xref: /linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision f2ee442115c9b6219083c019939a9cc0c9abb2f8)
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 #include <linux/delay.h>
27 #include <linux/crash_dump.h>
28 
29 #include <asm/uv/uv_mmrs.h>
30 #include <asm/uv/uv_hub.h>
31 #include <asm/current.h>
32 #include <asm/pgtable.h>
33 #include <asm/uv/bios.h>
34 #include <asm/uv/uv.h>
35 #include <asm/apic.h>
36 #include <asm/ipi.h>
37 #include <asm/smp.h>
38 #include <asm/x86_init.h>
39 #include <asm/emergency-restart.h>
40 #include <asm/nmi.h>
41 
42 /* BMC sets a bit this MMR non-zero before sending an NMI */
43 #define UVH_NMI_MMR				UVH_SCRATCH5
44 #define UVH_NMI_MMR_CLEAR			(UVH_NMI_MMR + 8)
45 #define UV_NMI_PENDING_MASK			(1UL << 63)
46 DEFINE_PER_CPU(unsigned long, cpu_last_nmi_count);
47 
48 DEFINE_PER_CPU(int, x2apic_extra_bits);
49 
50 #define PR_DEVEL(fmt, args...)	pr_devel("%s: " fmt, __func__, args)
51 
52 static enum uv_system_type uv_system_type;
53 static u64 gru_start_paddr, gru_end_paddr;
54 static union uvh_apicid uvh_apicid;
55 int uv_min_hub_revision_id;
56 EXPORT_SYMBOL_GPL(uv_min_hub_revision_id);
57 unsigned int uv_apicid_hibits;
58 EXPORT_SYMBOL_GPL(uv_apicid_hibits);
59 static DEFINE_SPINLOCK(uv_nmi_lock);
60 
61 static struct apic apic_x2apic_uv_x;
62 
63 static unsigned long __init uv_early_read_mmr(unsigned long addr)
64 {
65 	unsigned long val, *mmr;
66 
67 	mmr = early_ioremap(UV_LOCAL_MMR_BASE | addr, sizeof(*mmr));
68 	val = *mmr;
69 	early_iounmap(mmr, sizeof(*mmr));
70 	return val;
71 }
72 
73 static inline bool is_GRU_range(u64 start, u64 end)
74 {
75 	return start >= gru_start_paddr && end <= gru_end_paddr;
76 }
77 
78 static bool uv_is_untracked_pat_range(u64 start, u64 end)
79 {
80 	return is_ISA_range(start, end) || is_GRU_range(start, end);
81 }
82 
83 static int __init early_get_pnodeid(void)
84 {
85 	union uvh_node_id_u node_id;
86 	union uvh_rh_gam_config_mmr_u  m_n_config;
87 	int pnode;
88 
89 	/* Currently, all blades have same revision number */
90 	node_id.v = uv_early_read_mmr(UVH_NODE_ID);
91 	m_n_config.v = uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR);
92 	uv_min_hub_revision_id = node_id.s.revision;
93 
94 	if (node_id.s.part_number == UV2_HUB_PART_NUMBER)
95 		uv_min_hub_revision_id += UV2_HUB_REVISION_BASE - 1;
96 
97 	uv_hub_info->hub_revision = uv_min_hub_revision_id;
98 	pnode = (node_id.s.node_id >> 1) & ((1 << m_n_config.s.n_skt) - 1);
99 	return pnode;
100 }
101 
102 static void __init early_get_apic_pnode_shift(void)
103 {
104 	uvh_apicid.v = uv_early_read_mmr(UVH_APICID);
105 	if (!uvh_apicid.v)
106 		/*
107 		 * Old bios, use default value
108 		 */
109 		uvh_apicid.s.pnode_shift = UV_APIC_PNODE_SHIFT;
110 }
111 
112 /*
113  * Add an extra bit as dictated by bios to the destination apicid of
114  * interrupts potentially passing through the UV HUB.  This prevents
115  * a deadlock between interrupts and IO port operations.
