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/cache.h>
22 #include <linux/delay.h>
23 #include <linux/init.h>
24 #include <linux/interrupt.h>
25 #include <linux/spinlock.h>
26 #include <linux/threads.h>
27 #include <linux/module.h>
28 #include <linux/time.h>
29 #include <linux/timex.h>
30 #include <linux/sched.h>
31 #include <linux/cpumask.h>
33 #include <asm/atomic.h>
35 #include <asm/processor.h>
36 #include <asm/system.h>
37 #include <asm/mmu_context.h>
40 cpumask_t phys_cpu_present_map
; /* Bitmask of available CPUs */
41 volatile cpumask_t cpu_callin_map
; /* Bitmask of started secondaries */
42 cpumask_t cpu_online_map
; /* Bitmask of currently online CPUs */
43 int __cpu_number_map
[NR_CPUS
]; /* Map physical to logical */
44 int __cpu_logical_map
[NR_CPUS
]; /* Map logical to physical */
46 EXPORT_SYMBOL(phys_cpu_present_map
);
47 EXPORT_SYMBOL(cpu_online_map
);
49 static void smp_tune_scheduling (void)
51 struct cache_desc
*cd
= ¤t_cpu_data
.scache
;
52 unsigned long cachesize
; /* kB */
53 unsigned long bandwidth
= 350; /* MB/s */
54 unsigned long cpu_khz
;
57 * Crude estimate until we actually meassure ...
59 cpu_khz
= loops_per_jiffy
* 2 * HZ
/ 1000;
62 * Rough estimation for SMP scheduling, this is the number of
63 * cycles it takes for a fully memory-limited process to flush
64 * the SMP-local cache.
66 * (For a P5 this pretty much means we will choose another idle
67 * CPU almost always at wakeup time (this is due to the small
68 * L1 cache), on PIIs it's around 50-100 usecs, depending on
74 cachesize
= cd
->linesz
* cd
->sets
* cd
->ways
;
77 extern void __init
calibrate_delay(void);
78 extern ATTRIB_NORET
void cpu_idle(void);
81 * First C code run on the secondary CPUs after being started up by
84 asmlinkage
void start_secondary(void)
86 unsigned int cpu
= smp_processor_id();
91 prom_init_secondary();
94 * XXX parity protection should be folded in here when it's converted
95 * to an option instead of something based on .cputype
99 cpu_data
[cpu
].udelay_val
= loops_per_jiffy
;
103 cpu_set(cpu
, cpu_callin_map
);
108 DEFINE_SPINLOCK(smp_call_lock
);
110 struct call_data_struct
*call_data
;
113 * Run a function on all other CPUs.
114 * <func> The function to run. This must be fast and non-blocking.
115 * <info> An arbitrary pointer to pass to the function.
116 * <retry> If true, keep retrying until ready.
117 * <wait> If true, wait until function has completed on other CPUs.
118 * [RETURNS] 0 on success, else a negative status code.
120 * Does not return until remote CPUs are nearly ready to execute <func>
121 * or are or have executed.
123 * You must not call this function with disabled interrupts or from a
124 * hardware interrupt handler or from a bottom half handler:
128 * smp_call_function()
131 * Wait for all cpus to acknowledge IPI
132 * CPU A has not responded, spin waiting
133 * for cpu A to respond, holding call_lock
134 * smp_call_function()
135 * Spin waiting for call_lock
138 int smp_call_function (void (*func
) (void *info
), void *info
, int retry
,
141 struct call_data_struct data
;
142 int i
, cpus
= num_online_cpus() - 1;
143 int cpu
= smp_processor_id();
146 * Can die spectacularly if this CPU isn't yet marked online
148 BUG_ON(!cpu_online(cpu
));
153 /* Can deadlock when called with interrupts disabled */
154 WARN_ON(irqs_disabled());
158 atomic_set(&data
.started
, 0);
161 atomic_set(&data
.finished
, 0);
163 spin_lock(&smp_call_lock
);
167 /* Send a message to all other CPUs and wait for them to respond */
168 for (i
= 0; i
< NR_CPUS
; i
++)
169 if (cpu_online(i
) && i
!= cpu
)
170 core_send_ipi(i
, SMP_CALL_FUNCTION
);
172 /* Wait for response */
173 /* FIXME: lock-up detection, backtrace on lock-up */
174 while (atomic_read(&data
.started
) != cpus
)
178 while (atomic_read(&data
.finished
) != cpus
)
180 spin_unlock(&smp_call_lock
);
185 void smp_call_function_interrupt(void)
187 void (*func
) (void *info
) = call_data
->func
;
188 void *info
= call_data
->info
;
189 int wait
= call_data
->wait
;
192 * Notify initiating CPU that I've grabbed the data and am
193 * about to execute the function.
196 atomic_inc(&call_data
->started
);
199 * At this point the info structure may be out of scope unless wait==1.
207 atomic_inc(&call_data
->finished
);
211 static void stop_this_cpu(void *dummy
)
216 cpu_clear(smp_processor_id(), cpu_online_map
);
217 local_irq_enable(); /* May need to service _machine_restart IPI */
218 for (;;); /* Wait if available. */
221 void smp_send_stop(void)
223 smp_call_function(stop_this_cpu
, NULL
, 1, 0);
226 void __init
smp_cpus_done(unsigned int max_cpus
)
231 /* called from main before smp_init() */
232 void __init
smp_prepare_cpus(unsigned int max_cpus
)
234 init_new_context(current
, &init_mm
);
235 current_thread_info()->cpu
= 0;
236 smp_tune_scheduling();
237 prom_prepare_cpus(max_cpus
);
240 /* preload SMP state for boot cpu */
241 void __devinit
smp_prepare_boot_cpu(void)
244 * This assumes that bootup is always handled by the processor
245 * with the logic and physical number 0.
