[SPARC64]: Fix boot failures on SunBlade-150

The sequence to move over to the Linux trap tables from
the firmware ones needs to be more air tight.  It turns
out that to be %100 safe we do need to be able to translate
OBP mappings in our TLB miss handlers early.

In order not to eat up a lot of kernel image memory with
static page tables, just use the translations array in
the OBP TLB miss handlers.  That solves the bulk of the
problem.

Furthermore, to make sure the OBP TLB miss path will work
even before the fixed MMU globals are loaded, explicitly
load %g1 to TLB_SFSR at the beginning of the i-TLB and
d-TLB miss handlers.

To ease the OBP TLB miss walking of the prom_trans[] array,
we sort it then delete all of the non-OBP entries in there
(for example, there are entries for the kernel image itself
which we're not interested in at all).

We also save about 32K of kernel image size with this change.
Not a bad side effect :-)

There are still some reasons why trampoline.S can't use the
setup_trap_table() yet.  The most noteworthy are:

1) OBP boots secondary processors with non-bias'd stack for
   some reason.  This is easily fixed by using a small bootup
   stack in the kernel image explicitly for this purpose.

2) Doing a firmware call via the normal C call prom_set_trap_table()
   goes through the whole OBP enter/exit sequence that saves and
   restores OBP and Linux kernel state in the MMUs.  This path
   unfortunately does a "flush %g6" while loading up the OBP locked
   TLB entries for the firmware call.

   If we setup the %g6 in the trampoline.S code properly, that
   is in the PAGE_OFFSET linear mapping, but we're not on the
   kernel trap table yet so those addresses won't translate properly.

   One idea is to do a by-hand firmware call like we do in the
   early bootup code and elsewhere here in trampoline.S  But this
   fails as well, as aparently the secondary processors are not
   booted with OBP's special locked TLB entries loaded.  These
   are necessary for the firwmare to processes TLB misses correctly
   up until the point where we take over the trap table.

This does need to be resolved at some point.

Signed-off-by: David S. Miller <davem@davemloft.net>

6 files changed, 166 insertions, 226 deletions

diff --git a/arch/sparc64/kernel/dtlb_base.S b/arch/sparc64/kernel/dtlb_base.S
index 702d349c1e88f7..6528786840c06b 100644
--- a/arch/sparc64/kernel/dtlb_base.S
+++ b/arch/sparc64/kernel/dtlb_base.S
@@ -53,19 +53,18 @@
  * be guaranteed to be 0 ... mmu_context.h does guarantee this
  * by only using 10 bits in the hwcontext value.
  */
-#define CREATE_VPTE_OFFSET1(r1, r2)
+#define CREATE_VPTE_OFFSET1(r1, r2) nop
 #define CREATE_VPTE_OFFSET2(r1, r2) \
 				srax	r1, 10, r2
-#define CREATE_VPTE_NOP		nop
 #else
 #define CREATE_VPTE_OFFSET1(r1, r2) \
 				srax	r1, PAGE_SHIFT, r2
 #define CREATE_VPTE_OFFSET2(r1, r2) \
 				sllx	r2, 3, r2
-#define CREATE_VPTE_NOP
 #endif
 
 /* DTLB ** ICACHE line 1: Quick user TLB misses		*/
+	mov		TLB_SFSR, %g1
 	ldxa		[%g1 + %g1] ASI_DMMU, %g4	! Get TAG_ACCESS
 	andcc		%g4, TAG_CONTEXT_BITS, %g0	! From Nucleus?
 from_tl1_trap:
@@ -74,18 +73,16 @@ from_tl1_trap:
 	be,pn		%xcc, kvmap			! Yep, special processing
 	 CREATE_VPTE_OFFSET2(%g4, %g6)			! Create VPTE offset
 	cmp		%g5, 4				! Last trap level?
-	be,pn		%xcc, longpath			! Yep, cannot risk VPTE miss
-	 nop						! delay slot
 
 /* DTLB ** ICACHE line 2: User finish + quick kernel TLB misses	*/
+	be,pn		%xcc, longpath			! Yep, cannot risk VPTE miss
+	 nop						! delay slot
 	ldxa		[%g3 + %g6] ASI_S, %g5		! Load VPTE
 1:	brgez,pn	%g5, longpath			! Invalid, branch out
 	 nop						! Delay-slot
 9:	stxa		%g5, [%g0] ASI_DTLB_DATA_IN	! Reload TLB
 	retry						! Trap return
 	nop
-	nop
-	nop
 
