[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] [PATCH v3 10/24] x86/emul: Always use fault semantics for software events



The common case is already using fault semantics out of x86_emulate(), as that
is how VT-x/SVM expects to inject the event (given suitable hardware support).

However, x86_emulate() returning X86EMUL_EXCEPTION and also completing a
register writeback is problematic for callers.

Switch the logic to always using fault semantics, and leave svm_inject_trap()
to fix up %eip if necessary.

Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
---
CC: Jan Beulich <JBeulich@xxxxxxxx>
CC: Tim Deegan <tim@xxxxxxx>
CC: Boris Ostrovsky <boris.ostrovsky@xxxxxxxxxx>
CC: Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx>

v3:
 * New
---
 xen/arch/x86/hvm/svm/svm.c             | 44 ++++++++++++++++++++--------------
 xen/arch/x86/x86_emulate/x86_emulate.c |  2 --
 xen/arch/x86/x86_emulate/x86_emulate.h |  8 ++++++-
 3 files changed, 33 insertions(+), 21 deletions(-)

diff --git a/xen/arch/x86/hvm/svm/svm.c b/xen/arch/x86/hvm/svm/svm.c
index 912d871..65eeab7 100644
--- a/xen/arch/x86/hvm/svm/svm.c
+++ b/xen/arch/x86/hvm/svm/svm.c
@@ -1209,7 +1209,7 @@ static void svm_inject_event(const struct x86_event 
*event)
     struct vmcb_struct *vmcb = curr->arch.hvm_svm.vmcb;
     eventinj_t eventinj = vmcb->eventinj;
     struct x86_event _event = *event;
-    const struct cpu_user_regs *regs = guest_cpu_user_regs();
+    struct cpu_user_regs *regs = guest_cpu_user_regs();
 
     switch ( _event.vector )
     {
@@ -1242,27 +1242,38 @@ static void svm_inject_event(const struct x86_event 
*event)
     eventinj.fields.v = 1;
     eventinj.fields.vector = _event.vector;
 
-    /* Refer to AMD Vol 2: System Programming, 15.20 Event Injection. */
+    /*
+     * Refer to AMD Vol 2: System Programming, 15.20 Event Injection.
+     *
+     * On hardware lacking NextRIP support, and all hardware in the case of
+     * icebp, software events with trap semantics need emulating, so %eip in
+     * the trap frame points after the instruction.
+     *
+     * The x86 emulator (if requested by the x86_swint_emulate_* choice) will
+     * have performed checks such as presence/dpl/etc and believes that the
+     * event injection will succeed without faulting.
+     *
+     * The x86 emulator will always provide fault semantics for software
+     * events, with _trap.insn_len set appropriately.  If the injection
+     * requires emulation, move %eip forwards at this point.
+     */
     switch ( _event.type )
     {
     case X86_EVENTTYPE_SW_INTERRUPT: /* int $n */
-        /*
-         * Software interrupts (type 4) cannot be properly injected if the
-         * processor doesn't support NextRIP.  Without NextRIP, the emulator
-         * will have performed DPL and presence checks for us, and will have
-         * moved eip forward if appropriate.
-         */
         if ( cpu_has_svm_nrips )
             vmcb->nextrip = regs->eip + _event.insn_len;
+        else
+            regs->eip += _event.insn_len;
         eventinj.fields.type = X86_EVENTTYPE_SW_INTERRUPT;
         break;
 
     case X86_EVENTTYPE_PRI_SW_EXCEPTION: /* icebp */
         /*
-         * icebp's injection must always be emulated.  Software injection help
-         * in x86_emulate has moved eip forward, but NextRIP (if used) still
-         * needs setting or execution will resume from 0.
+         * icebp's injection must always be emulated, as hardware does not
+         * special case HW_EXCEPTION with vector 1 (#DB) as having trap
+         * semantics.
          */
+        regs->eip += _event.insn_len;
         if ( cpu_has_svm_nrips )
             vmcb->nextrip = regs->eip;
         eventinj.fields.type = X86_EVENTTYPE_HW_EXCEPTION;
@@ -1270,16 +1281,13 @@ static void svm_inject_event(const struct x86_event 
*event)
 
     case X86_EVENTTYPE_SW_EXCEPTION: /* int3, into */
         /*
-         * The AMD manual states that .type=3 (HW exception), .vector=3 or 4,
-         * will perform DPL checks.  Experimentally, DPL and presence checks
-         * are indeed performed, even without NextRIP support.
-         *
-         * However without NextRIP support, the event injection still needs
-         * fully emulating to get the correct eip in the trap frame, yet get
-         * the correct faulting eip should a fault occur.
+         * Hardware special cases HW_EXCEPTION with vectors 3 and 4 as having
+         * trap semantics, and will perform DPL checks.
          */
         if ( cpu_has_svm_nrips )
             vmcb->nextrip = regs->eip + _event.insn_len;
+        else
+            regs->eip += _event.insn_len;
         eventinj.fields.type = X86_EVENTTYPE_HW_EXCEPTION;
         break;
 
diff --git a/xen/arch/x86/x86_emulate/x86_emulate.c 
b/xen/arch/x86/x86_emulate/x86_emulate.c
index e4643a3..8a1f1f5 100644
--- a/xen/arch/x86/x86_emulate/x86_emulate.c
+++ b/xen/arch/x86/x86_emulate/x86_emulate.c
@@ -1694,8 +1694,6 @@ static int inject_swint(enum x86_swint_type type,
                     goto raise_exn;
             }
         }
-
-        ctxt->regs->eip += insn_len;
     }
 
     rc = ops->inject_sw_interrupt(type, vector, insn_len, ctxt);
diff --git a/xen/arch/x86/x86_emulate/x86_emulate.h 
b/xen/arch/x86/x86_emulate/x86_emulate.h
index f84ced2..fc28976 100644
--- a/xen/arch/x86/x86_emulate/x86_emulate.h
+++ b/xen/arch/x86/x86_emulate/x86_emulate.h
@@ -64,7 +64,13 @@ enum x86_swint_type {
     x86_swint_int,   /* 0xcd $n */
 };
 
-/* How much help is required with software event injection? */
+/*
+ * How much help is required with software event injection?
+ *
+ * All software events return from x86_emulate() with X86EMUL_EXCEPTION and
+ * fault-like semantics.  This just controls whether the emulator performs
+ * presence/dpl/etc checks and possibly raises excepions instead.
+ */
 enum x86_swint_emulation {
     x86_swint_emulate_none, /* Hardware supports all software injection 
properly */
     x86_swint_emulate_icebp,/* Help needed with `icebp` (0xf1) */
-- 
2.1.4


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.