1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Copyright (C) 2010, 2011, 2012, Lemote, Inc.
4 * Author: Chen Huacai, chenhc@lemote.com
5 */
6
7 #include <irq.h>
8 #include <linux/init.h>
9 #include <linux/cpu.h>
10 #include <linux/sched.h>
11 #include <linux/sched/hotplug.h>
12 #include <linux/sched/task_stack.h>
13 #include <linux/smp.h>
14 #include <linux/cpufreq.h>
15 #include <linux/kexec.h>
16 #include <asm/processor.h>
17 #include <asm/smp.h>
18 #include <asm/time.h>
19 #include <asm/tlbflush.h>
20 #include <asm/cacheflush.h>
21 #include <loongson.h>
22 #include <loongson_regs.h>
23 #include <workarounds.h>
24
25 #include "smp.h"
26
27 DEFINE_PER_CPU(int, cpu_state);
28
29 #define LS_IPI_IRQ (MIPS_CPU_IRQ_BASE + 6)
30
31 static void __iomem *ipi_set0_regs[16];
32 static void __iomem *ipi_clear0_regs[16];
33 static void __iomem *ipi_status0_regs[16];
34 static void __iomem *ipi_en0_regs[16];
35 static void __iomem *ipi_mailbox_buf[16];
36
37 static u32 (*ipi_read_clear)(int cpu);
38 static void (*ipi_write_action)(int cpu, u32 action);
39 static void (*ipi_write_enable)(int cpu);
40 static void (*ipi_clear_buf)(int cpu);
41 static void (*ipi_write_buf)(int cpu, struct task_struct *idle);
42
43 /* send mail via Mail_Send register for 3A4000+ CPU */
csr_mail_send(uint64_t data,int cpu,int mailbox)44 static void csr_mail_send(uint64_t data, int cpu, int mailbox)
45 {
46 uint64_t val;
47
48 /* send high 32 bits */
49 val = CSR_MAIL_SEND_BLOCK;
50 val |= (CSR_MAIL_SEND_BOX_HIGH(mailbox) << CSR_MAIL_SEND_BOX_SHIFT);
51 val |= (cpu << CSR_MAIL_SEND_CPU_SHIFT);
52 val |= (data & CSR_MAIL_SEND_H32_MASK);
53 csr_writeq(val, LOONGSON_CSR_MAIL_SEND);
54
55 /* send low 32 bits */
56 val = CSR_MAIL_SEND_BLOCK;
57 val |= (CSR_MAIL_SEND_BOX_LOW(mailbox) << CSR_MAIL_SEND_BOX_SHIFT);
58 val |= (cpu << CSR_MAIL_SEND_CPU_SHIFT);
59 val |= (data << CSR_MAIL_SEND_BUF_SHIFT);
60 csr_writeq(val, LOONGSON_CSR_MAIL_SEND);
61 };
62
csr_ipi_read_clear(int cpu)63 static u32 csr_ipi_read_clear(int cpu)
64 {
65 u32 action;
66
67 /* Load the ipi register to figure out what we're supposed to do */
68 action = csr_readl(LOONGSON_CSR_IPI_STATUS);
69 /* Clear the ipi register to clear the interrupt */
70 csr_writel(action, LOONGSON_CSR_IPI_CLEAR);
71
72 return action;
73 }
74
csr_ipi_write_action(int cpu,u32 action)75 static void csr_ipi_write_action(int cpu, u32 action)
76 {
77 unsigned int irq = 0;
78
79 while ((irq = ffs(action))) {
80 uint32_t val = CSR_IPI_SEND_BLOCK;
81 val |= (irq - 1);
82 val |= (cpu << CSR_IPI_SEND_CPU_SHIFT);
83 csr_writel(val, LOONGSON_CSR_IPI_SEND);
84 action &= ~BIT(irq - 1);
85 }
86 }
87
csr_ipi_write_enable(int cpu)88 static void csr_ipi_write_enable(int cpu)
89 {
90 csr_writel(0xffffffff, LOONGSON_CSR_IPI_EN);
91 }
92
csr_ipi_clear_buf(int cpu)93 static void csr_ipi_clear_buf(int cpu)
94 {
95 csr_writeq(0, LOONGSON_CSR_MAIL_BUF0);
96 }
97
csr_ipi_write_buf(int cpu,struct task_struct * idle)98 static void csr_ipi_write_buf(int cpu, struct task_struct *idle)
99 {
100 unsigned long startargs[4];
101
102 /* startargs[] are initial PC, SP and GP for secondary CPU */
103 startargs[0] = (unsigned long)&smp_bootstrap;
104 startargs[1] = (unsigned long)__KSTK_TOS(idle);
105 startargs[2] = (unsigned long)task_thread_info(idle);
106 startargs[3] = 0;
107
108 pr_debug("CPU#%d, func_pc=%lx, sp=%lx, gp=%lx\n",
109 cpu, startargs[0], startargs[1], startargs[2]);
110
111 csr_mail_send(startargs[3], cpu_logical_map(cpu), 3);
112 csr_mail_send(startargs[2], cpu_logical_map(cpu), 2);
113 csr_mail_send(startargs[1], cpu_logical_map(cpu), 1);
114 csr_mail_send(startargs[0], cpu_logical_map(cpu), 0);
