xref: /openbmc/linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  * SGI UV APIC functions (note: not an Intel compatible APIC)
7  *
8  * Copyright (C) 2007-2010 Silicon Graphics, Inc. All rights reserved.
9  */
10 #include <linux/cpumask.h>
11 #include <linux/hardirq.h>
12 #include <linux/proc_fs.h>
13 #include <linux/threads.h>
14 #include <linux/kernel.h>
15 #include <linux/module.h>
16 #include <linux/string.h>
17 #include <linux/ctype.h>
18 #include <linux/sched.h>
19 #include <linux/timer.h>
20 #include <linux/slab.h>
21 #include <linux/cpu.h>
22 #include <linux/init.h>
23 #include <linux/io.h>
24 #include <linux/pci.h>
25 #include <linux/kdebug.h>
26 
27 #include <asm/uv/uv_mmrs.h>
28 #include <asm/uv/uv_hub.h>
29 #include <asm/current.h>
30 #include <asm/pgtable.h>
31 #include <asm/uv/bios.h>
32 #include <asm/uv/uv.h>
33 #include <asm/apic.h>
34 #include <asm/ipi.h>
35 #include <asm/smp.h>
36 #include <asm/x86_init.h>
37 
38 DEFINE_PER_CPU(int, x2apic_extra_bits);
39 
40 #define PR_DEVEL(fmt, args...)	pr_devel("%s: " fmt, __func__, args)
41 
42 static enum uv_system_type uv_system_type;
43 static u64 gru_start_paddr, gru_end_paddr;
44 static union uvh_apicid uvh_apicid;
45 int uv_min_hub_revision_id;
46 EXPORT_SYMBOL_GPL(uv_min_hub_revision_id);
47 unsigned int uv_apicid_hibits;
48 EXPORT_SYMBOL_GPL(uv_apicid_hibits);
49 static DEFINE_SPINLOCK(uv_nmi_lock);
50 
51 static unsigned long __init uv_early_read_mmr(unsigned long addr)
52 {
53 	unsigned long val, *mmr;
54 
55 	mmr = early_ioremap(UV_LOCAL_MMR_BASE | addr, sizeof(*mmr));
56 	val = *mmr;
57 	early_iounmap(mmr, sizeof(*mmr));
58 	return val;
59 }
60 
61 static inline bool is_GRU_range(u64 start, u64 end)
62 {
63 	return start >= gru_start_paddr && end <= gru_end_paddr;
64 }
65 
66 static bool uv_is_untracked_pat_range(u64 start, u64 end)
67 {
68 	return is_ISA_range(start, end) || is_GRU_range(start, end);
69 }
70 
71 static int __init early_get_pnodeid(void)
72 {
73 	union uvh_node_id_u node_id;
74 	union uvh_rh_gam_config_mmr_u  m_n_config;
75 	int pnode;
76 
77 	/* Currently, all blades have same revision number */
78 	node_id.v = uv_early_read_mmr(UVH_NODE_ID);
79 	m_n_config.v = uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR);
80 	uv_min_hub_revision_id = node_id.s.revision;
81 
82 	pnode = (node_id.s.node_id >> 1) & ((1 << m_n_config.s.n_skt) - 1);
83 	return pnode;
84 }
85 
86 static void __init early_get_apic_pnode_shift(void)
87 {
88 	uvh_apicid.v = uv_early_read_mmr(UVH_APICID);
89 	if (!uvh_apicid.v)
90 		/*
91 		 * Old bios, use default value
92 		 */
93 		uvh_apicid.s.pnode_shift = UV_APIC_PNODE_SHIFT;
94 }
95 
96 /*
97  * Add an extra bit as dictated by bios to the destination apicid of
98  * interrupts potentially passing through the UV HUB.  This prevents
99  * a deadlock between interrupts and IO port operations.
