xref: /openbmc/linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision 81d67439)
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 	mdelay(10);
211 
212 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
213 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
214 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
215 	    APIC_DM_STARTUP;
216 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
217 
218 	atomic_set(&init_deasserted, 1);
219 #endif
220 	return 0;
221 }
222 
223 static void uv_send_IPI_one(int cpu, int vector)
224 {
225 	unsigned long apicid;
226 	int pnode;
227 
228 	apicid = per_cpu(x86_cpu_to_apicid, cpu);
229 	pnode = uv_apicid_to_pnode(apicid);
230 	uv_hub_send_ipi(pnode, apicid, vector);
231 }
232 
233 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
234 {
235 	unsigned int cpu;
236 
237 	for_each_cpu(cpu, mask)
238 		uv_send_IPI_one(cpu, vector);
239 }
240 
241 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
242 {
243 	unsigned int this_cpu = smp_processor_id();
244 	unsigned int cpu;
245 
246 	for_each_cpu(cpu, mask) {
247 		if (cpu != this_cpu)
248 			uv_send_IPI_one(cpu, vector);
249 	}
250 }
251 
252 static void uv_send_IPI_allbutself(int vector)
253 {
254 	unsigned int this_cpu = smp_processor_id();
255 	unsigned int cpu;
256 
257 	for_each_online_cpu(cpu) {
258 		if (cpu != this_cpu)
259 			uv_send_IPI_one(cpu, vector);
260 	}
261 }
262 
263 static void uv_send_IPI_all(int vector)
264 {
265 	uv_send_IPI_mask(cpu_online_mask, vector);
266 }
267 
268 static int uv_apic_id_registered(void)
269 {
270 	return 1;
271 }
272 
273 static void uv_init_apic_ldr(void)
274 {
275 }
276 
277 static unsigned int uv_cpu_mask_to_apicid(const struct cpumask *cpumask)
278 {
279 	/*
280 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
281 	 * May as well be the first.
282 	 */
283 	int cpu = cpumask_first(cpumask);
284 
285 	if ((unsigned)cpu < nr_cpu_ids)
286 		return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
287 	else
288 		return BAD_APICID;
289 }
290 
291 static unsigned int
292 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
293 			  const struct cpumask *andmask)
294 {
295 	int cpu;
296 
297 	/*
298 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
299 	 * May as well be the first.
300 	 */
301 	for_each_cpu_and(cpu, cpumask, andmask) {
302 		if (cpumask_test_cpu(cpu, cpu_online_mask))
303 			break;
304 	}
305 	return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
306 }
307 
308 static unsigned int x2apic_get_apic_id(unsigned long x)
309 {
310 	unsigned int id;
311 
312 	WARN_ON(preemptible() && num_online_cpus() > 1);
313 	id = x | __this_cpu_read(x2apic_extra_bits);
314 
315 	return id;
316 }
317 
318 static unsigned long set_apic_id(unsigned int id)
319 {
320 	unsigned long x;
321 
322 	/* maskout x2apic_extra_bits ? */
323 	x = id;
324 	return x;
325 }
326 
327 static unsigned int uv_read_apic_id(void)
328 {
329 
330 	return x2apic_get_apic_id(apic_read(APIC_ID));
331 }
332 
333 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
334 {
335 	return uv_read_apic_id() >> index_msb;
336 }
337 
338 static void uv_send_IPI_self(int vector)
339 {
340 	apic_write(APIC_SELF_IPI, vector);
341 }
342 
343 static int uv_probe(void)
344 {
345 	return apic == &apic_x2apic_uv_x;
346 }
347 
348 static struct apic __refdata apic_x2apic_uv_x = {
349 
350 	.name				= "UV large system",
351 	.probe				= uv_probe,
352 	.acpi_madt_oem_check		= uv_acpi_madt_oem_check,
353 	.apic_id_registered		= uv_apic_id_registered,
354 
355 	.irq_delivery_mode		= dest_Fixed,
356 	.irq_dest_mode			= 0, /* physical */
357 
358 	.target_cpus			= uv_target_cpus,
359 	.disable_esr			= 0,
360 	.dest_logical			= APIC_DEST_LOGICAL,
361 	.check_apicid_used		= NULL,
362 	.check_apicid_present		= NULL,
363 
364 	.vector_allocation_domain	= uv_vector_allocation_domain,
365 	.init_apic_ldr			= uv_init_apic_ldr,
366 
367 	.ioapic_phys_id_map		= NULL,
368 	.setup_apic_routing		= NULL,
369 	.multi_timer_check		= NULL,
370 	.cpu_present_to_apicid		= default_cpu_present_to_apicid,
371 	.apicid_to_cpu_present		= NULL,
372 	.setup_portio_remap		= NULL,
