xref: /openbmc/linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision bc5aa3a0)
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) 2007-2014 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/export.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 #include <linux/reboot.h>
29 
30 #include <asm/uv/uv_mmrs.h>
31 #include <asm/uv/uv_hub.h>
32 #include <asm/current.h>
33 #include <asm/pgtable.h>
34 #include <asm/uv/bios.h>
35 #include <asm/uv/uv.h>
36 #include <asm/apic.h>
37 #include <asm/ipi.h>
38 #include <asm/smp.h>
39 #include <asm/x86_init.h>
40 #include <asm/nmi.h>
41 
42 DEFINE_PER_CPU(int, x2apic_extra_bits);
43 
44 #define PR_DEVEL(fmt, args...)	pr_devel("%s: " fmt, __func__, args)
45 
46 static enum uv_system_type uv_system_type;
47 static u64 gru_start_paddr, gru_end_paddr;
48 static u64 gru_dist_base, gru_first_node_paddr = -1LL, gru_last_node_paddr;
49 static u64 gru_dist_lmask, gru_dist_umask;
50 static union uvh_apicid uvh_apicid;
51 
52 /* info derived from CPUID */
53 static struct {
54 	unsigned int apicid_shift;
55 	unsigned int apicid_mask;
56 	unsigned int socketid_shift;	/* aka pnode_shift for UV1/2/3 */
57 	unsigned int pnode_mask;
58 	unsigned int gpa_shift;
59 } uv_cpuid;
60 
61 int uv_min_hub_revision_id;
62 EXPORT_SYMBOL_GPL(uv_min_hub_revision_id);
63 unsigned int uv_apicid_hibits;
64 EXPORT_SYMBOL_GPL(uv_apicid_hibits);
65 
66 static struct apic apic_x2apic_uv_x;
67 static struct uv_hub_info_s uv_hub_info_node0;
68 
69 /* Set this to use hardware error handler instead of kernel panic */
70 static int disable_uv_undefined_panic = 1;
71 unsigned long uv_undefined(char *str)
72 {
73 	if (likely(!disable_uv_undefined_panic))
74 		panic("UV: error: undefined MMR: %s\n", str);
75 	else
76 		pr_crit("UV: error: undefined MMR: %s\n", str);
77 	return ~0ul;	/* cause a machine fault  */
78 }
79 EXPORT_SYMBOL(uv_undefined);
80 
81 static unsigned long __init uv_early_read_mmr(unsigned long addr)
82 {
83 	unsigned long val, *mmr;
84 
85 	mmr = early_ioremap(UV_LOCAL_MMR_BASE | addr, sizeof(*mmr));
86 	val = *mmr;
87 	early_iounmap(mmr, sizeof(*mmr));
88 	return val;
89 }
90 
91 static inline bool is_GRU_range(u64 start, u64 end)
92 {
93 	if (gru_dist_base) {
94 		u64 su = start & gru_dist_umask; /* upper (incl pnode) bits */
95 		u64 sl = start & gru_dist_lmask; /* base offset bits */
96 		u64 eu = end & gru_dist_umask;
97 		u64 el = end & gru_dist_lmask;
98 
99 		/* Must reside completely within a single GRU range */
100 		return (sl == gru_dist_base && el == gru_dist_base &&
101 			su >= gru_first_node_paddr &&
102 			su <= gru_last_node_paddr &&
103 			eu == su);
104 	} else {
105 		return start >= gru_start_paddr && end <= gru_end_paddr;
106 	}
107 }
108 
109 static bool uv_is_untracked_pat_range(u64 start, u64 end)
110 {
111 	return is_ISA_range(start, end) || is_GRU_range(start, end);
112 }
113 
114 static int __init early_get_pnodeid(void)
115 {
116 	union uvh_node_id_u node_id;
117 	union uvh_rh_gam_config_mmr_u  m_n_config;
118 	int pnode;
119 
120 	/* Currently, all blades have same revision number */
121 	node_id.v = uv_early_read_mmr(UVH_NODE_ID);
122 	m_n_config.v = uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR);
123 	uv_min_hub_revision_id = node_id.s.revision;
124 
125 	switch (node_id.s.part_number) {
126 	case UV2_HUB_PART_NUMBER:
127 	case UV2_HUB_PART_NUMBER_X:
128 		uv_min_hub_revision_id += UV2_HUB_REVISION_BASE - 1;
129 		break;
130 	case UV3_HUB_PART_NUMBER:
131 	case UV3_HUB_PART_NUMBER_X:
132 		uv_min_hub_revision_id += UV3_HUB_REVISION_BASE;
133 		break;
134 	case UV4_HUB_PART_NUMBER:
135 		uv_min_hub_revision_id += UV4_HUB_REVISION_BASE - 1;
136 		break;
137 	}
138 
139 	uv_hub_info->hub_revision = uv_min_hub_revision_id;
140 	uv_cpuid.pnode_mask = (1 << m_n_config.s.n_skt) - 1;
141 	pnode = (node_id.s.node_id >> 1) & uv_cpuid.pnode_mask;
142 	uv_cpuid.gpa_shift = 46;	/* default unless changed */
143 
144 	pr_info("UV: rev:%d part#:%x nodeid:%04x n_skt:%d pnmsk:%x pn:%x\n",
145 		node_id.s.revision, node_id.s.part_number, node_id.s.node_id,
146 		m_n_config.s.n_skt, uv_cpuid.pnode_mask, pnode);
147 	return pnode;
148 }
149 
150 /* [copied from arch/x86/kernel/cpu/topology.c:detect_extended_topology()] */
151 #define SMT_LEVEL	0	/* leaf 0xb SMT level */
152 #define INVALID_TYPE	0	/* leaf 0xb sub-leaf types */
153 #define SMT_TYPE	1
154 #define CORE_TYPE	2
155 #define LEAFB_SUBTYPE(ecx)		(((ecx) >> 8) & 0xff)
156 #define BITS_SHIFT_NEXT_LEVEL(eax)	((eax) & 0x1f)
157 
158 static void set_x2apic_bits(void)
159 {
160 	unsigned int eax, ebx, ecx, edx, sub_index;
161 	unsigned int sid_shift;
162 
163 	cpuid(0, &eax, &ebx, &ecx, &edx);
164 	if (eax < 0xb) {
165 		pr_info("UV: CPU does not have CPUID.11\n");
166 		return;
167 	}
168 	cpuid_count(0xb, SMT_LEVEL, &eax, &ebx, &ecx, &edx);
169 	if (ebx == 0 || (LEAFB_SUBTYPE(ecx) != SMT_TYPE)) {
170 		pr_info("UV: CPUID.11 not implemented\n");
171 		return;
172 	}
173 	sid_shift = BITS_SHIFT_NEXT_LEVEL(eax);
174 	sub_index = 1;
175 	do {
176 		cpuid_count(0xb, sub_index, &eax, &ebx, &ecx, &edx);
177 		if (LEAFB_SUBTYPE(ecx) == CORE_TYPE) {
178 			sid_shift = BITS_SHIFT_NEXT_LEVEL(eax);
179 			break;
180 		}
181 		sub_index++;
182 	} while (LEAFB_SUBTYPE(ecx) != INVALID_TYPE);
183 	uv_cpuid.apicid_shift = 0;
184 	uv_cpuid.apicid_mask = (~(-1 << sid_shift));
185 	uv_cpuid.socketid_shift = sid_shift;
186 }
187 
188 static void __init early_get_apic_socketid_shift(void)
189 {
190 	if (is_uv2_hub() || is_uv3_hub())
191 		uvh_apicid.v = uv_early_read_mmr(UVH_APICID);
192 
193 	set_x2apic_bits();
194 
195 	pr_info("UV: apicid_shift:%d apicid_mask:0x%x\n",
196 		uv_cpuid.apicid_shift, uv_cpuid.apicid_mask);
197 	pr_info("UV: socketid_shift:%d pnode_mask:0x%x\n",
198 		uv_cpuid.socketid_shift, uv_cpuid.pnode_mask);
199 }
200 
201 /*
202  * Add an extra bit as dictated by bios to the destination apicid of
203  * interrupts potentially passing through the UV HUB.  This prevents
204  * a deadlock between interrupts and IO port operations.
