xref: /openbmc/linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision e2f1cf25)
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/module.h>
16 #include <linux/string.h>
17 #include <linux/ctype.h>
18 #include <linux/sched.h>
19 #include <linux/timer.h>
20 #include <linux/slab.h>
21 #include <linux/cpu.h>
22 #include <linux/init.h>
23 #include <linux/io.h>
24 #include <linux/pci.h>
25 #include <linux/kdebug.h>
26 #include <linux/delay.h>
27 #include <linux/crash_dump.h>
28 #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 int uv_min_hub_revision_id;
52 EXPORT_SYMBOL_GPL(uv_min_hub_revision_id);
53 unsigned int uv_apicid_hibits;
54 EXPORT_SYMBOL_GPL(uv_apicid_hibits);
55 
56 static struct apic apic_x2apic_uv_x;
57 
58 static unsigned long __init uv_early_read_mmr(unsigned long addr)
59 {
60 	unsigned long val, *mmr;
61 
62 	mmr = early_ioremap(UV_LOCAL_MMR_BASE | addr, sizeof(*mmr));
63 	val = *mmr;
64 	early_iounmap(mmr, sizeof(*mmr));
65 	return val;
66 }
67 
68 static inline bool is_GRU_range(u64 start, u64 end)
69 {
70 	if (gru_dist_base) {
71 		u64 su = start & gru_dist_umask; /* upper (incl pnode) bits */
72 		u64 sl = start & gru_dist_lmask; /* base offset bits */
73 		u64 eu = end & gru_dist_umask;
74 		u64 el = end & gru_dist_lmask;
75 
76 		/* Must reside completely within a single GRU range */
77 		return (sl == gru_dist_base && el == gru_dist_base &&
78 			su >= gru_first_node_paddr &&
79 			su <= gru_last_node_paddr &&
80 			eu == su);
81 	} else {
82 		return start >= gru_start_paddr && end <= gru_end_paddr;
83 	}
84 }
85 
86 static bool uv_is_untracked_pat_range(u64 start, u64 end)
87 {
88 	return is_ISA_range(start, end) || is_GRU_range(start, end);
89 }
90 
91 static int __init early_get_pnodeid(void)
92 {
93 	union uvh_node_id_u node_id;
94 	union uvh_rh_gam_config_mmr_u  m_n_config;
95 	int pnode;
96 
97 	/* Currently, all blades have same revision number */
98 	node_id.v = uv_early_read_mmr(UVH_NODE_ID);
99 	m_n_config.v = uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR);
100 	uv_min_hub_revision_id = node_id.s.revision;
101 
102 	switch (node_id.s.part_number) {
103 	case UV2_HUB_PART_NUMBER:
104 	case UV2_HUB_PART_NUMBER_X:
105 		uv_min_hub_revision_id += UV2_HUB_REVISION_BASE - 1;
106 		break;
107 	case UV3_HUB_PART_NUMBER:
108 	case UV3_HUB_PART_NUMBER_X:
109 		uv_min_hub_revision_id += UV3_HUB_REVISION_BASE;
110 		break;
111 	}
112 
113 	uv_hub_info->hub_revision = uv_min_hub_revision_id;
114 	pnode = (node_id.s.node_id >> 1) & ((1 << m_n_config.s.n_skt) - 1);
115 	return pnode;
116 }
117 
118 static void __init early_get_apic_pnode_shift(void)
119 {
120 	uvh_apicid.v = uv_early_read_mmr(UVH_APICID);
121 	if (!uvh_apicid.v)
122 		/*
123 		 * Old bios, use default value
124 		 */
125 		uvh_apicid.s.pnode_shift = UV_APIC_PNODE_SHIFT;
126 }
127 
128 /*
129  * Add an extra bit as dictated by bios to the destination apicid of
130  * interrupts potentially passing through the UV HUB.  This prevents
131  * a deadlock between interrupts and IO port operations.
132  */
133 static void __init uv_set_apicid_hibit(void)
134 {
135 	union uv1h_lb_target_physical_apic_id_mask_u apicid_mask;
136 
137 	if (is_uv1_hub()) {
138 		apicid_mask.v =
139 			uv_early_read_mmr(UV1H_LB_TARGET_PHYSICAL_APIC_ID_MASK);
140 		uv_apicid_hibits =
141 			apicid_mask.s1.bit_enables & UV_APICID_HIBIT_MASK;
142 	}
143 }
144 
145 static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
146 {
147 	int pnodeid;
148 	int uv_apic;
149 
150 	if (strncmp(oem_id, "SGI", 3) != 0)
151 		return 0;
152 
153 	/*
154 	 * Determine UV arch type.