373 	.check_phys_apicid_present	= default_check_phys_apicid_present,
374 	.enable_apic_mode		= NULL,
375 	.phys_pkg_id			= uv_phys_pkg_id,
376 	.mps_oem_check			= NULL,
377 
378 	.get_apic_id			= x2apic_get_apic_id,
379 	.set_apic_id			= set_apic_id,
380 	.apic_id_mask			= 0xFFFFFFFFu,
381 
382 	.cpu_mask_to_apicid		= uv_cpu_mask_to_apicid,
383 	.cpu_mask_to_apicid_and		= uv_cpu_mask_to_apicid_and,
384 
385 	.send_IPI_mask			= uv_send_IPI_mask,
386 	.send_IPI_mask_allbutself	= uv_send_IPI_mask_allbutself,
387 	.send_IPI_allbutself		= uv_send_IPI_allbutself,
388 	.send_IPI_all			= uv_send_IPI_all,
389 	.send_IPI_self			= uv_send_IPI_self,
390 
391 	.wakeup_secondary_cpu		= uv_wakeup_secondary,
392 	.trampoline_phys_low		= DEFAULT_TRAMPOLINE_PHYS_LOW,
393 	.trampoline_phys_high		= DEFAULT_TRAMPOLINE_PHYS_HIGH,
394 	.wait_for_init_deassert		= NULL,
395 	.smp_callin_clear_local_apic	= NULL,
396 	.inquire_remote_apic		= NULL,
397 
398 	.read				= native_apic_msr_read,
399 	.write				= native_apic_msr_write,
400 	.icr_read			= native_x2apic_icr_read,
401 	.icr_write			= native_x2apic_icr_write,
402 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
403 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
404 };
405 
406 static __cpuinit void set_x2apic_extra_bits(int pnode)
407 {
408 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
409 }
410 
411 /*
412  * Called on boot cpu.
413  */
414 static __init int boot_pnode_to_blade(int pnode)
415 {
416 	int blade;
417 
418 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
419 		if (pnode == uv_blade_info[blade].pnode)
420 			return blade;
421 	BUG();
422 }
423 
424 struct redir_addr {
425 	unsigned long redirect;
426 	unsigned long alias;
427 };
428 
429 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
430 
431 static __initdata struct redir_addr redir_addrs[] = {
432 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
433 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
434 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
435 };
436 
437 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
438 {
439 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
440 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
441 	int i;
442 
443 	for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
444 		alias.v = uv_read_local_mmr(redir_addrs[i].alias);
445 		if (alias.s.enable && alias.s.base == 0) {
446 			*size = (1UL << alias.s.m_alias);
447 			redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
448 			*base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
449 			return;
450 		}
451 	}
452 	*base = *size = 0;
453 }
454 
455 enum map_type {map_wb, map_uc};
456 
457 static __init void map_high(char *id, unsigned long base, int pshift,
458 			int bshift, int max_pnode, enum map_type map_type)
459 {
460 	unsigned long bytes, paddr;
461 
462 	paddr = base << pshift;
463 	bytes = (1UL << bshift) * (max_pnode + 1);
464 	printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
465 						paddr + bytes);
466 	if (map_type == map_uc)
467 		init_extra_mapping_uc(paddr, bytes);
468 	else
469 		init_extra_mapping_wb(paddr, bytes);
470 
471 }
472 static __init void map_gru_high(int max_pnode)
473 {
474 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
475 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
476 
477 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
478 	if (gru.s.enable) {
479 		map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
480 		gru_start_paddr = ((u64)gru.s.base << shift);
481 		gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
482 
483 	}
484 }
485 
486 static __init void map_mmr_high(int max_pnode)
487 {
488 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
489 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
490 
491 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
492 	if (mmr.s.enable)
493 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
494 }
495 
496 static __init void map_mmioh_high(int max_pnode)
497 {
498 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
499 	int shift;
500 
501 	mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
502 	if (is_uv1_hub() && mmioh.s1.enable) {
