Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* smp.c: Sparc64 SMP support. |
| 2 | * |
| 3 | * Copyright (C) 1997 David S. Miller (davem@caip.rutgers.edu) |
| 4 | */ |
| 5 | |
| 6 | #include <linux/module.h> |
| 7 | #include <linux/kernel.h> |
| 8 | #include <linux/sched.h> |
| 9 | #include <linux/mm.h> |
| 10 | #include <linux/pagemap.h> |
| 11 | #include <linux/threads.h> |
| 12 | #include <linux/smp.h> |
| 13 | #include <linux/smp_lock.h> |
| 14 | #include <linux/interrupt.h> |
| 15 | #include <linux/kernel_stat.h> |
| 16 | #include <linux/delay.h> |
| 17 | #include <linux/init.h> |
| 18 | #include <linux/spinlock.h> |
| 19 | #include <linux/fs.h> |
| 20 | #include <linux/seq_file.h> |
| 21 | #include <linux/cache.h> |
| 22 | #include <linux/jiffies.h> |
| 23 | #include <linux/profile.h> |
| 24 | #include <linux/bootmem.h> |
| 25 | |
| 26 | #include <asm/head.h> |
| 27 | #include <asm/ptrace.h> |
| 28 | #include <asm/atomic.h> |
| 29 | #include <asm/tlbflush.h> |
| 30 | #include <asm/mmu_context.h> |
| 31 | #include <asm/cpudata.h> |
| 32 | |
| 33 | #include <asm/irq.h> |
| 34 | #include <asm/page.h> |
| 35 | #include <asm/pgtable.h> |
| 36 | #include <asm/oplib.h> |
| 37 | #include <asm/uaccess.h> |
| 38 | #include <asm/timer.h> |
| 39 | #include <asm/starfire.h> |
| 40 | #include <asm/tlb.h> |
David S. Miller | 56fb4df | 2006-02-26 23:24:22 -0800 | [diff] [blame] | 41 | #include <asm/sections.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | extern void calibrate_delay(void); |
| 44 | |
| 45 | /* Please don't make this stuff initdata!!! --DaveM */ |
| 46 | static unsigned char boot_cpu_id; |
| 47 | |
Andrew Morton | c12a828 | 2005-07-12 12:09:43 -0700 | [diff] [blame] | 48 | cpumask_t cpu_online_map __read_mostly = CPU_MASK_NONE; |
| 49 | cpumask_t phys_cpu_present_map __read_mostly = CPU_MASK_NONE; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | static cpumask_t smp_commenced_mask; |
| 51 | static cpumask_t cpu_callout_map; |
| 52 | |
| 53 | void smp_info(struct seq_file *m) |
| 54 | { |
| 55 | int i; |
| 56 | |
| 57 | seq_printf(m, "State:\n"); |
| 58 | for (i = 0; i < NR_CPUS; i++) { |
| 59 | if (cpu_online(i)) |
| 60 | seq_printf(m, |
| 61 | "CPU%d:\t\tonline\n", i); |
| 62 | } |
| 63 | } |
| 64 | |
| 65 | void smp_bogo(struct seq_file *m) |
| 66 | { |
| 67 | int i; |
| 68 | |
| 69 | for (i = 0; i < NR_CPUS; i++) |
| 70 | if (cpu_online(i)) |
| 71 | seq_printf(m, |
| 72 | "Cpu%dBogo\t: %lu.%02lu\n" |
| 73 | "Cpu%dClkTck\t: %016lx\n", |
| 74 | i, cpu_data(i).udelay_val / (500000/HZ), |
| 75 | (cpu_data(i).udelay_val / (5000/HZ)) % 100, |
| 76 | i, cpu_data(i).clock_tick); |
| 77 | } |
| 78 | |
| 79 | void __init smp_store_cpu_info(int id) |
| 80 | { |
David S. Miller | f03b8a5 | 2006-02-15 00:35:50 -0800 | [diff] [blame] | 81 | int cpu_node, def; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | |
| 83 | /* multiplier and counter set by |
| 84 | smp_setup_percpu_timer() */ |
| 85 | cpu_data(id).udelay_val = loops_per_jiffy; |
| 86 | |
| 87 | cpu_find_by_mid(id, &cpu_node); |
| 88 | cpu_data(id).clock_tick = prom_getintdefault(cpu_node, |
| 89 | "clock-frequency", 0); |
| 90 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 91 | cpu_data(id).idle_volume = 1; |
David S. Miller | 80dc0d6 | 2005-09-26 00:32:17 -0700 | [diff] [blame] | 92 | |
David S. Miller | f03b8a5 | 2006-02-15 00:35:50 -0800 | [diff] [blame] | 93 | def = ((tlb_type == hypervisor) ? (8 * 1024) : (16 * 1024)); |
David S. Miller | 80dc0d6 | 2005-09-26 00:32:17 -0700 | [diff] [blame] | 94 | cpu_data(id).dcache_size = prom_getintdefault(cpu_node, "dcache-size", |
David S. Miller | f03b8a5 | 2006-02-15 00:35:50 -0800 | [diff] [blame] | 95 | def); |
| 96 | |
| 97 | def = 32; |
David S. Miller | 80dc0d6 | 2005-09-26 00:32:17 -0700 | [diff] [blame] | 98 | cpu_data(id).dcache_line_size = |
David S. Miller | f03b8a5 | 2006-02-15 00:35:50 -0800 | [diff] [blame] | 99 | prom_getintdefault(cpu_node, "dcache-line-size", def); |
| 100 | |
| 101 | def = 16 * 1024; |
David S. Miller | 80dc0d6 | 2005-09-26 00:32:17 -0700 | [diff] [blame] | 102 | cpu_data(id).icache_size = prom_getintdefault(cpu_node, "icache-size", |
David S. Miller | f03b8a5 | 2006-02-15 00:35:50 -0800 | [diff] [blame] | 103 | def); |
| 104 | |
| 105 | def = 32; |
David S. Miller | 80dc0d6 | 2005-09-26 00:32:17 -0700 | [diff] [blame] | 106 | cpu_data(id).icache_line_size = |
David S. Miller | f03b8a5 | 2006-02-15 00:35:50 -0800 | [diff] [blame] | 107 | prom_getintdefault(cpu_node, "icache-line-size", def); |
| 108 | |
| 109 | def = ((tlb_type == hypervisor) ? |
| 110 | (3 * 1024 * 1024) : |
| 111 | (4 * 1024 * 1024)); |
David S. Miller | 80dc0d6 | 2005-09-26 00:32:17 -0700 | [diff] [blame] | 112 | cpu_data(id).ecache_size = prom_getintdefault(cpu_node, "ecache-size", |
David S. Miller | f03b8a5 | 2006-02-15 00:35:50 -0800 | [diff] [blame] | 113 | def); |
| 114 | |
| 115 | def = 64; |
David S. Miller | 80dc0d6 | 2005-09-26 00:32:17 -0700 | [diff] [blame] | 116 | cpu_data(id).ecache_line_size = |
David S. Miller | f03b8a5 | 2006-02-15 00:35:50 -0800 | [diff] [blame] | 117 | prom_getintdefault(cpu_node, "ecache-line-size", def); |
| 118 | |
David S. Miller | 80dc0d6 | 2005-09-26 00:32:17 -0700 | [diff] [blame] | 119 | printk("CPU[%d]: Caches " |
| 120 | "D[sz(%d):line_sz(%d)] " |
| 121 | "I[sz(%d):line_sz(%d)] " |
| 122 | "E[sz(%d):line_sz(%d)]\n", |
| 123 | id, |
| 124 | cpu_data(id).dcache_size, cpu_data(id).dcache_line_size, |
| 125 | cpu_data(id).icache_size, cpu_data(id).icache_line_size, |
| 126 | cpu_data(id).ecache_size, cpu_data(id).ecache_line_size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | } |
| 128 | |
| 129 | static void smp_setup_percpu_timer(void); |
| 130 | |
| 131 | static volatile unsigned long callin_flag = 0; |
| 132 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | void __init smp_callin(void) |
| 134 | { |
| 135 | int cpuid = hard_smp_processor_id(); |
| 136 | |
David S. Miller | 56fb4df | 2006-02-26 23:24:22 -0800 | [diff] [blame] | 137 | __local_per_cpu_offset = __per_cpu_offset(cpuid); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | |
David S. Miller | 4a07e64 | 2006-02-14 13:49:32 -0800 | [diff] [blame] | 139 | if (tlb_type == hypervisor) |
David S. Miller | 490384e | 2006-02-11 14:41:18 -0800 | [diff] [blame] | 140 | sun4v_ktsb_register(); |
David S. Miller | 481295f | 2006-02-07 21:51:08 -0800 | [diff] [blame] | 141 | |
David S. Miller | 56fb4df | 2006-02-26 23:24:22 -0800 | [diff] [blame] | 142 | __flush_tlb_all(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 143 | |
| 144 | smp_setup_percpu_timer(); |
| 145 | |
David S. Miller | 816242d | 2005-05-23 15:52:08 -0700 | [diff] [blame] | 146 | if (cheetah_pcache_forced_on) |
| 147 | cheetah_enable_pcache(); |