115 }
116
legacy_ipi_read_clear(int cpu)117 static u32 legacy_ipi_read_clear(int cpu)
118 {
119 u32 action;
120
121 /* Load the ipi register to figure out what we're supposed to do */
122 action = readl_relaxed(ipi_status0_regs[cpu_logical_map(cpu)]);
123 /* Clear the ipi register to clear the interrupt */
124 writel_relaxed(action, ipi_clear0_regs[cpu_logical_map(cpu)]);
125 nudge_writes();
126
127 return action;
128 }
129
legacy_ipi_write_action(int cpu,u32 action)130 static void legacy_ipi_write_action(int cpu, u32 action)
131 {
132 writel_relaxed((u32)action, ipi_set0_regs[cpu]);
133 nudge_writes();
134 }
135
legacy_ipi_write_enable(int cpu)136 static void legacy_ipi_write_enable(int cpu)
137 {
138 writel_relaxed(0xffffffff, ipi_en0_regs[cpu_logical_map(cpu)]);
139 }
140
legacy_ipi_clear_buf(int cpu)141 static void legacy_ipi_clear_buf(int cpu)
142 {
143 writeq_relaxed(0, ipi_mailbox_buf[cpu_logical_map(cpu)] + 0x0);
144 }
145
legacy_ipi_write_buf(int cpu,struct task_struct * idle)146 static void legacy_ipi_write_buf(int cpu, struct task_struct *idle)
147 {
148 unsigned long startargs[4];
149
150 /* startargs[] are initial PC, SP and GP for secondary CPU */
151 startargs[0] = (unsigned long)&smp_bootstrap;
152 startargs[1] = (unsigned long)__KSTK_TOS(idle);
153 startargs[2] = (unsigned long)task_thread_info(idle);
154 startargs[3] = 0;
155
156 pr_debug("CPU#%d, func_pc=%lx, sp=%lx, gp=%lx\n",
157 cpu, startargs[0], startargs[1], startargs[2]);
158
159 writeq_relaxed(startargs[3],
160 ipi_mailbox_buf[cpu_logical_map(cpu)] + 0x18);
161 writeq_relaxed(startargs[2],
162 ipi_mailbox_buf[cpu_logical_map(cpu)] + 0x10);
163 writeq_relaxed(startargs[1],
164 ipi_mailbox_buf[cpu_logical_map(cpu)] + 0x8);
165 writeq_relaxed(startargs[0],
166 ipi_mailbox_buf[cpu_logical_map(cpu)] + 0x0);
167 nudge_writes();
168 }
169
csr_ipi_probe(void)170 static void csr_ipi_probe(void)
171 {
172 if (cpu_has_csr() && csr_readl(LOONGSON_CSR_FEATURES) & LOONGSON_CSRF_IPI) {
173 ipi_read_clear = csr_ipi_read_clear;
174 ipi_write_action = csr_ipi_write_action;
175 ipi_write_enable = csr_ipi_write_enable;
176 ipi_clear_buf = csr_ipi_clear_buf;
177 ipi_write_buf = csr_ipi_write_buf;
178 } else {
179 ipi_read_clear = legacy_ipi_read_clear;
180 ipi_write_action = legacy_ipi_write_action;
181 ipi_write_enable = legacy_ipi_write_enable;
182 ipi_clear_buf = legacy_ipi_clear_buf;
183 ipi_write_buf = legacy_ipi_write_buf;
184 }
185 }
186
ipi_set0_regs_init(void)187 static void ipi_set0_regs_init(void)
188 {
189 ipi_set0_regs[0] = (void __iomem *)
190 (SMP_CORE_GROUP0_BASE + SMP_CORE0_OFFSET + SET0);
191 ipi_set0_regs[1] = (void __iomem *)
192 (SMP_CORE_GROUP0_BASE + SMP_CORE1_OFFSET + SET0);
193 ipi_set0_regs[2] = (void __iomem *)
194 (SMP_CORE_GROUP0_BASE + SMP_CORE2_OFFSET + SET0);
195 ipi_set0_regs[3] = (void __iomem *)
196 (SMP_CORE_GROUP0_BASE + SMP_CORE3_OFFSET + SET0);
197 ipi_set0_regs[4] = (void __iomem *)
198 (SMP_CORE_GROUP1_BASE + SMP_CORE0_OFFSET + SET0);
199 ipi_set0_regs[5] = (void __iomem *)
200 (SMP_CORE_GROUP1_BASE + SMP_CORE1_OFFSET + SET0);
201 ipi_set0_regs[6] = (void __iomem *)
202 (SMP_CORE_GROUP1_BASE + SMP_CORE2_OFFSET + SET0);
203 ipi_set0_regs[7] = (void __iomem *)
204 (SMP_CORE_GROUP1_BASE + SMP_CORE3_OFFSET + SET0);
205 ipi_set0_regs[8] = (void __iomem *)
206 (SMP_CORE_GROUP2_BASE + SMP_CORE0_OFFSET + SET0);
207 ipi_set0_regs[9] = (void __iomem *)
208 (SMP_CORE_GROUP2_BASE + SMP_CORE1_OFFSET + SET0);
209 ipi_set0_regs[10] = (void __iomem *)
210 (SMP_CORE_GROUP2_BASE + SMP_CORE2_OFFSET + SET0);
211 ipi_set0_regs[11] = (void __iomem *)
212 (SMP_CORE_GROUP2_BASE + SMP_CORE3_OFFSET + SET0);