205  */
206 static void __init uv_set_apicid_hibit(void)
207 {
208 	union uv1h_lb_target_physical_apic_id_mask_u apicid_mask;
209 
210 	if (is_uv1_hub()) {
211 		apicid_mask.v =
212 			uv_early_read_mmr(UV1H_LB_TARGET_PHYSICAL_APIC_ID_MASK);
213 		uv_apicid_hibits =
214 			apicid_mask.s1.bit_enables & UV_APICID_HIBIT_MASK;
215 	}
216 }
217 
218 static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
219 {
220 	int pnodeid;
221 	int uv_apic;
222 
223 	if (strncmp(oem_id, "SGI", 3) != 0)
224 		return 0;
225 
226 	if (numa_off) {
227 		pr_err("UV: NUMA is off, disabling UV support\n");
228 		return 0;
229 	}
230 
231 	/* Setup early hub type field in uv_hub_info for Node 0 */
232 	uv_cpu_info->p_uv_hub_info = &uv_hub_info_node0;
233 
234 	/*
235 	 * Determine UV arch type.
236 	 *   SGI: UV100/1000
237 	 *   SGI2: UV2000/3000
238 	 *   SGI3: UV300 (truncated to 4 chars because of different varieties)
239 	 *   SGI4: UV400 (truncated to 4 chars because of different varieties)
240 	 */
241 	uv_hub_info->hub_revision =
242 		!strncmp(oem_id, "SGI4", 4) ? UV4_HUB_REVISION_BASE :
243 		!strncmp(oem_id, "SGI3", 4) ? UV3_HUB_REVISION_BASE :
244 		!strcmp(oem_id, "SGI2") ? UV2_HUB_REVISION_BASE :
245 		!strcmp(oem_id, "SGI") ? UV1_HUB_REVISION_BASE : 0;
246 
247 	if (uv_hub_info->hub_revision == 0)
248 		goto badbios;
249 
250 	pnodeid = early_get_pnodeid();
251 	early_get_apic_socketid_shift();
252 	x86_platform.is_untracked_pat_range =  uv_is_untracked_pat_range;
253 	x86_platform.nmi_init = uv_nmi_init;
254 
255 	if (!strcmp(oem_table_id, "UVX")) {		/* most common */
256 		uv_system_type = UV_X2APIC;
257 		uv_apic = 0;
258 
259 	} else if (!strcmp(oem_table_id, "UVH")) {	/* only UV1 systems */
260 		uv_system_type = UV_NON_UNIQUE_APIC;
261 		__this_cpu_write(x2apic_extra_bits,
262 			pnodeid << uvh_apicid.s.pnode_shift);
263 		uv_set_apicid_hibit();
264 		uv_apic = 1;
265 
266 	} else	if (!strcmp(oem_table_id, "UVL")) {	/* only used for */
267 		uv_system_type = UV_LEGACY_APIC;	/* very small systems */
268 		uv_apic = 0;
269 
270 	} else {
271 		goto badbios;
272 	}
273 
274 	pr_info("UV: OEM IDs %s/%s, System/HUB Types %d/%d, uv_apic %d\n",
275 		oem_id, oem_table_id, uv_system_type,
276 		uv_min_hub_revision_id, uv_apic);
277 
278 	return uv_apic;
279 
280 badbios:
281 	pr_err("UV: OEM_ID:%s OEM_TABLE_ID:%s\n", oem_id, oem_table_id);
282 	pr_err("Current BIOS not supported, update kernel and/or BIOS\n");
283 	BUG();
284 }
285 
286 enum uv_system_type get_uv_system_type(void)
287 {
288 	return uv_system_type;
289 }
290 
291 int is_uv_system(void)
292 {
293 	return uv_system_type != UV_NONE;
294 }
295 EXPORT_SYMBOL_GPL(is_uv_system);
296 
297 void **__uv_hub_info_list;
298 EXPORT_SYMBOL_GPL(__uv_hub_info_list);
299 
300 DEFINE_PER_CPU(struct uv_cpu_info_s, __uv_cpu_info);
301 EXPORT_PER_CPU_SYMBOL_GPL(__uv_cpu_info);
302 
303 short uv_possible_blades;
304 EXPORT_SYMBOL_GPL(uv_possible_blades);
305 
306 unsigned long sn_rtc_cycles_per_second;
307 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
308 
309 /* the following values are used for the per node hub info struct */
310 static __initdata unsigned short *_node_to_pnode;
311 static __initdata unsigned short _min_socket, _max_socket;
312 static __initdata unsigned short _min_pnode, _max_pnode, _gr_table_len;
313 static __initdata struct uv_gam_range_entry *uv_gre_table;
314 static __initdata struct uv_gam_parameters *uv_gp_table;
315 static __initdata unsigned short *_socket_to_node;
316 static __initdata unsigned short *_socket_to_pnode;
317 static __initdata unsigned short *_pnode_to_socket;
318 static __initdata struct uv_gam_range_s *_gr_table;
319 #define	SOCK_EMPTY	((unsigned short)~0)
320 
321 extern int uv_hub_info_version(void)
322 {
323 	return UV_HUB_INFO_VERSION;
324 }
325 EXPORT_SYMBOL(uv_hub_info_version);
326 
327 /* Build GAM range lookup table */
328 static __init void build_uv_gr_table(void)
329 {
330 	struct uv_gam_range_entry *gre = uv_gre_table;
331 	struct uv_gam_range_s *grt;
332 	unsigned long last_limit = 0, ram_limit = 0;
333 	int bytes, i, sid, lsid = -1, indx = 0, lindx = -1;
334 
335 	if (!gre)
336 		return;
337 
338 	bytes = _gr_table_len * sizeof(struct uv_gam_range_s);
339 	grt = kzalloc(bytes, GFP_KERNEL);
340 	BUG_ON(!grt);
341 	_gr_table = grt;
342 
343 	for (; gre->type != UV_GAM_RANGE_TYPE_UNUSED; gre++) {
344 		if (gre->type == UV_GAM_RANGE_TYPE_HOLE) {
345 			if (!ram_limit) {   /* mark hole between ram/non-ram */
346 				ram_limit = last_limit;
347 				last_limit = gre->limit;
348 				lsid++;
349 				continue;
350 			}
351 			last_limit = gre->limit;
352 			pr_info("UV: extra hole in GAM RE table @%d\n",
353 				(int)(gre - uv_gre_table));
354 			continue;
355 		}
356 		if (_max_socket < gre->sockid) {
357 			pr_err("UV: GAM table sockid(%d) too large(>%d) @%d\n",
358 				gre->sockid, _max_socket,
359 				(int)(gre - uv_gre_table));
360 			continue;
361 		}
362 		sid = gre->sockid - _min_socket;
363 		if (lsid < sid) {		/* new range */
364 			grt = &_gr_table[indx];
365 			grt->base = lindx;
366 			grt->nasid = gre->nasid;
367 			grt->limit = last_limit = gre->limit;
368 			lsid = sid;
369 			lindx = indx++;
370 			continue;
371 		}
372 		if (lsid == sid && !ram_limit) {	/* update range */
373 			if (grt->limit == last_limit) {	/* .. if contiguous */
374 				grt->limit = last_limit = gre->limit;
375 				continue;
376 			}
377 		}
378 		if (!ram_limit) {		/* non-contiguous ram range */
379 			grt++;
380 			grt->base = lindx;
381 			grt->nasid = gre->nasid;
382 			grt->limit = last_limit = gre->limit;
383 			continue;
384 		}
385 		grt++;				/* non-contiguous/non-ram */
386 		grt->base = grt - _gr_table;	/* base is this entry */
387 		grt->nasid = gre->nasid;
388 		grt->limit = last_limit = gre->limit;
389 		lsid++;
390 	}
391 
392 	/* shorten table if possible */
393 	grt++;
394 	i = grt - _gr_table;
395 	if (i < _gr_table_len) {
396 		void *ret;
397 
398 		bytes = i * sizeof(struct uv_gam_range_s);
399 		ret = krealloc(_gr_table, bytes, GFP_KERNEL);
400 		if (ret) {
401 			_gr_table = ret;
