]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - arch/sparc/kernel/leon_smp.c
sparc32: refactor smp boot
[~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_cpu_pre_starting(void *arg)
73 {
74         leon_configure_cache_smp();
75 }
76
77 void __cpuinit 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"(&current_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
109 void __cpuinit 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
133 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
151 unsigned int leon_smp_getbroadcast(void)
152 {
153         unsigned int mask;
154         mask = LEON_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpbroadcast));
155         return mask;
156 }
157
158 int leon_smp_nrcpus(void)
159 {
160         int nrcpu =
161             ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >>
162               LEON3_IRQMPSTATUS_CPUNR) & 0xf) + 1;
163         return nrcpu;
164 }
165
166 void __init leon_boot_cpus(void)
167 {
168         int nrcpu = leon_smp_nrcpus();
169         int me = smp_processor_id();
170
171         /* Setup IPI */
172         leon_ipi_init();
173
174         printk(KERN_INFO "%d:(%d:%d) cpus mpirq at 0x%x\n", (unsigned int)me,
175                (unsigned int)nrcpu, (unsigned int)NR_CPUS,
176                (unsigned int)&(leon3_irqctrl_regs->mpstatus));
177
178         leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, me);
179         leon_enable_irq_cpu(LEON3_IRQ_TICKER, me);
180         leon_enable_irq_cpu(leon_ipi_irq, me);
181
182         leon_smp_setbroadcast(1 << LEON3_IRQ_TICKER);
183
184         leon_configure_cache_smp();
185         local_ops->cache_all();
186
187 }
188
189 int __cpuinit leon_boot_one_cpu(int i, struct task_struct *idle)
190 {
191         int timeout;
192
193         current_set[i] = task_thread_info(idle);
194
195         /* See trampoline.S:leon_smp_cpu_startup for details...
196          * Initialize the contexts table
197          * Since the call to prom_startcpu() trashes the structure,
198          * we need to re-initialize it for each cpu
199          */
200         smp_penguin_ctable.which_io = 0;
201         smp_penguin_ctable.phys_addr = (unsigned int)srmmu_ctx_table_phys;
202         smp_penguin_ctable.reg_size = 0;
203
204         /* whirrr, whirrr, whirrrrrrrrr... */
205         printk(KERN_INFO "Starting CPU %d : (irqmp: 0x%x)\n", (unsigned int)i,
206                (unsigned int)&leon3_irqctrl_regs->mpstatus);
207         local_ops->cache_all();
208
209         /* Make sure all IRQs are of from the start for this new CPU */
210         LEON_BYPASS_STORE_PA(&leon3_irqctrl_regs->mask[i], 0);
211
212         /* Wake one CPU */
213         LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpstatus), 1 << i);
214
215         /* wheee... it's going... */
216         for (timeout = 0; timeout < 10000; timeout++) {
217                 if (cpu_callin_map[i])
218                         break;
219                 udelay(200);
220         }
221         printk(KERN_INFO "Started CPU %d\n", (unsigned int)i);
222
223         if (!(cpu_callin_map[i])) {
224                 printk(KERN_ERR "Processor %d is stuck.\n", i);
225                 return -ENODEV;
226         } else {
227                 leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, i);
228                 leon_enable_irq_cpu(LEON3_IRQ_TICKER, i);
229                 leon_enable_irq_cpu(leon_ipi_irq, i);
230         }
231
232         local_ops->cache_all();
233         return 0;
234 }
235
236 void __init leon_smp_done(void)
237 {
238
239         int i, first;
240         int *prev;
241
242         /* setup cpu list for irq rotation */
243         first = 0;
244         prev = &first;
245         for (i = 0; i < NR_CPUS; i++) {
246                 if (cpu_online(i)) {
247                         *prev = i;
248                         prev = &cpu_data(i).next;
249                 }
250         }
251         *prev = first;
252         local_ops->cache_all();
253
254         /* Free unneeded trap tables */
255         if (!cpu_present(1)) {
256                 ClearPageReserved(virt_to_page(&trapbase_cpu1));
257                 init_page_count(virt_to_page(&trapbase_cpu1));
258                 free_page((unsigned long)&trapbase_cpu1);
