Merge davem@nuts.ninka.net:/home/davem/src/BK/sparc-2.5

into kernel.bkbits.net:/home/davem/sparc-2.5

arch/sparc64/kernel/cpu.c

@@ -12,6 +12,9 @@
 #include <asm/asi.h>
 #include <asm/system.h>
 #include <asm/fpumacro.h>
+#include <asm/cpudata.h>
+
+DEFINE_PER_CPU(cpuinfo_sparc, __cpu_data) = { 0 };
 
 struct cpu_iu_info {
   short manuf;
@@ -52,17 +55,15 @@ struct cpu_iu_info linux_sparc_chips[] = {
 
 #define NSPARCCHIPS  (sizeof(linux_sparc_chips)/sizeof(struct cpu_iu_info))
 
-char *sparc_cpu_type[NR_CPUS] = { "cpu-oops", };
-char *sparc_fpu_type[NR_CPUS] = { "fpu-oops", };
+char *sparc_cpu_type = "cpu-oops";
+char *sparc_fpu_type = "fpu-oops";
 
 unsigned int fsr_storage;
 
 void __init cpu_probe(void)
 {
 	unsigned long ver, fpu_vers, manuf, impl, fprs;
-	int i, cpuid;
-	
-	cpuid = hard_smp_processor_id();
+	int i;
 	
 	fprs = fprs_read();
 	fprs_write(FPRS_FEF);
@@ -80,7 +81,7 @@ void __init cpu_probe(void)
 	for (i = 0; i < NSPARCCHIPS; i++) {
 		if (linux_sparc_chips[i].manuf == manuf) {
 			if (linux_sparc_chips[i].impl == impl) {
-				sparc_cpu_type[cpuid] =
+				sparc_cpu_type =
 					linux_sparc_chips[i].cpu_name;
 				break;
 			}
@@ -99,14 +100,14 @@ void __init cpu_probe(void)
  			printk("DEBUG: manuf[%lx] impl[%lx]\n",
  			       manuf, impl);
  		}
-		sparc_cpu_type[cpuid] = "Unknown CPU";
+		sparc_cpu_type = "Unknown CPU";
 	}
 
 	for (i = 0; i < NSPARCFPU; i++) {
 		if (linux_sparc_fpu[i].manuf == manuf &&
 		    linux_sparc_fpu[i].impl == impl) {
 			if (linux_sparc_fpu[i].fpu_vers == fpu_vers) {
-				sparc_fpu_type[cpuid] =
+				sparc_fpu_type =
 					linux_sparc_fpu[i].fp_name;
 				break;
 			}
@@ -116,6 +117,6 @@ void __init cpu_probe(void)
 	if (i == NSPARCFPU) {
  		printk("DEBUG: manuf[%lx] impl[%lx] fsr.vers[%lx]\n",
  		       manuf, impl, fpu_vers);
-		sparc_fpu_type[cpuid] = "Unknown FPU";
+		sparc_fpu_type = "Unknown FPU";
 	}
 }

arch/sparc64/kernel/devices.c

@@ -11,6 +11,7 @@
 #include <linux/ioport.h>
 #include <linux/string.h>
 #include <linux/spinlock.h>
+#include <linux/errno.h>
 
 #include <asm/page.h>
 #include <asm/oplib.h>
@@ -18,77 +19,120 @@
 #include <asm/smp.h>
 #include <asm/spitfire.h>
 #include <asm/timer.h>
+#include <asm/cpudata.h>
 
 /* Used to synchronize acceses to NatSemi SUPER I/O chip configure
  * operations in asm/ns87303.h
  */
 spinlock_t ns87303_lock = SPIN_LOCK_UNLOCKED;
 
-struct prom_cpuinfo linux_cpus[NR_CPUS] __initdata = { { 0 } };
-unsigned prom_cpu_nodes[NR_CPUS];
-int linux_num_cpus = 0;
-
 extern void cpu_probe(void);
 extern void central_probe(void);
 
-void __init device_scan(void)
+static char *cpu_mid_prop(void)
 {
-	char node_str[128];
-	int nd, prom_node_cpu, thismid;
-	int cpu_nds[64];  /* One node for each cpu */
-	int cpu_ctr = 0;
+	if (tlb_type == spitfire)
+		return "upa-portid";
+	return "portid";
+}
 
-	/* FIX ME FAST... -DaveM */
-	ioport_resource.end = 0xffffffffffffffffUL;
+static int check_cpu_node(int nd, int *cur_inst,
+			  int (*compare)(int, int, void *), void *compare_arg,
+			  int *prom_node, int *mid)
+{
+	char node_str[128];
 
-	prom_getstring(prom_root_node, "device_type", node_str, sizeof(node_str));
+	prom_getstring(nd, "device_type", node_str, sizeof(node_str));
+	if (strcmp(node_str, "cpu"))
+		return -ENODEV;
 
