blob: aa36223497b931c9dc96c1ef2d536672aa36248d [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/* smp.c: Sparc64 SMP support.
2 *
David S. Millercf3d7c12008-03-26 01:11:55 -07003 * Copyright (C) 1997, 2007, 2008 David S. Miller (davem@davemloft.net)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004 */
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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070013#include <linux/interrupt.h>
14#include <linux/kernel_stat.h>
15#include <linux/delay.h>
16#include <linux/init.h>
17#include <linux/spinlock.h>
18#include <linux/fs.h>
19#include <linux/seq_file.h>
20#include <linux/cache.h>
21#include <linux/jiffies.h>
22#include <linux/profile.h>
David S. Miller73fffc02009-04-01 16:15:20 -070023#include <linux/bootmem.h>
David S. Miller4fd78a52009-04-08 20:32:02 -070024#include <linux/vmalloc.h>
David S. Miller82960b82008-10-12 20:55:24 -070025#include <linux/cpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
27#include <asm/head.h>
28#include <asm/ptrace.h>
29#include <asm/atomic.h>
30#include <asm/tlbflush.h>
31#include <asm/mmu_context.h>
32#include <asm/cpudata.h>
David S. Miller27a2ef32007-07-14 00:58:53 -070033#include <asm/hvtramp.h>
34#include <asm/io.h>
David S. Millercf3d7c12008-03-26 01:11:55 -070035#include <asm/timer.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
37#include <asm/irq.h>
Al Viro6d24c8d2006-10-08 08:23:28 -040038#include <asm/irq_regs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <asm/page.h>
40#include <asm/pgtable.h>
41#include <asm/oplib.h>
42#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <asm/starfire.h>
44#include <asm/tlb.h>
David S. Miller56fb4df2006-02-26 23:24:22 -080045#include <asm/sections.h>
David S. Miller07f8e5f2006-06-21 23:34:02 -070046#include <asm/prom.h>
David S. Miller5cbc3072007-05-25 15:49:59 -070047#include <asm/mdesc.h>
David S. Miller4f0234f2007-07-13 16:03:42 -070048#include <asm/ldc.h>
David S. Millere0204402007-07-16 03:49:40 -070049#include <asm/hypervisor.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050
Hong H. Pham280ff972009-06-04 02:10:11 -070051#include "cpumap.h"
52
David S. Millera2f9f6b2007-06-04 21:48:33 -070053int sparc64_multi_core __read_mostly;
54
Mike Travisd5a74302007-10-16 01:24:05 -070055DEFINE_PER_CPU(cpumask_t, cpu_sibling_map) = CPU_MASK_NONE;
David S. Millerf78eae22007-06-04 17:01:39 -070056cpumask_t cpu_core_map[NR_CPUS] __read_mostly =
57 { [0 ... NR_CPUS-1] = CPU_MASK_NONE };
David S. Miller4f0234f2007-07-13 16:03:42 -070058
Mike Travisd5a74302007-10-16 01:24:05 -070059EXPORT_PER_CPU_SYMBOL(cpu_sibling_map);
David S. Miller4f0234f2007-07-13 16:03:42 -070060EXPORT_SYMBOL(cpu_core_map);
61
Linus Torvalds1da177e2005-04-16 15:20:36 -070062static cpumask_t smp_commenced_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64void smp_info(struct seq_file *m)
65{
66 int i;
67
68 seq_printf(m, "State:\n");
Andrew Morton394e3902006-03-23 03:01:05 -080069 for_each_online_cpu(i)
70 seq_printf(m, "CPU%d:\t\tonline\n", i);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071}
72
73void smp_bogo(struct seq_file *m)
74{
75 int i;
76
Andrew Morton394e3902006-03-23 03:01:05 -080077 for_each_online_cpu(i)
78 seq_printf(m,
Andrew Morton394e3902006-03-23 03:01:05 -080079 "Cpu%dClkTck\t: %016lx\n",
Andrew Morton394e3902006-03-23 03:01:05 -080080 i, cpu_data(i).clock_tick);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081}
82
David S. Miller112f4872007-03-05 15:28:37 -080083extern void setup_sparc64_timer(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -070084
85static volatile unsigned long callin_flag = 0;
86
Sam Ravnborg0f7f22d2008-02-20 22:22:16 -080087void __cpuinit smp_callin(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
89 int cpuid = hard_smp_processor_id();
90
David S. Miller56fb4df2006-02-26 23:24:22 -080091 __local_per_cpu_offset = __per_cpu_offset(cpuid);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092
David S. Miller4a07e642006-02-14 13:49:32 -080093 if (tlb_type == hypervisor)
David S. Miller490384e2006-02-11 14:41:18 -080094 sun4v_ktsb_register();
David S. Miller481295f2006-02-07 21:51:08 -080095
David S. Miller56fb4df2006-02-26 23:24:22 -080096 __flush_tlb_all();
Linus Torvalds1da177e2005-04-16 15:20:36 -070097
David S. Miller112f4872007-03-05 15:28:37 -080098 setup_sparc64_timer();
Linus Torvalds1da177e2005-04-16 15:20:36 -070099
David S. Miller816242d2005-05-23 15:52:08 -0700100 if (cheetah_pcache_forced_on)
101 cheetah_enable_pcache();
102
Linus Torvalds1da177e2005-04-16 15:20:36 -0700103 local_irq_enable();
104
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105 callin_flag = 1;
106 __asm__ __volatile__("membar #Sync\n\t"
107 "flush %%g6" : : : "memory");
108
109 /* Clear this or we will die instantly when we
110 * schedule back to this idler...
111 */
David S. Millerdb7d9a42005-07-24 19:36:26 -0700112 current_thread_info()->new_child = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700113
114 /* Attach to the address space of init_task. */
115 atomic_inc(&init_mm.mm_count);
116 current->active_mm = &init_mm;
117
David S. Miller82960b82008-10-12 20:55:24 -0700118 /* inform the notifiers about the new cpu */
119 notify_cpu_starting(cpuid);
120
Linus Torvalds1da177e2005-04-16 15:20:36 -0700121 while (!cpu_isset(cpuid, smp_commenced_mask))
David S. Miller4f071182005-08-29 12:46:22 -0700122 rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123
David S. Miller8e255ba2009-04-08 21:06:35 -0700124 ipi_call_lock_irq();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125 cpu_set(cpuid, cpu_online_map);
David S. Miller8e255ba2009-04-08 21:06:35 -0700126 ipi_call_unlock_irq();
Nick Piggin5bfb5d62005-11-08 21:39:01 -0800127
128 /* idle thread is expected to have preempt disabled */
129 preempt_disable();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700130}
131
132void cpu_panic(void)
133{
134 printk("CPU[%d]: Returns from cpu_idle!\n", smp_processor_id());
135 panic("SMP bolixed\n");
136}
137
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138/* This tick register synchronization scheme is taken entirely from
139 * the ia64 port, see arch/ia64/kernel/smpboot.c for details and credit.