155 	 *   SGI: UV100/1000
156 	 *   SGI2: UV2000/3000
157 	 *   SGI3: UV300 (truncated to 4 chars because of different varieties)
158 	 */
159 	uv_hub_info->hub_revision =
160 		!strncmp(oem_id, "SGI3", 4) ? UV3_HUB_REVISION_BASE :
161 		!strcmp(oem_id, "SGI2") ? UV2_HUB_REVISION_BASE :
162 		!strcmp(oem_id, "SGI") ? UV1_HUB_REVISION_BASE : 0;
163 
164 	if (uv_hub_info->hub_revision == 0)
165 		goto badbios;
166 
167 	pnodeid = early_get_pnodeid();
168 	early_get_apic_pnode_shift();
169 	x86_platform.is_untracked_pat_range =  uv_is_untracked_pat_range;
170 	x86_platform.nmi_init = uv_nmi_init;
171 
172 	if (!strcmp(oem_table_id, "UVX")) {		/* most common */
173 		uv_system_type = UV_X2APIC;
174 		uv_apic = 0;
175 
176 	} else if (!strcmp(oem_table_id, "UVH")) {	/* only UV1 systems */
177 		uv_system_type = UV_NON_UNIQUE_APIC;
178 		__this_cpu_write(x2apic_extra_bits,
179 			pnodeid << uvh_apicid.s.pnode_shift);
180 		uv_set_apicid_hibit();
181 		uv_apic = 1;
182 
183 	} else	if (!strcmp(oem_table_id, "UVL")) {	/* only used for */
184 		uv_system_type = UV_LEGACY_APIC;	/* very small systems */
185 		uv_apic = 0;
186 
187 	} else {
188 		goto badbios;
189 	}
190 
191 	pr_info("UV: OEM IDs %s/%s, System/HUB Types %d/%d, uv_apic %d\n",
192 		oem_id, oem_table_id, uv_system_type,
193 		uv_min_hub_revision_id, uv_apic);
194 
195 	return uv_apic;
196 
197 badbios:
198 	pr_err("UV: OEM_ID:%s OEM_TABLE_ID:%s\n", oem_id, oem_table_id);
199 	pr_err("Current BIOS not supported, update kernel and/or BIOS\n");
200 	BUG();
201 }
202 
203 enum uv_system_type get_uv_system_type(void)
204 {
205 	return uv_system_type;
206 }
207 
208 int is_uv_system(void)
209 {
210 	return uv_system_type != UV_NONE;
211 }
212 EXPORT_SYMBOL_GPL(is_uv_system);
213 
214 DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
215 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
216 
217 struct uv_blade_info *uv_blade_info;
218 EXPORT_SYMBOL_GPL(uv_blade_info);
219 
220 short *uv_node_to_blade;
221 EXPORT_SYMBOL_GPL(uv_node_to_blade);
222 
223 short *uv_cpu_to_blade;
224 EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
225 
226 short uv_possible_blades;
227 EXPORT_SYMBOL_GPL(uv_possible_blades);
228 
229 unsigned long sn_rtc_cycles_per_second;
230 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
231 
232 static int uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
233 {
234 	unsigned long val;
235 	int pnode;
236 
237 	pnode = uv_apicid_to_pnode(phys_apicid);
238 	phys_apicid |= uv_apicid_hibits;
239 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
240 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
241 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
242 	    APIC_DM_INIT;
243 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
244 
245 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
246 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
247 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
248 	    APIC_DM_STARTUP;
249 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
250 
251 	atomic_set(&init_deasserted, 1);
252 	return 0;
253 }
254 
255 static void uv_send_IPI_one(int cpu, int vector)
256 {
257 	unsigned long apicid;
258 	int pnode;
259 
260 	apicid = per_cpu(x86_cpu_to_apicid, cpu);
261 	pnode = uv_apicid_to_pnode(apicid);
262 	uv_hub_send_ipi(pnode, apicid, vector);
263 }
264 
265 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
266 {
267 	unsigned int cpu;
268 
269 	for_each_cpu(cpu, mask)
270 		uv_send_IPI_one(cpu, vector);
271 }
272 
273 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
274 {
275 	unsigned int this_cpu = smp_processor_id();
276 	unsigned int cpu;
277 
278 	for_each_cpu(cpu, mask) {
279 		if (cpu != this_cpu)
280 			uv_send_IPI_one(cpu, vector);
281 	}
282 }
283 
284 static void uv_send_IPI_allbutself(int vector)
285 {
