xref: /linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision 071bf69a0220253a44acb8b2a27f7a262b9a46bf)
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 	/* Setup early hub type field in uv_hub_info for Node 0 */
227 	uv_cpu_info->p_uv_hub_info = &uv_hub_info_node0;
228 
229 	/*
230 	 * Determine UV arch type.
231 	 *   SGI: UV100/1000
232 	 *   SGI2: UV2000/3000
233 	 *   SGI3: UV300 (truncated to 4 chars because of different varieties)
234 	 *   SGI4: UV400 (truncated to 4 chars because of different varieties)
235 	 */
236 	uv_hub_info->hub_revision =
237 		!strncmp(oem_id, "SGI4", 4) ? UV4_HUB_REVISION_BASE :
238 		!strncmp(oem_id, "SGI3", 4) ? UV3_HUB_REVISION_BASE :
239 		!strcmp(oem_id, "SGI2") ? UV2_HUB_REVISION_BASE :
240 		!strcmp(oem_id, "SGI") ? UV1_HUB_REVISION_BASE : 0;
241 
242 	if (uv_hub_info->hub_revision == 0)
243 		goto badbios;
244 
245 	pnodeid = early_get_pnodeid();
246 	early_get_apic_socketid_shift();
247 	x86_platform.is_untracked_pat_range =  uv_is_untracked_pat_range;
248 	x86_platform.nmi_init = uv_nmi_init;
249 
250 	if (!strcmp(oem_table_id, "UVX")) {		/* most common */
251 		uv_system_type = UV_X2APIC;
252 		uv_apic = 0;
253 
254 	} else if (!strcmp(oem_table_id, "UVH")) {	/* only UV1 systems */
255 		uv_system_type = UV_NON_UNIQUE_APIC;
256 		__this_cpu_write(x2apic_extra_bits,
257 			pnodeid << uvh_apicid.s.pnode_shift);
258 		uv_set_apicid_hibit();
259 		uv_apic = 1;
260 
261 	} else	if (!strcmp(oem_table_id, "UVL")) {	/* only used for */
262 		uv_system_type = UV_LEGACY_APIC;	/* very small systems */
263 		uv_apic = 0;
264 
265 	} else {
266 		goto badbios;
267 	}
268 
269 	pr_info("UV: OEM IDs %s/%s, System/HUB Types %d/%d, uv_apic %d\n",
270 		oem_id, oem_table_id, uv_system_type,
271 		uv_min_hub_revision_id, uv_apic);
272 
273 	return uv_apic;
274 
275 badbios:
276 	pr_err("UV: OEM_ID:%s OEM_TABLE_ID:%s\n", oem_id, oem_table_id);
277 	pr_err("Current BIOS not supported, update kernel and/or BIOS\n");
278 	BUG();
279 }
280 
281 enum uv_system_type get_uv_system_type(void)
282 {
283 	return uv_system_type;
284 }
285 
286 int is_uv_system(void)
287 {
288 	return uv_system_type != UV_NONE;
289 }
290 EXPORT_SYMBOL_GPL(is_uv_system);
291 
292 void **__uv_hub_info_list;
293 EXPORT_SYMBOL_GPL(__uv_hub_info_list);
294 
295 DEFINE_PER_CPU(struct uv_cpu_info_s, __uv_cpu_info);
296 EXPORT_PER_CPU_SYMBOL_GPL(__uv_cpu_info);
297 
298 short uv_possible_blades;
299 EXPORT_SYMBOL_GPL(uv_possible_blades);
300 
301 unsigned long sn_rtc_cycles_per_second;
302 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
303 
304 /* the following values are used for the per node hub info struct */
305 static __initdata unsigned short *_node_to_pnode;
306 static __initdata unsigned short _min_socket, _max_socket;
307 static __initdata unsigned short _min_pnode, _max_pnode, _gr_table_len;
308 static __initdata struct uv_gam_range_entry *uv_gre_table;
309 static __initdata struct uv_gam_parameters *uv_gp_table;
310 static __initdata unsigned short *_socket_to_node;
311 static __initdata unsigned short *_socket_to_pnode;
312 static __initdata unsigned short *_pnode_to_socket;
313 static __initdata struct uv_gam_range_s *_gr_table;
314 #define	SOCK_EMPTY	((unsigned short)~0)
315 
316 extern int uv_hub_info_version(void)
317 {
318 	return UV_HUB_INFO_VERSION;
319 }
320 EXPORT_SYMBOL(uv_hub_info_version);
321 
322 /* Build GAM range lookup table */
323 static __init void build_uv_gr_table(void)
324 {
325 	struct uv_gam_range_entry *gre = uv_gre_table;
326 	struct uv_gam_range_s *grt;
327 	unsigned long last_limit = 0, ram_limit = 0;
328 	int bytes, i, sid, lsid = -1;
329 
330 	if (!gre)
331 		return;
332 
333 	bytes = _gr_table_len * sizeof(struct uv_gam_range_s);
334 	grt = kzalloc(bytes, GFP_KERNEL);
335 	BUG_ON(!grt);
336 	_gr_table = grt;
337 
338 	for (; gre->type != UV_GAM_RANGE_TYPE_UNUSED; gre++) {
339 		if (gre->type == UV_GAM_RANGE_TYPE_HOLE) {
340 			if (!ram_limit) {   /* mark hole between ram/non-ram */
341 				ram_limit = last_limit;
342 				last_limit = gre->limit;
343 				lsid++;
344 				continue;
345 			}
346 			last_limit = gre->limit;
347 			pr_info("UV: extra hole in GAM RE table @%d\n",
348 				(int)(gre - uv_gre_table));
349 			continue;
350 		}
351 		if (_max_socket < gre->sockid) {
352 			pr_err("UV: GAM table sockid(%d) too large(>%d) @%d\n",
353 				gre->sockid, _max_socket,
354 				(int)(gre - uv_gre_table));
355 			continue;
356 		}
357 		sid = gre->sockid - _min_socket;
358 		if (lsid < sid) {		/* new range */
359 			grt = &_gr_table[sid];
360 			grt->base = lsid;
361 			grt->nasid = gre->nasid;
362 			grt->limit = last_limit = gre->limit;
363 			lsid = sid;
364 			continue;
365 		}
366 		if (lsid == sid && !ram_limit) {	/* update range */
367 			if (grt->limit == last_limit) {	/* .. if contiguous */
368 				grt->limit = last_limit = gre->limit;
369 				continue;
370 			}
371 		}
372 		if (!ram_limit) {		/* non-contiguous ram range */
373 			grt++;
374 			grt->base = sid - 1;
375 			grt->nasid = gre->nasid;
376 			grt->limit = last_limit = gre->limit;
377 			continue;
378 		}
379 		grt++;				/* non-contiguous/non-ram */
380 		grt->base = grt - _gr_table;	/* base is this entry */
381 		grt->nasid = gre->nasid;