116  */
117 static void __init uv_set_apicid_hibit(void)
118 {
119 	union uv1h_lb_target_physical_apic_id_mask_u apicid_mask;
120 
121 	if (is_uv1_hub()) {
122 		apicid_mask.v =
123 			uv_early_read_mmr(UV1H_LB_TARGET_PHYSICAL_APIC_ID_MASK);
124 		uv_apicid_hibits =
125 			apicid_mask.s1.bit_enables & UV_APICID_HIBIT_MASK;
126 	}
127 }
128 
129 static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
130 {
131 	int pnodeid, is_uv1, is_uv2;
132 
133 	is_uv1 = !strcmp(oem_id, "SGI");
134 	is_uv2 = !strcmp(oem_id, "SGI2");
135 	if (is_uv1 || is_uv2) {
136 		uv_hub_info->hub_revision =
137 			is_uv1 ? UV1_HUB_REVISION_BASE : UV2_HUB_REVISION_BASE;
138 		pnodeid = early_get_pnodeid();
139 		early_get_apic_pnode_shift();
140 		x86_platform.is_untracked_pat_range =  uv_is_untracked_pat_range;
141 		x86_platform.nmi_init = uv_nmi_init;
142 		if (!strcmp(oem_table_id, "UVL"))
143 			uv_system_type = UV_LEGACY_APIC;
144 		else if (!strcmp(oem_table_id, "UVX"))
145 			uv_system_type = UV_X2APIC;
146 		else if (!strcmp(oem_table_id, "UVH")) {
147 			__this_cpu_write(x2apic_extra_bits,
148 				pnodeid << uvh_apicid.s.pnode_shift);
149 			uv_system_type = UV_NON_UNIQUE_APIC;
150 			uv_set_apicid_hibit();
151 			return 1;
152 		}
153 	}
154 	return 0;
155 }
156 
157 enum uv_system_type get_uv_system_type(void)
158 {
159 	return uv_system_type;
160 }
161 
162 int is_uv_system(void)
163 {
164 	return uv_system_type != UV_NONE;
165 }
166 EXPORT_SYMBOL_GPL(is_uv_system);
167 
168 DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
169 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
170 
171 struct uv_blade_info *uv_blade_info;
172 EXPORT_SYMBOL_GPL(uv_blade_info);
173 
174 short *uv_node_to_blade;
175 EXPORT_SYMBOL_GPL(uv_node_to_blade);
176 
177 short *uv_cpu_to_blade;
178 EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
179 
180 short uv_possible_blades;
181 EXPORT_SYMBOL_GPL(uv_possible_blades);
182 
183 unsigned long sn_rtc_cycles_per_second;
184 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
185 
186 static const struct cpumask *uv_target_cpus(void)
187 {
188 	return cpu_online_mask;
189 }
190 
191 static void uv_vector_allocation_domain(int cpu, struct cpumask *retmask)
192 {
193 	cpumask_clear(retmask);
194 	cpumask_set_cpu(cpu, retmask);
195 }
196 
197 static int __cpuinit uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
198 {
199 #ifdef CONFIG_SMP
200 	unsigned long val;
201 	int pnode;
202 
203 	pnode = uv_apicid_to_pnode(phys_apicid);
204 	phys_apicid |= uv_apicid_hibits;
205 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
206 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
207 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
208 	    APIC_DM_INIT;
209 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
210 
211 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
212 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
213 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
214 	    APIC_DM_STARTUP;
215 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
216 
217 	atomic_set(&init_deasserted, 1);
218 #endif
219 	return 0;
220 }
221 
222 static void uv_send_IPI_one(int cpu, int vector)
223 {
224 	unsigned long apicid;
225 	int pnode;
226 
227 	apicid = per_cpu(x86_cpu_to_apicid, cpu);
228 	pnode = uv_apicid_to_pnode(apicid);
229 	uv_hub_send_ipi(pnode, apicid, vector);
230 }
231 
232 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
233 {
234 	unsigned int cpu;
235 
236 	for_each_cpu(cpu, mask)
237 		uv_send_IPI_one(cpu, vector);
238 }
239 
240 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
241 {
242 	unsigned int this_cpu = smp_processor_id();
243 	unsigned int cpu;
244 
245 	for_each_cpu(cpu, mask) {
246 		if (cpu != this_cpu)
247 			uv_send_IPI_one(cpu, vector);
248 	}
249 }
250 
251 static void uv_send_IPI_allbutself(int vector)
252 {
253 	unsigned int this_cpu = smp_processor_id();
254 	unsigned int cpu;
255 
256 	for_each_online_cpu(cpu) {
257 		if (cpu != this_cpu)
258 			uv_send_IPI_one(cpu, vector);
259 	}
260 }
261 
262 static void uv_send_IPI_all(int vector)
263 {
264 	uv_send_IPI_mask(cpu_online_mask, vector);
265 }
266 
267 static int uv_apic_id_registered(void)
268 {
269 	return 1;
270 }
271 
272 static void uv_init_apic_ldr(void)
273 {
274 }
275 
276 static unsigned int uv_cpu_mask_to_apicid(const struct cpumask *cpumask)
277 {
278 	/*
279 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
280 	 * May as well be the first.
