Skip to contentSkip to conversation

x86: adjust reserved bit page fault handling

PATCH In review Jan Beulich 3 participants 5 messages Feb 1, 2008

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 x86: adjust reserved bit page fault handling

One could even debate whether reserved bit faults are always fatal (and should never be propagated to the guest)...

xen/arch/x86/traps.c +19 -2
Index: 2008-01-28/xen/arch/x86/traps.c
===================================================================
@@ -823,6 +823,17 @@ asmlinkage void do_machine_check(struct 
     machine_check_vector(regs, regs->error_code);
 }
 
+static inline void reserved_bit_page_fault(int guest, unsigned long addr,
+                                           struct cpu_user_regs *regs)
+{
+    if ( guest )
+        gdprintk(XENLOG_ERR, "reserved bit in page table entry: ");
+    else
+        dprintk(XENLOG_ERR, "reserved bit in page table entry: ");
+    show_page_walk(addr);
+    show_execution_state(regs);
+}
+
 void propagate_page_fault(unsigned long addr, u16 error_code)
 {
     struct trap_info *ti;
@@ -852,6 +863,8 @@ void propagate_page_fault(unsigned long 
                v->domain->domain_id, v->vcpu_id, error_code);
         show_page_walk(addr);
     }
+    else if ( unlikely(error_code & PFEC_reserved_bit) )
+        reserved_bit_page_fault(1, addr, guest_cpu_user_regs());
 }
 
 static int handle_gdt_ldt_mapping_fault(
@@ -1034,8 +1047,10 @@ static int fixup_page_fault(unsigned lon
     struct vcpu   *v = current;
     struct domain *d = v->domain;
 
-    /* No fixups in interrupt context or when interrupts are disabled. */
-    if ( in_irq() || !(regs->eflags & X86_EFLAGS_IF) )
+    /* No fixups in interrupt context, when interrupts are disabled, or
+     * when a reserved bit was found to be set. */
+    if ( in_irq() || !(regs->eflags & X86_EFLAGS_IF) ||
+         (regs->error_code & PFEC_reserved_bit) )
         return 0;
 
     if ( unlikely(IN_HYPERVISOR_RANGE(addr)) )
@@ -1101,6 +1116,8 @@ asmlinkage void do_page_fault(struct cpu
         if ( likely((fixup = search_exception_table(regs->eip)) != 0) )
         {
             perfc_incr(copy_user_faults);
+            if ( unlikely(regs->error_code & PFEC_reserved_bit) )
+                reserved_bit_page_fault(0, addr, regs);
             regs->eip = fixup;
             return;
         }
#1 lore raw

xen/arch/x86/traps.c static int fixup_page_fault(unsigned lon · line 1056

Quoted 13 lines · Jan Beulich
xen/arch/x86/traps.c +4 -2
@@ -1034,8 +1047,10 @@ static int fixup_page_fault(unsigned lon
     struct vcpu   *v = current;
     struct domain *d = v->domain;
 
-    /* No fixups in interrupt context or when interrupts are disabled. */
-    if ( in_irq() || !(regs->eflags & X86_EFLAGS_IF) )
+    /* No fixups in interrupt context, when interrupts are disabled, or
+     * when a reserved bit was found to be set. */
+    if ( in_irq() || !(regs->eflags & X86_EFLAGS_IF) ||
+         (regs->error_code & PFEC_reserved_bit) )
         return 0;
 
     if ( unlikely(IN_HYPERVISOR_RANGE(addr)) )

The shadow pagetable code deliberately introduces invalid pagetable entries as part of its fast-path treatment of MMIO and not-present entries in the guest tables, so paging_fault needs to be called for PFEC_reserved_bit faults.

Cheers,

Tim.

Signature · 3 lines
Tim Deegan <[email protected]>
Principal Software Engineer, Citrix Systems (R&D) Ltd.
[Company #02300071, SL9 0DZ, UK.]
#2 lore raw
Quoted 14 lines
Quoted 1 line
Quoted 1 line

Tim Deegan <[email protected]> 01.02.08 10:28 >>>

Quoted 12 lines · Jan Beulich
(quoted diff) +4 -2
@@ -1034,8 +1047,10 @@ static int fixup_page_fault(unsigned lon
     struct vcpu   *v = current;
     struct domain *d = v->domain;
 
-    /* No fixups in interrupt context or when interrupts are disabled. */
-    if ( in_irq() || !(regs->eflags & X86_EFLAGS_IF) )
+    /* No fixups in interrupt context, when interrupts are disabled, or
+     * when a reserved bit was found to be set. */
+    if ( in_irq() || !(regs->eflags & X86_EFLAGS_IF) ||
+         (regs->error_code & PFEC_reserved_bit) )
         return 0;
 
Quoted 6 lines

Review static int fixup_page_fault(unsigned lon · line 1056

Quoted 1 line
(quoted diff) +0 -0
     if ( unlikely(IN_HYPERVISOR_RANGE(addr)) )

The shadow pagetable code deliberately introduces invalid pagetable entries as part of its fast-path treatment of MMIO and not-present entries in the guest tables, so paging_fault needs to be called for PFEC_reserved_bit faults.

