Commit | Line | Data |
---|---|---|
8401707f KE |
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> | |
8401707f KE |
15 | #include <linux/interrupt.h> |
16 | #include <linux/kernel_stat.h> | |
17 | #include <linux/init.h> | |
18 | #include <linux/spinlock.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/swap.h> | |
21 | #include <linux/profile.h> | |
22 | #include <linux/pm.h> | |
23 | #include <linux/delay.h> | |
5a0e3ad6 | 24 | #include <linux/gfp.h> |
8401707f KE |
25 | |
26 | #include <asm/cacheflush.h> | |
27 | #include <asm/tlbflush.h> | |
28 | ||
29 | #include <asm/ptrace.h> | |
30 | #include <asm/atomic.h> | |
31 | #include <asm/irq_regs.h> | |
32 | ||
33 | #include <asm/delay.h> | |
34 | #include <asm/irq.h> | |
35 | #include <asm/page.h> | |
36 | #include <asm/pgalloc.h> | |
37 | #include <asm/pgtable.h> | |
38 | #include <asm/oplib.h> | |
39 | #include <asm/cpudata.h> | |
40 | #include <asm/asi.h> | |
41 | #include <asm/leon.h> | |
42 | #include <asm/leon_amba.h> | |
43 | ||
a2a211cb SR |
44 | #include "kernel.h" |
45 | ||
8401707f KE |
46 | #ifdef CONFIG_SPARC_LEON |
47 | ||
48 | #include "irq.h" | |
49 | ||
50 | extern ctxd_t *srmmu_ctx_table_phys; | |
51 | static int smp_processors_ready; | |
52 | extern volatile unsigned long cpu_callin_map[NR_CPUS]; | |
53 | extern unsigned char boot_cpu_id; | |
54 | extern cpumask_t smp_commenced_mask; | |
55 | void __init leon_configure_cache_smp(void); | |
56 | ||
57 | static inline unsigned long do_swap(volatile unsigned long *ptr, | |
58 | unsigned long val) | |
59 | { | |
502279a7 DH |
60 | __asm__ __volatile__("swapa [%2] %3, %0\n\t" : "=&r"(val) |
61 | : "0"(val), "r"(ptr), "i"(ASI_LEON_DCACHE_MISS) | |
8401707f KE |
62 | : "memory"); |
63 | return val; | |
64 | } | |
65 | ||
66 | static void smp_setup_percpu_timer(void); | |
67 | ||
68 | void __cpuinit leon_callin(void) | |
69 | { | |
70 | int cpuid = hard_smpleon_processor_id(); | |
71 | ||
72 | local_flush_cache_all(); | |
73 | local_flush_tlb_all(); | |
74 | leon_configure_cache_smp(); | |
75 | ||
76 | /* Get our local ticker going. */ | |
77 | smp_setup_percpu_timer(); | |
78 | ||
79 | calibrate_delay(); | |
80 | smp_store_cpu_info(cpuid); | |
81 | ||
82 | local_flush_cache_all(); | |
83 | local_flush_tlb_all(); | |
84 | ||
85 | /* | |
86 | * Unblock the master CPU _only_ when the scheduler state | |
87 | * of all secondary CPUs will be up-to-date, so after | |
88 | * the SMP initialization the master will be just allowed | |
89 | * to call the scheduler code. | |
90 | * Allow master to continue. | |
91 | */ | |
92 | do_swap(&cpu_callin_map[cpuid], 1); | |
93 | ||
94 | local_flush_cache_all(); | |
95 | local_flush_tlb_all(); | |
96 | ||
97 | cpu_probe(); | |
98 | ||
99 | /* Fix idle thread fields. */ | |
100 | __asm__ __volatile__("ld [%0], %%g6\n\t" : : "r"(¤t_set[cpuid]) | |
101 | : "memory" /* paranoid */); | |
102 | ||
103 | /* Attach to the address space of init_task. */ | |