281 	 */
282 	int cpu = cpumask_first(cpumask);
283 
284 	if ((unsigned)cpu < nr_cpu_ids)
285 		return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
286 	else
287 		return BAD_APICID;
288 }
289 
290 static unsigned int
291 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
292 			  const struct cpumask *andmask)
293 {
294 	int cpu;
295 
296 	/*
297 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
298 	 * May as well be the first.
299 	 */
300 	for_each_cpu_and(cpu, cpumask, andmask) {
301 		if (cpumask_test_cpu(cpu, cpu_online_mask))
302 			break;
303 	}
304 	return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
305 }
306 
307 static unsigned int x2apic_get_apic_id(unsigned long x)
308 {
309 	unsigned int id;
310 
311 	WARN_ON(preemptible() && num_online_cpus() > 1);
312 	id = x | __this_cpu_read(x2apic_extra_bits);
313 
314 	return id;
315 }
316 
317 static unsigned long set_apic_id(unsigned int id)
318 {
319 	unsigned long x;
320 
321 	/* maskout x2apic_extra_bits ? */
322 	x = id;
323 	return x;
324 }
325 
326 static unsigned int uv_read_apic_id(void)
327 {
328 
329 	return x2apic_get_apic_id(apic_read(APIC_ID));
330 }
331 
332 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
333 {
334 	return uv_read_apic_id() >> index_msb;
335 }
336 
337 static void uv_send_IPI_self(int vector)
338 {
339 	apic_write(APIC_SELF_IPI, vector);
340 }
341 
342 static int uv_probe(void)
343 {
344 	return apic == &apic_x2apic_uv_x;
345 }
346 
347 static struct apic __refdata apic_x2apic_uv_x = {
348 
349 	.name				= "UV large system",
350 	.probe				= uv_probe,
351 	.acpi_madt_oem_check		= uv_acpi_madt_oem_check,
352 	.apic_id_registered		= uv_apic_id_registered,
353 
354 	.irq_delivery_mode		= dest_Fixed,
355 	.irq_dest_mode			= 0, /* physical */
356 
357 	.target_cpus			= uv_target_cpus,
358 	.disable_esr			= 0,
359 	.dest_logical			= APIC_DEST_LOGICAL,
360 	.check_apicid_used		= NULL,
361 	.check_apicid_present		= NULL,
362 
363 	.vector_allocation_domain	= uv_vector_allocation_domain,
364 	.init_apic_ldr			= uv_init_apic_ldr,
365 
366 	.ioapic_phys_id_map		= NULL,
367 	.setup_apic_routing		= NULL,
368 	.multi_timer_check		= NULL,
369 	.cpu_present_to_apicid		= default_cpu_present_to_apicid,
370 	.apicid_to_cpu_present		= NULL,
371 	.setup_portio_remap		= NULL,
372 	.check_phys_apicid_present	= default_check_phys_apicid_present,
373 	.enable_apic_mode		= NULL,
374 	.phys_pkg_id			= uv_phys_pkg_id,
375 	.mps_oem_check			= NULL,
376 
377 	.get_apic_id			= x2apic_get_apic_id,
378 	.set_apic_id			= set_apic_id,
379 	.apic_id_mask			= 0xFFFFFFFFu,
380 
381 	.cpu_mask_to_apicid		= uv_cpu_mask_to_apicid,
382 	.cpu_mask_to_apicid_and		= uv_cpu_mask_to_apicid_and,
383 
384 	.send_IPI_mask			= uv_send_IPI_mask,
385 	.send_IPI_mask_allbutself	= uv_send_IPI_mask_allbutself,
386 	.send_IPI_allbutself		= uv_send_IPI_allbutself,
387 	.send_IPI_all			= uv_send_IPI_all,
388 	.send_IPI_self			= uv_send_IPI_self,
389 
390 	.wakeup_secondary_cpu		= uv_wakeup_secondary,
391 	.trampoline_phys_low		= DEFAULT_TRAMPOLINE_PHYS_LOW,
392 	.trampoline_phys_high		= DEFAULT_TRAMPOLINE_PHYS_HIGH,
393 	.wait_for_init_deassert		= NULL,
394 	.smp_callin_clear_local_apic	= NULL,
395 	.inquire_remote_apic		= NULL,
396 
397 	.read				= native_apic_msr_read,
398 	.write				= native_apic_msr_write,
399 	.icr_read			= native_x2apic_icr_read,
400 	.icr_write			= native_x2apic_icr_write,
401 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
402 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
403 };
404 