Here's a replacement patch then, it would still be questionable whether reserved bit faults should ever be propagated to a guest...

xen/arch/x86/traps.c +20 -2
Index: 2008-01-28/xen/arch/x86/traps.c
===================================================================
@@ -823,6 +823,18 @@ asmlinkage void do_machine_check(struct 
     machine_check_vector(regs, regs->error_code);
 }
 
+static inline void reserved_bit_page_fault(int guest, unsigned long addr,
+                                           struct cpu_user_regs *regs)
+{
+    if ( guest )
+        gdprintk(XENLOG_ERR, "");
+    else
+        dprintk(XENLOG_ERR, "");
+    printk("reserved bit in page table (ec=%04X): ", regs->error_code);
+    show_page_walk(addr);
+    show_execution_state(regs);
+}
+
 void propagate_page_fault(unsigned long addr, u16 error_code)
 {
     struct trap_info *ti;
@@ -852,6 +864,8 @@ void propagate_page_fault(unsigned long 
                v->domain->domain_id, v->vcpu_id, error_code);
         show_page_walk(addr);
     }
+    else if ( unlikely(error_code & PFEC_reserved_bit) )
+        reserved_bit_page_fault(1, addr, guest_cpu_user_regs());
 }
 
 static int handle_gdt_ldt_mapping_fault(
@@ -1047,7 +1061,8 @@ static int fixup_page_fault(unsigned lon
                 trace_trap_two_addr(TRC_PV_PAGING_FIXUP, regs->eip, addr);
             return ret;
         }
-        if ( (addr >= GDT_LDT_VIRT_START) && (addr < GDT_LDT_VIRT_END) )
+        if ( !(regs->error_code & PFEC_reserved_bit) &&
+             (addr >= GDT_LDT_VIRT_START) && (addr < GDT_LDT_VIRT_END) )
             return handle_gdt_ldt_mapping_fault(
                 addr - GDT_LDT_VIRT_START, regs);
         return 0;
@@ -1057,7 +1072,8 @@ static int fixup_page_fault(unsigned lon
          guest_kernel_mode(v, regs) &&
          /* Do not check if access-protection fault since the page may 
             legitimately be not present in shadow page tables */
-         ((regs->error_code & PFEC_write_access) == PFEC_write_access) &&
+         ((regs->error_code & (PFEC_write_access|PFEC_reserved_bit)) ==
+          PFEC_write_access) &&
          ptwr_do_page_fault(v, addr, regs) )
         return EXCRET_fault_fixed;
 
@@ -1101,6 +1117,8 @@ asmlinkage void do_page_fault(struct cpu
         if ( likely((fixup = search_exception_table(regs->eip)) != 0) )
         {
             perfc_incr(copy_user_faults);
+            if ( unlikely(regs->error_code & PFEC_reserved_bit) )
+                reserved_bit_page_fault(0, addr, regs);
             regs->eip = fixup;
             return;
         }
#3 lore raw
Quoted 9 lines · Jan Beulich
Quoted 4 lines

The shadow pagetable code deliberately introduces invalid pagetable entries as part of its fast-path treatment of MMIO and not-present entries in the guest tables, so paging_fault needs to be called for PFEC_reserved_bit faults.

Here's a replacement patch then, it would still be questionable whether reserved bit faults should ever be propagated to a guest...

What problem does this patch solve?

-- Keir

#4 lore raw
Quoted 8 lines
Quoted 1 line
Quoted 1 line

Keir Fraser <[email protected]> 01.02.08 11:54 >>>

Quoted 5 lines · Jan Beulich
Quoted 4 lines

The shadow pagetable code deliberately introduces invalid pagetable entries as part of its fast-path treatment of MMIO and not-present entries in the guest tables, so paging_fault needs to be called for PFEC_reserved_bit faults.

Quoted 6 lines
Quoted 4 lines

Here's a replacement patch then, it would still be questionable whether reserved bit faults should ever be propagated to a guest...

What problem does this patch solve?

When debugging the xen_change_pte_range() problem I mistakenly interpreted a page fault error code 15 as being decimal (it really was hex) and looked at the reserved bit fault handling. The patch is the outcome of this; despite it turned out to not be a fault of this kind I thought that tightening the handling (and adding the printing, since reserved bit faults are [except in the case mentioned by Tim] always bad) wouldn't be a bad idea.

Jan