503 		shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
504 		map_high("MMIOH", mmioh.s1.base, shift, mmioh.s1.m_io,
505 			max_pnode, map_uc);
506 	}
507 	if (is_uv2_hub() && mmioh.s2.enable) {
508 		shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
509 		map_high("MMIOH", mmioh.s2.base, shift, mmioh.s2.m_io,
510 			max_pnode, map_uc);
511 	}
512 }
513 
514 static __init void map_low_mmrs(void)
515 {
516 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
517 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
518 }
519 
520 static __init void uv_rtc_init(void)
521 {
522 	long status;
523 	u64 ticks_per_sec;
524 
525 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
526 					&ticks_per_sec);
527 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
528 		printk(KERN_WARNING
529 			"unable to determine platform RTC clock frequency, "
530 			"guessing.\n");
531 		/* BIOS gives wrong value for clock freq. so guess */
532 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
533 	} else
534 		sn_rtc_cycles_per_second = ticks_per_sec;
535 }
536 
537 /*
538  * percpu heartbeat timer
539  */
540 static void uv_heartbeat(unsigned long ignored)
541 {
542 	struct timer_list *timer = &uv_hub_info->scir.timer;
543 	unsigned char bits = uv_hub_info->scir.state;
544 
545 	/* flip heartbeat bit */
546 	bits ^= SCIR_CPU_HEARTBEAT;
547 
548 	/* is this cpu idle? */
549 	if (idle_cpu(raw_smp_processor_id()))
550 		bits &= ~SCIR_CPU_ACTIVITY;
551 	else
552 		bits |= SCIR_CPU_ACTIVITY;
553 
554 	/* update system controller interface reg */
555 	uv_set_scir_bits(bits);
556 
557 	/* enable next timer period */
558 	mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
559 }
560 
561 static void __cpuinit uv_heartbeat_enable(int cpu)
562 {
563 	while (!uv_cpu_hub_info(cpu)->scir.enabled) {
564 		struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
565 
566 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
567 		setup_timer(timer, uv_heartbeat, cpu);
568 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
569 		add_timer_on(timer, cpu);
570 		uv_cpu_hub_info(cpu)->scir.enabled = 1;
571 
572 		/* also ensure that boot cpu is enabled */
573 		cpu = 0;
574 	}
575 }
576 
577 #ifdef CONFIG_HOTPLUG_CPU
578 static void __cpuinit uv_heartbeat_disable(int cpu)
579 {
580 	if (uv_cpu_hub_info(cpu)->scir.enabled) {
581 		uv_cpu_hub_info(cpu)->scir.enabled = 0;
582 		del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
583 	}
584 	uv_set_cpu_scir_bits(cpu, 0xff);
585 }
586 
587 /*
588  * cpu hotplug notifier
589  */
590 static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
591 				       unsigned long action, void *hcpu)
592 {
593 	long cpu = (long)hcpu;
594 
595 	switch (action) {
596 	case CPU_ONLINE:
597 		uv_heartbeat_enable(cpu);
598 		break;
599 	case CPU_DOWN_PREPARE:
600 		uv_heartbeat_disable(cpu);
601 		break;
602 	default:
603 		break;
604 	}
605 	return NOTIFY_OK;
606 }
607 
608 static __init void uv_scir_register_cpu_notifier(void)
609 {
610 	hotcpu_notifier(uv_scir_cpu_notify, 0);
611 }
612 
613 #else /* !CONFIG_HOTPLUG_CPU */
614 
615 static __init void uv_scir_register_cpu_notifier(void)
616 {
617 }
618 
619 static __init int uv_init_heartbeat(void)
620 {
621 	int cpu;
622 
623 	if (is_uv_system())
624 		for_each_online_cpu(cpu)
625 			uv_heartbeat_enable(cpu);
626 	return 0;
627 }
628 
629 late_initcall(uv_init_heartbeat);
630 
631 #endif /* !CONFIG_HOTPLUG_CPU */
632 
633 /* Direct Legacy VGA I/O traffic to designated IOH */
634 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
635 		      unsigned int command_bits, u32 flags)
636 {
637 	int domain, bus, rc;
638 
639 	PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
640 			pdev->devfn, decode, command_bits, flags);
641 
642 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
643 		return 0;
644 
645 	if ((command_bits & PCI_COMMAND_IO) == 0)
646 		return 0;
647 
648 	domain = pci_domain_nr(pdev->bus);
649 	bus = pdev->bus->number;
650 
651 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
652 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
653 
654 	return rc;
655 }
656 
657 /*
658  * Called on each cpu to initialize the per_cpu UV data area.