 /* DTLB ** ICACHE line 3: winfixups+real_faults		*/
 longpath:
@@ -106,8 +103,7 @@ longpath:
 	nop
 	nop
 	nop
-	CREATE_VPTE_NOP
+	nop
 
 #undef CREATE_VPTE_OFFSET1
 #undef CREATE_VPTE_OFFSET2
-#undef CREATE_VPTE_NOP
diff --git a/arch/sparc64/kernel/dtlb_prot.S b/arch/sparc64/kernel/dtlb_prot.S
index d848bb7374bb4e..e0a920162604e3 100644
--- a/arch/sparc64/kernel/dtlb_prot.S
+++ b/arch/sparc64/kernel/dtlb_prot.S
@@ -14,14 +14,14 @@
  */
 
 /* PROT ** ICACHE line 1: User DTLB protection trap	*/
-	stxa		%g0, [%g1] ASI_DMMU		! Clear SFSR FaultValid bit
-	membar		#Sync				! Synchronize ASI stores
-	rdpr		%pstate, %g5			! Move into alternate globals
+	mov		TLB_SFSR, %g1
+	stxa		%g0, [%g1] ASI_DMMU		! Clear FaultValid bit
+	membar		#Sync				! Synchronize stores
+	rdpr		%pstate, %g5			! Move into alt-globals
 	wrpr		%g5, PSTATE_AG|PSTATE_MG, %pstate
-	rdpr		%tl, %g1			! Need to do a winfixup?
+	rdpr		%tl, %g1			! Need a winfixup?
 	cmp		%g1, 1				! Trap level >1?
-	mov		TLB_TAG_ACCESS, %g4		! Prepare reload of vaddr
-	nop
+	mov		TLB_TAG_ACCESS, %g4		! For reload of vaddr
 
 /* PROT ** ICACHE line 2: More real fault processing */
 	bgu,pn		%xcc, winfix_trampoline		! Yes, perform winfixup
diff --git a/arch/sparc64/kernel/head.S b/arch/sparc64/kernel/head.S
index 4c942f71184daa..b49dcd4504b029 100644
--- a/arch/sparc64/kernel/head.S
+++ b/arch/sparc64/kernel/head.S
@@ -28,19 +28,14 @@
 #include <asm/mmu.h>
 	
 /* This section from from _start to sparc64_boot_end should fit into
- * 0x0000.0000.0040.4000 to 0x0000.0000.0040.8000 and will be sharing space
- * with bootup_user_stack, which is from 0x0000.0000.0040.4000 to
- * 0x0000.0000.0040.6000 and empty_bad_page, which is from
- * 0x0000.0000.0040.6000 to 0x0000.0000.0040.8000. 
+ * 0x0000000000404000 to 0x0000000000408000.
  */
-
 	.text
 	.globl	start, _start, stext, _stext
 _start:
 start:
 _stext:
 stext:
-bootup_user_stack:
 ! 0x0000000000404000
 	b	sparc64_boot
 	 flushw					/* Flush register file.      */
@@ -392,31 +387,30 @@ tlb_fixup_done:
 	 * former does use this code, the latter does not yet due
 	 * to some complexities.  That should be fixed up at some
 	 * point.
+	 *
+	 * There used to be enormous complexity wrt. transferring
+	 * over from the firwmare's trap table to the Linux kernel's.
+	 * For example, there was a chicken & egg problem wrt. building
+	 * the OBP page tables, yet needing to be on the Linux kernel
+	 * trap table (to translate PAGE_OFFSET addresses) in order to
+	 * do that.
+	 *
+	 * We now handle OBP tlb misses differently, via linear lookups
+	 * into the prom_trans[] array.  So that specific problem no
+	 * longer exists.  Yet, unfortunately there are still some issues
+	 * preventing trampoline.S from using this code... ho hum.
 	 */
 	.globl	setup_trap_table
 setup_trap_table:
 	save	%sp, -192, %sp
 
