[PATCH v2 13/19] x86/shadow: Avoid raising faults behind the emulators back

[Andrew Cooper <andrew.cooper3@citrix.com>]

Use x86_emul_{hw_exception,pagefault}() rather than
{pv,hvm}_inject_page_fault() and hvm_inject_hw_exception() to cause raised
faults to be known to the emulator.  This requires altering the callers of
x86_emulate() to properly re-inject the event.

While fixing this, fix the singlestep behaviour.  Previously, an otherwise
successful emulation would fail if singlestepping was active, as the emulator
couldn't raise #DB.  This is unreasonable from the point of view of the guest.

We therefore tolerate #PF/#GP/SS and #DB being raised by the emulator, but
reject anything else as unexpected.

Signed-off-by: Andrew Cooper <andrew.cooper3@citrix.com>
---
CC: Jan Beulich <JBeulich@suse.com>
CC: Tim Deegan <tim@xen.org>

v2:
 * New
---
 xen/arch/x86/mm/shadow/common.c | 13 ++++-----
 xen/arch/x86/mm/shadow/multi.c  | 61 ++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 66 insertions(+), 8 deletions(-)

diff --git a/xen/arch/x86/mm/shadow/common.c b/xen/arch/x86/mm/shadow/common.c
index f07803b..e509cc1 100644
--- a/xen/arch/x86/mm/shadow/common.c
+++ b/xen/arch/x86/mm/shadow/common.c
@@ -162,8 +162,9 @@ static int hvm_translate_linear_addr(
 
     if ( !okay )
     {
-        hvm_inject_hw_exception(
-            (seg == x86_seg_ss) ? TRAP_stack_error : TRAP_gp_fault, 0);
+        x86_emul_hw_exception(
+            (seg == x86_seg_ss) ? TRAP_stack_error : TRAP_gp_fault,
+            0, &sh_ctxt->ctxt);
         return X86EMUL_EXCEPTION;
     }
 
@@ -323,7 +324,7 @@ pv_emulate_read(enum x86_segment seg,
 
     if ( (rc = copy_from_user(p_data, (void *)offset, bytes)) != 0 )
     {
-        pv_inject_page_fault(0, offset + bytes - rc); /* Read fault. */
+        x86_emul_pagefault(0, offset + bytes - rc, ctxt); /* Read fault. */
         return X86EMUL_EXCEPTION;
     }
 
@@ -1720,10 +1721,8 @@ static mfn_t emulate_gva_to_mfn(struct vcpu *v, unsigned long vaddr,
     gfn = paging_get_hostmode(v)->gva_to_gfn(v, NULL, vaddr, &pfec);
     if ( gfn == gfn_x(INVALID_GFN) )
     {
-        if ( is_hvm_vcpu(v) )
-            hvm_inject_page_fault(pfec, vaddr);
-        else
-            pv_inject_page_fault(pfec, vaddr);
+        x86_emul_pagefault(pfec, vaddr, &sh_ctxt->ctxt);
+
         return _mfn(BAD_GVA_TO_GFN);
     }
 
diff --git a/xen/arch/x86/mm/shadow/multi.c b/xen/arch/x86/mm/shadow/multi.c
index 13fa1bf..50705a0 100644
--- a/xen/arch/x86/mm/shadow/multi.c
+++ b/xen/arch/x86/mm/shadow/multi.c
@@ -3390,7 +3390,7 @@ static int sh_page_fault(struct vcpu *v,
      * would be a good unshadow hint. If we *do* decide to unshadow-on-fault
      * then it must be 'failable': we cannot require the unshadow to succeed.
      */
-    if ( r == X86EMUL_UNHANDLEABLE || emul_ctxt.ctxt.event_pending )
+    if ( r == X86EMUL_UNHANDLEABLE )
     {
         perfc_incr(shadow_fault_emulate_failed);
 #if SHADOW_OPTIMIZATIONS & SHOPT_FAST_EMULATION
@@ -3434,6 +3434,34 @@ static int sh_page_fault(struct vcpu *v,
         v->arch.paging.last_write_emul_ok = 0;
 #endif
 
+    if ( r == X86EMUL_EXCEPTION && emul_ctxt.ctxt.event_pending )
+    {
+        /*
+         * This emulation covers writes to shadow pagetables.  We tolerate #PF
+         * (from hitting adjacent pages), #GP/#SS (from segmentation errors),
+         * and #DB (from singlestepping).  Anything else is an emulation bug,
+         * or a guest playing with the instruction stream under Xen's feet.
+         */
+        if ( emul_ctxt.ctxt.event.type == X86_EVENTTYPE_HW_EXCEPTION &&
+             (emul_ctxt.ctxt.event.vector < 32) &&
+             ((1u << emul_ctxt.ctxt.event.vector) &
+              ((1u << TRAP_debug) | (1u << TRAP_stack_error) |
+               (1u << TRAP_gp_fault) | (1u << TRAP_page_fault))) )
+        {
+            if ( is_hvm_vcpu(v) )
+                hvm_inject_event(&emul_ctxt.ctxt.event);
+            else
+                pv_inject_event(&emul_ctxt.ctxt.event);
+        }
+        else
+        {
+            if ( is_hvm_vcpu(v) )
+                hvm_inject_hw_exception(TRAP_gp_fault, 0);
+            else
+                pv_inject_hw_exception(TRAP_gp_fault, 0);
+        }
+    }
+
 #if GUEST_PAGING_LEVELS == 3 /* PAE guest */
     if ( r == X86EMUL_OKAY ) {
         int i, emulation_count=0;
@@ -3475,6 +3503,37 @@ static int sh_page_fault(struct vcpu *v,
             {
                 perfc_incr(shadow_em_ex_fail);
                 TRACE_SHADOW_PATH_FLAG(TRCE_SFLAG_EMULATION_LAST_FAILED);
+
+                if ( r == X86EMUL_EXCEPTION && emul_ctxt.ctxt.event_pending )
+                {
+                    /*
+                     * This emulation covers writes to shadow pagetables.  We
+                     * tolerate #PF (from hitting adjacent pages), #GP/#SS
+                     * (from segmentation errors), and #DB (from
+                     * singlestepping).  Anything else is an emulation bug, or
+                     * a guest playing with the instruction stream under Xen's
+                     * feet.
+                     */
+                    if ( emul_ctxt.ctxt.event.type == X86_EVENTTYPE_HW_EXCEPTION &&
+                         (emul_ctxt.ctxt.event.vector < 32) &&
+                         ((1u << emul_ctxt.ctxt.event.vector) &
+                          ((1u << TRAP_debug) | (1u << TRAP_stack_error) |
+                           (1u << TRAP_gp_fault) | (1u << TRAP_page_fault))) )
+                    {
+                        if ( is_hvm_vcpu(v) )
+                            hvm_inject_event(&emul_ctxt.ctxt.event);
+                        else
+                            pv_inject_event(&emul_ctxt.ctxt.event);
+                    }
+                    else
+                    {
+                        if ( is_hvm_vcpu(v) )
+                            hvm_inject_hw_exception(TRAP_gp_fault, 0);
+                        else
+                            pv_inject_hw_exception(TRAP_gp_fault, 0);
+                    }
+                }
+
                 break; /* Don't emulate again if we failed! */
             }
         }
-- 
2.1.4


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