[SPARC64]: Detect sun4v early in boot process.

We look for "SUNW,sun4v" in the 'compatible' property
of the root OBP device tree node.

Protect every %ver register access, to make sure it is
not touched on sun4v, as %ver is hyperprivileged there.

Lock kernel TLB entries using hypervisor calls instead of
calls into OBP.

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

7 files changed, 204 insertions, 53 deletions

diff --git a/arch/sparc64/kernel/cpu.c b/arch/sparc64/kernel/cpu.c
index 00eed88ef2e8..c7a4fb397024 100644
--- a/arch/sparc64/kernel/cpu.c
+++ b/arch/sparc64/kernel/cpu.c
@@ -71,6 +71,12 @@ void __init cpu_probe(void)
 	unsigned long ver, fpu_vers, manuf, impl, fprs;
 	int i;
 	
+	if (tlb_type == hypervisor) {
+		sparc_cpu_type = "UltraSparc T1 (Niagara)";
+		sparc_fpu_type = "UltraSparc T1 integrated FPU";
+		return;
+	}
+
 	fprs = fprs_read();
 	fprs_write(FPRS_FEF);
 	__asm__ __volatile__ ("rdpr %%ver, %0; stx %%fsr, [%1]"
diff --git a/arch/sparc64/kernel/head.S b/arch/sparc64/kernel/head.S
index 01980014aead..d048f0dfd423 100644
--- a/arch/sparc64/kernel/head.S
+++ b/arch/sparc64/kernel/head.S
@@ -95,12 +95,17 @@ sparc64_boot:
 	wrpr	%g1, 0x0, %pstate
 	ba,a,pt	%xcc, 1f
 
-	.globl	prom_finddev_name, prom_chosen_path
-	.globl	prom_getprop_name, prom_mmu_name
-	.globl	prom_callmethod_name, prom_translate_name
+	.globl	prom_finddev_name, prom_chosen_path, prom_root_node
+	.globl	prom_getprop_name, prom_mmu_name, prom_peer_name
+	.globl	prom_callmethod_name, prom_translate_name, prom_root_compatible
 	.globl	prom_map_name, prom_unmap_name, prom_mmu_ihandle_cache
 	.globl	prom_boot_mapped_pc, prom_boot_mapping_mode
 	.globl	prom_boot_mapping_phys_high, prom_boot_mapping_phys_low
+	.globl	is_sun4v
+prom_peer_name:
+	.asciz	"peer"
+prom_compatible_name:
+	.asciz	"compatible"
 prom_finddev_name:
 	.asciz	"finddevice"
 prom_chosen_path:
@@ -117,7 +122,13 @@ prom_map_name:
 	.asciz	"map"
 prom_unmap_name:
 	.asciz	"unmap"
+prom_sun4v_name:
+	.asciz	"SUNW,sun4v"
 	.align	4
+prom_root_compatible:
+	.skip	64
+prom_root_node:
+	.word	0
 prom_mmu_ihandle_cache:
 	.word	0
 prom_boot_mapped_pc:
@@ -129,8 +140,54 @@ prom_boot_mapping_phys_high:
 	.xword	0
 prom_boot_mapping_phys_low:
 	.xword	0
+is_sun4v:
+	.word	0
 1:
 	rd	%pc, %l0
+
+	mov	(1b - prom_peer_name), %l1
+	sub	%l0, %l1, %l1
+	mov	0, %l2
+
+	/* prom_root_node = prom_peer(0) */
+	stx	%l1, [%sp + 2047 + 128 + 0x00]	! service, "peer"
+	mov	1, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x08]	! num_args, 1
+	stx	%l3, [%sp + 2047 + 128 + 0x10]	! num_rets, 1
+	stx	%l2, [%sp + 2047 + 128 + 0x18]	! arg1, 0
+	stx	%g0, [%sp + 2047 + 128 + 0x20]	! ret1
+	call	%l7
+	 add	%sp, (2047 + 128), %o0		! argument array
+
+	ldx	[%sp + 2047 + 128 + 0x20], %l4	! prom root node
+	mov	(1b - prom_root_node), %l1
+	sub	%l0, %l1, %l1
+	stw	%l4, [%l1]
+
+	mov	(1b - prom_getprop_name), %l1
+	mov	(1b - prom_compatible_name), %l2
+	mov	(1b - prom_root_compatible), %l5
+	sub	%l0, %l1, %l1
+	sub	%l0, %l2, %l2
+	sub	%l0, %l5, %l5
+
+	/* prom_getproperty(prom_root_node, "compatible",
+	 *                  &prom_root_compatible, 64)
+	 */
+	stx	%l1, [%sp + 2047 + 128 + 0x00]	! service, "getprop"
+	mov	4, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x08]	! num_args, 4
+	mov	1, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x10]	! num_rets, 1
+	stx	%l4, [%sp + 2047 + 128 + 0x18]	! arg1, prom_root_node
+	stx	%l2, [%sp + 2047 + 128 + 0x20]	! arg2, "compatible"
+	stx	%l5, [%sp + 2047 + 128 + 0x28]	! arg3, &prom_root_compatible
+	mov	64, %l3
+	stx	%l3, [%sp + 2047 + 128 + 0x30]	! arg4, size
+	stx	%g0, [%sp + 2047 + 128 + 0x38]	! ret1
+	call	%l7
+	 add	%sp, (2047 + 128), %o0		! argument array
+
 	mov	(1b - prom_finddev_name), %l1
 	mov	(1b - prom_chosen_path), %l2
 	mov	(1b - prom_boot_mapped_pc), %l3
@@ -239,6 +296,27 @@ prom_boot_mapping_phys_low:
 	add	%sp, (192 + 128), %sp
 