100  */
101 static void __init uv_set_apicid_hibit(void)
102 {
103 	union uvh_lb_target_physical_apic_id_mask_u apicid_mask;
104 
105 	apicid_mask.v = uv_early_read_mmr(UVH_LB_TARGET_PHYSICAL_APIC_ID_MASK);
106 	uv_apicid_hibits = apicid_mask.s.bit_enables & UV_APICID_HIBIT_MASK;
107 }
108 
109 static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
110 {
111 	int pnodeid;
112 
113 	if (!strcmp(oem_id, "SGI")) {
114 		pnodeid = early_get_pnodeid();
115 		early_get_apic_pnode_shift();
116 		x86_platform.is_untracked_pat_range =  uv_is_untracked_pat_range;
117 		x86_platform.nmi_init = uv_nmi_init;
118 		if (!strcmp(oem_table_id, "UVL"))
119 			uv_system_type = UV_LEGACY_APIC;
120 		else if (!strcmp(oem_table_id, "UVX"))
121 			uv_system_type = UV_X2APIC;
122 		else if (!strcmp(oem_table_id, "UVH")) {
123 			__this_cpu_write(x2apic_extra_bits,
124 				pnodeid << uvh_apicid.s.pnode_shift);
125 			uv_system_type = UV_NON_UNIQUE_APIC;
126 			uv_set_apicid_hibit();
127 			return 1;
128 		}
129 	}
130 	return 0;
131 }
132 
133 enum uv_system_type get_uv_system_type(void)
134 {
135 	return uv_system_type;
136 }
137 
138 int is_uv_system(void)
139 {
140 	return uv_system_type != UV_NONE;
141 }
142 EXPORT_SYMBOL_GPL(is_uv_system);
143 
144 DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
145 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
146 
147 struct uv_blade_info *uv_blade_info;
148 EXPORT_SYMBOL_GPL(uv_blade_info);
149 
150 short *uv_node_to_blade;
151 EXPORT_SYMBOL_GPL(uv_node_to_blade);
152 
153 short *uv_cpu_to_blade;
154 EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
155 
156 short uv_possible_blades;
157 EXPORT_SYMBOL_GPL(uv_possible_blades);
158 
159 unsigned long sn_rtc_cycles_per_second;
160 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
161 
162 static const struct cpumask *uv_target_cpus(void)
163 {
164 	return cpu_online_mask;
165 }
166 
167 static void uv_vector_allocation_domain(int cpu, struct cpumask *retmask)
168 {
169 	cpumask_clear(retmask);
170 	cpumask_set_cpu(cpu, retmask);
171 }
172 
173 static int __cpuinit uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
174 {
175 #ifdef CONFIG_SMP
176 	unsigned long val;
177 	int pnode;
178 
179 	pnode = uv_apicid_to_pnode(phys_apicid);
180 	phys_apicid |= uv_apicid_hibits;
181 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
182 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
183 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
184 	    APIC_DM_INIT;
185 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
186 	mdelay(10);
187 
188 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
189 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
190 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
191 	    APIC_DM_STARTUP;
192 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
193 
194 	atomic_set(&init_deasserted, 1);
195 #endif
196 	return 0;
197 }
198 
199 static void uv_send_IPI_one(int cpu, int vector)
200 {
201 	unsigned long apicid;
202 	int pnode;
203 
204 	apicid = per_cpu(x86_cpu_to_apicid, cpu);
205 	pnode = uv_apicid_to_pnode(apicid);
206 	uv_hub_send_ipi(pnode, apicid, vector);
207 }
208 
209 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
210 {
211 	unsigned int cpu;
212 
213 	for_each_cpu(cpu, mask)
214 		uv_send_IPI_one(cpu, vector);
215 }
216 
217 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
218 {
219 	unsigned int this_cpu = smp_processor_id();
220 	unsigned int cpu;
221 
222 	for_each_cpu(cpu, mask) {
223 		if (cpu != this_cpu)
224 			uv_send_IPI_one(cpu, vector);
225 	}
226 }
227 
228 static void uv_send_IPI_allbutself(int vector)
229 {
230 	unsigned int this_cpu = smp_processor_id();
231 	unsigned int cpu;
232 
233 	for_each_online_cpu(cpu) {
234 		if (cpu != this_cpu)
235 			uv_send_IPI_one(cpu, vector);
236 	}
237 }
238 
239 static void uv_send_IPI_all(int vector)
240 {
241 	uv_send_IPI_mask(cpu_online_mask, vector);
242 }
243 
244 static int uv_apic_id_registered(void)
245 {
246 	return 1;
247 }
248 
249 static void uv_init_apic_ldr(void)
250 {
251 }
252 
253 static unsigned int uv_cpu_mask_to_apicid(const struct cpumask *cpumask)
254 {
255 	/*
256 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
257 	 * May as well be the first.
