[SPARC64]: Refine code sequences to get the cpu id.

On uniprocessor, it's always zero for optimize that.

On SMP, the jmpl to the stub kills the return address stack in the cpu
branch prediction logic, so expand the code sequence inline and use a
code patching section to fix things up.  This also always better and
explicit register selection, which will be taken advantage of in a
future changeset.

The hard_smp_processor_id() function is big, so do not inline it.

Fix up tests for Jalapeno to also test for Serrano chips too.  These
tests want "jbus Ultra-IIIi" cases to match, so that is what we should
test for.

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

6 files changed, 73 insertions, 87 deletions

diff --git a/arch/sparc64/kernel/entry.S b/arch/sparc64/kernel/entry.S
index 563fa4e..b3511ff 100644
--- a/arch/sparc64/kernel/entry.S
+++ b/arch/sparc64/kernel/entry.S
@@ -1628,84 +1628,10 @@ __flushw_user:
 2:	retl
 	 nop
 
-	/* Read cpu ID from hardware, return in %g6.
-	 * (callers_pc - 4) is in %g1.  Patched at boot time.
-	 *
-	 * Default is spitfire implementation.
-	 *
-	 * The instruction sequence needs to be 5 instructions
-	 * in order to fit the longest implementation, which is
-	 * currently starfire.
-	 */
-	.align		32
-	.globl		__get_cpu_id
-__get_cpu_id:
-	ldxa		[%g0] ASI_UPA_CONFIG, %g6
-	srlx		%g6, 17, %g6
-	jmpl		%g1 + 0x4, %g0
-	 and		%g6, 0x1f, %g6
-	nop
-
-__get_cpu_id_cheetah_safari:
-	ldxa		[%g0] ASI_SAFARI_CONFIG, %g6
-	srlx		%g6, 17, %g6
-	jmpl		%g1 + 0x4, %g0
-	 and		%g6, 0x3ff, %g6
-	nop
-
-__get_cpu_id_cheetah_jbus:
-	ldxa		[%g0] ASI_JBUS_CONFIG, %g6
-	srlx		%g6, 17, %g6
-	jmpl		%g1 + 0x4, %g0
-	 and		%g6, 0x1f, %g6
-	nop
-
-__get_cpu_id_starfire:
-	sethi		%hi(0x1fff40000d0 >> 9), %g6
-	sllx		%g6, 9, %g6
-	or		%g6, 0xd0, %g6
-	jmpl		%g1 + 0x4, %g0
-	 lduwa		[%g6] ASI_PHYS_BYPASS_EC_E, %g6
-
-	.globl		per_cpu_patch
-per_cpu_patch:
-	sethi		%hi(this_is_starfire), %o0
-	lduw		[%o0 + %lo(this_is_starfire)], %o1
-	sethi		%hi(__get_cpu_id_starfire), %o0
-	brnz,pn		%o1, 10f
-	 or		%o0, %lo(__get_cpu_id_starfire), %o0
-	sethi		%hi(tlb_type), %o0
-	lduw		[%o0 + %lo(tlb_type)], %o1
-	brz,pt		%o1, 11f
-	 nop
-	rdpr		%ver, %o0
-	srlx		%o0, 32, %o0
-	sethi		%hi(0x003e0016), %o1
-	or		%o1, %lo(0x003e0016), %o1
-	cmp		%o0, %o1
-	sethi		%hi(__get_cpu_id_cheetah_jbus), %o0
-	be,pn		%icc, 10f
-	 or		%o0, %lo(__get_cpu_id_cheetah_jbus), %o0
-	sethi		%hi(__get_cpu_id_cheetah_safari), %o0
-	or		%o0, %lo(__get_cpu_id_cheetah_safari), %o0
-10:
-	sethi		%hi(__get_cpu_id), %o1
-	or		%o1, %lo(__get_cpu_id), %o1
-	lduw		[%o0 + 0x00], %o2
-	stw		%o2, [%o1 + 0x00]
-	flush		%o1 + 0x00
-	lduw		[%o0 + 0x04], %o2
-	stw		%o2, [%o1 + 0x04]
-	flush		%o1 + 0x04
-	lduw		[%o0 + 0x08], %o2
-	stw		%o2, [%o1 + 0x08]
-	flush		%o1 + 0x08
-	lduw		[%o0 + 0x0c], %o2
-	stw		%o2, [%o1 + 0x0c]
-	flush		%o1 + 0x0c
-	lduw		[%o0 + 0x10], %o2
-	stw		%o2, [%o1 + 0x10]
-	flush		%o1 + 0x10
-11:
+#ifdef CONFIG_SMP
+	.globl		hard_smp_processor_id
+hard_smp_processor_id:
+	__GET_CPUID(%o0)
 	retl
 	 nop
+#endif
diff --git a/arch/sparc64/kernel/irq.c b/arch/sparc64/kernel/irq.c
index 3e48af2..d069a6f 100644
--- a/arch/sparc64/kernel/irq.c
+++ b/arch/sparc64/kernel/irq.c
@@ -39,6 +39,7 @@
 #include <asm/cache.h>
 #include <asm/cpudata.h>
 #include <asm/auxio.h>
+#include <asm/head.h>
 
