| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* | 
 | 2 |  * SMP Support | 
 | 3 |  * | 
 | 4 |  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com> | 
 | 5 |  * Copyright (C) 1999, 2001, 2003 David Mosberger-Tang <davidm@hpl.hp.com> | 
 | 6 |  * | 
 | 7 |  * Lots of stuff stolen from arch/alpha/kernel/smp.c | 
 | 8 |  * | 
 | 9 |  * 01/05/16 Rohit Seth <rohit.seth@intel.com>  IA64-SMP functions. Reorganized | 
 | 10 |  * the existing code (on the lines of x86 port). | 
 | 11 |  * 00/09/11 David Mosberger <davidm@hpl.hp.com> Do loops_per_jiffy | 
 | 12 |  * calibration on each CPU. | 
 | 13 |  * 00/08/23 Asit Mallick <asit.k.mallick@intel.com> fixed logical processor id | 
 | 14 |  * 00/03/31 Rohit Seth <rohit.seth@intel.com>	Fixes for Bootstrap Processor | 
 | 15 |  * & cpu_online_map now gets done here (instead of setup.c) | 
 | 16 |  * 99/10/05 davidm	Update to bring it in sync with new command-line processing | 
 | 17 |  *  scheme. | 
 | 18 |  * 10/13/00 Goutham Rao <goutham.rao@intel.com> Updated smp_call_function and | 
 | 19 |  *		smp_call_function_single to resend IPI on timeouts | 
 | 20 |  */ | 
 | 21 | #include <linux/module.h> | 
 | 22 | #include <linux/kernel.h> | 
 | 23 | #include <linux/sched.h> | 
 | 24 | #include <linux/init.h> | 
 | 25 | #include <linux/interrupt.h> | 
 | 26 | #include <linux/smp.h> | 
 | 27 | #include <linux/kernel_stat.h> | 
 | 28 | #include <linux/mm.h> | 
 | 29 | #include <linux/cache.h> | 
 | 30 | #include <linux/delay.h> | 
 | 31 | #include <linux/efi.h> | 
 | 32 | #include <linux/bitops.h> | 
 | 33 |  | 
 | 34 | #include <asm/atomic.h> | 
 | 35 | #include <asm/current.h> | 
 | 36 | #include <asm/delay.h> | 
 | 37 | #include <asm/machvec.h> | 
 | 38 | #include <asm/io.h> | 
 | 39 | #include <asm/irq.h> | 
 | 40 | #include <asm/page.h> | 
 | 41 | #include <asm/pgalloc.h> | 
 | 42 | #include <asm/pgtable.h> | 
 | 43 | #include <asm/processor.h> | 
 | 44 | #include <asm/ptrace.h> | 
 | 45 | #include <asm/sal.h> | 
 | 46 | #include <asm/system.h> | 
 | 47 | #include <asm/tlbflush.h> | 
 | 48 | #include <asm/unistd.h> | 
 | 49 | #include <asm/mca.h> | 
 | 50 |  | 
 | 51 | /* | 
 | 52 |  * Structure and data for smp_call_function(). This is designed to minimise static memory | 
 | 53 |  * requirements. It also looks cleaner. | 
 | 54 |  */ | 
 | 55 | static  __cacheline_aligned DEFINE_SPINLOCK(call_lock); | 
 | 56 |  | 
 | 57 | struct call_data_struct { | 
 | 58 | 	void (*func) (void *info); | 
 | 59 | 	void *info; | 
 | 60 | 	long wait; | 
 | 61 | 	atomic_t started; | 
 | 62 | 	atomic_t finished; | 
 | 63 | }; | 
 | 64 |  | 
 | 65 | static volatile struct call_data_struct *call_data; | 
 | 66 |  | 
 | 67 | #define IPI_CALL_FUNC		0 | 
 | 68 | #define IPI_CPU_STOP		1 | 
 | 69 |  | 