258 	 */
259 	int cpu = cpumask_first(cpumask);
260 
261 	if ((unsigned)cpu < nr_cpu_ids)
262 		return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
263 	else
264 		return BAD_APICID;
265 }
266 
267 static unsigned int
268 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
269 			  const struct cpumask *andmask)
270 {
271 	int cpu;
272 
273 	/*
274 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
275 	 * May as well be the first.
276 	 */
277 	for_each_cpu_and(cpu, cpumask, andmask) {
278 		if (cpumask_test_cpu(cpu, cpu_online_mask))
279 			break;
280 	}
281 	return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
282 }
283 
284 static unsigned int x2apic_get_apic_id(unsigned long x)
285 {
286 	unsigned int id;
287 
288 	WARN_ON(preemptible() && num_online_cpus() > 1);
289 	id = x | __this_cpu_read(x2apic_extra_bits);
290 
291 	return id;
292 }
293 
294 static unsigned long set_apic_id(unsigned int id)
295 {
296 	unsigned long x;
297 
298 	/* maskout x2apic_extra_bits ? */
299 	x = id;
300 	return x;
301 }
302 
303 static unsigned int uv_read_apic_id(void)
304 {
305 
306 	return x2apic_get_apic_id(apic_read(APIC_ID));
307 }
308 
309 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
310 {
311 	return uv_read_apic_id() >> index_msb;
312 }
313 
314 static void uv_send_IPI_self(int vector)
315 {
316 	apic_write(APIC_SELF_IPI, vector);
317 }
318 
319 struct apic __refdata apic_x2apic_uv_x = {
320 
321 	.name				= "UV large system",
322 	.probe				= NULL,
323 	.acpi_madt_oem_check		= uv_acpi_madt_oem_check,
324 	.apic_id_registered		= uv_apic_id_registered,
325 
326 	.irq_delivery_mode		= dest_Fixed,
327 	.irq_dest_mode			= 0, /* physical */
328 
329 	.target_cpus			= uv_target_cpus,
330 	.disable_esr			= 0,
331 	.dest_logical			= APIC_DEST_LOGICAL,
332 	.check_apicid_used		= NULL,
333 	.check_apicid_present		= NULL,
334 
335 	.vector_allocation_domain	= uv_vector_allocation_domain,
336 	.init_apic_ldr			= uv_init_apic_ldr,
337 
338 	.ioapic_phys_id_map		= NULL,
339 	.setup_apic_routing		= NULL,
340 	.multi_timer_check		= NULL,
341 	.apicid_to_node			= NULL,
342 	.cpu_to_logical_apicid		= NULL,
343 	.cpu_present_to_apicid		= default_cpu_present_to_apicid,
344 	.apicid_to_cpu_present		= NULL,
345 	.setup_portio_remap		= NULL,
346 	.check_phys_apicid_present	= default_check_phys_apicid_present,
347 	.enable_apic_mode		= NULL,
348 	.phys_pkg_id			= uv_phys_pkg_id,
349 	.mps_oem_check			= NULL,
350 
351 	.get_apic_id			= x2apic_get_apic_id,
352 	.set_apic_id			= set_apic_id,
353 	.apic_id_mask			= 0xFFFFFFFFu,
354 
355 	.cpu_mask_to_apicid		= uv_cpu_mask_to_apicid,
356 	.cpu_mask_to_apicid_and		= uv_cpu_mask_to_apicid_and,
357 
358 	.send_IPI_mask			= uv_send_IPI_mask,
359 	.send_IPI_mask_allbutself	= uv_send_IPI_mask_allbutself,
360 	.send_IPI_allbutself		= uv_send_IPI_allbutself,
361 	.send_IPI_all			= uv_send_IPI_all,
362 	.send_IPI_self			= uv_send_IPI_self,
363 
364 	.wakeup_secondary_cpu		= uv_wakeup_secondary,
365 	.trampoline_phys_low		= DEFAULT_TRAMPOLINE_PHYS_LOW,
366 	.trampoline_phys_high		= DEFAULT_TRAMPOLINE_PHYS_HIGH,
367 	.wait_for_init_deassert		= NULL,
368 	.smp_callin_clear_local_apic	= NULL,
369 	.inquire_remote_apic		= NULL,
370 
371 	.read				= native_apic_msr_read,
372 	.write				= native_apic_msr_write,
373 	.icr_read			= native_x2apic_icr_read,
374 	.icr_write			= native_x2apic_icr_write,
375 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
376 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
377 };
378 
379 static __cpuinit void set_x2apic_extra_bits(int pnode)
380 {
381 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
382 }
383 
384 /*
385  * Called on boot cpu.