402 			_gr_table_len = i;
403 		}
404 	}
405 
406 	/* display resultant gam range table */
407 	for (i = 0, grt = _gr_table; i < _gr_table_len; i++, grt++) {
408 		int gb = grt->base;
409 		unsigned long start = gb < 0 ?  0 :
410 			(unsigned long)_gr_table[gb].limit << UV_GAM_RANGE_SHFT;
411 		unsigned long end =
412 			(unsigned long)grt->limit << UV_GAM_RANGE_SHFT;
413 
414 		pr_info("UV: GAM Range %2d %04x 0x%013lx-0x%013lx (%d)\n",
415 			i, grt->nasid, start, end, gb);
416 	}
417 }
418 
419 static int uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
420 {
421 	unsigned long val;
422 	int pnode;
423 
424 	pnode = uv_apicid_to_pnode(phys_apicid);
425 	phys_apicid |= uv_apicid_hibits;
426 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
427 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
428 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
429 	    APIC_DM_INIT;
430 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
431 
432 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
433 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
434 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
435 	    APIC_DM_STARTUP;
436 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
437 
438 	return 0;
439 }
440 
441 static void uv_send_IPI_one(int cpu, int vector)
442 {
443 	unsigned long apicid;
444 	int pnode;
445 
446 	apicid = per_cpu(x86_cpu_to_apicid, cpu);
447 	pnode = uv_apicid_to_pnode(apicid);
448 	uv_hub_send_ipi(pnode, apicid, vector);
449 }
450 
451 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
452 {
453 	unsigned int cpu;
454 
455 	for_each_cpu(cpu, mask)
456 		uv_send_IPI_one(cpu, vector);
457 }
458 
459 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
460 {
461 	unsigned int this_cpu = smp_processor_id();
462 	unsigned int cpu;
463 
464 	for_each_cpu(cpu, mask) {
465 		if (cpu != this_cpu)
466 			uv_send_IPI_one(cpu, vector);
467 	}
468 }
469 
470 static void uv_send_IPI_allbutself(int vector)
471 {
472 	unsigned int this_cpu = smp_processor_id();
473 	unsigned int cpu;
474 
475 	for_each_online_cpu(cpu) {
476 		if (cpu != this_cpu)
477 			uv_send_IPI_one(cpu, vector);
478 	}
479 }
480 
481 static void uv_send_IPI_all(int vector)
482 {
483 	uv_send_IPI_mask(cpu_online_mask, vector);
484 }
485 
486 static int uv_apic_id_valid(int apicid)
487 {
488 	return 1;
489 }
490 
491 static int uv_apic_id_registered(void)
492 {
493 	return 1;
494 }
495 
496 static void uv_init_apic_ldr(void)
497 {
498 }
499 
500 static int
501 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
502 			  const struct cpumask *andmask,
503 			  unsigned int *apicid)
504 {
505 	int unsigned cpu;
506 
507 	/*
508 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
509 	 * May as well be the first.
510 	 */
511 	for_each_cpu_and(cpu, cpumask, andmask) {
512 		if (cpumask_test_cpu(cpu, cpu_online_mask))
513 			break;
514 	}
515 
516 	if (likely(cpu < nr_cpu_ids)) {
517 		*apicid = per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
518 		return 0;
519 	}
520 
521 	return -EINVAL;
522 }
523 
524 static unsigned int x2apic_get_apic_id(unsigned long x)
525 {
526 	unsigned int id;
527 
528 	WARN_ON(preemptible() && num_online_cpus() > 1);
529 	id = x | __this_cpu_read(x2apic_extra_bits);
530 
531 	return id;
532 }
533 
534 static unsigned long set_apic_id(unsigned int id)
535 {
536 	unsigned long x;
537 
538 	/* maskout x2apic_extra_bits ? */
539 	x = id;
540 	return x;
541 }
542 
543 static unsigned int uv_read_apic_id(void)
544 {
545 	return x2apic_get_apic_id(apic_read(APIC_ID));
546 }
547 
548 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
549 {
550 	return uv_read_apic_id() >> index_msb;
551 }
552 
553 static void uv_send_IPI_self(int vector)
554 {
555 	apic_write(APIC_SELF_IPI, vector);
556 }
557 
558 static int uv_probe(void)
559 {
560 	return apic == &apic_x2apic_uv_x;
561 }
562 
563 static struct apic __refdata apic_x2apic_uv_x = {
564 
565 	.name				= "UV large system",
566 	.probe				= uv_probe,
567 	.acpi_madt_oem_check		= uv_acpi_madt_oem_check,
568 	.apic_id_valid			= uv_apic_id_valid,
569 	.apic_id_registered		= uv_apic_id_registered,
570 
571 	.irq_delivery_mode		= dest_Fixed,
572 	.irq_dest_mode			= 0, /* physical */
573 
574 	.target_cpus			= online_target_cpus,
575 	.disable_esr			= 0,
576 	.dest_logical			= APIC_DEST_LOGICAL,
577 	.check_apicid_used		= NULL,
578 
579 	.vector_allocation_domain	= default_vector_allocation_domain,
580 	.init_apic_ldr			= uv_init_apic_ldr,
581 
582 	.ioapic_phys_id_map		= NULL,
583 	.setup_apic_routing		= NULL,
584 	.cpu_present_to_apicid		= default_cpu_present_to_apicid,
585 	.apicid_to_cpu_present		= NULL,
586 	.check_phys_apicid_present	= default_check_phys_apicid_present,
587 	.phys_pkg_id			= uv_phys_pkg_id,
588 
589 	.get_apic_id			= x2apic_get_apic_id,
590 	.set_apic_id			= set_apic_id,
591 
592 	.cpu_mask_to_apicid_and		= uv_cpu_mask_to_apicid_and,
593 
594 	.send_IPI			= uv_send_IPI_one,
595 	.send_IPI_mask			= uv_send_IPI_mask,
596 	.send_IPI_mask_allbutself	= uv_send_IPI_mask_allbutself,
597 	.send_IPI_allbutself		= uv_send_IPI_allbutself,
598 	.send_IPI_all			= uv_send_IPI_all,
599 	.send_IPI_self			= uv_send_IPI_self,
600 
601 	.wakeup_secondary_cpu		= uv_wakeup_secondary,
602 	.inquire_remote_apic		= NULL,
603 
604 	.read				= native_apic_msr_read,
605 	.write				= native_apic_msr_write,
606 	.eoi_write			= native_apic_msr_eoi_write,
607 	.icr_read			= native_x2apic_icr_read,
608 	.icr_write			= native_x2apic_icr_write,
609 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
610 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
611 };
612 
613 static void set_x2apic_extra_bits(int pnode)
614 {
615 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
616 }
617 
618 #define	UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_LENGTH	3
619 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
620 
621 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
622 {
623 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
624 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
625 	unsigned long m_redirect;