213 ipi_set0_regs[12] = (void __iomem *)
214 (SMP_CORE_GROUP3_BASE + SMP_CORE0_OFFSET + SET0);
215 ipi_set0_regs[13] = (void __iomem *)
216 (SMP_CORE_GROUP3_BASE + SMP_CORE1_OFFSET + SET0);
217 ipi_set0_regs[14] = (void __iomem *)
218 (SMP_CORE_GROUP3_BASE + SMP_CORE2_OFFSET + SET0);
219 ipi_set0_regs[15] = (void __iomem *)
220 (SMP_CORE_GROUP3_BASE + SMP_CORE3_OFFSET + SET0);
221 }
222
ipi_clear0_regs_init(void)223 static void ipi_clear0_regs_init(void)
224 {
225 ipi_clear0_regs[0] = (void __iomem *)
226 (SMP_CORE_GROUP0_BASE + SMP_CORE0_OFFSET + CLEAR0);
227 ipi_clear0_regs[1] = (void __iomem *)
228 (SMP_CORE_GROUP0_BASE + SMP_CORE1_OFFSET + CLEAR0);
229 ipi_clear0_regs[2] = (void __iomem *)
230 (SMP_CORE_GROUP0_BASE + SMP_CORE2_OFFSET + CLEAR0);
231 ipi_clear0_regs[3] = (void __iomem *)
232 (SMP_CORE_GROUP0_BASE + SMP_CORE3_OFFSET + CLEAR0);
233 ipi_clear0_regs[4] = (void __iomem *)
234 (SMP_CORE_GROUP1_BASE + SMP_CORE0_OFFSET + CLEAR0);
235 ipi_clear0_regs[5] = (void __iomem *)
236 (SMP_CORE_GROUP1_BASE + SMP_CORE1_OFFSET + CLEAR0);
237 ipi_clear0_regs[6] = (void __iomem *)
238 (SMP_CORE_GROUP1_BASE + SMP_CORE2_OFFSET + CLEAR0);
239 ipi_clear0_regs[7] = (void __iomem *)
240 (SMP_CORE_GROUP1_BASE + SMP_CORE3_OFFSET + CLEAR0);
241 ipi_clear0_regs[8] = (void __iomem *)
242 (SMP_CORE_GROUP2_BASE + SMP_CORE0_OFFSET + CLEAR0);
243 ipi_clear0_regs[9] = (void __iomem *)
244 (SMP_CORE_GROUP2_BASE + SMP_CORE1_OFFSET + CLEAR0);
245 ipi_clear0_regs[10] = (void __iomem *)
246 (SMP_CORE_GROUP2_BASE + SMP_CORE2_OFFSET + CLEAR0);
247 ipi_clear0_regs[11] = (void __iomem *)
248 (SMP_CORE_GROUP2_BASE + SMP_CORE3_OFFSET + CLEAR0);
249 ipi_clear0_regs[12] = (void __iomem *)
250 (SMP_CORE_GROUP3_BASE + SMP_CORE0_OFFSET + CLEAR0);
251 ipi_clear0_regs[13] = (void __iomem *)
252 (SMP_CORE_GROUP3_BASE + SMP_CORE1_OFFSET + CLEAR0);
253 ipi_clear0_regs[14] = (void __iomem *)
254 (SMP_CORE_GROUP3_BASE + SMP_CORE2_OFFSET + CLEAR0);
255 ipi_clear0_regs[15] = (void __iomem *)
256 (SMP_CORE_GROUP3_BASE + SMP_CORE3_OFFSET + CLEAR0);
257 }
258
ipi_status0_regs_init(void)259 static void ipi_status0_regs_init(void)
260 {
261 ipi_status0_regs[0] = (void __iomem *)
262 (SMP_CORE_GROUP0_BASE + SMP_CORE0_OFFSET + STATUS0);
263 ipi_status0_regs[1] = (void __iomem *)
264 (SMP_CORE_GROUP0_BASE + SMP_CORE1_OFFSET + STATUS0);
265 ipi_status0_regs[2] = (void __iomem *)
266 (SMP_CORE_GROUP0_BASE + SMP_CORE2_OFFSET + STATUS0);
267 ipi_status0_regs[3] = (void __iomem *)
268 (SMP_CORE_GROUP0_BASE + SMP_CORE3_OFFSET + STATUS0);
269 ipi_status0_regs[4] = (void __iomem *)
270 (SMP_CORE_GROUP1_BASE + SMP_CORE0_OFFSET + STATUS0);
271 ipi_status0_regs[5] = (void __iomem *)
272 (SMP_CORE_GROUP1_BASE + SMP_CORE1_OFFSET + STATUS0);
273 ipi_status0_regs[6] = (void __iomem *)
274 (SMP_CORE_GROUP1_BASE + SMP_CORE2_OFFSET + STATUS0);
275 ipi_status0_regs[7] = (void __iomem *)
276 (SMP_CORE_GROUP1_BASE + SMP_CORE3_OFFSET + STATUS0);
277 ipi_status0_regs[8] = (void __iomem *)
278 (SMP_CORE_GROUP2_BASE + SMP_CORE0_OFFSET + STATUS0);
279 ipi_status0_regs[9] = (void __iomem *)
280 (SMP_CORE_GROUP2_BASE + SMP_CORE1_OFFSET + STATUS0);
281 ipi_status0_regs[10] = (void __iomem *)
282 (SMP_CORE_GROUP2_BASE + SMP_CORE2_OFFSET + STATUS0);
283 ipi_status0_regs[11] = (void __iomem *)
284 (SMP_CORE_GROUP2_BASE + SMP_CORE3_OFFSET + STATUS0);
285 ipi_status0_regs[12] = (void __iomem *)
286 (SMP_CORE_GROUP3_BASE + SMP_CORE0_OFFSET + STATUS0);
287 ipi_status0_regs[13] = (void __iomem *)
288 (SMP_CORE_GROUP3_BASE + SMP_CORE1_OFFSET + STATUS0);
289 ipi_status0_regs[14] = (void __iomem *)
290 (SMP_CORE_GROUP3_BASE + SMP_CORE2_OFFSET + STATUS0);
291 ipi_status0_regs[15] = (void __iomem *)