386  */
387 static __init int boot_pnode_to_blade(int pnode)
388 {
389 	int blade;
390 
391 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
392 		if (pnode == uv_blade_info[blade].pnode)
393 			return blade;
394 	BUG();
395 }
396 
397 struct redir_addr {
398 	unsigned long redirect;
399 	unsigned long alias;
400 };
401 
402 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
403 
404 static __initdata struct redir_addr redir_addrs[] = {
405 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
406 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
407 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
408 };
409 
410 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
411 {
412 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
413 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
414 	int i;
415 
416 	for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
417 		alias.v = uv_read_local_mmr(redir_addrs[i].alias);
418 		if (alias.s.enable && alias.s.base == 0) {
419 			*size = (1UL << alias.s.m_alias);
420 			redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
421 			*base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
422 			return;
423 		}
424 	}
425 	*base = *size = 0;
426 }
427 
428 enum map_type {map_wb, map_uc};
429 
430 static __init void map_high(char *id, unsigned long base, int pshift,
431 			int bshift, int max_pnode, enum map_type map_type)
432 {
433 	unsigned long bytes, paddr;
434 
435 	paddr = base << pshift;
436 	bytes = (1UL << bshift) * (max_pnode + 1);
437 	printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
438 						paddr + bytes);
439 	if (map_type == map_uc)
440 		init_extra_mapping_uc(paddr, bytes);
441 	else
442 		init_extra_mapping_wb(paddr, bytes);
443 
444 }
445 static __init void map_gru_high(int max_pnode)
446 {
447 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
448 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
449 
450 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
451 	if (gru.s.enable) {
452 		map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
453 		gru_start_paddr = ((u64)gru.s.base << shift);
454 		gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
455 
456 	}
457 }
458 
459 static __init void map_mmr_high(int max_pnode)
460 {
461 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
462 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
463 
464 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
465 	if (mmr.s.enable)
466 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
467 }
468 
469 static __init void map_mmioh_high(int max_pnode)
470 {
471 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
472 	int shift = UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
473 
474 	mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
475 	if (mmioh.s.enable)
476 		map_high("MMIOH", mmioh.s.base, shift, mmioh.s.m_io,
477 			max_pnode, map_uc);
478 }
479 
480 static __init void map_low_mmrs(void)
481 {
482 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
483 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
484 }
485 
486 static __init void uv_rtc_init(void)
487 {
488 	long status;
489 	u64 ticks_per_sec;
490 
491 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
492 					&ticks_per_sec);
493 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
494 		printk(KERN_WARNING
495 			"unable to determine platform RTC clock frequency, "
496 			"guessing.\n");
497 		/* BIOS gives wrong value for clock freq. so guess */
498 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
499 	} else
500 		sn_rtc_cycles_per_second = ticks_per_sec;
501 }
502 
503 /*
504  * percpu heartbeat timer
505  */
506 static void uv_heartbeat(unsigned long ignored)
507 {
508 	struct timer_list *timer = &uv_hub_info->scir.timer;
509 	unsigned char bits = uv_hub_info->scir.state;
510 
511 	/* flip heartbeat bit */
512 	bits ^= SCIR_CPU_HEARTBEAT;
513 
514 	/* is this cpu idle? */
515 	if (idle_cpu(raw_smp_processor_id()))
516 		bits &= ~SCIR_CPU_ACTIVITY;
517 	else
518 		bits |= SCIR_CPU_ACTIVITY;
519 
520 	/* update system controller interface reg */
521 	uv_set_scir_bits(bits);
522 
523 	/* enable next timer period */
524 	mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
525 }
526 
527 static void __cpuinit uv_heartbeat_enable(int cpu)
528 {
529 	while (!uv_cpu_hub_info(cpu)->scir.enabled) {
530 		struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
531 