-	prom_printf("Booting Linux...\n");
-	if(strcmp(node_str, "cpu") == 0) {
-		cpu_nds[0] = prom_root_node;
-		linux_cpus[0].prom_node = prom_root_node;
-		linux_cpus[0].mid = 0;
-		cpu_ctr++;
-	} else {
-		int scan;
-		scan = prom_getchild(prom_root_node);
-		/* prom_printf("root child is %08x\n", (unsigned) scan); */
-		nd = 0;
-		while((scan = prom_getsibling(scan)) != 0) {
-			prom_getstring(scan, "device_type", node_str, sizeof(node_str));
-			if(strcmp(node_str, "cpu") == 0) {
-				cpu_nds[cpu_ctr] = scan;
-				linux_cpus[cpu_ctr].prom_node = scan;
-				thismid = 0;
-				if (tlb_type == spitfire) {
-					prom_getproperty(scan, "upa-portid",
-							 (char *) &thismid, sizeof(thismid));
-				} else if (tlb_type == cheetah ||
-					   tlb_type == cheetah_plus) {
-					prom_getproperty(scan, "portid",
-							 (char *) &thismid, sizeof(thismid));
-				}
-				linux_cpus[cpu_ctr].mid = thismid;
-				printk("Found CPU %d (node=%08x,mid=%d)\n",
-				       cpu_ctr, (unsigned) scan, thismid);
-				cpu_ctr++;
-			}
-		};
-		if(cpu_ctr == 0) {
-			prom_printf("No CPU nodes found, cannot continue.\n");
-			prom_halt();
+	if (!compare(nd, *cur_inst, compare_arg)) {
+		if (prom_node)
+			*prom_node = nd;
+		if (mid)
+			*mid = prom_getintdefault(nd, cpu_mid_prop(), 0);
+		return 0;
 	}
-		printk("Found %d CPU prom device tree node(s).\n", cpu_ctr);
+
+	(*cur_inst)++;
+
+	return -ENODEV;
+}
+
+static int __cpu_find_by(int (*compare)(int, int, void *), void *compare_arg,
+			 int *prom_node, int *mid)
+{
+	int nd, cur_inst, err;
+
+	nd = prom_root_node;
+	cur_inst = 0;
+
+	err = check_cpu_node(nd, &cur_inst,
+			     compare, compare_arg,
+			     prom_node, mid);
+	if (err == 0)
+		return 0;
+
+	nd = prom_getchild(nd);
+	while ((nd = prom_getsibling(nd)) != 0) {
+		err = check_cpu_node(nd, &cur_inst,
+				     compare, compare_arg,
+				     prom_node, mid);
+		if (err == 0)
+			return 0;
 	}
-	prom_node_cpu = cpu_nds[0];
 
-	linux_num_cpus = cpu_ctr;
+	return -ENODEV;
+}
 
-	prom_cpu_nodes[0] = prom_node_cpu;
+static int cpu_instance_compare(int nd, int instance, void *_arg)
+{
+	int desired_instance = (int) (long) _arg;
+
+	if (instance == desired_instance)
+		return 0;
+	return -ENODEV;
+}
+
+int cpu_find_by_instance(int instance, int *prom_node, int *mid)
+{
+	return __cpu_find_by(cpu_instance_compare, (void *)(long)instance,
+			     prom_node, mid);
+}
+
+static int cpu_mid_compare(int nd, int instance, void *_arg)
+{
+	int desired_mid = (int) (long) _arg;
+	int this_mid;
+
+	this_mid = prom_getintdefault(nd, cpu_mid_prop(), 0);
+	if (this_mid == desired_mid)
+		return 0;
+	return -ENODEV;
+}
+
+int cpu_find_by_mid(int mid, int *prom_node)
+{
+	return __cpu_find_by(cpu_mid_compare, (void *)(long)mid,
+			     prom_node, NULL);
+}
+
+void __init device_scan(void)
+{
+	/* FIX ME FAST... -DaveM */
+	ioport_resource.end = 0xffffffffffffffffUL;
+
+	prom_printf("Booting Linux...\n");
 
 #ifndef CONFIG_SMP
 	{
-		up_clock_tick = prom_getintdefault(prom_node_cpu,
+		int err, cpu_node;
+		err = cpu_find_by_instance(0, &cpu_node, NULL);
+		if (err) {
+			prom_printf("No cpu nodes, cannot continue\n");
+			prom_halt();
+		}
+		cpu_data(0).clock_tick = prom_getintdefault(cpu_node,
 							    "clock-frequency",
 							    0);
 	}

arch/sparc64/kernel/ebus.c

@@ -13,6 +13,7 @@
 #include <linux/slab.h>
 #include <linux/string.h>
 #include <linux/interrupt.h>
+#include <linux/delay.h>
 
 #include <asm/system.h>
 #include <asm/page.h>

arch/sparc64/kernel/head.S

@@ -628,70 +628,21 @@ spitfire_vpte_base:
 #undef KERN_HIGHBITS
 #undef KERN_LOWBITS
 
-	/* Setup Interrupt globals */
-	wrpr	%o1, (PSTATE_IG|PSTATE_IE), %pstate
-#ifndef CONFIG_SMP
-	sethi	%hi(__up_workvec), %g5
-	or	%g5, %lo(__up_workvec), %g6
-#else
-	/* By definition of where we are, this is boot_cpu. */
-	brz,pt	%i0, not_starfire
-	 sethi	%hi(0x1fff4000), %g1
-	or	%g1, %lo(0x1fff4000), %g1
-	sllx	%g1, 12, %g1
-	or	%g1, 0xd0, %g1
-	lduwa	[%g1] ASI_PHYS_BYPASS_EC_E, %g1
-	b,pt	%xcc, set_worklist
-	 nop
-
-not_starfire:
-	BRANCH_IF_JALAPENO(g1,g5,is_jalapeno)
-	BRANCH_IF_ANY_CHEETAH(g1,g5,is_cheetah)
-
-	ba,pt	%xcc, not_cheetah
-	 nop
-
-is_jalapeno:
-	ldxa		[%g0] ASI_JBUS_CONFIG, %g1
-	srlx		%g1, 17, %g1
-	ba,pt		%xcc, set_worklist
-	 and		%g1, 0x1f, %g1		! 5bit JBUS ID
-
-is_cheetah:
-	ldxa		[%g0] ASI_SAFARI_CONFIG, %g1
-	srlx		%g1, 17, %g1
-	ba,pt		%xcc, set_worklist
-	 and		%g1, 0x3ff, %g1		! 10bit Safari Agent ID
-
-not_cheetah:
-	ldxa	[%g0] ASI_UPA_CONFIG, %g1
-	srlx	%g1, 17, %g1
-	and	%g1, 0x1f, %g1
-
-	/* In theory this is: &(cpu_data[boot_cpu_id].irq_worklists[0]) */
-set_worklist:
-	sethi	%hi(cpu_data), %g5
-	or	%g5, %lo(cpu_data), %g5
-	sllx	%g1, 7, %g1
-	add	%g5, %g1, %g5
-	add	%g5, 64, %g6
-#endif
-
 	/* Kill PROM timer */
-	sethi	%hi(0x80000000), %g1
-	sllx	%g1, 32, %g1
-	wr	%g1, 0, %tick_cmpr
+	sethi	%hi(0x80000000), %o2
+	sllx	%o2, 32, %o2
+	wr	%o2, 0, %tick_cmpr
 
