[SPARC64]: Verify vmalloc TLB misses more strictly.

Arrange the modules, OBP, and vmalloc areas such that a range
verification can be done quite minimally.
Signed-off-by: David S. Miller <davem@davemloft.net>

arch/sparc64/kernel/entry.S

@@ -42,19 +42,15 @@
  * executing (see inherit_locked_prom_mappings() rant).
  */
 sparc64_vpte_nucleus:
-	/* Load 0xf0000000, which is LOW_OBP_ADDRESS.  */
-	mov		0xf, %g5
-	sllx		%g5, 28, %g5
-
-	/* Is addr >= LOW_OBP_ADDRESS?  */
+	/* Note that kvmap below has verified that the address is
+	 * in the range MODULES_VADDR --> VMALLOC_END already.  So
+	 * here we need only check if it is an OBP address or not.
+	 */
+	sethi		%hi(LOW_OBP_ADDRESS), %g5
 	cmp		%g4, %g5
 	blu,pn		%xcc, sparc64_vpte_patchme1
 	 mov		0x1, %g5
-
-	/* Load 0x100000000, which is HI_OBP_ADDRESS.  */
 	sllx		%g5, 32, %g5
-
-	/* Is addr < HI_OBP_ADDRESS?  */
 	cmp		%g4, %g5
 	blu,pn		%xcc, obp_iaddr_patch
 	 nop
@@ -156,26 +152,29 @@ obp_daddr_patch:
  * rather, use information saved during inherit_prom_mappings() using 8k
  * pagesize.
  */
+	.align		32
 kvmap:
-	/* Load 0xf0000000, which is LOW_OBP_ADDRESS.  */
-	mov		0xf, %g5
-	sllx		%g5, 28, %g5
+	sethi		%hi(MODULES_VADDR), %g5
+	cmp		%g4, %g5
+	blu,pn		%xcc, longpath
+	 mov		(VMALLOC_END >> 24), %g5
+	sllx		%g5, 24, %g5
+	cmp		%g4, %g5
+	bgeu,pn		%xcc, longpath
+	 nop
 
-	/* Is addr >= LOW_OBP_ADDRESS?  */
+kvmap_check_obp:
+	sethi		%hi(LOW_OBP_ADDRESS), %g5
 	cmp		%g4, %g5
-	blu,pn		%xcc, vmalloc_addr
+	blu,pn		%xcc, kvmap_vmalloc_addr
 	 mov		0x1, %g5
-
-	/* Load 0x100000000, which is HI_OBP_ADDRESS.  */
 	sllx		%g5, 32, %g5
-
-	/* Is addr < HI_OBP_ADDRESS?  */
 	cmp		%g4, %g5
 	blu,pn		%xcc, obp_daddr_patch
 	 nop
 
-vmalloc_addr:
-	/* If we get here, a vmalloc addr accessed, load kernel VPTE.  */
+kvmap_vmalloc_addr:
+	/* If we get here, a vmalloc addr was accessed, load kernel VPTE.  */
 	ldxa		[%g3 + %g6] ASI_N, %g5
 	brgez,pn	%g5, longpath
 	 nop

include/asm-sparc64/pgtable.h

@@ -24,21 +24,23 @@
 #include <asm/processor.h>
 #include <asm/const.h>
 
-/* The kernel image occupies 0x4000000 to 0x1000000 (4MB --> 16MB).
- * The page copy blockops use 0x1000000 to 0x18000000 (16MB --> 24MB).
+/* The kernel image occupies 0x4000000 to 0x1000000 (4MB --> 32MB).
+ * The page copy blockops can use 0x2000000 to 0x10000000.
  * The PROM resides in an area spanning 0xf0000000 to 0x100000000.
- * The vmalloc area spans 0x140000000 to 0x200000000.
+ * The vmalloc area spans 0x100000000 to 0x200000000.
+ * Since modules need to be in the lowest 32-bits of the address space,
+ * we place them right before the OBP area from 0x10000000 to 0xf0000000.
  * There is a single static kernel PMD which maps from 0x0 to address
  * 0x400000000.
  */
-#define	TLBTEMP_BASE		_AC(0x0000000001000000,UL)
-#define MODULES_VADDR		_AC(0x0000000002000000,UL)
-#define MODULES_LEN		_AC(0x000000007e000000,UL)
-#define MODULES_END		_AC(0x0000000080000000,UL)
-#define VMALLOC_START		_AC(0x0000000140000000,UL)
-#define VMALLOC_END		_AC(0x0000000200000000,UL)
+#define	TLBTEMP_BASE		_AC(0x0000000002000000,UL)
+#define MODULES_VADDR		_AC(0x0000000010000000,UL)
+#define MODULES_LEN		_AC(0x00000000e0000000,UL)
+#define MODULES_END		_AC(0x00000000f0000000,UL)
 #define LOW_OBP_ADDRESS		_AC(0x00000000f0000000,UL)
 #define HI_OBP_ADDRESS		_AC(0x0000000100000000,UL)
+#define VMALLOC_START		_AC(0x0000000100000000,UL)
+#define VMALLOC_END		_AC(0x0000000200000000,UL)
 
 /* XXX All of this needs to be rethought so we can take advantage
  * XXX cheetah's full 64-bit virtual address space, ie. no more hole