xref: /openbmc/linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision bc000245)
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) 2007-2013 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, is_uv1, is_uv2, is_uv3;
148 
149 	is_uv1 = !strcmp(oem_id, "SGI");
150 	is_uv2 = !strcmp(oem_id, "SGI2");
151 	is_uv3 = !strncmp(oem_id, "SGI3", 4);	/* there are varieties of UV3 */
152 	if (is_uv1 || is_uv2 || is_uv3) {
153 		uv_hub_info->hub_revision =
154 			(is_uv1 ? UV1_HUB_REVISION_BASE :
155 			(is_uv2 ? UV2_HUB_REVISION_BASE :
156 				  UV3_HUB_REVISION_BASE));
157 		pnodeid = early_get_pnodeid();
158 		early_get_apic_pnode_shift();
159 		x86_platform.is_untracked_pat_range =  uv_is_untracked_pat_range;
160 		x86_platform.nmi_init = uv_nmi_init;
161 		if (!strcmp(oem_table_id, "UVL"))
162 			uv_system_type = UV_LEGACY_APIC;
163 		else if (!strcmp(oem_table_id, "UVX"))
164 			uv_system_type = UV_X2APIC;
165 		else if (!strcmp(oem_table_id, "UVH")) {
166 			__this_cpu_write(x2apic_extra_bits,
167 				pnodeid << uvh_apicid.s.pnode_shift);
168 			uv_system_type = UV_NON_UNIQUE_APIC;
169 			uv_set_apicid_hibit();
170 			return 1;
171 		}
172 	}
173 	return 0;
174 }
175 
176 enum uv_system_type get_uv_system_type(void)
177 {
178 	return uv_system_type;
179 }
180 
181 int is_uv_system(void)
182 {
183 	return uv_system_type != UV_NONE;
184 }
185 EXPORT_SYMBOL_GPL(is_uv_system);
186 
187 DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
188 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
189 
190 struct uv_blade_info *uv_blade_info;
191 EXPORT_SYMBOL_GPL(uv_blade_info);
192 
193 short *uv_node_to_blade;
194 EXPORT_SYMBOL_GPL(uv_node_to_blade);
195 
196 short *uv_cpu_to_blade;
197 EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
198 
199 short uv_possible_blades;
200 EXPORT_SYMBOL_GPL(uv_possible_blades);
201 
202 unsigned long sn_rtc_cycles_per_second;
203 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
204 
205 static int uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
206 {
207 #ifdef CONFIG_SMP
208 	unsigned long val;
209 	int pnode;
210 
211 	pnode = uv_apicid_to_pnode(phys_apicid);
212 	phys_apicid |= uv_apicid_hibits;
213 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
214 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
215 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
216 	    APIC_DM_INIT;
217 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
218 
219 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
220 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
221 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
222 	    APIC_DM_STARTUP;
223 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
224 
225 	atomic_set(&init_deasserted, 1);
226 #endif
227 	return 0;
228 }
229 
230 static void uv_send_IPI_one(int cpu, int vector)
231 {
232 	unsigned long apicid;
233 	int pnode;
234 
235 	apicid = per_cpu(x86_cpu_to_apicid, cpu);
236 	pnode = uv_apicid_to_pnode(apicid);
237 	uv_hub_send_ipi(pnode, apicid, vector);
238 }
239 
240 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
241 {
242 	unsigned int cpu;
243 
244 	for_each_cpu(cpu, mask)
245 		uv_send_IPI_one(cpu, vector);
246 }
247 
248 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
249 {
250 	unsigned int this_cpu = smp_processor_id();
251 	unsigned int cpu;
252 
253 	for_each_cpu(cpu, mask) {
254 		if (cpu != this_cpu)
255 			uv_send_IPI_one(cpu, vector);
256 	}
257 }
258 
259 static void uv_send_IPI_allbutself(int vector)
260 {
261 	unsigned int this_cpu = smp_processor_id();
262 	unsigned int cpu;
263 
264 	for_each_online_cpu(cpu) {
265 		if (cpu != this_cpu)
266 			uv_send_IPI_one(cpu, vector);
267 	}
268 }
269 
270 static void uv_send_IPI_all(int vector)
271 {
272 	uv_send_IPI_mask(cpu_online_mask, vector);
273 }
274 
275 static int uv_apic_id_valid(int apicid)
276 {
277 	return 1;
278 }
279 
280 static int uv_apic_id_registered(void)
281 {
282 	return 1;
283 }
284 
285 static void uv_init_apic_ldr(void)
286 {
287 }
288 
289 static int
290 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
291 			  const struct cpumask *andmask,
292 			  unsigned int *apicid)
293 {
294 	int unsigned cpu;
295 
296 	/*
297 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
298 	 * May as well be the first.
