xref: /openbmc/linux/arch/sparc/kernel/sun4m_smp.c (revision 545e4006)
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 	/* Get our local ticker going. */
75 	smp_setup_percpu_timer();
76 
77 	calibrate_delay();
78 	smp_store_cpu_info(cpuid);
79 
80 	local_flush_cache_all();
81 	local_flush_tlb_all();
82 
83 	/*
84 	 * Unblock the master CPU _only_ when the scheduler state
85 	 * of all secondary CPUs will be up-to-date, so after
86 	 * the SMP initialization the master will be just allowed
87 	 * to call the scheduler code.
88 	 */
89 	/* Allow master to continue. */
90 	swap(&cpu_callin_map[cpuid], 1);
91 
92 	/* XXX: What's up with all the flushes? */
93 	local_flush_cache_all();
94 	local_flush_tlb_all();
95 
96 	cpu_probe();
97 
98 	/* Fix idle thread fields. */
99 	__asm__ __volatile__("ld [%0], %%g6\n\t"
100 			     : : "r" (&current_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 
112 	cpu_set(cpuid, cpu_online_map);
113 }
114 
115 /*
116  *	Cycle through the processors asking the PROM to start each one.
117  */
118 
119 extern struct linux_prom_registers smp_penguin_ctable;
120 extern unsigned long trapbase_cpu1[];
121 extern unsigned long trapbase_cpu2[];
122 extern unsigned long trapbase_cpu3[];
123 
124 void __init smp4m_boot_cpus(void)
125 {
126 	smp_setup_percpu_timer();
127 	local_flush_cache_all();
128 }
129 
130 int __cpuinit smp4m_boot_one_cpu(int i)
131 {
132 	extern unsigned long sun4m_cpu_startup;
133 	unsigned long *entry = &sun4m_cpu_startup;
134 	struct task_struct *p;
135 	int timeout;
136 	int cpu_node;
137 
138 	cpu_find_by_mid(i, &cpu_node);
139 
140 	/* Cook up an idler for this guy. */
141 	p = fork_idle(i);
142 	current_set[i] = task_thread_info(p);
143 	/* See trampoline.S for details... */
144 	entry += ((i-1) * 3);
145 
146 	/*
147 	 * Initialize the contexts table
148 	 * Since the call to prom_startcpu() trashes the structure,
149 	 * we need to re-initialize it for each cpu
150 	 */
151 	smp_penguin_ctable.which_io = 0;
152 	smp_penguin_ctable.phys_addr = (unsigned int) srmmu_ctx_table_phys;
153 	smp_penguin_ctable.reg_size = 0;
154 
155 	/* whirrr, whirrr, whirrrrrrrrr... */
156 	printk("Starting CPU %d at %p\n", i, entry);
157 	local_flush_cache_all();
158 	prom_startcpu(cpu_node,
159 		      &smp_penguin_ctable, 0, (char *)entry);
160 
161 	/* wheee... it's going... */
162 	for(timeout = 0; timeout < 10000; timeout++) {
163 		if(cpu_callin_map[i])
164 			break;
165 		udelay(200);
166 	}
167 
168 	if (!(cpu_callin_map[i])) {
169 		printk("Processor %d is stuck.\n", i);
170 		return -ENODEV;
171 	}
172 
173 	local_flush_cache_all();
174 	return 0;
175 }
176 
177 void __init smp4m_smp_done(void)
178 {
179 	int i, first;
180 	int *prev;
181 
182 	/* setup cpu list for irq rotation */
183 	first = 0;
184 	prev = &first;
185 	for (i = 0; i < NR_CPUS; i++) {
186 		if (cpu_online(i)) {
187 			*prev = i;
188 			prev = &cpu_data(i).next;
189 		}
190 	}
191 	*prev = first;
192 	local_flush_cache_all();
193 
194 	/* Free unneeded trap tables */
195 	if (!cpu_isset(1, cpu_present_map)) {
196 		ClearPageReserved(virt_to_page(trapbase_cpu1));
197 		init_page_count(virt_to_page(trapbase_cpu1));
198 		free_page((unsigned long)trapbase_cpu1);
199 		totalram_pages++;
200 		num_physpages++;
201 	}
202 	if (!cpu_isset(2, cpu_present_map)) {
203 		ClearPageReserved(virt_to_page(trapbase_cpu2));
204 		init_page_count(virt_to_page(trapbase_cpu2));
205 		free_page((unsigned long)trapbase_cpu2);
206 		totalram_pages++;
207 		num_physpages++;
208 	}
209 	if (!cpu_isset(3, cpu_present_map)) {
210 		ClearPageReserved(virt_to_page(trapbase_cpu3));
211 		init_page_count(virt_to_page(trapbase_cpu3));
212 		free_page((unsigned long)trapbase_cpu3);
213 		totalram_pages++;
214 		num_physpages++;
215 	}
216 
217 	/* Ok, they are spinning and ready to go. */
218 }
219 
220 /* At each hardware IRQ, we get this called to forward IRQ reception
221  * to the next processor.  The caller must disable the IRQ level being
222  * serviced globally so that there are no double interrupts received.
223  *
224  * XXX See sparc64 irq.c.
