[AVR32] Optimize the TLB miss handler

Reorder some instructions and change the register usage to reduce
the number of pipeline stalls. Also use the bfextu and bfins
instructions for bitfield manipulations instead of shifting and
masking.

This makes gzipping a 80MB file approximately 2% faster.

Signed-off-by: Haavard Skinnemoen <hskinnemoen@atmel.com>

arch/avr32/kernel/entry-avr32b.S

@@ -100,55 +100,49 @@ dtlb_miss_write:
 
 	.global	tlb_miss_common
 tlb_miss_common:
-	mfsr	r0, SYSREG_PTBR
-	mfsr	r1, SYSREG_TLBEAR
+	mfsr	r0, SYSREG_TLBEAR
+	mfsr	r1, SYSREG_PTBR
 
 	/* Is it the vmalloc space? */
-	bld	r1, 31
+	bld	r0, 31
 	brcs	handle_vmalloc_miss
 
 	/* First level lookup */
 pgtbl_lookup:
-	lsr	r2, r1, PGDIR_SHIFT
-	ld.w	r0, r0[r2 << 2]
-	bld	r0, _PAGE_BIT_PRESENT
+	lsr	r2, r0, PGDIR_SHIFT
+	ld.w	r3, r1[r2 << 2]
+	bfextu	r1, r0, PAGE_SHIFT, PGDIR_SHIFT - PAGE_SHIFT
+	bld	r3, _PAGE_BIT_PRESENT
 	brcc	page_table_not_present
 
-	/* TODO: Check access rights on page table if necessary */
-
 	/* Translate to virtual address in P1. */
-	andl	r0, 0xf000
-	sbr	r0, 31
+	andl	r3, 0xf000
+	sbr	r3, 31
 
 	/* Second level lookup */
-	lsl	r1, (32 - PGDIR_SHIFT)
-	lsr	r1, (32 - PGDIR_SHIFT) + PAGE_SHIFT
-	add	r2, r0, r1 << 2
-	ld.w	r1, r2[0]
-	bld	r1, _PAGE_BIT_PRESENT
+	ld.w	r2, r3[r1 << 2]
+	mfsr	r0, SYSREG_TLBARLO
+	bld	r2, _PAGE_BIT_PRESENT
 	brcc	page_not_present
 
 	/* Mark the page as accessed */
-	sbr	r1, _PAGE_BIT_ACCESSED
-	st.w	r2[0], r1
+	sbr	r2, _PAGE_BIT_ACCESSED
+	st.w	r3[r1 << 2], r2
 
 	/* Drop software flags */
-	andl	r1, _PAGE_FLAGS_HARDWARE_MASK & 0xffff
-	mtsr	SYSREG_TLBELO, r1
+	andl	r2, _PAGE_FLAGS_HARDWARE_MASK & 0xffff
+	mtsr	SYSREG_TLBELO, r2
 
 	/* Figure out which entry we want to replace */
-	mfsr	r0, SYSREG_TLBARLO
+	mfsr	r1, SYSREG_MMUCR
 	clz	r2, r0
 	brcc	1f
-	mov	r1, -1			/* All entries have been accessed, */
-	mtsr	SYSREG_TLBARLO, r1	/* so reset TLBAR */
-	mov	r2, 0			/* and start at 0 */
-1:	mfsr	r1, SYSREG_MMUCR
-	lsl	r2, 14
-	andl	r1, 0x3fff, COH
-	or	r1, r2
-	mtsr	SYSREG_MMUCR, r1
+	mov	r3, -1			/* All entries have been accessed, */
+	mov	r2, 0			/* so start at 0 */
+	mtsr	SYSREG_TLBARLO, r3	/* and reset TLBAR */
 
+1:	bfins	r1, r2, SYSREG_DRP_OFFSET, SYSREG_DRP_SIZE
+	mtsr	SYSREG_MMUCR, r1
 	tlbw
 
 	tlbmiss_restore
@@ -156,8 +150,8 @@ pgtbl_lookup:
 
 handle_vmalloc_miss:
 	/* Simply do the lookup in init's page table */
-	mov	r0, lo(swapper_pg_dir)
-	orh	r0, hi(swapper_pg_dir)
+	mov	r1, lo(swapper_pg_dir)
+	orh	r1, hi(swapper_pg_dir)
 	rjmp	pgtbl_lookup