286 	unsigned int this_cpu = smp_processor_id();
287 	unsigned int cpu;
288 
289 	for_each_online_cpu(cpu) {
290 		if (cpu != this_cpu)
291 			uv_send_IPI_one(cpu, vector);
292 	}
293 }
294 
295 static void uv_send_IPI_all(int vector)
296 {
297 	uv_send_IPI_mask(cpu_online_mask, vector);
298 }
299 
300 static int uv_apic_id_valid(int apicid)
301 {
302 	return 1;
303 }
304 
305 static int uv_apic_id_registered(void)
306 {
307 	return 1;
308 }
309 
310 static void uv_init_apic_ldr(void)
311 {
312 }
313 
314 static int
315 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
316 			  const struct cpumask *andmask,
317 			  unsigned int *apicid)
318 {
319 	int unsigned cpu;
320 
321 	/*
322 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
323 	 * May as well be the first.
324 	 */
325 	for_each_cpu_and(cpu, cpumask, andmask) {
326 		if (cpumask_test_cpu(cpu, cpu_online_mask))
327 			break;
328 	}
329 
330 	if (likely(cpu < nr_cpu_ids)) {
331 		*apicid = per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
332 		return 0;
333 	}
334 
335 	return -EINVAL;
336 }
337 
338 static unsigned int x2apic_get_apic_id(unsigned long x)
339 {
340 	unsigned int id;
341 
342 	WARN_ON(preemptible() && num_online_cpus() > 1);
343 	id = x | __this_cpu_read(x2apic_extra_bits);
344 
345 	return id;
346 }
347 
348 static unsigned long set_apic_id(unsigned int id)
349 {
350 	unsigned long x;
351 
352 	/* maskout x2apic_extra_bits ? */
353 	x = id;
354 	return x;
355 }
356 
357 static unsigned int uv_read_apic_id(void)
358 {
359 
360 	return x2apic_get_apic_id(apic_read(APIC_ID));
361 }
362 
363 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
364 {
365 	return uv_read_apic_id() >> index_msb;
366 }
367 
368 static void uv_send_IPI_self(int vector)
369 {
370 	apic_write(APIC_SELF_IPI, vector);
371 }
372 
373 static int uv_probe(void)
374 {
375 	return apic == &apic_x2apic_uv_x;
376 }
377 
378 static struct apic __refdata apic_x2apic_uv_x = {
379 
380 	.name				= "UV large system",
381 	.probe				= uv_probe,
382 	.acpi_madt_oem_check		= uv_acpi_madt_oem_check,
383 	.apic_id_valid			= uv_apic_id_valid,
384 	.apic_id_registered		= uv_apic_id_registered,
385 
386 	.irq_delivery_mode		= dest_Fixed,
387 	.irq_dest_mode			= 0, /* physical */
388 
389 	.target_cpus			= online_target_cpus,
390 	.disable_esr			= 0,
391 	.dest_logical			= APIC_DEST_LOGICAL,
392 	.check_apicid_used		= NULL,
393 
394 	.vector_allocation_domain	= default_vector_allocation_domain,
395 	.init_apic_ldr			= uv_init_apic_ldr,
396 
397 	.ioapic_phys_id_map		= NULL,
398 	.setup_apic_routing		= NULL,
399 	.cpu_present_to_apicid		= default_cpu_present_to_apicid,
400 	.apicid_to_cpu_present		= NULL,
401 	.check_phys_apicid_present	= default_check_phys_apicid_present,
402 	.phys_pkg_id			= uv_phys_pkg_id,
403 
404 	.get_apic_id			= x2apic_get_apic_id,
405 	.set_apic_id			= set_apic_id,
406 	.apic_id_mask			= 0xFFFFFFFFu,
407 
408 	.cpu_mask_to_apicid_and		= uv_cpu_mask_to_apicid_and,
409 
410 	.send_IPI_mask			= uv_send_IPI_mask,
411 	.send_IPI_mask_allbutself	= uv_send_IPI_mask_allbutself,
412 	.send_IPI_allbutself		= uv_send_IPI_allbutself,
413 	.send_IPI_all			= uv_send_IPI_all,
414 	.send_IPI_self			= uv_send_IPI_self,
415 
416 	.wakeup_secondary_cpu		= uv_wakeup_secondary,
417 	.wait_for_init_deassert		= false,
418 	.inquire_remote_apic		= NULL,
419 
420 	.read				= native_apic_msr_read,
421 	.write				= native_apic_msr_write,
422 	.eoi_write			= native_apic_msr_eoi_write,
423 	.icr_read			= native_x2apic_icr_read,
424 	.icr_write			= native_x2apic_icr_write,
425 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
426 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
427 };
428 
429 static void set_x2apic_extra_bits(int pnode)
430 {
431 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
432 }
433 
434 /*
435  * Called on boot cpu.