382 		grt->limit = last_limit = gre->limit;
383 		lsid++;
384 	}
385 
386 	/* shorten table if possible */
387 	grt++;
388 	i = grt - _gr_table;
389 	if (i < _gr_table_len) {
390 		void *ret;
391 
392 		bytes = i * sizeof(struct uv_gam_range_s);
393 		ret = krealloc(_gr_table, bytes, GFP_KERNEL);
394 		if (ret) {
395 			_gr_table = ret;
396 			_gr_table_len = i;
397 		}
398 	}
399 
400 	/* display resultant gam range table */
401 	for (i = 0, grt = _gr_table; i < _gr_table_len; i++, grt++) {
402 		int gb = grt->base;
403 		unsigned long start = gb < 0 ?  0 :
404 			(unsigned long)_gr_table[gb].limit << UV_GAM_RANGE_SHFT;
405 		unsigned long end =
406 			(unsigned long)grt->limit << UV_GAM_RANGE_SHFT;
407 
408 		pr_info("UV: GAM Range %2d %04x 0x%013lx-0x%013lx (%d)\n",
409 			i, grt->nasid, start, end, gb);
410 	}
411 }
412 
413 static int uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
414 {
415 	unsigned long val;
416 	int pnode;
417 
418 	pnode = uv_apicid_to_pnode(phys_apicid);
419 	phys_apicid |= uv_apicid_hibits;
420 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
421 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
422 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
423 	    APIC_DM_INIT;
424 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
425 
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_STARTUP;
430 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
431 
432 	return 0;
433 }
434 
435 static void uv_send_IPI_one(int cpu, int vector)
436 {
437 	unsigned long apicid;
438 	int pnode;
439 
440 	apicid = per_cpu(x86_cpu_to_apicid, cpu);
441 	pnode = uv_apicid_to_pnode(apicid);
442 	uv_hub_send_ipi(pnode, apicid, vector);
443 }
444 
445 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
446 {
447 	unsigned int cpu;
448 
449 	for_each_cpu(cpu, mask)
450 		uv_send_IPI_one(cpu, vector);
451 }
452 
453 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
454 {
455 	unsigned int this_cpu = smp_processor_id();
456 	unsigned int cpu;
457 
458 	for_each_cpu(cpu, mask) {
459 		if (cpu != this_cpu)
460 			uv_send_IPI_one(cpu, vector);
461 	}
462 }
463 
464 static void uv_send_IPI_allbutself(int vector)
465 {
466 	unsigned int this_cpu = smp_processor_id();
467 	unsigned int cpu;
468 
469 	for_each_online_cpu(cpu) {
470 		if (cpu != this_cpu)
471 			uv_send_IPI_one(cpu, vector);
472 	}
473 }
474 
475 static void uv_send_IPI_all(int vector)
476 {
477 	uv_send_IPI_mask(cpu_online_mask, vector);
478 }
479 
480 static int uv_apic_id_valid(int apicid)
481 {
482 	return 1;
483 }
484 
485 static int uv_apic_id_registered(void)
486 {
487 	return 1;
488 }
489 
490 static void uv_init_apic_ldr(void)
491 {
492 }
493 
494 static int
495 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
496 			  const struct cpumask *andmask,
497 			  unsigned int *apicid)
498 {
499 	int unsigned cpu;
500 
501 	/*
502 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
503 	 * May as well be the first.
504 	 */
505 	for_each_cpu_and(cpu, cpumask, andmask) {
506 		if (cpumask_test_cpu(cpu, cpu_online_mask))
507 			break;
508 	}
509 
510 	if (likely(cpu < nr_cpu_ids)) {
511 		*apicid = per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
512 		return 0;
513 	}
514 
515 	return -EINVAL;
516 }
517 
518 static unsigned int x2apic_get_apic_id(unsigned long x)
519 {
520 	unsigned int id;
521 
522 	WARN_ON(preemptible() && num_online_cpus() > 1);
523 	id = x | __this_cpu_read(x2apic_extra_bits);
524 
525 	return id;
526 }
527 
528 static unsigned long set_apic_id(unsigned int id)
529 {
530 	unsigned long x;
531 
532 	/* maskout x2apic_extra_bits ? */
533 	x = id;
534 	return x;
535 }
536 
537 static unsigned int uv_read_apic_id(void)
538 {
539 	return x2apic_get_apic_id(apic_read(APIC_ID));
540 }
541 
542 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
543 {
544 	return uv_read_apic_id() >> index_msb;
545 }
546 
547 static void uv_send_IPI_self(int vector)
548 {
549 	apic_write(APIC_SELF_IPI, vector);
550 }
551 
552 static int uv_probe(void)
553 {
554 	return apic == &apic_x2apic_uv_x;
555 }
556 
557 static struct apic __refdata apic_x2apic_uv_x = {
558 
559 	.name				= "UV large system",
560 	.probe				= uv_probe,
561 	.acpi_madt_oem_check		= uv_acpi_madt_oem_check,
562 	.apic_id_valid			= uv_apic_id_valid,
563 	.apic_id_registered		= uv_apic_id_registered,
564 
565 	.irq_delivery_mode		= dest_Fixed,
566 	.irq_dest_mode			= 0, /* physical */
567 
568 	.target_cpus			= online_target_cpus,
569 	.disable_esr			= 0,
570 	.dest_logical			= APIC_DEST_LOGICAL,
571 	.check_apicid_used		= NULL,
572 
573 	.vector_allocation_domain	= default_vector_allocation_domain,
574 	.init_apic_ldr			= uv_init_apic_ldr,
575 
576 	.ioapic_phys_id_map		= NULL,
577 	.setup_apic_routing		= NULL,
578 	.cpu_present_to_apicid		= default_cpu_present_to_apicid,
579 	.apicid_to_cpu_present		= NULL,
580 	.check_phys_apicid_present	= default_check_phys_apicid_present,
581 	.phys_pkg_id			= uv_phys_pkg_id,
582 
583 	.get_apic_id			= x2apic_get_apic_id,
584 	.set_apic_id			= set_apic_id,
585 
586 	.cpu_mask_to_apicid_and		= uv_cpu_mask_to_apicid_and,
587 
588 	.send_IPI			= uv_send_IPI_one,
589 	.send_IPI_mask			= uv_send_IPI_mask,
590 	.send_IPI_mask_allbutself	= uv_send_IPI_mask_allbutself,