259                 totalram_pages++;
260                 num_physpages++;
261         }
262         if (!cpu_present(2)) {
263                 ClearPageReserved(virt_to_page(&trapbase_cpu2));
264                 init_page_count(virt_to_page(&trapbase_cpu2));
265                 free_page((unsigned long)&trapbase_cpu2);
266                 totalram_pages++;
267                 num_physpages++;
268         }
269         if (!cpu_present(3)) {
270                 ClearPageReserved(virt_to_page(&trapbase_cpu3));
271                 init_page_count(virt_to_page(&trapbase_cpu3));
272                 free_page((unsigned long)&trapbase_cpu3);
273                 totalram_pages++;
274                 num_physpages++;
275         }
276         /* Ok, they are spinning and ready to go. */
277         smp_processors_ready = 1;
278
279 }
280
281 void leon_irq_rotate(int cpu)
282 {
283 }
284
285 struct leon_ipi_work {
286         int single;
287         int msk;
288         int resched;
289 };
290
291 static DEFINE_PER_CPU_SHARED_ALIGNED(struct leon_ipi_work, leon_ipi_work);
292
293 /* Initialize IPIs on the LEON, in order to save IRQ resources only one IRQ
294  * is used for all three types of IPIs.
295  */
296 static void __init leon_ipi_init(void)
297 {
298         int cpu, len;
299         struct leon_ipi_work *work;
300         struct property *pp;
301         struct device_node *rootnp;
302         struct tt_entry *trap_table;
303         unsigned long flags;
304
305         /* Find IPI IRQ or stick with default value */
306         rootnp = of_find_node_by_path("/ambapp0");
307         if (rootnp) {
308                 pp = of_find_property(rootnp, "ipi_num", &len);
309                 if (pp && (*(int *)pp->value))
310                         leon_ipi_irq = *(int *)pp->value;
311         }
312         printk(KERN_INFO "leon: SMP IPIs at IRQ %d\n", leon_ipi_irq);
313
314         /* Adjust so that we jump directly to smpleon_ipi */
315         local_irq_save(flags);
316         trap_table = &sparc_ttable[SP_TRAP_IRQ1 + (leon_ipi_irq - 1)];
317         trap_table->inst_three += smpleon_ipi - real_irq_entry;
318         local_ops->cache_all();
319         local_irq_restore(flags);
320
321         for_each_possible_cpu(cpu) {
322                 work = &per_cpu(leon_ipi_work, cpu);
323                 work->single = work->msk = work->resched = 0;
324         }
325 }
326
327 static void leon_send_ipi(int cpu, int level)
328 {
329         unsigned long mask;
330         mask = leon_get_irqmask(level);
331         LEON3_BYPASS_STORE_PA(&leon3_irqctrl_regs->force[cpu], mask);
332 }
333
334 static void leon_ipi_single(int cpu)
335 {
336         struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
337
338         /* Mark work */
339         work->single = 1;
340
341         /* Generate IRQ on the CPU */
342         leon_send_ipi(cpu, leon_ipi_irq);
343 }
344
345 static void leon_ipi_mask_one(int cpu)
346 {
347         struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
348
349         /* Mark work */
350         work->msk = 1;
351
352         /* Generate IRQ on the CPU */
353         leon_send_ipi(cpu, leon_ipi_irq);
354 }
355
356 static void leon_ipi_resched(int cpu)
357 {
358         struct leon_ipi_work *work = &per_cpu(leon_ipi_work, cpu);
359
360         /* Mark work */
361         work->resched = 1;
362
363         /* Generate IRQ on the CPU (any IRQ will cause resched) */
364         leon_send_ipi(cpu, leon_ipi_irq);
365 }
366
367 void leonsmp_ipi_interrupt(void)
368 {
369         struct leon_ipi_work *work = &__get_cpu_var(leon_ipi_work);
370
371         if (work->single) {
372                 work->single = 0;
373                 smp_call_function_single_interrupt();
374         }
375         if (work->msk) {
376                 work->msk = 0;
377                 smp_call_function_interrupt();
378         }
379         if (work->resched) {
380                 work->resched = 0;
381                 smp_resched_interrupt();
382         }
383 }
384
385 static struct smp_funcall {
386         smpfunc_t func;
387         unsigned long arg1;
388         unsigned long arg2;
389         unsigned long arg3;
390         unsigned long arg4;
391         unsigned long arg5;
