xref: /linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision ca55b2fef3a9373fcfc30f82fd26bc7fccbda732)
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 	return 0;
252 }
253 
254 static void uv_send_IPI_one(int cpu, int vector)
255 {
256 	unsigned long apicid;
257 	int pnode;
258 
259 	apicid = per_cpu(x86_cpu_to_apicid, cpu);
260 	pnode = uv_apicid_to_pnode(apicid);
261 	uv_hub_send_ipi(pnode, apicid, vector);
262 }
263 
264 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
265 {
266 	unsigned int cpu;
267 
268 	for_each_cpu(cpu, mask)
269 		uv_send_IPI_one(cpu, vector);
270 }
271 
272 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
273 {
274 	unsigned int this_cpu = smp_processor_id();
275 	unsigned int cpu;
276 
277 	for_each_cpu(cpu, mask) {
278 		if (cpu != this_cpu)
279 			uv_send_IPI_one(cpu, vector);
280 	}
281 }
282 
283 static void uv_send_IPI_allbutself(int vector)
284 {
285 	unsigned int this_cpu = smp_processor_id();
286 	unsigned int cpu;
287 
288 	for_each_online_cpu(cpu) {
289 		if (cpu != this_cpu)
290 			uv_send_IPI_one(cpu, vector);
291 	}
292 }
293 
294 static void uv_send_IPI_all(int vector)
295 {
296 	uv_send_IPI_mask(cpu_online_mask, vector);
297 }
298 
299 static int uv_apic_id_valid(int apicid)
300 {
301 	return 1;
302 }
303 
304 static int uv_apic_id_registered(void)
305 {
306 	return 1;
307 }
308 
309 static void uv_init_apic_ldr(void)
310 {
311 }
312 
313 static int
314 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
315 			  const struct cpumask *andmask,
316 			  unsigned int *apicid)
317 {
318 	int unsigned cpu;
319 
320 	/*
321 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
322 	 * May as well be the first.
323 	 */
324 	for_each_cpu_and(cpu, cpumask, andmask) {
325 		if (cpumask_test_cpu(cpu, cpu_online_mask))
326 			break;
327 	}
328 
329 	if (likely(cpu < nr_cpu_ids)) {
330 		*apicid = per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
331 		return 0;
332 	}
333 
334 	return -EINVAL;
335 }
336 
337 static unsigned int x2apic_get_apic_id(unsigned long x)
338 {
339 	unsigned int id;
340 
341 	WARN_ON(preemptible() && num_online_cpus() > 1);
342 	id = x | __this_cpu_read(x2apic_extra_bits);
343 
344 	return id;
345 }
346 
347 static unsigned long set_apic_id(unsigned int id)
348 {
349 	unsigned long x;
350 
351 	/* maskout x2apic_extra_bits ? */
352 	x = id;
353 	return x;
354 }
355 
356 static unsigned int uv_read_apic_id(void)
357 {
358 
359 	return x2apic_get_apic_id(apic_read(APIC_ID));
360 }
361 
362 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
363 {
364 	return uv_read_apic_id() >> index_msb;
365 }
366 
367 static void uv_send_IPI_self(int vector)
368 {
369 	apic_write(APIC_SELF_IPI, vector);
370 }
371 
372 static int uv_probe(void)
373 {
374 	return apic == &apic_x2apic_uv_x;
375 }
376 
377 static struct apic __refdata apic_x2apic_uv_x = {
378 
379 	.name				= "UV large system",
380 	.probe				= uv_probe,
381 	.acpi_madt_oem_check		= uv_acpi_madt_oem_check,
382 	.apic_id_valid			= uv_apic_id_valid,
383 	.apic_id_registered		= uv_apic_id_registered,
384 
385 	.irq_delivery_mode		= dest_Fixed,
386 	.irq_dest_mode			= 0, /* physical */
387 
388 	.target_cpus			= online_target_cpus,
389 	.disable_esr			= 0,
390 	.dest_logical			= APIC_DEST_LOGICAL,
391 	.check_apicid_used		= NULL,
392 
393 	.vector_allocation_domain	= default_vector_allocation_domain,
394 	.init_apic_ldr			= uv_init_apic_ldr,
395 
396 	.ioapic_phys_id_map		= NULL,
397 	.setup_apic_routing		= NULL,
398 	.cpu_present_to_apicid		= default_cpu_present_to_apicid,
399 	.apicid_to_cpu_present		= NULL,
400 	.check_phys_apicid_present	= default_check_phys_apicid_present,
401 	.phys_pkg_id			= uv_phys_pkg_id,
402 
403 	.get_apic_id			= x2apic_get_apic_id,
404 	.set_apic_id			= set_apic_id,
405 	.apic_id_mask			= 0xFFFFFFFFu,
406 
407 	.cpu_mask_to_apicid_and		= uv_cpu_mask_to_apicid_and,
408 