225  */
226 void smp4m_irq_rotate(int cpu)
227 {
228 	int next = cpu_data(cpu).next;
229 	if (next != cpu)
230 		set_irq_udt(next);
231 }
232 
233 static struct smp_funcall {
234 	smpfunc_t func;
235 	unsigned long arg1;
236 	unsigned long arg2;
237 	unsigned long arg3;
238 	unsigned long arg4;
239 	unsigned long arg5;
240 	unsigned long processors_in[SUN4M_NCPUS];  /* Set when ipi entered. */
241 	unsigned long processors_out[SUN4M_NCPUS]; /* Set when ipi exited. */
242 } ccall_info;
243 
244 static DEFINE_SPINLOCK(cross_call_lock);
245 
246 /* Cross calls must be serialized, at least currently. */
247 static void smp4m_cross_call(smpfunc_t func, unsigned long arg1,
248 			     unsigned long arg2, unsigned long arg3,
249 			     unsigned long arg4, unsigned long arg5)
250 {
251 		register int ncpus = SUN4M_NCPUS;
252 		unsigned long flags;
253 
254 		spin_lock_irqsave(&cross_call_lock, flags);
255 
256 		/* Init function glue. */
257 		ccall_info.func = func;
258 		ccall_info.arg1 = arg1;
259 		ccall_info.arg2 = arg2;
260 		ccall_info.arg3 = arg3;
261 		ccall_info.arg4 = arg4;
262 		ccall_info.arg5 = arg5;
263 
264 		/* Init receive/complete mapping, plus fire the IPI's off. */
265 		{
266 			cpumask_t mask = cpu_online_map;
267 			register int i;
268 
269 			cpu_clear(smp_processor_id(), mask);
270 			for(i = 0; i < ncpus; i++) {
271 				if (cpu_isset(i, mask)) {
272 					ccall_info.processors_in[i] = 0;
273 					ccall_info.processors_out[i] = 0;
274 					set_cpu_int(i, IRQ_CROSS_CALL);
275 				} else {
276 					ccall_info.processors_in[i] = 1;
277 					ccall_info.processors_out[i] = 1;
278 				}
279 			}
280 		}
281 
282 		{
283 			register int i;
284 
285 			i = 0;
286 			do {
287 				while(!ccall_info.processors_in[i])
288 					barrier();
289 			} while(++i < ncpus);
290 
291 			i = 0;
292 			do {
293 				while(!ccall_info.processors_out[i])
294 					barrier();
295 			} while(++i < ncpus);
296 		}
297 
298 		spin_unlock_irqrestore(&cross_call_lock, flags);
299 }
300 
301 /* Running cross calls. */
302 void smp4m_cross_call_irq(void)
303 {
304 	int i = smp_processor_id();
305 
306 	ccall_info.processors_in[i] = 1;
307 	ccall_info.func(ccall_info.arg1, ccall_info.arg2, ccall_info.arg3,
308 			ccall_info.arg4, ccall_info.arg5);
309 	ccall_info.processors_out[i] = 1;
310 }
311 
312 void smp4m_percpu_timer_interrupt(struct pt_regs *regs)
313 {
314 	struct pt_regs *old_regs;
315 	int cpu = smp_processor_id();
316 
317 	old_regs = set_irq_regs(regs);
318 
319 	clear_profile_irq(cpu);
320 
321 	profile_tick(CPU_PROFILING);
322 
323 	if(!--prof_counter(cpu)) {
324 		int user = user_mode(regs);
325 
326 		irq_enter();
327 		update_process_times(user);
328 		irq_exit();
329 
330 		prof_counter(cpu) = prof_multiplier(cpu);
331 	}
332 	set_irq_regs(old_regs);
333 }
334 
335 extern unsigned int lvl14_resolution;
336 
337 static void __init smp_setup_percpu_timer(void)
338 {
339 	int cpu = smp_processor_id();
340 
341 	prof_counter(cpu) = prof_multiplier(cpu) = 1;
342 	load_profile_irq(cpu, lvl14_resolution);
343 
344 	if(cpu == boot_cpu_id)
345 		enable_pil_irq(14);
346 }
347 
348 static void __init smp4m_blackbox_id(unsigned *addr)
349 {
350 	int rd = *addr & 0x3e000000;
351 	int rs1 = rd >> 11;
352 
353 	addr[0] = 0x81580000 | rd;		/* rd %tbr, reg */
354 	addr[1] = 0x8130200c | rd | rs1;    	/* srl reg, 0xc, reg */
355 	addr[2] = 0x80082003 | rd | rs1;	/* and reg, 3, reg */
356 }
357 
358 static void __init smp4m_blackbox_current(unsigned *addr)
359 {
360 	int rd = *addr & 0x3e000000;
361 	int rs1 = rd >> 11;
362 
363 	addr[0] = 0x81580000 | rd;		/* rd %tbr, reg */
364 	addr[2] = 0x8130200a | rd | rs1;    	/* srl reg, 0xa, reg */
365 	addr[4] = 0x8008200c | rd | rs1;	/* and reg, 0xc, reg */
366 }
367 
368 void __init sun4m_init_smp(void)
369 {
370 	BTFIXUPSET_BLACKBOX(hard_smp_processor_id, smp4m_blackbox_id);
371 	BTFIXUPSET_BLACKBOX(load_current, smp4m_blackbox_current);
372 	BTFIXUPSET_CALL(smp_cross_call, smp4m_cross_call, BTFIXUPCALL_NORM);
373 	BTFIXUPSET_CALL(__hard_smp_processor_id, __smp4m_processor_id, BTFIXUPCALL_NORM);
374 }
375