-	BRANCH_IF_ANY_CHEETAH(g1,g5,1f)
+	BRANCH_IF_ANY_CHEETAH(o2,o3,1f)
 
 	ba,pt	%xcc, 2f
 	 nop
 
 	/* Disable STICK_INT interrupts. */
 1:
-	sethi	%hi(0x80000000), %g1
-	sllx	%g1, 32, %g1
-	wr	%g1, %asr25
+	sethi	%hi(0x80000000), %o2
+	sllx	%o2, 32, %o2
+	wr	%o2, %asr25
 
 	/* Ok, we're done setting up all the state our trap mechanims needs,
 	 * now get back into normal globals and let the PROM know what is up.
@@ -700,6 +651,9 @@ set_worklist:
 	wrpr	%g0, %g0, %wstate
 	wrpr	%o1, PSTATE_IE, %pstate
 
+	call	init_irqwork_curcpu
+	 nop
+
 	sethi	%hi(sparc64_ttable_tl0), %g5
 	call	prom_set_trap_table
 	 mov	%g5, %o0

arch/sparc64/kernel/irq.c

@@ -36,6 +36,7 @@
 #include <asm/starfire.h>
 #include <asm/uaccess.h>
 #include <asm/cache.h>
+#include <asm/cpudata.h>
 
 #ifdef CONFIG_SMP
 static void distribute_irqs(void);
@@ -56,12 +57,18 @@ static void distribute_irqs(void);
 
 struct ino_bucket ivector_table[NUM_IVECS] __attribute__ ((aligned (SMP_CACHE_BYTES)));
 
-#ifndef CONFIG_SMP
-unsigned int __up_workvec[16] __attribute__ ((aligned (SMP_CACHE_BYTES)));
-#define irq_work(__cpu, __pil)	&(__up_workvec[(void)(__cpu), (__pil)])
-#else
-#define irq_work(__cpu, __pil)	&(cpu_data[(__cpu)].irq_worklists[(__pil)])
-#endif
+/* This has to be in the main kernel image, it cannot be
+ * turned into per-cpu data.  The reason is that the main
+ * kernel image is locked into the TLB and this structure
+ * is accessed from the vectored interrupt trap handler.  If
+ * access to this structure takes a TLB miss it could cause
+ * the 5-level sparc v9 trap stack to overflow.
+ */
+struct irq_work_struct {
+	unsigned int	irq_worklists[16];
+};
+struct irq_work_struct __irq_work[NR_CPUS];
+#define irq_work(__cpu, __pil)	&(__irq_work[(__cpu)].irq_worklists[(__pil)])
 
 #ifdef CONFIG_PCI
 /* This is a table of physical addresses used to deal with IBF_DMA_SYNC.
@@ -706,7 +713,7 @@ static inline void redirect_intr(int cpu, struct ino_bucket *bp)
 		goto out;
 
 	/* Voo-doo programming. */
-	if (cpu_data[buddy].idle_volume < FORWARD_VOLUME)
+	if (cpu_data(buddy).idle_volume < FORWARD_VOLUME)
 		goto out;
 
 	/* This just so happens to be correct on Cheetah
@@ -1083,15 +1090,30 @@ void enable_prom_timer(void)
 	prom_timers->count0 = 0;
 }
 
+void init_irqwork_curcpu(void)
+{
+	register struct irq_work_struct *workp asm("o2");
+	unsigned long tmp;
+
+	memset(__irq_work + smp_processor_id(), 0, sizeof(*workp));
+
+	/* Set interrupt globals.  */
+	workp = &__irq_work[smp_processor_id()];
+	__asm__ __volatile__(
+	"rdpr	%%pstate, %0\n\t"
+	"wrpr	%0, %1, %%pstate\n\t"
+	"mov	%2, %%g6\n\t"
+	"wrpr	%0, 0x0, %%pstate\n\t"
+	: "=&r" (tmp)
+	: "i" (PSTATE_IG | PSTATE_IE), "r" (workp));
+}
+
 /* Only invoked on boot processor. */
 void __init init_IRQ(void)
 {
 	map_prom_timers();
 	kill_prom_timer();
 	memset(&ivector_table[0], 0, sizeof(ivector_table));
-#ifndef CONFIG_SMP
-	memset(&__up_workvec[0], 0, sizeof(__up_workvec));
-#endif
 
 	/* We need to clear any IRQ's pending in the soft interrupt
 	 * registers, a spurious one could be left around from the

arch/sparc64/kernel/pci_sabre.c

@@ -18,6 +18,7 @@
 #include <asm/iommu.h>
 #include <asm/irq.h>
 #include <asm/smp.h>
+#include <asm/oplib.h>
 
 #include "pci_impl.h"
 #include "iommu_common.h"
@@ -1518,11 +1519,12 @@ void __init sabre_init(int pnode, char *model_name)
 		    !strcmp(compat, "pci108e,a001")) {
 			hummingbird_p = 1;
 		} else {
-			int cpu_node = linux_cpus[0].prom_node;
+			int cpu_node;
 
 			/* Of course, Sun has to encode things a thousand
 			 * different ways, inconsistently.
 			 */
+			cpu_find_by_instance(0, &cpu_node, NULL);
 			if (prom_getproperty(cpu_node, "name",
 					     compat, sizeof(compat)) > 0 &&
 			    !strcmp(compat, "SUNW,UltraSPARC-IIe"))

arch/sparc64/kernel/process.c

@@ -41,6 +41,7 @@
 #include <asm/elf.h>
 #include <asm/fpumacro.h>
 #include <asm/head.h>
+#include <asm/cpudata.h>
 
 /* #define VERBOSE_SHOWREGS */
 
@@ -84,8 +85,8 @@ int cpu_idle(void)
 /*
  * the idle loop on a UltraMultiPenguin...
  */
-#define idle_me_harder()	(cpu_data[smp_processor_id()].idle_volume += 1)
-#define unidle_me()		(cpu_data[smp_processor_id()].idle_volume = 0)
+#define idle_me_harder()	(cpu_data(smp_processor_id()).idle_volume += 1)
+#define unidle_me()		(cpu_data(smp_processor_id()).idle_volume = 0)
 int cpu_idle(void)
 {
 	set_thread_flag(TIF_POLLING_NRFLAG);

arch/sparc64/kernel/setup.c

@@ -586,21 +586,15 @@ asmlinkage int sys_ioperm(unsigned long from, unsigned long num, int on)
 
