1 /* leon_smp.c: Sparc-Leon SMP support.
2 *
3 * based on sun4m_smp.c
4 * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
5 * Copyright (C) 2009 Daniel Hellstrom (daniel@gaisler.com) Aeroflex Gaisler AB
6 * Copyright (C) 2009 Konrad Eisele (konrad@gaisler.com) Aeroflex Gaisler AB
7 */
8
9 #include <asm/head.h>
10
11 #include <linux/kernel.h>
12 #include <linux/sched.h>
13 #include <linux/threads.h>
14 #include <linux/smp.h>
15 #include <linux/interrupt.h>
16 #include <linux/kernel_stat.h>
17 #include <linux/of.h>
18 #include <linux/init.h>
19 #include <linux/spinlock.h>
20 #include <linux/mm.h>
21 #include <linux/swap.h>
22 #include <linux/profile.h>
23 #include <linux/pm.h>
24 #include <linux/delay.h>
25 #include <linux/gfp.h>
26 #include <linux/cpu.h>
27 #include <linux/clockchips.h>
28
29 #include <asm/cacheflush.h>
30 #include <asm/tlbflush.h>
31
32 #include <asm/ptrace.h>
33 #include <linux/atomic.h>
34 #include <asm/irq_regs.h>
35 #include <asm/traps.h>
36
37 #include <asm/delay.h>
38 #include <asm/irq.h>
39 #include <asm/page.h>
40 #include <asm/pgalloc.h>
41 #include <asm/pgtable.h>
42 #include <asm/oplib.h>
43 #include <asm/cpudata.h>
44 #include <asm/asi.h>
45 #include <asm/leon.h>
46 #include <asm/leon_amba.h>
47 #include <asm/timer.h>
48
49 #include "kernel.h"
50
51 #include "irq.h"
52
53 extern ctxd_t *srmmu_ctx_table_phys;
54 static int smp_processors_ready;
55 extern volatile unsigned long cpu_callin_map[NR_CPUS];
56 extern cpumask_t smp_commenced_mask;
57 void leon_configure_cache_smp(void);
58 static void leon_ipi_init(void);
59
60 /* IRQ number of LEON IPIs */
61 int leon_ipi_irq = LEON3_IRQ_IPI_DEFAULT;
62
do_swap(volatile unsigned long * ptr,unsigned long val)63 static inline unsigned long do_swap(volatile unsigned long *ptr,
64 unsigned long val)
65 {
66 __asm__ __volatile__("swapa [%2] %3, %0\n\t" : "=&r"(val)
67 : "0"(val), "r"(ptr), "i"(ASI_LEON_DCACHE_MISS)
68 : "memory");
69 return val;
70 }
71
leon_cpu_pre_starting(void * arg)72 void leon_cpu_pre_starting(void *arg)
73 {
74 leon_configure_cache_smp();
75 }
76
leon_cpu_pre_online(void * arg)77 void leon_cpu_pre_online(void *arg)
78 {
79 int cpuid = hard_smp_processor_id();
80
81 /* Allow master to continue. The master will then give us the
82 * go-ahead by setting the smp_commenced_mask and will wait without
83 * timeouts until our setup is completed fully (signified by
84 * our bit being set in the cpu_online_mask).
85 */
86 do_swap(&cpu_callin_map[cpuid], 1);
87
88 local_ops->cache_all();
89 local_ops->tlb_all();
90
91 /* Fix idle thread fields. */
92 __asm__ __volatile__("ld [%0], %%g6\n\t" : : "r"(¤t_set[cpuid])
93 : "memory" /* paranoid */);
94
95 /* Attach to the address space of init_task. */
96 atomic_inc(&init_mm.mm_count);
97 current->active_mm = &init_mm;
98
99 while (!cpumask_test_cpu(cpuid, &smp_commenced_mask))
100 mb();
101 }
102
103 /*
104 * Cycle through the processors asking the PROM to start each one.
