x86/mm: accurately track which vCPU page-tables are loaded

Neither current nor curr_vcpu per-CPU fields accurately track which
page-tables are loaded.  There are corner cases when dealing with shadow
paging failures that switch to the idle vCPU page-tables without changing
current or curr_vcpu per-CPU fields.

Introduce a new per-CPU field that attempts to track which vCPU page-tables
are loaded.  Update such tracking when cr3 is changed, and do so in a
region with interrupts disabled, as to avoid handling interrupts with a
mismatch between the vCPU tracking field and the loaded page-tables.

As a result of this newly more accurate tracking the mapcache override
functionality can be removed: the dom0 PV builder was the only user of it,
and it's updated here to properly signal which vCPU page-tables are loaded
in the calls to switch_cr3_cr4().

Note the EFI page-tables have the Xen owned L4 slots copied from the idle
page-tables, so for the effects of the mapcache the EFI page-tables could
use the idle mapcache if it had one.  Pass the idle vCPU in the
switch_cr3_cr4() call that switches to the runtime EFI page-tables.

There are known issues with the use of mapcache in NMI context.  This patch
does not alter the behaviour.

This is CVE-2026-42488 / XSA-494.

Fixes: fb0ff49fe9 ("x86/shadow: defer releasing of PV's top-level shadow reference")
Signed-off-by: Roger Pau Monné <roger.pau@citrix.com>
Acked-by: Andrew Cooper <andrew.cooper3@citrix.com>

xen/arch/x86/domain_page.c

@@ -18,48 +18,40 @@
 #include <asm/hardirq.h>
 #include <asm/setup.h>
 
-static DEFINE_PER_CPU(struct vcpu *, override);
-
 static inline struct vcpu *mapcache_current_vcpu(void)
 {
-    /* In the common case we use the mapcache of the running VCPU. */
+    struct vcpu *v = this_cpu(pgtable_vcpu);
-    struct vcpu *v = this_cpu(override) ?: current;
+    struct vcpu *curr = current;
 
     /*
-     * When current isn't properly set up yet, this is equivalent to
+     * During early boot pgtable_vcpu is not set, callers must handle NULL.
-     * running in an idle vCPU (callers must check for NULL).
+     * Non-PV domains don't have a mapcache, the directmap covers all physical
+     * address space.
      */
-    if ( !v )
+    if ( !v || !is_pv_vcpu(v) )
         return NULL;
 
     /*
-     * When using efi runtime page tables, we have the equivalent of the idle
+     * If we are in a lazy context-switch state from a PV vCPU do a full switch
-     * domain's page tables but current may point at another domain's VCPU.
+     * to the idle vCPU now, otherwise an incoming FLUSH_VCPU_STATE IPI would
-     * Return NULL as though current is not properly set up yet.
+     * change the page tables under our feet an invalidate any in-use mapcache
+     * entries.
      */
-    if ( efi_rs_using_pgtables() )
+    if ( unlikely(this_cpu(curr_vcpu) != curr) )
-        return NULL;
-
-    /*
-     * If guest_table is NULL, and we are running a paravirtualised guest,
-     * then it means we are running on the idle domain's page table and must
-     * therefore use its mapcache.
-     */
-    if ( unlikely(pagetable_is_null(v->arch.guest_table)) && is_pv_vcpu(v) )
     {
-        /* If we really are idling, perform lazy context switch now. */
+        ASSERT(curr == idle_vcpu[smp_processor_id()]);
-        if ( (v = idle_vcpu[smp_processor_id()]) == current )
         sync_local_execstate();
         /* We must now be running on the idle page table. */
         ASSERT(cr3_pa(read_cr3()) == __pa(idle_pg_table));
     }
 
-    return v;
+    /*
-}
+     * At this point we can guarantee Xen is not in lazy context switch: either
-
+     * the code above will have synced the state, or an incoming
-void __init mapcache_override_current(struct vcpu *v)
+     * FLUSH_VCPU_STATE IPI has done so behind our back.  Use ACCESS_ONCE to
-{
+     * ensure the compiler never returns the locally cached pgtable_vcpu value.
-    this_cpu(override) = v;
+     */
+    return ACCESS_ONCE(this_cpu(pgtable_vcpu));
 }
 
 #define mapcache_l2_entry(e) ((e) >> PAGETABLE_ORDER)

xen/arch/x86/flushtlb.c

@@ -104,7 +104,9 @@ static void do_tlb_flush(void)
     local_irq_restore(flags);
 }
 
