xref: /illumos-gate/usr/src/uts/sun4u/os/mach_startup.c (revision c73a93f2f62aa040c7fc1755230b787b0ad03967)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 #include <sys/machsystm.h>
29 #include <sys/archsystm.h>
30 #include <sys/vm.h>
31 #include <sys/cpu.h>
32 #include <sys/cpupart.h>
33 #include <sys/atomic.h>
34 #include <sys/reboot.h>
35 #include <sys/kdi.h>
36 #include <sys/bootconf.h>
37 #include <sys/memlist_plat.h>
38 #include <sys/memlist_impl.h>
39 #include <sys/prom_plat.h>
40 #include <sys/prom_isa.h>
41 #include <sys/autoconf.h>
42 #include <sys/intreg.h>
43 #include <sys/ivintr.h>
44 #include <sys/fpu/fpusystm.h>
45 #include <sys/iommutsb.h>
46 #include <vm/vm_dep.h>
47 #include <vm/seg_kmem.h>
48 #include <vm/seg_kpm.h>
49 #include <vm/seg_map.h>
50 #include <vm/seg_kp.h>
51 #include <sys/sysconf.h>
52 #include <vm/hat_sfmmu.h>
53 #include <sys/kobj.h>
54 #include <sys/sun4asi.h>
55 #include <sys/clconf.h>
56 #include <sys/platform_module.h>
57 #include <sys/panic.h>
58 #include <sys/cpu_sgnblk_defs.h>
59 #include <sys/clock.h>
60 #include <sys/fpras_impl.h>
61 #include <sys/prom_debug.h>
62 #include <sys/traptrace.h>
63 #include <sys/memnode.h>
64 #include <sys/mem_cage.h>
65 
66 /*
67  * fpRAS implementation structures.
68  */
69 struct fpras_chkfn *fpras_chkfnaddrs[FPRAS_NCOPYOPS];
70 struct fpras_chkfngrp *fpras_chkfngrps;
71 struct fpras_chkfngrp *fpras_chkfngrps_base;
72 int fpras_frequency = -1;
73 int64_t fpras_interval = -1;
74 
75 /*
76  * Halt idling cpus optimization
77  *
78  * This optimation is only enabled in platforms that have
79  * the CPU halt support. The cpu_halt_cpu() support is provided
80  * in the cpu module and it is referenced here with a pragma weak.
81  * The presence of this routine automatically enable the halt idling
82  * cpus functionality if the global switch enable_halt_idle_cpus
83  * is set (default is set).
84  *
85  */
86 #pragma weak	cpu_halt_cpu
87 extern void	cpu_halt_cpu();
88 
89 int		enable_halt_idle_cpus = 1; /* global switch */
90 
91 void
92 setup_trap_table(void)
93 {
94 	intr_init(CPU);			/* init interrupt request free list */
95 	setwstate(WSTATE_KERN);
96 	prom_set_traptable(&trap_table);
97 }
98 
99 void
100 mach_fpras()
101 {
102 	if (fpras_implemented && !fpras_disable) {
103 		int i;
104 		struct fpras_chkfngrp *fcgp;
105 		size_t chkfngrpsallocsz;
106 
107 		/*
108 		 * Note that we size off of NCPU and setup for
109 		 * all those possibilities regardless of whether
110 		 * the cpu id is present or not.  We do this so that
111 		 * we don't have any construction or destruction
112 		 * activity to perform at DR time, and it's not
113 		 * costly in memory.  We require block alignment.
114 		 */
115 		chkfngrpsallocsz = NCPU * sizeof (struct fpras_chkfngrp);
116 		fpras_chkfngrps_base = kmem_alloc(chkfngrpsallocsz, KM_SLEEP);
117 		if (IS_P2ALIGNED((uintptr_t)fpras_chkfngrps_base, 64)) {
118 			fpras_chkfngrps = fpras_chkfngrps_base;
119 		} else {
120 			kmem_free(fpras_chkfngrps_base, chkfngrpsallocsz);
121 			chkfngrpsallocsz += 64;
122 			fpras_chkfngrps_base = kmem_alloc(chkfngrpsallocsz,
123 			    KM_SLEEP);
124 			fpras_chkfngrps = (struct fpras_chkfngrp *)
125 			    P2ROUNDUP((uintptr_t)fpras_chkfngrps_base, 64);
126 		}
127 
128 		/*
129 		 * Copy our check function into place for each copy operation
130 		 * and each cpu id.