659  * FIXME: hotplug not supported yet
660  */
661 void __cpuinit uv_cpu_init(void)
662 {
663 	/* CPU 0 initilization will be done via uv_system_init. */
664 	if (!uv_blade_info)
665 		return;
666 
667 	uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
668 
669 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
670 		set_x2apic_extra_bits(uv_hub_info->pnode);
671 }
672 
673 /*
674  * When NMI is received, print a stack trace.
675  */
676 int uv_handle_nmi(struct notifier_block *self, unsigned long reason, void *data)
677 {
678 	unsigned long real_uv_nmi;
679 	int bid;
680 
681 	if (reason != DIE_NMIUNKNOWN)
682 		return NOTIFY_OK;
683 
684 	if (in_crash_kexec)
685 		/* do nothing if entering the crash kernel */
686 		return NOTIFY_OK;
687 
688 	/*
689 	 * Each blade has an MMR that indicates when an NMI has been sent
690 	 * to cpus on the blade. If an NMI is detected, atomically
691 	 * clear the MMR and update a per-blade NMI count used to
692 	 * cause each cpu on the blade to notice a new NMI.
693 	 */
694 	bid = uv_numa_blade_id();
695 	real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
696 
697 	if (unlikely(real_uv_nmi)) {
698 		spin_lock(&uv_blade_info[bid].nmi_lock);
699 		real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
700 		if (real_uv_nmi) {
701 			uv_blade_info[bid].nmi_count++;
702 			uv_write_local_mmr(UVH_NMI_MMR_CLEAR, UV_NMI_PENDING_MASK);
703 		}
704 		spin_unlock(&uv_blade_info[bid].nmi_lock);
705 	}
706 
707 	if (likely(__get_cpu_var(cpu_last_nmi_count) == uv_blade_info[bid].nmi_count))
708 		return NOTIFY_DONE;
709 
710 	__get_cpu_var(cpu_last_nmi_count) = uv_blade_info[bid].nmi_count;
711 
712 	/*
713 	 * Use a lock so only one cpu prints at a time.
714 	 * This prevents intermixed output.
715 	 */
716 	spin_lock(&uv_nmi_lock);
717 	pr_info("UV NMI stack dump cpu %u:\n", smp_processor_id());
718 	dump_stack();
719 	spin_unlock(&uv_nmi_lock);
720 
721 	return NOTIFY_STOP;
722 }
723 
724 static struct notifier_block uv_dump_stack_nmi_nb = {
725 	.notifier_call	= uv_handle_nmi,
726 	.priority = NMI_LOCAL_LOW_PRIOR - 1,
727 };
728 
729 void uv_register_nmi_notifier(void)
730 {
731 	if (register_die_notifier(&uv_dump_stack_nmi_nb))
732 		printk(KERN_WARNING "UV NMI handler failed to register\n");
733 }
734 
735 void uv_nmi_init(void)
736 {
737 	unsigned int value;
738 
739 	/*
740 	 * Unmask NMI on all cpus
741 	 */
742 	value = apic_read(APIC_LVT1) | APIC_DM_NMI;
743 	value &= ~APIC_LVT_MASKED;
744 	apic_write(APIC_LVT1, value);
745 }
746 
747 void __init uv_system_init(void)
748 {
749 	union uvh_rh_gam_config_mmr_u  m_n_config;
750 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
751 	union uvh_node_id_u node_id;
752 	unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
753 	int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val, n_io;
754 	int gnode_extra, max_pnode = 0;
755 	unsigned long mmr_base, present, paddr;
756 	unsigned short pnode_mask, pnode_io_mask;
757 
758 	printk(KERN_INFO "UV: Found %s hub\n", is_uv1_hub() ? "UV1" : "UV2");
759 	map_low_mmrs();
760 
761 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
762 	m_val = m_n_config.s.m_skt;
763 	n_val = m_n_config.s.n_skt;
764 	mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
765 	n_io = is_uv1_hub() ? mmioh.s1.n_io : mmioh.s2.n_io;
766 	mmr_base =
767 	    uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
768 	    ~UV_MMR_ENABLE;
769 	pnode_mask = (1 << n_val) - 1;
770 	pnode_io_mask = (1 << n_io) - 1;
771 
772 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
773 	gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
774 	gnode_upper = ((unsigned long)gnode_extra  << m_val);
775 	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",
776 			n_val, m_val, n_io, gnode_upper, gnode_extra, pnode_mask, pnode_io_mask);
777 
778 	printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
779 
780 	for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
781 		uv_possible_blades +=