 sparc64_boot_after_remap:
+	sethi	%hi(prom_root_compatible), %g1
+	or	%g1, %lo(prom_root_compatible), %g1
+	sethi	%hi(prom_sun4v_name), %g7
+	or	%g7, %lo(prom_sun4v_name), %g7
+	mov	10, %g3
+1:	ldub	[%g7], %g2
+	ldub	[%g1], %g4
+	cmp	%g2, %g4
+	bne,pn	%icc, 2f
+	 add	%g7, 1, %g7
+	subcc	%g3, 1, %g3
+	bne,pt	%xcc, 1b
+	 add	%g1, 1, %g1
+
+	sethi	%hi(is_sun4v), %g1
+	or	%g1, %lo(is_sun4v), %g1
+	mov	1, %g7
+	stw	%g7, [%g1]
+
+2:
+	BRANCH_IF_SUN4V(g1, jump_to_sun4u_init)
 	BRANCH_IF_CHEETAH_BASE(g1,g7,cheetah_boot)
 	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,cheetah_plus_boot)
 	ba,pt	%xcc, spitfire_boot
@@ -323,14 +401,12 @@ sun4u_init:
 
 	membar	#Sync
 
-	BRANCH_IF_ANY_CHEETAH(g1,g7,cheetah_tlb_fixup)
+	BRANCH_IF_SUN4V(g1, niagara_tlb_fixup)
+	BRANCH_IF_ANY_CHEETAH(g1, g7, cheetah_tlb_fixup)
 
 	ba,pt	%xcc, spitfire_tlb_fixup
 	 nop
 
-	/* XXX Nothing branches to here yet, when %ver register indicates
-	 * XXX Niagara we should do this.
-	 */
 niagara_tlb_fixup:
 	mov	3, %g2		/* Set TLB type to hypervisor. */
 	sethi	%hi(tlb_type), %g1
@@ -346,6 +422,9 @@ niagara_tlb_fixup:
 	call	hypervisor_patch_cachetlbops
 	 nop
 
+	ba,pt	%xcc, tlb_fixup_done
+	 nop
+
 cheetah_tlb_fixup:
 	mov	2, %g2		/* Set TLB type to cheetah+. */
 	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g7,1f)
@@ -464,7 +543,8 @@ setup_trap_table:
 	sllx	%o2, 32, %o2
 	wr	%o2, 0, %tick_cmpr
 
