[Xen-devel] [PATCH v2.0 5/6] Handler m2p table and frametable fault in page fault handler

["Jiang, Yunhong" <yunhong.jiang@xxxxxxxxx>]

After memory added, the m2p table and frametable maybe increased, and the 
idle_page_table is updated for it.
These new mapping are propgated to other guest, when access to the new 
m2p/frame table. The access can happen either in HV or in guest for read-only 
mapping.
The propagation is needed for 32 bit Xen environment, or compatibility guest in 
64 bit Xen environment.

Signed-off-by: Jiang, Yunhong <yunhong.jiang@xxxxxxxxx>

diff -r 1e580fac45fb xen/arch/x86/traps.c
--- a/xen/arch/x86/traps.c      Thu Jul 02 23:03:22 2009 +0800
+++ b/xen/arch/x86/traps.c      Tue Jul 07 02:34:51 2009 +0800
@@ -1061,13 +1061,24 @@ static int handle_gdt_ldt_mapping_fault(
     return EXCRET_fault_fixed;
 }
 
+static inline int in_hypervisor_range(unsigned long va, struct vcpu *v)
+{
 #ifdef HYPERVISOR_VIRT_END
-#define IN_HYPERVISOR_RANGE(va) \
-    (((va) >= HYPERVISOR_VIRT_START) && ((va) < HYPERVISOR_VIRT_END))
+    if  (((va) >= HYPERVISOR_VIRT_START) && ((va) < HYPERVISOR_VIRT_END))
 #else
-#define IN_HYPERVISOR_RANGE(va) \
-    (((va) >= HYPERVISOR_VIRT_START))
+    if (((va) >= HYPERVISOR_VIRT_START))
 #endif
+        return 1;
+    else
+#if defined(__x86_64__)
+        return ( is_pv_32on64_domain(v->domain) ?
+                   ((va >=  HYPERVISOR_COMPAT_VIRT_START(v->domain) &&
+                     (va < MACH2PHYS_COMPAT_VIRT_END)) ) : 0 );
+#else
+        return 0;
+#endif
+}
+
 
 static int __spurious_page_fault(
     unsigned long addr, unsigned int error_code)
@@ -1207,12 +1218,15 @@ static int fixup_page_fault(unsigned lon
         return ret;
     }
 
-    if ( unlikely(IN_HYPERVISOR_RANGE(addr)) )
+    if ( unlikely(in_hypervisor_range(addr, v)) )
     {
         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);
+        if ( !(regs->error_code & PFEC_page_present) &&
+             (pagefault_by_memadd(addr, regs)) )
+             return handle_memadd_fault(addr, regs);
         return 0;
     }
 
diff -r 1e580fac45fb xen/arch/x86/x86_32/mm.c
--- a/xen/arch/x86/x86_32/mm.c  Thu Jul 02 23:03:22 2009 +0800
+++ b/xen/arch/x86/x86_32/mm.c  Tue Jul 07 01:48:03 2009 +0800
@@ -381,6 +381,76 @@ int check_descriptor(const struct domain
     return 0;
 }
 
+int pagefault_by_memadd(unsigned long addr, struct cpu_user_regs *regs)
+{
+    if ( guest_mode(regs) )
+    {
+        if ( ((addr >= RO_MPT_VIRT_START) && (addr < RO_MPT_VIRT_END)) )
+            return 1;
+    }
+    else
+    {
+        if ( (addr >= RO_MPT_VIRT_START) && (addr < RDWR_MPT_VIRT_END) )
+            return 1;
+    }
+
+    return 0;
+}
+
+int handle_memadd_fault(unsigned long addr, struct cpu_user_regs *regs)
+{
+    l3_pgentry_t  *pl3e = NULL;
+    l3_pgentry_t l3e;
+    l2_pgentry_t *pl2e = NULL;
+    l2_pgentry_t l2e, idle_l2e;
+    unsigned long mfn, cr3;
+    int rc = 0;
+
+    idle_l2e = idle_pg_table_l2[l2_linear_offset(addr)];
+    if (!(l2e_get_flags(idle_l2e) & _PAGE_PRESENT))
+        return 0;
+
+    cr3 = read_cr3();
+    mfn = cr3 >> PAGE_SHIFT;
+
+    /*
+     * No need get page type here and validate checking for xen mapping
+     */
+    pl3e = map_domain_page(mfn);
+    pl3e += (cr3 & 0xFE0UL) >> 3;
+    l3e = pl3e[3];
+
+    if ( !(l3e_get_flags(l3e) & _PAGE_PRESENT) )
+        goto done;
+
+    mfn = l3e_get_pfn(l3e);
+    pl2e = map_domain_page(mfn);
+
+    l2e = pl2e[l2_table_offset(addr)];
+
+    /*
+     * These entry will only be set by Xen HV.
+     * Other vCPU may setup it in another pCPU
+     */
+    if (l2e_get_flags(l2e) & _PAGE_PRESENT)
+    {
+        ASSERT(l2e_get_intpte(l2e) == l2e_get_intpte(idle_l2e));
+        rc = EXCRET_fault_fixed;
+        goto done;
+    }
+
+    pl2e[l2_table_offset(addr)] = idle_l2e;
+
+    rc = EXCRET_fault_fixed;
+
+done:
+    if (pl3e)
+        unmap_domain_page(pl3e);
+    if (pl2e)
+        unmap_domain_page(pl2e);
+    return rc;
+}
+
 /*
  * Local variables:
  * mode: C
diff -r 1e580fac45fb xen/arch/x86/x86_64/mm.c
--- a/xen/arch/x86/x86_64/mm.c  Thu Jul 02 23:03:22 2009 +0800
+++ b/xen/arch/x86/x86_64/mm.c  Mon Jul 06 23:47:48 2009 +0800
@@ -258,8 +258,12 @@ int setup_m2p_table(unsigned long spfn, 
     if ( mpt_size > RDWR_COMPAT_MPT_VIRT_END - RDWR_COMPAT_MPT_VIRT_START )
         mpt_size = RDWR_COMPAT_MPT_VIRT_END - RDWR_COMPAT_MPT_VIRT_START;
     mpt_size &= ~((1UL << L2_PAGETABLE_SHIFT) - 1UL);
+
+    /* We don't support change the m2p_compat_vstart with memory hotplug */
+#if !defined(CONFIG_MEMORY_HOTPLUG)
     if ( (m2p_compat_vstart + mpt_size) < MACH2PHYS_COMPAT_VIRT_END )
         m2p_compat_vstart = MACH2PHYS_COMPAT_VIRT_END - mpt_size;
+#endif
 
