xref: /openbmc/linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision e917ba44)
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, 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 
369 	.vector_allocation_domain	= default_vector_allocation_domain,
370 	.init_apic_ldr			= uv_init_apic_ldr,
371 
372 	.ioapic_phys_id_map		= NULL,
373 	.setup_apic_routing		= NULL,
374 	.cpu_present_to_apicid		= default_cpu_present_to_apicid,
375 	.apicid_to_cpu_present		= NULL,
376 	.check_phys_apicid_present	= default_check_phys_apicid_present,
377 	.phys_pkg_id			= uv_phys_pkg_id,
378 
379 	.get_apic_id			= x2apic_get_apic_id,
380 	.set_apic_id			= set_apic_id,
381 	.apic_id_mask			= 0xFFFFFFFFu,
382 
383 	.cpu_mask_to_apicid_and		= uv_cpu_mask_to_apicid_and,
384 
385 	.send_IPI_mask			= uv_send_IPI_mask,
386 	.send_IPI_mask_allbutself	= uv_send_IPI_mask_allbutself,
387 	.send_IPI_allbutself		= uv_send_IPI_allbutself,
388 	.send_IPI_all			= uv_send_IPI_all,
389 	.send_IPI_self			= uv_send_IPI_self,
390 
391 	.wakeup_secondary_cpu		= uv_wakeup_secondary,
392 	.wait_for_init_deassert		= false,
393 	.inquire_remote_apic		= NULL,
394 
395 	.read				= native_apic_msr_read,
396 	.write				= native_apic_msr_write,
397 	.eoi_write			= native_apic_msr_eoi_write,
398 	.icr_read			= native_x2apic_icr_read,
399 	.icr_write			= native_x2apic_icr_write,
400 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
401 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
402 };
403 
404 static void set_x2apic_extra_bits(int pnode)
405 {
406 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
407 }
408 
409 /*
410  * Called on boot cpu.
411  */
412 static __init int boot_pnode_to_blade(int pnode)
413 {
414 	int blade;
415 
416 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
417 		if (pnode == uv_blade_info[blade].pnode)
418 			return blade;
419 	BUG();
420 }
421 
422 struct redir_addr {
423 	unsigned long redirect;
424 	unsigned long alias;
425 };
426 
427 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
428 
429 static __initdata struct redir_addr redir_addrs[] = {
430 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
431 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
432 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
433 };
434 
435 static unsigned char get_n_lshift(int m_val)
436 {
437 	union uv3h_gr0_gam_gr_config_u m_gr_config;
438 
439 	if (is_uv1_hub())
440 		return m_val;
441 
442 	if (is_uv2_hub())
443 		return m_val == 40 ? 40 : 39;
444 
445 	m_gr_config.v = uv_read_local_mmr(UV3H_GR0_GAM_GR_CONFIG);
446 	return m_gr_config.s3.m_skt;
447 }
448 
449 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
450 {
451 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
452 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
453 	int i;
454 
455 	for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
456 		alias.v = uv_read_local_mmr(redir_addrs[i].alias);
457 		if (alias.s.enable && alias.s.base == 0) {
458 			*size = (1UL << alias.s.m_alias);
459 			redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
460 			*base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
461 			return;
462 		}
463 	}
464 	*base = *size = 0;
465 }
466 
467 enum map_type {map_wb, map_uc};
468 
469 static __init void map_high(char *id, unsigned long base, int pshift,
470 			int bshift, int max_pnode, enum map_type map_type)
471 {
472 	unsigned long bytes, paddr;
473 
474 	paddr = base << pshift;
475 	bytes = (1UL << bshift) * (max_pnode + 1);
476 	if (!paddr) {
477 		pr_info("UV: Map %s_HI base address NULL\n", id);
478 		return;
479 	}
480 	pr_debug("UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr, paddr + bytes);
481 	if (map_type == map_uc)
482 		init_extra_mapping_uc(paddr, bytes);
483 	else