409 	.send_IPI_mask			= uv_send_IPI_mask,
410 	.send_IPI_mask_allbutself	= uv_send_IPI_mask_allbutself,
411 	.send_IPI_allbutself		= uv_send_IPI_allbutself,
412 	.send_IPI_all			= uv_send_IPI_all,
413 	.send_IPI_self			= uv_send_IPI_self,
414 
415 	.wakeup_secondary_cpu		= uv_wakeup_secondary,
416 	.inquire_remote_apic		= NULL,
417 
418 	.read				= native_apic_msr_read,
419 	.write				= native_apic_msr_write,
420 	.eoi_write			= native_apic_msr_eoi_write,
421 	.icr_read			= native_x2apic_icr_read,
422 	.icr_write			= native_x2apic_icr_write,
423 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
424 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
425 };
426 
427 static void set_x2apic_extra_bits(int pnode)
428 {
429 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
430 }
431 
432 /*
433  * Called on boot cpu.
434  */
435 static __init int boot_pnode_to_blade(int pnode)
436 {
437 	int blade;
438 
439 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
440 		if (pnode == uv_blade_info[blade].pnode)
441 			return blade;
442 	BUG();
443 }
444 
445 struct redir_addr {
446 	unsigned long redirect;
447 	unsigned long alias;
448 };
449 
450 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
451 
452 static __initdata struct redir_addr redir_addrs[] = {
453 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
454 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
455 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
456 };
457 
458 static unsigned char get_n_lshift(int m_val)
459 {
460 	union uv3h_gr0_gam_gr_config_u m_gr_config;
461 
462 	if (is_uv1_hub())
463 		return m_val;
464 
465 	if (is_uv2_hub())
466 		return m_val == 40 ? 40 : 39;
467 
468 	m_gr_config.v = uv_read_local_mmr(UV3H_GR0_GAM_GR_CONFIG);
469 	return m_gr_config.s3.m_skt;
470 }
471 
472 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
473 {
474 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
475 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
476 	int i;
477 
478 	for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
479 		alias.v = uv_read_local_mmr(redir_addrs[i].alias);
480 		if (alias.s.enable && alias.s.base == 0) {
481 			*size = (1UL << alias.s.m_alias);
482 			redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
483 			*base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
484 			return;
485 		}
486 	}
487 	*base = *size = 0;
488 }
489 
490 enum map_type {map_wb, map_uc};
491 
492 static __init void map_high(char *id, unsigned long base, int pshift,
493 			int bshift, int max_pnode, enum map_type map_type)
494 {
495 	unsigned long bytes, paddr;
496 
497 	paddr = base << pshift;
498 	bytes = (1UL << bshift) * (max_pnode + 1);
499 	if (!paddr) {
500 		pr_info("UV: Map %s_HI base address NULL\n", id);
501 		return;
502 	}
503 	pr_debug("UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr, paddr + bytes);
504 	if (map_type == map_uc)
505 		init_extra_mapping_uc(paddr, bytes);
506 	else
507 		init_extra_mapping_wb(paddr, bytes);
508 }
509 
510 static __init void map_gru_distributed(unsigned long c)
511 {
512 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
513 	u64 paddr;
514 	unsigned long bytes;
515 	int nid;
516 
517 	gru.v = c;
518 	/* only base bits 42:28 relevant in dist mode */
519 	gru_dist_base = gru.v & 0x000007fff0000000UL;
520 	if (!gru_dist_base) {
521 		pr_info("UV: Map GRU_DIST base address NULL\n");
522 		return;
523 	}
524 	bytes = 1UL << UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
525 	gru_dist_lmask = ((1UL << uv_hub_info->m_val) - 1) & ~(bytes - 1);
526 	gru_dist_umask = ~((1UL << uv_hub_info->m_val) - 1);
527 	gru_dist_base &= gru_dist_lmask; /* Clear bits above M */
528 	for_each_online_node(nid) {
529 		paddr = ((u64)uv_node_to_pnode(nid) << uv_hub_info->m_val) |
530 				gru_dist_base;
531 		init_extra_mapping_wb(paddr, bytes);
532 		gru_first_node_paddr = min(paddr, gru_first_node_paddr);
