1/*
2 * Copyright (C) 2010, 2011, 2012, Lemote, Inc.
3 * Author: Chen Huacai, chenhc@lemote.com
4 *
5 * This program is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU General Public License
7 * as published by the Free Software Foundation; either version 2
8 * of the License, or (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
13 * GNU General Public License for more details.
14 *
15 */
16
17#include <linux/init.h>
18#include <linux/cpu.h>
19#include <linux/sched.h>
20#include <linux/smp.h>
21#include <linux/cpufreq.h>
22#include <asm/processor.h>
23#include <asm/time.h>
24#include <asm/clock.h>
25#include <asm/tlbflush.h>
26#include <asm/cacheflush.h>
27#include <loongson.h>
28#include <workarounds.h>
29
30#include "smp.h"
31
32DEFINE_PER_CPU(int, cpu_state);
33DEFINE_PER_CPU(uint32_t, core0_c0count);
34
35static void *ipi_set0_regs[16];
36static void *ipi_clear0_regs[16];
37static void *ipi_status0_regs[16];
38static void *ipi_en0_regs[16];
39static void *ipi_mailbox_buf[16];
40
41/* read a 32bit value from ipi register */
42#define loongson3_ipi_read32(addr) readl(addr)
43/* read a 64bit value from ipi register */
44#define loongson3_ipi_read64(addr) readq(addr)
45/* write a 32bit value to ipi register */
46#define loongson3_ipi_write32(action, addr)	\
47	do {					\
48		writel(action, addr);		\
49		__wbflush();			\
50	} while (0)
51/* write a 64bit value to ipi register */
52#define loongson3_ipi_write64(action, addr)	\
53	do {					\
54		writeq(action, addr);		\
55		__wbflush();			\
56	} while (0)
57
58static void ipi_set0_regs_init(void)
59{
60	ipi_set0_regs[0] = (void *)
61		(SMP_CORE_GROUP0_BASE + SMP_CORE0_OFFSET + SET0);
62	ipi_set0_regs[1] = (void *)
63		(SMP_CORE_GROUP0_BASE + SMP_CORE1_OFFSET + SET0);
64	ipi_set0_regs[2] = (void *)
65		(SMP_CORE_GROUP0_BASE + SMP_CORE2_OFFSET + SET0);
66	ipi_set0_regs[3] = (void *)
67		(SMP_CORE_GROUP0_BASE + SMP_CORE3_OFFSET + SET0);
68	ipi_set0_regs[4] = (void *)
69		(SMP_CORE_GROUP1_BASE + SMP_CORE0_OFFSET + SET0);
70	ipi_set0_regs[5] = (void *)
71		(SMP_CORE_GROUP1_BASE + SMP_CORE1_OFFSET + SET0);
72	ipi_set0_regs[6] = (void *)
73		(SMP_CORE_GROUP1_BASE + SMP_CORE2_OFFSET + SET0);
74	ipi_set0_regs[7] = (void *)
75		(SMP_CORE_GROUP1_BASE + SMP_CORE3_OFFSET + SET0);
76	ipi_set0_regs[8] = (void *)
77		(SMP_CORE_GROUP2_BASE + SMP_CORE0_OFFSET + SET0);
78	ipi_set0_regs[9] = (void *)
79		(SMP_CORE_GROUP2_BASE + SMP_CORE1_OFFSET + SET0);
80	ipi_set0_regs[10] = (void *)
81		(SMP_CORE_GROUP2_BASE + SMP_CORE2_OFFSET + SET0);
82	ipi_set0_regs[11] = (void *)
83		(SMP_CORE_GROUP2_BASE + SMP_CORE3_OFFSET + SET0);
