2 * This program is free software; you can redistribute it and/or
3 * modify it under the terms of the GNU General Public License
4 * as published by the Free Software Foundation; either version 2
5 * of the License, or (at your option) any later version.
7 * This program is distributed in the hope that it will be useful,
8 * but WITHOUT ANY WARRANTY; without even the implied warranty of
9 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
10 * GNU General Public License for more details.
12 * You should have received a copy of the GNU General Public License
13 * along with this program; if not, write to the Free Software
14 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
16 * Copyright (C) 2000, 2001 Kanoj Sarcar
17 * Copyright (C) 2000, 2001 Ralf Baechle
18 * Copyright (C) 2000, 2001 Silicon Graphics, Inc.
19 * Copyright (C) 2000, 2001, 2003 Broadcom Corporation
21 #include <linux/config.h>
22 #include <linux/cache.h>
23 #include <linux/delay.h>
24 #include <linux/init.h>
25 #include <linux/interrupt.h>
26 #include <linux/spinlock.h>
27 #include <linux/threads.h>
28 #include <linux/module.h>
29 #include <linux/time.h>
30 #include <linux/timex.h>
31 #include <linux/sched.h>
32 #include <linux/cpumask.h>
34 #include <asm/atomic.h>
36 #include <asm/processor.h>
37 #include <asm/system.h>
38 #include <asm/hardirq.h>
39 #include <asm/mmu_context.h>
42 cpumask_t phys_cpu_present_map; /* Bitmask of available CPUs */
43 volatile cpumask_t cpu_callin_map; /* Bitmask of started secondaries */
44 cpumask_t cpu_online_map; /* Bitmask of currently online CPUs */
45 int __cpu_number_map[NR_CPUS]; /* Map physical to logical */
46 int __cpu_logical_map[NR_CPUS]; /* Map logical to physical */
48 EXPORT_SYMBOL(cpu_online_map);
50 cycles_t cacheflush_time;
51 unsigned long cache_decay_ticks;
53 static void smp_tune_scheduling (void)
55 struct cache_desc *cd = ¤t_cpu_data.scache;
56 unsigned long cachesize; /* kB */
57 unsigned long bandwidth = 350; /* MB/s */
58 unsigned long cpu_khz;
61 * Crude estimate until we actually meassure ...
63 cpu_khz = loops_per_jiffy * 2 * HZ / 1000;
66 * Rough estimation for SMP scheduling, this is the number of
67 * cycles it takes for a fully memory-limited process to flush
68 * the SMP-local cache.
70 * (For a P5 this pretty much means we will choose another idle
71 * CPU almost always at wakeup time (this is due to the small
72 * L1 cache), on PIIs it's around 50-100 usecs, depending on
77 * This basically disables processor-affinity scheduling on SMP
78 * without a cycle counter. Currently all SMP capable MIPS
79 * processors have a cycle counter.
85 cachesize = cd->linesz * cd->sets * cd->ways;
86 cacheflush_time = (cpu_khz>>10) * (cachesize<<10) / bandwidth;
87 cache_decay_ticks = (long)cacheflush_time/cpu_khz * HZ / 1000;
89 printk("per-CPU timeslice cutoff: %ld.%02ld usecs.\n",
90 (long)cacheflush_time/(cpu_khz/1000),
91 ((long)cacheflush_time*100/(cpu_khz/1000)) % 100);
92 printk("task migration cache decay timeout: %ld msecs.\n",
93 (cache_decay_ticks + 1) * 1000 / HZ);
96 extern void __init calibrate_delay(void);
99 * First C code run on the secondary CPUs after being started up by
102 asmlinkage void start_secondary(void)
104 unsigned int cpu = smp_processor_id();
109 prom_init_secondary();
112 * XXX parity protection should be folded in here when it's converted
113 * to an option instead of something based on .cputype
117 cpu_data[cpu].udelay_val = loops_per_jiffy;
121 cpu_set(cpu, cpu_callin_map);
126 spinlock_t smp_call_lock = SPIN_LOCK_UNLOCKED;
128 struct call_data_struct *call_data;
131 * Run a function on all other CPUs.