131 		 */
132 		fcgp = &fpras_chkfngrps[0];
133 		for (i = 0; i < FPRAS_NCOPYOPS; ++i)
134 			bcopy((void *)fpras_chkfn_type1, &fcgp->fpras_fn[i],
135 			    sizeof (struct fpras_chkfn));
136 		for (i = 1; i < NCPU; ++i)
137 			*(&fpras_chkfngrps[i]) = *fcgp;
138 
139 		/*
140 		 * At definition fpras_frequency is set to -1, and it will
141 		 * still have that value unless changed in /etc/system (not
142 		 * strictly supported, but not preventable).  The following
143 		 * both sets the default and sanity checks anything from
144 		 * /etc/system.
145 		 */
146 		if (fpras_frequency < 0)
147 			fpras_frequency = FPRAS_DEFAULT_FREQUENCY;
148 
149 		/*
150 		 * Now calculate fpras_interval.  When fpras_interval
151 		 * becomes non-negative fpras checks will commence
152 		 * (copies before this point in boot will bypass fpras).
153 		 * Our stores of instructions must be visible; no need
154 		 * to flush as they're never been executed before.
155 		 */
156 		membar_producer();
157 		fpras_interval = (fpras_frequency == 0) ?
158 		    0 : sys_tick_freq / fpras_frequency;
159 	}
160 }
161 
162 void
163 mach_hw_copy_limit(void)
164 {
165 	if (!fpu_exists) {
166 		use_hw_bcopy = 0;
167 		hw_copy_limit_1 = 0;
168 		hw_copy_limit_2 = 0;
169 		hw_copy_limit_4 = 0;
170 		hw_copy_limit_8 = 0;
171 		use_hw_bzero = 0;
172 	}
173 }
174 
175 void
176 load_tod_module()
177 {
178 	/*
179 	 * Load tod driver module for the tod part found on this system.
180 	 * Recompute the cpu frequency/delays based on tod as tod part
181 	 * tends to keep time more accurately.
182 	 */
183 	if (tod_module_name == NULL || modload("tod", tod_module_name) == -1)
184 		halt("Can't load tod module");
185 }
186 
187 void
188 mach_memscrub(void)
189 {
190 	/*
191 	 * Startup memory scrubber, if not running fpu emulation code.
192 	 */
193 
194 #ifndef _HW_MEMSCRUB_SUPPORT
195 	if (fpu_exists) {
196 		if (memscrub_init()) {
197 			cmn_err(CE_WARN,
198 			    "Memory scrubber failed to initialize");
199 		}
200 	}
201 #endif /* _HW_MEMSCRUB_SUPPORT */
202 }
203 
204 /*
205  * Halt the calling CPU until awoken via an interrupt
206  * This routine should only be invoked if cpu_halt_cpu()
207  * exists and is supported, see mach_cpu_halt_idle()
208  */
209 static void
210 cpu_halt(void)
211 {
212 	cpu_t		*cpup = CPU;
213 	processorid_t	cpun = cpup->cpu_id;
214 	cpupart_t	*cp = cpup->cpu_part;
215 	int		hset_update = 1;
216 	uint_t		pstate;
217 	extern uint_t	getpstate(void);
218 	extern void	setpstate(uint_t);
219 
220 	/*
221 	 * If this CPU is online, and there's multiple CPUs
222 	 * in the system, then we should notate our halting
223 	 * by adding ourselves to the partition's halted CPU
224 	 * bitmap. This allows other CPUs to find/awaken us when
225 	 * work becomes available.
226 	 */
227 	if (CPU->cpu_flags & CPU_OFFLINE || ncpus == 1)
228 		hset_update = 0;
229 
230 	/*
231 	 * Add ourselves to the partition's halted CPUs bitmask
232 	 * and set our HALTED flag, if necessary.
233 	 *
234 	 * When a thread becomes runnable, it is placed on the queue
235 	 * and then the halted cpuset is checked to determine who
236 	 * (if anyone) should be awoken. We therefore need to first
237 	 * add ourselves to the halted cpuset, and then check if there
238 	 * is any work available.
239 	 */
240 	if (hset_update) {
241 		cpup->cpu_disp_flags |= CPU_DISP_HALTED;
242 		membar_producer();
243 		CPUSET_ATOMIC_ADD(cp->cp_haltset, cpun);
244 	}
245 
246 	/*
247 	 * Check to make sure there's really nothing to do.
248 	 * Work destined for this CPU may become available after
249 	 * this check. We'll be notified through the clearing of our
250 	 * bit in the halted CPU bitmask, and a poke.