104 | atomic_inc(&init_mm.mm_count); | |
105 | current->active_mm = &init_mm; | |
106 | ||
107 | while (!cpu_isset(cpuid, smp_commenced_mask)) | |
108 | mb(); | |
109 | ||
110 | local_irq_enable(); | |
111 | cpu_set(cpuid, cpu_online_map); | |
112 | } | |
113 | ||
114 | /* | |
115 | * Cycle through the processors asking the PROM to start each one. | |
116 | */ | |
117 | ||
118 | extern struct linux_prom_registers smp_penguin_ctable; | |
119 | ||
120 | void __init leon_configure_cache_smp(void) | |
121 | { | |
122 | unsigned long cfg = sparc_leon3_get_dcachecfg(); | |
123 | int me = smp_processor_id(); | |
124 | ||
125 | if (ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg) > 4) { | |
126 | printk(KERN_INFO "Note: SMP with snooping only works on 4k cache, found %dk(0x%x) on cpu %d, disabling caches\n", | |
127 | (unsigned int)ASI_LEON3_SYSCTRL_CFG_SSIZE(cfg), | |
128 | (unsigned int)cfg, (unsigned int)me); | |
129 | sparc_leon3_disable_cache(); | |
130 | } else { | |
131 | if (cfg & ASI_LEON3_SYSCTRL_CFG_SNOOPING) { | |
132 | sparc_leon3_enable_snooping(); | |
133 | } else { | |
134 | printk(KERN_INFO "Note: You have to enable snooping in the vhdl model cpu %d, disabling caches\n", | |
135 | me); | |
136 | sparc_leon3_disable_cache(); | |
137 | } | |
138 | } | |
139 | ||
140 | local_flush_cache_all(); | |
141 | local_flush_tlb_all(); | |
142 | } | |
143 | ||
144 | void leon_smp_setbroadcast(unsigned int mask) | |
145 | { | |
146 | int broadcast = | |
147 | ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >> | |
148 | LEON3_IRQMPSTATUS_BROADCAST) & 1); | |
149 | if (!broadcast) { | |
150 | prom_printf("######## !!!! The irqmp-ctrl must have broadcast enabled, smp wont work !!!!! ####### nr cpus: %d\n", | |
151 | leon_smp_nrcpus()); | |
152 | if (leon_smp_nrcpus() > 1) { | |
153 | BUG(); | |
154 | } else { | |
155 | prom_printf("continue anyway\n"); | |
156 | return; | |
157 | } | |
158 | } | |
159 | LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpbroadcast), mask); | |
160 | } | |
161 | ||
162 | unsigned int leon_smp_getbroadcast(void) | |
163 | { | |
164 | unsigned int mask; | |
165 | mask = LEON_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpbroadcast)); | |
166 | return mask; | |
167 | } | |
168 | ||
169 | int leon_smp_nrcpus(void) | |
170 | { | |
171 | int nrcpu = | |
172 | ((LEON3_BYPASS_LOAD_PA(&(leon3_irqctrl_regs->mpstatus)) >> | |
173 | LEON3_IRQMPSTATUS_CPUNR) & 0xf) + 1; | |
174 | return nrcpu; | |
175 | } | |
176 | ||
177 | void __init leon_boot_cpus(void) | |
178 | { | |
179 | int nrcpu = leon_smp_nrcpus(); | |
180 | int me = smp_processor_id(); | |
181 | ||
0da2b300 | 182 | printk(KERN_INFO "%d:(%d:%d) cpus mpirq at 0x%x\n", (unsigned int)me, |
8401707f KE |
183 | (unsigned int)nrcpu, (unsigned int)NR_CPUS, |
184 | (unsigned int)&(leon3_irqctrl_regs->mpstatus)); | |
185 | ||
186 | leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, me); | |