| 148 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | local_irq_enable(); |
| 150 | |
| 151 | calibrate_delay(); |
| 152 | smp_store_cpu_info(cpuid); |
| 153 | callin_flag = 1; |
| 154 | __asm__ __volatile__("membar #Sync\n\t" |
| 155 | "flush %%g6" : : : "memory"); |
| 156 | |
| 157 | /* Clear this or we will die instantly when we |
| 158 | * schedule back to this idler... |
| 159 | */ |
David S. Miller | db7d9a4 | 2005-07-24 19:36:26 -0700 | [diff] [blame] | 160 | current_thread_info()->new_child = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | |
| 162 | /* Attach to the address space of init_task. */ |
| 163 | atomic_inc(&init_mm.mm_count); |
| 164 | current->active_mm = &init_mm; |
| 165 | |
| 166 | while (!cpu_isset(cpuid, smp_commenced_mask)) |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 167 | rmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 168 | |
| 169 | cpu_set(cpuid, cpu_online_map); |
Nick Piggin | 5bfb5d6 | 2005-11-08 21:39:01 -0800 | [diff] [blame] | 170 | |
| 171 | /* idle thread is expected to have preempt disabled */ |
| 172 | preempt_disable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 173 | } |
| 174 | |
| 175 | void cpu_panic(void) |
| 176 | { |
| 177 | printk("CPU[%d]: Returns from cpu_idle!\n", smp_processor_id()); |
| 178 | panic("SMP bolixed\n"); |
| 179 | } |
| 180 | |
David S. Miller | d369ddd | 2005-07-10 15:45:11 -0700 | [diff] [blame] | 181 | static unsigned long current_tick_offset __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | |
| 183 | /* This tick register synchronization scheme is taken entirely from |
| 184 | * the ia64 port, see arch/ia64/kernel/smpboot.c for details and credit. |
| 185 | * |
| 186 | * The only change I've made is to rework it so that the master |
| 187 | * initiates the synchonization instead of the slave. -DaveM |
| 188 | */ |
| 189 | |
| 190 | #define MASTER 0 |
| 191 | #define SLAVE (SMP_CACHE_BYTES/sizeof(unsigned long)) |
| 192 | |
| 193 | #define NUM_ROUNDS 64 /* magic value */ |
| 194 | #define NUM_ITERS 5 /* likewise */ |
| 195 | |
| 196 | static DEFINE_SPINLOCK(itc_sync_lock); |
| 197 | static unsigned long go[SLAVE + 1]; |
| 198 | |
| 199 | #define DEBUG_TICK_SYNC 0 |
| 200 | |
| 201 | static inline long get_delta (long *rt, long *master) |
| 202 | { |
| 203 | unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0; |
| 204 | unsigned long tcenter, t0, t1, tm; |
| 205 | unsigned long i; |
| 206 | |
| 207 | for (i = 0; i < NUM_ITERS; i++) { |
| 208 | t0 = tick_ops->get_tick(); |
| 209 | go[MASTER] = 1; |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 210 | membar_storeload(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | while (!(tm = go[SLAVE])) |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 212 | rmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | go[SLAVE] = 0; |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 214 | wmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 215 | t1 = tick_ops->get_tick(); |
| 216 | |
| 217 | if (t1 - t0 < best_t1 - best_t0) |
| 218 | best_t0 = t0, best_t1 = t1, best_tm = tm; |
| 219 | } |
| 220 | |
| 221 | *rt = best_t1 - best_t0; |
| 222 | *master = best_tm - best_t0; |
| 223 | |
| 224 | /* average best_t0 and best_t1 without overflow: */ |
| 225 | tcenter = (best_t0/2 + best_t1/2); |
| 226 | if (best_t0 % 2 + best_t1 % 2 == 2) |
| 227 | tcenter++; |
| 228 | return tcenter - best_tm; |
| 229 | } |
| 230 | |
| 231 | void smp_synchronize_tick_client(void) |
| 232 | { |
| 233 | long i, delta, adj, adjust_latency = 0, done = 0; |
| 234 | unsigned long flags, rt, master_time_stamp, bound; |
| 235 | #if DEBUG_TICK_SYNC |
| 236 | struct { |
| 237 | long rt; /* roundtrip time */ |
| 238 | long master; /* master's timestamp */ |
| 239 | long diff; /* difference between midpoint and master's timestamp */ |
| 240 | long lat; /* estimate of itc adjustment latency */ |
| 241 | } t[NUM_ROUNDS]; |
| 242 | #endif |
| 243 | |
| 244 | go[MASTER] = 1; |
| 245 | |
| 246 | while (go[MASTER]) |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 247 | rmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 248 | |
| 249 | local_irq_save(flags); |
| 250 | { |
| 251 | for (i = 0; i < NUM_ROUNDS; i++) { |
| 252 | delta = get_delta(&rt, &master_time_stamp); |
| 253 | if (delta == 0) { |
| 254 | done = 1; /* let's lock on to this... */ |
| 255 | bound = rt; |
| 256 | } |
| 257 | |
| 258 | if (!done) { |
| 259 | if (i > 0) { |
| 260 | adjust_latency += -delta; |
| 261 | adj = -delta + adjust_latency/4; |
| 262 | } else |
| 263 | adj = -delta; |
| 264 | |
| 265 | tick_ops->add_tick(adj, current_tick_offset); |
| 266 | } |
| 267 | #if DEBUG_TICK_SYNC |
| 268 | t[i].rt = rt; |
| 269 | t[i].master = master_time_stamp; |
| 270 | t[i].diff = delta; |
| 271 | t[i].lat = adjust_latency/4; |
| 272 | #endif |
| 273 | } |
| 274 | } |
| 275 | local_irq_restore(flags); |
| 276 | |
| 277 | #if DEBUG_TICK_SYNC |
| 278 | for (i = 0; i < NUM_ROUNDS; i++) |
| 279 | printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n", |
| 280 | t[i].rt, t[i].master, t[i].diff, t[i].lat); |
| 281 | #endif |
| 282 | |
| 283 | printk(KERN_INFO "CPU %d: synchronized TICK with master CPU (last diff %ld cycles," |
| 284 | "maxerr %lu cycles)\n", smp_processor_id(), delta, rt); |
| 285 | } |
| 286 | |
| 287 | static void smp_start_sync_tick_client(int cpu); |
| 288 | |
| 289 | static void smp_synchronize_one_tick(int cpu) |
| 290 | { |
| 291 | unsigned long flags, i; |
| 292 | |
| 293 | go[MASTER] = 0; |
| 294 | |
| 295 | smp_start_sync_tick_client(cpu); |
| 296 | |
| 297 | /* wait for client to be ready */ |
| 298 | while (!go[MASTER]) |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 299 | rmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 300 | |
| 301 | /* now let the client proceed into his loop */ |
| 302 | go[MASTER] = 0; |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 303 | membar_storeload(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 304 | |
| 305 | spin_lock_irqsave(&itc_sync_lock, flags); |
| 306 | { |
| 307 | for (i = 0; i < NUM_ROUNDS*NUM_ITERS; i++) { |
| 308 | while (!go[MASTER]) |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 309 | rmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 310 | go[MASTER] = 0; |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 311 | wmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 312 | go[SLAVE] = tick_ops->get_tick(); |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 313 | membar_storeload(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 314 | } |
| 315 | } |
| 316 | spin_unlock_irqrestore(&itc_sync_lock, flags); |
| 317 | } |
| 318 | |
David S. Miller | 72aff53 | 2006-02-17 01:29:17 -0800 | [diff] [blame^] | 319 | extern void sun4v_init_mondo_queues(int use_bootmem, int cpu, int alloc, int load); |