132 * <func> The function to run. This must be fast and non-blocking.
133 * <info> An arbitrary pointer to pass to the function.
134 * <retry> If true, keep retrying until ready.
135 * <wait> If true, wait until function has completed on other CPUs.
136 * [RETURNS] 0 on success, else a negative status code.
138 * Does not return until remote CPUs are nearly ready to execute <func>
139 * or are or have executed.
141 * You must not call this function with disabled interrupts or from a
142 * hardware interrupt handler or from a bottom half handler.
144 int smp_call_function (void (*func) (void *info), void *info, int retry,
147 struct call_data_struct data;
148 int i, cpus = num_online_cpus() - 1;
149 int cpu = smp_processor_id();
154 /* Can deadlock when called with interrupts disabled */
155 WARN_ON(irqs_disabled());
159 atomic_set(&data.started, 0);
162 atomic_set(&data.finished, 0);
164 spin_lock(&smp_call_lock);
168 /* Send a message to all other CPUs and wait for them to respond */
169 for (i = 0; i < NR_CPUS; i++)
170 if (cpu_online(i) && i != cpu)
171 core_send_ipi(i, SMP_CALL_FUNCTION);
173 /* Wait for response */
174 /* FIXME: lock-up detection, backtrace on lock-up */
175 while (atomic_read(&data.started) != cpus)
179 while (atomic_read(&data.finished) != cpus)
181 spin_unlock(&smp_call_lock);
186 void smp_call_function_interrupt(void)
188 void (*func) (void *info) = call_data->func;
189 void *info = call_data->info;
190 int wait = call_data->wait;
193 * Notify initiating CPU that I've grabbed the data and am
194 * about to execute the function.
197 atomic_inc(&call_data->started);
200 * At this point the info structure may be out of scope unless wait==1.
208 atomic_inc(&call_data->finished);
212 static void stop_this_cpu(void *dummy)
217 cpu_clear(smp_processor_id(), cpu_online_map);
218 local_irq_enable(); /* May need to service _machine_restart IPI */
219 for (;;); /* Wait if available. */
222 void smp_send_stop(void)
224 smp_call_function(stop_this_cpu, NULL, 1, 0);
227 void __init smp_cpus_done(unsigned int max_cpus)
232 /* called from main before smp_init() */
233 void __init smp_prepare_cpus(unsigned int max_cpus)
235 cpu_data[0].udelay_val = loops_per_jiffy;
236 init_new_context(current, &init_mm);
237 current_thread_info()->cpu = 0;
238 smp_tune_scheduling();
239 prom_build_cpu_map();
240 prom_prepare_cpus(max_cpus);
243 /* preload SMP state for boot cpu */
244 void __devinit smp_prepare_boot_cpu(void)
247 * This assumes that bootup is always handled by the processor
248 * with the logic and physical number 0.
250 __cpu_number_map[0] = 0;
251 __cpu_logical_map[0] = 0;
252 cpu_set(0, phys_cpu_present_map);
253 cpu_set(0, cpu_online_map);
254 cpu_set(0, cpu_callin_map);
257 static struct task_struct * __init fork_by_hand(void)
261 * don't care about the eip and regs settings since
262 * we'll never reschedule the forked task.
264 return copy_process(CLONE_VM|CLONE_IDLETASK, 0, ®s, 0, NULL, NULL);
268 * Startup the CPU with this logical number
270 static int __init do_boot_cpu(int cpu)
272 struct task_struct *idle;
275 * The following code is purely to make sure
276 * Linux can schedule processes on this slave.
278 idle = fork_by_hand();
280 panic("failed fork for CPU %d\n", cpu);
282 wake_up_forked_process(idle);
285 * We remove it from the pidhash and the runqueue once we've
288 init_idle(idle, cpu);
290 unhash_process(idle);
292 prom_boot_secondary(cpu, idle);
295 while (!cpu_isset(cpu, cpu_callin_map))
298 cpu_set(cpu, cpu_online_map);
304 * Called once for each "cpu_possible(cpu)". Needs to spin up the cpu
305 * and keep control until "cpu_online(cpu)" is set. Note: cpu is
306 * physical, not logical.