140 *
141 * The only change I've made is to rework it so that the master
142 * initiates the synchonization instead of the slave. -DaveM
143 */
144
145#define MASTER 0
146#define SLAVE (SMP_CACHE_BYTES/sizeof(unsigned long))
147
148#define NUM_ROUNDS 64 /* magic value */
149#define NUM_ITERS 5 /* likewise */
150
151static DEFINE_SPINLOCK(itc_sync_lock);
152static unsigned long go[SLAVE + 1];
153
154#define DEBUG_TICK_SYNC 0
155
156static inline long get_delta (long *rt, long *master)
157{
158 unsigned long best_t0 = 0, best_t1 = ~0UL, best_tm = 0;
159 unsigned long tcenter, t0, t1, tm;
160 unsigned long i;
161
162 for (i = 0; i < NUM_ITERS; i++) {
163 t0 = tick_ops->get_tick();
164 go[MASTER] = 1;
David S. Miller293666b2008-11-15 13:33:25 -0800165 membar_safe("#StoreLoad");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 while (!(tm = go[SLAVE]))
David S. Miller4f071182005-08-29 12:46:22 -0700167 rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168 go[SLAVE] = 0;
David S. Miller4f071182005-08-29 12:46:22 -0700169 wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 t1 = tick_ops->get_tick();
171
172 if (t1 - t0 < best_t1 - best_t0)
173 best_t0 = t0, best_t1 = t1, best_tm = tm;
174 }
175
176 *rt = best_t1 - best_t0;
177 *master = best_tm - best_t0;
178
179 /* average best_t0 and best_t1 without overflow: */
180 tcenter = (best_t0/2 + best_t1/2);
181 if (best_t0 % 2 + best_t1 % 2 == 2)
182 tcenter++;
183 return tcenter - best_tm;
184}
185
186void smp_synchronize_tick_client(void)
187{
188 long i, delta, adj, adjust_latency = 0, done = 0;
189 unsigned long flags, rt, master_time_stamp, bound;
190#if DEBUG_TICK_SYNC
191 struct {
192 long rt; /* roundtrip time */
193 long master; /* master's timestamp */
194 long diff; /* difference between midpoint and master's timestamp */
195 long lat; /* estimate of itc adjustment latency */
196 } t[NUM_ROUNDS];
197#endif
198
199 go[MASTER] = 1;
200
201 while (go[MASTER])
David S. Miller4f071182005-08-29 12:46:22 -0700202 rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203
204 local_irq_save(flags);
205 {
206 for (i = 0; i < NUM_ROUNDS; i++) {
207 delta = get_delta(&rt, &master_time_stamp);
208 if (delta == 0) {
209 done = 1; /* let's lock on to this... */
210 bound = rt;
211 }
212
213 if (!done) {
214 if (i > 0) {
215 adjust_latency += -delta;
216 adj = -delta + adjust_latency/4;
217 } else
218 adj = -delta;
219
David S. Miller112f4872007-03-05 15:28:37 -0800220 tick_ops->add_tick(adj);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700221 }
222#if DEBUG_TICK_SYNC
223 t[i].rt = rt;
224 t[i].master = master_time_stamp;
225 t[i].diff = delta;
226 t[i].lat = adjust_latency/4;
227#endif
228 }
229 }
230 local_irq_restore(flags);
231
232#if DEBUG_TICK_SYNC
233 for (i = 0; i < NUM_ROUNDS; i++)
234 printk("rt=%5ld master=%5ld diff=%5ld adjlat=%5ld\n",
235 t[i].rt, t[i].master, t[i].diff, t[i].lat);
236#endif
237
Joe Perches519c4d22007-11-19 23:43:00 -0800238 printk(KERN_INFO "CPU %d: synchronized TICK with master CPU "
239 "(last diff %ld cycles, maxerr %lu cycles)\n",
240 smp_processor_id(), delta, rt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241}
242
243static void smp_start_sync_tick_client(int cpu);
244
245static void smp_synchronize_one_tick(int cpu)
246{
247 unsigned long flags, i;
248
249 go[MASTER] = 0;
250
251 smp_start_sync_tick_client(cpu);
252
253 /* wait for client to be ready */
254 while (!go[MASTER])
David S. Miller4f071182005-08-29 12:46:22 -0700255 rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256
257 /* now let the client proceed into his loop */
258 go[MASTER] = 0;
David S. Miller293666b2008-11-15 13:33:25 -0800259 membar_safe("#StoreLoad");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
261 spin_lock_irqsave(&itc_sync_lock, flags);
262 {
263 for (i = 0; i < NUM_ROUNDS*NUM_ITERS; i++) {
264 while (!go[MASTER])
David S. Miller4f071182005-08-29 12:46:22 -0700265 rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266 go[MASTER] = 0;
David S. Miller4f071182005-08-29 12:46:22 -0700267 wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 go[SLAVE] = tick_ops->get_tick();
David S. Miller293666b2008-11-15 13:33:25 -0800269 membar_safe("#StoreLoad");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 }
271 }
272 spin_unlock_irqrestore(&itc_sync_lock, flags);
273}
274
David S. Millerb14f5c12007-07-14 00:45:16 -0700275#if defined(CONFIG_SUN_LDOMS) && defined(CONFIG_HOTPLUG_CPU)
David S. Miller27a2ef32007-07-14 00:58:53 -0700276/* XXX Put this in some common place. XXX */
277static unsigned long kimage_addr_to_ra(void *p)
278{
279 unsigned long val = (unsigned long) p;
280
281 return kern_base + (val - KERNBASE);
282}
283
David S. Miller557fe0e2009-03-31 17:15:40 -0700284static void __cpuinit ldom_startcpu_cpuid(unsigned int cpu, unsigned long thread_reg, void **descrp)
David S. Millerb14f5c12007-07-14 00:45:16 -0700285{
286 extern unsigned long sparc64_ttable_tl0;
287 extern unsigned long kern_locked_tte_data;
David S. Millerb14f5c12007-07-14 00:45:16 -0700288 struct hvtramp_descr *hdesc;
289 unsigned long trampoline_ra;
290 struct trap_per_cpu *tb;
291 u64 tte_vaddr, tte_data;
292 unsigned long hv_err;
David S. Miller64658742008-03-21 17:01:38 -0700293 int i;
David S. Millerb14f5c12007-07-14 00:45:16 -0700294
David S. Miller64658742008-03-21 17:01:38 -0700295 hdesc = kzalloc(sizeof(*hdesc) +
296 (sizeof(struct hvtramp_mapping) *
297 num_kernel_image_mappings - 1),
298 GFP_KERNEL);
David S. Millerb14f5c12007-07-14 00:45:16 -0700299 if (!hdesc) {
David S. Miller27a2ef32007-07-14 00:58:53 -0700300 printk(KERN_ERR "ldom_startcpu_cpuid: Cannot allocate "
David S. Millerb14f5c12007-07-14 00:45:16 -0700301 "hvtramp_descr.\n");
302 return;
303 }
David S. Miller557fe0e2009-03-31 17:15:40 -0700304 *descrp = hdesc;
David S. Millerb14f5c12007-07-14 00:45:16 -0700305
306 hdesc->cpu = cpu;
David S. Miller64658742008-03-21 17:01:38 -0700307 hdesc->num_mappings = num_kernel_image_mappings;
David S. Millerb14f5c12007-07-14 00:45:16 -0700308
309 tb = &trap_block[cpu];
David S. Millerb14f5c12007-07-14 00:45:16 -0700310
311 hdesc->fault_info_va = (unsigned long) &tb->fault_info;
312 hdesc->fault_info_pa = kimage_addr_to_ra(&tb->fault_info);
313
314 hdesc->thread_reg = thread_reg;
315
316 tte_vaddr = (unsigned long) KERNBASE;
317 tte_data = kern_locked_tte_data;
318
David S. Miller64658742008-03-21 17:01:38 -0700319 for (i = 0; i < hdesc->num_mappings; i++) {
320 hdesc->maps[i].vaddr = tte_vaddr;
321 hdesc->maps[i].tte = tte_data;
David S. Millerb14f5c12007-07-14 00:45:16 -0700322 tte_vaddr += 0x400000;
323 tte_data += 0x400000;
David S. Millerb14f5c12007-07-14 00:45:16 -0700324 }
325
326 trampoline_ra = kimage_addr_to_ra(hv_cpu_startup);
327
328 hv_err = sun4v_cpu_start(cpu, trampoline_ra,
329 kimage_addr_to_ra(&sparc64_ttable_tl0),
330 __pa(hdesc));
David S. Millere0204402007-07-16 03:49:40 -0700331 if (hv_err)
332 printk(KERN_ERR "ldom_startcpu_cpuid: sun4v_cpu_start() "
333 "gives error %lu\n", hv_err);
David S. Millerb14f5c12007-07-14 00:45:16 -0700334}
335#endif
336
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337extern unsigned long sparc64_cpu_startup;
338
339/* The OBP cpu startup callback truncates the 3rd arg cookie to
340 * 32-bits (I think) so to be safe we have it read the pointer
341 * contained here so we work on >4GB machines. -DaveM
342 */
343static struct thread_info *cpu_new_thread = NULL;
344
Al Viro8c298902008-11-22 17:34:04 +0000345static int __cpuinit smp_boot_one_cpu(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346{
347 unsigned long entry =
348 (unsigned long)(&sparc64_cpu_startup);
349 unsigned long cookie =
350 (unsigned long)(&cpu_new_thread);
351 struct task_struct *p;
David S. Miller557fe0e2009-03-31 17:15:40 -0700352 void *descr = NULL;
David S. Miller7890f792006-02-15 02:26:54 -0800353 int timeout, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354
355 p = fork_idle(cpu);
Akinobu Mita1177bf92007-10-04 14:55:59 -0700356 if (IS_ERR(p))
357 return PTR_ERR(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 callin_flag = 0;
Al Virof3169642006-01-12 01:05:42 -0800359 cpu_new_thread = task_thread_info(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360
David S. Miller7890f792006-02-15 02:26:54 -0800361 if (tlb_type == hypervisor) {
David S. Millerb14f5c12007-07-14 00:45:16 -0700362#if defined(CONFIG_SUN_LDOMS) && defined(CONFIG_HOTPLUG_CPU)
David S. Miller4f0234f2007-07-13 16:03:42 -0700363 if (ldom_domaining_enabled)
364 ldom_startcpu_cpuid(cpu,
David S. Miller557fe0e2009-03-31 17:15:40 -0700365 (unsigned long) cpu_new_thread,
366 &descr);
David S. Miller4f0234f2007-07-13 16:03:42 -0700367 else
368#endif
369 prom_startcpu_cpuid(cpu, entry, cookie);
David S. Miller7890f792006-02-15 02:26:54 -0800370 } else {
David S. Miller5cbc3072007-05-25 15:49:59 -0700371 struct device_node *dp = of_find_node_by_cpuid(cpu);
David S. Miller7890f792006-02-15 02:26:54 -0800372
David S. Miller07f8e5f2006-06-21 23:34:02 -0700373 prom_startcpu(dp->node, entry, cookie);
David S. Miller7890f792006-02-15 02:26:54 -0800374 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375
David S. Miller4f0234f2007-07-13 16:03:42 -0700376 for (timeout = 0; timeout < 50000; timeout++) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 if (callin_flag)
378 break;
379 udelay(100);
380 }
David S. Miller72aff532006-02-17 01:29:17 -0800381
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 if (callin_flag) {
383 ret = 0;
384 } else {
385 printk("Processor %d is stuck.\n", cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 ret = -ENODEV;
387 }
388 cpu_new_thread = NULL;
389
David S. Miller557fe0e2009-03-31 17:15:40 -0700390 kfree(descr);
David S. Millerb37d40d2007-07-15 01:08:03 -0700391
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 return ret;
393}
394
395static void spitfire_xcall_helper(u64 data0, u64 data1, u64 data2, u64 pstate, unsigned long cpu)
396{
397 u64 result, target;
398 int stuck, tmp;
399
400 if (this_is_starfire) {
401 /* map to real upaid */
402 cpu = (((cpu & 0x3c) << 1) |
403 ((cpu & 0x40) >> 4) |
404 (cpu & 0x3));
405 }
406
407 target = (cpu << 14) | 0x70;
408again:
409 /* Ok, this is the real Spitfire Errata #54.