533 		gru_last_node_paddr = max(paddr, gru_last_node_paddr);
534 	}
535 	/* Save upper (63:M) bits of address only for is_GRU_range */
536 	gru_first_node_paddr &= gru_dist_umask;
537 	gru_last_node_paddr &= gru_dist_umask;
538 	pr_debug("UV: Map GRU_DIST base 0x%016llx  0x%016llx - 0x%016llx\n",
539 		gru_dist_base, gru_first_node_paddr, gru_last_node_paddr);
540 }
541 
542 static __init void map_gru_high(int max_pnode)
543 {
544 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
545 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
546 
547 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
548 	if (!gru.s.enable) {
549 		pr_info("UV: GRU disabled\n");
550 		return;
551 	}
552 
553 	if (is_uv3_hub() && gru.s3.mode) {
554 		map_gru_distributed(gru.v);
555 		return;
556 	}
557 	map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
558 	gru_start_paddr = ((u64)gru.s.base << shift);
559 	gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
560 }
561 
562 static __init void map_mmr_high(int max_pnode)
563 {
564 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
565 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
566 
567 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
568 	if (mmr.s.enable)
569 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
570 	else
571 		pr_info("UV: MMR disabled\n");
572 }
573 
574 /*
575  * This commonality works because both 0 & 1 versions of the MMIOH OVERLAY
576  * and REDIRECT MMR regs are exactly the same on UV3.
577  */
578 struct mmioh_config {
579 	unsigned long overlay;
580 	unsigned long redirect;
581 	char *id;
582 };
583 
584 static __initdata struct mmioh_config mmiohs[] = {
585 	{
586 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR,
587 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR,
588 		"MMIOH0"
589 	},
590 	{
591 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG1_MMR,
592 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG1_MMR,
593 		"MMIOH1"
594 	},
595 };
596 
597 static __init void map_mmioh_high_uv3(int index, int min_pnode, int max_pnode)
598 {
599 	union uv3h_rh_gam_mmioh_overlay_config0_mmr_u overlay;
600 	unsigned long mmr;
601 	unsigned long base;
602 	int i, n, shift, m_io, max_io;
603 	int nasid, lnasid, fi, li;
604 	char *id;
605 
606 	id = mmiohs[index].id;
607 	overlay.v = uv_read_local_mmr(mmiohs[index].overlay);
608 	pr_info("UV: %s overlay 0x%lx base:0x%x m_io:%d\n",
609 		id, overlay.v, overlay.s3.base, overlay.s3.m_io);
610 	if (!overlay.s3.enable) {
611 		pr_info("UV: %s disabled\n", id);
612 		return;
613 	}
614 
615 	shift = UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR_BASE_SHFT;
616 	base = (unsigned long)overlay.s3.base;
617 	m_io = overlay.s3.m_io;
618 	mmr = mmiohs[index].redirect;
619 	n = UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR_DEPTH;
620 	min_pnode *= 2;				/* convert to NASID */
621 	max_pnode *= 2;
622 	max_io = lnasid = fi = li = -1;
623 
624 	for (i = 0; i < n; i++) {
625 		union uv3h_rh_gam_mmioh_redirect_config0_mmr_u redirect;
626 
627 		redirect.v = uv_read_local_mmr(mmr + i * 8);
628 		nasid = redirect.s3.nasid;
629 		if (nasid < min_pnode || max_pnode < nasid)
630 			nasid = -1;		/* invalid NASID */
631 
632 		if (nasid == lnasid) {
633 			li = i;
634 			if (i != n-1)		/* last entry check */
635 				continue;
636 		}
637 
638 		/* check if we have a cached (or last) redirect to print */
639 		if (lnasid != -1 || (i == n-1 && nasid != -1))  {
640 			unsigned long addr1, addr2;
641 			int f, l;
642 
643 			if (lnasid == -1) {
644 				f = l = i;
645 				lnasid = nasid;
646 			} else {
647 				f = fi;
648 				l = li;
649 			}
650 			addr1 = (base << shift) +
651 				f * (unsigned long)(1 << m_io);
652 			addr2 = (base << shift) +
653 				(l + 1) * (unsigned long)(1 << m_io);
654 			pr_info("UV: %s[%03d..%03d] NASID 0x%04x ADDR 0x%016lx - 0x%016lx\n",