591 	.send_IPI_allbutself		= uv_send_IPI_allbutself,
592 	.send_IPI_all			= uv_send_IPI_all,
593 	.send_IPI_self			= uv_send_IPI_self,
594 
595 	.wakeup_secondary_cpu		= uv_wakeup_secondary,
596 	.inquire_remote_apic		= NULL,
597 
598 	.read				= native_apic_msr_read,
599 	.write				= native_apic_msr_write,
600 	.eoi_write			= native_apic_msr_eoi_write,
601 	.icr_read			= native_x2apic_icr_read,
602 	.icr_write			= native_x2apic_icr_write,
603 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
604 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
605 };
606 
607 static void set_x2apic_extra_bits(int pnode)
608 {
609 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
610 }
611 
612 #define	UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_LENGTH	3
613 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
614 
615 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
616 {
617 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
618 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
619 	unsigned long m_redirect;
620 	unsigned long m_overlay;
621 	int i;
622 
623 	for (i = 0; i < UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_LENGTH; i++) {
624 		switch (i) {
625 		case 0:
626 			m_redirect = UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR;
627 			m_overlay = UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR;
628 			break;
629 		case 1:
630 			m_redirect = UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR;
631 			m_overlay = UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR;
632 			break;
633 		case 2:
634 			m_redirect = UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR;
635 			m_overlay = UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR;
636 			break;
637 		}
638 		alias.v = uv_read_local_mmr(m_overlay);
639 		if (alias.s.enable && alias.s.base == 0) {
640 			*size = (1UL << alias.s.m_alias);
641 			redirect.v = uv_read_local_mmr(m_redirect);
642 			*base = (unsigned long)redirect.s.dest_base
643 							<< DEST_SHIFT;
644 			return;
645 		}
646 	}
647 	*base = *size = 0;
648 }
649 
650 enum map_type {map_wb, map_uc};
651 
652 static __init void map_high(char *id, unsigned long base, int pshift,
653 			int bshift, int max_pnode, enum map_type map_type)
654 {
655 	unsigned long bytes, paddr;
656 
657 	paddr = base << pshift;
658 	bytes = (1UL << bshift) * (max_pnode + 1);
659 	if (!paddr) {
660 		pr_info("UV: Map %s_HI base address NULL\n", id);
661 		return;
662 	}
663 	pr_debug("UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr, paddr + bytes);
664 	if (map_type == map_uc)
665 		init_extra_mapping_uc(paddr, bytes);
666 	else
667 		init_extra_mapping_wb(paddr, bytes);
668 }
669 
670 static __init void map_gru_distributed(unsigned long c)
671 {
672 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
673 	u64 paddr;
674 	unsigned long bytes;
675 	int nid;
676 
677 	gru.v = c;
678 	/* only base bits 42:28 relevant in dist mode */
679 	gru_dist_base = gru.v & 0x000007fff0000000UL;
680 	if (!gru_dist_base) {
681 		pr_info("UV: Map GRU_DIST base address NULL\n");
682 		return;
683 	}
684 	bytes = 1UL << UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
685 	gru_dist_lmask = ((1UL << uv_hub_info->m_val) - 1) & ~(bytes - 1);
686 	gru_dist_umask = ~((1UL << uv_hub_info->m_val) - 1);
687 	gru_dist_base &= gru_dist_lmask; /* Clear bits above M */
688 	for_each_online_node(nid) {
689 		paddr = ((u64)uv_node_to_pnode(nid) << uv_hub_info->m_val) |
690 				gru_dist_base;
691 		init_extra_mapping_wb(paddr, bytes);
692 		gru_first_node_paddr = min(paddr, gru_first_node_paddr);
693 		gru_last_node_paddr = max(paddr, gru_last_node_paddr);
694 	}
695 	/* Save upper (63:M) bits of address only for is_GRU_range */
696 	gru_first_node_paddr &= gru_dist_umask;
697 	gru_last_node_paddr &= gru_dist_umask;
698 	pr_debug("UV: Map GRU_DIST base 0x%016llx  0x%016llx - 0x%016llx\n",
699 		gru_dist_base, gru_first_node_paddr, gru_last_node_paddr);
700 }
701 
702 static __init void map_gru_high(int max_pnode)
703 {
704 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
705 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
706 	unsigned long mask = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_MASK;
707 	unsigned long base;
708 
709 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
710 	if (!gru.s.enable) {
711 		pr_info("UV: GRU disabled\n");
712 		return;
713 	}
714 
715 	if (is_uv3_hub() && gru.s3.mode) {
716 		map_gru_distributed(gru.v);
717 		return;
718 	}
719 	base = (gru.v & mask) >> shift;
720 	map_high("GRU", base, shift, shift, max_pnode, map_wb);
721 	gru_start_paddr = ((u64)base << shift);
722 	gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
723 }
724 
725 static __init void map_mmr_high(int max_pnode)
726 {
727 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
728 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
729 
730 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
731 	if (mmr.s.enable)
732 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
733 	else
734 		pr_info("UV: MMR disabled\n");
735 }
736 
737 /*
738  * This commonality works because both 0 & 1 versions of the MMIOH OVERLAY
739  * and REDIRECT MMR regs are exactly the same on UV3.