484 		init_extra_mapping_wb(paddr, bytes);
485 }
486 
487 static __init void map_gru_distributed(unsigned long c)
488 {
489 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
490 	u64 paddr;
491 	unsigned long bytes;
492 	int nid;
493 
494 	gru.v = c;
495 	/* only base bits 42:28 relevant in dist mode */
496 	gru_dist_base = gru.v & 0x000007fff0000000UL;
497 	if (!gru_dist_base) {
498 		pr_info("UV: Map GRU_DIST base address NULL\n");
499 		return;
500 	}
501 	bytes = 1UL << UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
502 	gru_dist_lmask = ((1UL << uv_hub_info->m_val) - 1) & ~(bytes - 1);
503 	gru_dist_umask = ~((1UL << uv_hub_info->m_val) - 1);
504 	gru_dist_base &= gru_dist_lmask; /* Clear bits above M */
505 	for_each_online_node(nid) {
506 		paddr = ((u64)uv_node_to_pnode(nid) << uv_hub_info->m_val) |
507 				gru_dist_base;
508 		init_extra_mapping_wb(paddr, bytes);
509 		gru_first_node_paddr = min(paddr, gru_first_node_paddr);
510 		gru_last_node_paddr = max(paddr, gru_last_node_paddr);
511 	}
512 	/* Save upper (63:M) bits of address only for is_GRU_range */
513 	gru_first_node_paddr &= gru_dist_umask;
514 	gru_last_node_paddr &= gru_dist_umask;
515 	pr_debug("UV: Map GRU_DIST base 0x%016llx  0x%016llx - 0x%016llx\n",
516 		gru_dist_base, gru_first_node_paddr, gru_last_node_paddr);
517 }
518 
519 static __init void map_gru_high(int max_pnode)
520 {
521 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
522 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
523 
524 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
525 	if (!gru.s.enable) {
526 		pr_info("UV: GRU disabled\n");
527 		return;
528 	}
529 
530 	if (is_uv3_hub() && gru.s3.mode) {
531 		map_gru_distributed(gru.v);
532 		return;
533 	}
534 	map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
535 	gru_start_paddr = ((u64)gru.s.base << shift);
536 	gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
537 }
538 
539 static __init void map_mmr_high(int max_pnode)
540 {
541 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
542 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
543 
544 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
545 	if (mmr.s.enable)
546 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
547 	else
548 		pr_info("UV: MMR disabled\n");
549 }
550 
551 /*
552  * This commonality works because both 0 & 1 versions of the MMIOH OVERLAY
553  * and REDIRECT MMR regs are exactly the same on UV3.
554  */
555 struct mmioh_config {
556 	unsigned long overlay;
557 	unsigned long redirect;
558 	char *id;
559 };
560 
561 static __initdata struct mmioh_config mmiohs[] = {
562 	{
563 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR,
564 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR,
565 		"MMIOH0"
566 	},
567 	{
568 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG1_MMR,
569 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG1_MMR,
570 		"MMIOH1"
571 	},
572 };
573 
574 static __init void map_mmioh_high_uv3(int index, int min_pnode, int max_pnode)
575 {
576 	union uv3h_rh_gam_mmioh_overlay_config0_mmr_u overlay;
577 	unsigned long mmr;
578 	unsigned long base;
579 	int i, n, shift, m_io, max_io;
580 	int nasid, lnasid, fi, li;
581 	char *id;
582 
583 	id = mmiohs[index].id;
584 	overlay.v = uv_read_local_mmr(mmiohs[index].overlay);
585 	pr_info("UV: %s overlay 0x%lx base:0x%x m_io:%d\n",
586 		id, overlay.v, overlay.s3.base, overlay.s3.m_io);
587 	if (!overlay.s3.enable) {
588 		pr_info("UV: %s disabled\n", id);
589 		return;
590 	}
591 
592 	shift = UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR_BASE_SHFT;
593 	base = (unsigned long)overlay.s3.base;
594 	m_io = overlay.s3.m_io;
595 	mmr = mmiohs[index].redirect;
596 	n = UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR_DEPTH;