 | 70 | /* This needs to be cacheline aligned because it is written to by *other* CPUs.  */ | 
 | 71 | static DEFINE_PER_CPU(u64, ipi_operation) ____cacheline_aligned; | 
 | 72 |  | 
 | 73 | extern void cpu_halt (void); | 
 | 74 |  | 
 | 75 | void | 
 | 76 | lock_ipi_calllock(void) | 
 | 77 | { | 
 | 78 | 	spin_lock_irq(&call_lock); | 
 | 79 | } | 
 | 80 |  | 
 | 81 | void | 
 | 82 | unlock_ipi_calllock(void) | 
 | 83 | { | 
 | 84 | 	spin_unlock_irq(&call_lock); | 
 | 85 | } | 
 | 86 |  | 
 | 87 | static void | 
 | 88 | stop_this_cpu (void) | 
 | 89 | { | 
 | 90 | 	/* | 
 | 91 | 	 * Remove this CPU: | 
 | 92 | 	 */ | 
 | 93 | 	cpu_clear(smp_processor_id(), cpu_online_map); | 
 | 94 | 	max_xtp(); | 
 | 95 | 	local_irq_disable(); | 
 | 96 | 	cpu_halt(); | 
 | 97 | } | 
 | 98 |  | 
 | 99 | void | 
 | 100 | cpu_die(void) | 
 | 101 | { | 
 | 102 | 	max_xtp(); | 
 | 103 | 	local_irq_disable(); | 
 | 104 | 	cpu_halt(); | 
 | 105 | 	/* Should never be here */ | 
 | 106 | 	BUG(); | 
 | 107 | 	for (;;); | 
 | 108 | } | 
 | 109 |  | 
 | 110 | irqreturn_t | 
 | 111 | handle_IPI (int irq, void *dev_id, struct pt_regs *regs) | 
 | 112 | { | 
 | 113 | 	int this_cpu = get_cpu(); | 
 | 114 | 	unsigned long *pending_ipis = &__ia64_per_cpu_var(ipi_operation); | 
 | 115 | 	unsigned long ops; | 
 | 116 |  | 
 | 117 | 	mb();	/* Order interrupt and bit testing. */ | 
 | 118 | 	while ((ops = xchg(pending_ipis, 0)) != 0) { | 
 | 119 | 		mb();	/* Order bit clearing and data access. */ | 
 | 120 | 		do { | 
 | 121 | 			unsigned long which; | 
 | 122 |  | 
 | 123 | 			which = ffz(~ops); | 
 | 124 | 			ops &= ~(1 << which); | 
 | 125 |  | 
 | 126 | 			switch (which) { | 
 | 127 | 			      case IPI_CALL_FUNC: | 
 | 128 | 			      { | 
 | 129 | 				      struct call_data_struct *data; | 
 | 130 | 				      void (*func)(void *info); | 
 | 131 | 				      void *info; | 
 | 132 | 				      int wait; | 
 | 133 |  | 
 | 134 | 				      /* release the 'pointer lock' */ | 
 | 135 | 				      data = (struct call_data_struct *) call_data; | 
 | 136 | 				      func = data->func; | 
 | 137 | 				      info = data->info; | 
 | 138 | 				      wait = data->wait; | 
 | 139 |  | 
 | 140 | 				      mb(); | 
 | 141 | 				      atomic_inc(&data->started); | 
 | 142 | 				      /* | 
 | 143 | 				       * At this point the structure may be gone unless | 
 | 144 | 				       * wait is true. | 
 | 145 | 				       */ | 
 | 146 | 				      (*func)(info); | 
 | 147 |  | 
 | 148 | 				      /* Notify the sending CPU that the task is done.  */ | 
 | 149 | 				      mb(); | 
 | 150 | 				      if (wait) | 
 | 151 | 					      atomic_inc(&data->finished); | 
 | 152 | 			      } | 
 | 153 | 			      break; | 
 | 154 |  | 
 | 155 | 			      case IPI_CPU_STOP: | 
 | 156 | 				stop_this_cpu(); | 