626 	unsigned long m_overlay;
627 	int i;
628 
629 	for (i = 0; i < UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_LENGTH; i++) {
630 		switch (i) {
631 		case 0:
632 			m_redirect = UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR;
633 			m_overlay = UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR;
634 			break;
635 		case 1:
636 			m_redirect = UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR;
637 			m_overlay = UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR;
638 			break;
639 		case 2:
640 			m_redirect = UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR;
641 			m_overlay = UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR;
642 			break;
643 		}
644 		alias.v = uv_read_local_mmr(m_overlay);
645 		if (alias.s.enable && alias.s.base == 0) {
646 			*size = (1UL << alias.s.m_alias);
647 			redirect.v = uv_read_local_mmr(m_redirect);
648 			*base = (unsigned long)redirect.s.dest_base
649 							<< DEST_SHIFT;
650 			return;
651 		}
652 	}
653 	*base = *size = 0;
654 }
655 
656 enum map_type {map_wb, map_uc};
657 
658 static __init void map_high(char *id, unsigned long base, int pshift,
659 			int bshift, int max_pnode, enum map_type map_type)
660 {
661 	unsigned long bytes, paddr;
662 
663 	paddr = base << pshift;
664 	bytes = (1UL << bshift) * (max_pnode + 1);
665 	if (!paddr) {
666 		pr_info("UV: Map %s_HI base address NULL\n", id);
667 		return;
668 	}
669 	pr_debug("UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr, paddr + bytes);
670 	if (map_type == map_uc)
671 		init_extra_mapping_uc(paddr, bytes);
672 	else
673 		init_extra_mapping_wb(paddr, bytes);
674 }
675 
676 static __init void map_gru_distributed(unsigned long c)
677 {
678 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
679 	u64 paddr;
680 	unsigned long bytes;
681 	int nid;
682 
683 	gru.v = c;
684 	/* only base bits 42:28 relevant in dist mode */
685 	gru_dist_base = gru.v & 0x000007fff0000000UL;
686 	if (!gru_dist_base) {
687 		pr_info("UV: Map GRU_DIST base address NULL\n");
688 		return;
689 	}
690 	bytes = 1UL << UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
691 	gru_dist_lmask = ((1UL << uv_hub_info->m_val) - 1) & ~(bytes - 1);
692 	gru_dist_umask = ~((1UL << uv_hub_info->m_val) - 1);
693 	gru_dist_base &= gru_dist_lmask; /* Clear bits above M */
694 	for_each_online_node(nid) {
695 		paddr = ((u64)uv_node_to_pnode(nid) << uv_hub_info->m_val) |
696 				gru_dist_base;
697 		init_extra_mapping_wb(paddr, bytes);
698 		gru_first_node_paddr = min(paddr, gru_first_node_paddr);
699 		gru_last_node_paddr = max(paddr, gru_last_node_paddr);
700 	}
701 	/* Save upper (63:M) bits of address only for is_GRU_range */
702 	gru_first_node_paddr &= gru_dist_umask;
703 	gru_last_node_paddr &= gru_dist_umask;
704 	pr_debug("UV: Map GRU_DIST base 0x%016llx  0x%016llx - 0x%016llx\n",
705 		gru_dist_base, gru_first_node_paddr, gru_last_node_paddr);
706 }
707 
708 static __init void map_gru_high(int max_pnode)
709 {
710 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
711 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
712 	unsigned long mask = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_MASK;
713 	unsigned long base;
714 
715 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
716 	if (!gru.s.enable) {
717 		pr_info("UV: GRU disabled\n");
718 		return;
719 	}
720 
721 	if (is_uv3_hub() && gru.s3.mode) {
722 		map_gru_distributed(gru.v);
723 		return;
724 	}
725 	base = (gru.v & mask) >> shift;
726 	map_high("GRU", base, shift, shift, max_pnode, map_wb);
727 	gru_start_paddr = ((u64)base << shift);
728 	gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
729 }
730 
731 static __init void map_mmr_high(int max_pnode)
732 {
733 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
734 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
735 
736 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
737 	if (mmr.s.enable)
738 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
739 	else
740 		pr_info("UV: MMR disabled\n");
741 }
742 
743 /*
744  * This commonality works because both 0 & 1 versions of the MMIOH OVERLAY
745  * and REDIRECT MMR regs are exactly the same on UV3.
746  */
747 struct mmioh_config {
748 	unsigned long overlay;
749 	unsigned long redirect;
750 	char *id;
751 };
752 
753 static __initdata struct mmioh_config mmiohs[] = {
754 	{
755 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR,
756 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR,
757 		"MMIOH0"
758 	},
759 	{
760 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG1_MMR,
761 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG1_MMR,
762 		"MMIOH1"
763 	},
764 };
765 
766 /* UV3 & UV4 have identical MMIOH overlay configs */
767 static __init void map_mmioh_high_uv3(int index, int min_pnode, int max_pnode)
768 {
769 	union uv3h_rh_gam_mmioh_overlay_config0_mmr_u overlay;
770 	unsigned long mmr;
771 	unsigned long base;
772 	int i, n, shift, m_io, max_io;
773 	int nasid, lnasid, fi, li;
774 	char *id;
775 
776 	id = mmiohs[index].id;
777 	overlay.v = uv_read_local_mmr(mmiohs[index].overlay);
778 	pr_info("UV: %s overlay 0x%lx base:0x%x m_io:%d\n",
779 		id, overlay.v, overlay.s3.base, overlay.s3.m_io);
780 	if (!overlay.s3.enable) {
781 		pr_info("UV: %s disabled\n", id);
782 		return;
783 	}
784 
785 	shift = UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR_BASE_SHFT;
786 	base = (unsigned long)overlay.s3.base;
787 	m_io = overlay.s3.m_io;
788 	mmr = mmiohs[index].redirect;
789 	n = UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR_DEPTH;
790 	min_pnode *= 2;				/* convert to NASID */
791 	max_pnode *= 2;
792 	max_io = lnasid = fi = li = -1;
793 
794 	for (i = 0; i < n; i++) {
795 		union uv3h_rh_gam_mmioh_redirect_config0_mmr_u redirect;
796 
797 		redirect.v = uv_read_local_mmr(mmr + i * 8);
798 		nasid = redirect.s3.nasid;
799 		if (nasid < min_pnode || max_pnode < nasid)
800 			nasid = -1;		/* invalid NASID */
801 
802 		if (nasid == lnasid) {
803 			li = i;
804 			if (i != n-1)		/* last entry check */
805 				continue;
806 		}
807 