532 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
533 		setup_timer(timer, uv_heartbeat, cpu);
534 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
535 		add_timer_on(timer, cpu);
536 		uv_cpu_hub_info(cpu)->scir.enabled = 1;
537 
538 		/* also ensure that boot cpu is enabled */
539 		cpu = 0;
540 	}
541 }
542 
543 #ifdef CONFIG_HOTPLUG_CPU
544 static void __cpuinit uv_heartbeat_disable(int cpu)
545 {
546 	if (uv_cpu_hub_info(cpu)->scir.enabled) {
547 		uv_cpu_hub_info(cpu)->scir.enabled = 0;
548 		del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
549 	}
550 	uv_set_cpu_scir_bits(cpu, 0xff);
551 }
552 
553 /*
554  * cpu hotplug notifier
555  */
556 static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
557 				       unsigned long action, void *hcpu)
558 {
559 	long cpu = (long)hcpu;
560 
561 	switch (action) {
562 	case CPU_ONLINE:
563 		uv_heartbeat_enable(cpu);
564 		break;
565 	case CPU_DOWN_PREPARE:
566 		uv_heartbeat_disable(cpu);
567 		break;
568 	default:
569 		break;
570 	}
571 	return NOTIFY_OK;
572 }
573 
574 static __init void uv_scir_register_cpu_notifier(void)
575 {
576 	hotcpu_notifier(uv_scir_cpu_notify, 0);
577 }
578 
579 #else /* !CONFIG_HOTPLUG_CPU */
580 
581 static __init void uv_scir_register_cpu_notifier(void)
582 {
583 }
584 
585 static __init int uv_init_heartbeat(void)
586 {
587 	int cpu;
588 
589 	if (is_uv_system())
590 		for_each_online_cpu(cpu)
591 			uv_heartbeat_enable(cpu);
592 	return 0;
593 }
594 
595 late_initcall(uv_init_heartbeat);
596 
597 #endif /* !CONFIG_HOTPLUG_CPU */
598 
599 /* Direct Legacy VGA I/O traffic to designated IOH */
600 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
601 		      unsigned int command_bits, bool change_bridge)
602 {
603 	int domain, bus, rc;
604 
605 	PR_DEVEL("devfn %x decode %d cmd %x chg_brdg %d\n",
606 			pdev->devfn, decode, command_bits, change_bridge);
607 
608 	if (!change_bridge)
609 		return 0;
610 
611 	if ((command_bits & PCI_COMMAND_IO) == 0)
612 		return 0;
613 
614 	domain = pci_domain_nr(pdev->bus);
615 	bus = pdev->bus->number;
616 
617 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
618 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
619 
620 	return rc;
621 }
622 
623 /*
624  * Called on each cpu to initialize the per_cpu UV data area.
625  * FIXME: hotplug not supported yet
626  */
627 void __cpuinit uv_cpu_init(void)
628 {
629 	/* CPU 0 initilization will be done via uv_system_init. */
630 	if (!uv_blade_info)
631 		return;
632 
633 	uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
634 
635 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
636 		set_x2apic_extra_bits(uv_hub_info->pnode);
637 }
638 
639 /*
640  * When NMI is received, print a stack trace.
641  */
642 int uv_handle_nmi(struct notifier_block *self, unsigned long reason, void *data)
643 {
644 	if (reason != DIE_NMIUNKNOWN)
645 		return NOTIFY_OK;
646 
647 	if (in_crash_kexec)
648 		/* do nothing if entering the crash kernel */
649 		return NOTIFY_OK;
650 	/*
651 	 * Use a lock so only one cpu prints at a time
652 	 * to prevent intermixed output.
653 	 */
654 	spin_lock(&uv_nmi_lock);
655 	pr_info("NMI stack dump cpu %u:\n", smp_processor_id());
656 	dump_stack();
657 	spin_unlock(&uv_nmi_lock);
658 
659 	return NOTIFY_STOP;
660 }
661 
662 static struct notifier_block uv_dump_stack_nmi_nb = {
663 	.notifier_call	= uv_handle_nmi
664 };
665 
666 void uv_register_nmi_notifier(void)
667 {
668 	if (register_die_notifier(&uv_dump_stack_nmi_nb))
669 		printk(KERN_WARNING "UV NMI handler failed to register\n");
670 }
671 
672 void uv_nmi_init(void)
673 {
674 	unsigned int value;
675 
676 	/*
677 	 * Unmask NMI on all cpus
678 	 */
679 	value = apic_read(APIC_LVT1) | APIC_DM_NMI;
680 	value &= ~APIC_LVT_MASKED;
681 	apic_write(APIC_LVT1, value);
682 }
683 
684 void __init uv_system_init(void)
685 {
686 	union uvh_rh_gam_config_mmr_u  m_n_config;
687 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
688 	union uvh_node_id_u node_id;
689 	unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
690 	int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val, n_io;
691 	int gnode_extra, max_pnode = 0;
692 	unsigned long mmr_base, present, paddr;
693 	unsigned short pnode_mask, pnode_io_mask;
694 
695 	map_low_mmrs();
696 
697 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
698 	m_val = m_n_config.s.m_skt;
699 	n_val = m_n_config.s.n_skt;