 /* BUFFER is PAGE_SIZE bytes long. */
 
-extern char *sparc_cpu_type[];
-extern char *sparc_fpu_type[];
+extern char *sparc_cpu_type;
+extern char *sparc_fpu_type;
 
 extern void smp_info(struct seq_file *);
 extern void smp_bogo(struct seq_file *);
 extern void mmu_info(struct seq_file *);
 
-#ifndef CONFIG_SMP
-unsigned long up_clock_tick;
-#endif
-
 static int show_cpuinfo(struct seq_file *m, void *__unused)
 {
-	int cpuid = smp_processor_id();
-
 	seq_printf(m, 
 		   "cpu\t\t: %s\n"
 		   "fpu\t\t: %s\n"
@@ -614,18 +608,18 @@ static int show_cpuinfo(struct seq_file *m, void *__unused)
 		   "Cpu0ClkTck\t: %016lx\n"
 #endif
 		   ,
-		   sparc_cpu_type[cpuid],
-		   sparc_fpu_type[cpuid],
+		   sparc_cpu_type,
+		   sparc_fpu_type,
 		   prom_rev,
 		   prom_prev >> 16,
 		   (prom_prev >> 8) & 0xff,
 		   prom_prev & 0xff,
-		   (long)linux_num_cpus,
+		   (long)num_possible_cpus(),
 		   (long)num_online_cpus()
 #ifndef CONFIG_SMP
-		   , loops_per_jiffy/(500000/HZ),
-		   (loops_per_jiffy/(5000/HZ)) % 100,
-		   up_clock_tick
+		   , cpu_data(0).udelay_val/(500000/HZ),
+		   (cpu_data(0).udelay_val/(5000/HZ)) % 100,
+		   cpu_data(0).clock_tick
 #endif
 		);
 #ifdef CONFIG_SMP
@@ -680,21 +674,24 @@ void sun_do_break(void)
 int serial_console;
 int stop_a_enabled = 1;
 
-static struct cpu *sparc64_cpus;
-
 static int __init topology_init(void)
 {
-	int i;
+	int i, err;
 
-	sparc64_cpus = kmalloc(NR_CPUS * sizeof(struct cpu), GFP_KERNEL);
-	if (!sparc64_cpus)
-		return -ENOMEM;
-	memset(sparc64_cpus, 0, NR_CPUS * sizeof(struct cpu));
+	err = -ENOMEM;
 	for (i = 0; i < NR_CPUS; i++) {
-		if (cpu_possible(i))
-			register_cpu(&sparc64_cpus[i], i, NULL);
+		if (cpu_possible(i)) {
+			struct cpu *p = kmalloc(sizeof(*p), GFP_KERNEL);
+
+			if (p) {
+				memset(p, 0, sizeof(*p));
+				register_cpu(p, i, NULL);
+				err = 0;
 			}
-	return 0;
+		}
+	}
+
+	return err;
 }
 
 subsys_initcall(topology_init);

arch/sparc64/kernel/smp.c

@@ -25,6 +25,7 @@
 #include <asm/atomic.h>
 #include <asm/tlbflush.h>
 #include <asm/mmu_context.h>
+#include <asm/cpudata.h>
 
 #include <asm/irq.h>
 #include <asm/page.h>
@@ -41,8 +42,6 @@
 extern int linux_num_cpus;
 extern void calibrate_delay(void);
 
-cpuinfo_sparc cpu_data[NR_CPUS];
-
 /* Please don't make this stuff initdata!!!  --DaveM */
 static unsigned char boot_cpu_id;
 
@@ -73,35 +72,29 @@ void smp_bogo(struct seq_file *m)
 			seq_printf(m,
 				   "Cpu%dBogo\t: %lu.%02lu\n"
 				   "Cpu%dClkTck\t: %016lx\n",
-				   i, cpu_data[i].udelay_val / (500000/HZ),
-				   (cpu_data[i].udelay_val / (5000/HZ)) % 100,
-				   i, cpu_data[i].clock_tick);
+				   i, cpu_data(i).udelay_val / (500000/HZ),
+				   (cpu_data(i).udelay_val / (5000/HZ)) % 100,
+				   i, cpu_data(i).clock_tick);
 }
 
 void __init smp_store_cpu_info(int id)
 {
-	int i, no;
+	int cpu_node;
 
 	/* multiplier and counter set by
 	   smp_setup_percpu_timer()  */
-	cpu_data[id].udelay_val			= loops_per_jiffy;
-
-	for (no = 0; no < linux_num_cpus; no++)
-		if (linux_cpus[no].mid == id)
-			break;
+	cpu_data(id).udelay_val			= loops_per_jiffy;
 
-	cpu_data[id].clock_tick = prom_getintdefault(linux_cpus[no].prom_node,
+	cpu_find_by_mid(id, &cpu_node);
+	cpu_data(id).clock_tick = prom_getintdefault(cpu_node,
 						     "clock-frequency", 0);
 
-	cpu_data[id].pgcache_size		= 0;
-	cpu_data[id].pte_cache[0]		= NULL;
-	cpu_data[id].pte_cache[1]		= NULL;
-	cpu_data[id].pgdcache_size		= 0;
-	cpu_data[id].pgd_cache			= NULL;
-	cpu_data[id].idle_volume		= 1;
-
-	for (i = 0; i < 16; i++)
-		cpu_data[id].irq_worklists[i] = 0;
+	cpu_data(id).pgcache_size		= 0;
+	cpu_data(id).pte_cache[0]		= NULL;
+	cpu_data(id).pte_cache[1]		= NULL;
+	cpu_data(id).pgdcache_size		= 0;
+	cpu_data(id).pgd_cache			= NULL;
+	cpu_data(id).idle_volume		= 1;
 }
 
 static void smp_setup_percpu_timer(void);
@@ -109,7 +102,6 @@ static void smp_setup_percpu_timer(void);
 static volatile unsigned long callin_flag = 0;
 
 extern void inherit_locked_prom_mappings(int save_p);
-extern void cpu_probe(void);
 
 void __init smp_callin(void)
 {
@@ -129,8 +121,6 @@ void __init smp_callin(void)
 	__flush_cache_all();
 	__flush_tlb_all();
 