410 * One must read back from a UDB internal register
411 * after writes to the UDB interrupt dispatch, but
412 * before the membar Sync for that write.
413 * So we use the high UDB control register (ASI 0x7f,
414 * ADDR 0x20) for the dummy read. -DaveM
415 */
416 tmp = 0x40;
417 __asm__ __volatile__(
418 "wrpr %1, %2, %%pstate\n\t"
419 "stxa %4, [%0] %3\n\t"
420 "stxa %5, [%0+%8] %3\n\t"
421 "add %0, %8, %0\n\t"
422 "stxa %6, [%0+%8] %3\n\t"
423 "membar #Sync\n\t"
424 "stxa %%g0, [%7] %3\n\t"
425 "membar #Sync\n\t"
426 "mov 0x20, %%g1\n\t"
427 "ldxa [%%g1] 0x7f, %%g0\n\t"
428 "membar #Sync"
429 : "=r" (tmp)
430 : "r" (pstate), "i" (PSTATE_IE), "i" (ASI_INTR_W),
431 "r" (data0), "r" (data1), "r" (data2), "r" (target),
432 "r" (0x10), "0" (tmp)
433 : "g1");
434
435 /* NOTE: PSTATE_IE is still clear. */
436 stuck = 100000;
437 do {
438 __asm__ __volatile__("ldxa [%%g0] %1, %0"
439 : "=r" (result)
440 : "i" (ASI_INTR_DISPATCH_STAT));
441 if (result == 0) {
442 __asm__ __volatile__("wrpr %0, 0x0, %%pstate"
443 : : "r" (pstate));
444 return;
445 }
446 stuck -= 1;
447 if (stuck == 0)
448 break;
449 } while (result & 0x1);
450 __asm__ __volatile__("wrpr %0, 0x0, %%pstate"
451 : : "r" (pstate));
452 if (stuck == 0) {
Sam Ravnborg90181132009-01-06 13:19:28 -0800453 printk("CPU[%d]: mondo stuckage result[%016llx]\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700454 smp_processor_id(), result);
455 } else {
456 udelay(2);
457 goto again;
458 }
459}
460
David S. Miller90f7ae82008-08-04 16:42:58 -0700461static void spitfire_xcall_deliver(struct trap_per_cpu *tb, int cnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462{
David S. Miller90f7ae82008-08-04 16:42:58 -0700463 u64 *mondo, data0, data1, data2;
464 u16 *cpu_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465 u64 pstate;
466 int i;
467
468 __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
David S. Miller90f7ae82008-08-04 16:42:58 -0700469 cpu_list = __va(tb->cpu_list_pa);
470 mondo = __va(tb->cpu_mondo_block_pa);
471 data0 = mondo[0];
472 data1 = mondo[1];
473 data2 = mondo[2];
474 for (i = 0; i < cnt; i++)
475 spitfire_xcall_helper(data0, data1, data2, pstate, cpu_list[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476}
477
478/* Cheetah now allows to send the whole 64-bytes of data in the interrupt
479 * packet, but we have no use for that. However we do take advantage of
480 * the new pipelining feature (ie. dispatch to multiple cpus simultaneously).
481 */
David S. Miller90f7ae82008-08-04 16:42:58 -0700482static void cheetah_xcall_deliver(struct trap_per_cpu *tb, int cnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
David S. Miller22adb352007-05-26 01:14:43 -0700484 int nack_busy_id, is_jbus, need_more;
David S. Miller90f7ae82008-08-04 16:42:58 -0700485 u64 *mondo, pstate, ver, busy_mask;
486 u16 *cpu_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487
David S. Miller90f7ae82008-08-04 16:42:58 -0700488 cpu_list = __va(tb->cpu_list_pa);
489 mondo = __va(tb->cpu_mondo_block_pa);
David S. Millercd5bc892008-08-03 23:24:26 -0700490
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 /* Unfortunately, someone at Sun had the brilliant idea to make the
492 * busy/nack fields hard-coded by ITID number for this Ultra-III
493 * derivative processor.
494 */
495 __asm__ ("rdpr %%ver, %0" : "=r" (ver));
David S. Miller92704a12006-02-26 23:27:19 -0800496 is_jbus = ((ver >> 32) == __JALAPENO_ID ||
497 (ver >> 32) == __SERRANO_ID);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498
499 __asm__ __volatile__("rdpr %%pstate, %0" : "=r" (pstate));
500
501retry:
David S. Miller22adb352007-05-26 01:14:43 -0700502 need_more = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 __asm__ __volatile__("wrpr %0, %1, %%pstate\n\t"
504 : : "r" (pstate), "i" (PSTATE_IE));
505
506 /* Setup the dispatch data registers. */
507 __asm__ __volatile__("stxa %0, [%3] %6\n\t"
508 "stxa %1, [%4] %6\n\t"
509 "stxa %2, [%5] %6\n\t"
510 "membar #Sync\n\t"
511 : /* no outputs */
David S. Miller90f7ae82008-08-04 16:42:58 -0700512 : "r" (mondo[0]), "r" (mondo[1]), "r" (mondo[2]),
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 "r" (0x40), "r" (0x50), "r" (0x60),
514 "i" (ASI_INTR_W));
515
516 nack_busy_id = 0;
David S. Miller0de56d12007-12-12 07:31:46 -0800517 busy_mask = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700518 {
519 int i;
520
David S. Miller90f7ae82008-08-04 16:42:58 -0700521 for (i = 0; i < cnt; i++) {
522 u64 target, nr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700523
David S. Miller90f7ae82008-08-04 16:42:58 -0700524 nr = cpu_list[i];
525 if (nr == 0xffff)
526 continue;
527
528 target = (nr << 14) | 0x70;
David S. Miller0de56d12007-12-12 07:31:46 -0800529 if (is_jbus) {
David S. Miller90f7ae82008-08-04 16:42:58 -0700530 busy_mask |= (0x1UL << (nr * 2));
David S. Miller0de56d12007-12-12 07:31:46 -0800531 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 target |= (nack_busy_id << 24);
David S. Miller0de56d12007-12-12 07:31:46 -0800533 busy_mask |= (0x1UL <<
534 (nack_busy_id * 2));
535 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700536 __asm__ __volatile__(
537 "stxa %%g0, [%0] %1\n\t"
538 "membar #Sync\n\t"
539 : /* no outputs */
540 : "r" (target), "i" (ASI_INTR_W));
541 nack_busy_id++;
David S. Miller22adb352007-05-26 01:14:43 -0700542 if (nack_busy_id == 32) {
543 need_more = 1;
544 break;
545 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 }
547 }
548
549 /* Now, poll for completion. */
550 {
David S. Miller0de56d12007-12-12 07:31:46 -0800551 u64 dispatch_stat, nack_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552 long stuck;
553
554 stuck = 100000 * nack_busy_id;
David S. Miller0de56d12007-12-12 07:31:46 -0800555 nack_mask = busy_mask << 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 do {
557 __asm__ __volatile__("ldxa [%%g0] %1, %0"
558 : "=r" (dispatch_stat)
559 : "i" (ASI_INTR_DISPATCH_STAT));
David S. Miller0de56d12007-12-12 07:31:46 -0800560 if (!(dispatch_stat & (busy_mask | nack_mask))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 __asm__ __volatile__("wrpr %0, 0x0, %%pstate"
562 : : "r" (pstate));
David S. Miller22adb352007-05-26 01:14:43 -0700563 if (unlikely(need_more)) {
David S. Miller90f7ae82008-08-04 16:42:58 -0700564 int i, this_cnt = 0;
565 for (i = 0; i < cnt; i++) {
566 if (cpu_list[i] == 0xffff)
567 continue;
568 cpu_list[i] = 0xffff;
569 this_cnt++;
570 if (this_cnt == 32)
David S. Miller22adb352007-05-26 01:14:43 -0700571 break;
572 }
573 goto retry;
574 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 return;
576 }
577 if (!--stuck)
578 break;
David S. Miller0de56d12007-12-12 07:31:46 -0800579 } while (dispatch_stat & busy_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580
581 __asm__ __volatile__("wrpr %0, 0x0, %%pstate"
582 : : "r" (pstate));
583
David S. Miller0de56d12007-12-12 07:31:46 -0800584 if (dispatch_stat & busy_mask) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 /* Busy bits will not clear, continue instead
586 * of freezing up on this cpu.
