xref: /openbmc/linux/arch/sparc/kernel/sun4m_smp.c (revision 384740dc)
1 /* sun4m_smp.c: Sparc SUN4M SMP support.
2  *
3  * Copyright (C) 1996 David S. Miller (davem@caip.rutgers.edu)
4  */
5 
6 #include <asm/head.h>
7 
8 #include <linux/kernel.h>
9 #include <linux/sched.h>
10 #include <linux/threads.h>
11 #include <linux/smp.h>
12 #include <linux/interrupt.h>
13 #include <linux/kernel_stat.h>
14 #include <linux/init.h>
15 #include <linux/spinlock.h>
16 #include <linux/mm.h>
17 #include <linux/swap.h>
18 #include <linux/profile.h>
19 #include <linux/delay.h>
20 
21 #include <asm/cacheflush.h>
22 #include <asm/tlbflush.h>
23 #include <asm/irq_regs.h>
24 
25 #include <asm/ptrace.h>
26 #include <asm/atomic.h>
27 
28 #include <asm/irq.h>
29 #include <asm/page.h>
30 #include <asm/pgalloc.h>
31 #include <asm/pgtable.h>
32 #include <asm/oplib.h>
33 #include <asm/cpudata.h>
34 
35 #include "irq.h"
36 
37 #define IRQ_CROSS_CALL		15
38 
39 extern ctxd_t *srmmu_ctx_table_phys;
40 
41 extern volatile unsigned long cpu_callin_map[NR_CPUS];
42 extern unsigned char boot_cpu_id;
43 
44 extern cpumask_t smp_commenced_mask;
45 
46 extern int __smp4m_processor_id(void);
47 
48 /*#define SMP_DEBUG*/
49 
50 #ifdef SMP_DEBUG
51 #define SMP_PRINTK(x)	printk x
52 #else
53 #define SMP_PRINTK(x)
54 #endif
55 
56 static inline unsigned long swap(volatile unsigned long *ptr, unsigned long val)
57 {
58 	__asm__ __volatile__("swap [%1], %0\n\t" :
59 			     "=&r" (val), "=&r" (ptr) :
60 			     "0" (val), "1" (ptr));
61 	return val;
62 }
63 
64 static void smp_setup_percpu_timer(void);
65 extern void cpu_probe(void);
66 
67 void __cpuinit smp4m_callin(void)
68 {
69 	int cpuid = hard_smp_processor_id();
70 
71 	local_flush_cache_all();
72 	local_flush_tlb_all();
73 
74 	notify_cpu_starting(cpuid);
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 	 */
91 	/* Allow master to continue. */
92 	swap(&cpu_callin_map[cpuid], 1);
93 
94 	/* XXX: What's up with all the flushes? */
95 	local_flush_cache_all();
96 	local_flush_tlb_all();
97 
98 	cpu_probe();
99 
100 	/* Fix idle thread fields. */
101 	__asm__ __volatile__("ld [%0], %%g6\n\t"
102 			     : : "r" (&current_set[cpuid])
103 			     : "memory" /* paranoid */);
104 
105 	/* Attach to the address space of init_task. */
106 	atomic_inc(&init_mm.mm_count);
107 	current->active_mm = &init_mm;
108 
109 	while (!cpu_isset(cpuid, smp_commenced_mask))
110 		mb();
111 
112 	local_irq_enable();
113 
114 	cpu_set(cpuid, cpu_online_map);
115 }
116 
117 /*
118  *	Cycle through the processors asking the PROM to start each one.