| 320 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 321 | extern unsigned long sparc64_cpu_startup; |
| 322 | |
| 323 | /* The OBP cpu startup callback truncates the 3rd arg cookie to |
| 324 | * 32-bits (I think) so to be safe we have it read the pointer |
| 325 | * contained here so we work on >4GB machines. -DaveM |
| 326 | */ |
| 327 | static struct thread_info *cpu_new_thread = NULL; |
| 328 | |
| 329 | static int __devinit smp_boot_one_cpu(unsigned int cpu) |
| 330 | { |
| 331 | unsigned long entry = |
| 332 | (unsigned long)(&sparc64_cpu_startup); |
| 333 | unsigned long cookie = |
| 334 | (unsigned long)(&cpu_new_thread); |
| 335 | struct task_struct *p; |
David S. Miller | 7890f79 | 2006-02-15 02:26:54 -0800 | [diff] [blame] | 336 | int timeout, ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 337 | |
| 338 | p = fork_idle(cpu); |
| 339 | callin_flag = 0; |
Al Viro | f316964 | 2006-01-12 01:05:42 -0800 | [diff] [blame] | 340 | cpu_new_thread = task_thread_info(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | cpu_set(cpu, cpu_callout_map); |
| 342 | |
David S. Miller | 7890f79 | 2006-02-15 02:26:54 -0800 | [diff] [blame] | 343 | if (tlb_type == hypervisor) { |
David S. Miller | 72aff53 | 2006-02-17 01:29:17 -0800 | [diff] [blame^] | 344 | /* Alloc the mondo queues, cpu will load them. */ |
| 345 | sun4v_init_mondo_queues(0, cpu, 1, 0); |
| 346 | |
David S. Miller | 7890f79 | 2006-02-15 02:26:54 -0800 | [diff] [blame] | 347 | prom_startcpu_cpuid(cpu, entry, cookie); |
| 348 | } else { |
| 349 | int cpu_node; |
| 350 | |
| 351 | cpu_find_by_mid(cpu, &cpu_node); |
| 352 | prom_startcpu(cpu_node, entry, cookie); |
| 353 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 354 | |
| 355 | for (timeout = 0; timeout < 5000000; timeout++) { |
| 356 | if (callin_flag) |
| 357 | break; |
| 358 | udelay(100); |
| 359 | } |
David S. Miller | 72aff53 | 2006-02-17 01:29:17 -0800 | [diff] [blame^] | 360 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 361 | if (callin_flag) { |
| 362 | ret = 0; |
| 363 | } else { |
| 364 | printk("Processor %d is stuck.\n", cpu); |
| 365 | cpu_clear(cpu, cpu_callout_map); |
| 366 | ret = -ENODEV; |
| 367 | } |
| 368 | cpu_new_thread = NULL; |
| 369 | |
| 370 | return ret; |
| 371 | } |
| 372 | |
| 373 | static void spitfire_xcall_helper(u64 data0, u64 data1, u64 data2, u64 pstate, unsigned long cpu) |
| 374 | { |
| 375 | u64 result, target; |
| 376 | int stuck, tmp; |
| 377 | |
| 378 | if (this_is_starfire) { |
| 379 | /* map to real upaid */ |
| 380 | cpu = (((cpu & 0x3c) << 1) | |
| 381 | ((cpu & 0x40) >> 4) | |
| 382 | (cpu & 0x3)); |
| 383 | } |
| 384 | |
| 385 | target = (cpu << 14) | 0x70; |
| 386 | again: |
| 387 | /* Ok, this is the real Spitfire Errata #54. |
| 388 | * One must read back from a UDB internal register |
| 389 | * after writes to the UDB interrupt dispatch, but |
| 390 | * before the membar Sync for that write. |
| 391 | * So we use the high UDB control register (ASI 0x7f, |
| 392 | * ADDR 0x20) for the dummy read. -DaveM |
| 393 | */ |
| 394 | tmp = 0x40; |
| 395 | __asm__ __volatile__( |
| 396 | "wrpr %1, %2, %%pstate\n\t" |
| 397 | "stxa %4, [%0] %3\n\t" |
| 398 | "stxa %5, [%0+%8] %3\n\t" |
| 399 | "add %0, %8, %0\n\t" |
| 400 | "stxa %6, [%0+%8] %3\n\t" |
| 401 | "membar #Sync\n\t" |
| 402 | "stxa %%g0, [%7] %3\n\t" |
| 403 | "membar #Sync\n\t" |
| 404 | "mov 0x20, %%g1\n\t" |
| 405 | "ldxa [%%g1] 0x7f, %%g0\n\t" |
| 406 | "membar #Sync" |
| 407 | : "=r" (tmp) |
| 408 | : "r" (pstate), "i" (PSTATE_IE), "i" (ASI_INTR_W), |
| 409 | "r" (data0), "r" (data1), "r" (data2), "r" (target), |
| 410 | "r" (0x10), "0" (tmp) |
| 411 | : "g1"); |
| 412 | |
| 413 | /* NOTE: PSTATE_IE is still clear. */ |
| 414 | stuck = 100000; |
| 415 | do { |
| 416 | __asm__ __volatile__("ldxa [%%g0] %1, %0" |
| 417 | : "=r" (result) |
| 418 | : "i" (ASI_INTR_DISPATCH_STAT)); |
| 419 | if (result == 0) { |
| 420 | __asm__ __volatile__("wrpr %0, 0x0, %%pstate" |
| 421 | : : "r" (pstate)); |
| 422 | return; |
| 423 | } |
| 424 | stuck -= 1; |
| 425 | if (stuck == 0) |
| 426 | break; |
| 427 | } while (result & 0x1); |
| 428 | __asm__ __volatile__("wrpr %0, 0x0, %%pstate" |
| 429 | : : "r" (pstate)); |
| 430 | if (stuck == 0) { |
| 431 | printk("CPU[%d]: mondo stuckage result[%016lx]\n", |
| 432 | smp_processor_id(), result); |
| 433 | } else { |
| 434 | udelay(2); |
| 435 | goto again; |
| 436 | } |
| 437 | } |
| 438 | |
| 439 | static __inline__ void spitfire_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask) |
| 440 | { |
| 441 | u64 pstate; |
| 442 | int i; |
| 443 | |
| 444 | __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate)); |
| 445 | for_each_cpu_mask(i, mask) |
| 446 | spitfire_xcall_helper(data0, data1, data2, pstate, i); |
| 447 | } |
| 448 | |
| 449 | /* Cheetah now allows to send the whole 64-bytes of data in the interrupt |
| 450 | * packet, but we have no use for that. However we do take advantage of |
| 451 | * the new pipelining feature (ie. dispatch to multiple cpus simultaneously). |
| 452 | */ |
| 453 | static void cheetah_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask) |
| 454 | { |
| 455 | u64 pstate, ver; |
David S. Miller | 92704a1 | 2006-02-26 23:27:19 -0800 | [diff] [blame] | 456 | int nack_busy_id, is_jbus; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 457 | |
| 458 | if (cpus_empty(mask)) |
| 459 | return; |
| 460 | |
| 461 | /* Unfortunately, someone at Sun had the brilliant idea to make the |
| 462 | * busy/nack fields hard-coded by ITID number for this Ultra-III |
| 463 | * derivative processor. |
| 464 | */ |
| 465 | __asm__ ("rdpr %%ver, %0" : "=r" (ver)); |
David S. Miller | 92704a1 | 2006-02-26 23:27:19 -0800 | [diff] [blame] | 466 | is_jbus = ((ver >> 32) == __JALAPENO_ID || |
| 467 | (ver >> 32) == __SERRANO_ID); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 468 | |
| 469 | __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate)); |
| 470 | |
| 471 | retry: |
| 472 | __asm__ __volatile__("wrpr %0, %1, %%pstate\n\t" |
| 473 | : : "r" (pstate), "i" (PSTATE_IE)); |
| 474 | |
| 475 | /* Setup the dispatch data registers. */ |
| 476 | __asm__ __volatile__("stxa %0, [%3] %6\n\t" |
| 477 | "stxa %1, [%4] %6\n\t" |
| 478 | "stxa %2, [%5] %6\n\t" |
| 479 | "membar #Sync\n\t" |
| 480 | : /* no outputs */ |
| 481 | : "r" (data0), "r" (data1), "r" (data2), |
| 482 | "r" (0x40), "r" (0x50), "r" (0x60), |
| 483 | "i" (ASI_INTR_W)); |
| 484 | |
| 485 | nack_busy_id = 0; |
| 486 | { |
| 487 | int i; |
| 488 | |
| 489 | for_each_cpu_mask(i, mask) { |
| 490 | u64 target = (i << 14) | 0x70; |
| 491 | |
