0f3fb6d9c8efc3e8cd0f12e48d3ff26d87a6699f
[~shefty/rdma-dev.git] / arch / sparc / kernel / leon_smp.c
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 __cpuinit 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
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
72 void __cpuinit leon_callin(void)
73 {
74         int cpuid = hard_smp_processor_id();
75
76         local_ops->cache_all();
77         local_ops->tlb_all();
78         leon_configure_cache_smp();
79
80         notify_cpu_starting(cpuid);
81
82         /* Get our local ticker going. */
83         register_percpu_ce(cpuid);
84
85         calibrate_delay();
86         smp_store_cpu_info(cpuid);
87
88         local_ops->cache_all();
89         local_ops->tlb_all();
90
91         /*
92          * Unblock the master CPU _only_ when the scheduler state
93          * of all secondary CPUs will be up-to-date, so after
94          * the SMP initialization the master will be just allowed
95          * to call the scheduler code.
96          * Allow master to continue.
97          */
98         do_swap(&cpu_callin_map[cpuid], 1);
99
100         local_ops->cache_all();
101         local_ops->tlb_all();
102
103         /* Fix idle thread fields. */
104         __asm__ __volatile__("ld [%0], %%g6\n\t" : : "r"(&current_set[cpuid])
105                              : "memory" /* paranoid */);
106
107         /* Attach to the address space of init_task. */
108         atomic_inc(&init_mm.mm_count);
109         current->active_mm = &init_mm;
110
111         while (!cpumask_test_cpu(cpuid, &smp_commenced_mask))
112                 mb();
113
114         local_irq_enable();
115         set_cpu_online(cpuid, true);
116 }
117
118 /*
119  *      Cycle through the processors asking the PROM to start each one.
120  */
121
122 extern struct linux_prom_registers smp_penguin_ctable;
123
124 void __cpuinit leon_configure_cache_smp(void)
125 {
126         unsigned long cfg = sparc_leon3_get_dcachecfg();
127         int me = smp_processor_id();
128
129         if (ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg) > 4) {
130                 printk(KERN_INFO "Note: SMP with snooping only works on 4k cache, found %dk(0x%x) on cpu %d, disabling caches\n",
131                      (unsigned int)ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg),
132                      (unsigned int)cfg, (unsigned int)me);
133                 sparc_leon3_disable_cache();
134         } else {
135                 if (cfg & ASI_LEON3_SYSCTRL_CFG_SNOOPING) {
136                         sparc_leon3_enable_snooping();
137                 } else {
138                         printk(KERN_INFO "Note: You have to enable snooping in the vhdl model cpu %d, disabling caches\n",
139                              me);
140                         sparc_leon3_disable_cache();
141                 }
142         }
143
144         local_ops->cache_all();
145         local_ops->tlb_all();
146 }
147
148 void leon_smp_setbroadcast(unsigned int mask)
149 {
150         int broadcast =
151             ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
152               LEON3_IRQMPSTATUS_BROADCAST) & 1);
153         if (!broadcast) {
154                 prom_printf("######## !!!! The irqmp-ctrl must have broadcast enabled, smp wont work !!!!! ####### nr cpus: %d\n",
155                      leon_smp_nrcpus());
156                 if (leon_smp_nrcpus() > 1) {
157                         BUG();
158                 } else {
159                         prom_printf("continue anyway\n");
160                         return;
161                 }
162         }
163         LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpbroadcast), mask);
164 }
165
166 unsigned int leon_smp_getbroadcast(void)
167 {
168         unsigned int mask;
169         mask = LEON_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpbroadcast));
170         return mask;
171 }
172
173 int leon_smp_nrcpus(void)
174 {
175         int nrcpu =
176             ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
177               LEON3_IRQMPSTATUS_CPUNR) & 0xf) + 1;
178         return nrcpu;
179 }
180
181 void __init leon_boot_cpus(void)
182 {
183         int nrcpu = leon_smp_nrcpus();
184         int me = smp_processor_id();
185
186         /* Setup IPI */
187         leon_ipi_init();
188
189         printk(KERN_INFO "%d:(%d:%d) cpus mpirq at 0x%x\n", (unsigned int)me,
190                (unsigned int)nrcpu, (unsigned int)NR_CPUS,
191                (unsigned int)&(leon3_irqctrl_regs->mpstatus));
192
193         leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, me);
194         leon_enable_irq_cpu(LEON3_IRQ_TICKER, me);
195         leon_enable_irq_cpu(leon_ipi_irq, me);
196
197         leon_smp_setbroadcast(1 << LEON3_IRQ_TICKER);
198
199         leon_configure_cache_smp();
200         local_ops->cache_all();
201
202 }
203
204 int __cpuinit leon_boot_one_cpu(int i, struct task_struct *idle)
205 {
206         int timeout;
207
208         current_set[i] = task_thread_info(idle);
209
210         /* See trampoline.S:leon_smp_cpu_startup for details...