-	/* Force interrupts to be disabled.  Transferring over to
-	 * the Linux trap table is a very delicate operation.
-	 * Until we are actually on the Linux trap table, we cannot
-	 * get the PAGE_OFFSET linear mappings translated.  We need
-	 * that mapping to be setup in order to initialize the firmware
-	 * page tables.
-	 *
-	 * So there is this window of time, from the return from
-	 * prom_set_trap_table() until inherit_prom_mappings_post()
-	 * (in arch/sparc64/mm/init.c) completes, during which no
-	 * firmware address space accesses can be made.
-	 */
+	/* Force interrupts to be disabled. */
 	rdpr	%pstate, %o1
 	andn	%o1, PSTATE_IE, %o1
 	wrpr	%o1, 0x0, %pstate
 	wrpr	%g0, 15, %pil
 
-	/* Ok, now make the final valid firmware call to jump over
-	 * to the Linux trap table.
-	 */
+	/* Make the firmware call to jump over to the Linux trap table.  */
 	call	prom_set_trap_table
 	 sethi	%hi(sparc64_ttable_tl0), %o0
 
@@ -540,15 +534,21 @@ setup_tba:	/* i0 = is_starfire */
 
 	ret
 	 restore
+sparc64_boot_end:
+
+#include "systbls.S"
+#include "ktlb.S"
+#include "etrap.S"
+#include "rtrap.S"
+#include "winfixup.S"
+#include "entry.S"
 
 /*
- * The following skips make sure the trap table in ttable.S is aligned
+ * The following skip makes sure the trap table in ttable.S is aligned
  * on a 32K boundary as required by the v9 specs for TBA register.
  */
-sparc64_boot_end:
-	.skip	0x2000 + _start - sparc64_boot_end
-bootup_user_stack_end:
-	.skip	0x2000
+1:
+	.skip	0x4000 + _start - 1b
 
 #ifdef CONFIG_SBUS
 /* This is just a hack to fool make depend config.h discovering
@@ -560,15 +560,6 @@ bootup_user_stack_end:
 ! 0x0000000000408000
 
 #include "ttable.S"
-#include "systbls.S"
-#include "ktlb.S"
-#include "etrap.S"
-#include "rtrap.S"
-#include "winfixup.S"
-#include "entry.S"
-
-	/* This is just anal retentiveness on my part... */
-	.align	16384
 
 	.data
 	.align	8
diff --git a/arch/sparc64/kernel/itlb_base.S b/arch/sparc64/kernel/itlb_base.S
index b5e32dfa4fbc7b..4951ff8f68775c 100644
--- a/arch/sparc64/kernel/itlb_base.S
+++ b/arch/sparc64/kernel/itlb_base.S
@@ -15,14 +15,12 @@
  */
 #define CREATE_VPTE_OFFSET1(r1, r2) \
 				srax	r1, 10, r2
-#define CREATE_VPTE_OFFSET2(r1, r2)
-#define CREATE_VPTE_NOP		nop
+#define CREATE_VPTE_OFFSET2(r1, r2) nop
 #else /* PAGE_SHIFT */
 #define CREATE_VPTE_OFFSET1(r1, r2) \
 				srax	r1, PAGE_SHIFT, r2
 #define CREATE_VPTE_OFFSET2(r1, r2) \
 				sllx	r2, 3, r2
-#define CREATE_VPTE_NOP
 #endif /* PAGE_SHIFT */
 