292 (SMP_CORE_GROUP3_BASE + SMP_CORE3_OFFSET + STATUS0);
293 }
294
ipi_en0_regs_init(void)295 static void ipi_en0_regs_init(void)
296 {
297 ipi_en0_regs[0] = (void __iomem *)
298 (SMP_CORE_GROUP0_BASE + SMP_CORE0_OFFSET + EN0);
299 ipi_en0_regs[1] = (void __iomem *)
300 (SMP_CORE_GROUP0_BASE + SMP_CORE1_OFFSET + EN0);
301 ipi_en0_regs[2] = (void __iomem *)
302 (SMP_CORE_GROUP0_BASE + SMP_CORE2_OFFSET + EN0);
303 ipi_en0_regs[3] = (void __iomem *)
304 (SMP_CORE_GROUP0_BASE + SMP_CORE3_OFFSET + EN0);
305 ipi_en0_regs[4] = (void __iomem *)
306 (SMP_CORE_GROUP1_BASE + SMP_CORE0_OFFSET + EN0);
307 ipi_en0_regs[5] = (void __iomem *)
308 (SMP_CORE_GROUP1_BASE + SMP_CORE1_OFFSET + EN0);
309 ipi_en0_regs[6] = (void __iomem *)
310 (SMP_CORE_GROUP1_BASE + SMP_CORE2_OFFSET + EN0);
311 ipi_en0_regs[7] = (void __iomem *)
312 (SMP_CORE_GROUP1_BASE + SMP_CORE3_OFFSET + EN0);
313 ipi_en0_regs[8] = (void __iomem *)
314 (SMP_CORE_GROUP2_BASE + SMP_CORE0_OFFSET + EN0);
315 ipi_en0_regs[9] = (void __iomem *)
316 (SMP_CORE_GROUP2_BASE + SMP_CORE1_OFFSET + EN0);
317 ipi_en0_regs[10] = (void __iomem *)
318 (SMP_CORE_GROUP2_BASE + SMP_CORE2_OFFSET + EN0);
319 ipi_en0_regs[11] = (void __iomem *)
320 (SMP_CORE_GROUP2_BASE + SMP_CORE3_OFFSET + EN0);
321 ipi_en0_regs[12] = (void __iomem *)
322 (SMP_CORE_GROUP3_BASE + SMP_CORE0_OFFSET + EN0);
323 ipi_en0_regs[13] = (void __iomem *)
324 (SMP_CORE_GROUP3_BASE + SMP_CORE1_OFFSET + EN0);
325 ipi_en0_regs[14] = (void __iomem *)
326 (SMP_CORE_GROUP3_BASE + SMP_CORE2_OFFSET + EN0);
327 ipi_en0_regs[15] = (void __iomem *)
328 (SMP_CORE_GROUP3_BASE + SMP_CORE3_OFFSET + EN0);
329 }
330
ipi_mailbox_buf_init(void)331 static void ipi_mailbox_buf_init(void)
332 {
333 ipi_mailbox_buf[0] = (void __iomem *)
334 (SMP_CORE_GROUP0_BASE + SMP_CORE0_OFFSET + BUF);
335 ipi_mailbox_buf[1] = (void __iomem *)
336 (SMP_CORE_GROUP0_BASE + SMP_CORE1_OFFSET + BUF);
337 ipi_mailbox_buf[2] = (void __iomem *)
338 (SMP_CORE_GROUP0_BASE + SMP_CORE2_OFFSET + BUF);
339 ipi_mailbox_buf[3] = (void __iomem *)
340 (SMP_CORE_GROUP0_BASE + SMP_CORE3_OFFSET + BUF);
341 ipi_mailbox_buf[4] = (void __iomem *)
342 (SMP_CORE_GROUP1_BASE + SMP_CORE0_OFFSET + BUF);
343 ipi_mailbox_buf[5] = (void __iomem *)
344 (SMP_CORE_GROUP1_BASE + SMP_CORE1_OFFSET + BUF);
345 ipi_mailbox_buf[6] = (void __iomem *)
346 (SMP_CORE_GROUP1_BASE + SMP_CORE2_OFFSET + BUF);
347 ipi_mailbox_buf[7] = (void __iomem *)
348 (SMP_CORE_GROUP1_BASE + SMP_CORE3_OFFSET + BUF);
349 ipi_mailbox_buf[8] = (void __iomem *)
350 (SMP_CORE_GROUP2_BASE + SMP_CORE0_OFFSET + BUF);
351 ipi_mailbox_buf[9] = (void __iomem *)
352 (SMP_CORE_GROUP2_BASE + SMP_CORE1_OFFSET + BUF);
353 ipi_mailbox_buf[10] = (void __iomem *)
354 (SMP_CORE_GROUP2_BASE + SMP_CORE2_OFFSET + BUF);
355 ipi_mailbox_buf[11] = (void __iomem *)
356 (SMP_CORE_GROUP2_BASE + SMP_CORE3_OFFSET + BUF);
357 ipi_mailbox_buf[12] = (void __iomem *)
358 (SMP_CORE_GROUP3_BASE + SMP_CORE0_OFFSET + BUF);
359 ipi_mailbox_buf[13] = (void __iomem *)
360 (SMP_CORE_GROUP3_BASE + SMP_CORE1_OFFSET + BUF);
361 ipi_mailbox_buf[14] = (void __iomem *)
362 (SMP_CORE_GROUP3_BASE + SMP_CORE2_OFFSET + BUF);
363 ipi_mailbox_buf[15] = (void __iomem *)
364 (SMP_CORE_GROUP3_BASE + SMP_CORE3_OFFSET + BUF);
365 }
366
367 /*
368 * Simple enough, just poke the appropriate ipi register
369 */
loongson3_send_ipi_single(int cpu,unsigned int action)370 static void loongson3_send_ipi_single(int cpu, unsigned int action)
371 {
372 ipi_write_action(cpu_logical_map(cpu), (u32)action);
373 }
374
375 static void
loongson3_send_ipi_mask(const struct cpumask * mask,unsigned int action)376 loongson3_send_ipi_mask(const struct cpumask *mask, unsigned int action)