299 	 */
300 	for_each_cpu_and(cpu, cpumask, andmask) {
301 		if (cpumask_test_cpu(cpu, cpu_online_mask))
302 			break;
303 	}
304 
305 	if (likely(cpu < nr_cpu_ids)) {
306 		*apicid = per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
307 		return 0;
308 	}
309 
310 	return -EINVAL;
311 }
312 
313 static unsigned int x2apic_get_apic_id(unsigned long x)
314 {
315 	unsigned int id;
316 
317 	WARN_ON(preemptible() && num_online_cpus() > 1);
318 	id = x | __this_cpu_read(x2apic_extra_bits);
319 
320 	return id;
321 }
322 
323 static unsigned long set_apic_id(unsigned int id)
324 {
325 	unsigned long x;
326 
327 	/* maskout x2apic_extra_bits ? */
328 	x = id;
329 	return x;
330 }
331 
332 static unsigned int uv_read_apic_id(void)
333 {
334 
335 	return x2apic_get_apic_id(apic_read(APIC_ID));
336 }
337 
338 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
339 {
340 	return uv_read_apic_id() >> index_msb;
341 }
342 
343 static void uv_send_IPI_self(int vector)
344 {
345 	apic_write(APIC_SELF_IPI, vector);
346 }
347 
348 static int uv_probe(void)
349 {
350 	return apic == &apic_x2apic_uv_x;
351 }
352 
353 static struct apic __refdata apic_x2apic_uv_x = {
354 
355 	.name				= "UV large system",
356 	.probe				= uv_probe,
357 	.acpi_madt_oem_check		= uv_acpi_madt_oem_check,
358 	.apic_id_valid			= uv_apic_id_valid,
359 	.apic_id_registered		= uv_apic_id_registered,
360 
361 	.irq_delivery_mode		= dest_Fixed,
362 	.irq_dest_mode			= 0, /* physical */
363 
364 	.target_cpus			= online_target_cpus,
365 	.disable_esr			= 0,
366 	.dest_logical			= APIC_DEST_LOGICAL,
367 	.check_apicid_used		= NULL,
368 	.check_apicid_present		= NULL,
369 
370 	.vector_allocation_domain	= default_vector_allocation_domain,
371 	.init_apic_ldr			= uv_init_apic_ldr,
372 
373 	.ioapic_phys_id_map		= NULL,
374 	.setup_apic_routing		= NULL,
375 	.multi_timer_check		= NULL,
376 	.cpu_present_to_apicid		= default_cpu_present_to_apicid,
377 	.apicid_to_cpu_present		= NULL,
378 	.setup_portio_remap		= NULL,
379 	.check_phys_apicid_present	= default_check_phys_apicid_present,
380 	.enable_apic_mode		= NULL,
381 	.phys_pkg_id			= uv_phys_pkg_id,
382 	.mps_oem_check			= NULL,
383 
384 	.get_apic_id			= x2apic_get_apic_id,
385 	.set_apic_id			= set_apic_id,
386 	.apic_id_mask			= 0xFFFFFFFFu,
387 
388 	.cpu_mask_to_apicid_and		= uv_cpu_mask_to_apicid_and,
389 
390 	.send_IPI_mask			= uv_send_IPI_mask,
391 	.send_IPI_mask_allbutself	= uv_send_IPI_mask_allbutself,
392 	.send_IPI_allbutself		= uv_send_IPI_allbutself,
393 	.send_IPI_all			= uv_send_IPI_all,
394 	.send_IPI_self			= uv_send_IPI_self,
395 
396 	.wakeup_secondary_cpu		= uv_wakeup_secondary,
397 	.trampoline_phys_low		= DEFAULT_TRAMPOLINE_PHYS_LOW,
398 	.trampoline_phys_high		= DEFAULT_TRAMPOLINE_PHYS_HIGH,
399 	.wait_for_init_deassert		= NULL,
400 	.smp_callin_clear_local_apic	= NULL,
401 	.inquire_remote_apic		= NULL,
402 
403 	.read				= native_apic_msr_read,
404 	.write				= native_apic_msr_write,
405 	.eoi_write			= native_apic_msr_eoi_write,
406 	.icr_read			= native_x2apic_icr_read,
407 	.icr_write			= native_x2apic_icr_write,
408 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
409 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
410 };
411 
412 static void set_x2apic_extra_bits(int pnode)
413 {
414 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
415 }
416 
417 /*
418  * Called on boot cpu.