308 int __devinit __cpu_up(unsigned int cpu)
312 /* Processor goes to start_secondary(), sets online flag */
313 ret = do_boot_cpu(cpu);
320 /* Not really SMP stuff ... */
321 int setup_profiling_timer(unsigned int multiplier)
326 static void flush_tlb_all_ipi(void *info)
328 local_flush_tlb_all();
331 void flush_tlb_all(void)
333 on_each_cpu(flush_tlb_all_ipi, 0, 1, 1);
336 static void flush_tlb_mm_ipi(void *mm)
338 local_flush_tlb_mm((struct mm_struct *)mm);
342 * The following tlb flush calls are invoked when old translations are
343 * being torn down, or pte attributes are changing. For single threaded
344 * address spaces, a new context is obtained on the current cpu, and tlb
345 * context on other cpus are invalidated to force a new context allocation
346 * at switch_mm time, should the mm ever be used on other cpus. For
347 * multithreaded address spaces, intercpu interrupts have to be sent.
348 * Another case where intercpu interrupts are required is when the target
349 * mm might be active on another cpu (eg debuggers doing the flushes on
350 * behalf of debugees, kswapd stealing pages from another process etc).
354 void flush_tlb_mm(struct mm_struct *mm)
358 if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
359 smp_call_function(flush_tlb_mm_ipi, (void *)mm, 1, 1);
362 for (i = 0; i < num_online_cpus(); i++)
363 if (smp_processor_id() != i)
364 cpu_context(i, mm) = 0;
366 local_flush_tlb_mm(mm);
371 struct flush_tlb_data {
372 struct vm_area_struct *vma;
377 static void flush_tlb_range_ipi(void *info)
379 struct flush_tlb_data *fd = (struct flush_tlb_data *)info;
381 local_flush_tlb_range(fd->vma, fd->addr1, fd->addr2);
384 void flush_tlb_range(struct vm_area_struct *vma, unsigned long start, unsigned long end)
386 struct mm_struct *mm = vma->vm_mm;
389 if ((atomic_read(&mm->mm_users) != 1) || (current->mm != mm)) {
390 struct flush_tlb_data fd;
395 smp_call_function(flush_tlb_range_ipi, (void *)&fd, 1, 1);
398 for (i = 0; i < num_online_cpus(); i++)
399 if (smp_processor_id() != i)
400 cpu_context(i, mm) = 0;
402 local_flush_tlb_range(vma, start, end);
406 static void flush_tlb_kernel_range_ipi(void *info)
408 struct flush_tlb_data *fd = (struct flush_tlb_data *)info;
410 local_flush_tlb_kernel_range(fd->addr1, fd->addr2);
413 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
415 struct flush_tlb_data fd;
419 on_each_cpu(flush_tlb_kernel_range_ipi, (void *)&fd, 1, 1);
422 static void flush_tlb_page_ipi(void *info)
424 struct flush_tlb_data *fd = (struct flush_tlb_data *)info;
426 local_flush_tlb_page(fd->vma, fd->addr1);
429 void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
432 if ((atomic_read(&vma->vm_mm->mm_users) != 1) || (current->mm != vma->vm_mm)) {
433 struct flush_tlb_data fd;
437 smp_call_function(flush_tlb_page_ipi, (void *)&fd, 1, 1);
440 for (i = 0; i < num_online_cpus(); i++)
441 if (smp_processor_id() != i)
442 cpu_context(i, vma->vm_mm) = 0;
444 local_flush_tlb_page(vma, page);
448 static void flush_tlb_one_ipi(void *info)
450 unsigned long vaddr = (unsigned long) info;
452 local_flush_tlb_one(vaddr);
455 void flush_tlb_one(unsigned long vaddr)
457 smp_call_function(flush_tlb_one_ipi, (void *) vaddr, 1, 1);
458 local_flush_tlb_one(vaddr);
461 EXPORT_SYMBOL(flush_tlb_page);
462 EXPORT_SYMBOL(flush_tlb_one);
463 EXPORT_SYMBOL(cpu_data);
464 EXPORT_SYMBOL(synchronize_irq);