782 		  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
783 	printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
784 
785 	bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
786 	uv_blade_info = kzalloc(bytes, GFP_KERNEL);
787 	BUG_ON(!uv_blade_info);
788 
789 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
790 		uv_blade_info[blade].memory_nid = -1;
791 
792 	get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
793 
794 	bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
795 	uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
796 	BUG_ON(!uv_node_to_blade);
797 	memset(uv_node_to_blade, 255, bytes);
798 
799 	bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
800 	uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
801 	BUG_ON(!uv_cpu_to_blade);
802 	memset(uv_cpu_to_blade, 255, bytes);
803 
804 	blade = 0;
805 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
806 		present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
807 		for (j = 0; j < 64; j++) {
808 			if (!test_bit(j, &present))
809 				continue;
810 			pnode = (i * 64 + j) & pnode_mask;
811 			uv_blade_info[blade].pnode = pnode;
812 			uv_blade_info[blade].nr_possible_cpus = 0;
813 			uv_blade_info[blade].nr_online_cpus = 0;
814 			spin_lock_init(&uv_blade_info[blade].nmi_lock);
815 			max_pnode = max(pnode, max_pnode);
816 			blade++;
817 		}
818 	}
819 
820 	uv_bios_init();
821 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
822 			    &sn_region_size, &system_serial_number);
823 	uv_rtc_init();
824 
825 	for_each_present_cpu(cpu) {
826 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
827 
828 		nid = cpu_to_node(cpu);
829 		/*
830 		 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
831 		 */
832 		uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
833 		uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
834 		uv_cpu_hub_info(cpu)->hub_revision = uv_hub_info->hub_revision;
835 
836 		pnode = uv_apicid_to_pnode(apicid);
837 		blade = boot_pnode_to_blade(pnode);
838 		lcpu = uv_blade_info[blade].nr_possible_cpus;
839 		uv_blade_info[blade].nr_possible_cpus++;
840 
841 		/* Any node on the blade, else will contain -1. */
842 		uv_blade_info[blade].memory_nid = nid;
843 
844 		uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
845 		uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
846 		uv_cpu_hub_info(cpu)->m_val = m_val;
847 		uv_cpu_hub_info(cpu)->n_val = n_val;
848 		uv_cpu_hub_info(cpu)->numa_blade_id = blade;
849 		uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
850 		uv_cpu_hub_info(cpu)->pnode = pnode;
851 		uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
852 		uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
853 		uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
854 		uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
855 		uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
856 		uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
857 		uv_node_to_blade[nid] = blade;
858 		uv_cpu_to_blade[cpu] = blade;
859 	}
860 
861 	/* Add blade/pnode info for nodes without cpus */
862 	for_each_online_node(nid) {
863 		if (uv_node_to_blade[nid] >= 0)
864 			continue;
865 		paddr = node_start_pfn(nid) << PAGE_SHIFT;
866 		paddr = uv_soc_phys_ram_to_gpa(paddr);
867 		pnode = (paddr >> m_val) & pnode_mask;
868 		blade = boot_pnode_to_blade(pnode);
869 		uv_node_to_blade[nid] = blade;
870 	}
871 
872 	map_gru_high(max_pnode);
873 	map_mmr_high(max_pnode);
874 	map_mmioh_high(max_pnode & pnode_io_mask);
875 
876 	uv_cpu_init();
877 	uv_scir_register_cpu_notifier();
878 	uv_register_nmi_notifier();
879 	proc_mkdir("sgi_uv", NULL);
880 
881 	/* register Legacy VGA I/O redirection handler */
882 	pci_register_set_vga_state(uv_set_vga_state);
883 
884 	/*
885 	 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
886 	 * EFI is not enabled in the kdump kernel.
887 	 */
888 	if (is_kdump_kernel())
889 		reboot_type = BOOT_ACPI;
890 }
891 
892 apic_driver(apic_x2apic_uv_x);
893