808 		/* check if we have a cached (or last) redirect to print */
809 		if (lnasid != -1 || (i == n-1 && nasid != -1))  {
810 			unsigned long addr1, addr2;
811 			int f, l;
812 
813 			if (lnasid == -1) {
814 				f = l = i;
815 				lnasid = nasid;
816 			} else {
817 				f = fi;
818 				l = li;
819 			}
820 			addr1 = (base << shift) +
821 				f * (unsigned long)(1 << m_io);
822 			addr2 = (base << shift) +
823 				(l + 1) * (unsigned long)(1 << m_io);
824 			pr_info("UV: %s[%03d..%03d] NASID 0x%04x ADDR 0x%016lx - 0x%016lx\n",
825 				id, fi, li, lnasid, addr1, addr2);
826 			if (max_io < l)
827 				max_io = l;
828 		}
829 		fi = li = i;
830 		lnasid = nasid;
831 	}
832 
833 	pr_info("UV: %s base:0x%lx shift:%d M_IO:%d MAX_IO:%d\n",
834 		id, base, shift, m_io, max_io);
835 
836 	if (max_io >= 0)
837 		map_high(id, base, shift, m_io, max_io, map_uc);
838 }
839 
840 static __init void map_mmioh_high(int min_pnode, int max_pnode)
841 {
842 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
843 	unsigned long mmr, base;
844 	int shift, enable, m_io, n_io;
845 
846 	if (is_uv3_hub() || is_uv4_hub()) {
847 		/* Map both MMIOH Regions */
848 		map_mmioh_high_uv3(0, min_pnode, max_pnode);
849 		map_mmioh_high_uv3(1, min_pnode, max_pnode);
850 		return;
851 	}
852 
853 	if (is_uv1_hub()) {
854 		mmr = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
855 		shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
856 		mmioh.v = uv_read_local_mmr(mmr);
857 		enable = !!mmioh.s1.enable;
858 		base = mmioh.s1.base;
859 		m_io = mmioh.s1.m_io;
860 		n_io = mmioh.s1.n_io;
861 	} else if (is_uv2_hub()) {
862 		mmr = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
863 		shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
864 		mmioh.v = uv_read_local_mmr(mmr);
865 		enable = !!mmioh.s2.enable;
866 		base = mmioh.s2.base;
867 		m_io = mmioh.s2.m_io;
868 		n_io = mmioh.s2.n_io;
869 	} else
870 		return;
871 
872 	if (enable) {
873 		max_pnode &= (1 << n_io) - 1;
874 		pr_info(
875 		    "UV: base:0x%lx shift:%d N_IO:%d M_IO:%d max_pnode:0x%x\n",
876 			base, shift, m_io, n_io, max_pnode);
877 		map_high("MMIOH", base, shift, m_io, max_pnode, map_uc);
878 	} else {
879 		pr_info("UV: MMIOH disabled\n");
880 	}
881 }
882 
883 static __init void map_low_mmrs(void)
884 {
885 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
886 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
887 }
888 
889 static __init void uv_rtc_init(void)
890 {
891 	long status;
892 	u64 ticks_per_sec;
893 
894 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
895 					&ticks_per_sec);
896 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
897 		printk(KERN_WARNING
898 			"unable to determine platform RTC clock frequency, "
899 			"guessing.\n");
900 		/* BIOS gives wrong value for clock freq. so guess */
901 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
902 	} else
903 		sn_rtc_cycles_per_second = ticks_per_sec;
904 }
905 
906 /*
907  * percpu heartbeat timer
908  */
909 static void uv_heartbeat(unsigned long ignored)
910 {
911 	struct timer_list *timer = &uv_scir_info->timer;
912 	unsigned char bits = uv_scir_info->state;
913 
914 	/* flip heartbeat bit */
915 	bits ^= SCIR_CPU_HEARTBEAT;
916 
917 	/* is this cpu idle? */
918 	if (idle_cpu(raw_smp_processor_id()))
919 		bits &= ~SCIR_CPU_ACTIVITY;
920 	else
921 		bits |= SCIR_CPU_ACTIVITY;
922 
923 	/* update system controller interface reg */
924 	uv_set_scir_bits(bits);
925 
926 	/* enable next timer period */
927 	mod_timer(timer, jiffies + SCIR_CPU_HB_INTERVAL);
928 }
929 
930 static void uv_heartbeat_enable(int cpu)
931 {
932 	while (!uv_cpu_scir_info(cpu)->enabled) {
933 		struct timer_list *timer = &uv_cpu_scir_info(cpu)->timer;
934 
935 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
936 		setup_pinned_timer(timer, uv_heartbeat, cpu);
937 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
938 		add_timer_on(timer, cpu);
939 		uv_cpu_scir_info(cpu)->enabled = 1;
940 
941 		/* also ensure that boot cpu is enabled */
942 		cpu = 0;
943 	}
944 }
945 
946 #ifdef CONFIG_HOTPLUG_CPU
947 static void uv_heartbeat_disable(int cpu)
948 {
949 	if (uv_cpu_scir_info(cpu)->enabled) {
950 		uv_cpu_scir_info(cpu)->enabled = 0;
951 		del_timer(&uv_cpu_scir_info(cpu)->timer);
952 	}
953 	uv_set_cpu_scir_bits(cpu, 0xff);
954 }
955 
956 /*
957  * cpu hotplug notifier
958  */
959 static int uv_scir_cpu_notify(struct notifier_block *self, unsigned long action,
960 			      void *hcpu)
961 {
962 	long cpu = (long)hcpu;
963 
964 	switch (action & ~CPU_TASKS_FROZEN) {
965 	case CPU_DOWN_FAILED:
966 	case CPU_ONLINE:
967 		uv_heartbeat_enable(cpu);
968 		break;
969 	case CPU_DOWN_PREPARE:
970 		uv_heartbeat_disable(cpu);
971 		break;
972 	default:
973 		break;
974 	}
975 	return NOTIFY_OK;
976 }
977 
978 static __init void uv_scir_register_cpu_notifier(void)
979 {
980 	hotcpu_notifier(uv_scir_cpu_notify, 0);
981 }
982 
983 #else /* !CONFIG_HOTPLUG_CPU */
984 
985 static __init void uv_scir_register_cpu_notifier(void)
986 {
987 }
988 
989 static __init int uv_init_heartbeat(void)
990 {
991 	int cpu;
992 
993 	if (is_uv_system())
994 		for_each_online_cpu(cpu)
995 			uv_heartbeat_enable(cpu);
996 	return 0;
997 }
998 
999 late_initcall(uv_init_heartbeat);
1000 
1001 #endif /* !CONFIG_HOTPLUG_CPU */
1002 
1003 /* Direct Legacy VGA I/O traffic to designated IOH */
1004 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
1005 		      unsigned int command_bits, u32 flags)
1006 {
1007 	int domain, bus, rc;
1008 
1009 	PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
1010 			pdev->devfn, decode, command_bits, flags);
1011 
1012 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
1013 		return 0;
1014 
1015 	if ((command_bits & PCI_COMMAND_IO) == 0)
1016 		return 0;
1017 
1018 	domain = pci_domain_nr(pdev->bus);
1019 	bus = pdev->bus->number;
1020 
1021 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
1022 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
1023 
1024 	return rc;
1025 }
1026 
1027 /*
1028  * Called on each cpu to initialize the per_cpu UV data area.