251 	 */
252 	if (disp_anywork()) {
253 		if (hset_update) {
254 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
255 			CPUSET_ATOMIC_DEL(cp->cp_haltset, cpun);
256 		}
257 		return;
258 	}
259 
260 	/*
261 	 * We're on our way to being halted.
262 	 *
263 	 * Disable interrupts now, so that we'll awaken immediately
264 	 * after halting if someone tries to poke us between now and
265 	 * the time we actually halt.
266 	 *
267 	 * We check for the presence of our bit after disabling interrupts.
268 	 * If it's cleared, we'll return. If the bit is cleared after
269 	 * we check then the poke will pop us out of the halted state.
270 	 *
271 	 * The ordering of the poke and the clearing of the bit by cpu_wakeup
272 	 * is important.
273 	 * cpu_wakeup() must clear, then poke.
274 	 * cpu_halt() must disable interrupts, then check for the bit.
275 	 */
276 	pstate = getpstate();
277 	setpstate(pstate & ~PSTATE_IE);
278 
279 	if (hset_update && !CPU_IN_SET(cp->cp_haltset, cpun)) {
280 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
281 		setpstate(pstate);
282 		return;
283 	}
284 
285 	/*
286 	 * The check for anything locally runnable is here for performance
287 	 * and isn't needed for correctness. disp_nrunnable ought to be
288 	 * in our cache still, so it's inexpensive to check, and if there
289 	 * is anything runnable we won't have to wait for the poke.
290 	 */
291 	if (cpup->cpu_disp->disp_nrunnable != 0) {
292 		if (hset_update) {
293 			cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
294 			CPUSET_ATOMIC_DEL(cp->cp_haltset, cpun);
295 		}
296 		setpstate(pstate);
297 		return;
298 	}
299 
300 	/*
301 	 * Halt the strand.
302 	 */
303 	if (&cpu_halt_cpu)
304 		cpu_halt_cpu();
305 
306 	/*
307 	 * We're no longer halted
308 	 */
309 	setpstate(pstate);
310 	if (hset_update) {
311 		cpup->cpu_disp_flags &= ~CPU_DISP_HALTED;
312 		CPUSET_ATOMIC_DEL(cp->cp_haltset, cpun);
313 	}
314 }
315 
316 /*
317  * If "cpu" is halted, then wake it up clearing its halted bit in advance.
318  * Otherwise, see if other CPUs in the cpu partition are halted and need to
319  * be woken up so that they can steal the thread we placed on this CPU.
320  * This function is only used on MP systems.
321  * This function should only be invoked if cpu_halt_cpu()
322  * exists and is supported, see mach_cpu_halt_idle()
323  */
324 static void
325 cpu_wakeup(cpu_t *cpu, int bound)
326 {
327 	uint_t		cpu_found;
328 	int		result;
329 	cpupart_t	*cp;
330 
331 	cp = cpu->cpu_part;
332 	if (CPU_IN_SET(cp->cp_haltset, cpu->cpu_id)) {
333 		/*
334 		 * Clear the halted bit for that CPU since it will be
335 		 * poked in a moment.
336 		 */
337 		CPUSET_ATOMIC_DEL(cp->cp_haltset, cpu->cpu_id);
338 		/*
339 		 * We may find the current CPU present in the halted cpuset
340 		 * if we're in the context of an interrupt that occurred
341 		 * before we had a chance to clear our bit in cpu_halt().
342 		 * Poking ourself is obviously unnecessary, since if
343 		 * we're here, we're not halted.
344 		 */
345 		if (cpu != CPU)
346 			poke_cpu(cpu->cpu_id);
347 		return;
348 	} else {
349 		/*
350 		 * This cpu isn't halted, but it's idle or undergoing a
351 		 * context switch. No need to awaken anyone else.
352 		 */
353 		if (cpu->cpu_thread == cpu->cpu_idle_thread ||
354 		    cpu->cpu_disp_flags & CPU_DISP_DONTSTEAL)
355 			return;
356 	}
357 
358 	/*
359 	 * No need to wake up other CPUs if the thread we just enqueued
360 	 * is bound.
361 	 */
362 	if (bound)
363 		return;
364 
365 	/*
366 	 * See if there's any other halted CPUs. If there are, then
367 	 * select one, and awaken it.
368 	 * It's possible that after we find a CPU, somebody else
369 	 * will awaken it before we get the chance.
370 	 * In that case, look again.