 #ifdef CONFIG_SMP
 static void distribute_irqs(void);
@@ -153,7 +154,8 @@ void enable_irq(unsigned int irq)
 		unsigned long ver;
 
 		__asm__ ("rdpr %%ver, %0" : "=r" (ver));
-		if ((ver >> 32) == 0x003e0016) {
+		if ((ver >> 32) == __JALAPENO_ID ||
+		    (ver >> 32) == __SERRANO_ID) {
 			/* We set it to our JBUS ID. */
 			__asm__ __volatile__("ldxa [%%g0] %1, %0"
 					     : "=r" (tid)
diff --git a/arch/sparc64/kernel/setup.c b/arch/sparc64/kernel/setup.c
index 59a7030..f751d11 100644
--- a/arch/sparc64/kernel/setup.c
+++ b/arch/sparc64/kernel/setup.c
@@ -490,6 +490,58 @@ void register_prom_callbacks(void)
 		   "' linux-.soft2 to .soft2");
 }
 
+static void __init per_cpu_patch(void)
+{
+#ifdef CONFIG_SMP
+	struct cpuid_patch_entry *p;
+	unsigned long ver;
+	int is_jbus;
+
+	if (tlb_type == spitfire && !this_is_starfire)
+		return;
+
+	__asm__ ("rdpr %%ver, %0" : "=r" (ver));
+	is_jbus = ((ver >> 32) == __JALAPENO_ID ||
+		   (ver >> 32) == __SERRANO_ID);
+
+	p = &__cpuid_patch;
+	while (p < &__cpuid_patch_end) {
+		unsigned long addr = p->addr;
+		unsigned int *insns;
+
+		switch (tlb_type) {
+		case spitfire:
+			insns = &p->starfire[0];
+			break;
+		case cheetah:
+		case cheetah_plus:
+			if (is_jbus)
+				insns = &p->cheetah_jbus[0];
+			else
+				insns = &p->cheetah_safari[0];
+			break;
+		default:
+			prom_printf("Unknown cpu type, halting.\n");
+			prom_halt();
+		};
+
+		*(unsigned int *) (addr +  0) = insns[0];
+		__asm__ __volatile__("flush	%0" : : "r" (addr +  0));
+
+		*(unsigned int *) (addr +  4) = insns[1];
+		__asm__ __volatile__("flush	%0" : : "r" (addr +  4));
+
+		*(unsigned int *) (addr +  8) = insns[2];
+		__asm__ __volatile__("flush	%0" : : "r" (addr +  8));
+
+		*(unsigned int *) (addr + 12) = insns[3];
+		__asm__ __volatile__("flush	%0" : : "r" (addr + 12));
+
+		p++;
+	}
+#endif
+}
+
 void __init setup_arch(char **cmdline_p)
 {
 	/* Initialize PROM console and command line. */
@@ -507,8 +559,8 @@ void __init setup_arch(char **cmdline_p)
 	/* Work out if we are starfire early on */
 	check_if_starfire();
 