436  */
437 static __init int boot_pnode_to_blade(int pnode)
438 {
439 	int blade;
440 
441 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
442 		if (pnode == uv_blade_info[blade].pnode)
443 			return blade;
444 	BUG();
445 }
446 
447 struct redir_addr {
448 	unsigned long redirect;
449 	unsigned long alias;
450 };
451 
452 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
453 
454 static __initdata struct redir_addr redir_addrs[] = {
455 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
456 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
457 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
458 };
459 
460 static unsigned char get_n_lshift(int m_val)
461 {
462 	union uv3h_gr0_gam_gr_config_u m_gr_config;
463 
464 	if (is_uv1_hub())
465 		return m_val;
466 
467 	if (is_uv2_hub())
468 		return m_val == 40 ? 40 : 39;
469 
470 	m_gr_config.v = uv_read_local_mmr(UV3H_GR0_GAM_GR_CONFIG);
471 	return m_gr_config.s3.m_skt;
472 }
473 
474 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
475 {
476 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
477 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
478 	int i;
479 
480 	for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
481 		alias.v = uv_read_local_mmr(redir_addrs[i].alias);
482 		if (alias.s.enable && alias.s.base == 0) {
483 			*size = (1UL << alias.s.m_alias);
484 			redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
485 			*base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
486 			return;
487 		}
488 	}
489 	*base = *size = 0;
490 }
491 
492 enum map_type {map_wb, map_uc};
493 
494 static __init void map_high(char *id, unsigned long base, int pshift,
495 			int bshift, int max_pnode, enum map_type map_type)
496 {
497 	unsigned long bytes, paddr;
498 
499 	paddr = base << pshift;
500 	bytes = (1UL << bshift) * (max_pnode + 1);
501 	if (!paddr) {
502 		pr_info("UV: Map %s_HI base address NULL\n", id);
503 		return;
504 	}
505 	pr_debug("UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr, paddr + bytes);
506 	if (map_type == map_uc)
507 		init_extra_mapping_uc(paddr, bytes);
508 	else
509 		init_extra_mapping_wb(paddr, bytes);
510 }
511 
512 static __init void map_gru_distributed(unsigned long c)
513 {
514 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
515 	u64 paddr;
516 	unsigned long bytes;
517 	int nid;
518 
519 	gru.v = c;
520 	/* only base bits 42:28 relevant in dist mode */
521 	gru_dist_base = gru.v & 0x000007fff0000000UL;
522 	if (!gru_dist_base) {
523 		pr_info("UV: Map GRU_DIST base address NULL\n");
524 		return;
525 	}
526 	bytes = 1UL << UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
527 	gru_dist_lmask = ((1UL << uv_hub_info->m_val) - 1) & ~(bytes - 1);
528 	gru_dist_umask = ~((1UL << uv_hub_info->m_val) - 1);
529 	gru_dist_base &= gru_dist_lmask; /* Clear bits above M */
530 	for_each_online_node(nid) {
531 		paddr = ((u64)uv_node_to_pnode(nid) << uv_hub_info->m_val) |
532 				gru_dist_base;
533 		init_extra_mapping_wb(paddr, bytes);
534 		gru_first_node_paddr = min(paddr, gru_first_node_paddr);
535 		gru_last_node_paddr = max(paddr, gru_last_node_paddr);
536 	}
537 	/* Save upper (63:M) bits of address only for is_GRU_range */
538 	gru_first_node_paddr &= gru_dist_umask;
539 	gru_last_node_paddr &= gru_dist_umask;
540 	pr_debug("UV: Map GRU_DIST base 0x%016llx  0x%016llx - 0x%016llx\n",
541 		gru_dist_base, gru_first_node_paddr, gru_last_node_paddr);
542 }
543 
544 static __init void map_gru_high(int max_pnode)
545 {
546 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
547 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
548 
549 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
550 	if (!gru.s.enable) {
551 		pr_info("UV: GRU disabled\n");