 
@@ -36,6 +34,7 @@
  */
 
 /* ITLB ** ICACHE line 1: Quick user TLB misses		*/
+	mov		TLB_SFSR, %g1
 	ldxa		[%g1 + %g1] ASI_IMMU, %g4	! Get TAG_ACCESS
 	CREATE_VPTE_OFFSET1(%g4, %g6)			! Create VPTE offset
 	CREATE_VPTE_OFFSET2(%g4, %g6)			! Create VPTE offset
@@ -43,41 +42,38 @@
 1:	brgez,pn	%g5, 3f				! Not valid, branch out
 	 sethi		%hi(_PAGE_EXEC), %g4		! Delay-slot
 	andcc		%g5, %g4, %g0			! Executable?
+
+/* ITLB ** ICACHE line 2: Real faults			*/
 	be,pn		%xcc, 3f			! Nope, branch.
 	 nop						! Delay-slot
 2:	stxa		%g5, [%g0] ASI_ITLB_DATA_IN	! Load PTE into TLB
 	retry						! Trap return
-3:	rdpr		%pstate, %g4			! Move into alternate globals
-
-/* ITLB ** ICACHE line 2: Real faults			*/
+3:	rdpr		%pstate, %g4			! Move into alt-globals
 	wrpr		%g4, PSTATE_AG|PSTATE_MG, %pstate
 	rdpr		%tpc, %g5			! And load faulting VA
 	mov		FAULT_CODE_ITLB, %g4		! It was read from ITLB
-sparc64_realfault_common:				! Called by TL0 dtlb_miss too
+
+/* ITLB ** ICACHE line 3: Finish faults	*/
+sparc64_realfault_common:				! Called by dtlb_miss
 	stb		%g4, [%g6 + TI_FAULT_CODE]
 	stx		%g5, [%g6 + TI_FAULT_ADDR]
 	ba,pt		%xcc, etrap			! Save state
 1:	 rd		%pc, %g7			! ...
-	nop
-
-/* ITLB ** ICACHE line 3: Finish faults + window fixups	*/
 	call		do_sparc64_fault		! Call fault handler
 	 add		%sp, PTREGS_OFF, %o0! Compute pt_regs arg
 	ba,pt		%xcc, rtrap_clr_l6		! Restore cpu state
 	 nop
+
+/* ITLB ** ICACHE line 4: Window fixups */
 winfix_trampoline:
 	rdpr		%tpc, %g3			! Prepare winfixup TNPC
-	or		%g3, 0x7c, %g3			! Compute offset to branch
+	or		%g3, 0x7c, %g3			! Compute branch offset
 	wrpr		%g3, %tnpc			! Write it into TNPC
 	done						! Do it to it
-
-/* ITLB ** ICACHE line 4: Unused...	*/
 	nop
 	nop
 	nop
 	nop
-	CREATE_VPTE_NOP
 
 #undef CREATE_VPTE_OFFSET1
 #undef CREATE_VPTE_OFFSET2
-#undef CREATE_VPTE_NOP
diff --git a/arch/sparc64/kernel/ktlb.S b/arch/sparc64/kernel/ktlb.S
index 7796b37f478cc6..d9244d3c9f73dd 100644
--- a/arch/sparc64/kernel/ktlb.S
+++ b/arch/sparc64/kernel/ktlb.S
@@ -58,9 +58,6 @@ vpte_noent:
 	done
 
 vpte_insn_obp:
-	sethi		%hi(prom_pmd_phys), %g5
-	ldx		[%g5 + %lo(prom_pmd_phys)], %g5
-
 	/* Behave as if we are at TL0.  */
 	wrpr		%g0, 1, %tl
 	rdpr		%tpc, %g4	/* Find original faulting iaddr */
@@ -71,58 +68,57 @@ vpte_insn_obp:
 	mov		TLB_SFSR, %g1
 	stxa		%g4, [%g1 + %g1] ASI_IMMU
 
-	/* Get PMD offset.  */
-	srlx		%g4, 23, %g6
-	and		%g6, 0x7ff, %g6
-	sllx		%g6, 2, %g6
-
-	/* Load PMD, is it valid?  */
-	lduwa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
-	brz,pn		%g5, longpath
-	 sllx		%g5, 11, %g5
-
-	/* Get PTE offset.  */
-	srlx		%g4, 13, %g6
-	and		%g6, 0x3ff, %g6
-	sllx		%g6, 3, %g6
-
-	/* Load PTE.  */
-	ldxa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
-	brgez,pn	%g5, longpath
-	 nop
-
-	/* TLB load and return from trap.  */
+	sethi		%hi(prom_trans), %g5
+	or		%g5, %lo(prom_trans), %g5
+
+1:	ldx		[%g5 + 0x00], %g6	! base
+	brz,a,pn	%g6, longpath		! no more entries, fail
+	 mov		TLB_SFSR, %g1		! and restore %g1
+	ldx		[%g5 + 0x08], %g1	! len
+	add		%g6, %g1, %g1		! end
+	cmp		%g6, %g4
+	bgu,pt		%xcc, 2f
+	 cmp		%g4, %g1
+	bgeu,pt		%xcc, 2f
+	 ldx		[%g5 + 0x10], %g1	! PTE
+
+	/* TLB load, restore %g1, and return from trap.  */
+	sub		%g4, %g6, %g6
+	add		%g1, %g6, %g5
+	mov		TLB_SFSR, %g1
 	stxa		%g5, [%g0] ASI_ITLB_DATA_IN
 	retry
 
