SPARC64: Port to new cpu hotplug startup sequence.

arch/sparc64/kernel/smp.c

@@ -50,24 +50,12 @@ static int smp_activated;
 /* Kernel spinlock */
 spinlock_t kernel_flag __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 
-volatile int smp_processors_ready = 0;
 atomic_t sparc64_num_cpus_online = ATOMIC_INIT(0);
 unsigned long cpu_online_map = 0;
-int smp_threads_ready = 0;
-
-void __init smp_setup(char *str, int *ints)
-{
-	/* XXX implement me XXX */
-}
-
-static int max_cpus = NR_CPUS;
-static int __init maxcpus(char *str)
-{
-	get_option(&str, &max_cpus);
-	return 1;
-}
-
-__setup("maxcpus=", maxcpus);
+atomic_t sparc64_num_cpus_possible = ATOMIC_INIT(0);
+unsigned long phys_cpu_present_map = 0;
+static unsigned long smp_commenced_mask;
+static unsigned long cpu_callout_map;
 
 void smp_info(struct seq_file *m)
 {
@@ -121,10 +109,6 @@ void __init smp_store_cpu_info(int id)
 		cpu_data[id].irq_worklists[i] = 0;
 }
 
-void __init smp_commence(void)
-{
-}
-
 static void smp_setup_percpu_timer(void);
 
 static volatile unsigned long callin_flag = 0;
@@ -216,8 +200,11 @@ void __init smp_callin(void)
 	atomic_inc(&init_mm.mm_count);
 	current->active_mm = &init_mm;
 
-	while (!smp_threads_ready)
+	while (!test_bit(cpuid, &smp_commenced_mask))
 		membar("#LoadLoad");
+
+	set_bit(cpuid, &cpu_online_map);
+	atomic_inc(&sparc64_num_cpus_online);
 }
 
 void cpu_panic(void)
@@ -236,9 +223,7 @@ extern unsigned long sparc64_cpu_startup;
  */
 static struct thread_info *cpu_new_thread = NULL;
 
-static void smp_tune_scheduling(void);
-
-void __init smp_boot_cpus(void)
+static void __init smp_boot_cpus(unsigned int max_cpus)
 {
 	int cpucount = 0, i;
 
@@ -246,10 +231,8 @@ void __init smp_boot_cpus(void)
 	local_irq_enable();
 	smp_store_cpu_info(boot_cpu_id);
 
-	if (linux_num_cpus == 1) {
-		smp_tune_scheduling();
+	if (linux_num_cpus == 1)
 		return;
-	}
 
 	for (i = 0; i < NR_CPUS; i++) {
 		if (i == boot_cpu_id)
@@ -257,7 +240,7 @@ void __init smp_boot_cpus(void)
 
 		if ((cpucount + 1) == max_cpus)
 			goto ignorecpu;
-		if (cpu_online(i)) {
+		if (test_bit(i, &phys_cpu_present_map)) {
 			unsigned long entry =
 				(unsigned long)(&sparc64_cpu_startup);
 			unsigned long cookie =
@@ -281,6 +264,7 @@ void __init smp_boot_cpus(void)
 				if (linux_cpus[no].mid == i)
 					break;
 			cpu_new_thread = p->thread_info;
+			set_bit(i, &cpu_callout_map);
 			prom_startcpu(linux_cpus[no].prom_node,
 				      entry, cookie);
 			for (timeout = 0; timeout < 5000000; timeout++) {
@@ -289,28 +273,21 @@ void __init smp_boot_cpus(void)
 				udelay(100);
 			}
 			if (callin_flag) {
-				atomic_inc(&sparc64_num_cpus_online);
 				prom_cpu_nodes[i] = linux_cpus[no].prom_node;
 				prom_printf("OK\n");
 			} else {
 				cpucount--;
 				printk("Processor %d is stuck.\n", i);
 				prom_printf("FAILED\n");
+				clear_bit(i, &cpu_callout_map);
 			}
-			if (!callin_flag) {
 ignorecpu:
-				clear_bit(i, &cpu_online_map);
-			}
-
 		}
 	}
 	cpu_new_thread = NULL;
 	if (cpucount == 0) {
 		if (max_cpus != 1)
 			printk("Error: only one processor found.\n");
-		memset(&cpu_online_map, 0, sizeof(cpu_online_map));
-		set_bit(smp_processor_id(), &cpu_online_map);
-		atomic_set(&sparc64_num_cpus_online, 1);
 	} else {
 		unsigned long bogosum = 0;
 