187 | leon_enable_irq_cpu(LEON3_IRQ_TICKER, me); | |
188 | leon_enable_irq_cpu(LEON3_IRQ_RESCHEDULE, me); | |
189 | ||
190 | leon_smp_setbroadcast(1 << LEON3_IRQ_TICKER); | |
191 | ||
192 | leon_configure_cache_smp(); | |
193 | smp_setup_percpu_timer(); | |
194 | local_flush_cache_all(); | |
195 | ||
196 | } | |
197 | ||
198 | int __cpuinit leon_boot_one_cpu(int i) | |
199 | { | |
200 | ||
201 | struct task_struct *p; | |
202 | int timeout; | |
203 | ||
204 | /* Cook up an idler for this guy. */ | |
205 | p = fork_idle(i); | |
206 | ||
207 | current_set[i] = task_thread_info(p); | |
208 | ||
209 | /* See trampoline.S:leon_smp_cpu_startup for details... | |
210 | * Initialize the contexts table | |
211 | * Since the call to prom_startcpu() trashes the structure, | |
212 | * we need to re-initialize it for each cpu | |
213 | */ | |
214 | smp_penguin_ctable.which_io = 0; | |
215 | smp_penguin_ctable.phys_addr = (unsigned int)srmmu_ctx_table_phys; | |
216 | smp_penguin_ctable.reg_size = 0; | |
217 | ||
218 | /* whirrr, whirrr, whirrrrrrrrr... */ | |
219 | printk(KERN_INFO "Starting CPU %d : (irqmp: 0x%x)\n", (unsigned int)i, | |
220 | (unsigned int)&leon3_irqctrl_regs->mpstatus); | |
221 | local_flush_cache_all(); | |
222 | ||
223 | LEON_BYPASS_STORE_PA(&(leon3_irqctrl_regs->mpstatus), 1 << i); | |
224 | ||
225 | /* wheee... it's going... */ | |
226 | for (timeout = 0; timeout < 10000; timeout++) { | |
227 | if (cpu_callin_map[i]) | |
228 | break; | |
229 | udelay(200); | |
230 | } | |
0da2b300 | 231 | printk(KERN_INFO "Started CPU %d\n", (unsigned int)i); |
8401707f KE |
232 | |
233 | if (!(cpu_callin_map[i])) { | |
234 | printk(KERN_ERR "Processor %d is stuck.\n", i); | |
235 | return -ENODEV; | |
236 | } else { | |
237 | leon_enable_irq_cpu(LEON3_IRQ_CROSS_CALL, i); | |
238 | leon_enable_irq_cpu(LEON3_IRQ_TICKER, i); | |
239 | leon_enable_irq_cpu(LEON3_IRQ_RESCHEDULE, i); | |
240 | } | |
241 | ||
242 | local_flush_cache_all(); | |
243 | return 0; | |
244 | } | |
245 | ||
246 | void __init leon_smp_done(void) | |
247 | { | |
248 | ||
249 | int i, first; | |
250 | int *prev; | |
251 | ||
252 | /* setup cpu list for irq rotation */ | |
253 | first = 0; | |
254 | prev = &first; | |
255 | for (i = 0; i < NR_CPUS; i++) { | |
256 | if (cpu_online(i)) { | |
257 | *prev = i; | |
258 | prev = &cpu_data(i).next; | |
259 | } | |
260 | } | |
261 | *prev = first; | |
262 | local_flush_cache_all(); | |
263 | ||
264 | /* Free unneeded trap tables */ | |
265 | if (!cpu_isset(1, cpu_present_map)) { | |
a2a211cb SR |
266 | ClearPageReserved(virt_to_page(&trapbase_cpu1)); |
267 | init_page_count(virt_to_page(&trapbase_cpu1)); | |
268 | free_page((unsigned long)&trapbase_cpu1); | |
8401707f KE |
269 | totalram_pages++; |
270 | num_physpages++; | |
271 | } | |
272 | if (!cpu_isset(2, cpu_present_map)) { | |
a2a211cb SR |
273 | ClearPageReserved(virt_to_page(&trapbase_cpu2)); |