84	ipi_set0_regs[12] = (void *)
85		(SMP_CORE_GROUP3_BASE + SMP_CORE0_OFFSET + SET0);
86	ipi_set0_regs[13] = (void *)
87		(SMP_CORE_GROUP3_BASE + SMP_CORE1_OFFSET + SET0);
88	ipi_set0_regs[14] = (void *)
89		(SMP_CORE_GROUP3_BASE + SMP_CORE2_OFFSET + SET0);
90	ipi_set0_regs[15] = (void *)
91		(SMP_CORE_GROUP3_BASE + SMP_CORE3_OFFSET + SET0);
92}
93
94static void ipi_clear0_regs_init(void)
95{
96	ipi_clear0_regs[0] = (void *)
97		(SMP_CORE_GROUP0_BASE + SMP_CORE0_OFFSET + CLEAR0);
98	ipi_clear0_regs[1] = (void *)
99		(SMP_CORE_GROUP0_BASE + SMP_CORE1_OFFSET + CLEAR0);
100	ipi_clear0_regs[2] = (void *)
101		(SMP_CORE_GROUP0_BASE + SMP_CORE2_OFFSET + CLEAR0);
102	ipi_clear0_regs[3] = (void *)
103		(SMP_CORE_GROUP0_BASE + SMP_CORE3_OFFSET + CLEAR0);
104	ipi_clear0_regs[4] = (void *)
105		(SMP_CORE_GROUP1_BASE + SMP_CORE0_OFFSET + CLEAR0);
106	ipi_clear0_regs[5] = (void *)
107		(SMP_CORE_GROUP1_BASE + SMP_CORE1_OFFSET + CLEAR0);
108	ipi_clear0_regs[6] = (void *)
109		(SMP_CORE_GROUP1_BASE + SMP_CORE2_OFFSET + CLEAR0);
110	ipi_clear0_regs[7] = (void *)
111		(SMP_CORE_GROUP1_BASE + SMP_CORE3_OFFSET + CLEAR0);
112	ipi_clear0_regs[8] = (void *)
113		(SMP_CORE_GROUP2_BASE + SMP_CORE0_OFFSET + CLEAR0);
114	ipi_clear0_regs[9] = (void *)
115		(SMP_CORE_GROUP2_BASE + SMP_CORE1_OFFSET + CLEAR0);
116	ipi_clear0_regs[10] = (void *)
117		(SMP_CORE_GROUP2_BASE + SMP_CORE2_OFFSET + CLEAR0);
118	ipi_clear0_regs[11] = (void *)
119		(SMP_CORE_GROUP2_BASE + SMP_CORE3_OFFSET + CLEAR0);
120	ipi_clear0_regs[12] = (void *)
121		(SMP_CORE_GROUP3_BASE + SMP_CORE0_OFFSET + CLEAR0);
122	ipi_clear0_regs[13] = (void *)
123		(SMP_CORE_GROUP3_BASE + SMP_CORE1_OFFSET + CLEAR0);
124	ipi_clear0_regs[14] = (void *)
125		(SMP_CORE_GROUP3_BASE + SMP_CORE2_OFFSET + CLEAR0);
126	ipi_clear0_regs[15] = (void *)
127		(SMP_CORE_GROUP3_BASE + SMP_CORE3_OFFSET + CLEAR0);
128}
129
130static void ipi_status0_regs_init(void)
131{
132	ipi_status0_regs[0] = (void *)
133		(SMP_CORE_GROUP0_BASE + SMP_CORE0_OFFSET + STATUS0);
134	ipi_status0_regs[1] = (void *)
135		(SMP_CORE_GROUP0_BASE + SMP_CORE1_OFFSET + STATUS0);
136	ipi_status0_regs[2] = (void *)
137		(SMP_CORE_GROUP0_BASE + SMP_CORE2_OFFSET + STATUS0);
138	ipi_status0_regs[3] = (void *)
139		(SMP_CORE_GROUP0_BASE + SMP_CORE3_OFFSET + STATUS0);
140	ipi_status0_regs[4] = (void *)
141		(SMP_CORE_GROUP1_BASE + SMP_CORE0_OFFSET + STATUS0);
142	ipi_status0_regs[5] = (void *)
143		(SMP_CORE_GROUP1_BASE + SMP_CORE1_OFFSET + STATUS0);
144	ipi_status0_regs[6] = (void *)
145		(SMP_CORE_GROUP1_BASE + SMP_CORE2_OFFSET + STATUS0);