1029  * FIXME: hotplug not supported yet
1030  */
1031 void uv_cpu_init(void)
1032 {
1033 	/* CPU 0 initialization will be done via uv_system_init. */
1034 	if (smp_processor_id() == 0)
1035 		return;
1036 
1037 	uv_hub_info->nr_online_cpus++;
1038 
1039 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
1040 		set_x2apic_extra_bits(uv_hub_info->pnode);
1041 }
1042 
1043 struct mn {
1044 	unsigned char	m_val;
1045 	unsigned char	n_val;
1046 	unsigned char	m_shift;
1047 	unsigned char	n_lshift;
1048 };
1049 
1050 static void get_mn(struct mn *mnp)
1051 {
1052 	union uvh_rh_gam_config_mmr_u m_n_config;
1053 	union uv3h_gr0_gam_gr_config_u m_gr_config;
1054 
1055 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR);
1056 	mnp->n_val = m_n_config.s.n_skt;
1057 	if (is_uv4_hub()) {
1058 		mnp->m_val = 0;
1059 		mnp->n_lshift = 0;
1060 	} else if (is_uv3_hub()) {
1061 		mnp->m_val = m_n_config.s3.m_skt;
1062 		m_gr_config.v = uv_read_local_mmr(UV3H_GR0_GAM_GR_CONFIG);
1063 		mnp->n_lshift = m_gr_config.s3.m_skt;
1064 	} else if (is_uv2_hub()) {
1065 		mnp->m_val = m_n_config.s2.m_skt;
1066 		mnp->n_lshift = mnp->m_val == 40 ? 40 : 39;
1067 	} else if (is_uv1_hub()) {
1068 		mnp->m_val = m_n_config.s1.m_skt;
1069 		mnp->n_lshift = mnp->m_val;
1070 	}
1071 	mnp->m_shift = mnp->m_val ? 64 - mnp->m_val : 0;
1072 }
1073 
1074 void __init uv_init_hub_info(struct uv_hub_info_s *hub_info)
1075 {
1076 	struct mn mn = {0};	/* avoid unitialized warnings */
1077 	union uvh_node_id_u node_id;
1078 
1079 	get_mn(&mn);
1080 	hub_info->m_val = mn.m_val;
1081 	hub_info->n_val = mn.n_val;
1082 	hub_info->m_shift = mn.m_shift;
1083 	hub_info->n_lshift = mn.n_lshift ? mn.n_lshift : 0;
1084 
1085 	hub_info->hub_revision = uv_hub_info->hub_revision;
1086 	hub_info->pnode_mask = uv_cpuid.pnode_mask;
1087 	hub_info->min_pnode = _min_pnode;
1088 	hub_info->min_socket = _min_socket;
1089 	hub_info->pnode_to_socket = _pnode_to_socket;
1090 	hub_info->socket_to_node = _socket_to_node;
1091 	hub_info->socket_to_pnode = _socket_to_pnode;
1092 	hub_info->gr_table_len = _gr_table_len;
1093 	hub_info->gr_table = _gr_table;
1094 	hub_info->gpa_mask = mn.m_val ?
1095 		(1UL << (mn.m_val + mn.n_val)) - 1 :
1096 		(1UL << uv_cpuid.gpa_shift) - 1;
1097 
1098 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
1099 	hub_info->gnode_extra =
1100 		(node_id.s.node_id & ~((1 << mn.n_val) - 1)) >> 1;
1101 
1102 	hub_info->gnode_upper =
1103 		((unsigned long)hub_info->gnode_extra << mn.m_val);
1104 
1105 	if (uv_gp_table) {
1106 		hub_info->global_mmr_base = uv_gp_table->mmr_base;
1107 		hub_info->global_mmr_shift = uv_gp_table->mmr_shift;
1108 		hub_info->global_gru_base = uv_gp_table->gru_base;
1109 		hub_info->global_gru_shift = uv_gp_table->gru_shift;
1110 		hub_info->gpa_shift = uv_gp_table->gpa_shift;
1111 		hub_info->gpa_mask = (1UL << hub_info->gpa_shift) - 1;
1112 	} else {
1113 		hub_info->global_mmr_base =
1114 			uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
1115 					~UV_MMR_ENABLE;
1116 		hub_info->global_mmr_shift = _UV_GLOBAL_MMR64_PNODE_SHIFT;
1117 	}
1118 
1119 	get_lowmem_redirect(
1120 		&hub_info->lowmem_remap_base, &hub_info->lowmem_remap_top);
1121 
1122 	hub_info->apic_pnode_shift = uv_cpuid.socketid_shift;
1123 
1124 	/* show system specific info */
1125 	pr_info("UV: N:%d M:%d m_shift:%d n_lshift:%d\n",
1126 		hub_info->n_val, hub_info->m_val,
1127 		hub_info->m_shift, hub_info->n_lshift);
1128 
1129 	pr_info("UV: gpa_mask/shift:0x%lx/%d pnode_mask:0x%x apic_pns:%d\n",
1130 		hub_info->gpa_mask, hub_info->gpa_shift,
1131 		hub_info->pnode_mask, hub_info->apic_pnode_shift);
1132 
1133 	pr_info("UV: mmr_base/shift:0x%lx/%ld gru_base/shift:0x%lx/%ld\n",
1134 		hub_info->global_mmr_base, hub_info->global_mmr_shift,
1135 		hub_info->global_gru_base, hub_info->global_gru_shift);
1136 
1137 	pr_info("UV: gnode_upper:0x%lx gnode_extra:0x%x\n",
1138 		hub_info->gnode_upper, hub_info->gnode_extra);
1139 }
1140 
1141 static void __init decode_gam_params(unsigned long ptr)
1142 {
1143 	uv_gp_table = (struct uv_gam_parameters *)ptr;
1144 
1145 	pr_info("UV: GAM Params...\n");
1146 	pr_info("UV: mmr_base/shift:0x%llx/%d gru_base/shift:0x%llx/%d gpa_shift:%d\n",
1147 		uv_gp_table->mmr_base, uv_gp_table->mmr_shift,
1148 		uv_gp_table->gru_base, uv_gp_table->gru_shift,
1149 		uv_gp_table->gpa_shift);
1150 }
1151 
1152 static void __init decode_gam_rng_tbl(unsigned long ptr)
1153 {
1154 	struct uv_gam_range_entry *gre = (struct uv_gam_range_entry *)ptr;
1155 	unsigned long lgre = 0;
1156 	int index = 0;
1157 	int sock_min = 999999, pnode_min = 99999;
1158 	int sock_max = -1, pnode_max = -1;
1159 
1160 	uv_gre_table = gre;
1161 	for (; gre->type != UV_GAM_RANGE_TYPE_UNUSED; gre++) {
1162 		if (!index) {
1163 			pr_info("UV: GAM Range Table...\n");
1164 			pr_info("UV:  # %20s %14s %5s %4s %5s %3s %2s\n",
1165 				"Range", "", "Size", "Type", "NASID",
1166 				"SID", "PN");
1167 		}
1168 		pr_info(
1169 		"UV: %2d: 0x%014lx-0x%014lx %5luG %3d   %04x  %02x %02x\n",
1170 			index++,
1171 			(unsigned long)lgre << UV_GAM_RANGE_SHFT,
1172 			(unsigned long)gre->limit << UV_GAM_RANGE_SHFT,
1173 			((unsigned long)(gre->limit - lgre)) >>
1174 				(30 - UV_GAM_RANGE_SHFT), /* 64M -> 1G */
1175 			gre->type, gre->nasid, gre->sockid, gre->pnode);
1176 
1177 		lgre = gre->limit;
1178 		if (sock_min > gre->sockid)
1179 			sock_min = gre->sockid;