419  */
420 static __init int boot_pnode_to_blade(int pnode)
421 {
422 	int blade;
423 
424 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
425 		if (pnode == uv_blade_info[blade].pnode)
426 			return blade;
427 	BUG();
428 }
429 
430 struct redir_addr {
431 	unsigned long redirect;
432 	unsigned long alias;
433 };
434 
435 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
436 
437 static __initdata struct redir_addr redir_addrs[] = {
438 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
439 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
440 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
441 };
442 
443 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
444 {
445 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
446 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
447 	int i;
448 
449 	for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
450 		alias.v = uv_read_local_mmr(redir_addrs[i].alias);
451 		if (alias.s.enable && alias.s.base == 0) {
452 			*size = (1UL << alias.s.m_alias);
453 			redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
454 			*base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
455 			return;
456 		}
457 	}
458 	*base = *size = 0;
459 }
460 
461 enum map_type {map_wb, map_uc};
462 
463 static __init void map_high(char *id, unsigned long base, int pshift,
464 			int bshift, int max_pnode, enum map_type map_type)
465 {
466 	unsigned long bytes, paddr;
467 
468 	paddr = base << pshift;
469 	bytes = (1UL << bshift) * (max_pnode + 1);
470 	if (!paddr) {
471 		pr_info("UV: Map %s_HI base address NULL\n", id);
472 		return;
473 	}
474 	pr_debug("UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr, paddr + bytes);
475 	if (map_type == map_uc)
476 		init_extra_mapping_uc(paddr, bytes);
477 	else
478 		init_extra_mapping_wb(paddr, bytes);
479 }
480 
481 static __init void map_gru_distributed(unsigned long c)
482 {
483 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
484 	u64 paddr;
485 	unsigned long bytes;
486 	int nid;
487 
488 	gru.v = c;
489 	/* only base bits 42:28 relevant in dist mode */
490 	gru_dist_base = gru.v & 0x000007fff0000000UL;
491 	if (!gru_dist_base) {
492 		pr_info("UV: Map GRU_DIST base address NULL\n");
493 		return;
494 	}
495 	bytes = 1UL << UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
496 	gru_dist_lmask = ((1UL << uv_hub_info->m_val) - 1) & ~(bytes - 1);
497 	gru_dist_umask = ~((1UL << uv_hub_info->m_val) - 1);
498 	gru_dist_base &= gru_dist_lmask; /* Clear bits above M */
499 	for_each_online_node(nid) {
500 		paddr = ((u64)uv_node_to_pnode(nid) << uv_hub_info->m_val) |
501 				gru_dist_base;
502 		init_extra_mapping_wb(paddr, bytes);
503 		gru_first_node_paddr = min(paddr, gru_first_node_paddr);
504 		gru_last_node_paddr = max(paddr, gru_last_node_paddr);
505 	}
506 	/* Save upper (63:M) bits of address only for is_GRU_range */
507 	gru_first_node_paddr &= gru_dist_umask;
508 	gru_last_node_paddr &= gru_dist_umask;
509 	pr_debug("UV: Map GRU_DIST base 0x%016llx  0x%016llx - 0x%016llx\n",
510 		gru_dist_base, gru_first_node_paddr, gru_last_node_paddr);
511 }
512 
513 static __init void map_gru_high(int max_pnode)
514 {
515 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
516 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
517 
518 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
519 	if (!gru.s.enable) {
520 		pr_info("UV: GRU disabled\n");
521 		return;
522 	}
523 
524 	if (is_uv3_hub() && gru.s3.mode) {
525 		map_gru_distributed(gru.v);
526 		return;
527 	}
528 	map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
529 	gru_start_paddr = ((u64)gru.s.base << shift);
530 	gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
531 }
532 
533 static __init void map_mmr_high(int max_pnode)
534 {
535 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
536 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
537 
538 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
539 	if (mmr.s.enable)
540 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
541 	else
542 		pr_info("UV: MMR disabled\n");
543 }
544 
545 /*
546  * This commonality works because both 0 & 1 versions of the MMIOH OVERLAY
547  * and REDIRECT MMR regs are exactly the same on UV3.