655 				id, fi, li, lnasid, addr1, addr2);
656 			if (max_io < l)
657 				max_io = l;
658 		}
659 		fi = li = i;
660 		lnasid = nasid;
661 	}
662 
663 	pr_info("UV: %s base:0x%lx shift:%d M_IO:%d MAX_IO:%d\n",
664 		id, base, shift, m_io, max_io);
665 
666 	if (max_io >= 0)
667 		map_high(id, base, shift, m_io, max_io, map_uc);
668 }
669 
670 static __init void map_mmioh_high(int min_pnode, int max_pnode)
671 {
672 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
673 	unsigned long mmr, base;
674 	int shift, enable, m_io, n_io;
675 
676 	if (is_uv3_hub()) {
677 		/* Map both MMIOH Regions */
678 		map_mmioh_high_uv3(0, min_pnode, max_pnode);
679 		map_mmioh_high_uv3(1, min_pnode, max_pnode);
680 		return;
681 	}
682 
683 	if (is_uv1_hub()) {
684 		mmr = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
685 		shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
686 		mmioh.v = uv_read_local_mmr(mmr);
687 		enable = !!mmioh.s1.enable;
688 		base = mmioh.s1.base;
689 		m_io = mmioh.s1.m_io;
690 		n_io = mmioh.s1.n_io;
691 	} else if (is_uv2_hub()) {
692 		mmr = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
693 		shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
694 		mmioh.v = uv_read_local_mmr(mmr);
695 		enable = !!mmioh.s2.enable;
696 		base = mmioh.s2.base;
697 		m_io = mmioh.s2.m_io;
698 		n_io = mmioh.s2.n_io;
699 	} else
700 		return;
701 
702 	if (enable) {
703 		max_pnode &= (1 << n_io) - 1;
704 		pr_info(
705 		    "UV: base:0x%lx shift:%d N_IO:%d M_IO:%d max_pnode:0x%x\n",
706 			base, shift, m_io, n_io, max_pnode);
707 		map_high("MMIOH", base, shift, m_io, max_pnode, map_uc);
708 	} else {
709 		pr_info("UV: MMIOH disabled\n");
710 	}
711 }
712 
713 static __init void map_low_mmrs(void)
714 {
715 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
716 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
717 }
718 
719 static __init void uv_rtc_init(void)
720 {
721 	long status;
722 	u64 ticks_per_sec;
723 
724 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
725 					&ticks_per_sec);
726 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
727 		printk(KERN_WARNING
728 			"unable to determine platform RTC clock frequency, "
729 			"guessing.\n");
730 		/* BIOS gives wrong value for clock freq. so guess */
731 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
732 	} else
733 		sn_rtc_cycles_per_second = ticks_per_sec;
734 }
735 
736 /*
737  * percpu heartbeat timer
738  */
739 static void uv_heartbeat(unsigned long ignored)
740 {
741 	struct timer_list *timer = &uv_hub_info->scir.timer;
742 	unsigned char bits = uv_hub_info->scir.state;
743 
744 	/* flip heartbeat bit */
745 	bits ^= SCIR_CPU_HEARTBEAT;
746 
747 	/* is this cpu idle? */
748 	if (idle_cpu(raw_smp_processor_id()))
749 		bits &= ~SCIR_CPU_ACTIVITY;
750 	else
751 		bits |= SCIR_CPU_ACTIVITY;
752 
753 	/* update system controller interface reg */
754 	uv_set_scir_bits(bits);
755 
756 	/* enable next timer period */
757 	mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
758 }
759 
760 static void uv_heartbeat_enable(int cpu)
761 {
762 	while (!uv_cpu_hub_info(cpu)->scir.enabled) {
763 		struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
764 
765 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
766 		setup_timer(timer, uv_heartbeat, cpu);
767 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
768 		add_timer_on(timer, cpu);
769 		uv_cpu_hub_info(cpu)->scir.enabled = 1;
770 
771 		/* also ensure that boot cpu is enabled */
772 		cpu = 0;
773 	}
774 }
775 
776 #ifdef CONFIG_HOTPLUG_CPU
777 static void uv_heartbeat_disable(int cpu)
778 {
779 	if (uv_cpu_hub_info(cpu)->scir.enabled) {
780 		uv_cpu_hub_info(cpu)->scir.enabled = 0;
781 		del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
782 	}
783 	uv_set_cpu_scir_bits(cpu, 0xff);
784 }
785 
786 /*
787  * cpu hotplug notifier
788  */
789 static int uv_scir_cpu_notify(struct notifier_block *self, unsigned long action,