@@ -325,14 +302,6 @@ void __init smp_boot_cpus(void)
 		       (bogosum/(5000/HZ))%100);
 		smp_activated = 1;
 	}
-
-	/* We want to run this with all the other cpus spinning
-	 * in the kernel.
-	 */
-	smp_tune_scheduling();
-
-	smp_processors_ready = 1;
-	membar("#StoreStore | #StoreLoad");
 }
 
 static void spitfire_xcall_helper(u64 data0, u64 data1, u64 data2, u64 pstate, unsigned long cpu)
@@ -532,7 +501,6 @@ static void cheetah_xcall_deliver(u64 data0, u64 data1, u64 data2, unsigned long
  */
 static void smp_cross_call_masked(unsigned long *func, u32 ctx, u64 data1, u64 data2, unsigned long mask)
 {
-	if (smp_processors_ready) {
 	u64 data0 = (((u64)ctx)<<32 | (((u64)func) & 0xffffffff));
 
 	mask &= cpu_online_map;
@@ -542,9 +510,7 @@ static void smp_cross_call_masked(unsigned long *func, u32 ctx, u64 data1, u64 d
 		spitfire_xcall_deliver(data0, data1, data2, mask);
 	else
 		cheetah_xcall_deliver(data0, data1, data2, mask);
-
 	/* NOTE: Caller runs local copy on master. */
-	}
 }
 
 /* Send cross call to all processors except self. */
@@ -660,7 +626,6 @@ static __inline__ void __local_flush_dcache_page(struct page *page)
 
 void smp_flush_dcache_page_impl(struct page *page, int cpu)
 {
-	if (smp_processors_ready) {
 	unsigned long mask = 1UL << cpu;
 
 #ifdef CONFIG_DEBUG_DCFLUSH
@@ -691,14 +656,11 @@ void smp_flush_dcache_page_impl(struct page *page, int cpu)
 		atomic_inc(&dcpage_flushes_xcall);
 #endif
 	}
-	}
 }
 
 void flush_dcache_page_all(struct mm_struct *mm, struct page *page)
 {
-	if (smp_processors_ready) {
-		unsigned long mask =
-			cpu_online_map & ~(1UL << smp_processor_id());
+	unsigned long mask = cpu_online_map & ~(1UL << smp_processor_id());
 	u64 data0;
 
 #ifdef CONFIG_DEBUG_DCFLUSH
@@ -725,24 +687,20 @@ void flush_dcache_page_all(struct mm_struct *mm, struct page *page)
 #endif
  flush_self:
 	__local_flush_dcache_page(page);
-	}
 }
 
 void smp_receive_signal(int cpu)
 {
-	if (smp_processors_ready) {
 	unsigned long mask = 1UL << cpu;
 
 	if ((cpu_online_map & mask) != 0) {
-			u64 data0 =
-			  (((u64)&xcall_receive_signal) & 0xffffffff);
+		u64 data0 = (((u64)&xcall_receive_signal) & 0xffffffff);
 
 		if (tlb_type == spitfire)
 			spitfire_xcall_deliver(data0, 0, 0, mask);
 		else
 			cheetah_xcall_deliver(data0, 0, 0, mask);
 	}
-	}
 }
 
 void smp_receive_signal_client(int irq, struct pt_regs *regs)
@@ -934,7 +892,6 @@ static unsigned long penguins_are_doing_time;
 
 void smp_capture(void)
 {
-	if (smp_processors_ready) {
 	int result = __atomic_add(1, &smp_capture_depth);
 
 	membar("#StoreStore | #LoadStore");
@@ -955,12 +912,10 @@ void smp_capture(void)
 		printk("done\n");
 #endif
 	}
-	}
 }
 
 void smp_release(void)
 {
-	if (smp_processors_ready) {
 	if (atomic_dec_and_test(&smp_capture_depth)) {
 #ifdef CAPTURE_DEBUG
 		printk("CPU[%d]: Giving pardon to "
@@ -971,7 +926,6 @@ void smp_release(void)
 		membar("#StoreStore | #StoreLoad");
 		atomic_dec(&smp_capture_registry);
 	}
-	}
 }
 