377 {
378 unsigned int i;
379
380 for_each_cpu(i, mask)
381 ipi_write_action(cpu_logical_map(i), (u32)action);
382 }
383
loongson3_ipi_interrupt(int irq,void * dev_id)384 static irqreturn_t loongson3_ipi_interrupt(int irq, void *dev_id)
385 {
386 int cpu = smp_processor_id();
387 unsigned int action;
388
389 action = ipi_read_clear(cpu);
390
391 if (action & SMP_RESCHEDULE_YOURSELF)
392 scheduler_ipi();
393
394 if (action & SMP_CALL_FUNCTION) {
395 irq_enter();
396 generic_smp_call_function_interrupt();
397 irq_exit();
398 }
399
400 return IRQ_HANDLED;
401 }
402
403 /*
404 * SMP init and finish on secondary CPUs
405 */
loongson3_init_secondary(void)406 static void loongson3_init_secondary(void)
407 {
408 unsigned int cpu = smp_processor_id();
409 unsigned int imask = STATUSF_IP7 | STATUSF_IP6 |
410 STATUSF_IP3 | STATUSF_IP2;
411
412 /* Set interrupt mask, but don't enable */
413 change_c0_status(ST0_IM, imask);
414 ipi_write_enable(cpu);
415
416 per_cpu(cpu_state, cpu) = CPU_ONLINE;
417 cpu_set_core(&cpu_data[cpu],
418 cpu_logical_map(cpu) % loongson_sysconf.cores_per_package);
419 cpu_data[cpu].package =
420 cpu_logical_map(cpu) / loongson_sysconf.cores_per_package;
421 }
422
loongson3_smp_finish(void)423 static void loongson3_smp_finish(void)
424 {
425 int cpu = smp_processor_id();
426
427 write_c0_compare(read_c0_count() + mips_hpt_frequency/HZ);
428 local_irq_enable();
429 ipi_clear_buf(cpu);
430
431 pr_info("CPU#%d finished, CP0_ST=%x\n",
432 smp_processor_id(), read_c0_status());
433 }
434
loongson3_smp_setup(void)435 static void __init loongson3_smp_setup(void)
436 {
437 int i = 0, num = 0; /* i: physical id, num: logical id */
438 int max_cpus = 0;
439
440 init_cpu_possible(cpu_none_mask);
441
442 for (i = 0; i < ARRAY_SIZE(smp_group); i++) {
443 if (!smp_group[i])
444 break;
445 max_cpus += loongson_sysconf.cores_per_node;
446 }
447
448 if (max_cpus < loongson_sysconf.nr_cpus) {
449 pr_err("SMP Groups are less than the number of CPUs\n");
450 loongson_sysconf.nr_cpus = max_cpus ? max_cpus : 1;
451 }
452
453 /* For unified kernel, NR_CPUS is the maximum possible value,
454 * loongson_sysconf.nr_cpus is the really present value
455 */
456 i = 0;
457 while (i < loongson_sysconf.nr_cpus) {
458 if (loongson_sysconf.reserved_cpus_mask & (1<<i)) {
459 /* Reserved physical CPU cores */
460 __cpu_number_map[i] = -1;
461 } else {
462 __cpu_number_map[i] = num;
463 __cpu_logical_map[num] = i;
464 set_cpu_possible(num, true);
465 /* Loongson processors are always grouped by 4 */
466 cpu_set_cluster(&cpu_data[num], i / 4);
467 num++;
468 }
469 i++;
470 }
471 pr_info("Detected %i available CPU(s)\n", num);
472
473 while (num < loongson_sysconf.nr_cpus) {
474 __cpu_logical_map[num] = -1;
475 num++;
476 }
477 csr_ipi_probe();
478 ipi_set0_regs_init();
479 ipi_clear0_regs_init();
480 ipi_status0_regs_init();
481 ipi_en0_regs_init();
482 ipi_mailbox_buf_init();
483 if (smp_group[0])
484 ipi_write_enable(0);
485
486 cpu_set_core(&cpu_data[0],
487 cpu_logical_map(0) % loongson_sysconf.cores_per_package);
488 cpu_data[0].package = cpu_logical_map(0) / loongson_sysconf.cores_per_package;
489 }
490
loongson3_prepare_cpus(unsigned int max_cpus)491 static void __init loongson3_prepare_cpus(unsigned int max_cpus)
492 {
493 if (request_irq(LS_IPI_IRQ, loongson3_ipi_interrupt,
494 IRQF_PERCPU | IRQF_NO_SUSPEND, "SMP_IPI", NULL))
495 pr_err("Failed to request IPI IRQ\n");
496 init_cpu_present(cpu_possible_mask);
497 per_cpu(cpu_state, smp_processor_id()) = CPU_ONLINE;
498 }
499
500 /*
501 * Setup the PC, SP, and GP of a secondary processor and start it running!