-kvmap_do_obp:
-	sethi		%hi(prom_pmd_phys), %g5
-	ldx		[%g5 + %lo(prom_pmd_phys)], %g5
-
-	/* Get PMD offset.  */
-	srlx		%g4, 23, %g6
-	and		%g6, 0x7ff, %g6
-	sllx		%g6, 2, %g6
-
-	/* Load PMD, is it valid?  */
-	lduwa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
-	brz,pn		%g5, longpath
-	 sllx		%g5, 11, %g5
-
-	/* Get PTE offset.  */
-	srlx		%g4, 13, %g6
-	and		%g6, 0x3ff, %g6
-	sllx		%g6, 3, %g6
-
-	/* Load PTE.  */
-	ldxa		[%g5 + %g6] ASI_PHYS_USE_EC, %g5
-	brgez,pn	%g5, longpath
-	 nop
+2:	ba,pt		%xcc, 1b
+	 add		%g5, (3 * 8), %g5	! next entry
 
-	/* TLB load and return from trap.  */
+kvmap_do_obp:
+	sethi		%hi(prom_trans), %g5
+	or		%g5, %lo(prom_trans), %g5
+	srlx		%g4, 13, %g4
+	sllx		%g4, 13, %g4
+
+1:	ldx		[%g5 + 0x00], %g6	! base
+	brz,a,pn	%g6, longpath		! no more entries, fail
+	 mov		TLB_SFSR, %g1		! and restore %g1
+	ldx		[%g5 + 0x08], %g1	! len
+	add		%g6, %g1, %g1		! end
+	cmp		%g6, %g4
+	bgu,pt		%xcc, 2f
+	 cmp		%g4, %g1
+	bgeu,pt		%xcc, 2f
+	 ldx		[%g5 + 0x10], %g1	! PTE
+
+	/* TLB load, restore %g1, and return from trap.  */
+	sub		%g4, %g6, %g6
+	add		%g1, %g6, %g5
+	mov		TLB_SFSR, %g1
 	stxa		%g5, [%g0] ASI_DTLB_DATA_IN
 	retry
 
+2:	ba,pt		%xcc, 1b
+	 add		%g5, (3 * 8), %g5	! next entry
+
 /*
  * On a first level data miss, check whether this is to the OBP range (note
  * that such accesses can be made by prom, as well as by kernel using
diff --git a/arch/sparc64/mm/init.c b/arch/sparc64/mm/init.c
index 0d2e967c7200b3..1e44ee26cee847 100644
--- a/arch/sparc64/mm/init.c
+++ b/arch/sparc64/mm/init.c
@@ -105,7 +105,7 @@ static void __init read_obp_memory(const char *property,
 		regs[i].phys_addr = base;
 		regs[i].reg_size = size;
 	}
-	sort(regs, ents, sizeof(struct  linux_prom64_registers),
+	sort(regs, ents, sizeof(struct linux_prom64_registers),
 	     cmp_p64, NULL);
 }
 
@@ -367,8 +367,11 @@ struct linux_prom_translation {
 	unsigned long size;
 	unsigned long data;
 };
-static struct linux_prom_translation prom_trans[512] __initdata;
-static unsigned int prom_trans_ents __initdata;
+
+/* Exported for kernel TLB miss handling in ktlb.S */
+struct linux_prom_translation prom_trans[512] __read_mostly;
+unsigned int prom_trans_ents __read_mostly;
+unsigned int swapper_pgd_zero __read_mostly;
 
 extern unsigned long prom_boot_page;
 extern void prom_remap(unsigned long physpage, unsigned long virtpage, int mmu_ihandle);
@@ -378,122 +381,57 @@ extern void register_prom_callbacks(void);
 /* Exported for SMP bootup purposes. */
 unsigned long kern_locked_tte_data;
 