105 */
106
107 extern struct linux_prom_registers smp_penguin_ctable;
108
leon_configure_cache_smp(void)109 void leon_configure_cache_smp(void)
110 {
111 unsigned long cfg = sparc_leon3_get_dcachecfg();
112 int me = smp_processor_id();
113
114 if (ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg) > 4) {
115 printk(KERN_INFO "Note: SMP with snooping only works on 4k cache, found %dk(0x%x) on cpu %d, disabling caches\n",
116 (unsigned int)ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg),
117 (unsigned int)cfg, (unsigned int)me);
118 sparc_leon3_disable_cache();
119 } else {
120 if (cfg & ASI_LEON3_SYSCTRL_CFG_SNOOPING) {
121 sparc_leon3_enable_snooping();
122 } else {
123 printk(KERN_INFO "Note: You have to enable snooping in the vhdl model cpu %d, disabling caches\n",
124 me);
125 sparc_leon3_disable_cache();
126 }
127 }
128
129 local_ops->cache_all();
130 local_ops->tlb_all();
131 }
132
leon_smp_setbroadcast(unsigned int mask)133 static void leon_smp_setbroadcast(unsigned int mask)
134 {
135 int broadcast =
136 ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
137 LEON3_IRQMPSTATUS_BROADCAST) & 1);
138 if (!broadcast) {
139 prom_printf("######## !!!! The irqmp-ctrl must have broadcast enabled, smp wont work !!!!! ####### nr cpus: %d\n",
140 leon_smp_nrcpus());
141 if (leon_smp_nrcpus() > 1) {
142 BUG();
143 } else {
144 prom_printf("continue anyway\n");
145 return;
146 }
147 }
148 LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpbroadcast), mask);
149 }
150
leon_smp_nrcpus(void)151 int leon_smp_nrcpus(void)
152 {
153 int nrcpu =
154 ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
155 LEON3_IRQMPSTATUS_CPUNR) & 0xf) + 1;
156 return nrcpu;
157 }
158
leon_boot_cpus(void)159 void __init leon_boot_cpus(void)
160 {
161 int nrcpu = leon_smp_nrcpus();
162 int me = smp_processor_id();
163
164 /* Setup IPI */
165 leon_ipi_init();
166
167 printk(KERN_INFO "%d:(%d:%d) cpus mpirq at 0x%x\n", (unsigned int)me,
168 (unsigned int)nrcpu, (unsigned int)NR_CPUS,
169 (unsigned int)&(leon3_irqctrl_regs->mpstatus));
170
171 leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, me);
172 leon_enable_irq_cpu(LEON3_IRQ_TICKER, me);
173 leon_enable_irq_cpu(leon_ipi_irq, me);
174
175 leon_smp_setbroadcast(1 << LEON3_IRQ_TICKER);
176
177 leon_configure_cache_smp();
178 local_ops->cache_all();
179
180 }
181
leon_boot_one_cpu(int i,struct task_struct * idle)182 int leon_boot_one_cpu(int i, struct task_struct *idle)
183 {
184 int timeout;
185
186 current_set[i] = task_thread_info(idle);
187
188 /* See trampoline.S:leon_smp_cpu_startup for details...
189 * Initialize the contexts table
190 * Since the call to prom_startcpu() trashes the structure,
191 * we need to re-initialize it for each cpu
192 */
193 smp_penguin_ctable.which_io = 0;
194 smp_penguin_ctable.phys_addr = (unsigned int)srmmu_ctx_table_phys;
195 smp_penguin_ctable.reg_size = 0;
196
197 /* whirrr, whirrr, whirrrrrrrrr... */
198 printk(KERN_INFO "Starting CPU %d : (irqmp: 0x%x)\n", (unsigned int)i,
199 (unsigned int)&leon3_irqctrl_regs->mpstatus);
200 local_ops->cache_all();
201
202 /* Make sure all IRQs are of from the start for this new CPU */
203 LEON_BYPASS_STORE_PA(&leon3_irqctrl_regs->mask[i], 0);
204
205 /* Wake one CPU */
206 LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpstatus), 1 << i);
207
208 /* wheee... it's going... */
209 for (timeout = 0; timeout < 10000; timeout++) {
210 if (cpu_callin_map[i])
211 break;
212 udelay(200);