740  */
741 struct mmioh_config {
742 	unsigned long overlay;
743 	unsigned long redirect;
744 	char *id;
745 };
746 
747 static __initdata struct mmioh_config mmiohs[] = {
748 	{
749 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR,
750 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR,
751 		"MMIOH0"
752 	},
753 	{
754 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG1_MMR,
755 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG1_MMR,
756 		"MMIOH1"
757 	},
758 };
759 
760 /* UV3 & UV4 have identical MMIOH overlay configs */
761 static __init void map_mmioh_high_uv3(int index, int min_pnode, int max_pnode)
762 {
763 	union uv3h_rh_gam_mmioh_overlay_config0_mmr_u overlay;
764 	unsigned long mmr;
765 	unsigned long base;
766 	int i, n, shift, m_io, max_io;
767 	int nasid, lnasid, fi, li;
768 	char *id;
769 
770 	id = mmiohs[index].id;
771 	overlay.v = uv_read_local_mmr(mmiohs[index].overlay);
772 	pr_info("UV: %s overlay 0x%lx base:0x%x m_io:%d\n",
773 		id, overlay.v, overlay.s3.base, overlay.s3.m_io);
774 	if (!overlay.s3.enable) {
775 		pr_info("UV: %s disabled\n", id);
776 		return;
777 	}
778 
779 	shift = UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR_BASE_SHFT;
780 	base = (unsigned long)overlay.s3.base;
781 	m_io = overlay.s3.m_io;
782 	mmr = mmiohs[index].redirect;
783 	n = UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR_DEPTH;
784 	min_pnode *= 2;				/* convert to NASID */
785 	max_pnode *= 2;
786 	max_io = lnasid = fi = li = -1;
787 
788 	for (i = 0; i < n; i++) {
789 		union uv3h_rh_gam_mmioh_redirect_config0_mmr_u redirect;
790 
791 		redirect.v = uv_read_local_mmr(mmr + i * 8);
792 		nasid = redirect.s3.nasid;
793 		if (nasid < min_pnode || max_pnode < nasid)
794 			nasid = -1;		/* invalid NASID */
795 
796 		if (nasid == lnasid) {
797 			li = i;
798 			if (i != n-1)		/* last entry check */
799 				continue;
800 		}
801 
802 		/* check if we have a cached (or last) redirect to print */
803 		if (lnasid != -1 || (i == n-1 && nasid != -1))  {
804 			unsigned long addr1, addr2;
805 			int f, l;
806 
807 			if (lnasid == -1) {
808 				f = l = i;
809 				lnasid = nasid;
810 			} else {
811 				f = fi;
812 				l = li;
813 			}
814 			addr1 = (base << shift) +
815 				f * (unsigned long)(1 << m_io);
816 			addr2 = (base << shift) +
817 				(l + 1) * (unsigned long)(1 << m_io);
818 			pr_info("UV: %s[%03d..%03d] NASID 0x%04x ADDR 0x%016lx - 0x%016lx\n",
819 				id, fi, li, lnasid, addr1, addr2);
820 			if (max_io < l)
821 				max_io = l;
822 		}
823 		fi = li = i;
824 		lnasid = nasid;
825 	}
826 
827 	pr_info("UV: %s base:0x%lx shift:%d M_IO:%d MAX_IO:%d\n",
828 		id, base, shift, m_io, max_io);
829 
830 	if (max_io >= 0)
831 		map_high(id, base, shift, m_io, max_io, map_uc);
832 }
833 
834 static __init void map_mmioh_high(int min_pnode, int max_pnode)
835 {
836 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
837 	unsigned long mmr, base;
838 	int shift, enable, m_io, n_io;
839 
840 	if (is_uv3_hub() || is_uv4_hub()) {
841 		/* Map both MMIOH Regions */
842 		map_mmioh_high_uv3(0, min_pnode, max_pnode);
843 		map_mmioh_high_uv3(1, min_pnode, max_pnode);
844 		return;
845 	}
846 
847 	if (is_uv1_hub()) {
848 		mmr = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
849 		shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
850 		mmioh.v = uv_read_local_mmr(mmr);
851 		enable = !!mmioh.s1.enable;
852 		base = mmioh.s1.base;
853 		m_io = mmioh.s1.m_io;
854 		n_io = mmioh.s1.n_io;
855 	} else if (is_uv2_hub()) {
856 		mmr = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
857 		shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
858 		mmioh.v = uv_read_local_mmr(mmr);
859 		enable = !!mmioh.s2.enable;
860 		base = mmioh.s2.base;
861 		m_io = mmioh.s2.m_io;
862 		n_io = mmioh.s2.n_io;
863 	} else
864 		return;
865 
866 	if (enable) {
867 		max_pnode &= (1 << n_io) - 1;
868 		pr_info(
869 		    "UV: base:0x%lx shift:%d N_IO:%d M_IO:%d max_pnode:0x%x\n",
870 			base, shift, m_io, n_io, max_pnode);
871 		map_high("MMIOH", base, shift, m_io, max_pnode, map_uc);
872 	} else {
873 		pr_info("UV: MMIOH disabled\n");
874 	}
875 }
876 
877 static __init void map_low_mmrs(void)
878 {
879 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
880 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
881 }
882 
883 static __init void uv_rtc_init(void)
884 {
885 	long status;
886 	u64 ticks_per_sec;
887 
888 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
889 					&ticks_per_sec);
890 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
891 		printk(KERN_WARNING
892 			"unable to determine platform RTC clock frequency, "
893 			"guessing.\n");
894 		/* BIOS gives wrong value for clock freq. so guess */
895 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
896 	} else
897 		sn_rtc_cycles_per_second = ticks_per_sec;
898 }
899 
900 /*
901  * percpu heartbeat timer
902  */
903 static void uv_heartbeat(unsigned long ignored)
904 {
905 	struct timer_list *timer = &uv_scir_info->timer;
906 	unsigned char bits = uv_scir_info->state;
907 