-	/* Now we know enough to patch the __get_cpu_id()
-	 * trampoline used by trap code.
+	/* Now we know enough to patch the get_cpuid sequences
+	 * used by trap code.
 	 */
 	per_cpu_patch();
 
diff --git a/arch/sparc64/kernel/smp.c b/arch/sparc64/kernel/smp.c
index 0e75525..16b8eca 100644
--- a/arch/sparc64/kernel/smp.c
+++ b/arch/sparc64/kernel/smp.c
@@ -424,7 +424,7 @@ static __inline__ void spitfire_xcall_deliver(u64 data0, u64 data1, u64 data2, c
 static void cheetah_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask)
 {
 	u64 pstate, ver;
-	int nack_busy_id, is_jalapeno;
+	int nack_busy_id, is_jbus;
 
 	if (cpus_empty(mask))
 		return;
@@ -434,7 +434,8 @@ static void cheetah_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mas
 	 * derivative processor.
 	 */
 	__asm__ ("rdpr %%ver, %0" : "=r" (ver));
-	is_jalapeno = ((ver >> 32) == 0x003e0016);
+	is_jbus = ((ver >> 32) == __JALAPENO_ID ||
+		   (ver >> 32) == __SERRANO_ID);
 
 	__asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
 
@@ -459,7 +460,7 @@ retry:
 		for_each_cpu_mask(i, mask) {
 			u64 target = (i << 14) | 0x70;
 
-			if (!is_jalapeno)
+			if (!is_jbus)
 				target |= (nack_busy_id << 24);
 			__asm__ __volatile__(
 				"stxa	%%g0, [%0] %1\n\t"
@@ -512,7 +513,7 @@ retry:
 			for_each_cpu_mask(i, mask) {
 				u64 check_mask;
 
-				if (is_jalapeno)
+				if (is_jbus)
 					check_mask = (0x2UL << (2*i));
 				else
 					check_mask = (0x2UL <<
diff --git a/arch/sparc64/kernel/traps.c b/arch/sparc64/kernel/traps.c
index 7e52e89..1c4744c 100644
--- a/arch/sparc64/kernel/traps.c
+++ b/arch/sparc64/kernel/traps.c
@@ -38,6 +38,7 @@
 #include <asm/processor.h>
 #include <asm/timer.h>
 #include <asm/kdebug.h>
+#include <asm/head.h>
 #ifdef CONFIG_KMOD
 #include <linux/kmod.h>
 #endif
@@ -788,7 +789,8 @@ void __init cheetah_ecache_flush_init(void)
 		cheetah_error_log[i].afsr = CHAFSR_INVALID;
 
 	__asm__ ("rdpr %%ver, %0" : "=r" (ver));
-	if ((ver >> 32) == 0x003e0016) {
+	if ((ver >> 32) == __JALAPENO_ID ||
+	    (ver >> 32) == __SERRANO_ID) {
 		cheetah_error_table = &__jalapeno_error_table[0];
 		cheetah_afsr_errors = JPAFSR_ERRORS;
 	} else if ((ver >> 32) == 0x003e0015) {
diff --git a/arch/sparc64/kernel/vmlinux.lds.S b/arch/sparc64/kernel/vmlinux.lds.S
index 71b943f..1639d9c 100644
--- a/arch/sparc64/kernel/vmlinux.lds.S
+++ b/arch/sparc64/kernel/vmlinux.lds.S
@@ -74,6 +74,9 @@ SECTIONS
   __tsb_phys_patch = .;
   .tsb_phys_patch : { *(.tsb_phys_patch) }
   __tsb_phys_patch_end = .;
+  __cpuid_patch = .;
+  .cpuid_patch : { *(.cpuid_patch) }
+  __cpuid_patch_end = .;
   . = ALIGN(8192); 
   __initramfs_start = .;
   .init.ramfs : { *(.init.ramfs) }