-	BRANCH_IF_ANY_CHEETAH(o2,o3,1f)
+	BRANCH_IF_SUN4V(o2, 1f)
+	BRANCH_IF_ANY_CHEETAH(o2, o3, 1f)
 
 	ba,pt	%xcc, 2f
 	 nop
diff --git a/arch/sparc64/kernel/irq.c b/arch/sparc64/kernel/irq.c
index c80d2531ec46..1f6455503f24 100644
--- a/arch/sparc64/kernel/irq.c
+++ b/arch/sparc64/kernel/irq.c
@@ -150,47 +150,53 @@ void enable_irq(unsigned int irq)
 
 	preempt_disable();
 
-	if (tlb_type == cheetah || tlb_type == cheetah_plus) {
-		unsigned long ver;
-
-		__asm__ ("rdpr %%ver, %0" : "=r" (ver));
-		if ((ver >> 32) == __JALAPENO_ID ||
-		    (ver >> 32) == __SERRANO_ID) {
-			/* We set it to our JBUS ID. */
+	if (tlb_type == hypervisor) {
+		/* XXX SUN4V: implement me... XXX */
+	} else {
+		if (tlb_type == cheetah || tlb_type == cheetah_plus) {
+			unsigned long ver;
+
+			__asm__ ("rdpr %%ver, %0" : "=r" (ver));
+			if ((ver >> 32) == __JALAPENO_ID ||
+			    (ver >> 32) == __SERRANO_ID) {
+				/* We set it to our JBUS ID. */
+				__asm__ __volatile__("ldxa [%%g0] %1, %0"
+						     : "=r" (tid)
+						     : "i" (ASI_JBUS_CONFIG));
+				tid = ((tid & (0x1fUL<<17)) << 9);
+				tid &= IMAP_TID_JBUS;
+			} else {
+				/* We set it to our Safari AID. */
+				__asm__ __volatile__("ldxa [%%g0] %1, %0"
+						     : "=r" (tid)
+						     : "i"(ASI_SAFARI_CONFIG));
+				tid = ((tid & (0x3ffUL<<17)) << 9);
+				tid &= IMAP_AID_SAFARI;
+			}
+		} else if (this_is_starfire == 0) {
+			/* We set it to our UPA MID. */
 			__asm__ __volatile__("ldxa [%%g0] %1, %0"
 					     : "=r" (tid)
-					     : "i" (ASI_JBUS_CONFIG));
-			tid = ((tid & (0x1fUL<<17)) << 9);
-			tid &= IMAP_TID_JBUS;
+					     : "i" (ASI_UPA_CONFIG));
+			tid = ((tid & UPA_CONFIG_MID) << 9);
+			tid &= IMAP_TID_UPA;
 		} else {
-			/* We set it to our Safari AID. */
-			__asm__ __volatile__("ldxa [%%g0] %1, %0"
-					     : "=r" (tid)
-					     : "i" (ASI_SAFARI_CONFIG));
-			tid = ((tid & (0x3ffUL<<17)) << 9);
-			tid &= IMAP_AID_SAFARI;
+			tid = (starfire_translate(imap,
+						  smp_processor_id()) << 26);
+			tid &= IMAP_TID_UPA;
 		}
-	} else if (this_is_starfire == 0) {
-		/* We set it to our UPA MID. */
-		__asm__ __volatile__("ldxa [%%g0] %1, %0"
-				     : "=r" (tid)
-				     : "i" (ASI_UPA_CONFIG));
-		tid = ((tid & UPA_CONFIG_MID) << 9);
-		tid &= IMAP_TID_UPA;
-	} else {
-		tid = (starfire_translate(imap, smp_processor_id()) << 26);
-		tid &= IMAP_TID_UPA;
-	}
 