146	ipi_status0_regs[7] = (void *)
147		(SMP_CORE_GROUP1_BASE + SMP_CORE3_OFFSET + STATUS0);
148	ipi_status0_regs[8] = (void *)
149		(SMP_CORE_GROUP2_BASE + SMP_CORE0_OFFSET + STATUS0);
150	ipi_status0_regs[9] = (void *)
151		(SMP_CORE_GROUP2_BASE + SMP_CORE1_OFFSET + STATUS0);
152	ipi_status0_regs[10] = (void *)
153		(SMP_CORE_GROUP2_BASE + SMP_CORE2_OFFSET + STATUS0);
154	ipi_status0_regs[11] = (void *)
155		(SMP_CORE_GROUP2_BASE + SMP_CORE3_OFFSET + STATUS0);
156	ipi_status0_regs[12] = (void *)
157		(SMP_CORE_GROUP3_BASE + SMP_CORE0_OFFSET + STATUS0);
158	ipi_status0_regs[13] = (void *)
159		(SMP_CORE_GROUP3_BASE + SMP_CORE1_OFFSET + STATUS0);
160	ipi_status0_regs[14] = (void *)
161		(SMP_CORE_GROUP3_BASE + SMP_CORE2_OFFSET + STATUS0);
162	ipi_status0_regs[15] = (void *)
163		(SMP_CORE_GROUP3_BASE + SMP_CORE3_OFFSET + STATUS0);
164}
165
166static void ipi_en0_regs_init(void)
167{
168	ipi_en0_regs[0] = (void *)
169		(SMP_CORE_GROUP0_BASE + SMP_CORE0_OFFSET + EN0);
170	ipi_en0_regs[1] = (void *)
171		(SMP_CORE_GROUP0_BASE + SMP_CORE1_OFFSET + EN0);
172	ipi_en0_regs[2] = (void *)
173		(SMP_CORE_GROUP0_BASE + SMP_CORE2_OFFSET + EN0);
174	ipi_en0_regs[3] = (void *)
175		(SMP_CORE_GROUP0_BASE + SMP_CORE3_OFFSET + EN0);
176	ipi_en0_regs[4] = (void *)
177		(SMP_CORE_GROUP1_BASE + SMP_CORE0_OFFSET + EN0);
178	ipi_en0_regs[5] = (void *)
179		(SMP_CORE_GROUP1_BASE + SMP_CORE1_OFFSET + EN0);
180	ipi_en0_regs[6] = (void *)
181		(SMP_CORE_GROUP1_BASE + SMP_CORE2_OFFSET + EN0);
182	ipi_en0_regs[7] = (void *)
183		(SMP_CORE_GROUP1_BASE + SMP_CORE3_OFFSET + EN0);
184	ipi_en0_regs[8] = (void *)
185		(SMP_CORE_GROUP2_BASE + SMP_CORE0_OFFSET + EN0);
186	ipi_en0_regs[9] = (void *)
187		(SMP_CORE_GROUP2_BASE + SMP_CORE1_OFFSET + EN0);
188	ipi_en0_regs[10] = (void *)
189		(SMP_CORE_GROUP2_BASE + SMP_CORE2_OFFSET + EN0);
190	ipi_en0_regs[11] = (void *)
191		(SMP_CORE_GROUP2_BASE + SMP_CORE3_OFFSET + EN0);
192	ipi_en0_regs[12] = (void *)
193		(SMP_CORE_GROUP3_BASE + SMP_CORE0_OFFSET + EN0);
194	ipi_en0_regs[13] = (void *)
195		(SMP_CORE_GROUP3_BASE + SMP_CORE1_OFFSET + EN0);
196	ipi_en0_regs[14] = (void *)
197		(SMP_CORE_GROUP3_BASE + SMP_CORE2_OFFSET + EN0);
198	ipi_en0_regs[15] = (void *)
199		(SMP_CORE_GROUP3_BASE + SMP_CORE3_OFFSET + EN0);
200}
201
202static void ipi_mailbox_buf_init(void)
203{
204	ipi_mailbox_buf[0] = (void *)
205		(SMP_CORE_GROUP0_BASE + SMP_CORE0_OFFSET + BUF);
206	ipi_mailbox_buf[1] = (void *)
207		(SMP_CORE_GROUP0_BASE + SMP_CORE1_OFFSET + BUF);
208	ipi_mailbox_buf[2] = (void *)
209		(SMP_CORE_GROUP0_BASE + SMP_CORE2_OFFSET + BUF);