552 		return;
553 	}
554 
555 	if (is_uv3_hub() && gru.s3.mode) {
556 		map_gru_distributed(gru.v);
557 		return;
558 	}
559 	map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
560 	gru_start_paddr = ((u64)gru.s.base << shift);
561 	gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
562 }
563 
564 static __init void map_mmr_high(int max_pnode)
565 {
566 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
567 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
568 
569 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
570 	if (mmr.s.enable)
571 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
572 	else
573 		pr_info("UV: MMR disabled\n");
574 }
575 
576 /*
577  * This commonality works because both 0 & 1 versions of the MMIOH OVERLAY
578  * and REDIRECT MMR regs are exactly the same on UV3.
579  */
580 struct mmioh_config {
581 	unsigned long overlay;
582 	unsigned long redirect;
583 	char *id;
584 };
585 
586 static __initdata struct mmioh_config mmiohs[] = {
587 	{
588 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR,
589 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR,
590 		"MMIOH0"
591 	},
592 	{
593 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG1_MMR,
594 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG1_MMR,
595 		"MMIOH1"
596 	},
597 };
598 
599 static __init void map_mmioh_high_uv3(int index, int min_pnode, int max_pnode)
600 {
601 	union uv3h_rh_gam_mmioh_overlay_config0_mmr_u overlay;
602 	unsigned long mmr;
603 	unsigned long base;
604 	int i, n, shift, m_io, max_io;
605 	int nasid, lnasid, fi, li;
606 	char *id;
607 
608 	id = mmiohs[index].id;
609 	overlay.v = uv_read_local_mmr(mmiohs[index].overlay);
610 	pr_info("UV: %s overlay 0x%lx base:0x%x m_io:%d\n",
611 		id, overlay.v, overlay.s3.base, overlay.s3.m_io);
612 	if (!overlay.s3.enable) {
613 		pr_info("UV: %s disabled\n", id);
614 		return;
615 	}
616 
617 	shift = UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR_BASE_SHFT;
618 	base = (unsigned long)overlay.s3.base;
619 	m_io = overlay.s3.m_io;
620 	mmr = mmiohs[index].redirect;
621 	n = UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR_DEPTH;
622 	min_pnode *= 2;				/* convert to NASID */
623 	max_pnode *= 2;
624 	max_io = lnasid = fi = li = -1;
625 
626 	for (i = 0; i < n; i++) {
627 		union uv3h_rh_gam_mmioh_redirect_config0_mmr_u redirect;
628 
629 		redirect.v = uv_read_local_mmr(mmr + i * 8);
630 		nasid = redirect.s3.nasid;
631 		if (nasid < min_pnode || max_pnode < nasid)
632 			nasid = -1;		/* invalid NASID */
633 
634 		if (nasid == lnasid) {
635 			li = i;
636 			if (i != n-1)		/* last entry check */
637 				continue;
638 		}
639 
640 		/* check if we have a cached (or last) redirect to print */
641 		if (lnasid != -1 || (i == n-1 && nasid != -1))  {
642 			unsigned long addr1, addr2;
643 			int f, l;
644 
645 			if (lnasid == -1) {
646 				f = l = i;
647 				lnasid = nasid;
648 			} else {
649 				f = fi;
650 				l = li;
651 			}
652 			addr1 = (base << shift) +
653 				f * (unsigned long)(1 << m_io);
654 			addr2 = (base << shift) +
655 				(l + 1) * (unsigned long)(1 << m_io);
656 			pr_info("UV: %s[%03d..%03d] NASID 0x%04x ADDR 0x%016lx - 0x%016lx\n",
657 				id, fi, li, lnasid, addr1, addr2);
658 			if (max_io < l)
659 				max_io = l;
660 		}
661 		fi = li = i;
662 		lnasid = nasid;
663 	}
664 
665 	pr_info("UV: %s base:0x%lx shift:%d M_IO:%d MAX_IO:%d\n",
666 		id, base, shift, m_io, max_io);
667 
668 	if (max_io >= 0)
669 		map_high(id, base, shift, m_io, max_io, map_uc);
670 }
671 
672 static __init void map_mmioh_high(int min_pnode, int max_pnode)
673 {
674 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
675 	unsigned long mmr, base;