908 	/* flip heartbeat bit */
909 	bits ^= SCIR_CPU_HEARTBEAT;
910 
911 	/* is this cpu idle? */
912 	if (idle_cpu(raw_smp_processor_id()))
913 		bits &= ~SCIR_CPU_ACTIVITY;
914 	else
915 		bits |= SCIR_CPU_ACTIVITY;
916 
917 	/* update system controller interface reg */
918 	uv_set_scir_bits(bits);
919 
920 	/* enable next timer period */
921 	mod_timer(timer, jiffies + SCIR_CPU_HB_INTERVAL);
922 }
923 
924 static void uv_heartbeat_enable(int cpu)
925 {
926 	while (!uv_cpu_scir_info(cpu)->enabled) {
927 		struct timer_list *timer = &uv_cpu_scir_info(cpu)->timer;
928 
929 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
930 		setup_pinned_timer(timer, uv_heartbeat, cpu);
931 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
932 		add_timer_on(timer, cpu);
933 		uv_cpu_scir_info(cpu)->enabled = 1;
934 
935 		/* also ensure that boot cpu is enabled */
936 		cpu = 0;
937 	}
938 }
939 
940 #ifdef CONFIG_HOTPLUG_CPU
941 static void uv_heartbeat_disable(int cpu)
942 {
943 	if (uv_cpu_scir_info(cpu)->enabled) {
944 		uv_cpu_scir_info(cpu)->enabled = 0;
945 		del_timer(&uv_cpu_scir_info(cpu)->timer);
946 	}
947 	uv_set_cpu_scir_bits(cpu, 0xff);
948 }
949 
950 /*
951  * cpu hotplug notifier
952  */
953 static int uv_scir_cpu_notify(struct notifier_block *self, unsigned long action,
954 			      void *hcpu)
955 {
956 	long cpu = (long)hcpu;
957 
958 	switch (action & ~CPU_TASKS_FROZEN) {
959 	case CPU_DOWN_FAILED:
960 	case CPU_ONLINE:
961 		uv_heartbeat_enable(cpu);
962 		break;
963 	case CPU_DOWN_PREPARE:
964 		uv_heartbeat_disable(cpu);
965 		break;
966 	default:
967 		break;
968 	}
969 	return NOTIFY_OK;
970 }
971 
972 static __init void uv_scir_register_cpu_notifier(void)
973 {
974 	hotcpu_notifier(uv_scir_cpu_notify, 0);
975 }
976 
977 #else /* !CONFIG_HOTPLUG_CPU */
978 
979 static __init void uv_scir_register_cpu_notifier(void)
980 {
981 }
982 
983 static __init int uv_init_heartbeat(void)
984 {
985 	int cpu;
986 
987 	if (is_uv_system())
988 		for_each_online_cpu(cpu)
989 			uv_heartbeat_enable(cpu);
990 	return 0;
991 }
992 
993 late_initcall(uv_init_heartbeat);
994 
995 #endif /* !CONFIG_HOTPLUG_CPU */
996 
997 /* Direct Legacy VGA I/O traffic to designated IOH */
998 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
999 		      unsigned int command_bits, u32 flags)
1000 {
1001 	int domain, bus, rc;
1002 
1003 	PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
1004 			pdev->devfn, decode, command_bits, flags);
1005 
1006 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
1007 		return 0;
1008 
1009 	if ((command_bits & PCI_COMMAND_IO) == 0)
1010 		return 0;
1011 
1012 	domain = pci_domain_nr(pdev->bus);
1013 	bus = pdev->bus->number;
1014 
1015 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
1016 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
1017 
1018 	return rc;
1019 }
1020 
1021 /*
1022  * Called on each cpu to initialize the per_cpu UV data area.
1023  * FIXME: hotplug not supported yet
1024  */
1025 void uv_cpu_init(void)
1026 {
1027 	/* CPU 0 initialization will be done via uv_system_init. */
1028 	if (smp_processor_id() == 0)
1029 		return;
1030 
1031 	uv_hub_info->nr_online_cpus++;
1032 
1033 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
1034 		set_x2apic_extra_bits(uv_hub_info->pnode);
1035 }
1036 
1037 struct mn {
1038 	unsigned char	m_val;
1039 	unsigned char	n_val;
1040 	unsigned char	m_shift;
1041 	unsigned char	n_lshift;
1042 };
1043 
1044 static void get_mn(struct mn *mnp)
1045 {
1046 	union uvh_rh_gam_config_mmr_u m_n_config;
1047 	union uv3h_gr0_gam_gr_config_u m_gr_config;
1048 
1049 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR);
1050 	mnp->n_val = m_n_config.s.n_skt;
1051 	if (is_uv4_hub()) {
1052 		mnp->m_val = 0;
1053 		mnp->n_lshift = 0;
1054 	} else if (is_uv3_hub()) {
1055 		mnp->m_val = m_n_config.s3.m_skt;
1056 		m_gr_config.v = uv_read_local_mmr(UV3H_GR0_GAM_GR_CONFIG);
1057 		mnp->n_lshift = m_gr_config.s3.m_skt;
1058 	} else if (is_uv2_hub()) {
1059 		mnp->m_val = m_n_config.s2.m_skt;
1060 		mnp->n_lshift = mnp->m_val == 40 ? 40 : 39;
1061 	} else if (is_uv1_hub()) {
1062 		mnp->m_val = m_n_config.s1.m_skt;
1063 		mnp->n_lshift = mnp->m_val;
1064 	}
1065 	mnp->m_shift = mnp->m_val ? 64 - mnp->m_val : 0;
1066 }
1067 
1068 void __init uv_init_hub_info(struct uv_hub_info_s *hub_info)
1069 {
1070 	struct mn mn = {0};	/* avoid unitialized warnings */
1071 	union uvh_node_id_u node_id;
1072 
1073 	get_mn(&mn);
1074 	hub_info->m_val = mn.m_val;
1075 	hub_info->n_val = mn.n_val;
1076 	hub_info->m_shift = mn.m_shift;
1077 	hub_info->n_lshift = mn.n_lshift ? mn.n_lshift : 0;
1078 
1079 	hub_info->hub_revision = uv_hub_info->hub_revision;
1080 	hub_info->pnode_mask = uv_cpuid.pnode_mask;
1081 	hub_info->min_pnode = _min_pnode;
1082 	hub_info->min_socket = _min_socket;
1083 	hub_info->pnode_to_socket = _pnode_to_socket;
1084 	hub_info->socket_to_node = _socket_to_node;
1085 	hub_info->socket_to_pnode = _socket_to_pnode;
1086 	hub_info->gr_table_len = _gr_table_len;
1087 	hub_info->gr_table = _gr_table;
1088 	hub_info->gpa_mask = mn.m_val ?