-	/* NOTE NOTE NOTE, IGN and INO are read-only, IGN is a product
-	 * of this SYSIO's preconfigured IGN in the SYSIO Control
-	 * Register, the hardware just mirrors that value here.
-	 * However for Graphics and UPA Slave devices the full
-	 * IMAP_INR field can be set by the programmer here.
-	 *
-	 * Things like FFB can now be handled via the new IRQ mechanism.
-	 */
-	upa_writel(tid | IMAP_VALID, imap);
+		/* NOTE NOTE NOTE, IGN and INO are read-only, IGN is a product
+		 * of this SYSIO's preconfigured IGN in the SYSIO Control
+		 * Register, the hardware just mirrors that value here.
+		 * However for Graphics and UPA Slave devices the full
+		 * IMAP_INR field can be set by the programmer here.
+		 *
+		 * Things like FFB can now be handled via the new IRQ
+		 * mechanism.
+		 */
+		upa_writel(tid | IMAP_VALID, imap);
+	}
 
 	preempt_enable();
 }
diff --git a/arch/sparc64/kernel/setup.c b/arch/sparc64/kernel/setup.c
index 2d64320d3a4d..7f02c8f71df6 100644
--- a/arch/sparc64/kernel/setup.c
+++ b/arch/sparc64/kernel/setup.c
@@ -504,9 +504,12 @@ static void __init per_cpu_patch(void)
 	if (tlb_type == spitfire && !this_is_starfire)
 		return;
 
-	__asm__ ("rdpr %%ver, %0" : "=r" (ver));
-	is_jbus = ((ver >> 32) == __JALAPENO_ID ||
-		   (ver >> 32) == __SERRANO_ID);
+	is_jbus = 0;
+	if (tlb_type != hypervisor) {
+		__asm__ ("rdpr %%ver, %0" : "=r" (ver));
+		is_jbus = ((ver >> 32) == __JALAPENO_ID ||
+			   (ver >> 32) == __SERRANO_ID);
+	}
 
 	p = &__cpuid_patch;
 	while (p < &__cpuid_patch_end) {
diff --git a/arch/sparc64/kernel/trampoline.S b/arch/sparc64/kernel/trampoline.S
index fbf844f84a49..ffa8b79632cf 100644
--- a/arch/sparc64/kernel/trampoline.S
+++ b/arch/sparc64/kernel/trampoline.S
@@ -16,6 +16,7 @@
 #include <asm/processor.h>
 #include <asm/thread_info.h>
 #include <asm/mmu.h>
+#include <asm/hypervisor.h>
 
 	.data
 	.align	8
@@ -34,8 +35,9 @@ dtlb_load:
 sparc64_cpu_startup:
 	flushw
 
-	BRANCH_IF_CHEETAH_BASE(g1,g5,cheetah_startup)
-	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1,g5,cheetah_plus_startup)
+	BRANCH_IF_SUN4V(g1, niagara_startup)
+	BRANCH_IF_CHEETAH_BASE(g1, g5, cheetah_startup)
+	BRANCH_IF_CHEETAH_PLUS_OR_FOLLOWON(g1, g5, cheetah_plus_startup)
 
 	ba,pt	%xcc, spitfire_startup
 	 nop
@@ -70,7 +72,9 @@ cheetah_generic_startup:
 	stxa	%g0, [%g3] ASI_DMMU
 	stxa	%g0, [%g3] ASI_IMMU
 	membar	#Sync
+	/* fallthru */
 
+niagara_startup:
 	/* Disable STICK_INT interrupts. */
 	sethi		%hi(0x80000000), %g5
 	sllx		%g5, 32, %g5
@@ -91,6 +95,8 @@ startup_continue:
 	sllx		%g2, 32, %g2
 	wr		%g2, 0, %tick_cmpr
 
+	BRANCH_IF_SUN4V(g1, niagara_lock_tlb)
+
 	/* Call OBP by hand to lock KERNBASE into i/d tlbs.
 	 * We lock 2 consequetive entries if we are 'bigkernel'.
 	 */
@@ -142,8 +148,7 @@ startup_continue:
 
 	sethi		%hi(bigkernel), %g2
 	lduw		[%g2 + %lo(bigkernel)], %g2
-	cmp		%g2, 0
-	be,pt		%icc, do_dtlb
+	brz,pt		%g2, do_dtlb
 	 nop
 
 	sethi		%hi(call_method), %g2
@@ -214,8 +219,7 @@ do_dtlb:
 
 	sethi		%hi(bigkernel), %g2
 	lduw		[%g2 + %lo(bigkernel)], %g2
-	cmp		%g2, 0
-	be,pt		%icc, do_unlock
+	brz,pt		%g2, do_unlock
 	 nop
 
 	sethi		%hi(call_method), %g2
@@ -257,6 +261,52 @@ do_unlock:
 	stb		%g0, [%g2 + %lo(prom_entry_lock)]
 	membar		#StoreStore | #StoreLoad
 