210	ipi_mailbox_buf[3] = (void *)
211		(SMP_CORE_GROUP0_BASE + SMP_CORE3_OFFSET + BUF);
212	ipi_mailbox_buf[4] = (void *)
213		(SMP_CORE_GROUP1_BASE + SMP_CORE0_OFFSET + BUF);
214	ipi_mailbox_buf[5] = (void *)
215		(SMP_CORE_GROUP1_BASE + SMP_CORE1_OFFSET + BUF);
216	ipi_mailbox_buf[6] = (void *)
217		(SMP_CORE_GROUP1_BASE + SMP_CORE2_OFFSET + BUF);
218	ipi_mailbox_buf[7] = (void *)
219		(SMP_CORE_GROUP1_BASE + SMP_CORE3_OFFSET + BUF);
220	ipi_mailbox_buf[8] = (void *)
221		(SMP_CORE_GROUP2_BASE + SMP_CORE0_OFFSET + BUF);
222	ipi_mailbox_buf[9] = (void *)
223		(SMP_CORE_GROUP2_BASE + SMP_CORE1_OFFSET + BUF);
224	ipi_mailbox_buf[10] = (void *)
225		(SMP_CORE_GROUP2_BASE + SMP_CORE2_OFFSET + BUF);
226	ipi_mailbox_buf[11] = (void *)
227		(SMP_CORE_GROUP2_BASE + SMP_CORE3_OFFSET + BUF);
228	ipi_mailbox_buf[12] = (void *)
229		(SMP_CORE_GROUP3_BASE + SMP_CORE0_OFFSET + BUF);
230	ipi_mailbox_buf[13] = (void *)
231		(SMP_CORE_GROUP3_BASE + SMP_CORE1_OFFSET + BUF);
232	ipi_mailbox_buf[14] = (void *)
233		(SMP_CORE_GROUP3_BASE + SMP_CORE2_OFFSET + BUF);
234	ipi_mailbox_buf[15] = (void *)
235		(SMP_CORE_GROUP3_BASE + SMP_CORE3_OFFSET + BUF);
236}
237
238/*
239 * Simple enough, just poke the appropriate ipi register
240 */
241static void loongson3_send_ipi_single(int cpu, unsigned int action)
242{
243	loongson3_ipi_write32((u32)action, ipi_set0_regs[cpu_logical_map(cpu)]);
244}
245
246static void
247loongson3_send_ipi_mask(const struct cpumask *mask, unsigned int action)
248{
249	unsigned int i;
250
251	for_each_cpu(i, mask)
252		loongson3_ipi_write32((u32)action, ipi_set0_regs[cpu_logical_map(i)]);
253}
254
255void loongson3_ipi_interrupt(struct pt_regs *regs)
256{
257	int i, cpu = smp_processor_id();
258	unsigned int action, c0count;
259
260	/* Load the ipi register to figure out what we're supposed to do */
261	action = loongson3_ipi_read32(ipi_status0_regs[cpu_logical_map(cpu)]);
262
263	/* Clear the ipi register to clear the interrupt */
264	loongson3_ipi_write32((u32)action, ipi_clear0_regs[cpu_logical_map(cpu)]);
265
266	if (action & SMP_RESCHEDULE_YOURSELF)
267		scheduler_ipi();
268
269	if (action & SMP_CALL_FUNCTION)
270		smp_call_function_interrupt();
271
272	if (action & SMP_ASK_C0COUNT) {
273		BUG_ON(cpu != 0);
274		c0count = read_c0_count();
275		for (i = 1; i < num_possible_cpus(); i++)
276			per_cpu(core0_c0count, i) = c0count;
277	}
278}
279
280#define MAX_LOOPS 1111
281/*
282 * SMP init and finish on secondary CPUs
283 */
284static void loongson3_init_secondary(void)
285{
286	int i;
287	uint32_t initcount;