-/* Exported for kernel TLB miss handling in ktlb.S */
-unsigned long prom_pmd_phys __read_mostly;
-unsigned int swapper_pgd_zero __read_mostly;
-
-static pmd_t *prompmd __read_mostly;
-
-#define BASE_PAGE_SIZE 8192
-
 /*
  * Translate PROM's mapping we capture at boot time into physical address.
  * The second parameter is only set from prom_callback() invocations.
  */
 unsigned long prom_virt_to_phys(unsigned long promva, int *error)
 {
-	pmd_t *pmdp = prompmd + ((promva >> 23) & 0x7ff);
-	pte_t *ptep;
-	unsigned long base;
-
-	if (pmd_none(*pmdp)) {
-		if (error)
-			*error = 1;
-		return 0;
-	}
-	ptep = (pte_t *)__pmd_page(*pmdp) + ((promva >> 13) & 0x3ff);
-	if (!pte_present(*ptep)) {
-		if (error)
-			*error = 1;
-		return 0;
-	}
-	if (error) {
-		*error = 0;
-		return pte_val(*ptep);
-	}
-	base = pte_val(*ptep) & _PAGE_PADDR;
-
-	return base + (promva & (BASE_PAGE_SIZE - 1));
-}
+	int i;
 
-/* The obp translations are saved based on 8k pagesize, since obp can
- * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
- * HI_OBP_ADDRESS range are handled in entry.S and do not use the vpte
- * scheme (also, see rant in inherit_locked_prom_mappings()).
- */
-static void __init build_obp_range(unsigned long start, unsigned long end, unsigned long data)
-{
-	unsigned long vaddr;
+	for (i = 0; i < prom_trans_ents; i++) {
+		struct linux_prom_translation *p = &prom_trans[i];
 
-	for (vaddr = start; vaddr < end; vaddr += BASE_PAGE_SIZE) {
-		unsigned long val;
-		pmd_t *pmd;
-		pte_t *pte;
+		if (promva >= p->virt &&
+		    promva < (p->virt + p->size)) {
+			unsigned long base = p->data & _PAGE_PADDR;
 
-		pmd = prompmd + ((vaddr >> 23) & 0x7ff);
-		if (pmd_none(*pmd)) {
-			pte = __alloc_bootmem(BASE_PAGE_SIZE, BASE_PAGE_SIZE,
-					      PAGE_SIZE);
-			if (!pte)
-				prom_halt();
-			memset(pte, 0, BASE_PAGE_SIZE);
-			pmd_set(pmd, pte);
+			if (error)
+				*error = 0;
+			return base + (promva & (8192 - 1));
 		}
-		pte = (pte_t *) __pmd_page(*pmd) + ((vaddr >> 13) & 0x3ff);
-
-		val = data;
-
-		/* Clear diag TTE bits. */
-		if (tlb_type == spitfire)
-			val &= ~0x0003fe0000000000UL;
-
-		set_pte_at(&init_mm, vaddr, pte,
-			   __pte(val | _PAGE_MODIFIED));
-
-		data += BASE_PAGE_SIZE;
 	}
+	if (error)
+		*error = 1;
+	return 0UL;
 }
 
+/* The obp translations are saved based on 8k pagesize, since obp can
+ * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
+ * HI_OBP_ADDRESS range are handled in ktlb.S and do not use the vpte
+ * scheme (also, see rant in inherit_locked_prom_mappings()).
+ */
 static inline int in_obp_range(unsigned long vaddr)
 {
 	return (vaddr >= LOW_OBP_ADDRESS &&
 		vaddr < HI_OBP_ADDRESS);
 }
 
-#define OBP_PMD_SIZE 2048
-static void __init build_obp_pgtable(void)
+static int cmp_ptrans(const void *a, const void *b)
 {
-	unsigned long i;
-
-	prompmd = __alloc_bootmem(OBP_PMD_SIZE, OBP_PMD_SIZE, PAGE_SIZE);
-	if (!prompmd)
-		prom_halt();
-
-	memset(prompmd, 0, OBP_PMD_SIZE);
-
-	prom_pmd_phys = __pa(prompmd);
-
-	for (i = 0; i < prom_trans_ents; i++) {
-		unsigned long start, end;
-
-		if (!in_obp_range(prom_trans[i].virt))
-			continue;
+	const struct linux_prom_translation *x = a, *y = b;
 