274 | init_page_count(virt_to_page(&trapbase_cpu2)); | |
275 | free_page((unsigned long)&trapbase_cpu2); | |
8401707f KE |
276 | totalram_pages++; |
277 | num_physpages++; | |
278 | } | |
279 | if (!cpu_isset(3, cpu_present_map)) { | |
a2a211cb SR |
280 | ClearPageReserved(virt_to_page(&trapbase_cpu3)); |
281 | init_page_count(virt_to_page(&trapbase_cpu3)); | |
282 | free_page((unsigned long)&trapbase_cpu3); | |
8401707f KE |
283 | totalram_pages++; |
284 | num_physpages++; | |
285 | } | |
286 | /* Ok, they are spinning and ready to go. */ | |
287 | smp_processors_ready = 1; | |
288 | ||
289 | } | |
290 | ||
291 | void leon_irq_rotate(int cpu) | |
292 | { | |
293 | } | |
294 | ||
295 | static struct smp_funcall { | |
296 | smpfunc_t func; | |
297 | unsigned long arg1; | |
298 | unsigned long arg2; | |
299 | unsigned long arg3; | |
300 | unsigned long arg4; | |
301 | unsigned long arg5; | |
302 | unsigned long processors_in[NR_CPUS]; /* Set when ipi entered. */ | |
303 | unsigned long processors_out[NR_CPUS]; /* Set when ipi exited. */ | |
304 | } ccall_info; | |
305 | ||
306 | static DEFINE_SPINLOCK(cross_call_lock); | |
307 | ||
308 | /* Cross calls must be serialized, at least currently. */ | |
309 | static void leon_cross_call(smpfunc_t func, cpumask_t mask, unsigned long arg1, | |
310 | unsigned long arg2, unsigned long arg3, | |
311 | unsigned long arg4) | |
312 | { | |
313 | if (smp_processors_ready) { | |
314 | register int high = NR_CPUS - 1; | |
315 | unsigned long flags; | |
316 | ||
317 | spin_lock_irqsave(&cross_call_lock, flags); | |
318 | ||
319 | { | |
320 | /* If you make changes here, make sure gcc generates proper code... */ | |
321 | register smpfunc_t f asm("i0") = func; | |
322 | register unsigned long a1 asm("i1") = arg1; | |
323 | register unsigned long a2 asm("i2") = arg2; | |
324 | register unsigned long a3 asm("i3") = arg3; | |
325 | register unsigned long a4 asm("i4") = arg4; | |
326 | register unsigned long a5 asm("i5") = 0; | |
327 | ||
328 | __asm__ __volatile__("std %0, [%6]\n\t" | |
329 | "std %2, [%6 + 8]\n\t" | |
330 | "std %4, [%6 + 16]\n\t" : : | |
331 | "r"(f), "r"(a1), "r"(a2), "r"(a3), | |
332 | "r"(a4), "r"(a5), | |
333 | "r"(&ccall_info.func)); | |
334 | } | |
335 | ||
336 | /* Init receive/complete mapping, plus fire the IPI's off. */ | |
337 | { | |
338 | register int i; | |
339 | ||
340 | cpu_clear(smp_processor_id(), mask); | |
341 | cpus_and(mask, cpu_online_map, mask); | |
342 | for (i = 0; i <= high; i++) { | |
343 | if (cpu_isset(i, mask)) { | |
344 | ccall_info.processors_in[i] = 0; | |
345 | ccall_info.processors_out[i] = 0; | |
346 | set_cpu_int(i, LEON3_IRQ_CROSS_CALL); | |
347 | ||
348 | } | |
349 | } | |
350 | } | |
351 | ||
352 | { | |
353 | register int i; | |
354 | ||
355 | i = 0; | |
356 | do { | |
357 | if (!cpu_isset(i, mask)) | |
358 | continue; | |
359 | ||
360 | while (!ccall_info.processors_in[i]) | |
361 | barrier(); | |