David S. Miller | 92704a1 | 2006-02-26 23:27:19 -0800 | [diff] [blame] | 492 | if (!is_jbus) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 493 | target |= (nack_busy_id << 24); |
| 494 | __asm__ __volatile__( |
| 495 | "stxa %%g0, [%0] %1\n\t" |
| 496 | "membar #Sync\n\t" |
| 497 | : /* no outputs */ |
| 498 | : "r" (target), "i" (ASI_INTR_W)); |
| 499 | nack_busy_id++; |
| 500 | } |
| 501 | } |
| 502 | |
| 503 | /* Now, poll for completion. */ |
| 504 | { |
| 505 | u64 dispatch_stat; |
| 506 | long stuck; |
| 507 | |
| 508 | stuck = 100000 * nack_busy_id; |
| 509 | do { |
| 510 | __asm__ __volatile__("ldxa [%%g0] %1, %0" |
| 511 | : "=r" (dispatch_stat) |
| 512 | : "i" (ASI_INTR_DISPATCH_STAT)); |
| 513 | if (dispatch_stat == 0UL) { |
| 514 | __asm__ __volatile__("wrpr %0, 0x0, %%pstate" |
| 515 | : : "r" (pstate)); |
| 516 | return; |
| 517 | } |
| 518 | if (!--stuck) |
| 519 | break; |
| 520 | } while (dispatch_stat & 0x5555555555555555UL); |
| 521 | |
| 522 | __asm__ __volatile__("wrpr %0, 0x0, %%pstate" |
| 523 | : : "r" (pstate)); |
| 524 | |
| 525 | if ((dispatch_stat & ~(0x5555555555555555UL)) == 0) { |
| 526 | /* Busy bits will not clear, continue instead |
| 527 | * of freezing up on this cpu. |
| 528 | */ |
| 529 | printk("CPU[%d]: mondo stuckage result[%016lx]\n", |
| 530 | smp_processor_id(), dispatch_stat); |
| 531 | } else { |
| 532 | int i, this_busy_nack = 0; |
| 533 | |
| 534 | /* Delay some random time with interrupts enabled |
| 535 | * to prevent deadlock. |
| 536 | */ |
| 537 | udelay(2 * nack_busy_id); |
| 538 | |
| 539 | /* Clear out the mask bits for cpus which did not |
| 540 | * NACK us. |
| 541 | */ |
| 542 | for_each_cpu_mask(i, mask) { |
| 543 | u64 check_mask; |
| 544 | |
David S. Miller | 92704a1 | 2006-02-26 23:27:19 -0800 | [diff] [blame] | 545 | if (is_jbus) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 546 | check_mask = (0x2UL << (2*i)); |
| 547 | else |
| 548 | check_mask = (0x2UL << |
| 549 | this_busy_nack); |
| 550 | if ((dispatch_stat & check_mask) == 0) |
| 551 | cpu_clear(i, mask); |
| 552 | this_busy_nack += 2; |
| 553 | } |
| 554 | |
| 555 | goto retry; |
| 556 | } |
| 557 | } |
| 558 | } |
| 559 | |
David S. Miller | 1d2f1f9 | 2006-02-08 16:41:20 -0800 | [diff] [blame] | 560 | #if 0 |
| 561 | /* Multi-cpu list version. */ |
| 562 | static int init_cpu_list(u16 *list, cpumask_t mask) |
| 563 | { |
| 564 | int i, cnt; |
| 565 | |
| 566 | cnt = 0; |
| 567 | for_each_cpu_mask(i, mask) |
| 568 | list[cnt++] = i; |
| 569 | |
| 570 | return cnt; |
| 571 | } |
| 572 | |
| 573 | static int update_cpu_list(u16 *list, int orig_cnt, cpumask_t mask) |
| 574 | { |
| 575 | int i; |
| 576 | |
| 577 | for (i = 0; i < orig_cnt; i++) { |
| 578 | if (list[i] == 0xffff) |
| 579 | cpu_clear(i, mask); |
| 580 | } |
| 581 | |
| 582 | return init_cpu_list(list, mask); |
| 583 | } |
| 584 | |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 585 | static void hypervisor_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask) |
| 586 | { |
David S. Miller | 1d2f1f9 | 2006-02-08 16:41:20 -0800 | [diff] [blame] | 587 | int this_cpu = get_cpu(); |
| 588 | struct trap_per_cpu *tb = &trap_block[this_cpu]; |
| 589 | u64 *mondo = __va(tb->cpu_mondo_block_pa); |
| 590 | u16 *cpu_list = __va(tb->cpu_list_pa); |
| 591 | int cnt, retries; |
| 592 | |
| 593 | mondo[0] = data0; |
| 594 | mondo[1] = data1; |
| 595 | mondo[2] = data2; |
| 596 | wmb(); |
| 597 | |
| 598 | retries = 0; |
| 599 | cnt = init_cpu_list(cpu_list, mask); |
| 600 | do { |
David S. Miller | 164c220 | 2006-02-09 22:57:21 -0800 | [diff] [blame] | 601 | register unsigned long func __asm__("%o5"); |
| 602 | register unsigned long arg0 __asm__("%o0"); |
| 603 | register unsigned long arg1 __asm__("%o1"); |
| 604 | register unsigned long arg2 __asm__("%o2"); |
David S. Miller | 1d2f1f9 | 2006-02-08 16:41:20 -0800 | [diff] [blame] | 605 | |
| 606 | func = HV_FAST_CPU_MONDO_SEND; |
| 607 | arg0 = cnt; |
| 608 | arg1 = tb->cpu_list_pa; |
| 609 | arg2 = tb->cpu_mondo_block_pa; |
| 610 | |
| 611 | __asm__ __volatile__("ta %8" |
| 612 | : "=&r" (func), "=&r" (arg0), |
| 613 | "=&r" (arg1), "=&r" (arg2) |
| 614 | : "0" (func), "1" (arg0), |
| 615 | "2" (arg1), "3" (arg2), |
| 616 | "i" (HV_FAST_TRAP) |
| 617 | : "memory"); |
David S. Miller | b5a37e9 | 2006-02-11 23:07:13 -0800 | [diff] [blame] | 618 | if (likely(arg0 == HV_EOK)) |
David S. Miller | 1d2f1f9 | 2006-02-08 16:41:20 -0800 | [diff] [blame] | 619 | break; |
| 620 | |
| 621 | if (unlikely(++retries > 100)) { |
| 622 | printk("CPU[%d]: sun4v mondo error %lu\n", |
| 623 | this_cpu, func); |
| 624 | break; |
| 625 | } |
| 626 | |
| 627 | cnt = update_cpu_list(cpu_list, cnt, mask); |
| 628 | |
| 629 | udelay(2 * cnt); |
| 630 | } while (1); |
| 631 | |
| 632 | put_cpu(); |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 633 | } |
David S. Miller | 1d2f1f9 | 2006-02-08 16:41:20 -0800 | [diff] [blame] | 634 | #else |
| 635 | /* Single-cpu list version. */ |
| 636 | static void hypervisor_xcall_deliver(u64 data0, u64 data1, u64 data2, cpumask_t mask) |
| 637 | { |
| 638 | int this_cpu = get_cpu(); |
| 639 | struct trap_per_cpu *tb = &trap_block[this_cpu]; |
| 640 | u64 *mondo = __va(tb->cpu_mondo_block_pa); |
| 641 | u16 *cpu_list = __va(tb->cpu_list_pa); |
| 642 | int i; |
| 643 | |
| 644 | mondo[0] = data0; |
| 645 | mondo[1] = data1; |
| 646 | mondo[2] = data2; |
| 647 | wmb(); |
| 648 | |
| 649 | for_each_cpu_mask(i, mask) { |
| 650 | int retries = 0; |
| 651 | |
| 652 | do { |
David S. Miller | 164c220 | 2006-02-09 22:57:21 -0800 | [diff] [blame] | 653 | register unsigned long func __asm__("%o5"); |
| 654 | register unsigned long arg0 __asm__("%o0"); |
| 655 | register unsigned long arg1 __asm__("%o1"); |
| 656 | register unsigned long arg2 __asm__("%o2"); |
David S. Miller | 1d2f1f9 | 2006-02-08 16:41:20 -0800 | [diff] [blame] | 657 | |
| 658 | cpu_list[0] = i; |
| 659 | func = HV_FAST_CPU_MONDO_SEND; |
| 660 | arg0 = 1; |
| 661 | arg1 = tb->cpu_list_pa; |
| 662 | arg2 = tb->cpu_mondo_block_pa; |
| 663 | |
| 664 | __asm__ __volatile__("ta %8" |
| 665 | : "=&r" (func), "=&r" (arg0), |
| 666 | "=&r" (arg1), "=&r" (arg2) |
| 667 | : "0" (func), "1" (arg0), |
| 668 | "2" (arg1), "3" (arg2), |
| 669 | "i" (HV_FAST_TRAP) |
| 670 | : "memory"); |
David S. Miller | b5a37e9 | 2006-02-11 23:07:13 -0800 | [diff] [blame] | 671 | if (likely(arg0 == HV_EOK)) |
David S. Miller | 1d2f1f9 | 2006-02-08 16:41:20 -0800 | [diff] [blame] | 672 | break; |
| 673 | |
| 674 | if (unlikely(++retries > 100)) { |
| 675 | printk("CPU[%d]: sun4v mondo error %lu\n", |
| 676 | this_cpu, func); |
| 677 | break; |
| 678 | } |
| 679 | |
| 680 | udelay(2 * i); |
| 681 | } while (1); |
| 682 | } |
| 683 | |
| 684 | put_cpu(); |
| 685 | } |