1089 		(1UL << (mn.m_val + mn.n_val)) - 1 :
1090 		(1UL << uv_cpuid.gpa_shift) - 1;
1091 
1092 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
1093 	hub_info->gnode_extra =
1094 		(node_id.s.node_id & ~((1 << mn.n_val) - 1)) >> 1;
1095 
1096 	hub_info->gnode_upper =
1097 		((unsigned long)hub_info->gnode_extra << mn.m_val);
1098 
1099 	if (uv_gp_table) {
1100 		hub_info->global_mmr_base = uv_gp_table->mmr_base;
1101 		hub_info->global_mmr_shift = uv_gp_table->mmr_shift;
1102 		hub_info->global_gru_base = uv_gp_table->gru_base;
1103 		hub_info->global_gru_shift = uv_gp_table->gru_shift;
1104 		hub_info->gpa_shift = uv_gp_table->gpa_shift;
1105 		hub_info->gpa_mask = (1UL << hub_info->gpa_shift) - 1;
1106 	} else {
1107 		hub_info->global_mmr_base =
1108 			uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
1109 					~UV_MMR_ENABLE;
1110 		hub_info->global_mmr_shift = _UV_GLOBAL_MMR64_PNODE_SHIFT;
1111 	}
1112 
1113 	get_lowmem_redirect(
1114 		&hub_info->lowmem_remap_base, &hub_info->lowmem_remap_top);
1115 
1116 	hub_info->apic_pnode_shift = uv_cpuid.socketid_shift;
1117 
1118 	/* show system specific info */
1119 	pr_info("UV: N:%d M:%d m_shift:%d n_lshift:%d\n",
1120 		hub_info->n_val, hub_info->m_val,
1121 		hub_info->m_shift, hub_info->n_lshift);
1122 
1123 	pr_info("UV: gpa_mask/shift:0x%lx/%d pnode_mask:0x%x apic_pns:%d\n",
1124 		hub_info->gpa_mask, hub_info->gpa_shift,
1125 		hub_info->pnode_mask, hub_info->apic_pnode_shift);
1126 
1127 	pr_info("UV: mmr_base/shift:0x%lx/%ld gru_base/shift:0x%lx/%ld\n",
1128 		hub_info->global_mmr_base, hub_info->global_mmr_shift,
1129 		hub_info->global_gru_base, hub_info->global_gru_shift);
1130 
1131 	pr_info("UV: gnode_upper:0x%lx gnode_extra:0x%x\n",
1132 		hub_info->gnode_upper, hub_info->gnode_extra);
1133 }
1134 
1135 static void __init decode_gam_params(unsigned long ptr)
1136 {
1137 	uv_gp_table = (struct uv_gam_parameters *)ptr;
1138 
1139 	pr_info("UV: GAM Params...\n");
1140 	pr_info("UV: mmr_base/shift:0x%llx/%d gru_base/shift:0x%llx/%d gpa_shift:%d\n",
1141 		uv_gp_table->mmr_base, uv_gp_table->mmr_shift,
1142 		uv_gp_table->gru_base, uv_gp_table->gru_shift,
1143 		uv_gp_table->gpa_shift);
1144 }
1145 
1146 static void __init decode_gam_rng_tbl(unsigned long ptr)
1147 {
1148 	struct uv_gam_range_entry *gre = (struct uv_gam_range_entry *)ptr;
1149 	unsigned long lgre = 0;
1150 	int index = 0;
1151 	int sock_min = 999999, pnode_min = 99999;
1152 	int sock_max = -1, pnode_max = -1;
1153 
1154 	uv_gre_table = gre;
1155 	for (; gre->type != UV_GAM_RANGE_TYPE_UNUSED; gre++) {
1156 		if (!index) {
1157 			pr_info("UV: GAM Range Table...\n");
1158 			pr_info("UV:  # %20s %14s %5s %4s %5s %3s %2s %3s\n",
1159 				"Range", "", "Size", "Type", "NASID",
1160 				"SID", "PN", "PXM");
1161 		}
1162 		pr_info(
1163 		"UV: %2d: 0x%014lx-0x%014lx %5luG %3d   %04x  %02x %02x %3d\n",
1164 			index++,
1165 			(unsigned long)lgre << UV_GAM_RANGE_SHFT,
1166 			(unsigned long)gre->limit << UV_GAM_RANGE_SHFT,
1167 			((unsigned long)(gre->limit - lgre)) >>
1168 				(30 - UV_GAM_RANGE_SHFT), /* 64M -> 1G */
1169 			gre->type, gre->nasid, gre->sockid,
1170 			gre->pnode, gre->pxm);
1171 
1172 		lgre = gre->limit;
1173 		if (sock_min > gre->sockid)
1174 			sock_min = gre->sockid;
1175 		if (sock_max < gre->sockid)
1176 			sock_max = gre->sockid;
1177 		if (pnode_min > gre->pnode)
1178 			pnode_min = gre->pnode;
1179 		if (pnode_max < gre->pnode)
1180 			pnode_max = gre->pnode;
1181 	}
1182 	_min_socket = sock_min;
1183 	_max_socket = sock_max;
1184 	_min_pnode = pnode_min;
1185 	_max_pnode = pnode_max;
1186 	_gr_table_len = index;
1187 	pr_info(
1188 	"UV: GRT: %d entries, sockets(min:%x,max:%x) pnodes(min:%x,max:%x)\n",
1189 		index, _min_socket, _max_socket, _min_pnode, _max_pnode);
1190 }
1191 
1192 static void __init decode_uv_systab(void)
1193 {
1194 	struct uv_systab *st;
1195 	int i;
1196 
1197 	st = uv_systab;
1198 	if ((!st || st->revision < UV_SYSTAB_VERSION_UV4) && !is_uv4_hub())
1199 		return;
1200 	if (st->revision != UV_SYSTAB_VERSION_UV4_LATEST) {
1201 		pr_crit(
1202 		"UV: BIOS UVsystab version(%x) mismatch, expecting(%x)\n",
1203 			st->revision, UV_SYSTAB_VERSION_UV4_LATEST);
1204 		BUG();
1205 	}
1206 
1207 	for (i = 0; st->entry[i].type != UV_SYSTAB_TYPE_UNUSED; i++) {
1208 		unsigned long ptr = st->entry[i].offset;
1209 
1210 		if (!ptr)
1211 			continue;
1212 
1213 		ptr = ptr + (unsigned long)st;
1214 
1215 		switch (st->entry[i].type) {
