xen/x86: replace order-based range checking of M2P table by linear one
authorJan Beulich <JBeulich@novell.com>
Tue, 16 Aug 2011 14:07:41 +0000 (15:07 +0100)
committerKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Wed, 17 Aug 2011 14:26:48 +0000 (10:26 -0400)
The order-based approach is not only less efficient (requiring a shift
and a compare, typical generated code looking like this

mov eax, [machine_to_phys_order]
mov ecx, eax
shr ebx, cl
test ebx, ebx
jnz ...

whereas a direct check requires just a compare, like in

cmp ebx, [machine_to_phys_nr]
jae ...

), but also slightly dangerous in the 32-on-64 case - the element
address calculation can wrap if the next power of two boundary is
sufficiently far away from the actual upper limit of the table, and
hence can result in user space addresses being accessed (with it being
unknown what may actually be mapped there).

Additionally, the elimination of the mistaken use of fls() here (should
have been __fls()) fixes a latent issue on x86-64 that would trigger
if the code was run on a system with memory extending beyond the 44-bit
boundary.

CC: stable@kernel.org
Signed-off-by: Jan Beulich <jbeulich@novell.com>
[v1: Based on Jeremy's feedback]
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
arch/x86/include/asm/xen/page.h
arch/x86/xen/enlighten.c
arch/x86/xen/mmu.c

index 64a619d47d341739ff34b5236e6bf85db0a79b06..7ff4669580cfc8bd0db462d3d0c07cf94cfddca0 100644 (file)
@@ -39,7 +39,7 @@ typedef struct xpaddr {
     ((unsigned long)((u64)CONFIG_XEN_MAX_DOMAIN_MEMORY * 1024 * 1024 * 1024 / PAGE_SIZE))
 
 extern unsigned long *machine_to_phys_mapping;
-extern unsigned int   machine_to_phys_order;
+extern unsigned long  machine_to_phys_nr;
 
 extern unsigned long get_phys_to_machine(unsigned long pfn);
 extern bool set_phys_to_machine(unsigned long pfn, unsigned long mfn);
@@ -87,7 +87,7 @@ static inline unsigned long mfn_to_pfn(unsigned long mfn)
        if (xen_feature(XENFEAT_auto_translated_physmap))
                return mfn;
 
-       if (unlikely((mfn >> machine_to_phys_order) != 0)) {
+       if (unlikely(mfn >= machine_to_phys_nr)) {
                pfn = ~0;
                goto try_override;
        }
index 974a528458a04d3257234ceb70991b7f01eb3748..b960429d5b65d9e024adbe8655d556071f45bc2d 100644 (file)
@@ -77,8 +77,8 @@ EXPORT_SYMBOL_GPL(xen_domain_type);
 
 unsigned long *machine_to_phys_mapping = (void *)MACH2PHYS_VIRT_START;
 EXPORT_SYMBOL(machine_to_phys_mapping);
-unsigned int   machine_to_phys_order;
-EXPORT_SYMBOL(machine_to_phys_order);
+unsigned long  machine_to_phys_nr;
+EXPORT_SYMBOL(machine_to_phys_nr);
 
 struct start_info *xen_start_info;
 EXPORT_SYMBOL_GPL(xen_start_info);
index f987bde77c490666fa27e728b509ce9d8d0ab1b6..24abc1f50dc5c671ae2e4a9973efd1d9762472a4 100644 (file)
@@ -1713,15 +1713,19 @@ static void __init xen_map_identity_early(pmd_t *pmd, unsigned long max_pfn)
 void __init xen_setup_machphys_mapping(void)
 {
        struct xen_machphys_mapping mapping;
-       unsigned long machine_to_phys_nr_ents;
 
        if (HYPERVISOR_memory_op(XENMEM_machphys_mapping, &mapping) == 0) {
                machine_to_phys_mapping = (unsigned long *)mapping.v_start;
-               machine_to_phys_nr_ents = mapping.max_mfn + 1;
+               machine_to_phys_nr = mapping.max_mfn + 1;
        } else {
-               machine_to_phys_nr_ents = MACH2PHYS_NR_ENTRIES;
+               machine_to_phys_nr = MACH2PHYS_NR_ENTRIES;
        }
-       machine_to_phys_order = fls(machine_to_phys_nr_ents - 1);
+#ifdef CONFIG_X86_32
+       if ((machine_to_phys_mapping + machine_to_phys_nr)
+           < machine_to_phys_mapping)
+               machine_to_phys_nr = (unsigned long *)NULL
+                                    - machine_to_phys_mapping;
+#endif
 }
 
 #ifdef CONFIG_X86_64