-	cpu_probe();
-
 	smp_setup_percpu_timer();
 
 	local_irq_enable();
@@ -300,8 +290,6 @@ static void smp_synchronize_one_tick(int cpu)
 	spin_unlock_irqrestore(&itc_sync_lock, flags);
 }
 
-extern struct prom_cpuinfo linux_cpus[NR_CPUS];
-
 extern unsigned long sparc64_cpu_startup;
 
 /* The OBP cpu startup callback truncates the 3rd arg cookie to
@@ -317,7 +305,7 @@ static int __devinit smp_boot_one_cpu(unsigned int cpu)
 	unsigned long cookie =
 		(unsigned long)(&cpu_new_thread);
 	struct task_struct *p;
-	int timeout, no, ret;
+	int timeout, ret, cpu_node;
 
 	kernel_thread(NULL, NULL, CLONE_IDLETASK);
 
@@ -328,19 +316,18 @@ static int __devinit smp_boot_one_cpu(unsigned int cpu)
 	unhash_process(p);
 
 	callin_flag = 0;
-	for (no = 0; no < linux_num_cpus; no++)
-		if (linux_cpus[no].mid == cpu)
-			break;
 	cpu_new_thread = p->thread_info;
 	cpu_set(cpu, cpu_callout_map);
-	prom_startcpu(linux_cpus[no].prom_node, entry, cookie);
+
+	cpu_find_by_mid(cpu, &cpu_node);
+	prom_startcpu(cpu_node, entry, cookie);
+
 	for (timeout = 0; timeout < 5000000; timeout++) {
 		if (callin_flag)
 			break;
 		udelay(100);
 	}
 	if (callin_flag) {
-		prom_cpu_nodes[cpu] = linux_cpus[no].prom_node;
 		ret = 0;
 	} else {
 		printk("Processor %d is stuck.\n", cpu);
@@ -1048,8 +1035,8 @@ void smp_promstop_others(void)
 
 extern void sparc64_do_profile(struct pt_regs *regs);
 
-#define prof_multiplier(__cpu)		cpu_data[(__cpu)].multiplier
-#define prof_counter(__cpu)		cpu_data[(__cpu)].counter
+#define prof_multiplier(__cpu)		cpu_data(__cpu).multiplier
+#define prof_counter(__cpu)		cpu_data(__cpu).counter
 
 void smp_percpu_timer_interrupt(struct pt_regs *regs)
 {
@@ -1141,7 +1128,6 @@ void __init smp_tick_init(void)
 	}
 
 	cpu_set(boot_cpu_id, cpu_online_map);
-	prom_cpu_nodes[boot_cpu_id] = linux_cpus[0].prom_node;
 	prof_counter(boot_cpu_id) = prof_multiplier(boot_cpu_id) = 1;
 }
 
@@ -1155,6 +1141,7 @@ static void __init smp_tune_scheduling(void)
 	unsigned long orig_flush_base, flush_base, flags, *p;
 	unsigned int ecache_size, order;
 	cycles_t tick1, tick2, raw;
+	int cpu_node;
 
 	/* Approximate heuristic for SMP scheduling.  It is an
 	 * estimation of the time it takes to flush the L2 cache
@@ -1172,7 +1159,8 @@ static void __init smp_tune_scheduling(void)
 		goto report;
 	}
 
-	ecache_size = prom_getintdefault(linux_cpus[0].prom_node,
+	cpu_find_by_instance(0, &cpu_node, NULL);
+	ecache_size = prom_getintdefault(cpu_node,
 					 "ecache-size", (512 * 1024));
 	if (ecache_size > (4 * 1024 * 1024))
 		ecache_size = (4 * 1024 * 1024);
@@ -1254,22 +1242,27 @@ int setup_profiling_timer(unsigned int multiplier)
 
 void __init smp_prepare_cpus(unsigned int max_cpus)
 {
-	int i;
+	int instance, mid;
 
-	for (i = 0; i < linux_num_cpus; i++) {
-		if (linux_cpus[i].mid < max_cpus) {
-			cpu_set(linux_cpus[i].mid, phys_cpu_present_map);
+	instance = 0;
+	while (!cpu_find_by_instance(instance, NULL, &mid)) {
+		if (mid < max_cpus) {
+			cpu_set(mid, phys_cpu_present_map);
 			atomic_inc(&sparc64_num_cpus_possible);
 		}
+		instance++;
 	}
+
 	if (atomic_read(&sparc64_num_cpus_possible) > max_cpus) {
-		for (i = linux_num_cpus - 1; i >= 0; i--) {
-			if (linux_cpus[i].mid != boot_cpu_id) {
-				cpu_clear(linux_cpus[i].mid, phys_cpu_present_map);
+		instance = 0;
+		while (!cpu_find_by_instance(instance, NULL, &mid)) {
+			if (mid != boot_cpu_id) {
+				cpu_clear(mid, phys_cpu_present_map);
 				atomic_dec(&sparc64_num_cpus_possible);
 				if (atomic_read(&sparc64_num_cpus_possible) <= max_cpus)
 					break;
 			}
+			instance++;
 		}
 	}
 
@@ -1306,7 +1299,7 @@ void __init smp_cpus_done(unsigned int max_cpus)
 
 	for (i = 0; i < NR_CPUS; i++) {
 		if (cpu_online(i))
-			bogosum += cpu_data[i].udelay_val;
+			bogosum += cpu_data(i).udelay_val;
 	}
 	printk("Total of %ld processors activated "
 	       "(%lu.%02lu BogoMIPS).\n",
@@ -1319,3 +1312,27 @@ void __init smp_cpus_done(unsigned int max_cpus)
 	 */
 	smp_tune_scheduling();
 }
+
+/* This needn't do anything as we do not sleep the cpu
+ * inside of the idler task, so an interrupt is not needed
+ * to get a clean fast response.
+ *
+ * XXX Reverify this assumption... -DaveM
+ *
+ * Addendum: We do want it to do something for the signal
+ *           delivery case, we detect that by just seeing
+ *           if we are trying to send this to an idler or not.
+ */
+void smp_send_reschedule(int cpu)
+{
+	if (cpu_data(cpu).idle_volume == 0)
+		smp_receive_signal(cpu);
+}
+
+/* This is a nop because we capture all other cpus
+ * anyways when making the PROM active.
+ */
+void smp_send_stop(void)
+{
+}
+

arch/sparc64/kernel/sparc64_ksyms.c

@@ -21,6 +21,7 @@
 #include <linux/fs_struct.h>
 #include <linux/mm.h>
 #include <linux/socket.h>
+#include <linux/percpu.h>
 #include <net/compat.h>
 