1216 		case UV_SYSTAB_TYPE_GAM_PARAMS:
1217 			decode_gam_params(ptr);
1218 			break;
1219 
1220 		case UV_SYSTAB_TYPE_GAM_RNG_TBL:
1221 			decode_gam_rng_tbl(ptr);
1222 			break;
1223 		}
1224 	}
1225 }
1226 
1227 /*
1228  * Setup physical blade translations from UVH_NODE_PRESENT_TABLE
1229  * .. NB: UVH_NODE_PRESENT_TABLE is going away,
1230  * .. being replaced by GAM Range Table
1231  */
1232 static __init void boot_init_possible_blades(struct uv_hub_info_s *hub_info)
1233 {
1234 	int i, uv_pb = 0;
1235 
1236 	pr_info("UV: NODE_PRESENT_DEPTH = %d\n", UVH_NODE_PRESENT_TABLE_DEPTH);
1237 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
1238 		unsigned long np;
1239 
1240 		np = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
1241 		if (np)
1242 			pr_info("UV: NODE_PRESENT(%d) = 0x%016lx\n", i, np);
1243 
1244 		uv_pb += hweight64(np);
1245 	}
1246 	if (uv_possible_blades != uv_pb)
1247 		uv_possible_blades = uv_pb;
1248 }
1249 
1250 static void __init build_socket_tables(void)
1251 {
1252 	struct uv_gam_range_entry *gre = uv_gre_table;
1253 	int num, nump;
1254 	int cpu, i, lnid;
1255 	int minsock = _min_socket;
1256 	int maxsock = _max_socket;
1257 	int minpnode = _min_pnode;
1258 	int maxpnode = _max_pnode;
1259 	size_t bytes;
1260 
1261 	if (!gre) {
1262 		if (is_uv1_hub() || is_uv2_hub() || is_uv3_hub()) {
1263 			pr_info("UV: No UVsystab socket table, ignoring\n");
1264 			return;		/* not required */
1265 		}
1266 		pr_crit(
1267 		"UV: Error: UVsystab address translations not available!\n");
1268 		BUG();
1269 	}
1270 
1271 	/* build socket id -> node id, pnode */
1272 	num = maxsock - minsock + 1;
1273 	bytes = num * sizeof(_socket_to_node[0]);
1274 	_socket_to_node = kmalloc(bytes, GFP_KERNEL);
1275 	_socket_to_pnode = kmalloc(bytes, GFP_KERNEL);
1276 
1277 	nump = maxpnode - minpnode + 1;
1278 	bytes = nump * sizeof(_pnode_to_socket[0]);
1279 	_pnode_to_socket = kmalloc(bytes, GFP_KERNEL);
1280 	BUG_ON(!_socket_to_node || !_socket_to_pnode || !_pnode_to_socket);
1281 
1282 	for (i = 0; i < num; i++)
1283 		_socket_to_node[i] = _socket_to_pnode[i] = SOCK_EMPTY;
1284 
1285 	for (i = 0; i < nump; i++)
1286 		_pnode_to_socket[i] = SOCK_EMPTY;
1287 
1288 	/* fill in pnode/node/addr conversion list values */
1289 	pr_info("UV: GAM Building socket/pnode/pxm conversion tables\n");
1290 	for (; gre->type != UV_GAM_RANGE_TYPE_UNUSED; gre++) {
1291 		if (gre->type == UV_GAM_RANGE_TYPE_HOLE)
1292 			continue;
1293 		i = gre->sockid - minsock;
1294 		if (_socket_to_pnode[i] != SOCK_EMPTY)
1295 			continue;	/* duplicate */
1296 		_socket_to_pnode[i] = gre->pnode;
1297 		_socket_to_node[i] = gre->pxm;
1298 
1299 		i = gre->pnode - minpnode;
1300 		_pnode_to_socket[i] = gre->sockid;
1301 
1302 		pr_info(
1303 		"UV: sid:%02x type:%d nasid:%04x pn:%02x pxm:%2d pn2s:%2x\n",
1304 			gre->sockid, gre->type, gre->nasid,
1305 			_socket_to_pnode[gre->sockid - minsock],
1306 			_socket_to_node[gre->sockid - minsock],
1307 			_pnode_to_socket[gre->pnode - minpnode]);
1308 	}
1309 
1310 	/* check socket -> node values */
1311 	lnid = -1;
1312 	for_each_present_cpu(cpu) {
1313 		int nid = cpu_to_node(cpu);
1314 		int apicid, sockid;
1315 
1316 		if (lnid == nid)
1317 			continue;
1318 		lnid = nid;
1319 		apicid = per_cpu(x86_cpu_to_apicid, cpu);
1320 		sockid = apicid >> uv_cpuid.socketid_shift;
1321 		i = sockid - minsock;
1322 
1323 		if (nid != _socket_to_node[i]) {
1324 			pr_warn(
1325 			"UV: %02x: type:%d socket:%02x PXM:%02x != node:%2d\n",
1326 				i, sockid, gre->type, _socket_to_node[i], nid);
1327 			_socket_to_node[i] = nid;
1328 		}
1329 	}
1330 
1331 	/* Setup physical blade to pnode translation from GAM Range Table */
1332 	bytes = num_possible_nodes() * sizeof(_node_to_pnode[0]);
1333 	_node_to_pnode = kmalloc(bytes, GFP_KERNEL);
1334 	BUG_ON(!_node_to_pnode);
1335 
1336 	for (lnid = 0; lnid < num_possible_nodes(); lnid++) {
1337 		unsigned short sockid;
1338 
1339 		for (sockid = minsock; sockid <= maxsock; sockid++) {
1340 			if (lnid == _socket_to_node[sockid - minsock]) {
1341 				_node_to_pnode[lnid] =
1342 					_socket_to_pnode[sockid - minsock];
1343 				break;
1344 			}
1345 		}
1346 		if (sockid > maxsock) {
1347 			pr_err("UV: socket for node %d not found!\n", lnid);
1348 			BUG();
1349 		}
1350 	}
1351 
1352 	/*
1353 	 * If socket id == pnode or socket id == node for all nodes,
1354 	 *   system runs faster by removing corresponding conversion table.