548  */
549 struct mmioh_config {
550 	unsigned long overlay;
551 	unsigned long redirect;
552 	char *id;
553 };
554 
555 static __initdata struct mmioh_config mmiohs[] = {
556 	{
557 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR,
558 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR,
559 		"MMIOH0"
560 	},
561 	{
562 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG1_MMR,
563 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG1_MMR,
564 		"MMIOH1"
565 	},
566 };
567 
568 static __init void map_mmioh_high_uv3(int index, int min_pnode, int max_pnode)
569 {
570 	union uv3h_rh_gam_mmioh_overlay_config0_mmr_u overlay;
571 	unsigned long mmr;
572 	unsigned long base;
573 	int i, n, shift, m_io, max_io;
574 	int nasid, lnasid, fi, li;
575 	char *id;
576 
577 	id = mmiohs[index].id;
578 	overlay.v = uv_read_local_mmr(mmiohs[index].overlay);
579 	pr_info("UV: %s overlay 0x%lx base:0x%x m_io:%d\n",
580 		id, overlay.v, overlay.s3.base, overlay.s3.m_io);
581 	if (!overlay.s3.enable) {
582 		pr_info("UV: %s disabled\n", id);
583 		return;
584 	}
585 
586 	shift = UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR_BASE_SHFT;
587 	base = (unsigned long)overlay.s3.base;
588 	m_io = overlay.s3.m_io;
589 	mmr = mmiohs[index].redirect;
590 	n = UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR_DEPTH;
591 	min_pnode *= 2;				/* convert to NASID */
592 	max_pnode *= 2;
593 	max_io = lnasid = fi = li = -1;
594 
595 	for (i = 0; i < n; i++) {
596 		union uv3h_rh_gam_mmioh_redirect_config0_mmr_u redirect;
597 
598 		redirect.v = uv_read_local_mmr(mmr + i * 8);
599 		nasid = redirect.s3.nasid;
600 		if (nasid < min_pnode || max_pnode < nasid)
601 			nasid = -1;		/* invalid NASID */
602 
603 		if (nasid == lnasid) {
604 			li = i;
605 			if (i != n-1)		/* last entry check */
606 				continue;
607 		}
608 
609 		/* check if we have a cached (or last) redirect to print */
610 		if (lnasid != -1 || (i == n-1 && nasid != -1))  {
611 			unsigned long addr1, addr2;
612 			int f, l;
613 
614 			if (lnasid == -1) {
615 				f = l = i;
616 				lnasid = nasid;
617 			} else {
618 				f = fi;
619 				l = li;
620 			}
621 			addr1 = (base << shift) +
622 				f * (unsigned long)(1 << m_io);
623 			addr2 = (base << shift) +
624 				(l + 1) * (unsigned long)(1 << m_io);
625 			pr_info("UV: %s[%03d..%03d] NASID 0x%04x ADDR 0x%016lx - 0x%016lx\n",
626 				id, fi, li, lnasid, addr1, addr2);
627 			if (max_io < l)
628 				max_io = l;
629 		}
630 		fi = li = i;
631 		lnasid = nasid;
632 	}
633 
634 	pr_info("UV: %s base:0x%lx shift:%d M_IO:%d MAX_IO:%d\n",
635 		id, base, shift, m_io, max_io);
636 
637 	if (max_io >= 0)
638 		map_high(id, base, shift, m_io, max_io, map_uc);
639 }
640 
641 static __init void map_mmioh_high(int min_pnode, int max_pnode)
642 {
643 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
644 	unsigned long mmr, base;
645 	int shift, enable, m_io, n_io;
646 
647 	if (is_uv3_hub()) {
648 		/* Map both MMIOH Regions */
649 		map_mmioh_high_uv3(0, min_pnode, max_pnode);
650 		map_mmioh_high_uv3(1, min_pnode, max_pnode);
651 		return;
652 	}
653 
654 	if (is_uv1_hub()) {
655 		mmr = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
656 		shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
657 		mmioh.v = uv_read_local_mmr(mmr);
658 		enable = !!mmioh.s1.enable;
659 		base = mmioh.s1.base;
660 		m_io = mmioh.s1.m_io;
661 		n_io = mmioh.s1.n_io;
662 	} else if (is_uv2_hub()) {