288	unsigned int cpu = smp_processor_id();
289	unsigned int imask = STATUSF_IP7 | STATUSF_IP6 |
290			     STATUSF_IP3 | STATUSF_IP2;
291
292	/* Set interrupt mask, but don't enable */
293	change_c0_status(ST0_IM, imask);
294
295	for (i = 0; i < num_possible_cpus(); i++)
296		loongson3_ipi_write32(0xffffffff, ipi_en0_regs[cpu_logical_map(i)]);
297
298	per_cpu(cpu_state, cpu) = CPU_ONLINE;
299	cpu_data[cpu].core =
300		cpu_logical_map(cpu) % loongson_sysconf.cores_per_package;
301	cpu_data[cpu].package =
302		cpu_logical_map(cpu) / loongson_sysconf.cores_per_package;
303
304	i = 0;
305	__this_cpu_write(core0_c0count, 0);
306	loongson3_send_ipi_single(0, SMP_ASK_C0COUNT);
307	while (!__this_cpu_read(core0_c0count)) {
308		i++;
309		cpu_relax();
310	}
311
312	if (i > MAX_LOOPS)
313		i = MAX_LOOPS;
314	initcount = __this_cpu_read(core0_c0count) + i;
315	write_c0_count(initcount);
316}
317
318static void loongson3_smp_finish(void)
319{
320	int cpu = smp_processor_id();
321
322	write_c0_compare(read_c0_count() + mips_hpt_frequency/HZ);
323	local_irq_enable();
324	loongson3_ipi_write64(0,
325			(void *)(ipi_mailbox_buf[cpu_logical_map(cpu)]+0x0));
326	pr_info("CPU#%d finished, CP0_ST=%x\n",
327			smp_processor_id(), read_c0_status());
328}
329
330static void __init loongson3_smp_setup(void)
331{
332	int i = 0, num = 0; /* i: physical id, num: logical id */
333
334	init_cpu_possible(cpu_none_mask);
335
336	/* For unified kernel, NR_CPUS is the maximum possible value,
337	 * loongson_sysconf.nr_cpus is the really present value */
338	while (i < loongson_sysconf.nr_cpus) {
339		if (loongson_sysconf.reserved_cpus_mask & (1<<i)) {
340			/* Reserved physical CPU cores */
341			__cpu_number_map[i] = -1;
342		} else {
343			__cpu_number_map[i] = num;
344			__cpu_logical_map[num] = i;
345			set_cpu_possible(num, true);
346			num++;
347		}
348		i++;
349	}
350	pr_info("Detected %i available CPU(s)\n", num);
351
352	while (num < loongson_sysconf.nr_cpus) {
353		__cpu_logical_map[num] = -1;
354		num++;
355	}
356
357	ipi_set0_regs_init();
358	ipi_clear0_regs_init();
359	ipi_status0_regs_init();
360	ipi_en0_regs_init();
361	ipi_mailbox_buf_init();
362	cpu_data[0].core = cpu_logical_map(0) % loongson_sysconf.cores_per_package;
363	cpu_data[0].package = cpu_logical_map(0) / loongson_sysconf.cores_per_package;
364}
365
366static void __init loongson3_prepare_cpus(unsigned int max_cpus)
367{
368	init_cpu_present(cpu_possible_mask);
369	per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
370}
371
372/*
373 * Setup the PC, SP, and GP of a secondary processor and start it runing!