 #include <asm/oplib.h>
@@ -55,6 +56,7 @@
 #include <asm/a.out.h>
 #include <asm/ns87303.h>
 #include <asm/timer.h>
+#include <asm/cpudata.h>
 
 struct poll {
 	int fd;
@@ -124,6 +126,9 @@ extern void xor_vis_4(unsigned long, unsigned long *, unsigned long *,
 extern void xor_vis_5(unsigned long, unsigned long *, unsigned long *,
 		      unsigned long *, unsigned long *, unsigned long *);
 
+/* Per-CPU information table */
+EXPORT_PER_CPU_SYMBOL(__cpu_data);
+
 /* used by various drivers */
 #ifdef CONFIG_SMP
 #ifndef CONFIG_DEBUG_SPINLOCK
@@ -142,9 +147,6 @@ extern void mcount(void);
 EXPORT_SYMBOL_NOVERS(mcount);
 #endif
 
-/* Per-CPU information table */
-EXPORT_SYMBOL(cpu_data);
-
 /* CPU online map and active count.  */
 EXPORT_SYMBOL(cpu_online_map);
 EXPORT_SYMBOL(phys_cpu_present_map);
@@ -320,7 +322,6 @@ EXPORT_SYMBOL(sys_getegid);
 EXPORT_SYMBOL(sys_getgid);
 EXPORT_SYMBOL(svr4_getcontext);
 EXPORT_SYMBOL(svr4_setcontext);
-EXPORT_SYMBOL(prom_cpu_nodes);
 EXPORT_SYMBOL(sys_ioctl);
 EXPORT_SYMBOL(compat_sys_ioctl);
 EXPORT_SYMBOL(sparc32_open);

arch/sparc64/kernel/time.c

@@ -42,6 +42,7 @@
 #include <asm/starfire.h>
 #include <asm/smp.h>
 #include <asm/sections.h>
+#include <asm/cpudata.h>
 
 spinlock_t mostek_lock = SPIN_LOCK_UNLOCKED;
 spinlock_t rtc_lock = SPIN_LOCK_UNLOCKED;
@@ -956,7 +957,7 @@ static unsigned long sparc64_init_timers(irqreturn_t (*cfunc)(int, void *, struc
 			clock = prom_getint(node, "stick-frequency");
 		} else {
 			tick_ops = &tick_operations;
-			node = linux_cpus[0].prom_node;
+			cpu_find_by_instance(0, &node, NULL);
 			clock = prom_getint(node, "clock-frequency");
 		}
 	} else {
@@ -1012,11 +1013,7 @@ unsigned long sparc64_get_clock_tick(unsigned int cpu)
 
 	if (ft->clock_tick_ref)
 		return ft->clock_tick_ref;
-#ifdef CONFIG_SMP
-	return cpu_data[cpu].clock_tick;
-#else
-	return up_clock_tick;
-#endif
+	return cpu_data(cpu).clock_tick;
 }
 
 #ifdef CONFIG_CPU_FREQ
@@ -1028,35 +1025,22 @@ static int sparc64_cpufreq_notifier(struct notifier_block *nb, unsigned long val
 	unsigned int cpu = freq->cpu;
 	struct freq_table *ft = &per_cpu(sparc64_freq_table, cpu);
 
-#ifdef CONFIG_SMP
 	if (!ft->ref_freq) {
 		ft->ref_freq = freq->old;
-		ft->udelay_val_ref = cpu_data[cpu].udelay_val;
-		ft->clock_tick_ref = cpu_data[cpu].clock_tick;
+		ft->udelay_val_ref = cpu_data(cpu).udelay_val;
+		ft->clock_tick_ref = cpu_data(cpu).clock_tick;
 	}
 	if ((val == CPUFREQ_PRECHANGE  && freq->old < freq->new) ||
 	    (val == CPUFREQ_POSTCHANGE && freq->old > freq->new)) {
-		cpu_data[cpu].udelay_val =
+		cpu_data(cpu).udelay_val =
 			cpufreq_scale(ft->udelay_val_ref,
 				      ft->ref_freq,
 				      freq->new);
-		cpu_data[cpu].clock_tick =
+		cpu_data(cpu).clock_tick =
 			cpufreq_scale(ft->clock_tick_ref,
 				      ft->ref_freq,
 				      freq->new);
 	}
-#else
-	/* In the non-SMP case, kernel/cpufreq.c takes care of adjusting
-	 * loops_per_jiffy.
-	 */
-	if (!ft->ref_freq) {
-		ft->ref_freq = freq->old;
-		ft->clock_tick_ref = up_clock_tick;
-	}
-	if ((val == CPUFREQ_PRECHANGE  && freq->old < freq->new) ||
-	    (val == CPUFREQ_POSTCHANGE && freq->old > freq->new))
-		up_clock_tick = cpufreq_scale(ft->clock_tick_ref, ft->ref_freq, freq->new);
-#endif
 
 	return 0;
 }

arch/sparc64/kernel/trampoline.S

@@ -246,20 +246,12 @@ startup_continue:
 	wrpr		%o1, 0x0, %pstate
 	ldx		[%g6 + TI_TASK], %g4
 
-	/* Setup interrupt globals, we are always SMP. */
-	wrpr		%o1, PSTATE_IG, %pstate
-
-	/* Get our UPA MID. */
-	ldub		[%o2 + TI_CPU], %g1
-	sethi		%hi(cpu_data), %g5
-	or		%g5, %lo(cpu_data), %g5
+	wrpr		%g0, 0, %wstate
 