663 		mmr = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
664 		shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
665 		mmioh.v = uv_read_local_mmr(mmr);
666 		enable = !!mmioh.s2.enable;
667 		base = mmioh.s2.base;
668 		m_io = mmioh.s2.m_io;
669 		n_io = mmioh.s2.n_io;
670 	} else
671 		return;
672 
673 	if (enable) {
674 		max_pnode &= (1 << n_io) - 1;
675 		pr_info(
676 		    "UV: base:0x%lx shift:%d N_IO:%d M_IO:%d max_pnode:0x%x\n",
677 			base, shift, m_io, n_io, max_pnode);
678 		map_high("MMIOH", base, shift, m_io, max_pnode, map_uc);
679 	} else {
680 		pr_info("UV: MMIOH disabled\n");
681 	}
682 }
683 
684 static __init void map_low_mmrs(void)
685 {
686 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
687 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
688 }
689 
690 static __init void uv_rtc_init(void)
691 {
692 	long status;
693 	u64 ticks_per_sec;
694 
695 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
696 					&ticks_per_sec);
697 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
698 		printk(KERN_WARNING
699 			"unable to determine platform RTC clock frequency, "
700 			"guessing.\n");
701 		/* BIOS gives wrong value for clock freq. so guess */
702 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
703 	} else
704 		sn_rtc_cycles_per_second = ticks_per_sec;
705 }
706 
707 /*
708  * percpu heartbeat timer
709  */
710 static void uv_heartbeat(unsigned long ignored)
711 {
712 	struct timer_list *timer = &uv_hub_info->scir.timer;
713 	unsigned char bits = uv_hub_info->scir.state;
714 
715 	/* flip heartbeat bit */
716 	bits ^= SCIR_CPU_HEARTBEAT;
717 
718 	/* is this cpu idle? */
719 	if (idle_cpu(raw_smp_processor_id()))
720 		bits &= ~SCIR_CPU_ACTIVITY;
721 	else
722 		bits |= SCIR_CPU_ACTIVITY;
723 
724 	/* update system controller interface reg */
725 	uv_set_scir_bits(bits);
726 
727 	/* enable next timer period */
728 	mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
729 }
730 
731 static void uv_heartbeat_enable(int cpu)
732 {
733 	while (!uv_cpu_hub_info(cpu)->scir.enabled) {
734 		struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
735 
736 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
737 		setup_timer(timer, uv_heartbeat, cpu);
738 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
739 		add_timer_on(timer, cpu);
740 		uv_cpu_hub_info(cpu)->scir.enabled = 1;
741 
742 		/* also ensure that boot cpu is enabled */
743 		cpu = 0;
744 	}
745 }
746 
747 #ifdef CONFIG_HOTPLUG_CPU
748 static void uv_heartbeat_disable(int cpu)
749 {
750 	if (uv_cpu_hub_info(cpu)->scir.enabled) {
751 		uv_cpu_hub_info(cpu)->scir.enabled = 0;
752 		del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
753 	}
754 	uv_set_cpu_scir_bits(cpu, 0xff);
755 }
756 
757 /*
758  * cpu hotplug notifier
759  */
760 static int uv_scir_cpu_notify(struct notifier_block *self, unsigned long action,
761 			      void *hcpu)
762 {
763 	long cpu = (long)hcpu;
764 
765 	switch (action) {
766 	case CPU_ONLINE:
767 		uv_heartbeat_enable(cpu);
768 		break;
769 	case CPU_DOWN_PREPARE:
770 		uv_heartbeat_disable(cpu);
771 		break;
772 	default:
773 		break;
774 	}
775 	return NOTIFY_OK;
776 }
777 
778 static __init void uv_scir_register_cpu_notifier(void)
779 {
780 	hotcpu_notifier(uv_scir_cpu_notify, 0);
781 }
782 
783 #else /* !CONFIG_HOTPLUG_CPU */
784 
785 static __init void uv_scir_register_cpu_notifier(void)
786 {
787 }
788 
789 static __init int uv_init_heartbeat(void)
790 {
791 	int cpu;
792 
793 	if (is_uv_system())