502 */
loongson3_boot_secondary(int cpu,struct task_struct * idle)503 static int loongson3_boot_secondary(int cpu, struct task_struct *idle)
504 {
505 pr_info("Booting CPU#%d...\n", cpu);
506
507 ipi_write_buf(cpu, idle);
508
509 return 0;
510 }
511
512 #ifdef CONFIG_HOTPLUG_CPU
513
loongson3_cpu_disable(void)514 static int loongson3_cpu_disable(void)
515 {
516 unsigned long flags;
517 unsigned int cpu = smp_processor_id();
518
519 set_cpu_online(cpu, false);
520 calculate_cpu_foreign_map();
521 local_irq_save(flags);
522 clear_c0_status(ST0_IM);
523 local_irq_restore(flags);
524 local_flush_tlb_all();
525
526 return 0;
527 }
528
529
loongson3_cpu_die(unsigned int cpu)530 static void loongson3_cpu_die(unsigned int cpu)
531 {
532 while (per_cpu(cpu_state, cpu) != CPU_DEAD)
533 cpu_relax();
534
535 mb();
536 }
537
538 /* To shutdown a core in Loongson 3, the target core should go to CKSEG1 and
539 * flush all L1 entries at first. Then, another core (usually Core 0) can
540 * safely disable the clock of the target core. loongson3_play_dead() is
541 * called via CKSEG1 (uncached and unmmaped)
542 */
loongson3_type1_play_dead(int * state_addr)543 static void loongson3_type1_play_dead(int *state_addr)
544 {
545 register int val;
546 register long cpuid, core, node, count;
547 register void *addr, *base, *initfunc;
548
549 __asm__ __volatile__(
550 " .set push \n"
551 " .set noreorder \n"
552 " li %[addr], 0x80000000 \n" /* KSEG0 */
553 "1: cache 0, 0(%[addr]) \n" /* flush L1 ICache */
554 " cache 0, 1(%[addr]) \n"
555 " cache 0, 2(%[addr]) \n"
556 " cache 0, 3(%[addr]) \n"
557 " cache 1, 0(%[addr]) \n" /* flush L1 DCache */
558 " cache 1, 1(%[addr]) \n"
559 " cache 1, 2(%[addr]) \n"
560 " cache 1, 3(%[addr]) \n"
561 " addiu %[sets], %[sets], -1 \n"
562 " bnez %[sets], 1b \n"
563 " addiu %[addr], %[addr], 0x20 \n"
564 " li %[val], 0x7 \n" /* *state_addr = CPU_DEAD; */
565 " sw %[val], (%[state_addr]) \n"
566 " sync \n"
567 " cache 21, (%[state_addr]) \n" /* flush entry of *state_addr */
568 " .set pop \n"
569 : [addr] "=&r" (addr), [val] "=&r" (val)
570 : [state_addr] "r" (state_addr),
571 [sets] "r" (cpu_data[smp_processor_id()].dcache.sets));
572
573 __asm__ __volatile__(
574 " .set push \n"
575 " .set noreorder \n"
576 " .set mips64 \n"
577 " mfc0 %[cpuid], $15, 1 \n"
578 " andi %[cpuid], 0x3ff \n"
579 " dli %[base], 0x900000003ff01000 \n"
580 " andi %[core], %[cpuid], 0x3 \n"
581 " sll %[core], 8 \n" /* get core id */
582 " or %[base], %[base], %[core] \n"
583 " andi %[node], %[cpuid], 0xc \n"
584 " dsll %[node], 42 \n" /* get node id */
585 " or %[base], %[base], %[node] \n"
586 "1: li %[count], 0x100 \n" /* wait for init loop */
587 "2: bnez %[count], 2b \n" /* limit mailbox access */
588 " addiu %[count], -1 \n"
589 " ld %[initfunc], 0x20(%[base]) \n" /* get PC via mailbox */
590 " beqz %[initfunc], 1b \n"
591 " nop \n"
592 " ld $sp, 0x28(%[base]) \n" /* get SP via mailbox */
593 " ld $gp, 0x30(%[base]) \n" /* get GP via mailbox */
594 " ld $a1, 0x38(%[base]) \n"
595 " jr %[initfunc] \n" /* jump to initial PC */
596 " nop \n"
597 " .set pop \n"
598 : [core] "=&r" (core), [node] "=&r" (node),
599 [base] "=&r" (base), [cpuid] "=&r" (cpuid),
600 [count] "=&r" (count), [initfunc] "=&r" (initfunc)
601 : /* No Input */
602 : "a1");
603 }
604
loongson3_type2_play_dead(int * state_addr)605 static void loongson3_type2_play_dead(int *state_addr)
606 {
607 register int val;
608 register long cpuid, core, node, count;
609 register void *addr, *base, *initfunc;
610
611 __asm__ __volatile__(
612 " .set push \n"
613 " .set noreorder \n"
614 " li %[addr], 0x80000000 \n" /* KSEG0 */
615 "1: cache 0, 0(%[addr]) \n" /* flush L1 ICache */
616 " cache 0, 1(%[addr]) \n"
617 " cache 0, 2(%[addr]) \n"
618 " cache 0, 3(%[addr]) \n"
619 " cache 1, 0(%[addr]) \n" /* flush L1 DCache */
620 " cache 1, 1(%[addr]) \n"
621 " cache 1, 2(%[addr]) \n"
622 " cache 1, 3(%[addr]) \n"
623 " addiu %[sets], %[sets], -1 \n"