371 	 */
372 	do {
373 		CPUSET_FIND(cp->cp_haltset, cpu_found);
374 		if (cpu_found == CPUSET_NOTINSET)
375 			return;
376 
377 		ASSERT(cpu_found >= 0 && cpu_found < NCPU);
378 		CPUSET_ATOMIC_XDEL(cp->cp_haltset, cpu_found, result);
379 	} while (result < 0);
380 
381 	if (cpu_found != CPU->cpu_id)
382 		poke_cpu(cpu_found);
383 }
384 
385 void
386 mach_cpu_halt_idle()
387 {
388 	if (enable_halt_idle_cpus) {
389 		if (&cpu_halt_cpu) {
390 			idle_cpu = cpu_halt;
391 			disp_enq_thread = cpu_wakeup;
392 		}
393 	}
394 }
395 
396 /*ARGSUSED*/
397 void
398 cpu_intrq_setup(struct cpu *cp)
399 {
400 	/* Interrupt mondo queues not applicable to sun4u */
401 }
402 
403 /*ARGSUSED*/
404 void
405 cpu_intrq_cleanup(struct cpu *cp)
406 {
407 	/* Interrupt mondo queues not applicable to sun4u */
408 }
409 
410 /*ARGSUSED*/
411 void
412 cpu_intrq_register(struct cpu *cp)
413 {
414 	/* Interrupt/error queues not applicable to sun4u */
415 }
416 
417 /*ARGSUSED*/
418 void
419 mach_htraptrace_setup(int cpuid)
420 {
421 	/* Setup hypervisor traptrace buffer, not applicable to sun4u */
422 }
423 
424 /*ARGSUSED*/
425 void
426 mach_htraptrace_configure(int cpuid)
427 {
428 	/* enable/ disable hypervisor traptracing, not applicable to sun4u */
429 }
430 
431 /*ARGSUSED*/
432 void
433 mach_htraptrace_cleanup(int cpuid)
434 {
435 	/* cleanup hypervisor traptrace buffer, not applicable to sun4u */
436 }
437 
438 void
439 mach_descrip_startup_init(void)
440 {
441 	/*
442 	 * Only for sun4v.
443 	 * Initialize Machine description framework during startup.
444 	 */
445 }
446 void
447 mach_descrip_startup_fini(void)
448 {
449 	/*
450 	 * Only for sun4v.
451 	 * Clean up Machine Description framework during startup.
452 	 */
453 }
454 
455 void
456 mach_descrip_init(void)
457 {
458 	/*
459 	 * Only for sun4v.
460 	 * Initialize Machine description framework.
461 	 */
462 }
463 
464 void
465 hsvc_setup(void)
466 {
467 	/* Setup hypervisor services, not applicable to sun4u */
468 }
469 
470 void
471 load_mach_drivers(void)
472 {
473 	/* Currently no machine class (sun4u) specific drivers to load */
474 }
475 
476 /*
477  * Return true if the machine we're running on is a Positron.
478  * (Positron is an unsupported developers platform.)
479  */
480 int
481 iam_positron(void)
482 {
483 	char model[32];
484 	const char proto_model[] = "SUNW,501-2732";
485 	pnode_t root = prom_rootnode();
486 
487 	if (prom_getproplen(root, "model") != sizeof (proto_model))
488 		return (0);
489 
490 	(void) prom_getprop(root, "model", model);
491 	if (strcmp(model, proto_model) == 0)
492 		return (1);
493 	return (0);
494 }
495 
496 /*
497  * Find a physically contiguous area of twice the largest ecache size
498  * to be used while doing displacement flush of ecaches.
499  */
500 uint64_t
501 ecache_flush_address(void)
502 {
503 	struct memlist *pmem;
504 	uint64_t flush_size;
505 	uint64_t ret_val;
506 
507 	flush_size = ecache_size * 2;
508 	for (pmem = phys_install; pmem; pmem = pmem->next) {
509 		ret_val = P2ROUNDUP(pmem->address, ecache_size);
510 		if (ret_val + flush_size <= pmem->address + pmem->size)
511 			return (ret_val);
512 	}
513 	return ((uint64_t)-1);
514 }
515 
516 /*
517  * Called with the memlist lock held to say that phys_install has
518  * changed.
519  */
520 void
521 phys_install_has_changed(void)
522 {
523 	/*
524 	 * Get the new address into a temporary just in case panicking
525 	 * involves use of ecache_flushaddr.
526 	 */
527 	uint64_t new_addr;
528 
529 	new_addr = ecache_flush_address();
530 	if (new_addr == (uint64_t)-1) {
531 		cmn_err(CE_PANIC,
532 		    "ecache_flush_address(): failed, ecache_size=%x",
533 		    ecache_size);
534 		/*NOTREACHED*/
535 	}
536 	ecache_flushaddr = new_addr;
537 	membar_producer();
538 }
539