| 686 | #endif |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 687 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | /* Send cross call to all processors mentioned in MASK |
| 689 | * except self. |
| 690 | */ |
| 691 | static void smp_cross_call_masked(unsigned long *func, u32 ctx, u64 data1, u64 data2, cpumask_t mask) |
| 692 | { |
| 693 | u64 data0 = (((u64)ctx)<<32 | (((u64)func) & 0xffffffff)); |
| 694 | int this_cpu = get_cpu(); |
| 695 | |
| 696 | cpus_and(mask, mask, cpu_online_map); |
| 697 | cpu_clear(this_cpu, mask); |
| 698 | |
| 699 | if (tlb_type == spitfire) |
| 700 | spitfire_xcall_deliver(data0, data1, data2, mask); |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 701 | else if (tlb_type == cheetah || tlb_type == cheetah_plus) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 702 | cheetah_xcall_deliver(data0, data1, data2, mask); |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 703 | else |
| 704 | hypervisor_xcall_deliver(data0, data1, data2, mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 705 | /* NOTE: Caller runs local copy on master. */ |
| 706 | |
| 707 | put_cpu(); |
| 708 | } |
| 709 | |
| 710 | extern unsigned long xcall_sync_tick; |
| 711 | |
| 712 | static void smp_start_sync_tick_client(int cpu) |
| 713 | { |
| 714 | cpumask_t mask = cpumask_of_cpu(cpu); |
| 715 | |
| 716 | smp_cross_call_masked(&xcall_sync_tick, |
| 717 | 0, 0, 0, mask); |
| 718 | } |
| 719 | |
| 720 | /* Send cross call to all processors except self. */ |
| 721 | #define smp_cross_call(func, ctx, data1, data2) \ |
| 722 | smp_cross_call_masked(func, ctx, data1, data2, cpu_online_map) |
| 723 | |
| 724 | struct call_data_struct { |
| 725 | void (*func) (void *info); |
| 726 | void *info; |
| 727 | atomic_t finished; |
| 728 | int wait; |
| 729 | }; |
| 730 | |
| 731 | static DEFINE_SPINLOCK(call_lock); |
| 732 | static struct call_data_struct *call_data; |
| 733 | |
| 734 | extern unsigned long xcall_call_function; |
| 735 | |
| 736 | /* |
| 737 | * You must not call this function with disabled interrupts or from a |
| 738 | * hardware interrupt handler or from a bottom half handler. |
| 739 | */ |
David S. Miller | bd40791 | 2006-01-31 18:31:38 -0800 | [diff] [blame] | 740 | static int smp_call_function_mask(void (*func)(void *info), void *info, |
| 741 | int nonatomic, int wait, cpumask_t mask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 742 | { |
| 743 | struct call_data_struct data; |
David S. Miller | bd40791 | 2006-01-31 18:31:38 -0800 | [diff] [blame] | 744 | int cpus = cpus_weight(mask) - 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 745 | long timeout; |
| 746 | |
| 747 | if (!cpus) |
| 748 | return 0; |
| 749 | |
| 750 | /* Can deadlock when called with interrupts disabled */ |
| 751 | WARN_ON(irqs_disabled()); |
| 752 | |
| 753 | data.func = func; |
| 754 | data.info = info; |
| 755 | atomic_set(&data.finished, 0); |
| 756 | data.wait = wait; |
| 757 | |
| 758 | spin_lock(&call_lock); |
| 759 | |
| 760 | call_data = &data; |
| 761 | |
David S. Miller | bd40791 | 2006-01-31 18:31:38 -0800 | [diff] [blame] | 762 | smp_cross_call_masked(&xcall_call_function, 0, 0, 0, mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 763 | |
| 764 | /* |
| 765 | * Wait for other cpus to complete function or at |
| 766 | * least snap the call data. |
| 767 | */ |
| 768 | timeout = 1000000; |
| 769 | while (atomic_read(&data.finished) != cpus) { |
| 770 | if (--timeout <= 0) |
| 771 | goto out_timeout; |
| 772 | barrier(); |
| 773 | udelay(1); |
| 774 | } |
| 775 | |
| 776 | spin_unlock(&call_lock); |
| 777 | |
| 778 | return 0; |
| 779 | |
| 780 | out_timeout: |
| 781 | spin_unlock(&call_lock); |
| 782 | printk("XCALL: Remote cpus not responding, ncpus=%ld finished=%ld\n", |
| 783 | (long) num_online_cpus() - 1L, |
| 784 | (long) atomic_read(&data.finished)); |
| 785 | return 0; |
| 786 | } |
| 787 | |
David S. Miller | bd40791 | 2006-01-31 18:31:38 -0800 | [diff] [blame] | 788 | int smp_call_function(void (*func)(void *info), void *info, |
| 789 | int nonatomic, int wait) |
| 790 | { |
| 791 | return smp_call_function_mask(func, info, nonatomic, wait, |
| 792 | cpu_online_map); |
| 793 | } |
| 794 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 795 | void smp_call_function_client(int irq, struct pt_regs *regs) |
| 796 | { |
| 797 | void (*func) (void *info) = call_data->func; |
| 798 | void *info = call_data->info; |
| 799 | |
| 800 | clear_softint(1 << irq); |
| 801 | if (call_data->wait) { |
| 802 | /* let initiator proceed only after completion */ |
| 803 | func(info); |
| 804 | atomic_inc(&call_data->finished); |
| 805 | } else { |
| 806 | /* let initiator proceed after getting data */ |
| 807 | atomic_inc(&call_data->finished); |
| 808 | func(info); |
| 809 | } |
| 810 | } |
| 811 | |
David S. Miller | bd40791 | 2006-01-31 18:31:38 -0800 | [diff] [blame] | 812 | static void tsb_sync(void *info) |
| 813 | { |
| 814 | struct mm_struct *mm = info; |
| 815 | |
| 816 | if (current->active_mm == mm) |
| 817 | tsb_context_switch(mm); |
| 818 | } |
| 819 | |
| 820 | void smp_tsb_sync(struct mm_struct *mm) |
| 821 | { |
| 822 | smp_call_function_mask(tsb_sync, mm, 0, 1, mm->cpu_vm_mask); |
| 823 | } |
| 824 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 825 | extern unsigned long xcall_flush_tlb_mm; |
| 826 | extern unsigned long xcall_flush_tlb_pending; |
| 827 | extern unsigned long xcall_flush_tlb_kernel_range; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | extern unsigned long xcall_report_regs; |
| 829 | extern unsigned long xcall_receive_signal; |
| 830 | |
| 831 | #ifdef DCACHE_ALIASING_POSSIBLE |
| 832 | extern unsigned long xcall_flush_dcache_page_cheetah; |
| 833 | #endif |
| 834 | extern unsigned long xcall_flush_dcache_page_spitfire; |
| 835 | |
| 836 | #ifdef CONFIG_DEBUG_DCFLUSH |
| 837 | extern atomic_t dcpage_flushes; |
| 838 | extern atomic_t dcpage_flushes_xcall; |
| 839 | #endif |
| 840 | |
| 841 | static __inline__ void __local_flush_dcache_page(struct page *page) |
| 842 | { |
| 843 | #ifdef DCACHE_ALIASING_POSSIBLE |
| 844 | __flush_dcache_page(page_address(page), |
| 845 | ((tlb_type == spitfire) && |
| 846 | page_mapping(page) != NULL)); |
| 847 | #else |
| 848 | if (page_mapping(page) != NULL && |
| 849 | tlb_type == spitfire) |
| 850 | __flush_icache_page(__pa(page_address(page))); |
| 851 | #endif |
| 852 | } |
| 853 | |
| 854 | void smp_flush_dcache_page_impl(struct page *page, int cpu) |
| 855 | { |
| 856 | cpumask_t mask = cpumask_of_cpu(cpu); |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 857 | int this_cpu; |
| 858 | |
| 859 | if (tlb_type == hypervisor) |
| 860 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 861 | |
| 862 | #ifdef CONFIG_DEBUG_DCFLUSH |
| 863 | atomic_inc(&dcpage_flushes); |