790 			      void *hcpu)
791 {
792 	long cpu = (long)hcpu;
793 
794 	switch (action) {
795 	case CPU_ONLINE:
796 		uv_heartbeat_enable(cpu);
797 		break;
798 	case CPU_DOWN_PREPARE:
799 		uv_heartbeat_disable(cpu);
800 		break;
801 	default:
802 		break;
803 	}
804 	return NOTIFY_OK;
805 }
806 
807 static __init void uv_scir_register_cpu_notifier(void)
808 {
809 	hotcpu_notifier(uv_scir_cpu_notify, 0);
810 }
811 
812 #else /* !CONFIG_HOTPLUG_CPU */
813 
814 static __init void uv_scir_register_cpu_notifier(void)
815 {
816 }
817 
818 static __init int uv_init_heartbeat(void)
819 {
820 	int cpu;
821 
822 	if (is_uv_system())
823 		for_each_online_cpu(cpu)
824 			uv_heartbeat_enable(cpu);
825 	return 0;
826 }
827 
828 late_initcall(uv_init_heartbeat);
829 
830 #endif /* !CONFIG_HOTPLUG_CPU */
831 
832 /* Direct Legacy VGA I/O traffic to designated IOH */
833 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
834 		      unsigned int command_bits, u32 flags)
835 {
836 	int domain, bus, rc;
837 
838 	PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
839 			pdev->devfn, decode, command_bits, flags);
840 
841 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
842 		return 0;
843 
844 	if ((command_bits & PCI_COMMAND_IO) == 0)
845 		return 0;
846 
847 	domain = pci_domain_nr(pdev->bus);
848 	bus = pdev->bus->number;
849 
850 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
851 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
852 
853 	return rc;
854 }
855 
856 /*
857  * Called on each cpu to initialize the per_cpu UV data area.
858  * FIXME: hotplug not supported yet
859  */
860 void uv_cpu_init(void)
861 {
862 	/* CPU 0 initilization will be done via uv_system_init. */
863 	if (!uv_blade_info)
864 		return;
865 
866 	uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
867 
868 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
869 		set_x2apic_extra_bits(uv_hub_info->pnode);
870 }
871 
872 void __init uv_system_init(void)
873 {
874 	union uvh_rh_gam_config_mmr_u  m_n_config;
875 	union uvh_node_id_u node_id;
876 	unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
877 	int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val;
878 	int gnode_extra, min_pnode = 999999, max_pnode = -1;
879 	unsigned long mmr_base, present, paddr;
880 	unsigned short pnode_mask;
881 	unsigned char n_lshift;
882 	char *hub = (is_uv1_hub() ? "UV100/1000" :
883 		    (is_uv2_hub() ? "UV2000/3000" :
884 		    (is_uv3_hub() ? "UV300" : NULL)));
885 
886 	if (!hub) {
887 		pr_err("UV: Unknown/unsupported UV hub\n");
888 		return;
889 	}
890 	pr_info("UV: Found %s hub\n", hub);
891 	map_low_mmrs();
892 
893 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
894 	m_val = m_n_config.s.m_skt;
895 	n_val = m_n_config.s.n_skt;
896 	pnode_mask = (1 << n_val) - 1;
897 	n_lshift = get_n_lshift(m_val);
898 	mmr_base =
899 	    uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
900 	    ~UV_MMR_ENABLE;
901 
902 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
903 	gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
904 	gnode_upper = ((unsigned long)gnode_extra  << m_val);
905 	pr_info("UV: N:%d M:%d pnode_mask:0x%x gnode_upper/extra:0x%lx/0x%x n_lshift 0x%x\n",
906 			n_val, m_val, pnode_mask, gnode_upper, gnode_extra,
907 			n_lshift);
908 
909 	pr_info("UV: global MMR base 0x%lx\n", mmr_base);
910 
911 	for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
912 		uv_possible_blades +=
913 		  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
914 
915 	/* uv_num_possible_blades() is really the hub count */
916 	pr_info("UV: Found %d blades, %d hubs\n",
917 			is_uv1_hub() ? uv_num_possible_blades() :
918 			(uv_num_possible_blades() + 1) / 2,
919 			uv_num_possible_blades());
920 
921 	bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
922 	uv_blade_info = kzalloc(bytes, GFP_KERNEL);