392         unsigned long processors_in[NR_CPUS];   /* Set when ipi entered. */
393         unsigned long processors_out[NR_CPUS];  /* Set when ipi exited. */
394 } ccall_info;
395
396 static DEFINE_SPINLOCK(cross_call_lock);
397
398 /* Cross calls must be serialized, at least currently. */
399 static void leon_cross_call(smpfunc_t func, cpumask_t mask, unsigned long arg1,
400                             unsigned long arg2, unsigned long arg3,
401                             unsigned long arg4)
402 {
403         if (smp_processors_ready) {
404                 register int high = NR_CPUS - 1;
405                 unsigned long flags;
406
407                 spin_lock_irqsave(&cross_call_lock, flags);
408
409                 {
410                         /* If you make changes here, make sure gcc generates proper code... */
411                         register smpfunc_t f asm("i0") = func;
412                         register unsigned long a1 asm("i1") = arg1;
413                         register unsigned long a2 asm("i2") = arg2;
414                         register unsigned long a3 asm("i3") = arg3;
415                         register unsigned long a4 asm("i4") = arg4;
416                         register unsigned long a5 asm("i5") = 0;
417
418                         __asm__ __volatile__("std %0, [%6]\n\t"
419                                              "std %2, [%6 + 8]\n\t"
420                                              "std %4, [%6 + 16]\n\t" : :
421                                              "r"(f), "r"(a1), "r"(a2), "r"(a3),
422                                              "r"(a4), "r"(a5),
423                                              "r"(&ccall_info.func));
424                 }
425
426                 /* Init receive/complete mapping, plus fire the IPI's off. */
427                 {
428                         register int i;
429
430                         cpumask_clear_cpu(smp_processor_id(), &mask);
431                         cpumask_and(&mask, cpu_online_mask, &mask);
432                         for (i = 0; i <= high; i++) {
433                                 if (cpumask_test_cpu(i, &mask)) {
434                                         ccall_info.processors_in[i] = 0;
435                                         ccall_info.processors_out[i] = 0;
436                                         leon_send_ipi(i, LEON3_IRQ_CROSS_CALL);
437
438                                 }
439                         }
440                 }
441
442                 {
443                         register int i;
444
445                         i = 0;
446                         do {
447                                 if (!cpumask_test_cpu(i, &mask))
448                                         continue;
449
450                                 while (!ccall_info.processors_in[i])
451                                         barrier();
452                         } while (++i <= high);
453
454                         i = 0;
455                         do {
456                                 if (!cpumask_test_cpu(i, &mask))
457                                         continue;
458
459                                 while (!ccall_info.processors_out[i])
460                                         barrier();
461                         } while (++i <= high);
462                 }
463
464                 spin_unlock_irqrestore(&cross_call_lock, flags);
465         }
466 }
467
468 /* Running cross calls. */
469 void leon_cross_call_irq(void)
470 {
471         int i = smp_processor_id();
472
473         ccall_info.processors_in[i] = 1;
474         ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
475                         ccall_info.arg4, ccall_info.arg5);
476         ccall_info.processors_out[i] = 1;
477 }
478
479 static const struct sparc32_ipi_ops leon_ipi_ops = {
480         .cross_call = leon_cross_call,
481         .resched    = leon_ipi_resched,
482         .single     = leon_ipi_single,
483         .mask_one   = leon_ipi_mask_one,
484 };
485
486 void __init leon_init_smp(void)
487 {
488         /* Patch ipi15 trap table */
489         t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_leon - linux_trap_ipi15_sun4m);
490
491         sparc32_ipi_ops = &leon_ipi_ops;
492 }