587 */
Sam Ravnborg90181132009-01-06 13:19:28 -0800588 printk("CPU[%d]: mondo stuckage result[%016llx]\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 smp_processor_id(), dispatch_stat);
590 } else {
591 int i, this_busy_nack = 0;
592
593 /* Delay some random time with interrupts enabled
594 * to prevent deadlock.
595 */
596 udelay(2 * nack_busy_id);
597
598 /* Clear out the mask bits for cpus which did not
599 * NACK us.
600 */
David S. Miller90f7ae82008-08-04 16:42:58 -0700601 for (i = 0; i < cnt; i++) {
602 u64 check_mask, nr;
603
604 nr = cpu_list[i];
605 if (nr == 0xffff)
606 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607
David S. Miller92704a12006-02-26 23:27:19 -0800608 if (is_jbus)
David S. Miller90f7ae82008-08-04 16:42:58 -0700609 check_mask = (0x2UL << (2*nr));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 else
611 check_mask = (0x2UL <<
612 this_busy_nack);
613 if ((dispatch_stat & check_mask) == 0)
David S. Miller90f7ae82008-08-04 16:42:58 -0700614 cpu_list[i] = 0xffff;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 this_busy_nack += 2;
David S. Miller22adb352007-05-26 01:14:43 -0700616 if (this_busy_nack == 64)
617 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 }
619
620 goto retry;
621 }
622 }
623}
624
David S. Miller1d2f1f92006-02-08 16:41:20 -0800625/* Multi-cpu list version. */
David S. Miller90f7ae82008-08-04 16:42:58 -0700626static void hypervisor_xcall_deliver(struct trap_per_cpu *tb, int cnt)
David S. Millera43fe0e2006-02-04 03:10:53 -0800627{
David S. Millered4d9c62008-08-04 16:47:57 -0700628 int retries, this_cpu, prev_sent, i, saw_cpu_error;
David S. Millerc02a5112008-08-04 16:18:40 -0700629 unsigned long status;
David S. Millerb830ab62006-02-28 15:10:26 -0800630 u16 *cpu_list;
David S. Miller17f34f02007-05-14 02:01:52 -0700631
David S. Millerb830ab62006-02-28 15:10:26 -0800632 this_cpu = smp_processor_id();
David S. Miller1d2f1f92006-02-08 16:41:20 -0800633
David S. Millerb830ab62006-02-28 15:10:26 -0800634 cpu_list = __va(tb->cpu_list_pa);
635
David S. Millered4d9c62008-08-04 16:47:57 -0700636 saw_cpu_error = 0;
David S. Miller1d2f1f92006-02-08 16:41:20 -0800637 retries = 0;
David S. Miller3cab0c32006-03-02 21:50:47 -0800638 prev_sent = 0;
David S. Miller1d2f1f92006-02-08 16:41:20 -0800639 do {
David S. Miller3cab0c32006-03-02 21:50:47 -0800640 int forward_progress, n_sent;
David S. Miller1d2f1f92006-02-08 16:41:20 -0800641
David S. Millerb830ab62006-02-28 15:10:26 -0800642 status = sun4v_cpu_mondo_send(cnt,
643 tb->cpu_list_pa,
644 tb->cpu_mondo_block_pa);
David S. Miller1d2f1f92006-02-08 16:41:20 -0800645
David S. Millerb830ab62006-02-28 15:10:26 -0800646 /* HV_EOK means all cpus received the xcall, we're done. */
647 if (likely(status == HV_EOK))
David S. Miller1d2f1f92006-02-08 16:41:20 -0800648 break;
649
David S. Miller3cab0c32006-03-02 21:50:47 -0800650 /* First, see if we made any forward progress.
651 *
652 * The hypervisor indicates successful sends by setting
653 * cpu list entries to the value 0xffff.
David S. Millerb830ab62006-02-28 15:10:26 -0800654 */
David S. Miller3cab0c32006-03-02 21:50:47 -0800655 n_sent = 0;
David S. Millerb830ab62006-02-28 15:10:26 -0800656 for (i = 0; i < cnt; i++) {
David S. Miller3cab0c32006-03-02 21:50:47 -0800657 if (likely(cpu_list[i] == 0xffff))
658 n_sent++;
David S. Miller1d2f1f92006-02-08 16:41:20 -0800659 }
660
David S. Miller3cab0c32006-03-02 21:50:47 -0800661 forward_progress = 0;
662 if (n_sent > prev_sent)
663 forward_progress = 1;
664
665 prev_sent = n_sent;
666
David S. Millerb830ab62006-02-28 15:10:26 -0800667 /* If we get a HV_ECPUERROR, then one or more of the cpus
668 * in the list are in error state. Use the cpu_state()
669 * hypervisor call to find out which cpus are in error state.
670 */
671 if (unlikely(status == HV_ECPUERROR)) {
672 for (i = 0; i < cnt; i++) {
673 long err;
674 u16 cpu;
David S. Miller1d2f1f92006-02-08 16:41:20 -0800675
David S. Millerb830ab62006-02-28 15:10:26 -0800676 cpu = cpu_list[i];
677 if (cpu == 0xffff)
678 continue;
679
680 err = sun4v_cpu_state(cpu);
David S. Millered4d9c62008-08-04 16:47:57 -0700681 if (err == HV_CPU_STATE_ERROR) {
682 saw_cpu_error = (cpu + 1);
David S. Miller3cab0c32006-03-02 21:50:47 -0800683 cpu_list[i] = 0xffff;
David S. Millerb830ab62006-02-28 15:10:26 -0800684 }
685 }
686 } else if (unlikely(status != HV_EWOULDBLOCK))
687 goto fatal_mondo_error;
688
David S. Miller3cab0c32006-03-02 21:50:47 -0800689 /* Don't bother rewriting the CPU list, just leave the
690 * 0xffff and non-0xffff entries in there and the
691 * hypervisor will do the right thing.
692 *
693 * Only advance timeout state if we didn't make any
694 * forward progress.
695 */
David S. Millerb830ab62006-02-28 15:10:26 -0800696 if (unlikely(!forward_progress)) {
697 if (unlikely(++retries > 10000))
698 goto fatal_mondo_timeout;
699
700 /* Delay a little bit to let other cpus catch up
701 * on their cpu mondo queue work.
702 */
703 udelay(2 * cnt);
704 }
David S. Miller1d2f1f92006-02-08 16:41:20 -0800705 } while (1);
706
David S. Millered4d9c62008-08-04 16:47:57 -0700707 if (unlikely(saw_cpu_error))
David S. Millerb830ab62006-02-28 15:10:26 -0800708 goto fatal_mondo_cpu_error;
709
710 return;
711
712fatal_mondo_cpu_error:
713 printk(KERN_CRIT "CPU[%d]: SUN4V mondo cpu error, some target cpus "
David S. Millered4d9c62008-08-04 16:47:57 -0700714 "(including %d) were in error state\n",
715 this_cpu, saw_cpu_error - 1);
David S. Millerb830ab62006-02-28 15:10:26 -0800716 return;
717
718fatal_mondo_timeout:
David S. Millerb830ab62006-02-28 15:10:26 -0800719 printk(KERN_CRIT "CPU[%d]: SUN4V mondo timeout, no forward "
720 " progress after %d retries.\n",
721 this_cpu, retries);
722 goto dump_cpu_list_and_out;
723
724fatal_mondo_error:
David S. Millerb830ab62006-02-28 15:10:26 -0800725 printk(KERN_CRIT "CPU[%d]: Unexpected SUN4V mondo error %lu\n",
726 this_cpu, status);
727 printk(KERN_CRIT "CPU[%d]: Args were cnt(%d) cpulist_pa(%lx) "
728 "mondo_block_pa(%lx)\n",
729 this_cpu, cnt, tb->cpu_list_pa, tb->cpu_mondo_block_pa);
730
731dump_cpu_list_and_out:
732 printk(KERN_CRIT "CPU[%d]: CPU list [ ", this_cpu);
733 for (i = 0; i < cnt; i++)
734 printk("%u ", cpu_list[i]);
735 printk("]\n");
David S. Millera43fe0e2006-02-04 03:10:53 -0800736}
737
David S. Miller90f7ae82008-08-04 16:42:58 -0700738static void (*xcall_deliver_impl)(struct trap_per_cpu *, int);
David S. Millerdeb16992008-08-04 16:16:20 -0700739
740static void xcall_deliver(u64 data0, u64 data1, u64 data2, const cpumask_t *mask)
741{
David S. Miller90f7ae82008-08-04 16:42:58 -0700742 struct trap_per_cpu *tb;
743 int this_cpu, i, cnt;
David S. Millerc02a5112008-08-04 16:18:40 -0700744 unsigned long flags;
David S. Miller90f7ae82008-08-04 16:42:58 -0700745 u16 *cpu_list;
746 u64 *mondo;
David S. Millerc02a5112008-08-04 16:18:40 -0700747
748 /* We have to do this whole thing with interrupts fully disabled.