676 	int shift, enable, m_io, n_io;
677 
678 	if (is_uv3_hub()) {
679 		/* Map both MMIOH Regions */
680 		map_mmioh_high_uv3(0, min_pnode, max_pnode);
681 		map_mmioh_high_uv3(1, min_pnode, max_pnode);
682 		return;
683 	}
684 
685 	if (is_uv1_hub()) {
686 		mmr = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
687 		shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
688 		mmioh.v = uv_read_local_mmr(mmr);
689 		enable = !!mmioh.s1.enable;
690 		base = mmioh.s1.base;
691 		m_io = mmioh.s1.m_io;
692 		n_io = mmioh.s1.n_io;
693 	} else if (is_uv2_hub()) {
694 		mmr = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
695 		shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
696 		mmioh.v = uv_read_local_mmr(mmr);
697 		enable = !!mmioh.s2.enable;
698 		base = mmioh.s2.base;
699 		m_io = mmioh.s2.m_io;
700 		n_io = mmioh.s2.n_io;
701 	} else
702 		return;
703 
704 	if (enable) {
705 		max_pnode &= (1 << n_io) - 1;
706 		pr_info(
707 		    "UV: base:0x%lx shift:%d N_IO:%d M_IO:%d max_pnode:0x%x\n",
708 			base, shift, m_io, n_io, max_pnode);
709 		map_high("MMIOH", base, shift, m_io, max_pnode, map_uc);
710 	} else {
711 		pr_info("UV: MMIOH disabled\n");
712 	}
713 }
714 
715 static __init void map_low_mmrs(void)
716 {
717 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
718 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
719 }
720 
721 static __init void uv_rtc_init(void)
722 {
723 	long status;
724 	u64 ticks_per_sec;
725 
726 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
727 					&ticks_per_sec);
728 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
729 		printk(KERN_WARNING
730 			"unable to determine platform RTC clock frequency, "
731 			"guessing.\n");
732 		/* BIOS gives wrong value for clock freq. so guess */
733 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
734 	} else
735 		sn_rtc_cycles_per_second = ticks_per_sec;
736 }
737 
738 /*
739  * percpu heartbeat timer
740  */
741 static void uv_heartbeat(unsigned long ignored)
742 {
743 	struct timer_list *timer = &uv_hub_info->scir.timer;
744 	unsigned char bits = uv_hub_info->scir.state;
745 
746 	/* flip heartbeat bit */
747 	bits ^= SCIR_CPU_HEARTBEAT;
748 
749 	/* is this cpu idle? */
750 	if (idle_cpu(raw_smp_processor_id()))
751 		bits &= ~SCIR_CPU_ACTIVITY;
752 	else
753 		bits |= SCIR_CPU_ACTIVITY;
754 
755 	/* update system controller interface reg */
756 	uv_set_scir_bits(bits);
757 
758 	/* enable next timer period */
759 	mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
760 }
761 
762 static void uv_heartbeat_enable(int cpu)
763 {
764 	while (!uv_cpu_hub_info(cpu)->scir.enabled) {
765 		struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
766 
767 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
768 		setup_timer(timer, uv_heartbeat, cpu);
769 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
770 		add_timer_on(timer, cpu);
771 		uv_cpu_hub_info(cpu)->scir.enabled = 1;
772 
773 		/* also ensure that boot cpu is enabled */
774 		cpu = 0;
775 	}
776 }
777 
778 #ifdef CONFIG_HOTPLUG_CPU
779 static void uv_heartbeat_disable(int cpu)
780 {
781 	if (uv_cpu_hub_info(cpu)->scir.enabled) {
782 		uv_cpu_hub_info(cpu)->scir.enabled = 0;
783 		del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
784 	}
785 	uv_set_cpu_scir_bits(cpu, 0xff);
786 }
787 
788 /*
789  * cpu hotplug notifier
790  */
791 static int uv_scir_cpu_notify(struct notifier_block *self, unsigned long action,
792 			      void *hcpu)
793 {
794 	long cpu = (long)hcpu;
795 
796 	switch (action) {
797 	case CPU_ONLINE:
798 		uv_heartbeat_enable(cpu);
799 		break;
800 	case CPU_DOWN_PREPARE:
801 		uv_heartbeat_disable(cpu);
802 		break;
803 	default:
804 		break;
805 	}
806 	return NOTIFY_OK;
807 }
808 