-	/* In theory this is: &(cpu_data[this_upamid].irq_worklists[0]) */
-	sllx		%g1, 7, %g1
-	add		%g5, %g1, %g1
-	add		%g1, 64, %g6
+	call		init_irqwork_curcpu
+	 nop
 
-	wrpr		%g0, 0, %wstate
+	rdpr		%pstate, %o1
 	or		%o1, PSTATE_IE, %o1
 	wrpr		%o1, 0, %pstate
 

arch/sparc64/kernel/traps.c

@@ -613,8 +613,7 @@ extern unsigned int cheetah_deferred_trap_vector[], cheetah_deferred_trap_vector
 void __init cheetah_ecache_flush_init(void)
 {
 	unsigned long largest_size, smallest_linesize, order, ver;
-	char type[16];
-	int node, i;
+	int node, i, instance;
 
 	/* Scan all cpu device tree nodes, note two values:
 	 * 1) largest E-cache size
@@ -622,10 +621,9 @@ void __init cheetah_ecache_flush_init(void)
 	 */
 	largest_size = 0UL;
 	smallest_linesize = ~0UL;
-	node = prom_getchild(prom_root_node);
-	while ((node = prom_getsibling(node)) != 0) {
-		prom_getstring(node, "device_type", type, sizeof(type));
-		if (!strcmp(type, "cpu")) {
+
+	instance = 0;
+	while (!cpu_find_by_instance(instance, &node, NULL)) {
 		unsigned long val;
 
 		val = prom_getintdefault(node, "ecache-size",
@@ -635,8 +633,9 @@ void __init cheetah_ecache_flush_init(void)
 		val = prom_getintdefault(node, "ecache-line-size", 64);
 		if (val < smallest_linesize)
 			smallest_linesize = val;
+		instance++;
 	}
-	}
+
 	if (largest_size == 0UL || smallest_linesize == ~0UL) {
 		prom_printf("cheetah_ecache_flush_init: Cannot probe cpu E-cache "
 			    "parameters.\n");

arch/sparc64/mm/init.c

@@ -1706,6 +1706,10 @@ void __init mem_init(void)
 	unsigned long addr, last;
 	int i;
 
+#ifndef CONFIG_SMP
+	cpu_data(0).udelay_val = loops_per_jiffy;
+#endif
+
 	i = last_valid_pfn >> ((22 - PAGE_SHIFT) + 6);
 	i += 1;
 	sparc64_valid_addr_bitmap = (unsigned long *)

arch/sparc64/solaris/misc.c

@@ -349,8 +349,7 @@ asmlinkage int solaris_sysconf(int id)
 	case SOLARIS_CONFIG_XOPEN_VER:	return 3;
 	case SOLARIS_CONFIG_CLK_TCK:
 	case SOLARIS_CONFIG_PROF_TCK:
-		return prom_getintdefault(prom_cpu_nodes[smp_processor_id()],
-					  "clock-frequency", 167000000);
+		return sparc64_get_clock_tick(smp_processor_id());
 #ifdef CONFIG_SMP	
 	case SOLARIS_CONFIG_NPROC_CONF:	return NR_CPUS;
 	case SOLARIS_CONFIG_NPROC_ONLN:	return num_online_cpus();

drivers/serial/sunsab.c

@@ -32,6 +32,7 @@
 #include <linux/console.h>
 #include <linux/spinlock.h>
 #include <linux/slab.h>
+#include <linux/delay.h>
 #include <linux/init.h>
 
 #include <asm/io.h>

include/asm-generic/cpumask_up.h

@@ -3,10 +3,10 @@
 
 #define cpus_coerce(map)	(map)
 
-#define cpu_set(cpu, map)		do { cpus_coerce(map) = 1UL; } while (0)
-#define cpu_clear(cpu, map)		do { cpus_coerce(map) = 0UL; } while (0)
-#define cpu_isset(cpu, map)		(cpus_coerce(map) != 0UL)
-#define cpu_test_and_set(cpu, map)	test_and_set_bit(0, (map).mask)
+#define cpu_set(cpu, map)		do { (void)(cpu); cpus_coerce(map) = 1UL; } while (0)
+#define cpu_clear(cpu, map)		do { (void)(cpu); cpus_coerce(map) = 0UL; } while (0)
+#define cpu_isset(cpu, map)		((void)(cpu), cpus_coerce(map) != 0UL)
+#define cpu_test_and_set(cpu, map)	((void)(cpu), test_and_set_bit(0, (map).mask))
 
 #define cpus_and(dst, src1, src2)					\
 	do {								\
@@ -46,7 +46,7 @@
 		cpus_coerce(__tmp__) = map;				\
 		__tmp__;						\
 	})
-#define cpumask_of_cpu(cpu)		cpus_promote(1)
+#define cpumask_of_cpu(cpu)		((void)(cpu), cpus_promote(1))
 #define any_online_cpu(map)		(cpus_coerce(map) ? 0 : 1)
 
 /*

include/asm-sparc64/cpudata.h

+/* cpudata.h: Per-cpu parameters.
+ *
+ * Copyright (C) 2003 David S. Miller (davem@redhat.com)
+ */
+
+#ifndef _SPARC64_CPUDATA_H
+#define _SPARC64_CPUDATA_H
+
+#include <linux/percpu.h>
+
+typedef struct {
+	/* Dcache line 1 */
+	unsigned int	__pad0;		/* bh_count moved to irq_stat for consistency. KAO */
+	unsigned int	multiplier;
+	unsigned int	counter;
+	unsigned int	idle_volume;
+	unsigned long	clock_tick;	/* %tick's per second */
+	unsigned long	udelay_val;
+
+	/* Dcache line 2 */
+	unsigned int	pgcache_size;
+	unsigned int	pgdcache_size;
+	unsigned long	*pte_cache[2];
+	unsigned long	*pgd_cache;
+} cpuinfo_sparc;
+
+DECLARE_PER_CPU(cpuinfo_sparc, __cpu_data);
+#define cpu_data(__cpu)	per_cpu(__cpu_data, (__cpu))
+
+#endif /* _SPARC64_CPUDATA_H */

include/asm-sparc64/delay.h

@@ -9,6 +9,7 @@
 
 #include <linux/config.h>
 #include <linux/param.h>
+#include <asm/cpudata.h>
 
 #ifndef __ASSEMBLY__
 
@@ -58,11 +59,7 @@ extern __inline__ void __ndelay(unsigned long usecs, unsigned long lps)
 	__delay(usecs * HZ);
 }
 
-#ifdef CONFIG_SMP
-#define __udelay_val cpu_data[smp_processor_id()].udelay_val
-#else
-#define __udelay_val loops_per_jiffy
-#endif
+#define __udelay_val cpu_data(smp_processor_id()).udelay_val
 
 #define udelay(usecs) __udelay((usecs),__udelay_val)
 #define ndelay(usecs) __ndelay((usecs),__udelay_val)

include/asm-sparc64/irq.h

@@ -121,13 +121,6 @@ extern void disable_irq(unsigned int);
 extern void enable_irq(unsigned int);
 extern unsigned int build_irq(int pil, int inofixup, unsigned long iclr, unsigned long imap);
 extern unsigned int sbus_build_irq(void *sbus, unsigned int ino);
-extern unsigned int psycho_build_irq(void *psycho, int imap_off, int ino, int need_dma_sync);
-
-#ifdef CONFIG_SMP
-extern void set_cpu_int(int, int);
-extern void clear_cpu_int(int, int);
-extern void set_irq_udt(int);
-#endif
 
 extern int request_fast_irq(unsigned int irq,
 			    irqreturn_t (*handler)(int, void *, struct pt_regs *),

include/asm-sparc64/oplib.h

@@ -327,6 +327,10 @@ extern int prom_setprop(int node, char *prop_name, char *prop_value,
 extern int prom_pathtoinode(char *path);
 extern int prom_inst2pkg(int);
 
+/* CPU probing helpers.  */
+int cpu_find_by_instance(int instance, int *prom_node, int *mid);
+int cpu_find_by_mid(int mid, int *prom_node);
+
 /* Client interface level routines. */
 extern void prom_set_trap_table(unsigned long tba);
 

include/asm-sparc64/pgalloc.h

@@ -10,11 +10,12 @@
 #include <asm/page.h>
 #include <asm/spitfire.h>
 #include <asm/pgtable.h>
+#include <asm/cpudata.h>
 
 /* Page table allocation/freeing. */
 #ifdef CONFIG_SMP
 /* Sliiiicck */
-#define pgt_quicklists	cpu_data[smp_processor_id()]
+#define pgt_quicklists	cpu_data(smp_processor_id())
 #else
 extern struct pgtable_cache_struct {
 	unsigned long *pgd_cache;

include/asm-sparc64/processor.h

@@ -20,7 +20,6 @@
 #include <asm/ptrace.h>
 #include <asm/segment.h>
 #include <asm/page.h>
-#include <asm/delay.h>
 
 /* Bus types */
 #define EISA_bus 0

include/asm-sparc64/smp.h

@@ -17,49 +17,12 @@
 #include <linux/cpumask.h>
 #include <linux/cache.h>
 
-/* PROM provided per-processor information we need
- * to start them all up.
- */
-
-struct prom_cpuinfo {
-	int prom_node;
-	int mid;
-};
-
-extern int linux_num_cpus;	/* number of CPUs probed  */
-extern struct prom_cpuinfo linux_cpus[NR_CPUS];
-extern unsigned int prom_cpu_nodes[NR_CPUS];
-
 #endif /* !(__ASSEMBLY__) */
 
 #ifdef CONFIG_SMP
 
 #ifndef __ASSEMBLY__
 
-/* Per processor Sparc parameters we need. */
-
-/* Keep this a multiple of 64-bytes for cache reasons. */
-typedef struct {
-	/* Dcache line 1 */
-	unsigned int	__pad0;		/* bh_count moved to irq_stat for consistency. KAO */
-	unsigned int	multiplier;
-	unsigned int	counter;
-	unsigned int	idle_volume;
-	unsigned long	clock_tick;	/* %tick's per second */
-	unsigned long	udelay_val;
-
-	/* Dcache line 2 */
-	unsigned int	pgcache_size;
-	unsigned int	pgdcache_size;
-	unsigned long	*pte_cache[2];
-	unsigned long	*pgd_cache;
-
-	/* Dcache lines 3 and 4 */
-	unsigned int	irq_worklists[16];
-} ____cacheline_aligned cpuinfo_sparc;
-
-extern cpuinfo_sparc cpu_data[NR_CPUS];
-
 /*
  *	Private routines/data
  */
@@ -110,29 +73,11 @@ static __inline__ int hard_smp_processor_id(void)
 
 #define smp_processor_id() (current_thread_info()->cpu)
 
-/* This needn't do anything as we do not sleep the cpu
- * inside of the idler task, so an interrupt is not needed
- * to get a clean fast response.
- *
- * XXX Reverify this assumption... -DaveM
- *
- * Addendum: We do want it to do something for the signal
- *           delivery case, we detect that by just seeing
- *           if we are trying to send this to an idler or not.
- */
-static __inline__ void smp_send_reschedule(int cpu)
-{
-	extern void smp_receive_signal(int);
-	if (cpu_data[cpu].idle_volume == 0)
-		smp_receive_signal(cpu);
-}
+#endif /* !(__ASSEMBLY__) */
 
-/* This is a nop as well because we capture all other cpus
- * anyways when making the PROM active.
- */
-static __inline__ void smp_send_stop(void) { }
+#else
 
-#endif /* !(__ASSEMBLY__) */
+#define num_possible_cpus()	(1)
 
 #endif /* !(CONFIG_SMP) */
 

include/asm-sparc64/timer.h

@@ -69,10 +69,6 @@ struct pt_regs;
 extern void timer_tick_interrupt(struct pt_regs *);
 #endif
 
-#ifndef CONFIG_SMP
-extern unsigned long up_clock_tick;
-#endif
-
 extern unsigned long sparc64_get_clock_tick(unsigned int cpu);
 
 #endif /* _SPARC64_TIMER_H */