749 * Otherwise if we send an xcall from interrupt context it will
750 * corrupt both our mondo block and cpu list state.
751 *
752 * One consequence of this is that we cannot use timeout mechanisms
753 * that depend upon interrupts being delivered locally. So, for
754 * example, we cannot sample jiffies and expect it to advance.
755 *
756 * Fortunately, udelay() uses %stick/%tick so we can use that.
757 */
758 local_irq_save(flags);
David S. Miller90f7ae82008-08-04 16:42:58 -0700759
760 this_cpu = smp_processor_id();
761 tb = &trap_block[this_cpu];
762
763 mondo = __va(tb->cpu_mondo_block_pa);
764 mondo[0] = data0;
765 mondo[1] = data1;
766 mondo[2] = data2;
767 wmb();
768
769 cpu_list = __va(tb->cpu_list_pa);
770
771 /* Setup the initial cpu list. */
772 cnt = 0;
Rusty Russell8e757282008-12-08 01:10:08 -0800773 for_each_cpu(i, mask) {
David S. Miller90f7ae82008-08-04 16:42:58 -0700774 if (i == this_cpu || !cpu_online(i))
775 continue;
776 cpu_list[cnt++] = i;
777 }
778
779 if (cnt)
780 xcall_deliver_impl(tb, cnt);
781
David S. Millerc02a5112008-08-04 16:18:40 -0700782 local_irq_restore(flags);
David S. Millerdeb16992008-08-04 16:16:20 -0700783}
David S. Miller5e0797e2008-08-03 22:52:41 -0700784
David S. Miller91a42312008-08-04 00:51:18 -0700785/* Send cross call to all processors mentioned in MASK_P
786 * except self. Really, there are only two cases currently,
787 * "&cpu_online_map" and "&mm->cpu_vm_mask".
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 */
David S. Millerae583882008-08-04 16:56:15 -0700789static void smp_cross_call_masked(unsigned long *func, u32 ctx, u64 data1, u64 data2, const cpumask_t *mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790{
791 u64 data0 = (((u64)ctx)<<32 | (((u64)func) & 0xffffffff));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792
David S. Millerae583882008-08-04 16:56:15 -0700793 xcall_deliver(data0, data1, data2, mask);
794}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795
David S. Millerae583882008-08-04 16:56:15 -0700796/* Send cross call to all processors except self. */
797static void smp_cross_call(unsigned long *func, u32 ctx, u64 data1, u64 data2)
798{
799 smp_cross_call_masked(func, ctx, data1, data2, &cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800}
801
802extern unsigned long xcall_sync_tick;
803
804static void smp_start_sync_tick_client(int cpu)
805{
David S. Miller24445a42008-08-04 00:02:31 -0700806 xcall_deliver((u64) &xcall_sync_tick, 0, 0,
807 &cpumask_of_cpu(cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808}
809
David S. Millerd172ad12008-07-17 23:44:50 -0700810extern unsigned long xcall_call_function;
811
Rusty Russellf46df022009-03-16 14:40:22 +1030812void arch_send_call_function_ipi_mask(const struct cpumask *mask)
David S. Millerd172ad12008-07-17 23:44:50 -0700813{
Rusty Russellf46df022009-03-16 14:40:22 +1030814 xcall_deliver((u64) &xcall_call_function, 0, 0, mask);
David S. Millerd172ad12008-07-17 23:44:50 -0700815}
816
817extern unsigned long xcall_call_function_single;
818
819void arch_send_call_function_single_ipi(int cpu)
820{
David S. Miller19926632008-08-03 23:56:28 -0700821 xcall_deliver((u64) &xcall_call_function_single, 0, 0,
822 &cpumask_of_cpu(cpu));
David S. Millerd172ad12008-07-17 23:44:50 -0700823}
824
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825void smp_call_function_client(int irq, struct pt_regs *regs)
826{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 clear_softint(1 << irq);
David S. Millerd172ad12008-07-17 23:44:50 -0700828 generic_smp_call_function_interrupt();
829}
830
831void smp_call_function_single_client(int irq, struct pt_regs *regs)
832{
833 clear_softint(1 << irq);
834 generic_smp_call_function_single_interrupt();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700835}
836
David S. Millerbd407912006-01-31 18:31:38 -0800837static void tsb_sync(void *info)
838{
David S. Miller6f25f392006-03-28 13:29:26 -0800839 struct trap_per_cpu *tp = &trap_block[raw_smp_processor_id()];
David S. Millerbd407912006-01-31 18:31:38 -0800840 struct mm_struct *mm = info;
841
David S. Miller6f25f392006-03-28 13:29:26 -0800842 /* It is not valid to test "currrent->active_mm == mm" here.
843 *
844 * The value of "current" is not changed atomically with
845 * switch_mm(). But that's OK, we just need to check the
846 * current cpu's trap block PGD physical address.
847 */
848 if (tp->pgd_paddr == __pa(mm->pgd))
David S. Millerbd407912006-01-31 18:31:38 -0800849 tsb_context_switch(mm);
850}
851
852void smp_tsb_sync(struct mm_struct *mm)
853{
Rusty Russell81f1adf2009-03-16 14:40:39 +1030854 smp_call_function_many(mm_cpumask(mm), tsb_sync, mm, 1);
David S. Millerbd407912006-01-31 18:31:38 -0800855}
856
Linus Torvalds1da177e2005-04-16 15:20:36 -0700857extern unsigned long xcall_flush_tlb_mm;
858extern unsigned long xcall_flush_tlb_pending;
859extern unsigned long xcall_flush_tlb_kernel_range;
David S. Miller93dae5b2008-05-19 23:46:00 -0700860extern unsigned long xcall_fetch_glob_regs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861extern unsigned long xcall_receive_signal;
David S. Milleree290742006-03-06 22:50:44 -0800862extern unsigned long xcall_new_mmu_context_version;
David S. Millere2fdd7f2008-04-29 02:38:50 -0700863#ifdef CONFIG_KGDB
864extern unsigned long xcall_kgdb_capture;
865#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866
867#ifdef DCACHE_ALIASING_POSSIBLE
868extern unsigned long xcall_flush_dcache_page_cheetah;
869#endif
870extern unsigned long xcall_flush_dcache_page_spitfire;
871
872#ifdef CONFIG_DEBUG_DCFLUSH
873extern atomic_t dcpage_flushes;
874extern atomic_t dcpage_flushes_xcall;
875#endif
876
David S. Millerd979f172007-10-27 00:13:04 -0700877static inline void __local_flush_dcache_page(struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878{
879#ifdef DCACHE_ALIASING_POSSIBLE
880 __flush_dcache_page(page_address(page),
881 ((tlb_type == spitfire) &&
882 page_mapping(page) != NULL));
883#else
884 if (page_mapping(page) != NULL &&
885 tlb_type == spitfire)
886 __flush_icache_page(__pa(page_address(page)));
887#endif
888}
889
890void smp_flush_dcache_page_impl(struct page *page, int cpu)
891{
David S. Millera43fe0e2006-02-04 03:10:53 -0800892 int this_cpu;
893
894 if (tlb_type == hypervisor)
895 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700896
897#ifdef CONFIG_DEBUG_DCFLUSH
898 atomic_inc(&dcpage_flushes);
899#endif
David S. Millera43fe0e2006-02-04 03:10:53 -0800900
901 this_cpu = get_cpu();
902
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 if (cpu == this_cpu) {
904 __local_flush_dcache_page(page);
905 } else if (cpu_online(cpu)) {
906 void *pg_addr = page_address(page);
David S. Miller622824d2008-08-03 23:07:18 -0700907 u64 data0 = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700908
909 if (tlb_type == spitfire) {
David S. Miller622824d2008-08-03 23:07:18 -0700910 data0 = ((u64)&xcall_flush_dcache_page_spitfire);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700911 if (page_mapping(page) != NULL)
912 data0 |= ((u64)1 << 32);
David S. Millera43fe0e2006-02-04 03:10:53 -0800913 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914#ifdef DCACHE_ALIASING_POSSIBLE
David S. Miller622824d2008-08-03 23:07:18 -0700915 data0 = ((u64)&xcall_flush_dcache_page_cheetah);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916#endif
917 }
David S. Miller622824d2008-08-03 23:07:18 -0700918 if (data0) {
919 xcall_deliver(data0, __pa(pg_addr),
David S. Millerae583882008-08-04 16:56:15 -0700920 (u64) pg_addr, &cpumask_of_cpu(cpu));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700921#ifdef CONFIG_DEBUG_DCFLUSH
David S. Miller622824d2008-08-03 23:07:18 -0700922 atomic_inc(&dcpage_flushes_xcall);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923#endif
David S. Miller622824d2008-08-03 23:07:18 -0700924 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 }
926
927 put_cpu();
928}
929
930void flush_dcache_page_all(struct mm_struct *mm, struct page *page)
931{
David S. Miller622824d2008-08-03 23:07:18 -0700932 void *pg_addr;
David S. Millera43fe0e2006-02-04 03:10:53 -0800933 int this_cpu;
David S. Miller622824d2008-08-03 23:07:18 -0700934 u64 data0;
David S. Millera43fe0e2006-02-04 03:10:53 -0800935
936 if (tlb_type == hypervisor)
937 return;
938
939 this_cpu = get_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941#ifdef CONFIG_DEBUG_DCFLUSH
942 atomic_inc(&dcpage_flushes);
943#endif
David S. Miller622824d2008-08-03 23:07:18 -0700944 data0 = 0;
945 pg_addr = page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946 if (tlb_type == spitfire) {
947 data0 = ((u64)&xcall_flush_dcache_page_spitfire);
948 if (page_mapping(page) != NULL)
949 data0 |= ((u64)1 << 32);