405 static __cpuinit void set_x2apic_extra_bits(int pnode)
406 {
407 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
408 }
409 
410 /*
411  * Called on boot cpu.
412  */
413 static __init int boot_pnode_to_blade(int pnode)
414 {
415 	int blade;
416 
417 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
418 		if (pnode == uv_blade_info[blade].pnode)
419 			return blade;
420 	BUG();
421 }
422 
423 struct redir_addr {
424 	unsigned long redirect;
425 	unsigned long alias;
426 };
427 
428 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
429 
430 static __initdata struct redir_addr redir_addrs[] = {
431 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
432 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
433 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
434 };
435 
436 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
437 {
438 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
439 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
440 	int i;
441 
442 	for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
443 		alias.v = uv_read_local_mmr(redir_addrs[i].alias);
444 		if (alias.s.enable && alias.s.base == 0) {
445 			*size = (1UL << alias.s.m_alias);
446 			redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
447 			*base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
448 			return;
449 		}
450 	}
451 	*base = *size = 0;
452 }
453 
454 enum map_type {map_wb, map_uc};
455 
456 static __init void map_high(char *id, unsigned long base, int pshift,
457 			int bshift, int max_pnode, enum map_type map_type)
458 {
459 	unsigned long bytes, paddr;
460 
461 	paddr = base << pshift;
462 	bytes = (1UL << bshift) * (max_pnode + 1);
463 	printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
464 						paddr + bytes);
465 	if (map_type == map_uc)
466 		init_extra_mapping_uc(paddr, bytes);
467 	else
468 		init_extra_mapping_wb(paddr, bytes);
469 
470 }
471 static __init void map_gru_high(int max_pnode)
472 {
473 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
474 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
475 
476 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
477 	if (gru.s.enable) {
478 		map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
479 		gru_start_paddr = ((u64)gru.s.base << shift);
480 		gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
481 
482 	}
483 }
484 
485 static __init void map_mmr_high(int max_pnode)
486 {
487 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
488 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
489 
490 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
491 	if (mmr.s.enable)
492 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
493 }
494 
495 static __init void map_mmioh_high(int max_pnode)
496 {
497 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
498 	int shift;
499 
500 	mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
501 	if (is_uv1_hub() && mmioh.s1.enable) {
502 		shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
503 		map_high("MMIOH", mmioh.s1.base, shift, mmioh.s1.m_io,
504 			max_pnode, map_uc);
505 	}
506 	if (is_uv2_hub() && mmioh.s2.enable) {
507 		shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
508 		map_high("MMIOH", mmioh.s2.base, shift, mmioh.s2.m_io,
509 			max_pnode, map_uc);
510 	}
511 }
512 
513 static __init void map_low_mmrs(void)
514 {
515 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
516 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
517 }
518 
519 static __init void uv_rtc_init(void)
520 {
521 	long status;
522 	u64 ticks_per_sec;
523 
