 100644
> > --- a/arch/x86/kvm/emulate.c
> > +++ b/arch/x86/kvm/emulate.c
> > @@ -2100,6 +2100,7 @@ static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
> > X86_EFLAGS_FIXED;
> > unsigned long vm86_mask = X86_EFLAGS_VM | X86_EFLAGS_VIF |
> > X86_EFLAGS_VIP;
> > + u8 cpl = ctxt->ops->cpl(ctxt);
> >
> > /* TODO: Add stack limit check */
> >
> > @@ -2121,7 +2122,8 @@ static int emulate_iret_real(struct x86_emulate_ctxt *ctxt)
> > if (rc != X86EMUL_CONTINUE)
> > return rc;
> >
> > - rc = load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS);
> > + rc = __load_segment_descriptor(ctxt, (u16)cs, VCPU_SREG_CS, cpl,
> > + X86_TRANSFER_RET, NULL);
> >
> > if (rc != X86EMUL_CONTINUE)
> > return rc;
> > --
> > 2.31.1
> >
On Thu, Oct 13, 2022, Hou Wenlong wrote:
> On Thu, Oct 13, 2022 at 12:34:57AM +0800, Sean Christopherson wrote:
> > diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
> > index 3b27622d4642..fe735e18c419 100644
> > --- a/arch/x86/kvm/emulate.c
> > +++ b/arch/x86/kvm/emulate.c
> > @@ -1641,6 +1641,14 @@ static int __load_segment_descriptor(struct x86_emulate_ctxt *ctxt,
> > goto exception;
> > break;
> > case VCPU_SREG_CS:
> > + /*
> > + * KVM uses "none" when loading CS as part of emulating Real
> > + * Mode exceptions and IRET (handled above). In all other
> > + * cases, loading CS without a control transfer is a KVM bug.
> > + */
> > + if (WARN_ON_ONCE(transfer == X86_TRANSFER_NONE))
> > + goto exception;
> > +
> > if (!(seg_desc.type & 8))
> > goto exception;
> >
> Do I need to prepare this patch or you will add this directly?
No preference. Feel free to post a patch, if not I'll get to it soon-ish.