David S. Millera43fe0e2006-02-04 03:10:53 -0800950 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951#ifdef DCACHE_ALIASING_POSSIBLE
952 data0 = ((u64)&xcall_flush_dcache_page_cheetah);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953#endif
954 }
David S. Miller622824d2008-08-03 23:07:18 -0700955 if (data0) {
956 xcall_deliver(data0, __pa(pg_addr),
David S. Millerae583882008-08-04 16:56:15 -0700957 (u64) pg_addr, &cpu_online_map);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958#ifdef CONFIG_DEBUG_DCFLUSH
David S. Miller622824d2008-08-03 23:07:18 -0700959 atomic_inc(&dcpage_flushes_xcall);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960#endif
David S. Miller622824d2008-08-03 23:07:18 -0700961 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962 __local_flush_dcache_page(page);
963
964 put_cpu();
965}
966
David S. Milleree290742006-03-06 22:50:44 -0800967void smp_new_mmu_context_version_client(int irq, struct pt_regs *regs)
968{
David S. Millera0663a72006-02-23 14:19:28 -0800969 struct mm_struct *mm;
David S. Milleree290742006-03-06 22:50:44 -0800970 unsigned long flags;
David S. Millera0663a72006-02-23 14:19:28 -0800971
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 clear_softint(1 << irq);
David S. Millera0663a72006-02-23 14:19:28 -0800973
974 /* See if we need to allocate a new TLB context because
975 * the version of the one we are using is now out of date.
976 */
977 mm = current->active_mm;
David S. Milleree290742006-03-06 22:50:44 -0800978 if (unlikely(!mm || (mm == &init_mm)))
979 return;
David S. Millera0663a72006-02-23 14:19:28 -0800980
David S. Milleree290742006-03-06 22:50:44 -0800981 spin_lock_irqsave(&mm->context.lock, flags);
David S. Milleraac0aad2006-02-27 17:56:51 -0800982
David S. Milleree290742006-03-06 22:50:44 -0800983 if (unlikely(!CTX_VALID(mm->context)))
984 get_new_mmu_context(mm);
David S. Milleraac0aad2006-02-27 17:56:51 -0800985
David S. Milleree290742006-03-06 22:50:44 -0800986 spin_unlock_irqrestore(&mm->context.lock, flags);
David S. Milleraac0aad2006-02-27 17:56:51 -0800987
David S. Milleree290742006-03-06 22:50:44 -0800988 load_secondary_context(mm);
989 __flush_tlb_mm(CTX_HWBITS(mm->context),
990 SECONDARY_CONTEXT);
David S. Millera0663a72006-02-23 14:19:28 -0800991}
992
993void smp_new_mmu_context_version(void)
994{
David S. Milleree290742006-03-06 22:50:44 -0800995 smp_cross_call(&xcall_new_mmu_context_version, 0, 0, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996}
997
David S. Millere2fdd7f2008-04-29 02:38:50 -0700998#ifdef CONFIG_KGDB
999void kgdb_roundup_cpus(unsigned long flags)
1000{
1001 smp_cross_call(&xcall_kgdb_capture, 0, 0, 0);
1002}
1003#endif
1004
David S. Miller93dae5b2008-05-19 23:46:00 -07001005void smp_fetch_global_regs(void)
1006{
1007 smp_cross_call(&xcall_fetch_glob_regs, 0, 0, 0);
1008}
David S. Miller93dae5b2008-05-19 23:46:00 -07001009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010/* We know that the window frames of the user have been flushed
1011 * to the stack before we get here because all callers of us
1012 * are flush_tlb_*() routines, and these run after flush_cache_*()
1013 * which performs the flushw.
1014 *
1015 * The SMP TLB coherency scheme we use works as follows:
1016 *
1017 * 1) mm->cpu_vm_mask is a bit mask of which cpus an address
1018 * space has (potentially) executed on, this is the heuristic
1019 * we use to avoid doing cross calls.
1020 *
1021 * Also, for flushing from kswapd and also for clones, we
1022 * use cpu_vm_mask as the list of cpus to make run the TLB.
1023 *
1024 * 2) TLB context numbers are shared globally across all processors
1025 * in the system, this allows us to play several games to avoid
1026 * cross calls.
1027 *
1028 * One invariant is that when a cpu switches to a process, and
1029 * that processes tsk->active_mm->cpu_vm_mask does not have the
1030 * current cpu's bit set, that tlb context is flushed locally.
1031 *
1032 * If the address space is non-shared (ie. mm->count == 1) we avoid
1033 * cross calls when we want to flush the currently running process's
1034 * tlb state. This is done by clearing all cpu bits except the current
David S. Millerf9384d42009-03-27 01:09:17 -07001035 * processor's in current->mm->cpu_vm_mask and performing the
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036 * flush locally only. This will force any subsequent cpus which run
1037 * this task to flush the context from the local tlb if the process
1038 * migrates to another cpu (again).
1039 *
1040 * 3) For shared address spaces (threads) and swapping we bite the
1041 * bullet for most cases and perform the cross call (but only to
1042 * the cpus listed in cpu_vm_mask).
1043 *
1044 * The performance gain from "optimizing" away the cross call for threads is
1045 * questionable (in theory the big win for threads is the massive sharing of
1046 * address space state across processors).
1047 */
David S. Miller62dbec72005-11-07 14:09:58 -08001048
1049/* This currently is only used by the hugetlb arch pre-fault
1050 * hook on UltraSPARC-III+ and later when changing the pagesize
1051 * bits of the context register for an address space.
1052 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053void smp_flush_tlb_mm(struct mm_struct *mm)
1054{
David S. Miller62dbec72005-11-07 14:09:58 -08001055 u32 ctx = CTX_HWBITS(mm->context);
1056 int cpu = get_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
David S. Miller62dbec72005-11-07 14:09:58 -08001058 if (atomic_read(&mm->mm_users) == 1) {
Rusty Russell81f1adf2009-03-16 14:40:39 +10301059 cpumask_copy(mm_cpumask(mm), cpumask_of(cpu));
David S. Miller62dbec72005-11-07 14:09:58 -08001060 goto local_flush_and_out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 }
David S. Miller62dbec72005-11-07 14:09:58 -08001062
1063 smp_cross_call_masked(&xcall_flush_tlb_mm,
1064 ctx, 0, 0,
Rusty Russell81f1adf2009-03-16 14:40:39 +10301065 mm_cpumask(mm));
David S. Miller62dbec72005-11-07 14:09:58 -08001066
1067local_flush_and_out:
1068 __flush_tlb_mm(ctx, SECONDARY_CONTEXT);
1069
1070 put_cpu();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071}
1072
1073void smp_flush_tlb_pending(struct mm_struct *mm, unsigned long nr, unsigned long *vaddrs)
1074{
1075 u32 ctx = CTX_HWBITS(mm->context);
1076 int cpu = get_cpu();
1077
David S. Millerf9384d42009-03-27 01:09:17 -07001078 if (mm == current->mm && atomic_read(&mm->mm_users) == 1)
Rusty Russell81f1adf2009-03-16 14:40:39 +10301079 cpumask_copy(mm_cpumask(mm), cpumask_of(cpu));
Hugh Dickinsdedeb002005-11-07 14:09:01 -08001080 else
1081 smp_cross_call_masked(&xcall_flush_tlb_pending,
1082 ctx, nr, (unsigned long) vaddrs,
Rusty Russell81f1adf2009-03-16 14:40:39 +10301083 mm_cpumask(mm));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085 __flush_tlb_pending(ctx, nr, vaddrs);
1086
1087 put_cpu();
1088}
1089
1090void smp_flush_tlb_kernel_range(unsigned long start, unsigned long end)
1091{
1092 start &= PAGE_MASK;
1093 end = PAGE_ALIGN(end);
1094 if (start != end) {
1095 smp_cross_call(&xcall_flush_tlb_kernel_range,
1096 0, start, end);
1097
1098 __flush_tlb_kernel_range(start, end);
1099 }
1100}
1101
1102/* CPU capture. */
1103/* #define CAPTURE_DEBUG */
1104extern unsigned long xcall_capture;
1105
1106static atomic_t smp_capture_depth = ATOMIC_INIT(0);
1107static atomic_t smp_capture_registry = ATOMIC_INIT(0);
1108static unsigned long penguins_are_doing_time;
1109
1110void smp_capture(void)
1111{
1112 int result = atomic_add_ret(1, &smp_capture_depth);
1113
1114 if (result == 1) {
1115 int ncpus = num_online_cpus();
1116
1117#ifdef CAPTURE_DEBUG
1118 printk("CPU[%d]: Sending penguins to jail...",
1119 smp_processor_id());
1120#endif
1121 penguins_are_doing_time = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122 atomic_inc(&smp_capture_registry);
1123 smp_cross_call(&xcall_capture, 0, 0, 0);
1124 while (atomic_read(&smp_capture_registry) != ncpus)
David S. Miller4f071182005-08-29 12:46:22 -07001125 rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126#ifdef CAPTURE_DEBUG
1127 printk("done\n");
1128#endif
1129 }
1130}
1131
1132void smp_release(void)
1133{
1134 if (atomic_dec_and_test(&smp_capture_depth)) {
1135#ifdef CAPTURE_DEBUG
1136 printk("CPU[%d]: Giving pardon to "
1137 "imprisoned penguins\n",
1138 smp_processor_id());
1139#endif
1140 penguins_are_doing_time = 0;
David S. Miller293666b2008-11-15 13:33:25 -08001141 membar_safe("#StoreLoad");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 atomic_dec(&smp_capture_registry);
1143 }
1144}
1145
David S. Millerb4f43722008-11-23 21:55:29 -08001146/* Imprisoned penguins run with %pil == PIL_NORMAL_MAX, but PSTATE_IE
1147 * set, so they can service tlb flush xcalls...