524 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
525 					&ticks_per_sec);
526 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
527 		printk(KERN_WARNING
528 			"unable to determine platform RTC clock frequency, "
529 			"guessing.\n");
530 		/* BIOS gives wrong value for clock freq. so guess */
531 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
532 	} else
533 		sn_rtc_cycles_per_second = ticks_per_sec;
534 }
535 
536 /*
537  * percpu heartbeat timer
538  */
539 static void uv_heartbeat(unsigned long ignored)
540 {
541 	struct timer_list *timer = &uv_hub_info->scir.timer;
542 	unsigned char bits = uv_hub_info->scir.state;
543 
544 	/* flip heartbeat bit */
545 	bits ^= SCIR_CPU_HEARTBEAT;
546 
547 	/* is this cpu idle? */
548 	if (idle_cpu(raw_smp_processor_id()))
549 		bits &= ~SCIR_CPU_ACTIVITY;
550 	else
551 		bits |= SCIR_CPU_ACTIVITY;
552 
553 	/* update system controller interface reg */
554 	uv_set_scir_bits(bits);
555 
556 	/* enable next timer period */
557 	mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
558 }
559 
560 static void __cpuinit uv_heartbeat_enable(int cpu)
561 {
562 	while (!uv_cpu_hub_info(cpu)->scir.enabled) {
563 		struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
564 
565 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
566 		setup_timer(timer, uv_heartbeat, cpu);
567 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
568 		add_timer_on(timer, cpu);
569 		uv_cpu_hub_info(cpu)->scir.enabled = 1;
570 
571 		/* also ensure that boot cpu is enabled */
572 		cpu = 0;
573 	}
574 }
575 
576 #ifdef CONFIG_HOTPLUG_CPU
577 static void __cpuinit uv_heartbeat_disable(int cpu)
578 {
579 	if (uv_cpu_hub_info(cpu)->scir.enabled) {
580 		uv_cpu_hub_info(cpu)->scir.enabled = 0;
581 		del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
582 	}
583 	uv_set_cpu_scir_bits(cpu, 0xff);
584 }
585 
586 /*
587  * cpu hotplug notifier
588  */
589 static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
590 				       unsigned long action, void *hcpu)
591 {
592 	long cpu = (long)hcpu;
593 
594 	switch (action) {
595 	case CPU_ONLINE:
596 		uv_heartbeat_enable(cpu);
597 		break;
598 	case CPU_DOWN_PREPARE:
599 		uv_heartbeat_disable(cpu);
600 		break;
601 	default:
602 		break;
603 	}
604 	return NOTIFY_OK;
605 }
606 
607 static __init void uv_scir_register_cpu_notifier(void)
608 {
609 	hotcpu_notifier(uv_scir_cpu_notify, 0);
610 }
611 
612 #else /* !CONFIG_HOTPLUG_CPU */
613 
614 static __init void uv_scir_register_cpu_notifier(void)
615 {
616 }
617 
618 static __init int uv_init_heartbeat(void)
619 {
620 	int cpu;
621 
622 	if (is_uv_system())
623 		for_each_online_cpu(cpu)
624 			uv_heartbeat_enable(cpu);
625 	return 0;
626 }
627 
628 late_initcall(uv_init_heartbeat);
629 
630 #endif /* !CONFIG_HOTPLUG_CPU */
631 
632 /* Direct Legacy VGA I/O traffic to designated IOH */
633 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
634 		      unsigned int command_bits, u32 flags)
635 {
636 	int domain, bus, rc;
637 
638 	PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
639 			pdev->devfn, decode, command_bits, flags);
640 
641 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
642 		return 0;
643 
644 	if ((command_bits & PCI_COMMAND_IO) == 0)
645 		return 0;
646 
647 	domain = pci_domain_nr(pdev->bus);
648 	bus = pdev->bus->number;
649 
650 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
651 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
652 
653 	return rc;
654 }
655 
656 /*
657  * Called on each cpu to initialize the per_cpu UV data area.