 | 157 | 				break; | 
 | 158 |  | 
 | 159 | 			      default: | 
 | 160 | 				printk(KERN_CRIT "Unknown IPI on CPU %d: %lu\n", this_cpu, which); | 
 | 161 | 				break; | 
 | 162 | 			} | 
 | 163 | 		} while (ops); | 
 | 164 | 		mb();	/* Order data access and bit testing. */ | 
 | 165 | 	} | 
 | 166 | 	put_cpu(); | 
 | 167 | 	return IRQ_HANDLED; | 
 | 168 | } | 
 | 169 |  | 
 | 170 | /* | 
 | 171 |  * Called with preeemption disabled. | 
 | 172 |  */ | 
 | 173 | static inline void | 
 | 174 | send_IPI_single (int dest_cpu, int op) | 
 | 175 | { | 
 | 176 | 	set_bit(op, &per_cpu(ipi_operation, dest_cpu)); | 
 | 177 | 	platform_send_ipi(dest_cpu, IA64_IPI_VECTOR, IA64_IPI_DM_INT, 0); | 
 | 178 | } | 
 | 179 |  | 
 | 180 | /* | 
 | 181 |  * Called with preeemption disabled. | 
 | 182 |  */ | 
 | 183 | static inline void | 
 | 184 | send_IPI_allbutself (int op) | 
 | 185 | { | 
 | 186 | 	unsigned int i; | 
 | 187 |  | 
| hawkes@sgi.com | dc565b5 | 2005-10-10 08:43:26 -0700 | [diff] [blame] | 188 | 	for_each_online_cpu(i) { | 
 | 189 | 		if (i != smp_processor_id()) | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 190 | 			send_IPI_single(i, op); | 
 | 191 | 	} | 
 | 192 | } | 
 | 193 |  | 
 | 194 | /* | 
 | 195 |  * Called with preeemption disabled. | 
 | 196 |  */ | 
 | 197 | static inline void | 
 | 198 | send_IPI_all (int op) | 
 | 199 | { | 
 | 200 | 	int i; | 
 | 201 |  | 
| hawkes@sgi.com | dc565b5 | 2005-10-10 08:43:26 -0700 | [diff] [blame] | 202 | 	for_each_online_cpu(i) { | 
 | 203 | 		send_IPI_single(i, op); | 
 | 204 | 	} | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | } | 
 | 206 |  | 
 | 207 | /* | 
 | 208 |  * Called with preeemption disabled. | 
 | 209 |  */ | 
 | 210 | static inline void | 
 | 211 | send_IPI_self (int op) | 
 | 212 | { | 
 | 213 | 	send_IPI_single(smp_processor_id(), op); | 
 | 214 | } | 
 | 215 |  | 
 | 216 | /* | 
 | 217 |  * Called with preeemption disabled. | 
 | 218 |  */ | 
 | 219 | void | 
 | 220 | smp_send_reschedule (int cpu) | 
 | 221 | { | 
 | 222 | 	platform_send_ipi(cpu, IA64_IPI_RESCHEDULE, IA64_IPI_DM_INT, 0); | 
 | 223 | } | 
 | 224 |  | 
 | 225 | void | 
 | 226 | smp_flush_tlb_all (void) | 
 | 227 | { | 
 | 228 | 	on_each_cpu((void (*)(void *))local_flush_tlb_all, NULL, 1, 1); | 
 | 229 | } | 
 | 230 |  | 
 | 231 | void | 
 | 232 | smp_flush_tlb_mm (struct mm_struct *mm) | 
 | 233 | { | 
| Peter Chubb | a68db76 | 2005-06-23 21:14:00 -0700 | [diff] [blame] | 234 | 	preempt_disable(); | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 235 | 	/* this happens for the common case of a single-threaded fork():  */ | 
 | 236 | 	if (likely(mm == current->active_mm && atomic_read(&mm->mm_users) == 1)) | 
 | 237 | 	{ | 
 | 238 | 		local_finish_flush_tlb_mm(mm); | 