809 static __init void uv_scir_register_cpu_notifier(void)
810 {
811 	hotcpu_notifier(uv_scir_cpu_notify, 0);
812 }
813 
814 #else /* !CONFIG_HOTPLUG_CPU */
815 
816 static __init void uv_scir_register_cpu_notifier(void)
817 {
818 }
819 
820 static __init int uv_init_heartbeat(void)
821 {
822 	int cpu;
823 
824 	if (is_uv_system())
825 		for_each_online_cpu(cpu)
826 			uv_heartbeat_enable(cpu);
827 	return 0;
828 }
829 
830 late_initcall(uv_init_heartbeat);
831 
832 #endif /* !CONFIG_HOTPLUG_CPU */
833 
834 /* Direct Legacy VGA I/O traffic to designated IOH */
835 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
836 		      unsigned int command_bits, u32 flags)
837 {
838 	int domain, bus, rc;
839 
840 	PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
841 			pdev->devfn, decode, command_bits, flags);
842 
843 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
844 		return 0;
845 
846 	if ((command_bits & PCI_COMMAND_IO) == 0)
847 		return 0;
848 
849 	domain = pci_domain_nr(pdev->bus);
850 	bus = pdev->bus->number;
851 
852 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
853 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
854 
855 	return rc;
856 }
857 
858 /*
859  * Called on each cpu to initialize the per_cpu UV data area.
860  * FIXME: hotplug not supported yet
861  */
862 void uv_cpu_init(void)
863 {
864 	/* CPU 0 initilization will be done via uv_system_init. */
865 	if (!uv_blade_info)
866 		return;
867 
868 	uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
869 
870 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
871 		set_x2apic_extra_bits(uv_hub_info->pnode);
872 }
873 
874 void __init uv_system_init(void)
875 {
876 	union uvh_rh_gam_config_mmr_u  m_n_config;
877 	union uvh_node_id_u node_id;
878 	unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
879 	int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val;
880 	int gnode_extra, min_pnode = 999999, max_pnode = -1;
881 	unsigned long mmr_base, present, paddr;
882 	unsigned short pnode_mask;
883 	unsigned char n_lshift;
884 	char *hub = (is_uv1_hub() ? "UV100/1000" :
885 		    (is_uv2_hub() ? "UV2000/3000" :
886 		    (is_uv3_hub() ? "UV300" : NULL)));
887 
888 	if (!hub) {
889 		pr_err("UV: Unknown/unsupported UV hub\n");
890 		return;
891 	}
892 	pr_info("UV: Found %s hub\n", hub);
893 	map_low_mmrs();
894 
895 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
896 	m_val = m_n_config.s.m_skt;
897 	n_val = m_n_config.s.n_skt;
898 	pnode_mask = (1 << n_val) - 1;
899 	n_lshift = get_n_lshift(m_val);
900 	mmr_base =
901 	    uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
902 	    ~UV_MMR_ENABLE;
903 
904 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
905 	gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
906 	gnode_upper = ((unsigned long)gnode_extra  << m_val);
907 	pr_info("UV: N:%d M:%d pnode_mask:0x%x gnode_upper/extra:0x%lx/0x%x n_lshift 0x%x\n",
908 			n_val, m_val, pnode_mask, gnode_upper, gnode_extra,
909 			n_lshift);
910 
911 	pr_info("UV: global MMR base 0x%lx\n", mmr_base);
912 
913 	for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
914 		uv_possible_blades +=
915 		  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
916 
917 	/* uv_num_possible_blades() is really the hub count */
918 	pr_info("UV: Found %d blades, %d hubs\n",
919 			is_uv1_hub() ? uv_num_possible_blades() :
920 			(uv_num_possible_blades() + 1) / 2,
921 			uv_num_possible_blades());
922 
923 	bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
924 	uv_blade_info = kzalloc(bytes, GFP_KERNEL);
925 	BUG_ON(!uv_blade_info);
926 
927 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
928 		uv_blade_info[blade].memory_nid = -1;
929 