211          * Initialize the contexts table
212          * Since the call to prom_startcpu() trashes the structure,
213          * we need to re-initialize it for each cpu
214          */
215         smp_penguin_ctable.which_io = 0;
216         smp_penguin_ctable.phys_addr = (unsigned int)srmmu_ctx_table_phys;
217         smp_penguin_ctable.reg_size = 0;
218
219         /* whirrr, whirrr, whirrrrrrrrr... */
220         printk(KERN_INFO "Starting CPU %d : (irqmp: 0x%x)\n", (unsigned int)i,
221                (unsigned int)&leon3_irqctrl_regs->mpstatus);
222         local_ops->cache_all();
223
224         /* Make sure all IRQs are of from the start for this new CPU */
225         LEON_BYPASS_STORE_PA(&leon3_irqctrl_regs->mask[i], 0);
226
227         /* Wake one CPU */
228         LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpstatus), 1 << i);
229
230         /* wheee... it's going... */
231         for (timeout = 0; timeout < 10000; timeout++) {
232                 if (cpu_callin_map[i])
233                         break;
234                 udelay(200);
235         }
236         printk(KERN_INFO "Started CPU %d\n", (unsigned int)i);
237
238         if (!(cpu_callin_map[i])) {
239                 printk(KERN_ERR "Processor %d is stuck.\n", i);
240                 return -ENODEV;
241         } else {
242                 leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, i);
243                 leon_enable_irq_cpu(LEON3_IRQ_TICKER, i);
244                 leon_enable_irq_cpu(leon_ipi_irq, i);
245         }
246
247         local_ops->cache_all();
248         return 0;
249 }
250
251 void __init leon_smp_done(void)
252 {
253
254         int i, first;
255         int *prev;
256
257         /* setup cpu list for irq rotation */
258         first = 0;
259         prev = &first;
260         for (i = 0; i < NR_CPUS; i++) {
261                 if (cpu_online(i)) {
262                         *prev = i;
263                         prev = &cpu_data(i).next;
264                 }
265         }
266         *prev = first;
267         local_ops->cache_all();
268
269         /* Free unneeded trap tables */
270         if (!cpu_present(1)) {
271                 ClearPageReserved(virt_to_page(&trapbase_cpu1));
272                 init_page_count(virt_to_page(&trapbase_cpu1));
273                 free_page((unsigned long)&trapbase_cpu1);
274                 totalram_pages++;
275                 num_physpages++;
276         }
277         if (!cpu_present(2)) {
278                 ClearPageReserved(virt_to_page(&trapbase_cpu2));
279                 init_page_count(virt_to_page(&trapbase_cpu2));
280                 free_page((unsigned long)&trapbase_cpu2);
281                 totalram_pages++;
282                 num_physpages++;
283         }
284         if (!cpu_present(3)) {
285                 ClearPageReserved(virt_to_page(&trapbase_cpu3));
286                 init_page_count(virt_to_page(&trapbase_cpu3));
287                 free_page((unsigned long)&trapbase_cpu3);
288                 totalram_pages++;
289                 num_physpages++;
290         }
291         /* Ok, they are spinning and ready to go. */
292         smp_processors_ready = 1;
293
294 }
295
296 void leon_irq_rotate(int cpu)
297 {
298 }
299
300 struct leon_ipi_work {
301         int single;
302         int msk;
303         int resched;
304 };
305
306 static DEFINE_PER_CPU_SHARED_ALIGNED(struct leon_ipi_work, leon_ipi_work);
307
308 /* Initialize IPIs on the LEON, in order to save IRQ resources only one IRQ
309  * is used for all three types of IPIs.
310  */
311 static void __init leon_ipi_init(void)
312 {
313         int cpu, len;
314         struct leon_ipi_work *work;
315         struct property *pp;
316         struct device_node *rootnp;
317         struct tt_entry *trap_table;
318         unsigned long flags;
319
320         /* Find IPI IRQ or stick with default value */
321         rootnp = of_find_node_by_path("/ambapp0");
322         if (rootnp) {
323                 pp = of_find_property(rootnp, "ipi_num", &len);
324                 if (pp && (*(int *)pp->value))
325                         leon_ipi_irq = *(int *)pp->value;
326         }
327         printk(KERN_INFO "leon: SMP IPIs at IRQ %d\n", leon_ipi_irq);
328
329         /* Adjust so that we jump directly to smpleon_ipi */
330         local_irq_save(flags);
331         trap_table = &sparc_ttable[SP_TRAP_IRQ1 + (leon_ipi_irq - 1)];
332         trap_table->inst_three += smpleon_ipi - real_irq_entry;
333         local_ops->cache_all();
334         local_irq_restore(flags);
335
336         for_each_possible_cpu(cpu) {
337                 work = &per_cpu(leon_ipi_work, cpu);
338                 work->single = work->msk = work->resched = 0;
339         }
340 }
341
342 static void leon_send_ipi(int cpu, int level)
343 {
344         unsigned long mask;
345         mask = leon_get_irqmask(level);
346         LEON3_BYPASS_STORE_PA(&leon3_irqctrl_regs->force[cpu], mask);
347 }
348
349 static void leon_ipi_single(int cpu)
350 {
351         struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
352
353         /* Mark work */
354         work->single = 1;
355
356         /* Generate IRQ on the CPU */
357         leon_send_ipi(cpu, leon_ipi_irq);
358 }
359
360 static void leon_ipi_mask_one(int cpu)
361 {
362         struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
363
364         /* Mark work */