| Peter Chubb | a68db76 | 2005-06-23 21:14:00 -0700 | [diff] [blame] | 239 | 		preempt_enable(); | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 240 | 		return; | 
 | 241 | 	} | 
 | 242 |  | 
| Peter Chubb | a68db76 | 2005-06-23 21:14:00 -0700 | [diff] [blame] | 243 | 	preempt_enable(); | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | 	/* | 
 | 245 | 	 * We could optimize this further by using mm->cpu_vm_mask to track which CPUs | 
 | 246 | 	 * have been running in the address space.  It's not clear that this is worth the | 
 | 247 | 	 * trouble though: to avoid races, we have to raise the IPI on the target CPU | 
 | 248 | 	 * anyhow, and once a CPU is interrupted, the cost of local_flush_tlb_all() is | 
 | 249 | 	 * rather trivial. | 
 | 250 | 	 */ | 
 | 251 | 	on_each_cpu((void (*)(void *))local_finish_flush_tlb_mm, mm, 1, 1); | 
 | 252 | } | 
 | 253 |  | 
 | 254 | /* | 
 | 255 |  * Run a function on another CPU | 
 | 256 |  *  <func>	The function to run. This must be fast and non-blocking. | 
 | 257 |  *  <info>	An arbitrary pointer to pass to the function. | 
 | 258 |  *  <nonatomic>	Currently unused. | 
 | 259 |  *  <wait>	If true, wait until function has completed on other CPUs. | 
 | 260 |  *  [RETURNS]   0 on success, else a negative status code. | 
 | 261 |  * | 
 | 262 |  * Does not return until the remote CPU is nearly ready to execute <func> | 
 | 263 |  * or is or has executed. | 
 | 264 |  */ | 
 | 265 |  | 
 | 266 | int | 
 | 267 | smp_call_function_single (int cpuid, void (*func) (void *info), void *info, int nonatomic, | 
 | 268 | 			  int wait) | 
 | 269 | { | 
 | 270 | 	struct call_data_struct data; | 
 | 271 | 	int cpus = 1; | 
 | 272 | 	int me = get_cpu(); /* prevent preemption and reschedule on another processor */ | 
 | 273 |  | 
 | 274 | 	if (cpuid == me) { | 
| Christophe Lucas | 52a0de2 | 2005-06-20 14:50:00 -0700 | [diff] [blame] | 275 | 		printk(KERN_INFO "%s: trying to call self\n", __FUNCTION__); | 
| Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 276 | 		put_cpu(); | 
 | 277 | 		return -EBUSY; | 
 | 278 | 	} | 
 | 279 |  | 
 | 280 | 	data.func = func; | 
 | 281 | 	data.info = info; | 
 | 282 | 	atomic_set(&data.started, 0); | 
 | 283 | 	data.wait = wait; | 
 | 284 | 	if (wait) | 
 | 285 | 		atomic_set(&data.finished, 0); | 
 | 286 |  | 
 | 287 | 	spin_lock_bh(&call_lock); | 
 | 288 |  | 
 | 289 | 	call_data = &data; | 
 | 290 | 	mb();	/* ensure store to call_data precedes setting of IPI_CALL_FUNC */ | 
 | 291 |   	send_IPI_single(cpuid, IPI_CALL_FUNC); | 
 | 292 |  | 
 | 293 | 	/* Wait for response */ | 
 | 294 | 	while (atomic_read(&data.started) != cpus) | 
 | 295 | 		cpu_relax(); | 
 | 296 |  | 
 | 297 | 	if (wait) | 
 | 298 | 		while (atomic_read(&data.finished) != cpus) | 
 | 299 | 			cpu_relax(); | 