624 " bnez %[sets], 1b \n"
625 " addiu %[addr], %[addr], 0x20 \n"
626 " li %[val], 0x7 \n" /* *state_addr = CPU_DEAD; */
627 " sw %[val], (%[state_addr]) \n"
628 " sync \n"
629 " cache 21, (%[state_addr]) \n" /* flush entry of *state_addr */
630 " .set pop \n"
631 : [addr] "=&r" (addr), [val] "=&r" (val)
632 : [state_addr] "r" (state_addr),
633 [sets] "r" (cpu_data[smp_processor_id()].dcache.sets));
634
635 __asm__ __volatile__(
636 " .set push \n"
637 " .set noreorder \n"
638 " .set mips64 \n"
639 " mfc0 %[cpuid], $15, 1 \n"
640 " andi %[cpuid], 0x3ff \n"
641 " dli %[base], 0x900000003ff01000 \n"
642 " andi %[core], %[cpuid], 0x3 \n"
643 " sll %[core], 8 \n" /* get core id */
644 " or %[base], %[base], %[core] \n"
645 " andi %[node], %[cpuid], 0xc \n"
646 " dsll %[node], 42 \n" /* get node id */
647 " or %[base], %[base], %[node] \n"
648 " dsrl %[node], 30 \n" /* 15:14 */
649 " or %[base], %[base], %[node] \n"
650 "1: li %[count], 0x100 \n" /* wait for init loop */
651 "2: bnez %[count], 2b \n" /* limit mailbox access */
652 " addiu %[count], -1 \n"
653 " ld %[initfunc], 0x20(%[base]) \n" /* get PC via mailbox */
654 " beqz %[initfunc], 1b \n"
655 " nop \n"
656 " ld $sp, 0x28(%[base]) \n" /* get SP via mailbox */
657 " ld $gp, 0x30(%[base]) \n" /* get GP via mailbox */
658 " ld $a1, 0x38(%[base]) \n"
659 " jr %[initfunc] \n" /* jump to initial PC */
660 " nop \n"
661 " .set pop \n"
662 : [core] "=&r" (core), [node] "=&r" (node),
663 [base] "=&r" (base), [cpuid] "=&r" (cpuid),
664 [count] "=&r" (count), [initfunc] "=&r" (initfunc)
665 : /* No Input */
666 : "a1");
667 }
668
loongson3_type3_play_dead(int * state_addr)669 static void loongson3_type3_play_dead(int *state_addr)
670 {
671 register int val;
672 register long cpuid, core, node, count;
673 register void *addr, *base, *initfunc;
674
675 __asm__ __volatile__(
676 " .set push \n"
677 " .set noreorder \n"
678 " li %[addr], 0x80000000 \n" /* KSEG0 */
679 "1: cache 0, 0(%[addr]) \n" /* flush L1 ICache */
680 " cache 0, 1(%[addr]) \n"
681 " cache 0, 2(%[addr]) \n"
682 " cache 0, 3(%[addr]) \n"
683 " cache 1, 0(%[addr]) \n" /* flush L1 DCache */
684 " cache 1, 1(%[addr]) \n"
685 " cache 1, 2(%[addr]) \n"
686 " cache 1, 3(%[addr]) \n"
687 " addiu %[sets], %[sets], -1 \n"
688 " bnez %[sets], 1b \n"
689 " addiu %[addr], %[addr], 0x40 \n"
690 " li %[addr], 0x80000000 \n" /* KSEG0 */
691 "2: cache 2, 0(%[addr]) \n" /* flush L1 VCache */
692 " cache 2, 1(%[addr]) \n"
693 " cache 2, 2(%[addr]) \n"
694 " cache 2, 3(%[addr]) \n"
695 " cache 2, 4(%[addr]) \n"
696 " cache 2, 5(%[addr]) \n"
697 " cache 2, 6(%[addr]) \n"
698 " cache 2, 7(%[addr]) \n"
699 " cache 2, 8(%[addr]) \n"
700 " cache 2, 9(%[addr]) \n"
701 " cache 2, 10(%[addr]) \n"
702 " cache 2, 11(%[addr]) \n"
703 " cache 2, 12(%[addr]) \n"
704 " cache 2, 13(%[addr]) \n"
705 " cache 2, 14(%[addr]) \n"
706 " cache 2, 15(%[addr]) \n"
707 " addiu %[vsets], %[vsets], -1 \n"
708 " bnez %[vsets], 2b \n"
709 " addiu %[addr], %[addr], 0x40 \n"
710 " li %[val], 0x7 \n" /* *state_addr = CPU_DEAD; */
711 " sw %[val], (%[state_addr]) \n"
712 " sync \n"
713 " cache 21, (%[state_addr]) \n" /* flush entry of *state_addr */
714 " .set pop \n"
715 : [addr] "=&r" (addr), [val] "=&r" (val)
716 : [state_addr] "r" (state_addr),
717 [sets] "r" (cpu_data[smp_processor_id()].dcache.sets),
718 [vsets] "r" (cpu_data[smp_processor_id()].vcache.sets));
719
720 __asm__ __volatile__(
721 " .set push \n"
722 " .set noreorder \n"
723 " .set mips64 \n"
724 " mfc0 %[cpuid], $15, 1 \n"
725 " andi %[cpuid], 0x3ff \n"
726 " dli %[base], 0x900000003ff01000 \n"
727 " andi %[core], %[cpuid], 0x3 \n"
728 " sll %[core], 8 \n" /* get core id */
729 " or %[base], %[base], %[core] \n"
730 " andi %[node], %[cpuid], 0xc \n"
731 " dsll %[node], 42 \n" /* get node id */
732 " or %[base], %[base], %[node] \n"
733 "1: li %[count], 0x100 \n" /* wait for init loop */
734 "2: bnez %[count], 2b \n" /* limit mailbox access */