| 864 | #endif |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 865 | |
| 866 | this_cpu = get_cpu(); |
| 867 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 868 | if (cpu == this_cpu) { |
| 869 | __local_flush_dcache_page(page); |
| 870 | } else if (cpu_online(cpu)) { |
| 871 | void *pg_addr = page_address(page); |
| 872 | u64 data0; |
| 873 | |
| 874 | if (tlb_type == spitfire) { |
| 875 | data0 = |
| 876 | ((u64)&xcall_flush_dcache_page_spitfire); |
| 877 | if (page_mapping(page) != NULL) |
| 878 | data0 |= ((u64)1 << 32); |
| 879 | spitfire_xcall_deliver(data0, |
| 880 | __pa(pg_addr), |
| 881 | (u64) pg_addr, |
| 882 | mask); |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 883 | } else if (tlb_type == cheetah || tlb_type == cheetah_plus) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 884 | #ifdef DCACHE_ALIASING_POSSIBLE |
| 885 | data0 = |
| 886 | ((u64)&xcall_flush_dcache_page_cheetah); |
| 887 | cheetah_xcall_deliver(data0, |
| 888 | __pa(pg_addr), |
| 889 | 0, mask); |
| 890 | #endif |
| 891 | } |
| 892 | #ifdef CONFIG_DEBUG_DCFLUSH |
| 893 | atomic_inc(&dcpage_flushes_xcall); |
| 894 | #endif |
| 895 | } |
| 896 | |
| 897 | put_cpu(); |
| 898 | } |
| 899 | |
| 900 | void flush_dcache_page_all(struct mm_struct *mm, struct page *page) |
| 901 | { |
| 902 | void *pg_addr = page_address(page); |
| 903 | cpumask_t mask = cpu_online_map; |
| 904 | u64 data0; |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 905 | int this_cpu; |
| 906 | |
| 907 | if (tlb_type == hypervisor) |
| 908 | return; |
| 909 | |
| 910 | this_cpu = get_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 911 | |
| 912 | cpu_clear(this_cpu, mask); |
| 913 | |
| 914 | #ifdef CONFIG_DEBUG_DCFLUSH |
| 915 | atomic_inc(&dcpage_flushes); |
| 916 | #endif |
| 917 | if (cpus_empty(mask)) |
| 918 | goto flush_self; |
| 919 | if (tlb_type == spitfire) { |
| 920 | data0 = ((u64)&xcall_flush_dcache_page_spitfire); |
| 921 | if (page_mapping(page) != NULL) |
| 922 | data0 |= ((u64)1 << 32); |
| 923 | spitfire_xcall_deliver(data0, |
| 924 | __pa(pg_addr), |
| 925 | (u64) pg_addr, |
| 926 | mask); |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 927 | } else if (tlb_type == cheetah || tlb_type == cheetah_plus) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 928 | #ifdef DCACHE_ALIASING_POSSIBLE |
| 929 | data0 = ((u64)&xcall_flush_dcache_page_cheetah); |
| 930 | cheetah_xcall_deliver(data0, |
| 931 | __pa(pg_addr), |
| 932 | 0, mask); |
| 933 | #endif |
| 934 | } |
| 935 | #ifdef CONFIG_DEBUG_DCFLUSH |
| 936 | atomic_inc(&dcpage_flushes_xcall); |
| 937 | #endif |
| 938 | flush_self: |
| 939 | __local_flush_dcache_page(page); |
| 940 | |
| 941 | put_cpu(); |
| 942 | } |
| 943 | |
| 944 | void smp_receive_signal(int cpu) |
| 945 | { |
| 946 | cpumask_t mask = cpumask_of_cpu(cpu); |
| 947 | |
| 948 | if (cpu_online(cpu)) { |
| 949 | u64 data0 = (((u64)&xcall_receive_signal) & 0xffffffff); |
| 950 | |
| 951 | if (tlb_type == spitfire) |
| 952 | spitfire_xcall_deliver(data0, 0, 0, mask); |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 953 | else if (tlb_type == cheetah || tlb_type == cheetah_plus) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 954 | cheetah_xcall_deliver(data0, 0, 0, mask); |
David S. Miller | a43fe0e | 2006-02-04 03:10:53 -0800 | [diff] [blame] | 955 | else if (tlb_type == hypervisor) |
| 956 | hypervisor_xcall_deliver(data0, 0, 0, mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 957 | } |
| 958 | } |
| 959 | |
| 960 | void smp_receive_signal_client(int irq, struct pt_regs *regs) |
| 961 | { |
| 962 | /* Just return, rtrap takes care of the rest. */ |
| 963 | clear_softint(1 << irq); |
| 964 | } |
| 965 | |
| 966 | void smp_report_regs(void) |
| 967 | { |
| 968 | smp_cross_call(&xcall_report_regs, 0, 0, 0); |
| 969 | } |
| 970 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 971 | /* We know that the window frames of the user have been flushed |
| 972 | * to the stack before we get here because all callers of us |
| 973 | * are flush_tlb_*() routines, and these run after flush_cache_*() |
| 974 | * which performs the flushw. |
| 975 | * |
| 976 | * The SMP TLB coherency scheme we use works as follows: |
| 977 | * |
| 978 | * 1) mm->cpu_vm_mask is a bit mask of which cpus an address |
| 979 | * space has (potentially) executed on, this is the heuristic |
| 980 | * we use to avoid doing cross calls. |
| 981 | * |
| 982 | * Also, for flushing from kswapd and also for clones, we |
| 983 | * use cpu_vm_mask as the list of cpus to make run the TLB. |
| 984 | * |
| 985 | * 2) TLB context numbers are shared globally across all processors |
| 986 | * in the system, this allows us to play several games to avoid |
| 987 | * cross calls. |
| 988 | * |
| 989 | * One invariant is that when a cpu switches to a process, and |
| 990 | * that processes tsk->active_mm->cpu_vm_mask does not have the |
| 991 | * current cpu's bit set, that tlb context is flushed locally. |
| 992 | * |
| 993 | * If the address space is non-shared (ie. mm->count == 1) we avoid |
| 994 | * cross calls when we want to flush the currently running process's |
| 995 | * tlb state. This is done by clearing all cpu bits except the current |
| 996 | * processor's in current->active_mm->cpu_vm_mask and performing the |
| 997 | * flush locally only. This will force any subsequent cpus which run |
| 998 | * this task to flush the context from the local tlb if the process |
| 999 | * migrates to another cpu (again). |
| 1000 | * |
| 1001 | * 3) For shared address spaces (threads) and swapping we bite the |
| 1002 | * bullet for most cases and perform the cross call (but only to |
| 1003 | * the cpus listed in cpu_vm_mask). |
| 1004 | * |
| 1005 | * The performance gain from "optimizing" away the cross call for threads is |
| 1006 | * questionable (in theory the big win for threads is the massive sharing of |
| 1007 | * address space state across processors). |
| 1008 | */ |
David S. Miller | 62dbec7 | 2005-11-07 14:09:58 -0800 | [diff] [blame] | 1009 | |
| 1010 | /* This currently is only used by the hugetlb arch pre-fault |
| 1011 | * hook on UltraSPARC-III+ and later when changing the pagesize |
| 1012 | * bits of the context register for an address space. |
| 1013 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1014 | void smp_flush_tlb_mm(struct mm_struct *mm) |
| 1015 | { |
David S. Miller | 62dbec7 | 2005-11-07 14:09:58 -0800 | [diff] [blame] | 1016 | u32 ctx = CTX_HWBITS(mm->context); |
| 1017 | int cpu = get_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1018 | |
David S. Miller | 62dbec7 | 2005-11-07 14:09:58 -0800 | [diff] [blame] | 1019 | if (atomic_read(&mm->mm_users) == 1) { |
| 1020 | mm->cpu_vm_mask = cpumask_of_cpu(cpu); |
| 1021 | goto local_flush_and_out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | } |
David S. Miller | 62dbec7 | 2005-11-07 14:09:58 -0800 | [diff] [blame] | 1023 | |
| 1024 | smp_cross_call_masked(&xcall_flush_tlb_mm, |
| 1025 | ctx, 0, 0, |
| 1026 | mm->cpu_vm_mask); |
| 1027 | |
| 1028 | local_flush_and_out: |
| 1029 | __flush_tlb_mm(ctx, SECONDARY_CONTEXT); |
| 1030 | |
| 1031 | put_cpu(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1032 | } |
| 1033 | |
| 1034 | void smp_flush_tlb_pending(struct mm_struct *mm, unsigned long nr, unsigned long *vaddrs) |
| 1035 | { |
| 1036 | u32 ctx = CTX_HWBITS(mm->context); |
| 1037 | int cpu = get_cpu(); |
| 1038 | |
Hugh Dickins | dedeb00 | 2005-11-07 14:09:01 -0800 | [diff] [blame] | 1039 | if (mm == current->active_mm && atomic_read(&mm->mm_users) == 1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1040 | mm->cpu_vm_mask = cpumask_of_cpu(cpu); |
Hugh Dickins | dedeb00 | 2005-11-07 14:09:01 -0800 | [diff] [blame] | 1041 | else |
| 1042 | smp_cross_call_masked(&xcall_flush_tlb_pending, |
| 1043 | ctx, nr, (unsigned long) vaddrs, |
| 1044 | mm->cpu_vm_mask); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1045 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1046 | __flush_tlb_pending(ctx, nr, vaddrs); |
| 1047 | |
| 1048 | put_cpu(); |
| 1049 | } |
| 1050 | |
| 1051 | void smp_flush_tlb_kernel_range(unsigned long start, unsigned long end) |
| 1052 | { |
| 1053 | start &= PAGE_MASK; |
| 1054 | end = PAGE_ALIGN(end); |
| 1055 | if (start != end) { |
| 1056 | smp_cross_call(&xcall_flush_tlb_kernel_range, |
| 1057 | 0, start, end); |
| 1058 | |
| 1059 | __flush_tlb_kernel_range(start, end); |
| 1060 | } |
| 1061 | } |
| 1062 | |
| 1063 | /* CPU capture. */ |
| 1064 | /* #define CAPTURE_DEBUG */ |
| 1065 | extern unsigned long xcall_capture; |
| 1066 | |
| 1067 | static atomic_t smp_capture_depth = ATOMIC_INIT(0); |
| 1068 | static atomic_t smp_capture_registry = ATOMIC_INIT(0); |
| 1069 | static unsigned long penguins_are_doing_time; |
| 1070 | |
| 1071 | void smp_capture(void) |
| 1072 | { |
| 1073 | int result = atomic_add_ret(1, &smp_capture_depth); |
| 1074 | |
| 1075 | if (result == 1) { |
| 1076 | int ncpus = num_online_cpus(); |
| 1077 | |
| 1078 | #ifdef CAPTURE_DEBUG |
| 1079 | printk("CPU[%d]: Sending penguins to jail...", |
| 1080 | smp_processor_id()); |
| 1081 | #endif |
| 1082 | penguins_are_doing_time = 1; |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 1083 | membar_storestore_loadstore(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1084 | atomic_inc(&smp_capture_registry); |
| 1085 | smp_cross_call(&xcall_capture, 0, 0, 0); |
| 1086 | while (atomic_read(&smp_capture_registry) != ncpus) |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 1087 | rmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1088 | #ifdef CAPTURE_DEBUG |
| 1089 | printk("done\n"); |
| 1090 | #endif |
| 1091 | } |
| 1092 | } |
| 1093 | |
| 1094 | void smp_release(void) |
| 1095 | { |
| 1096 | if (atomic_dec_and_test(&smp_capture_depth)) { |
| 1097 | #ifdef CAPTURE_DEBUG |
| 1098 | printk("CPU[%d]: Giving pardon to " |
| 1099 | "imprisoned penguins\n", |
| 1100 | smp_processor_id()); |
| 1101 | #endif |
| 1102 | penguins_are_doing_time = 0; |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 1103 | membar_storeload_storestore(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1104 | atomic_dec(&smp_capture_registry); |
| 1105 | } |
| 1106 | } |
| 1107 | |
| 1108 | /* Imprisoned penguins run with %pil == 15, but PSTATE_IE set, so they |
| 1109 | * can service tlb flush xcalls... |
| 1110 | */ |
| 1111 | extern void prom_world(int); |
David S. Miller | 96c6e0d | 2006-01-31 18:32:29 -0800 | [diff] [blame] | 1112 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1113 | void smp_penguin_jailcell(int irq, struct pt_regs *regs) |
| 1114 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1115 | clear_softint(1 << irq); |
| 1116 | |
| 1117 | preempt_disable(); |
| 1118 | |
| 1119 | __asm__ __volatile__("flushw"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1120 | prom_world(1); |
| 1121 | atomic_inc(&smp_capture_registry); |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 1122 | membar_storeload_storestore(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1123 | while (penguins_are_doing_time) |
David S. Miller | 4f07118 | 2005-08-29 12:46:22 -0700 | [diff] [blame] | 1124 | rmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1125 | atomic_dec(&smp_capture_registry); |
| 1126 | prom_world(0); |
| 1127 | |
| 1128 | preempt_enable(); |
| 1129 | } |
| 1130 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1131 | #define prof_multiplier(__cpu) cpu_data(__cpu).multiplier |
| 1132 | #define prof_counter(__cpu) cpu_data(__cpu).counter |
| 1133 | |
| 1134 | void smp_percpu_timer_interrupt(struct pt_regs *regs) |
| 1135 | { |
| 1136 | unsigned long compare, tick, pstate; |
| 1137 | int cpu = smp_processor_id(); |
| 1138 | int user = user_mode(regs); |
| 1139 | |
| 1140 | /* |
| 1141 | * Check for level 14 softint. |
| 1142 | */ |
| 1143 | { |
| 1144 | unsigned long tick_mask = tick_ops->softint_mask; |
| 1145 | |
| 1146 | if (!(get_softint() & tick_mask)) { |
| 1147 | extern void handler_irq(int, struct pt_regs *); |
| 1148 | |
| 1149 | handler_irq(14, regs); |
| 1150 | return; |
| 1151 | } |
| 1152 | clear_softint(tick_mask); |
| 1153 | } |
| 1154 | |
| 1155 | do { |
| 1156 | profile_tick(CPU_PROFILING, regs); |
| 1157 | if (!--prof_counter(cpu)) { |
| 1158 | irq_enter(); |
| 1159 | |
| 1160 | if (cpu == boot_cpu_id) { |
| 1161 | kstat_this_cpu.irqs[0]++; |
| 1162 | timer_tick_interrupt(regs); |
| 1163 | } |
| 1164 | |
| 1165 | update_process_times(user); |
| 1166 | |
| 1167 | irq_exit(); |
| 1168 | |
| 1169 | prof_counter(cpu) = prof_multiplier(cpu); |
| 1170 | } |
| 1171 | |
| 1172 | /* Guarantee that the following sequences execute |
| 1173 | * uninterrupted. |
| 1174 | */ |
| 1175 | __asm__ __volatile__("rdpr %%pstate, %0\n\t" |
| 1176 | "wrpr %0, %1, %%pstate" |
| 1177 | : "=r" (pstate) |
| 1178 | : "i" (PSTATE_IE)); |
| 1179 | |
| 1180 | compare = tick_ops->add_compare(current_tick_offset); |
| 1181 | tick = tick_ops->get_tick(); |
| 1182 | |
| 1183 | /* Restore PSTATE_IE. */ |
| 1184 | __asm__ __volatile__("wrpr %0, 0x0, %%pstate" |
| 1185 | : /* no outputs */ |
| 1186 | : "r" (pstate)); |
| 1187 | } while (time_after_eq(tick, compare)); |
| 1188 | } |
| 1189 | |
| 1190 | static void __init smp_setup_percpu_timer(void) |
| 1191 | { |
| 1192 | int cpu = smp_processor_id(); |
| 1193 | unsigned long pstate; |
| 1194 | |
| 1195 | prof_counter(cpu) = prof_multiplier(cpu) = 1; |
| 1196 | |
| 1197 | /* Guarantee that the following sequences execute |
| 1198 | * uninterrupted. |
| 1199 | */ |
| 1200 | __asm__ __volatile__("rdpr %%pstate, %0\n\t" |