-		start = prom_trans[i].virt;
-		end = start + prom_trans[i].size;
-		if (end > HI_OBP_ADDRESS)
-			end = HI_OBP_ADDRESS;
-
-		build_obp_range(start, end, prom_trans[i].data);
-	}
+	if (x->virt > y->virt)
+		return 1;
+	if (x->virt < y->virt)
+		return -1;
+	return 0;
 }
 
-/* Read OBP translations property into 'prom_trans[]'.
- * Return the number of entries.
- */
+/* Read OBP translations property into 'prom_trans[]'.  */
 static void __init read_obp_translations(void)
 {
-	int n, node;
+	int n, node, ents, first, last, i;
 
 	node = prom_finddevice("/virtual-memory");
 	n = prom_getproplen(node, "translations");
@@ -515,7 +453,41 @@ static void __init read_obp_translations(void)
 
 	n = n / sizeof(struct linux_prom_translation);
 
-	prom_trans_ents = n;
+	ents = n;
+
+	sort(prom_trans, ents, sizeof(struct linux_prom_translation),
+	     cmp_ptrans, NULL);
+
+	/* Now kick out all the non-OBP entries.  */
+	for (i = 0; i < ents; i++) {
+		if (in_obp_range(prom_trans[i].virt))
+			break;
+	}
+	first = i;
+	for (; i < ents; i++) {
+		if (!in_obp_range(prom_trans[i].virt))
+			break;
+	}
+	last = i;
+
+	for (i = 0; i < (last - first); i++) {
+		struct linux_prom_translation *src = &prom_trans[i + first];
+		struct linux_prom_translation *dest = &prom_trans[i];
+
+		*dest = *src;
+	}
+	for (; i < ents; i++) {
+		struct linux_prom_translation *dest = &prom_trans[i];
+		dest->virt = dest->size = dest->data = 0x0UL;
+	}
+
+	prom_trans_ents = last - first;
+
+	if (tlb_type == spitfire) {
+		/* Clear diag TTE bits. */
+		for (i = 0; i < prom_trans_ents; i++)
+			prom_trans[i].data &= ~0x0003fe0000000000UL;
+	}
 }
 
 static void __init remap_kernel(void)
@@ -553,21 +525,18 @@ static void __init remap_kernel(void)
 }
 
 
-static void __init inherit_prom_mappings_pre(void)
+static void __init inherit_prom_mappings(void)
 {
 	read_obp_translations();
 
 	/* Now fixup OBP's idea about where we really are mapped. */
 	prom_printf("Remapping the kernel... ");
 	remap_kernel();
-
 	prom_printf("done.\n");
-}
 
-static void __init inherit_prom_mappings_post(void)
-{
-	build_obp_pgtable();
+	prom_printf("Registering callbacks... ");
 	register_prom_callbacks();
+	prom_printf("done.\n");
 }
 
 /* The OBP specifications for sun4u mark 0xfffffffc00000000 and
@@ -1519,7 +1488,7 @@ void __init paging_init(void)
 	
 	swapper_pgd_zero = pgd_val(swapper_pg_dir[0]);
 	
-	inherit_prom_mappings_pre();
+	inherit_prom_mappings();
 	
 	/* Ok, we can use our TLB miss and window trap handlers safely.
 	 * We need to do a quick peek here to see if we are on StarFire
@@ -1530,23 +1499,15 @@ void __init paging_init(void)
 		extern void setup_tba(int);
 		setup_tba(this_is_starfire);
 	}
-	__flush_tlb_all();
 
-	/* Everything from this point forward, until we are done with
-	 * inherit_prom_mappings_post(), must complete successfully
-	 * without calling into the firmware.  The firwmare page tables
-	 * have not been built, but we are running on the Linux kernel's
-	 * trap table.
-	 */
+	inherit_locked_prom_mappings(1);
+
+	__flush_tlb_all();
 
 	/* Setup bootmem... */
 	pages_avail = 0;
 	last_valid_pfn = end_pfn = bootmem_init(&pages_avail);
 
-	inherit_prom_mappings_post();
-
-	inherit_locked_prom_mappings(1);
-
 #ifdef CONFIG_DEBUG_PAGEALLOC
 	kernel_physical_mapping_init();
 #endif