1180 		if (sock_max < gre->sockid)
1181 			sock_max = gre->sockid;
1182 		if (pnode_min > gre->pnode)
1183 			pnode_min = gre->pnode;
1184 		if (pnode_max < gre->pnode)
1185 			pnode_max = gre->pnode;
1186 	}
1187 	_min_socket = sock_min;
1188 	_max_socket = sock_max;
1189 	_min_pnode = pnode_min;
1190 	_max_pnode = pnode_max;
1191 	_gr_table_len = index;
1192 	pr_info(
1193 	"UV: GRT: %d entries, sockets(min:%x,max:%x) pnodes(min:%x,max:%x)\n",
1194 		index, _min_socket, _max_socket, _min_pnode, _max_pnode);
1195 }
1196 
1197 static void __init decode_uv_systab(void)
1198 {
1199 	struct uv_systab *st;
1200 	int i;
1201 
1202 	st = uv_systab;
1203 	if ((!st || st->revision < UV_SYSTAB_VERSION_UV4) && !is_uv4_hub())
1204 		return;
1205 	if (st->revision != UV_SYSTAB_VERSION_UV4_LATEST) {
1206 		pr_crit(
1207 		"UV: BIOS UVsystab version(%x) mismatch, expecting(%x)\n",
1208 			st->revision, UV_SYSTAB_VERSION_UV4_LATEST);
1209 		BUG();
1210 	}
1211 
1212 	for (i = 0; st->entry[i].type != UV_SYSTAB_TYPE_UNUSED; i++) {
1213 		unsigned long ptr = st->entry[i].offset;
1214 
1215 		if (!ptr)
1216 			continue;
1217 
1218 		ptr = ptr + (unsigned long)st;
1219 
1220 		switch (st->entry[i].type) {
1221 		case UV_SYSTAB_TYPE_GAM_PARAMS:
1222 			decode_gam_params(ptr);
1223 			break;
1224 
1225 		case UV_SYSTAB_TYPE_GAM_RNG_TBL:
1226 			decode_gam_rng_tbl(ptr);
1227 			break;
1228 		}
1229 	}
1230 }
1231 
1232 /*
1233  * Setup physical blade translations from UVH_NODE_PRESENT_TABLE
1234  * .. NB: UVH_NODE_PRESENT_TABLE is going away,
1235  * .. being replaced by GAM Range Table
1236  */
1237 static __init void boot_init_possible_blades(struct uv_hub_info_s *hub_info)
1238 {
1239 	int i, uv_pb = 0;
1240 
1241 	pr_info("UV: NODE_PRESENT_DEPTH = %d\n", UVH_NODE_PRESENT_TABLE_DEPTH);
1242 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
1243 		unsigned long np;
1244 
1245 		np = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
1246 		if (np)
1247 			pr_info("UV: NODE_PRESENT(%d) = 0x%016lx\n", i, np);
1248 
1249 		uv_pb += hweight64(np);
1250 	}
1251 	if (uv_possible_blades != uv_pb)
1252 		uv_possible_blades = uv_pb;
1253 }
1254 
1255 static void __init build_socket_tables(void)
1256 {
1257 	struct uv_gam_range_entry *gre = uv_gre_table;
1258 	int num, nump;
1259 	int cpu, i, lnid;
1260 	int minsock = _min_socket;
1261 	int maxsock = _max_socket;
1262 	int minpnode = _min_pnode;
1263 	int maxpnode = _max_pnode;
1264 	size_t bytes;
1265 
1266 	if (!gre) {
1267 		if (is_uv1_hub() || is_uv2_hub() || is_uv3_hub()) {
1268 			pr_info("UV: No UVsystab socket table, ignoring\n");
1269 			return;		/* not required */
1270 		}
1271 		pr_crit(
1272 		"UV: Error: UVsystab address translations not available!\n");
1273 		BUG();
1274 	}
1275 
1276 	/* build socket id -> node id, pnode */
1277 	num = maxsock - minsock + 1;
1278 	bytes = num * sizeof(_socket_to_node[0]);
1279 	_socket_to_node = kmalloc(bytes, GFP_KERNEL);
1280 	_socket_to_pnode = kmalloc(bytes, GFP_KERNEL);
1281 
1282 	nump = maxpnode - minpnode + 1;
1283 	bytes = nump * sizeof(_pnode_to_socket[0]);
1284 	_pnode_to_socket = kmalloc(bytes, GFP_KERNEL);
1285 	BUG_ON(!_socket_to_node || !_socket_to_pnode || !_pnode_to_socket);
1286 
1287 	for (i = 0; i < num; i++)
1288 		_socket_to_node[i] = _socket_to_pnode[i] = SOCK_EMPTY;
1289 
1290 	for (i = 0; i < nump; i++)
1291 		_pnode_to_socket[i] = SOCK_EMPTY;
1292 
1293 	/* fill in pnode/node/addr conversion list values */
1294 	pr_info("UV: GAM Building socket/pnode conversion tables\n");
1295 	for (; gre->type != UV_GAM_RANGE_TYPE_UNUSED; gre++) {
1296 		if (gre->type == UV_GAM_RANGE_TYPE_HOLE)
1297 			continue;
1298 		i = gre->sockid - minsock;
1299 		if (_socket_to_pnode[i] != SOCK_EMPTY)
1300 			continue;	/* duplicate */
1301 		_socket_to_pnode[i] = gre->pnode;
1302 
1303 		i = gre->pnode - minpnode;
1304 		_pnode_to_socket[i] = gre->sockid;
1305 
1306 		pr_info(
1307 		"UV: sid:%02x type:%d nasid:%04x pn:%02x pn2s:%2x\n",
1308 			gre->sockid, gre->type, gre->nasid,
1309 			_socket_to_pnode[gre->sockid - minsock],
1310 			_pnode_to_socket[gre->pnode - minpnode]);
1311 	}
1312 
1313 	/* Set socket -> node values */
1314 	lnid = -1;
1315 	for_each_present_cpu(cpu) {
1316 		int nid = cpu_to_node(cpu);
1317 		int apicid, sockid;
1318 
1319 		if (lnid == nid)
1320 			continue;
1321 		lnid = nid;
1322 		apicid = per_cpu(x86_cpu_to_apicid, cpu);
1323 		sockid = apicid >> uv_cpuid.socketid_shift;
1324 		_socket_to_node[sockid - minsock] = nid;
1325 		pr_info("UV: sid:%02x: apicid:%04x node:%2d\n",
1326 			sockid, apicid, nid);
1327 	}
1328 
1329 	/* Setup physical blade to pnode translation from GAM Range Table */
1330 	bytes = num_possible_nodes() * sizeof(_node_to_pnode[0]);
1331 	_node_to_pnode = kmalloc(bytes, GFP_KERNEL);
1332 	BUG_ON(!_node_to_pnode);
1333 
1334 	for (lnid = 0; lnid < num_possible_nodes(); lnid++) {
1335 		unsigned short sockid;
1336 
1337 		for (sockid = minsock; sockid <= maxsock; sockid++) {
1338 			if (lnid == _socket_to_node[sockid - minsock]) {
1339 				_node_to_pnode[lnid] =
1340 					_socket_to_pnode[sockid - minsock];
1341 				break;
1342 			}
1343 		}
1344 		if (sockid > maxsock) {
1345 			pr_err("UV: socket for node %d not found!\n", lnid);
1346 			BUG();
1347 		}
1348 	}
1349 
1350 	/*
1351 	 * If socket id == pnode or socket id == node for all nodes,
1352 	 *   system runs faster by removing corresponding conversion table.