923 	BUG_ON(!uv_blade_info);
924 
925 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
926 		uv_blade_info[blade].memory_nid = -1;
927 
928 	get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
929 
930 	bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
931 	uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
932 	BUG_ON(!uv_node_to_blade);
933 	memset(uv_node_to_blade, 255, bytes);
934 
935 	bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
936 	uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
937 	BUG_ON(!uv_cpu_to_blade);
938 	memset(uv_cpu_to_blade, 255, bytes);
939 
940 	blade = 0;
941 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
942 		present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
943 		for (j = 0; j < 64; j++) {
944 			if (!test_bit(j, &present))
945 				continue;
946 			pnode = (i * 64 + j) & pnode_mask;
947 			uv_blade_info[blade].pnode = pnode;
948 			uv_blade_info[blade].nr_possible_cpus = 0;
949 			uv_blade_info[blade].nr_online_cpus = 0;
950 			spin_lock_init(&uv_blade_info[blade].nmi_lock);
951 			min_pnode = min(pnode, min_pnode);
952 			max_pnode = max(pnode, max_pnode);
953 			blade++;
954 		}
955 	}
956 
957 	uv_bios_init();
958 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
959 			    &sn_region_size, &system_serial_number);
960 	uv_rtc_init();
961 
962 	for_each_present_cpu(cpu) {
963 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
964 
965 		nid = cpu_to_node(cpu);
966 		/*
967 		 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
968 		 */
969 		uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
970 		uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
971 		uv_cpu_hub_info(cpu)->hub_revision = uv_hub_info->hub_revision;
972 
973 		uv_cpu_hub_info(cpu)->m_shift = 64 - m_val;
974 		uv_cpu_hub_info(cpu)->n_lshift = n_lshift;
975 
976 		pnode = uv_apicid_to_pnode(apicid);
977 		blade = boot_pnode_to_blade(pnode);
978 		lcpu = uv_blade_info[blade].nr_possible_cpus;
979 		uv_blade_info[blade].nr_possible_cpus++;
980 
981 		/* Any node on the blade, else will contain -1. */
982 		uv_blade_info[blade].memory_nid = nid;
983 
984 		uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
985 		uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
986 		uv_cpu_hub_info(cpu)->m_val = m_val;
987 		uv_cpu_hub_info(cpu)->n_val = n_val;
988 		uv_cpu_hub_info(cpu)->numa_blade_id = blade;
989 		uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
990 		uv_cpu_hub_info(cpu)->pnode = pnode;
991 		uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
992 		uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
993 		uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
994 		uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
995 		uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
996 		uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
997 		uv_node_to_blade[nid] = blade;
998 		uv_cpu_to_blade[cpu] = blade;
999 	}
1000 
1001 	/* Add blade/pnode info for nodes without cpus */
1002 	for_each_online_node(nid) {
1003 		if (uv_node_to_blade[nid] >= 0)
1004 			continue;
1005 		paddr = node_start_pfn(nid) << PAGE_SHIFT;
1006 		pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
1007 		blade = boot_pnode_to_blade(pnode);
1008 		uv_node_to_blade[nid] = blade;
1009 	}
1010 
1011 	map_gru_high(max_pnode);
1012 	map_mmr_high(max_pnode);
1013 	map_mmioh_high(min_pnode, max_pnode);
1014 
1015 	uv_nmi_setup();
1016 	uv_cpu_init();
1017 	uv_scir_register_cpu_notifier();
1018 	proc_mkdir("sgi_uv", NULL);
1019 
1020 	/* register Legacy VGA I/O redirection handler */
1021 	pci_register_set_vga_state(uv_set_vga_state);
1022 
1023 	/*
1024 	 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
1025 	 * EFI is not enabled in the kdump kernel.
1026 	 */
1027 	if (is_kdump_kernel())
1028 		reboot_type = BOOT_ACPI;
1029 }
1030 
1031 apic_driver(apic_x2apic_uv_x);
1032