365         work->msk = 1;
366
367         /* Generate IRQ on the CPU */
368         leon_send_ipi(cpu, leon_ipi_irq);
369 }
370
371 static void leon_ipi_resched(int cpu)
372 {
373         struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
374
375         /* Mark work */
376         work->resched = 1;
377
378         /* Generate IRQ on the CPU (any IRQ will cause resched) */
379         leon_send_ipi(cpu, leon_ipi_irq);
380 }
381
382 void leonsmp_ipi_interrupt(void)
383 {
384         struct leon_ipi_work *work = &__get_cpu_var(leon_ipi_work);
385
386         if (work->single) {
387                 work->single = 0;
388                 smp_call_function_single_interrupt();
389         }
390         if (work->msk) {
391                 work->msk = 0;
392                 smp_call_function_interrupt();
393         }
394         if (work->resched) {
395                 work->resched = 0;
396                 smp_resched_interrupt();
397         }
398 }
399
400 static struct smp_funcall {
401         smpfunc_t func;
402         unsigned long arg1;
403         unsigned long arg2;
404         unsigned long arg3;
405         unsigned long arg4;
406         unsigned long arg5;
407         unsigned long processors_in[NR_CPUS];   /* Set when ipi entered. */
408         unsigned long processors_out[NR_CPUS];  /* Set when ipi exited. */
409 } ccall_info;
410
411 static DEFINE_SPINLOCK(cross_call_lock);
412
413 /* Cross calls must be serialized, at least currently. */
414 static void leon_cross_call(smpfunc_t func, cpumask_t mask, unsigned long arg1,
415                             unsigned long arg2, unsigned long arg3,
416                             unsigned long arg4)
417 {
418         if (smp_processors_ready) {
419                 register int high = NR_CPUS - 1;
420                 unsigned long flags;
421
422                 spin_lock_irqsave(&cross_call_lock, flags);
423
424                 {
425                         /* If you make changes here, make sure gcc generates proper code... */
426                         register smpfunc_t f asm("i0") = func;
427                         register unsigned long a1 asm("i1") = arg1;
428                         register unsigned long a2 asm("i2") = arg2;
429                         register unsigned long a3 asm("i3") = arg3;
430                         register unsigned long a4 asm("i4") = arg4;
431                         register unsigned long a5 asm("i5") = 0;
432
433                         __asm__ __volatile__("std %0, [%6]\n\t"
434                                              "std %2, [%6 + 8]\n\t"
435                                              "std %4, [%6 + 16]\n\t" : :
436                                              "r"(f), "r"(a1), "r"(a2), "r"(a3),
437                                              "r"(a4), "r"(a5),
438                                              "r"(&ccall_info.func));
439                 }
440
441                 /* Init receive/complete mapping, plus fire the IPI's off. */
442                 {
443                         register int i;
444
445                         cpumask_clear_cpu(smp_processor_id(), &mask);
446                         cpumask_and(&mask, cpu_online_mask, &mask);
447                         for (i = 0; i <= high; i++) {
448                                 if (cpumask_test_cpu(i, &mask)) {
449                                         ccall_info.processors_in[i] = 0;
450                                         ccall_info.processors_out[i] = 0;
451                                         leon_send_ipi(i, LEON3_IRQ_CROSS_CALL);
452
453                                 }
454                         }
455                 }
456
457                 {
458                         register int i;
459
460                         i = 0;
461                         do {
462                                 if (!cpumask_test_cpu(i, &mask))
463                                         continue;
464
465                                 while (!ccall_info.processors_in[i])
466                                         barrier();
467                         } while (++i <= high);
468
469                         i = 0;
470                         do {
471                                 if (!cpumask_test_cpu(i, &mask))
472                                         continue;
473
474                                 while (!ccall_info.processors_out[i])
475                                         barrier();
476                         } while (++i <= high);
477                 }
478
479                 spin_unlock_irqrestore(&cross_call_lock, flags);
480         }
481 }
482
483 /* Running cross calls. */
484 void leon_cross_call_irq(void)
485 {
486         int i = smp_processor_id();
487
488         ccall_info.processors_in[i] = 1;
489         ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
490                         ccall_info.arg4, ccall_info.arg5);
491         ccall_info.processors_out[i] = 1;
492 }
493
494 static const struct sparc32_ipi_ops leon_ipi_ops = {
495         .cross_call = leon_cross_call,
496         .resched    = leon_ipi_resched,
497         .single     = leon_ipi_single,
498         .mask_one   = leon_ipi_mask_one,
499 };
500
501 void __init leon_init_smp(void)
502 {
503         /* Patch ipi15 trap table */
504         t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_leon - linux_trap_ipi15_sun4m);
505
506         sparc32_ipi_ops = &leon_ipi_ops;
507 }