213 }
214 printk(KERN_INFO "Started CPU %d\n", (unsigned int)i);
215
216 if (!(cpu_callin_map[i])) {
217 printk(KERN_ERR "Processor %d is stuck.\n", i);
218 return -ENODEV;
219 } else {
220 leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, i);
221 leon_enable_irq_cpu(LEON3_IRQ_TICKER, i);
222 leon_enable_irq_cpu(leon_ipi_irq, i);
223 }
224
225 local_ops->cache_all();
226 return 0;
227 }
228
leon_smp_done(void)229 void __init leon_smp_done(void)
230 {
231
232 int i, first;
233 int *prev;
234
235 /* setup cpu list for irq rotation */
236 first = 0;
237 prev = &first;
238 for (i = 0; i < NR_CPUS; i++) {
239 if (cpu_online(i)) {
240 *prev = i;
241 prev = &cpu_data(i).next;
242 }
243 }
244 *prev = first;
245 local_ops->cache_all();
246
247 /* Free unneeded trap tables */
248 if (!cpu_present(1)) {
249 free_reserved_page(virt_to_page(&trapbase_cpu1));
250 }
251 if (!cpu_present(2)) {
252 free_reserved_page(virt_to_page(&trapbase_cpu2));
253 }
254 if (!cpu_present(3)) {
255 free_reserved_page(virt_to_page(&trapbase_cpu3));
256 }
257 /* Ok, they are spinning and ready to go. */
258 smp_processors_ready = 1;
259
260 }
261
262 struct leon_ipi_work {
263 int single;
264 int msk;
265 int resched;
266 };
267
268 static DEFINE_PER_CPU_SHARED_ALIGNED(struct leon_ipi_work, leon_ipi_work);
269
270 /* Initialize IPIs on the LEON, in order to save IRQ resources only one IRQ
271 * is used for all three types of IPIs.
272 */
leon_ipi_init(void)273 static void __init leon_ipi_init(void)
274 {
275 int cpu, len;
276 struct leon_ipi_work *work;
277 struct property *pp;
278 struct device_node *rootnp;
279 struct tt_entry *trap_table;
280 unsigned long flags;
281
282 /* Find IPI IRQ or stick with default value */
283 rootnp = of_find_node_by_path("/ambapp0");
284 if (rootnp) {
285 pp = of_find_property(rootnp, "ipi_num", &len);
286 if (pp && (*(int *)pp->value))
287 leon_ipi_irq = *(int *)pp->value;
288 }
289 printk(KERN_INFO "leon: SMP IPIs at IRQ %d\n", leon_ipi_irq);
290
291 /* Adjust so that we jump directly to smpleon_ipi */
292 local_irq_save(flags);
293 trap_table = &sparc_ttable[SP_TRAP_IRQ1 + (leon_ipi_irq - 1)];
294 trap_table->inst_three += smpleon_ipi - real_irq_entry;
295 local_ops->cache_all();
296 local_irq_restore(flags);
297
298 for_each_possible_cpu(cpu) {
299 work = &per_cpu(leon_ipi_work, cpu);
300 work->single = work->msk = work->resched = 0;
301 }
302 }
303
leon_send_ipi(int cpu,int level)304 static void leon_send_ipi(int cpu, int level)
305 {
306 unsigned long mask;
307 mask = leon_get_irqmask(level);
308 LEON3_BYPASS_STORE_PA(&leon3_irqctrl_regs->force[cpu], mask);
309 }
310
leon_ipi_single(int cpu)311 static void leon_ipi_single(int cpu)
312 {
313 struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
314
315 /* Mark work */
316 work->single = 1;
317
318 /* Generate IRQ on the CPU */
319 leon_send_ipi(cpu, leon_ipi_irq);
320 }
321
leon_ipi_mask_one(int cpu)322 static void leon_ipi_mask_one(int cpu)
323 {
324 struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
325
326 /* Mark work */
327 work->msk = 1;
328
329 /* Generate IRQ on the CPU */
330 leon_send_ipi(cpu, leon_ipi_irq);
331 }
332
leon_ipi_resched(int cpu)333 static void leon_ipi_resched(int cpu)
334 {
335 struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
336
337 /* Mark work */
338 work->resched = 1;
339
340 /* Generate IRQ on the CPU (any IRQ will cause resched) */
341 leon_send_ipi(cpu, leon_ipi_irq);
342 }
343