1353 	 */
1354 	pr_info("UV: Checking socket->node/pnode for identity maps\n");
1355 	if (minsock == 0) {
1356 		for (i = 0; i < num; i++)
1357 			if (_socket_to_node[i] == SOCK_EMPTY ||
1358 				i != _socket_to_node[i])
1359 				break;
1360 		if (i >= num) {
1361 			kfree(_socket_to_node);
1362 			_socket_to_node = NULL;
1363 			pr_info("UV: 1:1 socket_to_node table removed\n");
1364 		}
1365 	}
1366 	if (minsock == minpnode) {
1367 		for (i = 0; i < num; i++)
1368 			if (_socket_to_pnode[i] != SOCK_EMPTY &&
1369 				_socket_to_pnode[i] != i + minpnode)
1370 				break;
1371 		if (i >= num) {
1372 			kfree(_socket_to_pnode);
1373 			_socket_to_pnode = NULL;
1374 			pr_info("UV: 1:1 socket_to_pnode table removed\n");
1375 		}
1376 	}
1377 }
1378 
1379 void __init uv_system_init(void)
1380 {
1381 	struct uv_hub_info_s hub_info = {0};
1382 	int bytes, cpu, nodeid;
1383 	unsigned short min_pnode = 9999, max_pnode = 0;
1384 	char *hub = is_uv4_hub() ? "UV400" :
1385 		    is_uv3_hub() ? "UV300" :
1386 		    is_uv2_hub() ? "UV2000/3000" :
1387 		    is_uv1_hub() ? "UV100/1000" : NULL;
1388 
1389 	if (!hub) {
1390 		pr_err("UV: Unknown/unsupported UV hub\n");
1391 		return;
1392 	}
1393 	pr_info("UV: Found %s hub\n", hub);
1394 
1395 	map_low_mmrs();
1396 
1397 	uv_bios_init();			/* get uv_systab for decoding */
1398 	decode_uv_systab();
1399 	build_socket_tables();
1400 	build_uv_gr_table();
1401 	uv_init_hub_info(&hub_info);
1402 	uv_possible_blades = num_possible_nodes();
1403 	if (!_node_to_pnode)
1404 		boot_init_possible_blades(&hub_info);
1405 
1406 	/* uv_num_possible_blades() is really the hub count */
1407 	pr_info("UV: Found %d hubs, %d nodes, %d cpus\n",
1408 			uv_num_possible_blades(),
1409 			num_possible_nodes(),
1410 			num_possible_cpus());
1411 
1412 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
1413 			    &sn_region_size, &system_serial_number);
1414 	hub_info.coherency_domain_number = sn_coherency_id;
1415 	uv_rtc_init();
1416 
1417 	bytes = sizeof(void *) * uv_num_possible_blades();
1418 	__uv_hub_info_list = kzalloc(bytes, GFP_KERNEL);
1419 	BUG_ON(!__uv_hub_info_list);
1420 
1421 	bytes = sizeof(struct uv_hub_info_s);
1422 	for_each_node(nodeid) {
1423 		struct uv_hub_info_s *new_hub;
1424 
1425 		if (__uv_hub_info_list[nodeid]) {
1426 			pr_err("UV: Node %d UV HUB already initialized!?\n",
1427 				nodeid);
1428 			BUG();
1429 		}
1430 
1431 		/* Allocate new per hub info list */
1432 		new_hub = (nodeid == 0) ?
1433 			&uv_hub_info_node0 :
1434 			kzalloc_node(bytes, GFP_KERNEL, nodeid);
1435 		BUG_ON(!new_hub);
1436 		__uv_hub_info_list[nodeid] = new_hub;
1437 		new_hub = uv_hub_info_list(nodeid);
1438 		BUG_ON(!new_hub);
1439 		*new_hub = hub_info;
1440 
1441 		/* Use information from GAM table if available */
1442 		if (_node_to_pnode)
1443 			new_hub->pnode = _node_to_pnode[nodeid];
1444 		else	/* Fill in during cpu loop */
1445 			new_hub->pnode = 0xffff;
1446 		new_hub->numa_blade_id = uv_node_to_blade_id(nodeid);
1447 		new_hub->memory_nid = -1;
1448 		new_hub->nr_possible_cpus = 0;
1449 		new_hub->nr_online_cpus = 0;
1450 	}
1451 
1452 	/* Initialize per cpu info */
1453 	for_each_possible_cpu(cpu) {
1454 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
1455 		int numa_node_id;
1456 		unsigned short pnode;
1457 
1458 		nodeid = cpu_to_node(cpu);
1459 		numa_node_id = numa_cpu_node(cpu);
1460 		pnode = uv_apicid_to_pnode(apicid);
1461 
1462 		uv_cpu_info_per(cpu)->p_uv_hub_info = uv_hub_info_list(nodeid);
1463 		uv_cpu_info_per(cpu)->blade_cpu_id =
1464 			uv_cpu_hub_info(cpu)->nr_possible_cpus++;
1465 		if (uv_cpu_hub_info(cpu)->memory_nid == -1)
1466 			uv_cpu_hub_info(cpu)->memory_nid = cpu_to_node(cpu);
1467 		if (nodeid != numa_node_id &&	/* init memoryless node */
1468 		    uv_hub_info_list(numa_node_id)->pnode == 0xffff)
1469 			uv_hub_info_list(numa_node_id)->pnode = pnode;
1470 		else if (uv_cpu_hub_info(cpu)->pnode == 0xffff)
1471 			uv_cpu_hub_info(cpu)->pnode = pnode;
1472 		uv_cpu_scir_info(cpu)->offset = uv_scir_offset(apicid);
1473 	}
1474 
1475 	for_each_node(nodeid) {
1476 		unsigned short pnode = uv_hub_info_list(nodeid)->pnode;
1477 
1478 		/* Add pnode info for pre-GAM list nodes without cpus */
1479 		if (pnode == 0xffff) {
1480 			unsigned long paddr;
1481 
1482 			paddr = node_start_pfn(nodeid) << PAGE_SHIFT;
1483 			pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
1484 			uv_hub_info_list(nodeid)->pnode = pnode;
1485 		}
1486 		min_pnode = min(pnode, min_pnode);
1487 		max_pnode = max(pnode, max_pnode);
1488 		pr_info("UV: UVHUB node:%2d pn:%02x nrcpus:%d\n",
1489 			nodeid,
1490 			uv_hub_info_list(nodeid)->pnode,
1491 			uv_hub_info_list(nodeid)->nr_possible_cpus);
1492 	}
1493 
1494 	pr_info("UV: min_pnode:%02x max_pnode:%02x\n", min_pnode, max_pnode);
1495 	map_gru_high(max_pnode);
1496 	map_mmr_high(max_pnode);
1497 	map_mmioh_high(min_pnode, max_pnode);
1498 
1499 	uv_nmi_setup();
1500 	uv_cpu_init();
1501 	uv_scir_register_cpu_notifier();
1502 	proc_mkdir("sgi_uv", NULL);
1503 
1504 	/* register Legacy VGA I/O redirection handler */
1505 	pci_register_set_vga_state(uv_set_vga_state);
1506 
1507 	/*
1508 	 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
1509 	 * EFI is not enabled in the kdump kernel.
1510 	 */
1511 	if (is_kdump_kernel())
1512 		reboot_type = BOOT_ACPI;
1513 }
1514 
1515 apic_driver(apic_x2apic_uv_x);
1516