119  */
120 
121 extern struct linux_prom_registers smp_penguin_ctable;
122 extern unsigned long trapbase_cpu1[];
123 extern unsigned long trapbase_cpu2[];
124 extern unsigned long trapbase_cpu3[];
125 
126 void __init smp4m_boot_cpus(void)
127 {
128 	smp_setup_percpu_timer();
129 	local_flush_cache_all();
130 }
131 
132 int __cpuinit smp4m_boot_one_cpu(int i)
133 {
134 	extern unsigned long sun4m_cpu_startup;
135 	unsigned long *entry = &sun4m_cpu_startup;
136 	struct task_struct *p;
137 	int timeout;
138 	int cpu_node;
139 
140 	cpu_find_by_mid(i, &cpu_node);
141 
142 	/* Cook up an idler for this guy. */
143 	p = fork_idle(i);
144 	current_set[i] = task_thread_info(p);
145 	/* See trampoline.S for details... */
146 	entry += ((i-1) * 3);
147 
148 	/*
149 	 * Initialize the contexts table
150 	 * Since the call to prom_startcpu() trashes the structure,
151 	 * we need to re-initialize it for each cpu
152 	 */
153 	smp_penguin_ctable.which_io = 0;
154 	smp_penguin_ctable.phys_addr = (unsigned int) srmmu_ctx_table_phys;
155 	smp_penguin_ctable.reg_size = 0;
156 
157 	/* whirrr, whirrr, whirrrrrrrrr... */
158 	printk("Starting CPU %d at %p\n", i, entry);
159 	local_flush_cache_all();
160 	prom_startcpu(cpu_node,
161 		      &smp_penguin_ctable, 0, (char *)entry);
162 
163 	/* wheee... it's going... */
164 	for(timeout = 0; timeout < 10000; timeout++) {
165 		if(cpu_callin_map[i])
166 			break;
167 		udelay(200);
168 	}
169 
170 	if (!(cpu_callin_map[i])) {
171 		printk("Processor %d is stuck.\n", i);
172 		return -ENODEV;
173 	}
174 
175 	local_flush_cache_all();
176 	return 0;
177 }
178 
179 void __init smp4m_smp_done(void)
180 {
181 	int i, first;
182 	int *prev;
183 
184 	/* setup cpu list for irq rotation */
185 	first = 0;
186 	prev = &first;
187 	for (i = 0; i < NR_CPUS; i++) {
188 		if (cpu_online(i)) {
189 			*prev = i;
190 			prev = &cpu_data(i).next;
191 		}
192 	}
193 	*prev = first;
194 	local_flush_cache_all();
195 
196 	/* Free unneeded trap tables */
197 	if (!cpu_isset(1, cpu_present_map)) {
198 		ClearPageReserved(virt_to_page(trapbase_cpu1));
199 		init_page_count(virt_to_page(trapbase_cpu1));
200 		free_page((unsigned long)trapbase_cpu1);
201 		totalram_pages++;
202 		num_physpages++;
203 	}
204 	if (!cpu_isset(2, cpu_present_map)) {
205 		ClearPageReserved(virt_to_page(trapbase_cpu2));
206 		init_page_count(virt_to_page(trapbase_cpu2));
207 		free_page((unsigned long)trapbase_cpu2);
208 		totalram_pages++;
209 		num_physpages++;
210 	}
211 	if (!cpu_isset(3, cpu_present_map)) {
212 		ClearPageReserved(virt_to_page(trapbase_cpu3));
213 		init_page_count(virt_to_page(trapbase_cpu3));
214 		free_page((unsigned long)trapbase_cpu3);
215 		totalram_pages++;
216 		num_physpages++;
217 	}
218 
219 	/* Ok, they are spinning and ready to go. */
220 }
221 
222 /* At each hardware IRQ, we get this called to forward IRQ reception
223  * to the next processor.  The caller must disable the IRQ level being
224  * serviced globally so that there are no double interrupts received.
225  *
226  * XXX See sparc64 irq.c.