+	ba,pt		%xcc, after_lock_tlb
+	 nop
+
+niagara_lock_tlb:
+	mov		HV_FAST_MMU_MAP_PERM_ADDR, %o0
+	sethi		%hi(KERNBASE), %o1
+	clr		%o2
+	sethi		%hi(kern_locked_tte_data), %o3
+	ldx		[%o3 + %lo(kern_locked_tte_data)], %o3
+	mov		HV_MMU_IMMU, %o4
+	ta		HV_FAST_TRAP
+
+	mov		HV_FAST_MMU_MAP_PERM_ADDR, %o0
+	sethi		%hi(KERNBASE), %o1
+	clr		%o2
+	sethi		%hi(kern_locked_tte_data), %o3
+	ldx		[%o3 + %lo(kern_locked_tte_data)], %o3
+	mov		HV_MMU_DMMU, %o4
+	ta		HV_FAST_TRAP
+
+	sethi		%hi(bigkernel), %g2
+	lduw		[%g2 + %lo(bigkernel)], %g2
+	brz,pt		%g2, after_lock_tlb
+	 nop
+
+	mov		HV_FAST_MMU_MAP_PERM_ADDR, %o0
+	sethi		%hi(KERNBASE + 0x400000), %o1
+	clr		%o2
+	sethi		%hi(kern_locked_tte_data), %o3
+	ldx		[%o3 + %lo(kern_locked_tte_data)], %o3
+	sethi		%hi(0x400000), %o4
+	add		%o3, %o4, %o3
+	mov		HV_MMU_IMMU, %o4
+	ta		HV_FAST_TRAP
+
+	mov		HV_FAST_MMU_MAP_PERM_ADDR, %o0
+	sethi		%hi(KERNBASE + 0x400000), %o1
+	clr		%o2
+	sethi		%hi(kern_locked_tte_data), %o3
+	ldx		[%o3 + %lo(kern_locked_tte_data)], %o3
+	sethi		%hi(0x400000), %o4
+	add		%o3, %o4, %o3
+	mov		HV_MMU_DMMU, %o4
+	ta		HV_FAST_TRAP
+
+after_lock_tlb:
 	mov		%l1, %sp
 	flushw
 
diff --git a/arch/sparc64/kernel/us2e_cpufreq.c b/arch/sparc64/kernel/us2e_cpufreq.c
index b35dc8dc995a..669fb83dd4f5 100644
--- a/arch/sparc64/kernel/us2e_cpufreq.c
+++ b/arch/sparc64/kernel/us2e_cpufreq.c
@@ -346,6 +346,9 @@ static int __init us2e_freq_init(void)
 	unsigned long manuf, impl, ver;
 	int ret;
 
+	if (tlb_type != spitfire)
+		return -ENODEV;
+
 	__asm__("rdpr %%ver, %0" : "=r" (ver));
 	manuf = ((ver >> 48) & 0xffff);
 	impl  = ((ver >> 32) & 0xffff);
diff --git a/arch/sparc64/kernel/us3_cpufreq.c b/arch/sparc64/kernel/us3_cpufreq.c
index 6d1f9a3c464f..a912c45bdc05 100644
--- a/arch/sparc64/kernel/us3_cpufreq.c
+++ b/arch/sparc64/kernel/us3_cpufreq.c
@@ -203,6 +203,9 @@ static int __init us3_freq_init(void)
 	unsigned long manuf, impl, ver;
 	int ret;
 
+	if (tlb_type != cheetah && tlb_type != cheetah_plus)
+		return -ENODEV;
+
 	__asm__("rdpr %%ver, %0" : "=r" (ver));
 	manuf = ((ver >> 48) & 0xffff);
 	impl  = ((ver >> 32) & 0xffff);