374 */
375static void loongson3_boot_secondary(int cpu, struct task_struct *idle)
376{
377	unsigned long startargs[4];
378
379	pr_info("Booting CPU#%d...\n", cpu);
380
381	/* startargs[] are initial PC, SP and GP for secondary CPU */
382	startargs[0] = (unsigned long)&smp_bootstrap;
383	startargs[1] = (unsigned long)__KSTK_TOS(idle);
384	startargs[2] = (unsigned long)task_thread_info(idle);
385	startargs[3] = 0;
386
387	pr_debug("CPU#%d, func_pc=%lx, sp=%lx, gp=%lx\n",
388			cpu, startargs[0], startargs[1], startargs[2]);
389
390	loongson3_ipi_write64(startargs[3],
391			(void *)(ipi_mailbox_buf[cpu_logical_map(cpu)]+0x18));
392	loongson3_ipi_write64(startargs[2],
393			(void *)(ipi_mailbox_buf[cpu_logical_map(cpu)]+0x10));
394	loongson3_ipi_write64(startargs[1],
395			(void *)(ipi_mailbox_buf[cpu_logical_map(cpu)]+0x8));
396	loongson3_ipi_write64(startargs[0],
397			(void *)(ipi_mailbox_buf[cpu_logical_map(cpu)]+0x0));
398}
399
400#ifdef CONFIG_HOTPLUG_CPU
401
402static int loongson3_cpu_disable(void)
403{
404	unsigned long flags;
405	unsigned int cpu = smp_processor_id();
406
407	if (cpu == 0)
408		return -EBUSY;
409
410	set_cpu_online(cpu, false);
411	cpumask_clear_cpu(cpu, &cpu_callin_map);
412	local_irq_save(flags);
413	fixup_irqs();
414	local_irq_restore(flags);
415	flush_cache_all();
416	local_flush_tlb_all();
417
418	return 0;
419}
420
421
422static void loongson3_cpu_die(unsigned int cpu)
423{
424	while (per_cpu(cpu_state, cpu) != CPU_DEAD)
425		cpu_relax();
426
427	mb();
428}
429
430/* To shutdown a core in Loongson 3, the target core should go to CKSEG1 and
431 * flush all L1 entries at first. Then, another core (usually Core 0) can
432 * safely disable the clock of the target core. loongson3_play_dead() is
433 * called via CKSEG1 (uncached and unmmaped) */
434static void loongson3a_play_dead(int *state_addr)
435{
436	register int val;
437	register long cpuid, core, node, count;
438	register void *addr, *base, *initfunc;
439
440	__asm__ __volatile__(
441		"   .set push                     \n"
442		"   .set noreorder                \n"
443		"   li %[addr], 0x80000000        \n" /* KSEG0 */
444		"1: cache 0, 0(%[addr])           \n" /* flush L1 ICache */
445		"   cache 0, 1(%[addr])           \n"
446		"   cache 0, 2(%[addr])           \n"
447		"   cache 0, 3(%[addr])           \n"
448		"   cache 1, 0(%[addr])           \n" /* flush L1 DCache */
449		"   cache 1, 1(%[addr])           \n"
450		"   cache 1, 2(%[addr])           \n"
451		"   cache 1, 3(%[addr])           \n"
452		"   addiu %[sets], %[sets], -1    \n"
453		"   bnez  %[sets], 1b             \n"
454		"   addiu %[addr], %[addr], 0x20  \n"
455		"   li    %[val], 0x7             \n" /* *state_addr = CPU_DEAD; */
456		"   sw    %[val], (%[state_addr]) \n"
457		"   sync                          \n"
458		"   cache 21, (%[state_addr])     \n" /* flush entry of *state_addr */
459		"   .set pop                      \n"
460		: [addr] "=&r" (addr), [val] "=&r" (val)
461		: [state_addr] "r" (state_addr),
462		  [sets] "r" (cpu_data[smp_processor_id()].dcache.sets));
463
464	__asm__ __volatile__(
465		"   .set push                         \n"
466		"   .set noreorder                    \n"
467		"   .set mips64                       \n"
468		"   mfc0  %[cpuid], $15, 1            \n"
469		"   andi  %[cpuid], 0x3ff             \n"
470		"   dli   %[base], 0x900000003ff01000 \n"
471		"   andi  %[core], %[cpuid], 0x3      \n"