-void switch_cr3_cr4(unsigned long cr3, unsigned long cr4)
+DEFINE_PER_CPU(struct vcpu *, pgtable_vcpu);
+
+void switch_cr3_cr4(struct vcpu *v, unsigned long cr3, unsigned long cr4)
 {
     unsigned long flags, old_cr4;
     u32 t = 0;
@@ -148,6 +150,7 @@ void switch_cr3_cr4(unsigned long cr3, unsigned long cr4)
     if ( (old_cr4 & X86_CR4_PCIDE) > (cr4 & X86_CR4_PCIDE) )
         cr3 |= X86_CR3_NOFLUSH;
     write_cr3(cr3);
+    this_cpu(pgtable_vcpu) = v;
 
     if ( old_cr4 != cr4 )
         write_cr4(cr4);

xen/arch/x86/include/asm/domain.h

@@ -90,7 +90,6 @@ struct mapcache_domain {
 
 int mapcache_domain_init(struct domain *d);
 int mapcache_vcpu_init(struct vcpu *v);
-void mapcache_override_current(struct vcpu *v);
 
 /* x86/64: toggle guest between kernel and user modes. */
 void toggle_guest_mode(struct vcpu *v);

xen/arch/x86/include/asm/flushtlb.h

@@ -104,7 +104,7 @@ static inline void invlpg(const void *p)
 }
 
 /* Write pagetable base and implicitly tick the tlbflush clock. */
-void switch_cr3_cr4(unsigned long cr3, unsigned long cr4);
+void switch_cr3_cr4(struct vcpu *v, unsigned long cr3, unsigned long cr4);
 
 /* flush_* flag fields: */
  /*

xen/arch/x86/include/asm/processor.h

@@ -329,6 +329,9 @@ DECLARE_PER_CPU(struct tss_page, tss_page);
 
 DECLARE_PER_CPU(root_pgentry_t *, root_pgt);
 
+/* vCPU of the currently loaded page-tables. */
+DECLARE_PER_CPU(struct vcpu *, pgtable_vcpu);
+
 extern void write_ptbase(struct vcpu *v);
 
 /* PAUSE (encoding: REP NOP) is a good thing to insert into busy-wait loops. */

xen/arch/x86/mm.c

@@ -535,7 +535,7 @@ void write_ptbase(struct vcpu *v)
         cpu_info->pv_cr3 = __pa(this_cpu(root_pgt));
         if ( new_cr4 & X86_CR4_PCIDE )
             cpu_info->pv_cr3 |= get_pcid_bits(v, true);
-        switch_cr3_cr4(v->arch.cr3, new_cr4);
+        switch_cr3_cr4(v, v->arch.cr3, new_cr4);
     }
     else
     {
@@ -543,7 +543,7 @@ void write_ptbase(struct vcpu *v)
         cpu_info->use_pv_cr3 = false;
         cpu_info->xen_cr3 = 0;
         /* switch_cr3_cr4() serializes. */
-        switch_cr3_cr4(v->arch.cr3, new_cr4);
+        switch_cr3_cr4(v, v->arch.cr3, new_cr4);
         cpu_info->pv_cr3 = 0;
     }
 }

xen/arch/x86/pv/dom0_build.c

@@ -831,8 +831,7 @@ static int __init dom0_construct(const struct boot_domain *bd)
     update_cr3(v);
 
     /* We run on dom0's page tables for the final part of the build process. */
-    switch_cr3_cr4(cr3_pa(v->arch.cr3), read_cr4());
+    switch_cr3_cr4(v, cr3_pa(v->arch.cr3), read_cr4());
-    mapcache_override_current(v);
 
     /* Copy the OS image and free temporary buffer. */
     elf.dest_base = (void*)vkern_start;
@@ -841,8 +840,7 @@ static int __init dom0_construct(const struct boot_domain *bd)
     rc = elf_load_binary(&elf);
     if ( rc < 0 )
     {
-        mapcache_override_current(NULL);
+        switch_cr3_cr4(current, current->arch.cr3, read_cr4());
-        switch_cr3_cr4(current->arch.cr3, read_cr4());
         printk("Failed to load the kernel binary\n");
         goto out;
     }
@@ -853,8 +851,7 @@ static int __init dom0_construct(const struct boot_domain *bd)
         if ( (parms.virt_hypercall < v_start) ||
              (parms.virt_hypercall >= v_end) )
         {
-            mapcache_override_current(NULL);
+            switch_cr3_cr4(current, current->arch.cr3, read_cr4());
-            switch_cr3_cr4(current->arch.cr3, read_cr4());
             printk("Invalid HYPERCALL_PAGE field in ELF notes.\n");
             return -EINVAL;
         }
@@ -995,8 +992,7 @@ static int __init dom0_construct(const struct boot_domain *bd)
 #endif
 
     /* Return to idle domain's page tables. */
-    mapcache_override_current(NULL);
+    switch_cr3_cr4(current, current->arch.cr3, read_cr4());
-    switch_cr3_cr4(current->arch.cr3, read_cr4());
 
     update_domain_wallclock_time(d);
 

xen/arch/x86/pv/domain.c

@@ -457,6 +457,8 @@ static void _toggle_guest_pt(struct vcpu *v)
     pagetable_t old_shadow;
     unsigned long cr3;
 
+    ASSERT(local_irq_is_enabled());
+
     v->arch.flags ^= TF_kernel_mode;
     guest_update = v->arch.flags & TF_kernel_mode;
     old_shadow = update_cr3(v);
@@ -479,14 +481,21 @@ static void _toggle_guest_pt(struct vcpu *v)
     {
         cr3 &= ~X86_CR3_NOFLUSH;
 