     /* Has already setup */
     if ( (mpt_size >> PAGE_SHIFT) <= compat_mpt_pages )
@@ -645,6 +649,87 @@ unsigned int domain_clamp_alloc_bitsize(
     return min(d->arch.physaddr_bitsize, bits);
 }
 
+int pagefault_by_memadd(unsigned long addr, struct cpu_user_regs *regs)
+{
+    struct domain *d = current->domain;
+#ifdef CONFIG_COMPAT
+    if ( is_pv_32on64_domain(d) &&
+         (addr >= HYPERVISOR_COMPAT_VIRT_START(d)) &&
+           (addr < MACH2PHYS_COMPAT_VIRT_END) )
+            return 1;
+#endif
+    return 0;
+}
+
+int handle_memadd_fault(unsigned long addr, struct cpu_user_regs *regs)
+{
+#ifdef CONFIG_COMPAT
+    struct domain *d = current->domain;
+    l4_pgentry_t *pl4e = NULL;
+    l4_pgentry_t l4e;
+    l3_pgentry_t  *pl3e = NULL;
+    l3_pgentry_t l3e;
+    l2_pgentry_t *pl2e = NULL;
+    l2_pgentry_t l2e, idle_l2e;
+    unsigned long mfn, idle_index;
+    int rc = 0;
+
+    if (!is_pv_32on64_domain(d))
+        return 0;
+
+    idle_index = l2_table_offset(addr) -
+                        COMPAT_L2_PAGETABLE_FIRST_XEN_SLOT(d);
+    idle_l2e = compat_idle_pg_table_l2[idle_index];
+    if (!(l2e_get_flags(idle_l2e) & _PAGE_PRESENT))
+        return 0;
+
+    mfn = (read_cr3()) >> PAGE_SHIFT;
+
+    pl4e = map_domain_page(mfn);
+
+    l4e = pl4e[0];
+
+    if (!(l4e_get_flags(l4e) & _PAGE_PRESENT))
+        goto done;
+
+    mfn = l4e_get_pfn(l4e);
+    /* We don't need get page type here since this is xen mapping */
+    pl3e = map_domain_page(mfn);
+
+    l3e = pl3e[3];
+
+    if ( !(l3e_get_flags(l3e) & _PAGE_PRESENT) )
+        goto done;
+
+    mfn = l3e_get_pfn(l3e);
+    pl2e = map_domain_page(mfn);
+
+    l2e = pl2e[l2_table_offset(addr)];
+
+    /*
+     * Race happen that other pCPU may have fixed it
+     */
+    if (l2e_get_flags(l2e) & _PAGE_PRESENT)
+    {
+        ASSERT(l2e_get_intpte(l2e) == l2e_get_intpte(idle_l2e));
+        rc = EXCRET_fault_fixed;
+        goto done;
+    }
+
+    pl2e[l2_table_offset(addr)] = idle_l2e;
+
+    rc = EXCRET_fault_fixed;
+#endif
+done:
+    if (pl4e)
+        unmap_domain_page(pl4e);
+    if (pl3e)
+        unmap_domain_page(pl3e);
+    if (pl2e)
+        unmap_domain_page(pl2e);
+    return rc;
+}
+
 #include "compat/mm.c"
 
 /*
diff -r 1e580fac45fb xen/include/asm-x86/mm.h
--- a/xen/include/asm-x86/mm.h  Thu Jul 02 23:03:22 2009 +0800
+++ b/xen/include/asm-x86/mm.h  Thu Jul 02 23:03:22 2009 +0800
@@ -518,6 +518,8 @@ int setup_m2p_table(unsigned long spfn, 
 int setup_m2p_table(unsigned long spfn, unsigned long epfn, int hotplug);
 unsigned long domain_get_maximum_gpfn(struct domain *d);
 
+int handle_memadd_fault(unsigned long addr, struct cpu_user_regs *regs);
+int pagefault_by_memadd(unsigned long addr, struct cpu_user_regs *regs);
 extern struct domain *dom_xen, *dom_io;        /* for vmcoreinfo */
 
 #endif /* __ASM_X86_MM_H__ */

Attachment: page_fault.patch
Description: page_fault.patch

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel