ppc64: non linear cpu support

arch/ppc64/kernel/irq.c

@@ -346,8 +346,10 @@ int show_interrupts(struct seq_file *p, void *v)
 	struct irqaction * action;
 
 	seq_printf(p, "           ");
-	for (j=0; j<smp_num_cpus; j++)
+	for (j=0; j<NR_CPUS; j++) {
+		if (cpu_online(j))
 			seq_printf(p, "CPU%d       ",j);
+	}
 	seq_putc(p, '\n');
 
 	for (i = 0 ; i < NR_IRQS ; i++) {
@@ -356,9 +358,10 @@ int show_interrupts(struct seq_file *p, void *v)
 			continue;
 		seq_printf(p, "%3d: ", i);		
 #ifdef CONFIG_SMP
-		for (j = 0; j < smp_num_cpus; j++)
-			seq_printf(p, "%10u ",
-				kstat.irqs[cpu_logical_map(j)][i]);
+		for (j = 0; j < NR_CPUS; j++) {
+			if (cpu_online(j))
+				seq_printf(p, "%10u ", kstat.irqs[j][i]);
+		}
 #else		
 		seq_printf(p, "%10u ", kstat_irqs(i));
 #endif /* CONFIG_SMP */
@@ -425,14 +428,14 @@ static unsigned long move(unsigned long curr_cpu, unsigned long allowed_mask,
 inside:
 		if (direction == 1) {
 			cpu++;
-			if (cpu >= smp_num_cpus)
+			if (cpu >= NR_CPUS)
 				cpu = 0;
 		} else {
 			cpu--;
 			if (cpu == -1)
-				cpu = smp_num_cpus-1;
+				cpu = NR_CPUS-1;
 		}
-	} while (!IRQ_ALLOWED(cpu,allowed_mask) ||
+	} while (!cpu_online(cpu) || !IRQ_ALLOWED(cpu,allowed_mask) ||
 			(search_idle && !IDLE_ENOUGH(cpu,now)));
 
 	return cpu;
@@ -653,12 +656,16 @@ static void show(char * str)
 
 	printk("\n%s, CPU %d:\n", str, cpu);
 	printk("irq:  %d [ ", irqs_running());
-	for (i = 0; i < smp_num_cpus; i++)
+	for (i = 0; i < NR_CPUS; i++) {
+		if (cpu_online(i))
 			printk("%u ", __brlock_array[i][BR_GLOBALIRQ_LOCK]);
+	}
 	printk("]\nbh:   %d [ ",
 		(spin_is_locked(&global_bh_lock) ? 1 : 0));
-	for (i = 0; i < smp_num_cpus; i++)
+	for (i = 0; i < NR_CPUS; i++) {
+		if (cpu_online(i))
 			printk("%u ", local_bh_count(i));
+	}
 	printk("]\n");
 }
 

arch/ppc64/kernel/open_pic.c

@@ -85,10 +85,10 @@ unsigned int openpic_vec_spurious;
  */
 #ifdef CONFIG_SMP
 #define THIS_CPU		Processor[cpu]
-#define DECL_THIS_CPU		int cpu = hard_smp_processor_id()
+#define DECL_THIS_CPU		int cpu = smp_processor_id()
 #define CHECK_THIS_CPU		check_arg_cpu(cpu)
 #else
-#define THIS_CPU		Processor[hard_smp_processor_id()]
+#define THIS_CPU		Processor[smp_processor_id()]
 #define DECL_THIS_CPU
 #define CHECK_THIS_CPU
 #endif /* CONFIG_SMP */
@@ -356,7 +356,7 @@ void __init openpic_init(int main_pic, int offset, unsigned char* chrp_ack,
 	/* SIOint (8259 cascade) is special */
 	if (offset) {
 		openpic_initirq(0, 8, offset, 1, 1);
-		openpic_mapirq(0, 1<<get_hard_smp_processor_id(0));
+		openpic_mapirq(0, 1 << boot_cpuid);
 	}
 
 	/* Init all external sources */
@@ -374,7 +374,7 @@ void __init openpic_init(int main_pic, int offset, unsigned char* chrp_ack,
 		/* Enabled, Priority 8 or 9 */
 		openpic_initirq(i, pri, i+offset, !sense, sense);
 		/* Processor 0 */
-		openpic_mapirq(i, 1<<get_hard_smp_processor_id(0));
+		openpic_mapirq(i, 1 << boot_cpuid);
 	}
 
 	/* Init descriptors */
@@ -503,23 +503,10 @@ static void openpic_set_spurious(u_int vec)
 			   vec);
 }
 
-/*
- * Convert a cpu mask from logical to physical cpu numbers.
- */
-static inline u32 physmask(u32 cpumask)
-{
-	int i;
-	u32 mask = 0;
-
-	for (i = 0; i < smp_num_cpus; ++i, cpumask >>= 1)
-		mask |= (cpumask & 1) << get_hard_smp_processor_id(i);
-	return mask;
-}
-
 void openpic_init_processor(u_int cpumask)
 {
 	openpic_write(&OpenPIC->Global.Processor_Initialization,
-		      physmask(cpumask));
+		      cpumask & cpu_online_map);
 }
 
 #ifdef CONFIG_SMP
@@ -553,7 +540,7 @@ void openpic_cause_IPI(u_int ipi, u_int cpumask)
 	CHECK_THIS_CPU;
 	check_arg_ipi(ipi);
 	openpic_write(&OpenPIC->THIS_CPU.IPI_Dispatch(ipi),
-		      physmask(cpumask));
+		      cpumask & cpu_online_map);
 }
 
 void openpic_request_IPIs(void)
@@ -593,7 +580,7 @@ void __init do_openpic_setup_cpu(void)
 {
 #ifdef CONFIG_IRQ_ALL_CPUS
  	int i;
-	u32 msk = 1 << hard_smp_processor_id();
+	u32 msk = 1 << smp_processor_id();
 #endif
 
 	spin_lock(&openpic_setup_lock);
@@ -638,7 +625,7 @@ static void __init openpic_maptimer(u_int timer, u_int cpumask)
 {
 	check_arg_timer(timer);
 	openpic_write(&OpenPIC->Global.Timer[timer].Destination,
-		      physmask(cpumask));
+		      cpumask & cpu_online_map);
 }
 
 
@@ -761,7 +748,7 @@ static void openpic_end_irq(unsigned int irq_nr)
 
 static void openpic_set_affinity(unsigned int irq_nr, unsigned long cpumask)
 {
-	openpic_mapirq(irq_nr - open_pic_irq_offset, physmask(cpumask));
+	openpic_mapirq(irq_nr - open_pic_irq_offset, cpumask & cpu_online_map);
 }
 
 #ifdef CONFIG_SMP

arch/ppc64/kernel/pSeries_lpar.c

@@ -309,7 +309,7 @@ static void pSeriesLP_qirr_info(int n_cpu , u8 value)
 {
 	unsigned long lpar_rc;
 
-	lpar_rc = plpar_ipi(get_hard_smp_processor_id(n_cpu),value);
+	lpar_rc = plpar_ipi(n_cpu, value);
 	if (lpar_rc != H_Success) {
 		panic(" bad return code qirr -ipi  - rc = %lx \n", lpar_rc); 
 	}

arch/ppc64/kernel/pmc.c

@@ -59,7 +59,9 @@ ppc64_pmc_stab(int file)
 	stab_faults = stab_capacity_castouts = stab_invalidations = n = 0;
 
 	if (file == -1) {
-		for (i = 0;  i < smp_num_cpus; i++) {
+		for (i = 0;  i < NR_CPUS; i++) {
+			if (!cpu_online(i))
+				continue;
 			stab_faults += pmc_sw_cpu[i].stab_faults;
 			stab_capacity_castouts += pmc_sw_cpu[i].stab_capacity_castouts;
 			stab_invalidations += pmc_sw_cpu[i].stab_invalidations;

arch/ppc64/kernel/ppc_ksyms.c

@@ -87,7 +87,6 @@ EXPORT_SYMBOL(disable_irq_nosync);
 #ifdef CONFIG_SMP
 EXPORT_SYMBOL(kernel_flag);
 EXPORT_SYMBOL(synchronize_irq);
-EXPORT_SYMBOL(smp_num_cpus);
 #endif /* CONFIG_SMP */
 
 EXPORT_SYMBOL(register_ioctl32_conversion);

arch/ppc64/kernel/proc_pmc.c

@@ -109,14 +109,19 @@ void proc_ppc64_init(void)
 	proc_ppc64_pmc_root = proc_mkdir("pmc", proc_ppc64_root);
 
 	proc_ppc64_pmc_system_root = proc_mkdir("system", proc_ppc64_pmc_root);
-	for (i = 0; i < naca->processorCount; i++) {
+	for (i = 0; i < NR_CPUS; i++) {
+		if (cpu_online(i)) {
 			sprintf(buf, "cpu%ld", i); 
-		proc_ppc64_pmc_cpu_root[i] = proc_mkdir(buf, proc_ppc64_pmc_root);
+			proc_ppc64_pmc_cpu_root[i] =
+				proc_mkdir(buf, proc_ppc64_pmc_root);
+		}
 	}
 
 
 	/* Create directories for the software counters. */
-	for (i = 0; i < naca->processorCount; i++) {
+	for (i = 0; i < NR_CPUS; i++) {
+		if (!cpu_online(i))
+			continue;
 		ent = create_proc_entry("stab", S_IRUGO | S_IWUSR, 
 					proc_ppc64_pmc_cpu_root[i]);
 		if (ent) {
@@ -155,7 +160,9 @@ void proc_ppc64_init(void)
 	}
 
 	/* Create directories for the hardware counters. */
-	for (i = 0; i < naca->processorCount; i++) {
+	for (i = 0; i < NR_CPUS; i++) {
+		if (!cpu_online(i))
+			continue;
 		ent = create_proc_entry("hardware", S_IRUGO | S_IWUSR, 
 					proc_ppc64_pmc_cpu_root[i]);
 		if (ent) {
@@ -191,7 +198,9 @@ int proc_ppc64_pmc_find_file(void *data)
 	   (unsigned long) proc_ppc64_pmc_system_root) {
 		return(-1); 
 	} else {
-		for (i = 0; i < naca->processorCount; i++) {
+		for (i = 0; i < NR_CPUS; i++) {
+			if (!cpu_online(i))
+				continue;
 			if ((unsigned long)data ==
 			   (unsigned long)proc_ppc64_pmc_cpu_root[i]) {
 				return(i); 
@@ -383,7 +392,8 @@ int proc_get_lpevents
 			(unsigned long)xItLpQueue.xLpIntCountByType[i] );
 	}
 	len += sprintf( page+len, "\n  events processed by processor:\n" );
-	for (i=0; i<naca->processorCount; ++i) {
+	for (i = 0; i < NR_CPUS; ++i) {
+		if (cpu_online(i))
 			len += sprintf( page+len, "    CPU%02d  %10u\n",
 				i, paca[i].lpEvent_count );
 	}

arch/ppc64/kernel/process.c

@@ -304,7 +304,9 @@ void initialize_paca_hardware_interrupt_stack(void)
 	unsigned long stack;
 	unsigned long end_of_stack =0;
 
-	for (i=1; i < naca->processorCount; i++) {
+	for (i=1; i < NR_CPUS; i++) {
+		if (!cpu_possible(i))
+			continue;
 		/* Carve out storage for the hardware interrupt stack */
 		stack = __get_free_pages(GFP_KERNEL, get_order(8*PAGE_SIZE));
 
@@ -327,7 +329,9 @@ void initialize_paca_hardware_interrupt_stack(void)
 	if (__is_processor(PV_POWER4))
 		return;
 
-	for (i=0; i < naca->processorCount; i++) {
+	for (i=0; i < NR_CPUS; i++) {
+		if (!cpu_possible(i))
+			continue;
 		/* set page at the top of stack to be protected - prevent overflow */
 		end_of_stack = paca[i].xHrdIntStack - (8*PAGE_SIZE - STACK_FRAME_OVERHEAD);
 		ppc_md.hpte_updateboltedpp(PP_RXRX,end_of_stack);

arch/ppc64/kernel/prom.c

@@ -445,9 +445,12 @@ prom_initialize_naca(unsigned long mem)
 	}
 
 	/* We gotta have at least 1 cpu... */
-        if ( (_naca->processorCount = num_cpus) < 1 )
+        if (num_cpus < 1)
                 PROM_BUG();
 
+	if (num_cpus > 1)
+		RELOC(ppc64_is_smp) = 1;
+
 	_naca->physicalMemorySize = lmb_phys_mem_size();
 
 	if (_naca->platform == PLATFORM_PSERIES) {
@@ -477,10 +480,6 @@ prom_initialize_naca(unsigned long mem)
 	_naca->slb_size = 64;
 
 #ifdef DEBUG_PROM
-        prom_print(RELOC("naca->processorCount       = 0x"));
-        prom_print_hex(_naca->processorCount);
-        prom_print_nl();
-
         prom_print(RELOC("naca->physicalMemorySize   = 0x"));
         prom_print_hex(_naca->physicalMemorySize);
         prom_print_nl();
@@ -1044,20 +1043,15 @@ prom_hold_cpus(unsigned long mem)
 	phandle node;
 	unsigned long offset = reloc_offset();
 	char type[64], *path;
-	int cpuid = 0;
 	extern void __secondary_hold(void);
         extern unsigned long __secondary_hold_spinloop;
         extern unsigned long __secondary_hold_acknowledge;
         unsigned long *spinloop     = __v2a(&__secondary_hold_spinloop);
         unsigned long *acknowledge  = __v2a(&__secondary_hold_acknowledge);
         unsigned long secondary_hold = (unsigned long)__v2a(*PTRRELOC((unsigned long *)__secondary_hold));
-        struct naca_struct *_naca = RELOC(naca);
 	struct paca_struct *_xPaca = PTRRELOC(&paca[0]);
 	struct prom_t *_prom = PTRRELOC(&prom);
 
-	/* Initially, we must have one active CPU. */
-	_naca->processorCount = 1;
-
 #ifdef DEBUG_PROM
 	prom_print(RELOC("prom_hold_cpus: start...\n"));
 	prom_print(RELOC("    1) spinloop       = 0x"));
@@ -1117,19 +1111,12 @@ prom_hold_cpus(unsigned long mem)
 				     node, path, 255) < 0)
 			continue;
 
-		cpuid++;
-
 #ifdef DEBUG_PROM
-		prom_print_nl();
-		prom_print(RELOC("cpuid        = 0x"));
-		prom_print_hex(cpuid);
 		prom_print_nl();
 		prom_print(RELOC("cpu hw idx   = 0x"));
 		prom_print_hex(reg);
 		prom_print_nl();
 #endif
-		_xPaca[cpuid].xHwProcNum = reg;
-
 		prom_print(RELOC("starting cpu "));
 		prom_print(path);
 
@@ -1155,11 +1142,9 @@ prom_hold_cpus(unsigned long mem)
 		prom_print(RELOC("    3) secondary_hold = 0x"));
 		prom_print_hex(secondary_hold);
 		prom_print_nl();
-		prom_print(RELOC("    3) cpuid = 0x"));
-		prom_print_hex(cpuid);
 		prom_print_nl();
 #endif
-		call_prom(RELOC("start-cpu"), 3, 0, node, secondary_hold, cpuid);
+		call_prom(RELOC("start-cpu"), 3, 0, node, secondary_hold, reg);
 		prom_print(RELOC("..."));
 		for ( i = 0 ; (i < 100000000) && 
 			      (*acknowledge == ((unsigned long)-1)); i++ ) ;
@@ -1171,10 +1156,10 @@ prom_hold_cpus(unsigned long mem)
 			prom_print_nl();
 		}
 #endif
-		if (*acknowledge == cpuid) {
+		if (*acknowledge == reg) {
 			prom_print(RELOC("ok\n"));
 			/* Set the number of active processors. */
-			_naca->processorCount++;
+			_xPaca[reg].active = 1;
 		} else {
 			prom_print(RELOC("failed: "));
 			prom_print_hex(*acknowledge);
@@ -1188,10 +1173,11 @@ prom_hold_cpus(unsigned long mem)
 	    __is_processor(PV_SSTAR)) {
 		prom_print(RELOC("    starting secondary threads\n"));
 
-		for (i=0; i < _naca->processorCount ;i++) {
-			unsigned long threadid = _naca->processorCount*2-1-i;
+		for (i = 0; i < NR_CPUS; i += 2) {
+			if (!_xPaca[i].active)
+				continue;
 
-			if (i == 0) {
+			if (i == boot_cpuid) {
 				unsigned long pir = _get_PIR();
 				if (__is_processor(PV_PULSAR)) {
 					RELOC(hmt_thread_data)[i].pir = 
@@ -1201,21 +1187,9 @@ prom_hold_cpus(unsigned long mem)
 						pir & 0x3ff;
 				}
 			}
-			
-			RELOC(hmt_thread_data)[i].threadid = threadid;
-#ifdef DEBUG_PROM
-			prom_print(RELOC("        cpuid 0x"));
-			prom_print_hex(i);
-			prom_print(RELOC(" maps to threadid 0x"));
-			prom_print_hex(threadid);
-			prom_print_nl();
-			prom_print(RELOC(" pir 0x"));
-			prom_print_hex(RELOC(hmt_thread_data)[i].pir);
-			prom_print_nl();
-#endif
-			_xPaca[threadid].xHwProcNum = _xPaca[i].xHwProcNum+1;
+			_xPaca[i+1].active = 1;
+			RELOC(hmt_thread_data)[i].threadid = i+1;
 		}
-		_naca->processorCount *= 2;
 	} else {
 		prom_print(RELOC("Processor is not HMT capable\n"));
 	}
@@ -1372,7 +1346,9 @@ prom_init(unsigned long r3, unsigned long r4, unsigned long pp,
 		cpu_pkg, RELOC("reg"),
 		&getprop_rval, sizeof(getprop_rval));
 	_prom->cpu = (int)(unsigned long)getprop_rval;
-	_xPaca[0].xHwProcNum = _prom->cpu;
+	_xPaca[_prom->cpu].active = 1;
+	RELOC(cpu_online_map) = 1 << _prom->cpu;
+	RELOC(boot_cpuid) = _prom->cpu;
 
 #ifdef DEBUG_PROM
   	prom_print(RELOC("Booting CPU hw index = 0x"));
@@ -1408,7 +1384,7 @@ prom_init(unsigned long r3, unsigned long r4, unsigned long pp,
          * following, regardless of whether we have an SMP
          * kernel or not.
          */
-        if ( _naca->processorCount > 1 )
+        if (RELOC(ppc64_is_smp))
 	        prom_hold_cpus(mem);
 
 	mem = check_display(mem);

arch/ppc64/kernel/rtasd.c

@@ -220,10 +220,15 @@ static int rtasd(void *unused)
 	current->nice = sys_sched_get_priority_max(SCHED_FIFO) + 1;
 #endif
 
-	cpu = 0;
-	set_cpus_allowed(current, 1UL << cpu_logical_map(cpu));
+repeat:
+	for (cpu = 0; cpu < NR_CPUS; cpu++) {
+		if (!cpu_online(cpu))
+			continue;
+
+		DEBUG("scheduling on %d\n", cpu);
+		set_cpus_allowed(current, 1UL << cpu);
+		DEBUG("watchdog scheduled on cpu %d\n", smp_processor_id());
 
-	while(1) {
 		do {
 			memset(logdata, 0, rtas_error_log_max);
 			error = rtas_call(event_scan, 4, 1, NULL,
@@ -239,31 +244,23 @@ static int rtasd(void *unused)
 
 		} while(error == 0);
 
-		DEBUG("watchdog scheduled on cpu %d\n", smp_processor_id());
-
-		cpu++;
-		if (cpu >= smp_num_cpus) {
+		/* Check all cpus for pending events before sleeping*/
+		if (!first_pass) {
+			set_current_state(TASK_INTERRUPTIBLE);
+			schedule_timeout((HZ*60/rtas_event_scan_rate) / 2);
+		}
+	}
 
 	if (first_pass && surveillance_requested) {
 		DEBUG("enabling surveillance\n");
 		if (enable_surveillance())
 			goto error_vfree;
 		DEBUG("surveillance enabled\n");
-			}
-
+	} else {
 		first_pass = 0;
-			cpu = 0;
 	}
 
-		set_cpus_allowed(current, 1UL << cpu_logical_map(cpu));
-
-
-		/* Check all cpus for pending events before sleeping*/
-		if (!first_pass) {
-			set_current_state(TASK_INTERRUPTIBLE);
-			schedule_timeout((HZ*60/rtas_event_scan_rate) / 2);
-		}
-	}
+	goto repeat;
 
 error_vfree:
 	vfree(rtas_log_buf);

arch/ppc64/kernel/setup.c

@@ -168,10 +168,6 @@ void setup_system(unsigned long r3, unsigned long r4, unsigned long r5,
 	udbg_puthex((unsigned long)naca);
 	udbg_putc('\n');
 
-	udbg_puts("naca->processorCount       = 0x");
-	udbg_puthex(naca->processorCount);
-	udbg_putc('\n');
-
 	udbg_puts("naca->physicalMemorySize   = 0x");
 	udbg_puthex(naca->physicalMemorySize);
 	udbg_putc('\n');

arch/ppc64/kernel/smp.c

@@ -54,13 +54,15 @@
 
 int smp_threads_ready = 0;
 volatile int smp_commenced = 0;
-int smp_num_cpus = 1;
 int smp_tb_synchronized = 0;
 spinlock_t kernel_flag __cacheline_aligned = SPIN_LOCK_UNLOCKED;
 unsigned long cache_decay_ticks;
 static int max_cpus __initdata = NR_CPUS;
 
-unsigned long cpu_online_map;
+/* initialised so it doesnt end up in bss */
+unsigned long cpu_online_map = 0;
+int boot_cpuid = 0;
+int ppc64_is_smp = 0;
 
 volatile unsigned long cpu_callin_map[NR_CPUS] = {0,};
 
@@ -98,7 +100,7 @@ void iSeries_smp_message_recv( struct pt_regs * regs )
 	int cpu = smp_processor_id();
 	int msg;
 
-	if ( smp_num_cpus < 2 )
+	if ( num_online_cpus() < 2 )
 		return;
 
 	for ( msg = 0; msg < 4; ++msg )
@@ -110,11 +112,16 @@ void iSeries_smp_message_recv( struct pt_regs * regs )
 static void smp_iSeries_message_pass(int target, int msg, unsigned long data, int wait)
 {
 	int i;
-	for (i = 0; i < smp_num_cpus; ++i) {
-		if ( (target == MSG_ALL) || 
+
+	for (i = 0; i < NR_CPUS; ++i) {
+		if (!cpu_online(i))
+			continue;
+
+		if ((target == MSG_ALL) || 
 		    (target == i) || 
-                    ((target == MSG_ALL_BUT_SELF) && (i != smp_processor_id())) ) {
-			set_bit( msg, &iSeries_smp_message[i] );
+		    ((target == MSG_ALL_BUT_SELF) &&
+		     (i != smp_processor_id())) ) {
+			set_bit(msg, &iSeries_smp_message[i]);
 			HvCall_sendIPI(&(paca[i]));
 		}
 	}
@@ -135,7 +142,7 @@ static int smp_iSeries_numProcs(void)
 	return np;
 }
 
-static int smp_iSeries_probe(void)
+static void smp_iSeries_probe(void)
 {
 	unsigned i;
 	unsigned np;
@@ -151,7 +158,6 @@ static int smp_iSeries_probe(void)
 	}
 
 	smp_tb_synchronized = 1;
-	return np;
 }
 
 static void smp_iSeries_kick_cpu(int nr)
@@ -191,8 +197,10 @@ void smp_init_iSeries(void)
 	ppc_md.smp_probe        = smp_iSeries_probe;
 	ppc_md.smp_kick_cpu     = smp_iSeries_kick_cpu;
 	ppc_md.smp_setup_cpu    = smp_iSeries_setup_cpu;
-
+#ifdef CONFIG_PPC_ISERIES
+#warning fix for iseries
 	naca->processorCount	= smp_iSeries_numProcs();
+#endif
 }
 
 
@@ -220,13 +228,10 @@ smp_openpic_message_pass(int target, int msg, unsigned long data, int wait)
 	}
 }
 
-static int
-smp_chrp_probe(void)
+static void smp_chrp_probe(void)
 {
-	if (naca->processorCount > 1)
+	if (ppc64_is_smp)
 		openpic_request_IPIs();
-
-	return naca->processorCount;
 }
 
 static void
@@ -253,14 +258,14 @@ smp_kick_cpu(int nr)
 
 extern struct gettimeofday_struct do_gtod;
 
-static void smp_space_timers( unsigned nr )
+static void smp_space_timers()
 {
-	unsigned long offset, i;
+	int i;
+	unsigned long offset = tb_ticks_per_jiffy / NR_CPUS;
 
-	offset = tb_ticks_per_jiffy / nr;
-	for ( i=1; i<nr; ++i ) {
-		paca[i].next_jiffy_update_tb = paca[i-1].next_jiffy_update_tb + offset;
-	}
+	for (i = 1; i < NR_CPUS; ++i)
+		paca[i].next_jiffy_update_tb =
+			paca[i-1].next_jiffy_update_tb + offset;
 }
 
 static void
@@ -272,7 +277,7 @@ smp_chrp_setup_cpu(int cpu_nr)
 	if (naca->platform == PLATFORM_PSERIES_LPAR) {
 		/* timebases already synced under the hypervisor. */
 		paca[cpu_nr].next_jiffy_update_tb = tb_last_stamp = get_tb();
-		if (cpu_nr == 0) {
+		if (cpu_nr == boot_cpuid) {
 			do_gtod.tb_orig_stamp = tb_last_stamp;
 			/* Should update do_gtod.stamp_xsec.
 			 * For now we leave it which means the time can be some
@@ -281,9 +286,9 @@ smp_chrp_setup_cpu(int cpu_nr)
 		}
 		smp_tb_synchronized = 1;
 	} else {
-		if (cpu_nr == 0) {
+		if (cpu_nr == boot_cpuid) {
 			/* wait for all the others */
-			while (atomic_read(&ready) < smp_num_cpus)
+			while (atomic_read(&ready) < num_online_cpus())
 				barrier();
 			atomic_set(&ready, 1);
 			/* freeze the timebase */
@@ -291,9 +296,9 @@ smp_chrp_setup_cpu(int cpu_nr)
 			mb();
 			frozen = 1;
 			set_tb(0, 0);
-			paca[0].next_jiffy_update_tb = 0;
-			smp_space_timers(smp_num_cpus);
-			while (atomic_read(&ready) < smp_num_cpus)
+			paca[boot_cpuid].next_jiffy_update_tb = 0;
+			smp_space_timers();
+			while (atomic_read(&ready) < num_online_cpus())
 				barrier();
 			/* thaw the timebase again */
 			rtas_call(rtas_token("thaw-time-base"), 0, 1, NULL);
@@ -317,7 +322,7 @@ smp_chrp_setup_cpu(int cpu_nr)
 	if (OpenPIC_Addr) {
 		do_openpic_setup_cpu();
 	} else {
-	  if (cpu_nr > 0)
+		if (cpu_nr != boot_cpuid)
 			xics_setup_cpu();
 	}
 }
@@ -327,7 +332,10 @@ smp_xics_message_pass(int target, int msg, unsigned long data, int wait)
 {
 	int i;
 
-	for (i = 0; i < smp_num_cpus; ++i) {
+	for (i = 0; i < NR_CPUS; ++i) {
+		if (!cpu_online(i))
+			continue;
+
 		if (target == MSG_ALL || target == i
 		    || (target == MSG_ALL_BUT_SELF
 			&& i != smp_processor_id())) {
@@ -338,10 +346,8 @@ smp_xics_message_pass(int target, int msg, unsigned long data, int wait)
 	}
 }
 
-static int
-smp_xics_probe(void)
+static void smp_xics_probe(void)
 {
-	return naca->processorCount;
 }
 
 /* This is called very early */
@@ -429,7 +435,6 @@ static void stop_this_cpu(void *dummy)
 void smp_send_stop(void)
 {
 	smp_call_function(stop_this_cpu, NULL, 1, 0);
-	smp_num_cpus = 1;
 }
 
 /*
@@ -467,7 +472,7 @@ int smp_call_function (void (*func) (void *info), void *info, int nonatomic,
 
 { 
 	struct call_data_struct data;
-	int ret = -1, cpus = smp_num_cpus-1;
+	int ret = -1, cpus = num_online_cpus()-1;
 	int timeout;
 
 	if (!cpus)
@@ -554,29 +559,23 @@ struct thread_struct *current_set[NR_CPUS] = {&init_thread_union, 0};
 
 void __init smp_boot_cpus(void)
 {
-	int i, cpu_nr;
+	int i, cpu_nr = 0;
 	struct task_struct *p;
 
 	printk("Entering SMP Mode...\n");
 
-	smp_num_cpus = 1;
-        smp_store_cpu_info(0);
-	cpu_online_map = 1UL;
+        smp_store_cpu_info(boot_cpuid);
+	cpu_callin_map[boot_cpuid] = 1;
 
-	/*
-	 * assume for now that the first cpu booted is
-	 * cpu 0, the master -- Cort
-	 */
-	cpu_callin_map[0] = 1;
 	/* XXX buggy - Anton */
 	current_thread_info()->cpu = 0;
 
 	for (i = 0; i < NR_CPUS; i++) {
 		paca[i].prof_counter = 1;
 		paca[i].prof_multiplier = 1;
-		if(i != 0) {
+		if (i != boot_cpuid) {
 		        /*
-			 * Processor 0's segment table is statically 
+			 * the boot cpu segment table is statically 
 			 * initialized to real address 0x5000.  The
 			 * Other processor's tables are created and
 			 * initialized here.
@@ -593,28 +592,33 @@ void __init smp_boot_cpus(void)
 	 */
 	cache_decay_ticks = HZ/100;
 
-	/* Probe arch for CPUs */
-	cpu_nr = ppc_md.smp_probe();
+	ppc_md.smp_probe();
 
+	for (i = 0; i < NR_CPUS; i++) {
+		if (paca[i].active)
+			cpu_nr++;
+	}
 	printk("Probe found %d CPUs\n", cpu_nr);
 
-	/*
-	 * only check for cpus we know exist.  We keep the callin map
-	 * with cpus at the bottom -- Cort
-	 */
-	if (cpu_nr > max_cpus)
-		cpu_nr = max_cpus;
-
 #ifdef CONFIG_ISERIES
-	smp_space_timers( cpu_nr );
+	smp_space_timers();
 #endif
 
 	printk("Waiting for %d CPUs\n", cpu_nr-1);
 
-	for ( i = 1 ; i < cpu_nr; i++ ) {
+	for (i = 1 ; i < NR_CPUS; i++) {
 		int c;
 		struct pt_regs regs;
 
+		if (!paca[i].active)
+			continue;
+
+		if (i == boot_cpuid)
+			continue;
+
+		if (num_online_cpus() >= max_cpus)
+			break;
+
 		/* create a process for the processor */
 		/* we don't care about the values in regs since we'll
 		   never reschedule the forked task. */
@@ -658,16 +662,15 @@ void __init smp_boot_cpus(void)
 		{
 			printk("Processor %d found.\n", i);
 			/* this sync's the decr's -- Cort */
-			smp_num_cpus++;
 		} else {
 			printk("Processor %d is stuck.\n", i);
 		}
 	}
 
-	/* Setup CPU 0 last (important) */
-	ppc_md.smp_setup_cpu(0);
+	/* Setup boot cpu last (important) */
+	ppc_md.smp_setup_cpu(boot_cpuid);
 
-	if (smp_num_cpus < 2) {
+	if (num_online_cpus() < 2) {
 	        tb_last_stamp = get_tb();
 		smp_tb_synchronized = 1;
 	}
@@ -689,8 +692,9 @@ void __init smp_callin(void)
 
         smp_store_cpu_info(cpu);
 	set_dec(paca[cpu].default_decr);
-	cpu_callin_map[cpu] = 1;
 	set_bit(smp_processor_id(), &cpu_online_map);
+	smp_mb();
+	cpu_callin_map[cpu] = 1;
 
 	ppc_md.smp_setup_cpu(cpu);
 	

arch/ppc64/kernel/xics.c

@@ -340,7 +340,7 @@ xics_init_IRQ( void )
 	/* Find the server numbers for the boot cpu. */
 	for (np = find_type_devices("cpu"); np; np = np->next) {
 		ireg = (uint *)get_property(np, "reg", &ilen);
-		if (ireg && ireg[0] == hard_smp_processor_id()) {
+		if (ireg && ireg[0] == smp_processor_id()) {
 			ireg = (uint *)get_property(np, "ibm,ppc-interrupt-gserver#s", &ilen);
 			i = ilen / sizeof(int);
 			if (ireg && i > 0) {
@@ -371,10 +371,12 @@ xics_init_IRQ( void )
 
 	if (naca->platform == PLATFORM_PSERIES) {
 #ifdef CONFIG_SMP
-		for (i = 0; i < naca->processorCount; ++i) {
+		for (i = 0; i < NR_CPUS; ++i) {
+			if (!paca[i].active)
+				continue;
 			xics_info.per_cpu[i] =
-			  __ioremap((ulong)inodes[get_hard_smp_processor_id(i)].addr, 
-				  (ulong)inodes[get_hard_smp_processor_id(i)].size, _PAGE_NO_CACHE);
+			  __ioremap((ulong)inodes[i].addr, 
+				  (ulong)inodes[i].size, _PAGE_NO_CACHE);
 		}
 #else
 		xics_info.per_cpu[0] = __ioremap((ulong)intr_base, intr_size, _PAGE_NO_CACHE);
@@ -395,7 +397,7 @@ xics_init_IRQ( void )
 	for (; i < NR_IRQS; ++i)
 		irq_desc[i].handler = &xics_pic;
 
-	ops->cppr_info(0, 0xff);
+	ops->cppr_info(boot_cpuid, 0xff);
 	iosync();
 	if (xics_irq_8259_cascade != -1) {
 		if (request_irq(xics_irq_8259_cascade + XICS_IRQ_OFFSET, no_action,
@@ -420,23 +422,6 @@ void xics_isa_init(void)
 	i8259_init();
 }
 
-/*
- * Find first logical cpu and return its physical cpu number
- */
-static inline u32 physmask(u32 cpumask)
-{
-	int i;
-
-	for (i = 0; i < smp_num_cpus; ++i, cpumask >>= 1) {
-		if (cpumask & 1)
-			return get_hard_smp_processor_id(i);
-	}
-
-	printk(KERN_ERR "xics_set_affinity: invalid irq mask\n");
-
-	return default_distrib_server;
-}
-
 void xics_set_affinity(unsigned int virq, unsigned long cpumask)
 {
         irq_desc_t *desc = irq_desc + virq;
@@ -462,10 +447,13 @@ void xics_set_affinity(unsigned int virq, unsigned long cpumask)
 	}
 
 	/* For the moment only implement delivery to all cpus or one cpu */
-	if (cpumask == 0xffffffff)
+	if (cpumask == 0xffffffff) {
 		newmask = default_distrib_server;
-	else
-		newmask = physmask(cpumask);
+	} else {
+		if (!(cpumask & cpu_online_map))
+			goto out;
+		newmask = find_first_bit(&cpumask, 32);
+	}
 
 	status = rtas_call(ibm_set_xive, 3, 1, NULL,
 				irq, newmask, xics_status[1]);

include/asm-ppc64/hardirq.h

@@ -15,7 +15,7 @@
 #include <linux/config.h>
 #include <linux/brlock.h>
 #include <linux/spinlock.h>
-
+#include <asm/smp.h>
 
 typedef struct {
 	unsigned long __softirq_pending;
@@ -67,8 +67,8 @@ static __inline__ int irqs_running(void)
 {
 	int i;
 
-	for (i = 0; i < smp_num_cpus; i++)
-		if (local_irq_count(cpu_logical_map(i)))
+	for (i = 0; i < NR_CPUS; i++)
+		if (local_irq_count(i))
 			return 1;
 	return 0;
 }

include/asm-ppc64/machdep.h

@@ -62,7 +62,7 @@ struct machdep_calls {
 					    int msg, 
 					    unsigned long data,
 					    int wait);
-	int		(*smp_probe)(void);
+	void		(*smp_probe)(void);
 	void		(*smp_kick_cpu)(int nr);
 	void		(*smp_setup_cpu)(int nr);
 

include/asm-ppc64/naca.h

@@ -18,7 +18,6 @@ struct naca_struct {
 	u64 xRamDiskSize;		/* In pages */
 	struct paca_struct *paca;	/* Ptr to an array of pacas */
 	u64 debug_switch;		/* Bits to control debug printing */
-	u16 processorCount;		/* # of physical processors */
 	u16 dCacheL1LineSize;		/* Line size of L1 DCache in bytes */
 	u16 dCacheL1LogLineSize;	/* Log-2 of DCache line size */
 	u16 dCacheL1LinesPerPage;	/* DCache lines per page */

include/asm-ppc64/paca.h

@@ -71,7 +71,7 @@ struct paca_struct {
 	struct ItLpRegSave *xLpRegSavePtr; /* Pointer to LpRegSave for PLIC	0x08 */
 	u64 xCurrent;  		        /* Pointer to current			0x10 */
 	u16 xPacaIndex;			/* Logical processor number		0x18 */
-	u16 xHwProcNum;			/* Actual Hardware Processor Number	0x1a */
+	u16 active;			/* Is this cpu active?			0x1a */
 	u32 default_decr;		/* Default decrementer value		0x1c */	
 	u64 xHrdIntStack;		/* Stack for hardware interrupts	0x20 */
 	u64 xKsave;			/* Saved Kernel stack addr or zero	0x28 */

include/asm-ppc64/smp.h

@@ -39,16 +39,27 @@ extern void smp_send_reschedule_all(void);
 
 #define NO_PROC_ID		0xFF            /* No processor magic marker */
 
-/* 1 to 1 mapping on PPC -- Cort */
-#define cpu_logical_map(cpu) (cpu)
-#define cpu_number_map(x) (x)
+#define cpu_online(cpu)	test_bit((cpu), &cpu_online_map)
+
+#define cpu_possible(cpu)	paca[cpu].active
+
+static inline int num_online_cpus(void)
+{
+	int i, nr = 0;
+
+	for (i = 0; i < NR_CPUS; i++)
+		nr += test_bit(i, &cpu_online_map);
+
+	return nr;
+}
+
 extern volatile unsigned long cpu_callin_map[NR_CPUS];
 
 #define smp_processor_id() (get_paca()->xPacaIndex)
-#define hard_smp_processor_id() (get_paca()->xHwProcNum)
-#define get_hard_smp_processor_id(CPU) (paca[(CPU)].xHwProcNum)
-
 
+/* remove when the boot sequence gets rewritten to use hotplug interface */
+extern int boot_cpuid;
+extern int ppc64_is_smp;
 
 /* Since OpenPIC has only 4 IPIs, we use slightly different message numbers.
  *