658  * FIXME: hotplug not supported yet
659  */
660 void __cpuinit uv_cpu_init(void)
661 {
662 	/* CPU 0 initilization will be done via uv_system_init. */
663 	if (!uv_blade_info)
664 		return;
665 
666 	uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
667 
668 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
669 		set_x2apic_extra_bits(uv_hub_info->pnode);
670 }
671 
672 /*
673  * When NMI is received, print a stack trace.
674  */
675 int uv_handle_nmi(unsigned int reason, struct pt_regs *regs)
676 {
677 	unsigned long real_uv_nmi;
678 	int bid;
679 
680 	/*
681 	 * Each blade has an MMR that indicates when an NMI has been sent
682 	 * to cpus on the blade. If an NMI is detected, atomically
683 	 * clear the MMR and update a per-blade NMI count used to
684 	 * cause each cpu on the blade to notice a new NMI.
685 	 */
686 	bid = uv_numa_blade_id();
687 	real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
688 
689 	if (unlikely(real_uv_nmi)) {
690 		spin_lock(&uv_blade_info[bid].nmi_lock);
691 		real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
692 		if (real_uv_nmi) {
693 			uv_blade_info[bid].nmi_count++;
694 			uv_write_local_mmr(UVH_NMI_MMR_CLEAR, UV_NMI_PENDING_MASK);
695 		}
696 		spin_unlock(&uv_blade_info[bid].nmi_lock);
697 	}
698 
699 	if (likely(__get_cpu_var(cpu_last_nmi_count) == uv_blade_info[bid].nmi_count))
700 		return NMI_DONE;
701 
702 	__get_cpu_var(cpu_last_nmi_count) = uv_blade_info[bid].nmi_count;
703 
704 	/*
705 	 * Use a lock so only one cpu prints at a time.
706 	 * This prevents intermixed output.
707 	 */
708 	spin_lock(&uv_nmi_lock);
709 	pr_info("UV NMI stack dump cpu %u:\n", smp_processor_id());
710 	dump_stack();
711 	spin_unlock(&uv_nmi_lock);
712 
713 	return NMI_HANDLED;
714 }
715 
716 void uv_register_nmi_notifier(void)
717 {
718 	if (register_nmi_handler(NMI_UNKNOWN, uv_handle_nmi, 0, "uv"))
719 		printk(KERN_WARNING "UV NMI handler failed to register\n");
720 }
721 
722 void uv_nmi_init(void)
723 {
724 	unsigned int value;
725 
726 	/*
727 	 * Unmask NMI on all cpus
728 	 */
729 	value = apic_read(APIC_LVT1) | APIC_DM_NMI;
730 	value &= ~APIC_LVT_MASKED;
731 	apic_write(APIC_LVT1, value);
732 }
733 
734 void __init uv_system_init(void)
735 {
736 	union uvh_rh_gam_config_mmr_u  m_n_config;
737 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
738 	union uvh_node_id_u node_id;
739 	unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
740 	int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val, n_io;
741 	int gnode_extra, max_pnode = 0;
742 	unsigned long mmr_base, present, paddr;
743 	unsigned short pnode_mask, pnode_io_mask;
744 
745 	printk(KERN_INFO "UV: Found %s hub\n", is_uv1_hub() ? "UV1" : "UV2");
746 	map_low_mmrs();
747 
748 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
749 	m_val = m_n_config.s.m_skt;
750 	n_val = m_n_config.s.n_skt;
751 	mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
752 	n_io = is_uv1_hub() ? mmioh.s1.n_io : mmioh.s2.n_io;
753 	mmr_base =
754 	    uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
755 	    ~UV_MMR_ENABLE;
756 	pnode_mask = (1 << n_val) - 1;
757 	pnode_io_mask = (1 << n_io) - 1;
758 
759 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
760 	gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
761 	gnode_upper = ((unsigned long)gnode_extra  << m_val);
762 	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",
763 			n_val, m_val, n_io, gnode_upper, gnode_extra, pnode_mask, pnode_io_mask);
764 
765 	printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
766 
767 	for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
768 		uv_possible_blades +=
769 		  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
770 	printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
771 
772 	bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
773 	uv_blade_info = kzalloc(bytes, GFP_KERNEL);