472		"   sll   %[core], 8                  \n" /* get core id */
473		"   or    %[base], %[base], %[core]   \n"
474		"   andi  %[node], %[cpuid], 0xc      \n"
475		"   dsll  %[node], 42                 \n" /* get node id */
476		"   or    %[base], %[base], %[node]   \n"
477		"1: li    %[count], 0x100             \n" /* wait for init loop */
478		"2: bnez  %[count], 2b                \n" /* limit mailbox access */
479		"   addiu %[count], -1                \n"
480		"   ld    %[initfunc], 0x20(%[base])  \n" /* get PC via mailbox */
481		"   beqz  %[initfunc], 1b             \n"
482		"   nop                               \n"
483		"   ld    $sp, 0x28(%[base])          \n" /* get SP via mailbox */
484		"   ld    $gp, 0x30(%[base])          \n" /* get GP via mailbox */
485		"   ld    $a1, 0x38(%[base])          \n"
486		"   jr    %[initfunc]                 \n" /* jump to initial PC */
487		"   nop                               \n"
488		"   .set pop                          \n"
489		: [core] "=&r" (core), [node] "=&r" (node),
490		  [base] "=&r" (base), [cpuid] "=&r" (cpuid),
491		  [count] "=&r" (count), [initfunc] "=&r" (initfunc)
492		: /* No Input */
493		: "a1");
494}
495
496static void loongson3b_play_dead(int *state_addr)
497{
498	register int val;
499	register long cpuid, core, node, count;
500	register void *addr, *base, *initfunc;
501
502	__asm__ __volatile__(
503		"   .set push                     \n"
504		"   .set noreorder                \n"
505		"   li %[addr], 0x80000000        \n" /* KSEG0 */
506		"1: cache 0, 0(%[addr])           \n" /* flush L1 ICache */
507		"   cache 0, 1(%[addr])           \n"
508		"   cache 0, 2(%[addr])           \n"
509		"   cache 0, 3(%[addr])           \n"
510		"   cache 1, 0(%[addr])           \n" /* flush L1 DCache */
511		"   cache 1, 1(%[addr])           \n"
512		"   cache 1, 2(%[addr])           \n"
513		"   cache 1, 3(%[addr])           \n"
514		"   addiu %[sets], %[sets], -1    \n"
515		"   bnez  %[sets], 1b             \n"
516		"   addiu %[addr], %[addr], 0x20  \n"
517		"   li    %[val], 0x7             \n" /* *state_addr = CPU_DEAD; */
518		"   sw    %[val], (%[state_addr]) \n"
519		"   sync                          \n"
520		"   cache 21, (%[state_addr])     \n" /* flush entry of *state_addr */
521		"   .set pop                      \n"
522		: [addr] "=&r" (addr), [val] "=&r" (val)
523		: [state_addr] "r" (state_addr),
524		  [sets] "r" (cpu_data[smp_processor_id()].dcache.sets));
525
526	__asm__ __volatile__(
527		"   .set push                         \n"
528		"   .set noreorder                    \n"
529		"   .set mips64                       \n"
530		"   mfc0  %[cpuid], $15, 1            \n"
531		"   andi  %[cpuid], 0x3ff             \n"
532		"   dli   %[base], 0x900000003ff01000 \n"
533		"   andi  %[core], %[cpuid], 0x3      \n"
534		"   sll   %[core], 8                  \n" /* get core id */
535		"   or    %[base], %[base], %[core]   \n"
536		"   andi  %[node], %[cpuid], 0xc      \n"
537		"   dsll  %[node], 42                 \n" /* get node id */
538		"   or    %[base], %[base], %[node]   \n"
539		"   dsrl  %[node], 30                 \n" /* 15:14 */
540		"   or    %[base], %[base], %[node]   \n"
541		"1: li    %[count], 0x100             \n" /* wait for init loop */
542		"2: bnez  %[count], 2b                \n" /* limit mailbox access */
543		"   addiu %[count], -1                \n"