 /* Imprisoned penguins run with %pil == 15, but PSTATE_IE set, so they
@@ -1006,7 +960,6 @@ extern unsigned long xcall_promstop;
 
 void smp_promstop_others(void)
 {
-	if (smp_processors_ready)
 	smp_cross_call(&xcall_promstop, 0, 0, 0);
 }
 
@@ -1178,13 +1131,17 @@ void __init smp_tick_init(void)
 
 	atomic_set(&sparc64_num_cpus_online, 1);
 	memset(&cpu_online_map, 0, sizeof(cpu_online_map));
-	for (i = 0; i < linux_num_cpus; i++) {
-		if (linux_cpus[i].mid < NR_CPUS)
-			set_bit(linux_cpus[i].mid, &cpu_online_map);
-	}
-
+	set_bit(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;
+
+	for (i = 0; i < linux_num_cpus; i++) {
+		if (linux_cpus[i].mid < NR_CPUS) {
+			set_bit(linux_cpus[i].mid,
+				&phys_cpu_present_map);
+			atomic_inc(&sparc64_num_cpus_possible);
+		}
+	}
 }
 
 cycles_t cacheflush_time;
@@ -1312,3 +1269,24 @@ int setup_profiling_timer(unsigned int multiplier)
 
 	return 0;
 }
+
+void __init smp_prepare_cpus(unsigned int max_cpus)
+{
+	smp_boot_cpus(max_cpus);
+}
+
+int __devinit __cpu_up(unsigned int cpu)
+{
+	set_bit(cpu, &smp_commenced_mask);
+	while (!test_bit(cpu, &cpu_online_map))
+		mb();
+	return 0;
+}
+
+void __init smp_cpus_done(unsigned int max_cpus)
+{
+	/* We want to run this with all the other cpus spinning
+	 * in the kernel.
+	 */
+	smp_tune_scheduling();
+}

arch/sparc64/kernel/sparc64_ksyms.c

@@ -145,6 +145,8 @@ EXPORT_SYMBOL(cpu_data);
 /* CPU online map and active count.  */
 EXPORT_SYMBOL(cpu_online_map);
 EXPORT_SYMBOL(sparc64_num_cpus_online);
+EXPORT_SYMBOL(phys_cpu_present_map);
+EXPORT_SYMBOL(sparc64_num_cpus_possible);
 
 /* Spinlock debugging library, optional. */
 #ifdef CONFIG_DEBUG_SPINLOCK

include/asm-sparc64/smp.h

@@ -65,11 +65,19 @@ extern cpuinfo_sparc cpu_data[NR_CPUS];
 #include <asm/atomic.h>
 
 extern unsigned char boot_cpu_id;
+
+extern unsigned long phys_cpu_present_map;
+#define cpu_possible(cpu)	(phys_cpu_present_map & (1UL << (cpu)))
+
 extern unsigned long cpu_online_map;
 #define cpu_online(cpu)		(cpu_online_map & (1UL << (cpu)))
+
 extern atomic_t sparc64_num_cpus_online;
 #define num_online_cpus()	(atomic_read(&sparc64_num_cpus_online))
 
+extern atomic_t sparc64_num_cpus_possible;
+#define num_possible_cpus()	(atomic_read(&sparc64_num_cpus_possible))
+
 static inline int any_online_cpu(unsigned long mask)
 {
 	if ((mask &= cpu_online_map) != 0UL)
@@ -81,10 +89,6 @@ static inline int any_online_cpu(unsigned long mask)
  *	General functions that each host system must provide.
  */
 
-extern void smp_callin(void);
-extern void smp_boot_cpus(void);
-extern void smp_store_cpu_info(int id);
-
 extern __inline__ int hard_smp_processor_id(void)
 {
 	if (tlb_type == cheetah) {