362 | } while (++i <= high); | |
363 | ||
364 | i = 0; | |
365 | do { | |
366 | if (!cpu_isset(i, mask)) | |
367 | continue; | |
368 | ||
369 | while (!ccall_info.processors_out[i]) | |
370 | barrier(); | |
371 | } while (++i <= high); | |
372 | } | |
373 | ||
374 | spin_unlock_irqrestore(&cross_call_lock, flags); | |
375 | } | |
376 | } | |
377 | ||
378 | /* Running cross calls. */ | |
379 | void leon_cross_call_irq(void) | |
380 | { | |
381 | int i = smp_processor_id(); | |
382 | ||
383 | ccall_info.processors_in[i] = 1; | |
384 | ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3, | |
385 | ccall_info.arg4, ccall_info.arg5); | |
386 | ccall_info.processors_out[i] = 1; | |
387 | } | |
388 | ||
389 | void leon_percpu_timer_interrupt(struct pt_regs *regs) | |
390 | { | |
391 | struct pt_regs *old_regs; | |
392 | int cpu = smp_processor_id(); | |
393 | ||
394 | old_regs = set_irq_regs(regs); | |
395 | ||
396 | leon_clear_profile_irq(cpu); | |
397 | ||
398 | profile_tick(CPU_PROFILING); | |
399 | ||
400 | if (!--prof_counter(cpu)) { | |
401 | int user = user_mode(regs); | |
402 | ||
403 | irq_enter(); | |
404 | update_process_times(user); | |
405 | irq_exit(); | |
406 | ||
407 | prof_counter(cpu) = prof_multiplier(cpu); | |
408 | } | |
409 | set_irq_regs(old_regs); | |
410 | } | |
411 | ||
412 | static void __init smp_setup_percpu_timer(void) | |
413 | { | |
414 | int cpu = smp_processor_id(); | |
415 | ||
416 | prof_counter(cpu) = prof_multiplier(cpu) = 1; | |
417 | } | |
418 | ||
419 | void __init leon_blackbox_id(unsigned *addr) | |
420 | { | |
421 | int rd = *addr & 0x3e000000; | |
422 | int rs1 = rd >> 11; | |
423 | ||
424 | /* patch places where ___b_hard_smp_processor_id appears */ | |
425 | addr[0] = 0x81444000 | rd; /* rd %asr17, reg */ | |
426 | addr[1] = 0x8130201c | rd | rs1; /* srl reg, 0x1c, reg */ | |
427 | addr[2] = 0x01000000; /* nop */ | |
428 | } | |
429 | ||
430 | void __init leon_blackbox_current(unsigned *addr) | |
431 | { | |
432 | int rd = *addr & 0x3e000000; | |
433 | int rs1 = rd >> 11; | |
434 | ||
435 | /* patch LOAD_CURRENT macro where ___b_load_current appears */ | |
436 | addr[0] = 0x81444000 | rd; /* rd %asr17, reg */ | |
437 | addr[2] = 0x8130201c | rd | rs1; /* srl reg, 0x1c, reg */ | |
438 | addr[4] = 0x81282002 | rd | rs1; /* sll reg, 0x2, reg */ | |
439 | ||
440 | } | |
441 | ||
8401707f KE |
442 | void __init leon_init_smp(void) |
443 | { | |
444 | /* Patch ipi15 trap table */ | |
445 | t_nmi[1] = t_nmi[1] + (linux_trap_ipi15_leon - linux_trap_ipi15_sun4m); | |
446 | ||
447 | BTFIXUPSET_BLACKBOX(hard_smp_processor_id, leon_blackbox_id); | |
448 | BTFIXUPSET_BLACKBOX(load_current, leon_blackbox_current); | |
449 | BTFIXUPSET_CALL(smp_cross_call, leon_cross_call, BTFIXUPCALL_NORM); | |
450 | BTFIXUPSET_CALL(__hard_smp_processor_id, __leon_processor_id, | |
451 | BTFIXUPCALL_NORM); | |
8401707f KE |
452 | } |
453 | ||
454 | #endif /* CONFIG_SPARC_LEON */ |