+        local_irq_disable();
         if ( unlikely(mfn_eq(pagetable_get_mfn(old_shadow),
                              maddr_to_mfn(cr3))) )
         {
-            cr3 = idle_vcpu[v->processor]->arch.cr3;
             /* Also suppress runstate/time area updates below. */
             guest_update = false;
+
+            cr3 = idle_vcpu[v->processor]->arch.cr3;
+            this_cpu(pgtable_vcpu) = idle_vcpu[v->processor];
         }
+
+        write_cr3(cr3);
+        local_irq_enable();
     }
+    else
         write_cr3(cr3);
 
     if ( !pagetable_is_null(old_shadow) )

xen/arch/x86/smpboot.c

@@ -1063,6 +1063,7 @@ static int cpu_smpboot_alloc(unsigned int cpu)
 
     info->current_vcpu = idle_vcpu[cpu]; /* set_current() */
     per_cpu(curr_vcpu, cpu) = idle_vcpu[cpu];
+    per_cpu(pgtable_vcpu, cpu) = idle_vcpu[cpu];
 
     gdt = per_cpu(gdt, cpu) ?: alloc_xenheap_pages(0, memflags);
     if ( gdt == NULL )

xen/common/efi/common-stub.c

@@ -7,11 +7,6 @@ bool efi_enabled(unsigned int feature)
     return false;
 }
 
-bool efi_rs_using_pgtables(void)
-{
-    return false;
-}
-
 unsigned long efi_get_time(void)
 {
     BUG();

xen/common/efi/runtime.c

@@ -50,7 +50,6 @@ const CHAR16 *__read_mostly efi_fw_vendor;
 const EFI_RUNTIME_SERVICES *__read_mostly efi_rs;
 #ifndef CONFIG_ARM /* TODO - disabled until implemented on ARM */
 static DEFINE_SPINLOCK(efi_rs_lock);
-static unsigned int efi_rs_on_cpu = NR_CPUS;
 #endif
 
 UINTN __read_mostly efi_memmap_size;
@@ -93,6 +92,11 @@ struct efi_rs_state efi_rs_enter(void)
     if ( mfn_eq(efi_l4_mfn, INVALID_MFN) )
         return state;
 
+    /*
+     * If in lazy idle context switch state sync now to avoid an incoming
+     * FLUSH_VCPU_STATE IPI changing the loaded page-tables.
+     */
+    sync_local_execstate();
     state.cr3 = read_cr3();
     vcpu_save_fpu(current);
     asm volatile ( "fnclex; fldcw %0" :: "m" (fcw) );
@@ -100,8 +104,6 @@ struct efi_rs_state efi_rs_enter(void)
 
     spin_lock(&efi_rs_lock);
 
-    efi_rs_on_cpu = smp_processor_id();
-
     /* prevent fixup_page_fault() from doing anything */
     irq_enter();
 
@@ -116,7 +118,8 @@ struct efi_rs_state efi_rs_enter(void)
         lgdt(&gdt_desc);
     }
 
-    switch_cr3_cr4(mfn_to_maddr(efi_l4_mfn), read_cr4());
+    switch_cr3_cr4(idle_vcpu[smp_processor_id()], mfn_to_maddr(efi_l4_mfn),
+                   read_cr4());
 
     /*
      * At the time of writing (2022), no UEFI firwmare is CET-IBT compatible.
@@ -144,7 +147,7 @@ void efi_rs_leave(struct efi_rs_state *state)
     if ( state->msr_s_cet )
         wrmsrl(MSR_S_CET, state->msr_s_cet);
 
-    switch_cr3_cr4(state->cr3, read_cr4());
+    switch_cr3_cr4(curr, state->cr3, read_cr4());
     if ( is_pv_vcpu(curr) && !is_idle_vcpu(curr) )
     {
         struct desc_ptr gdt_desc = {
@@ -155,18 +158,10 @@ void efi_rs_leave(struct efi_rs_state *state)
         lgdt(&gdt_desc);
     }
     irq_exit();
-    efi_rs_on_cpu = NR_CPUS;
     spin_unlock(&efi_rs_lock);
     vcpu_restore_fpu(curr);
 }
 
-bool efi_rs_using_pgtables(void)
-{
-    return !mfn_eq(efi_l4_mfn, INVALID_MFN) &&
-           (smp_processor_id() == efi_rs_on_cpu) &&
-           (read_cr3() == mfn_to_maddr(efi_l4_mfn));
-}
-
 unsigned long efi_get_time(void)
 {
     EFI_TIME time;

xen/include/xen/efi.h

@@ -40,7 +40,6 @@ extern bool efi_secure_boot;
 
 void efi_init_memory(void);
 bool efi_boot_mem_unused(unsigned long *start, unsigned long *end);
-bool efi_rs_using_pgtables(void);
 unsigned long efi_get_time(void);
 void efi_reset_system(bool warm);
 #ifndef COMPAT