735 " addiu %[count], -1 \n"
736 " lw %[initfunc], 0x20(%[base]) \n" /* check lower 32-bit as jump indicator */
737 " beqz %[initfunc], 1b \n"
738 " nop \n"
739 " ld %[initfunc], 0x20(%[base]) \n" /* get PC (whole 64-bit) via mailbox */
740 " ld $sp, 0x28(%[base]) \n" /* get SP via mailbox */
741 " ld $gp, 0x30(%[base]) \n" /* get GP via mailbox */
742 " ld $a1, 0x38(%[base]) \n"
743 " jr %[initfunc] \n" /* jump to initial PC */
744 " nop \n"
745 " .set pop \n"
746 : [core] "=&r" (core), [node] "=&r" (node),
747 [base] "=&r" (base), [cpuid] "=&r" (cpuid),
748 [count] "=&r" (count), [initfunc] "=&r" (initfunc)
749 : /* No Input */
750 : "a1");
751 }
752
play_dead(void)753 void play_dead(void)
754 {
755 int prid_imp, prid_rev, *state_addr;
756 unsigned int cpu = smp_processor_id();
757 void (*play_dead_at_ckseg1)(int *);
758
759 idle_task_exit();
760 cpuhp_ap_report_dead();
761
762 prid_imp = read_c0_prid() & PRID_IMP_MASK;
763 prid_rev = read_c0_prid() & PRID_REV_MASK;
764
765 if (prid_imp == PRID_IMP_LOONGSON_64G) {
766 play_dead_at_ckseg1 =
767 (void *)CKSEG1ADDR((unsigned long)loongson3_type3_play_dead);
768 goto out;
769 }
770
771 switch (prid_rev) {
772 case PRID_REV_LOONGSON3A_R1:
773 default:
774 play_dead_at_ckseg1 =
775 (void *)CKSEG1ADDR((unsigned long)loongson3_type1_play_dead);
776 break;
777 case PRID_REV_LOONGSON3B_R1:
778 case PRID_REV_LOONGSON3B_R2:
779 play_dead_at_ckseg1 =
780 (void *)CKSEG1ADDR((unsigned long)loongson3_type2_play_dead);
781 break;
782 case PRID_REV_LOONGSON3A_R2_0:
783 case PRID_REV_LOONGSON3A_R2_1:
784 case PRID_REV_LOONGSON3A_R3_0:
785 case PRID_REV_LOONGSON3A_R3_1:
786 play_dead_at_ckseg1 =
787 (void *)CKSEG1ADDR((unsigned long)loongson3_type3_play_dead);
788 break;
789 }
790
791 out:
792 state_addr = &per_cpu(cpu_state, cpu);
793 mb();
794 play_dead_at_ckseg1(state_addr);
795 BUG();
796 }
797
loongson3_disable_clock(unsigned int cpu)798 static int loongson3_disable_clock(unsigned int cpu)
799 {
800 uint64_t core_id = cpu_core(&cpu_data[cpu]);
801 uint64_t package_id = cpu_data[cpu].package;
802
803 if (!loongson_chipcfg[package_id] || !loongson_freqctrl[package_id])
804 return 0;
805
806 if ((read_c0_prid() & PRID_REV_MASK) == PRID_REV_LOONGSON3A_R1) {
807 LOONGSON_CHIPCFG(package_id) &= ~(1 << (12 + core_id));
808 } else {
809 if (!(loongson_sysconf.workarounds & WORKAROUND_CPUHOTPLUG))
810 LOONGSON_FREQCTRL(package_id) &= ~(1 << (core_id * 4 + 3));
811 }
812 return 0;
813 }
814
loongson3_enable_clock(unsigned int cpu)815 static int loongson3_enable_clock(unsigned int cpu)
816 {
817 uint64_t core_id = cpu_core(&cpu_data[cpu]);
818 uint64_t package_id = cpu_data[cpu].package;
819
820 if (!loongson_chipcfg[package_id] || !loongson_freqctrl[package_id])
821 return 0;
822
823 if ((read_c0_prid() & PRID_REV_MASK) == PRID_REV_LOONGSON3A_R1) {
824 LOONGSON_CHIPCFG(package_id) |= 1 << (12 + core_id);
825 } else {
826 if (!(loongson_sysconf.workarounds & WORKAROUND_CPUHOTPLUG))
827 LOONGSON_FREQCTRL(package_id) |= 1 << (core_id * 4 + 3);
828 }
829 return 0;
830 }
831
register_loongson3_notifier(void)832 static int register_loongson3_notifier(void)
833 {
834 return cpuhp_setup_state_nocalls(CPUHP_MIPS_SOC_PREPARE,
835 "mips/loongson:prepare",
836 loongson3_enable_clock,
837 loongson3_disable_clock);
838 }
839 early_initcall(register_loongson3_notifier);
840
841 #endif
842
843 const struct plat_smp_ops loongson3_smp_ops = {
844 .send_ipi_single = loongson3_send_ipi_single,
845 .send_ipi_mask = loongson3_send_ipi_mask,
846 .init_secondary = loongson3_init_secondary,
847 .smp_finish = loongson3_smp_finish,
848 .boot_secondary = loongson3_boot_secondary,
849 .smp_setup = loongson3_smp_setup,
850 .prepare_cpus = loongson3_prepare_cpus,
851 #ifdef CONFIG_HOTPLUG_CPU
852 .cpu_disable = loongson3_cpu_disable,
853 .cpu_die = loongson3_cpu_die,
854 #endif
855 #ifdef CONFIG_KEXEC_CORE
856 .kexec_nonboot_cpu = kexec_nonboot_cpu_jump,
857 #endif
858 };
859