774 	BUG_ON(!uv_blade_info);
775 
776 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
777 		uv_blade_info[blade].memory_nid = -1;
778 
779 	get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
780 
781 	bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
782 	uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
783 	BUG_ON(!uv_node_to_blade);
784 	memset(uv_node_to_blade, 255, bytes);
785 
786 	bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
787 	uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
788 	BUG_ON(!uv_cpu_to_blade);
789 	memset(uv_cpu_to_blade, 255, bytes);
790 
791 	blade = 0;
792 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
793 		present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
794 		for (j = 0; j < 64; j++) {
795 			if (!test_bit(j, &present))
796 				continue;
797 			pnode = (i * 64 + j) & pnode_mask;
798 			uv_blade_info[blade].pnode = pnode;
799 			uv_blade_info[blade].nr_possible_cpus = 0;
800 			uv_blade_info[blade].nr_online_cpus = 0;
801 			spin_lock_init(&uv_blade_info[blade].nmi_lock);
802 			max_pnode = max(pnode, max_pnode);
803 			blade++;
804 		}
805 	}
806 
807 	uv_bios_init();
808 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
809 			    &sn_region_size, &system_serial_number);
810 	uv_rtc_init();
811 
812 	for_each_present_cpu(cpu) {
813 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
814 
815 		nid = cpu_to_node(cpu);
816 		/*
817 		 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
818 		 */
819 		uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
820 		uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
821 		uv_cpu_hub_info(cpu)->hub_revision = uv_hub_info->hub_revision;
822 
823 		uv_cpu_hub_info(cpu)->m_shift = 64 - m_val;
824 		uv_cpu_hub_info(cpu)->n_lshift = is_uv2_1_hub() ?
825 				(m_val == 40 ? 40 : 39) : m_val;
826 
827 		pnode = uv_apicid_to_pnode(apicid);
828 		blade = boot_pnode_to_blade(pnode);
829 		lcpu = uv_blade_info[blade].nr_possible_cpus;
830 		uv_blade_info[blade].nr_possible_cpus++;
831 
832 		/* Any node on the blade, else will contain -1. */
833 		uv_blade_info[blade].memory_nid = nid;
834 
835 		uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
836 		uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
837 		uv_cpu_hub_info(cpu)->m_val = m_val;
838 		uv_cpu_hub_info(cpu)->n_val = n_val;
839 		uv_cpu_hub_info(cpu)->numa_blade_id = blade;
840 		uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
841 		uv_cpu_hub_info(cpu)->pnode = pnode;
842 		uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
843 		uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
844 		uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
845 		uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
846 		uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
847 		uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
848 		uv_node_to_blade[nid] = blade;
849 		uv_cpu_to_blade[cpu] = blade;
850 	}
851 
852 	/* Add blade/pnode info for nodes without cpus */
853 	for_each_online_node(nid) {
854 		if (uv_node_to_blade[nid] >= 0)
855 			continue;
856 		paddr = node_start_pfn(nid) << PAGE_SHIFT;
857 		pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
858 		blade = boot_pnode_to_blade(pnode);
859 		uv_node_to_blade[nid] = blade;
860 	}
861 
862 	map_gru_high(max_pnode);
863 	map_mmr_high(max_pnode);
864 	map_mmioh_high(max_pnode & pnode_io_mask);
865 
866 	uv_cpu_init();
867 	uv_scir_register_cpu_notifier();
868 	uv_register_nmi_notifier();
869 	proc_mkdir("sgi_uv", NULL);
870 
871 	/* register Legacy VGA I/O redirection handler */
872 	pci_register_set_vga_state(uv_set_vga_state);
873 
874 	/*
875 	 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
876 	 * EFI is not enabled in the kdump kernel.
877 	 */
878 	if (is_kdump_kernel())
879 		reboot_type = BOOT_ACPI;
880 }
881 
882 apic_driver(apic_x2apic_uv_x);
883