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 */
1149extern void prom_world(int);
David S. Miller96c6e0d2006-01-31 18:32:29 -08001150
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151void smp_penguin_jailcell(int irq, struct pt_regs *regs)
1152{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 clear_softint(1 << irq);
1154
1155 preempt_disable();
1156
1157 __asm__ __volatile__("flushw");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 prom_world(1);
1159 atomic_inc(&smp_capture_registry);
David S. Miller293666b2008-11-15 13:33:25 -08001160 membar_safe("#StoreLoad");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 while (penguins_are_doing_time)
David S. Miller4f071182005-08-29 12:46:22 -07001162 rmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 atomic_dec(&smp_capture_registry);
1164 prom_world(0);
1165
1166 preempt_enable();
1167}
1168
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169/* /proc/profile writes can call this, don't __init it please. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001170int setup_profiling_timer(unsigned int multiplier)
1171{
David S. Miller777a4472007-02-22 06:24:10 -08001172 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173}
1174
1175void __init smp_prepare_cpus(unsigned int max_cpus)
1176{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177}
1178
1179void __devinit smp_prepare_boot_cpu(void)
1180{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
1182
David S. Miller5e0797e2008-08-03 22:52:41 -07001183void __init smp_setup_processor_id(void)
1184{
1185 if (tlb_type == spitfire)
David S. Millerdeb16992008-08-04 16:16:20 -07001186 xcall_deliver_impl = spitfire_xcall_deliver;
David S. Miller5e0797e2008-08-03 22:52:41 -07001187 else if (tlb_type == cheetah || tlb_type == cheetah_plus)
David S. Millerdeb16992008-08-04 16:16:20 -07001188 xcall_deliver_impl = cheetah_xcall_deliver;
David S. Miller5e0797e2008-08-03 22:52:41 -07001189 else
David S. Millerdeb16992008-08-04 16:16:20 -07001190 xcall_deliver_impl = hypervisor_xcall_deliver;
David S. Miller5e0797e2008-08-03 22:52:41 -07001191}
1192
David S. Miller5cbc3072007-05-25 15:49:59 -07001193void __devinit smp_fill_in_sib_core_maps(void)
1194{
1195 unsigned int i;
1196
David S. Millere0204402007-07-16 03:49:40 -07001197 for_each_present_cpu(i) {
David S. Miller5cbc3072007-05-25 15:49:59 -07001198 unsigned int j;
1199
David S. Miller39dd9922007-07-15 01:29:24 -07001200 cpus_clear(cpu_core_map[i]);
David S. Miller5cbc3072007-05-25 15:49:59 -07001201 if (cpu_data(i).core_id == 0) {
David S. Millerf78eae22007-06-04 17:01:39 -07001202 cpu_set(i, cpu_core_map[i]);
David S. Miller5cbc3072007-05-25 15:49:59 -07001203 continue;
1204 }
1205
David S. Millere0204402007-07-16 03:49:40 -07001206 for_each_present_cpu(j) {
David S. Miller5cbc3072007-05-25 15:49:59 -07001207 if (cpu_data(i).core_id ==
1208 cpu_data(j).core_id)
David S. Millerf78eae22007-06-04 17:01:39 -07001209 cpu_set(j, cpu_core_map[i]);
1210 }
1211 }
1212
David S. Millere0204402007-07-16 03:49:40 -07001213 for_each_present_cpu(i) {
David S. Millerf78eae22007-06-04 17:01:39 -07001214 unsigned int j;
1215
Mike Travisd5a74302007-10-16 01:24:05 -07001216 cpus_clear(per_cpu(cpu_sibling_map, i));
David S. Millerf78eae22007-06-04 17:01:39 -07001217 if (cpu_data(i).proc_id == -1) {
Mike Travisd5a74302007-10-16 01:24:05 -07001218 cpu_set(i, per_cpu(cpu_sibling_map, i));
David S. Millerf78eae22007-06-04 17:01:39 -07001219 continue;
1220 }
1221
David S. Millere0204402007-07-16 03:49:40 -07001222 for_each_present_cpu(j) {
David S. Millerf78eae22007-06-04 17:01:39 -07001223 if (cpu_data(i).proc_id ==
1224 cpu_data(j).proc_id)
Mike Travisd5a74302007-10-16 01:24:05 -07001225 cpu_set(j, per_cpu(cpu_sibling_map, i));
David S. Miller5cbc3072007-05-25 15:49:59 -07001226 }
1227 }
1228}
1229
Gautham R Shenoyb282b6f2007-01-10 23:15:34 -08001230int __cpuinit __cpu_up(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231{
1232 int ret = smp_boot_one_cpu(cpu);
1233
1234 if (!ret) {
1235 cpu_set(cpu, smp_commenced_mask);
1236 while (!cpu_isset(cpu, cpu_online_map))
1237 mb();
1238 if (!cpu_isset(cpu, cpu_online_map)) {
1239 ret = -ENODEV;
1240 } else {
David S. Miller02fead72006-02-11 23:22:47 -08001241 /* On SUN4V, writes to %tick and %stick are
1242 * not allowed.