794 		for_each_online_cpu(cpu)
795 			uv_heartbeat_enable(cpu);
796 	return 0;
797 }
798 
799 late_initcall(uv_init_heartbeat);
800 
801 #endif /* !CONFIG_HOTPLUG_CPU */
802 
803 /* Direct Legacy VGA I/O traffic to designated IOH */
804 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
805 		      unsigned int command_bits, u32 flags)
806 {
807 	int domain, bus, rc;
808 
809 	PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
810 			pdev->devfn, decode, command_bits, flags);
811 
812 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
813 		return 0;
814 
815 	if ((command_bits & PCI_COMMAND_IO) == 0)
816 		return 0;
817 
818 	domain = pci_domain_nr(pdev->bus);
819 	bus = pdev->bus->number;
820 
821 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
822 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
823 
824 	return rc;
825 }
826 
827 /*
828  * Called on each cpu to initialize the per_cpu UV data area.
829  * FIXME: hotplug not supported yet
830  */
831 void uv_cpu_init(void)
832 {
833 	/* CPU 0 initilization will be done via uv_system_init. */
834 	if (!uv_blade_info)
835 		return;
836 
837 	uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
838 
839 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
840 		set_x2apic_extra_bits(uv_hub_info->pnode);
841 }
842 
843 void __init uv_system_init(void)
844 {
845 	union uvh_rh_gam_config_mmr_u  m_n_config;
846 	union uvh_node_id_u node_id;
847 	unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
848 	int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val;
849 	int gnode_extra, min_pnode = 999999, max_pnode = -1;
850 	unsigned long mmr_base, present, paddr;
851 	unsigned short pnode_mask;
852 	char *hub = (is_uv1_hub() ? "UV1" :
853 		    (is_uv2_hub() ? "UV2" :
854 				    "UV3"));
855 
856 	pr_info("UV: Found %s hub\n", hub);
857 	map_low_mmrs();
858 
859 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
860 	m_val = m_n_config.s.m_skt;
861 	n_val = m_n_config.s.n_skt;
862 	pnode_mask = (1 << n_val) - 1;
863 	mmr_base =
864 	    uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
865 	    ~UV_MMR_ENABLE;
866 
867 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
868 	gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
869 	gnode_upper = ((unsigned long)gnode_extra  << m_val);
870 	pr_info("UV: N:%d M:%d pnode_mask:0x%x gnode_upper/extra:0x%lx/0x%x\n",
871 			n_val, m_val, pnode_mask, gnode_upper, gnode_extra);
872 
873 	pr_info("UV: global MMR base 0x%lx\n", mmr_base);
874 
875 	for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
876 		uv_possible_blades +=
877 		  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
878 
879 	/* uv_num_possible_blades() is really the hub count */
880 	pr_info("UV: Found %d blades, %d hubs\n",
881 			is_uv1_hub() ? uv_num_possible_blades() :
882 			(uv_num_possible_blades() + 1) / 2,
883 			uv_num_possible_blades());
884 
885 	bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
886 	uv_blade_info = kzalloc(bytes, GFP_KERNEL);
887 	BUG_ON(!uv_blade_info);
888 
889 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
890 		uv_blade_info[blade].memory_nid = -1;
891 
892 	get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
893 
894 	bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
895 	uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
896 	BUG_ON(!uv_node_to_blade);
897 	memset(uv_node_to_blade, 255, bytes);
898 
899 	bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