247 __cpu_number_map
[0] = 0;
248 __cpu_logical_map
[0] = 0;
249 cpu_set(0, phys_cpu_present_map
);
250 cpu_set(0, cpu_online_map
);
251 cpu_set(0, cpu_callin_map
);
255 * Called once for each "cpu_possible(cpu)". Needs to spin up the cpu
256 * and keep control until "cpu_online(cpu)" is set. Note: cpu is
257 * physical, not logical.
259 int __devinit
__cpu_up(unsigned int cpu
)
261 struct task_struct
*idle
;
264 * Processor goes to start_secondary(), sets online flag
265 * The following code is purely to make sure
266 * Linux can schedule processes on this slave.
268 idle
= fork_idle(cpu
);
270 panic(KERN_ERR
"Fork failed for CPU %d", cpu
);
272 prom_boot_secondary(cpu
, idle
);
275 * Trust is futile. We should really have timeouts ...
277 while (!cpu_isset(cpu
, cpu_callin_map
))
280 cpu_set(cpu
, cpu_online_map
);
285 /* Not really SMP stuff ... */
286 int setup_profiling_timer(unsigned int multiplier
)
291 static void flush_tlb_all_ipi(void *info
)
293 local_flush_tlb_all();
296 void flush_tlb_all(void)
298 on_each_cpu(flush_tlb_all_ipi
, 0, 1, 1);
301 static void flush_tlb_mm_ipi(void *mm
)
303 local_flush_tlb_mm((struct mm_struct
*)mm
);
307 * The following tlb flush calls are invoked when old translations are
308 * being torn down, or pte attributes are changing. For single threaded
309 * address spaces, a new context is obtained on the current cpu, and tlb
310 * context on other cpus are invalidated to force a new context allocation
311 * at switch_mm time, should the mm ever be used on other cpus. For
312 * multithreaded address spaces, intercpu interrupts have to be sent.
313 * Another case where intercpu interrupts are required is when the target
314 * mm might be active on another cpu (eg debuggers doing the flushes on
315 * behalf of debugees, kswapd stealing pages from another process etc).
319 void flush_tlb_mm(struct mm_struct
*mm
)
323 if ((atomic_read(&mm
->mm_users
) != 1) || (current
->mm
!= mm
)) {
324 smp_call_function(flush_tlb_mm_ipi
, (void *)mm
, 1, 1);
327 for (i
= 0; i
< num_online_cpus(); i
++)
328 if (smp_processor_id() != i
)
329 cpu_context(i
, mm
) = 0;
331 local_flush_tlb_mm(mm
);
336 struct flush_tlb_data
{
337 struct vm_area_struct
*vma
;
342 static void flush_tlb_range_ipi(void *info
)
344 struct flush_tlb_data
*fd
= (struct flush_tlb_data
*)info
;
346 local_flush_tlb_range(fd
->vma
, fd
->addr1
, fd
->addr2
);
349 void flush_tlb_range(struct vm_area_struct
*vma
, unsigned long start
, unsigned long end
)
351 struct mm_struct
*mm
= vma
->vm_mm
;
354 if ((atomic_read(&mm
->mm_users
) != 1) || (current
->mm
!= mm
)) {
355 struct flush_tlb_data fd
;
360 smp_call_function(flush_tlb_range_ipi
, (void *)&fd
, 1, 1);
363 for (i
= 0; i
< num_online_cpus(); i
++)
364 if (smp_processor_id() != i
)
365 cpu_context(i
, mm
) = 0;
367 local_flush_tlb_range(vma
, start
, end
);
371 static void flush_tlb_kernel_range_ipi(void *info
)
373 struct flush_tlb_data
*fd
= (struct flush_tlb_data
*)info
;
375 local_flush_tlb_kernel_range(fd
->addr1
, fd
->addr2
);
378 void flush_tlb_kernel_range(unsigned long start
, unsigned long end
)
380 struct flush_tlb_data fd
;
384 on_each_cpu(flush_tlb_kernel_range_ipi
, (void *)&fd
, 1, 1);
387 static void flush_tlb_page_ipi(void *info
)
389 struct flush_tlb_data
*fd
= (struct flush_tlb_data
*)info
;
391 local_flush_tlb_page(fd
->vma
, fd
->addr1
);
394 void flush_tlb_page(struct vm_area_struct
*vma
, unsigned long page
)
397 if ((atomic_read(&vma
->vm_mm
->mm_users
) != 1) || (current
->mm
!= vma
->vm_mm
)) {
398 struct flush_tlb_data fd
;
402 smp_call_function(flush_tlb_page_ipi
, (void *)&fd
, 1, 1);
405 for (i
= 0; i
< num_online_cpus(); i
++)
406 if (smp_processor_id() != i
)
407 cpu_context(i
, vma
->vm_mm
) = 0;
409 local_flush_tlb_page(vma
, page
);
413 static void flush_tlb_one_ipi(void *info
)
415 unsigned long vaddr
= (unsigned long) info
;
417 local_flush_tlb_one(vaddr
);
420 void flush_tlb_one(unsigned long vaddr
)
422 smp_call_function(flush_tlb_one_ipi
, (void *) vaddr
, 1, 1);
423 local_flush_tlb_one(vaddr
);
426 EXPORT_SYMBOL(flush_tlb_page
);
427 EXPORT_SYMBOL(flush_tlb_one
);
428 EXPORT_SYMBOL(cpu_data
);
429 EXPORT_SYMBOL(synchronize_irq
);