544		"   ld    %[initfunc], 0x20(%[base])  \n" /* get PC via mailbox */
545		"   beqz  %[initfunc], 1b             \n"
546		"   nop                               \n"
547		"   ld    $sp, 0x28(%[base])          \n" /* get SP via mailbox */
548		"   ld    $gp, 0x30(%[base])          \n" /* get GP via mailbox */
549		"   ld    $a1, 0x38(%[base])          \n"
550		"   jr    %[initfunc]                 \n" /* jump to initial PC */
551		"   nop                               \n"
552		"   .set pop                          \n"
553		: [core] "=&r" (core), [node] "=&r" (node),
554		  [base] "=&r" (base), [cpuid] "=&r" (cpuid),
555		  [count] "=&r" (count), [initfunc] "=&r" (initfunc)
556		: /* No Input */
557		: "a1");
558}
559
560void play_dead(void)
561{
562	int *state_addr;
563	unsigned int cpu = smp_processor_id();
564	void (*play_dead_at_ckseg1)(int *);
565
566	idle_task_exit();
567	switch (loongson_sysconf.cputype) {
568	case Loongson_3A:
569	default:
570		play_dead_at_ckseg1 =
571			(void *)CKSEG1ADDR((unsigned long)loongson3a_play_dead);
572		break;
573	case Loongson_3B:
574		play_dead_at_ckseg1 =
575			(void *)CKSEG1ADDR((unsigned long)loongson3b_play_dead);
576		break;
577	}
578	state_addr = &per_cpu(cpu_state, cpu);
579	mb();
580	play_dead_at_ckseg1(state_addr);
581}
582
583void loongson3_disable_clock(int cpu)
584{
585	uint64_t core_id = cpu_data[cpu].core;
586	uint64_t package_id = cpu_data[cpu].package;
587
588	if (loongson_sysconf.cputype == Loongson_3A) {
589		LOONGSON_CHIPCFG(package_id) &= ~(1 << (12 + core_id));
590	} else if (loongson_sysconf.cputype == Loongson_3B) {
591		if (!(loongson_sysconf.workarounds & WORKAROUND_CPUHOTPLUG))
592			LOONGSON_FREQCTRL(package_id) &= ~(1 << (core_id * 4 + 3));
593	}
594}
595
596void loongson3_enable_clock(int cpu)
597{
598	uint64_t core_id = cpu_data[cpu].core;
599	uint64_t package_id = cpu_data[cpu].package;
600
601	if (loongson_sysconf.cputype == Loongson_3A) {
602		LOONGSON_CHIPCFG(package_id) |= 1 << (12 + core_id);
603	} else if (loongson_sysconf.cputype == Loongson_3B) {
604		if (!(loongson_sysconf.workarounds & WORKAROUND_CPUHOTPLUG))
605			LOONGSON_FREQCTRL(package_id) |= 1 << (core_id * 4 + 3);
606	}
607}
608
609#define CPU_POST_DEAD_FROZEN	(CPU_POST_DEAD | CPU_TASKS_FROZEN)
610static int loongson3_cpu_callback(struct notifier_block *nfb,
611	unsigned long action, void *hcpu)
612{
613	unsigned int cpu = (unsigned long)hcpu;
614
615	switch (action) {
616	case CPU_POST_DEAD:
617	case CPU_POST_DEAD_FROZEN:
618		pr_info("Disable clock for CPU#%d\n", cpu);
619		loongson3_disable_clock(cpu);
620		break;
621	case CPU_UP_PREPARE:
622	case CPU_UP_PREPARE_FROZEN:
623		pr_info("Enable clock for CPU#%d\n", cpu);
624		loongson3_enable_clock(cpu);
625		break;
626	}
627
628	return NOTIFY_OK;
629}
630
631static int register_loongson3_notifier(void)
632{
633	hotcpu_notifier(loongson3_cpu_callback, 0);
634	return 0;
635}
636early_initcall(register_loongson3_notifier);
637
638#endif
639
640struct plat_smp_ops loongson3_smp_ops = {
641	.send_ipi_single = loongson3_send_ipi_single,
642	.send_ipi_mask = loongson3_send_ipi_mask,
643	.init_secondary = loongson3_init_secondary,
644	.smp_finish = loongson3_smp_finish,
645	.boot_secondary = loongson3_boot_secondary,
646	.smp_setup = loongson3_smp_setup,
647	.prepare_cpus = loongson3_prepare_cpus,
648#ifdef CONFIG_HOTPLUG_CPU
649	.cpu_disable = loongson3_cpu_disable,
650	.cpu_die = loongson3_cpu_die,
651#endif
652};
653