700 	mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
701 	n_io = mmioh.s.n_io;
702 	mmr_base =
703 	    uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
704 	    ~UV_MMR_ENABLE;
705 	pnode_mask = (1 << n_val) - 1;
706 	pnode_io_mask = (1 << n_io) - 1;
707 
708 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
709 	gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
710 	gnode_upper = ((unsigned long)gnode_extra  << m_val);
711 	printk(KERN_INFO "UV: N %d, M %d, N_IO: %d, gnode_upper 0x%lx, gnode_extra 0x%x, pnode_mask 0x%x, pnode_io_mask 0x%x\n",
712 			n_val, m_val, n_io, gnode_upper, gnode_extra, pnode_mask, pnode_io_mask);
713 
714 	printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
715 
716 	for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
717 		uv_possible_blades +=
718 		  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
719 	printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
720 
721 	bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
722 	uv_blade_info = kmalloc(bytes, GFP_KERNEL);
723 	BUG_ON(!uv_blade_info);
724 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
725 		uv_blade_info[blade].memory_nid = -1;
726 
727 	get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
728 
729 	bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
730 	uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
731 	BUG_ON(!uv_node_to_blade);
732 	memset(uv_node_to_blade, 255, bytes);
733 
734 	bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
735 	uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
736 	BUG_ON(!uv_cpu_to_blade);
737 	memset(uv_cpu_to_blade, 255, bytes);
738 
739 	blade = 0;
740 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
741 		present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
742 		for (j = 0; j < 64; j++) {
743 			if (!test_bit(j, &present))
744 				continue;
745 			pnode = (i * 64 + j) & pnode_mask;
746 			uv_blade_info[blade].pnode = pnode;
747 			uv_blade_info[blade].nr_possible_cpus = 0;
748 			uv_blade_info[blade].nr_online_cpus = 0;
749 			max_pnode = max(pnode, max_pnode);
750 			blade++;
751 		}
752 	}
753 
754 	uv_bios_init();
755 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
756 			    &sn_region_size, &system_serial_number);
757 	uv_rtc_init();
758 
759 	for_each_present_cpu(cpu) {
760 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
761 
762 		nid = cpu_to_node(cpu);
763 		/*
764 		 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
765 		 */
766 		uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
767 		uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
768 		pnode = uv_apicid_to_pnode(apicid);
769 		blade = boot_pnode_to_blade(pnode);
770 		lcpu = uv_blade_info[blade].nr_possible_cpus;
771 		uv_blade_info[blade].nr_possible_cpus++;
772 
773 		/* Any node on the blade, else will contain -1. */
774 		uv_blade_info[blade].memory_nid = nid;
775 
776 		uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
777 		uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
778 		uv_cpu_hub_info(cpu)->m_val = m_val;
779 		uv_cpu_hub_info(cpu)->n_val = n_val;
780 		uv_cpu_hub_info(cpu)->numa_blade_id = blade;
781 		uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
782 		uv_cpu_hub_info(cpu)->pnode = pnode;
783 		uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
784 		uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
785 		uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
786 		uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
787 		uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
788 		uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
789 		uv_node_to_blade[nid] = blade;
790 		uv_cpu_to_blade[cpu] = blade;
791 	}
792 
793 	/* Add blade/pnode info for nodes without cpus */
794 	for_each_online_node(nid) {
795 		if (uv_node_to_blade[nid] >= 0)
796 			continue;
797 		paddr = node_start_pfn(nid) << PAGE_SHIFT;
798 		paddr = uv_soc_phys_ram_to_gpa(paddr);
799 		pnode = (paddr >> m_val) & pnode_mask;
800 		blade = boot_pnode_to_blade(pnode);
801 		uv_node_to_blade[nid] = blade;
802 	}
803 
804 	map_gru_high(max_pnode);
805 	map_mmr_high(max_pnode);
806 	map_mmioh_high(max_pnode & pnode_io_mask);
807 
808 	uv_cpu_init();
809 	uv_scir_register_cpu_notifier();
810 	uv_register_nmi_notifier();
811 	proc_mkdir("sgi_uv", NULL);
812 
813 	/* register Legacy VGA I/O redirection handler */
814 	pci_register_set_vga_state(uv_set_vga_state);
815 }
816