227  */
228 void smp4m_irq_rotate(int cpu)
229 {
230 	int next = cpu_data(cpu).next;
231 	if (next != cpu)
232 		set_irq_udt(next);
233 }
234 
235 static struct smp_funcall {
236 	smpfunc_t func;
237 	unsigned long arg1;
238 	unsigned long arg2;
239 	unsigned long arg3;
240 	unsigned long arg4;
241 	unsigned long arg5;
242 	unsigned long processors_in[SUN4M_NCPUS];  /* Set when ipi entered. */
243 	unsigned long processors_out[SUN4M_NCPUS]; /* Set when ipi exited. */
244 } ccall_info;
245 
246 static DEFINE_SPINLOCK(cross_call_lock);
247 
248 /* Cross calls must be serialized, at least currently. */
249 static void smp4m_cross_call(smpfunc_t func, cpumask_t mask, unsigned long arg1,
250 			     unsigned long arg2, unsigned long arg3,
251 			     unsigned long arg4)
252 {
253 		register int ncpus = SUN4M_NCPUS;
254 		unsigned long flags;
255 
256 		spin_lock_irqsave(&cross_call_lock, flags);
257 
258 		/* Init function glue. */
259 		ccall_info.func = func;
260 		ccall_info.arg1 = arg1;
261 		ccall_info.arg2 = arg2;
262 		ccall_info.arg3 = arg3;
263 		ccall_info.arg4 = arg4;
264 		ccall_info.arg5 = 0;
265 
266 		/* Init receive/complete mapping, plus fire the IPI's off. */
267 		{
268 			register int i;
269 
270 			cpu_clear(smp_processor_id(), mask);
271 			cpus_and(mask, cpu_online_map, mask);
272 			for(i = 0; i < ncpus; i++) {
273 				if (cpu_isset(i, mask)) {
274 					ccall_info.processors_in[i] = 0;
275 					ccall_info.processors_out[i] = 0;
276 					set_cpu_int(i, IRQ_CROSS_CALL);
277 				} else {
278 					ccall_info.processors_in[i] = 1;
279 					ccall_info.processors_out[i] = 1;
280 				}
281 			}
282 		}
283 
284 		{
285 			register int i;
286 
287 			i = 0;
288 			do {
289 				if (!cpu_isset(i, mask))
290 					continue;
291 				while(!ccall_info.processors_in[i])
292 					barrier();
293 			} while(++i < ncpus);
294 
295 			i = 0;
296 			do {
297 				if (!cpu_isset(i, mask))
298 					continue;
299 				while(!ccall_info.processors_out[i])
300 					barrier();
301 			} while(++i < ncpus);
302 		}
303 
304 		spin_unlock_irqrestore(&cross_call_lock, flags);
305 }
306 
307 /* Running cross calls. */
308 void smp4m_cross_call_irq(void)
309 {
310 	int i = smp_processor_id();
311 
312 	ccall_info.processors_in[i] = 1;
313 	ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
314 			ccall_info.arg4, ccall_info.arg5);
315 	ccall_info.processors_out[i] = 1;
316 }
317 
318 void smp4m_percpu_timer_interrupt(struct pt_regs *regs)
319 {
320 	struct pt_regs *old_regs;
321 	int cpu = smp_processor_id();
322 
323 	old_regs = set_irq_regs(regs);
324 
325 	clear_profile_irq(cpu);
326 
327 	profile_tick(CPU_PROFILING);
328 
329 	if(!--prof_counter(cpu)) {
330 		int user = user_mode(regs);
331 
332 		irq_enter();
333 		update_process_times(user);
334 		irq_exit();
335 
336 		prof_counter(cpu) = prof_multiplier(cpu);
337 	}
338 	set_irq_regs(old_regs);
339 }
340 
341 extern unsigned int lvl14_resolution;
342 
343 static void __init smp_setup_percpu_timer(void)
344 {
345 	int cpu = smp_processor_id();
346 
347 	prof_counter(cpu) = prof_multiplier(cpu) = 1;
348 	load_profile_irq(cpu, lvl14_resolution);
349 
350 	if(cpu == boot_cpu_id)
351 		enable_pil_irq(14);
352 }
353 
354 static void __init smp4m_blackbox_id(unsigned *addr)
355 {
356 	int rd = *addr & 0x3e000000;
357 	int rs1 = rd >> 11;
358 
359 	addr[0] = 0x81580000 | rd;		/* rd %tbr, reg */
360 	addr[1] = 0x8130200c | rd | rs1;    	/* srl reg, 0xc, reg */
361 	addr[2] = 0x80082003 | rd | rs1;	/* and reg, 3, reg */
362 }
363 
364 static void __init smp4m_blackbox_current(unsigned *addr)
365 {
366 	int rd = *addr & 0x3e000000;
367 	int rs1 = rd >> 11;
368 
369 	addr[0] = 0x81580000 | rd;		/* rd %tbr, reg */
370 	addr[2] = 0x8130200a | rd | rs1;    	/* srl reg, 0xa, reg */
371 	addr[4] = 0x8008200c | rd | rs1;	/* and reg, 0xc, reg */
372 }
373 
374 void __init sun4m_init_smp(void)
375 {
376 	BTFIXUPSET_BLACKBOX(hard_smp_processor_id, smp4m_blackbox_id);
377 	BTFIXUPSET_BLACKBOX(load_current, smp4m_blackbox_current);
378 	BTFIXUPSET_CALL(smp_cross_call, smp4m_cross_call, BTFIXUPCALL_NORM);
379 	BTFIXUPSET_CALL(__hard_smp_processor_id, __smp4m_processor_id, BTFIXUPCALL_NORM);
380 }
381