1355 	 */
1356 	pr_info("UV: Checking socket->node/pnode for identity maps\n");
1357 	if (minsock == 0) {
1358 		for (i = 0; i < num; i++)
1359 			if (_socket_to_node[i] == SOCK_EMPTY ||
1360 				i != _socket_to_node[i])
1361 				break;
1362 		if (i >= num) {
1363 			kfree(_socket_to_node);
1364 			_socket_to_node = NULL;
1365 			pr_info("UV: 1:1 socket_to_node table removed\n");
1366 		}
1367 	}
1368 	if (minsock == minpnode) {
1369 		for (i = 0; i < num; i++)
1370 			if (_socket_to_pnode[i] != SOCK_EMPTY &&
1371 				_socket_to_pnode[i] != i + minpnode)
1372 				break;
1373 		if (i >= num) {
1374 			kfree(_socket_to_pnode);
1375 			_socket_to_pnode = NULL;
1376 			pr_info("UV: 1:1 socket_to_pnode table removed\n");
1377 		}
1378 	}
1379 }
1380 
1381 void __init uv_system_init(void)
1382 {
1383 	struct uv_hub_info_s hub_info = {0};
1384 	int bytes, cpu, nodeid;
1385 	unsigned short min_pnode = 9999, max_pnode = 0;
1386 	char *hub = is_uv4_hub() ? "UV400" :
1387 		    is_uv3_hub() ? "UV300" :
1388 		    is_uv2_hub() ? "UV2000/3000" :
1389 		    is_uv1_hub() ? "UV100/1000" : NULL;
1390 
1391 	if (!hub) {
1392 		pr_err("UV: Unknown/unsupported UV hub\n");
1393 		return;
1394 	}
1395 	pr_info("UV: Found %s hub\n", hub);
1396 
1397 	map_low_mmrs();
1398 
1399 	uv_bios_init();			/* get uv_systab for decoding */
1400 	decode_uv_systab();
1401 	build_socket_tables();
1402 	build_uv_gr_table();
1403 	uv_init_hub_info(&hub_info);
1404 	uv_possible_blades = num_possible_nodes();
1405 	if (!_node_to_pnode)
1406 		boot_init_possible_blades(&hub_info);
1407 
1408 	/* uv_num_possible_blades() is really the hub count */
1409 	pr_info("UV: Found %d hubs, %d nodes, %d cpus\n",
1410 			uv_num_possible_blades(),
1411 			num_possible_nodes(),
1412 			num_possible_cpus());
1413 
1414 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
1415 			    &sn_region_size, &system_serial_number);
1416 	hub_info.coherency_domain_number = sn_coherency_id;
1417 	uv_rtc_init();
1418 
1419 	bytes = sizeof(void *) * uv_num_possible_blades();
1420 	__uv_hub_info_list = kzalloc(bytes, GFP_KERNEL);
1421 	BUG_ON(!__uv_hub_info_list);
1422 
1423 	bytes = sizeof(struct uv_hub_info_s);
1424 	for_each_node(nodeid) {
1425 		struct uv_hub_info_s *new_hub;
1426 
1427 		if (__uv_hub_info_list[nodeid]) {
1428 			pr_err("UV: Node %d UV HUB already initialized!?\n",
1429 				nodeid);
1430 			BUG();
1431 		}
1432 
1433 		/* Allocate new per hub info list */
1434 		new_hub = (nodeid == 0) ?
1435 			&uv_hub_info_node0 :
1436 			kzalloc_node(bytes, GFP_KERNEL, nodeid);
1437 		BUG_ON(!new_hub);
1438 		__uv_hub_info_list[nodeid] = new_hub;
1439 		new_hub = uv_hub_info_list(nodeid);
1440 		BUG_ON(!new_hub);
1441 		*new_hub = hub_info;
1442 
1443 		/* Use information from GAM table if available */
1444 		if (_node_to_pnode)
1445 			new_hub->pnode = _node_to_pnode[nodeid];
1446 		else	/* Fill in during cpu loop */
1447 			new_hub->pnode = 0xffff;
1448 		new_hub->numa_blade_id = uv_node_to_blade_id(nodeid);
1449 		new_hub->memory_nid = -1;
1450 		new_hub->nr_possible_cpus = 0;
1451 		new_hub->nr_online_cpus = 0;
1452 	}
1453 
1454 	/* Initialize per cpu info */
1455 	for_each_possible_cpu(cpu) {
1456 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
1457 		int numa_node_id;
1458 		unsigned short pnode;
1459 
1460 		nodeid = cpu_to_node(cpu);
1461 		numa_node_id = numa_cpu_node(cpu);
1462 		pnode = uv_apicid_to_pnode(apicid);
1463 
1464 		uv_cpu_info_per(cpu)->p_uv_hub_info = uv_hub_info_list(nodeid);
1465 		uv_cpu_info_per(cpu)->blade_cpu_id =
1466 			uv_cpu_hub_info(cpu)->nr_possible_cpus++;
1467 		if (uv_cpu_hub_info(cpu)->memory_nid == -1)
1468 			uv_cpu_hub_info(cpu)->memory_nid = cpu_to_node(cpu);
1469 		if (nodeid != numa_node_id &&	/* init memoryless node */
1470 		    uv_hub_info_list(numa_node_id)->pnode == 0xffff)
1471 			uv_hub_info_list(numa_node_id)->pnode = pnode;
1472 		else if (uv_cpu_hub_info(cpu)->pnode == 0xffff)
1473 			uv_cpu_hub_info(cpu)->pnode = pnode;
1474 		uv_cpu_scir_info(cpu)->offset = uv_scir_offset(apicid);
1475 	}
1476 
1477 	for_each_node(nodeid) {
1478 		unsigned short pnode = uv_hub_info_list(nodeid)->pnode;
1479 
1480 		/* Add pnode info for pre-GAM list nodes without cpus */
1481 		if (pnode == 0xffff) {
1482 			unsigned long paddr;
1483 
1484 			paddr = node_start_pfn(nodeid) << PAGE_SHIFT;
1485 			pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
1486 			uv_hub_info_list(nodeid)->pnode = pnode;
1487 		}
1488 		min_pnode = min(pnode, min_pnode);
1489 		max_pnode = max(pnode, max_pnode);
1490 		pr_info("UV: UVHUB node:%2d pn:%02x nrcpus:%d\n",
1491 			nodeid,
1492 			uv_hub_info_list(nodeid)->pnode,
1493 			uv_hub_info_list(nodeid)->nr_possible_cpus);
1494 	}
1495 
1496 	pr_info("UV: min_pnode:%02x max_pnode:%02x\n", min_pnode, max_pnode);
1497 	map_gru_high(max_pnode);
1498 	map_mmr_high(max_pnode);
1499 	map_mmioh_high(min_pnode, max_pnode);
1500 
1501 	uv_nmi_setup();
1502 	uv_cpu_init();
1503 	uv_scir_register_cpu_notifier();
1504 	proc_mkdir("sgi_uv", NULL);
1505 
1506 	/* register Legacy VGA I/O redirection handler */
1507 	pci_register_set_vga_state(uv_set_vga_state);
1508 
1509 	/*
1510 	 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
1511 	 * EFI is not enabled in the kdump kernel.
1512 	 */
1513 	if (is_kdump_kernel())
1514 		reboot_type = BOOT_ACPI;
1515 }
1516 
1517 apic_driver(apic_x2apic_uv_x);
1518