leonsmp_ipi_interrupt(void)344 void leonsmp_ipi_interrupt(void)
345 {
346 struct leon_ipi_work *work = this_cpu_ptr(&leon_ipi_work);
347
348 if (work->single) {
349 work->single = 0;
350 smp_call_function_single_interrupt();
351 }
352 if (work->msk) {
353 work->msk = 0;
354 smp_call_function_interrupt();
355 }
356 if (work->resched) {
357 work->resched = 0;
358 smp_resched_interrupt();
359 }
360 }
361
362 static struct smp_funcall {
363 smpfunc_t func;
364 unsigned long arg1;
365 unsigned long arg2;
366 unsigned long arg3;
367 unsigned long arg4;
368 unsigned long arg5;
369 unsigned long processors_in[NR_CPUS]; /* Set when ipi entered. */
370 unsigned long processors_out[NR_CPUS]; /* Set when ipi exited. */
371 } ccall_info __attribute__((aligned(8)));
372
373 static DEFINE_SPINLOCK(cross_call_lock);
374
375 /* Cross calls must be serialized, at least currently. */
leon_cross_call(smpfunc_t func,cpumask_t mask,unsigned long arg1,unsigned long arg2,unsigned long arg3,unsigned long arg4)376 static void leon_cross_call(smpfunc_t func, cpumask_t mask, unsigned long arg1,
377 unsigned long arg2, unsigned long arg3,
378 unsigned long arg4)
379 {
380 if (smp_processors_ready) {
381 register int high = NR_CPUS - 1;
382 unsigned long flags;
383
384 spin_lock_irqsave(&cross_call_lock, flags);
385
386 {
387 /* If you make changes here, make sure gcc generates proper code... */
388 register smpfunc_t f asm("i0") = func;
389 register unsigned long a1 asm("i1") = arg1;
390 register unsigned long a2 asm("i2") = arg2;
391 register unsigned long a3 asm("i3") = arg3;
392 register unsigned long a4 asm("i4") = arg4;
393 register unsigned long a5 asm("i5") = 0;
394
395 __asm__ __volatile__("std %0, [%6]\n\t"
396 "std %2, [%6 + 8]\n\t"
397 "std %4, [%6 + 16]\n\t" : :
398 "r"(f), "r"(a1), "r"(a2), "r"(a3),
399 "r"(a4), "r"(a5),
400 "r"(&ccall_info.func));
401 }
402
403 /* Init receive/complete mapping, plus fire the IPI's off. */
404 {
405 register int i;
406
407 cpumask_clear_cpu(smp_processor_id(), &mask);
408 cpumask_and(&mask, cpu_online_mask, &mask);
409 for (i = 0; i <= high; i++) {
410 if (cpumask_test_cpu(i, &mask)) {
411 ccall_info.processors_in[i] = 0;
412 ccall_info.processors_out[i] = 0;
413 leon_send_ipi(i, LEON3_IRQ_CROSS_CALL);
414
415 }
416 }
417 }
418
419 {
420 register int i;
421
422 i = 0;
423 do {
424 if (!cpumask_test_cpu(i, &mask))
425 continue;
426
427 while (!ccall_info.processors_in[i])
428 barrier();
429 } while (++i <= high);
430
431 i = 0;
432 do {
433 if (!cpumask_test_cpu(i, &mask))
434 continue;
435
436 while (!ccall_info.processors_out[i])
437 barrier();
438 } while (++i <= high);
439 }
440
441 spin_unlock_irqrestore(&cross_call_lock, flags);
442 }
443 }
444
445 /* Running cross calls. */
leon_cross_call_irq(void)446 void leon_cross_call_irq(void)
447 {
448 int i = smp_processor_id();
449
450 ccall_info.processors_in[i] = 1;
451 ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
452 ccall_info.arg4, ccall_info.arg5);
453 ccall_info.processors_out[i] = 1;
454 }
455
456 static const struct sparc32_ipi_ops leon_ipi_ops = {
457 .cross_call = leon_cross_call,
458 .resched = leon_ipi_resched,
459 .single = leon_ipi_single,
460 .mask_one = leon_ipi_mask_one,
461 };
462
leon_init_smp(void)463 void __init leon_init_smp(void)
464 {
465 /* Patch ipi15 trap table */
466 t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_leon - linux_trap_ipi15_sun4m);
467
468 sparc32_ipi_ops = &leon_ipi_ops;
469 }
470