900 	uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
901 	BUG_ON(!uv_cpu_to_blade);
902 	memset(uv_cpu_to_blade, 255, bytes);
903 
904 	blade = 0;
905 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
906 		present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
907 		for (j = 0; j < 64; j++) {
908 			if (!test_bit(j, &present))
909 				continue;
910 			pnode = (i * 64 + j) & pnode_mask;
911 			uv_blade_info[blade].pnode = pnode;
912 			uv_blade_info[blade].nr_possible_cpus = 0;
913 			uv_blade_info[blade].nr_online_cpus = 0;
914 			spin_lock_init(&uv_blade_info[blade].nmi_lock);
915 			min_pnode = min(pnode, min_pnode);
916 			max_pnode = max(pnode, max_pnode);
917 			blade++;
918 		}
919 	}
920 
921 	uv_bios_init();
922 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
923 			    &sn_region_size, &system_serial_number);
924 	uv_rtc_init();
925 
926 	for_each_present_cpu(cpu) {
927 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
928 
929 		nid = cpu_to_node(cpu);
930 		/*
931 		 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
932 		 */
933 		uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
934 		uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
935 		uv_cpu_hub_info(cpu)->hub_revision = uv_hub_info->hub_revision;
936 
937 		uv_cpu_hub_info(cpu)->m_shift = 64 - m_val;
938 		uv_cpu_hub_info(cpu)->n_lshift = is_uv2_1_hub() ?
939 				(m_val == 40 ? 40 : 39) : m_val;
940 
941 		pnode = uv_apicid_to_pnode(apicid);
942 		blade = boot_pnode_to_blade(pnode);
943 		lcpu = uv_blade_info[blade].nr_possible_cpus;
944 		uv_blade_info[blade].nr_possible_cpus++;
945 
946 		/* Any node on the blade, else will contain -1. */
947 		uv_blade_info[blade].memory_nid = nid;
948 
949 		uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
950 		uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
951 		uv_cpu_hub_info(cpu)->m_val = m_val;
952 		uv_cpu_hub_info(cpu)->n_val = n_val;
953 		uv_cpu_hub_info(cpu)->numa_blade_id = blade;
954 		uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
955 		uv_cpu_hub_info(cpu)->pnode = pnode;
956 		uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
957 		uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
958 		uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
959 		uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
960 		uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
961 		uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
962 		uv_node_to_blade[nid] = blade;
963 		uv_cpu_to_blade[cpu] = blade;
964 	}
965 
966 	/* Add blade/pnode info for nodes without cpus */
967 	for_each_online_node(nid) {
968 		if (uv_node_to_blade[nid] >= 0)
969 			continue;
970 		paddr = node_start_pfn(nid) << PAGE_SHIFT;
971 		pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
972 		blade = boot_pnode_to_blade(pnode);
973 		uv_node_to_blade[nid] = blade;
974 	}
975 
976 	map_gru_high(max_pnode);
977 	map_mmr_high(max_pnode);
978 	map_mmioh_high(min_pnode, max_pnode);
979 
980 	uv_nmi_setup();
981 	uv_cpu_init();
982 	uv_scir_register_cpu_notifier();
983 	uv_register_nmi_notifier();
984 	proc_mkdir("sgi_uv", NULL);
985 
986 	/* register Legacy VGA I/O redirection handler */
987 	pci_register_set_vga_state(uv_set_vga_state);
988 
989 	/*
990 	 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
991 	 * EFI is not enabled in the kdump kernel.
992 	 */
993 	if (is_kdump_kernel())
994 		reboot_type = BOOT_ACPI;
995 }
996 
997 apic_driver(apic_x2apic_uv_x);
998