1243 */
1244 if (tlb_type != hypervisor)
1245 smp_synchronize_one_tick(cpu);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 }
1247 }
1248 return ret;
1249}
1250
David S. Miller4f0234f2007-07-13 16:03:42 -07001251#ifdef CONFIG_HOTPLUG_CPU
David S. Millere0204402007-07-16 03:49:40 -07001252void cpu_play_dead(void)
1253{
1254 int cpu = smp_processor_id();
1255 unsigned long pstate;
1256
1257 idle_task_exit();
1258
1259 if (tlb_type == hypervisor) {
1260 struct trap_per_cpu *tb = &trap_block[cpu];
1261
1262 sun4v_cpu_qconf(HV_CPU_QUEUE_CPU_MONDO,
1263 tb->cpu_mondo_pa, 0);
1264 sun4v_cpu_qconf(HV_CPU_QUEUE_DEVICE_MONDO,
1265 tb->dev_mondo_pa, 0);
1266 sun4v_cpu_qconf(HV_CPU_QUEUE_RES_ERROR,
1267 tb->resum_mondo_pa, 0);
1268 sun4v_cpu_qconf(HV_CPU_QUEUE_NONRES_ERROR,
1269 tb->nonresum_mondo_pa, 0);
1270 }
1271
1272 cpu_clear(cpu, smp_commenced_mask);
1273 membar_safe("#Sync");
1274
1275 local_irq_disable();
1276
1277 __asm__ __volatile__(
1278 "rdpr %%pstate, %0\n\t"
1279 "wrpr %0, %1, %%pstate"
1280 : "=r" (pstate)
1281 : "i" (PSTATE_IE));
1282
1283 while (1)
1284 barrier();
1285}
1286
David S. Miller4f0234f2007-07-13 16:03:42 -07001287int __cpu_disable(void)
1288{
David S. Millere0204402007-07-16 03:49:40 -07001289 int cpu = smp_processor_id();
1290 cpuinfo_sparc *c;
1291 int i;
1292
1293 for_each_cpu_mask(i, cpu_core_map[cpu])
1294 cpu_clear(cpu, cpu_core_map[i]);
1295 cpus_clear(cpu_core_map[cpu]);
1296
Mike Travisd5a74302007-10-16 01:24:05 -07001297 for_each_cpu_mask(i, per_cpu(cpu_sibling_map, cpu))
1298 cpu_clear(cpu, per_cpu(cpu_sibling_map, i));
1299 cpus_clear(per_cpu(cpu_sibling_map, cpu));
David S. Millere0204402007-07-16 03:49:40 -07001300
1301 c = &cpu_data(cpu);
1302
1303 c->core_id = 0;
1304 c->proc_id = -1;
1305
David S. Millere0204402007-07-16 03:49:40 -07001306 smp_wmb();
1307
1308 /* Make sure no interrupts point to this cpu. */
1309 fixup_irqs();
1310
1311 local_irq_enable();
1312 mdelay(1);
1313 local_irq_disable();
1314
Paul E. McKenney4d084612008-09-03 02:15:30 -07001315 ipi_call_lock();
1316 cpu_clear(cpu, cpu_online_map);
1317 ipi_call_unlock();
1318
Hong H. Pham280ff972009-06-04 02:10:11 -07001319 cpu_map_rebuild();
1320
David S. Millere0204402007-07-16 03:49:40 -07001321 return 0;
David S. Miller4f0234f2007-07-13 16:03:42 -07001322}
1323
1324void __cpu_die(unsigned int cpu)
1325{
David S. Millere0204402007-07-16 03:49:40 -07001326 int i;
1327
1328 for (i = 0; i < 100; i++) {
1329 smp_rmb();
1330 if (!cpu_isset(cpu, smp_commenced_mask))
1331 break;
1332 msleep(100);
1333 }
1334 if (cpu_isset(cpu, smp_commenced_mask)) {
1335 printk(KERN_ERR "CPU %u didn't die...\n", cpu);
1336 } else {
1337#if defined(CONFIG_SUN_LDOMS)
1338 unsigned long hv_err;
1339 int limit = 100;
1340
1341 do {
1342 hv_err = sun4v_cpu_stop(cpu);
1343 if (hv_err == HV_EOK) {
1344 cpu_clear(cpu, cpu_present_map);
1345 break;
1346 }
1347 } while (--limit > 0);
1348 if (limit <= 0) {
1349 printk(KERN_ERR "sun4v_cpu_stop() fails err=%lu\n",
1350 hv_err);
1351 }
1352#endif
1353 }
David S. Miller4f0234f2007-07-13 16:03:42 -07001354}
1355#endif
1356
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357void __init smp_cpus_done(unsigned int max_cpus)
1358{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001359}
1360
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361void smp_send_reschedule(int cpu)
1362{
David S. Miller19926632008-08-03 23:56:28 -07001363 xcall_deliver((u64) &xcall_receive_signal, 0, 0,
1364 &cpumask_of_cpu(cpu));
1365}
1366
1367void smp_receive_signal_client(int irq, struct pt_regs *regs)
1368{
1369 clear_softint(1 << irq);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370}
1371
1372/* This is a nop because we capture all other cpus
1373 * anyways when making the PROM active.
1374 */
1375void smp_send_stop(void)
1376{
1377}
1378
David S. Miller4fd78a52009-04-08 20:32:02 -07001379/**
1380 * pcpu_alloc_bootmem - NUMA friendly alloc_bootmem wrapper for percpu
1381 * @cpu: cpu to allocate for
1382 * @size: size allocation in bytes
1383 * @align: alignment
1384 *
1385 * Allocate @size bytes aligned at @align for cpu @cpu. This wrapper
1386 * does the right thing for NUMA regardless of the current
1387 * configuration.
1388 *
1389 * RETURNS:
1390 * Pointer to the allocated area on success, NULL on failure.
1391 */
Tejun Heobcb21072009-08-14 15:00:53 +09001392static void * __init pcpu_alloc_bootmem(unsigned int cpu, size_t size,
1393 size_t align)
David S. Miller4fd78a52009-04-08 20:32:02 -07001394{
1395 const unsigned long goal = __pa(MAX_DMA_ADDRESS);
1396#ifdef CONFIG_NEED_MULTIPLE_NODES
1397 int node = cpu_to_node(cpu);
1398 void *ptr;
1399
1400 if (!node_online(node) || !NODE_DATA(node)) {
1401 ptr = __alloc_bootmem(size, align, goal);
1402 pr_info("cpu %d has no node %d or node-local memory\n",
1403 cpu, node);
1404 pr_debug("per cpu data for cpu%d %lu bytes at %016lx\n",
1405 cpu, size, __pa(ptr));
1406 } else {
1407 ptr = __alloc_bootmem_node(NODE_DATA(node),
1408 size, align, goal);
1409 pr_debug("per cpu data for cpu%d %lu bytes on node%d at "
1410 "%016lx\n", cpu, size, node, __pa(ptr));
1411 }
1412 return ptr;
1413#else
1414 return __alloc_bootmem(size, align, goal);
1415#endif
1416}
1417
Tejun Heobcb21072009-08-14 15:00:53 +09001418static void __init pcpu_free_bootmem(void *ptr, size_t size)
David S. Miller4fd78a52009-04-08 20:32:02 -07001419{
Tejun Heobcb21072009-08-14 15:00:53 +09001420 free_bootmem(__pa(ptr), size);
1421}
David S. Miller4fd78a52009-04-08 20:32:02 -07001422
Tejun Heoa70c6912009-09-24 18:18:55 +09001423static int __init pcpu_cpu_distance(unsigned int from, unsigned int to)
Tejun Heobcb21072009-08-14 15:00:53 +09001424{
1425 if (cpu_to_node(from) == cpu_to_node(to))
1426 return LOCAL_DISTANCE;
1427 else
1428 return REMOTE_DISTANCE;
David S. Miller4fd78a52009-04-08 20:32:02 -07001429}
1430
Tejun Heoa70c6912009-09-24 18:18:55 +09001431static void __init pcpu_populate_pte(unsigned long addr)
1432{
1433 pgd_t *pgd = pgd_offset_k(addr);
1434 pud_t *pud;
1435 pmd_t *pmd;
1436
1437 pud = pud_offset(pgd, addr);
1438 if (pud_none(*pud)) {
1439 pmd_t *new;
1440
1441 new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
1442 pud_populate(&init_mm, pud, new);
1443 }
1444
1445 pmd = pmd_offset(pud, addr);
1446 if (!pmd_present(*pmd)) {
1447 pte_t *new;
1448
1449 new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
1450 pmd_populate_kernel(&init_mm, pmd, new);
1451 }
1452}
1453
David S. Miller73fffc02009-04-01 16:15:20 -07001454void __init setup_per_cpu_areas(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455{
Tejun Heobcb21072009-08-14 15:00:53 +09001456 unsigned long delta;
1457 unsigned int cpu;
Tejun Heoa70c6912009-09-24 18:18:55 +09001458 int rc = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459
Tejun Heoa70c6912009-09-24 18:18:55 +09001460 if (pcpu_chosen_fc != PCPU_FC_PAGE) {
1461 rc = pcpu_embed_first_chunk(PERCPU_MODULE_RESERVE,
1462 PERCPU_DYNAMIC_RESERVE, 4 << 20,
1463 pcpu_cpu_distance,
1464 pcpu_alloc_bootmem,
1465 pcpu_free_bootmem);
1466 if (rc)
1467 pr_warning("PERCPU: %s allocator failed (%d), "
1468 "falling back to page size\n",
1469 pcpu_fc_names[pcpu_chosen_fc], rc);
1470 }
1471 if (rc < 0)
1472 rc = pcpu_page_first_chunk(PERCPU_MODULE_RESERVE,
1473 pcpu_alloc_bootmem,
1474 pcpu_free_bootmem,
1475 pcpu_populate_pte);
1476 if (rc < 0)
1477 panic("cannot initialize percpu area (err=%d)", rc);
David S. Miller4fd78a52009-04-08 20:32:02 -07001478
1479 delta = (unsigned long)pcpu_base_addr - (unsigned long)__per_cpu_start;
Tejun Heofb435d52009-08-14 15:00:51 +09001480 for_each_possible_cpu(cpu)
1481 __per_cpu_offset(cpu) = delta + pcpu_unit_offsets[cpu];
David S. Miller951bc822006-05-31 01:24:02 -07001482
1483 /* Setup %g5 for the boot cpu. */
1484 __local_per_cpu_offset = __per_cpu_offset(smp_processor_id());
David S. Millerb696fdc2009-05-26 22:37:25 -07001485
1486 of_fill_in_cpu_data();
1487 if (tlb_type == hypervisor)
Stephen Rothwell6ac5c612009-06-15 03:06:18 -07001488 mdesc_fill_in_cpu_data(cpu_all_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489}