 | 300 | 	call_data = NULL; | 
 | 301 |  | 
 | 302 | 	spin_unlock_bh(&call_lock); | 
 | 303 | 	put_cpu(); | 
 | 304 | 	return 0; | 
 | 305 | } | 
 | 306 | EXPORT_SYMBOL(smp_call_function_single); | 
 | 307 |  | 
 | 308 | /* | 
 | 309 |  * this function sends a 'generic call function' IPI to all other CPUs | 
 | 310 |  * in the system. | 
 | 311 |  */ | 
 | 312 |  | 
 | 313 | /* | 
 | 314 |  *  [SUMMARY]	Run a function on all other CPUs. | 
 | 315 |  *  <func>	The function to run. This must be fast and non-blocking. | 
 | 316 |  *  <info>	An arbitrary pointer to pass to the function. | 
 | 317 |  *  <nonatomic>	currently unused. | 
 | 318 |  *  <wait>	If true, wait (atomically) until function has completed on other CPUs. | 
 | 319 |  *  [RETURNS]   0 on success, else a negative status code. | 
 | 320 |  * | 
 | 321 |  * Does not return until remote CPUs are nearly ready to execute <func> or are or have | 
 | 322 |  * executed. | 
 | 323 |  * | 
 | 324 |  * You must not call this function with disabled interrupts or from a | 
 | 325 |  * hardware interrupt handler or from a bottom half handler. | 
 | 326 |  */ | 
 | 327 | int | 
 | 328 | smp_call_function (void (*func) (void *info), void *info, int nonatomic, int wait) | 
 | 329 | { | 
 | 330 | 	struct call_data_struct data; | 
 | 331 | 	int cpus = num_online_cpus()-1; | 
 | 332 |  | 
 | 333 | 	if (!cpus) | 
 | 334 | 		return 0; | 
 | 335 |  | 
 | 336 | 	/* Can deadlock when called with interrupts disabled */ | 
 | 337 | 	WARN_ON(irqs_disabled()); | 
 | 338 |  | 
 | 339 | 	data.func = func; | 
 | 340 | 	data.info = info; | 
 | 341 | 	atomic_set(&data.started, 0); | 
 | 342 | 	data.wait = wait; | 
 | 343 | 	if (wait) | 
 | 344 | 		atomic_set(&data.finished, 0); | 
 | 345 |  | 
 | 346 | 	spin_lock(&call_lock); | 
 | 347 |  | 
 | 348 | 	call_data = &data; | 
 | 349 | 	mb();	/* ensure store to call_data precedes setting of IPI_CALL_FUNC */ | 
 | 350 | 	send_IPI_allbutself(IPI_CALL_FUNC); | 
 | 351 |  | 
 | 352 | 	/* Wait for response */ | 
 | 353 | 	while (atomic_read(&data.started) != cpus) | 
 | 354 | 		cpu_relax(); | 
 | 355 |  | 
 | 356 | 	if (wait) | 
 | 357 | 		while (atomic_read(&data.finished) != cpus) | 
 | 358 | 			cpu_relax(); | 
 | 359 | 	call_data = NULL; | 
 | 360 |  | 
 | 361 | 	spin_unlock(&call_lock); | 
 | 362 | 	return 0; | 
 | 363 | } | 
 | 364 | EXPORT_SYMBOL(smp_call_function); | 
 | 365 |  | 
 | 366 | /* | 
 | 367 |  * this function calls the 'stop' function on all other CPUs in the system. | 
 | 368 |  */ | 
 | 369 | void | 
 | 370 | smp_send_stop (void) | 
 | 371 | { | 
 | 372 | 	send_IPI_allbutself(IPI_CPU_STOP); | 
 | 373 | } | 
 | 374 |  | 
 | 375 | int __init | 
 | 376 | setup_profiling_timer (unsigned int multiplier) | 
 | 377 | { | 
 | 378 | 	return -EINVAL; | 
 | 379 | } |