| 1201 | "wrpr %0, %1, %%pstate" |
| 1202 | : "=r" (pstate) |
| 1203 | : "i" (PSTATE_IE)); |
| 1204 | |
| 1205 | tick_ops->init_tick(current_tick_offset); |
| 1206 | |
| 1207 | /* Restore PSTATE_IE. */ |
| 1208 | __asm__ __volatile__("wrpr %0, 0x0, %%pstate" |
| 1209 | : /* no outputs */ |
| 1210 | : "r" (pstate)); |
| 1211 | } |
| 1212 | |
| 1213 | void __init smp_tick_init(void) |
| 1214 | { |
| 1215 | boot_cpu_id = hard_smp_processor_id(); |
| 1216 | current_tick_offset = timer_tick_offset; |
| 1217 | |
| 1218 | cpu_set(boot_cpu_id, cpu_online_map); |
| 1219 | prof_counter(boot_cpu_id) = prof_multiplier(boot_cpu_id) = 1; |
| 1220 | } |
| 1221 | |
| 1222 | /* /proc/profile writes can call this, don't __init it please. */ |
| 1223 | static DEFINE_SPINLOCK(prof_setup_lock); |
| 1224 | |
| 1225 | int setup_profiling_timer(unsigned int multiplier) |
| 1226 | { |
| 1227 | unsigned long flags; |
| 1228 | int i; |
| 1229 | |
| 1230 | if ((!multiplier) || (timer_tick_offset / multiplier) < 1000) |
| 1231 | return -EINVAL; |
| 1232 | |
| 1233 | spin_lock_irqsave(&prof_setup_lock, flags); |
| 1234 | for (i = 0; i < NR_CPUS; i++) |
| 1235 | prof_multiplier(i) = multiplier; |
| 1236 | current_tick_offset = (timer_tick_offset / multiplier); |
| 1237 | spin_unlock_irqrestore(&prof_setup_lock, flags); |
| 1238 | |
| 1239 | return 0; |
| 1240 | } |
| 1241 | |
David S. Miller | 7abea92 | 2006-02-25 13:39:56 -0800 | [diff] [blame] | 1242 | /* Constrain the number of cpus to max_cpus. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1243 | void __init smp_prepare_cpus(unsigned int max_cpus) |
| 1244 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1245 | if (num_possible_cpus() > max_cpus) { |
David S. Miller | 7abea92 | 2006-02-25 13:39:56 -0800 | [diff] [blame] | 1246 | int instance, mid; |
| 1247 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1248 | instance = 0; |
| 1249 | while (!cpu_find_by_instance(instance, NULL, &mid)) { |
| 1250 | if (mid != boot_cpu_id) { |
| 1251 | cpu_clear(mid, phys_cpu_present_map); |
| 1252 | if (num_possible_cpus() <= max_cpus) |
| 1253 | break; |
| 1254 | } |
| 1255 | instance++; |
| 1256 | } |
| 1257 | } |
| 1258 | |
| 1259 | smp_store_cpu_info(boot_cpu_id); |
| 1260 | } |
| 1261 | |
David S. Miller | 7abea92 | 2006-02-25 13:39:56 -0800 | [diff] [blame] | 1262 | /* Set this up early so that things like the scheduler can init |
| 1263 | * properly. We use the same cpu mask for both the present and |
| 1264 | * possible cpu map. |
| 1265 | */ |
| 1266 | void __init smp_setup_cpu_possible_map(void) |
| 1267 | { |
| 1268 | int instance, mid; |
| 1269 | |
| 1270 | instance = 0; |
| 1271 | while (!cpu_find_by_instance(instance, NULL, &mid)) { |
| 1272 | if (mid < NR_CPUS) |
| 1273 | cpu_set(mid, phys_cpu_present_map); |
| 1274 | instance++; |
| 1275 | } |
| 1276 | } |
| 1277 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1278 | void __devinit smp_prepare_boot_cpu(void) |
| 1279 | { |
David S. Miller | 56fb4df | 2006-02-26 23:24:22 -0800 | [diff] [blame] | 1280 | int cpu = hard_smp_processor_id(); |
| 1281 | |
| 1282 | if (cpu >= NR_CPUS) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1283 | prom_printf("Serious problem, boot cpu id >= NR_CPUS\n"); |
| 1284 | prom_halt(); |
| 1285 | } |
| 1286 | |
David S. Miller | 56fb4df | 2006-02-26 23:24:22 -0800 | [diff] [blame] | 1287 | current_thread_info()->cpu = cpu; |
| 1288 | __local_per_cpu_offset = __per_cpu_offset(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1289 | |
| 1290 | cpu_set(smp_processor_id(), cpu_online_map); |
| 1291 | cpu_set(smp_processor_id(), phys_cpu_present_map); |
| 1292 | } |
| 1293 | |
| 1294 | int __devinit __cpu_up(unsigned int cpu) |
| 1295 | { |
| 1296 | int ret = smp_boot_one_cpu(cpu); |
| 1297 | |
| 1298 | if (!ret) { |
| 1299 | cpu_set(cpu, smp_commenced_mask); |
| 1300 | while (!cpu_isset(cpu, cpu_online_map)) |
| 1301 | mb(); |
| 1302 | if (!cpu_isset(cpu, cpu_online_map)) { |
| 1303 | ret = -ENODEV; |
| 1304 | } else { |
David S. Miller | 02fead7 | 2006-02-11 23:22:47 -0800 | [diff] [blame] | 1305 | /* On SUN4V, writes to %tick and %stick are |
| 1306 | * not allowed. |
| 1307 | */ |
| 1308 | if (tlb_type != hypervisor) |
| 1309 | smp_synchronize_one_tick(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1310 | } |
| 1311 | } |
| 1312 | return ret; |
| 1313 | } |
| 1314 | |
| 1315 | void __init smp_cpus_done(unsigned int max_cpus) |
| 1316 | { |
| 1317 | unsigned long bogosum = 0; |
| 1318 | int i; |
| 1319 | |
| 1320 | for (i = 0; i < NR_CPUS; i++) { |
| 1321 | if (cpu_online(i)) |
| 1322 | bogosum += cpu_data(i).udelay_val; |
| 1323 | } |
| 1324 | printk("Total of %ld processors activated " |
| 1325 | "(%lu.%02lu BogoMIPS).\n", |
| 1326 | (long) num_online_cpus(), |
| 1327 | bogosum/(500000/HZ), |
| 1328 | (bogosum/(5000/HZ))%100); |
| 1329 | } |
| 1330 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1331 | void smp_send_reschedule(int cpu) |
| 1332 | { |
Nick Piggin | 64c7c8f | 2005-11-08 21:39:04 -0800 | [diff] [blame] | 1333 | smp_receive_signal(cpu); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1334 | } |
| 1335 | |
| 1336 | /* This is a nop because we capture all other cpus |
| 1337 | * anyways when making the PROM active. |
| 1338 | */ |
| 1339 | void smp_send_stop(void) |
| 1340 | { |
| 1341 | } |
| 1342 | |
David S. Miller | d369ddd | 2005-07-10 15:45:11 -0700 | [diff] [blame] | 1343 | unsigned long __per_cpu_base __read_mostly; |
| 1344 | unsigned long __per_cpu_shift __read_mostly; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1345 | |
| 1346 | EXPORT_SYMBOL(__per_cpu_base); |
| 1347 | EXPORT_SYMBOL(__per_cpu_shift); |
| 1348 | |
| 1349 | void __init setup_per_cpu_areas(void) |
| 1350 | { |
| 1351 | unsigned long goal, size, i; |
| 1352 | char *ptr; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1353 | |
| 1354 | /* Copy section for each CPU (we discard the original) */ |
David S. Miller | 56fb4df | 2006-02-26 23:24:22 -0800 | [diff] [blame] | 1355 | goal = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1356 | #ifdef CONFIG_MODULES |
| 1357 | if (goal < PERCPU_ENOUGH_ROOM) |
| 1358 | goal = PERCPU_ENOUGH_ROOM; |
| 1359 | #endif |
| 1360 | __per_cpu_shift = 0; |
| 1361 | for (size = 1UL; size < goal; size <<= 1UL) |
| 1362 | __per_cpu_shift++; |
| 1363 | |
David S. Miller | 56fb4df | 2006-02-26 23:24:22 -0800 | [diff] [blame] | 1364 | ptr = alloc_bootmem(size * NR_CPUS); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1365 | |
| 1366 | __per_cpu_base = ptr - __per_cpu_start; |
| 1367 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1368 | for (i = 0; i < NR_CPUS; i++, ptr += size) |
| 1369 | memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1370 | } |