930 	get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
931 
932 	bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
933 	uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
934 	BUG_ON(!uv_node_to_blade);
935 	memset(uv_node_to_blade, 255, bytes);
936 
937 	bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
938 	uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
939 	BUG_ON(!uv_cpu_to_blade);
940 	memset(uv_cpu_to_blade, 255, bytes);
941 
942 	blade = 0;
943 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
944 		present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
945 		for (j = 0; j < 64; j++) {
946 			if (!test_bit(j, &present))
947 				continue;
948 			pnode = (i * 64 + j) & pnode_mask;
949 			uv_blade_info[blade].pnode = pnode;
950 			uv_blade_info[blade].nr_possible_cpus = 0;
951 			uv_blade_info[blade].nr_online_cpus = 0;
952 			spin_lock_init(&uv_blade_info[blade].nmi_lock);
953 			min_pnode = min(pnode, min_pnode);
954 			max_pnode = max(pnode, max_pnode);
955 			blade++;
956 		}
957 	}
958 
959 	uv_bios_init();
960 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
961 			    &sn_region_size, &system_serial_number);
962 	uv_rtc_init();
963 
964 	for_each_present_cpu(cpu) {
965 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
966 
967 		nid = cpu_to_node(cpu);
968 		/*
969 		 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
970 		 */
971 		uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
972 		uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
973 		uv_cpu_hub_info(cpu)->hub_revision = uv_hub_info->hub_revision;
974 
975 		uv_cpu_hub_info(cpu)->m_shift = 64 - m_val;
976 		uv_cpu_hub_info(cpu)->n_lshift = n_lshift;
977 
978 		pnode = uv_apicid_to_pnode(apicid);
979 		blade = boot_pnode_to_blade(pnode);
980 		lcpu = uv_blade_info[blade].nr_possible_cpus;
981 		uv_blade_info[blade].nr_possible_cpus++;
982 
983 		/* Any node on the blade, else will contain -1. */
984 		uv_blade_info[blade].memory_nid = nid;
985 
986 		uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
987 		uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
988 		uv_cpu_hub_info(cpu)->m_val = m_val;
989 		uv_cpu_hub_info(cpu)->n_val = n_val;
990 		uv_cpu_hub_info(cpu)->numa_blade_id = blade;
991 		uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
992 		uv_cpu_hub_info(cpu)->pnode = pnode;
993 		uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
994 		uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
995 		uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
996 		uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
997 		uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
998 		uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
999 		uv_node_to_blade[nid] = blade;
1000 		uv_cpu_to_blade[cpu] = blade;
1001 	}
1002 
1003 	/* Add blade/pnode info for nodes without cpus */
1004 	for_each_online_node(nid) {
1005 		if (uv_node_to_blade[nid] >= 0)
1006 			continue;
1007 		paddr = node_start_pfn(nid) << PAGE_SHIFT;
1008 		pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
1009 		blade = boot_pnode_to_blade(pnode);
1010 		uv_node_to_blade[nid] = blade;
1011 	}
1012 
1013 	map_gru_high(max_pnode);
1014 	map_mmr_high(max_pnode);
1015 	map_mmioh_high(min_pnode, max_pnode);
1016 
1017 	uv_nmi_setup();
1018 	uv_cpu_init();
1019 	uv_scir_register_cpu_notifier();
1020 	proc_mkdir("sgi_uv", NULL);
1021 
1022 	/* register Legacy VGA I/O redirection handler */
1023 	pci_register_set_vga_state(uv_set_vga_state);
1024 
1025 	/*
1026 	 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
1027 	 * EFI is not enabled in the kdump kernel.
1028 	 */
1029 	if (is_kdump_kernel())
1030 		reboot_type = BOOT_ACPI;
1031 }
1032 
1033 apic_driver(apic_x2apic_uv_x);
1034