597 	min_pnode *= 2;				/* convert to NASID */
598 	max_pnode *= 2;
599 	max_io = lnasid = fi = li = -1;
600 
601 	for (i = 0; i < n; i++) {
602 		union uv3h_rh_gam_mmioh_redirect_config0_mmr_u redirect;
603 
604 		redirect.v = uv_read_local_mmr(mmr + i * 8);
605 		nasid = redirect.s3.nasid;
606 		if (nasid < min_pnode || max_pnode < nasid)
607 			nasid = -1;		/* invalid NASID */
608 
609 		if (nasid == lnasid) {
610 			li = i;
611 			if (i != n-1)		/* last entry check */
612 				continue;
613 		}
614 
615 		/* check if we have a cached (or last) redirect to print */
616 		if (lnasid != -1 || (i == n-1 && nasid != -1))  {
617 			unsigned long addr1, addr2;
618 			int f, l;
619 
620 			if (lnasid == -1) {
621 				f = l = i;
622 				lnasid = nasid;
623 			} else {
624 				f = fi;
625 				l = li;
626 			}
627 			addr1 = (base << shift) +
628 				f * (unsigned long)(1 << m_io);
629 			addr2 = (base << shift) +
630 				(l + 1) * (unsigned long)(1 << m_io);
631 			pr_info("UV: %s[%03d..%03d] NASID 0x%04x ADDR 0x%016lx - 0x%016lx\n",
632 				id, fi, li, lnasid, addr1, addr2);
633 			if (max_io < l)
634 				max_io = l;
635 		}
636 		fi = li = i;
637 		lnasid = nasid;
638 	}
639 
640 	pr_info("UV: %s base:0x%lx shift:%d M_IO:%d MAX_IO:%d\n",
641 		id, base, shift, m_io, max_io);
642 
643 	if (max_io >= 0)
644 		map_high(id, base, shift, m_io, max_io, map_uc);
645 }
646 
647 static __init void map_mmioh_high(int min_pnode, int max_pnode)
648 {
649 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
650 	unsigned long mmr, base;
651 	int shift, enable, m_io, n_io;
652 
653 	if (is_uv3_hub()) {
654 		/* Map both MMIOH Regions */
655 		map_mmioh_high_uv3(0, min_pnode, max_pnode);
656 		map_mmioh_high_uv3(1, min_pnode, max_pnode);
657 		return;
658 	}
659 
660 	if (is_uv1_hub()) {
661 		mmr = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
662 		shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
663 		mmioh.v = uv_read_local_mmr(mmr);
664 		enable = !!mmioh.s1.enable;
665 		base = mmioh.s1.base;
666 		m_io = mmioh.s1.m_io;
667 		n_io = mmioh.s1.n_io;
668 	} else if (is_uv2_hub()) {
669 		mmr = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
670 		shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
671 		mmioh.v = uv_read_local_mmr(mmr);
672 		enable = !!mmioh.s2.enable;
673 		base = mmioh.s2.base;
674 		m_io = mmioh.s2.m_io;
675 		n_io = mmioh.s2.n_io;
676 	} else
677 		return;
678 
679 	if (enable) {
680 		max_pnode &= (1 << n_io) - 1;
681 		pr_info(
682 		    "UV: base:0x%lx shift:%d N_IO:%d M_IO:%d max_pnode:0x%x\n",
683 			base, shift, m_io, n_io, max_pnode);
684 		map_high("MMIOH", base, shift, m_io, max_pnode, map_uc);
685 	} else {
686 		pr_info("UV: MMIOH disabled\n");
687 	}
688 }
689 
690 static __init void map_low_mmrs(void)
691 {
692 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
693 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
694 }
695 
696 static __init void uv_rtc_init(void)
697 {
698 	long status;
699 	u64 ticks_per_sec;
700 
701 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
702 					&ticks_per_sec);
703 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
704 		printk(KERN_WARNING
705 			"unable to determine platform RTC clock frequency, "
706 			"guessing.\n");
707 		/* BIOS gives wrong value for clock freq. so guess */
708 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
709 	} else
710 		sn_rtc_cycles_per_second = ticks_per_sec;
711 }
712 
713 /*
714  * percpu heartbeat timer
715  */
716 static void uv_heartbeat(unsigned long ignored)
717 {
718 	struct timer_list *timer = &uv_hub_info->scir.timer;
719 	unsigned char bits = uv_hub_info->scir.state;
720 
721 	/* flip heartbeat bit */
722 	bits ^= SCIR_CPU_HEARTBEAT;
723 
724 	/* is this cpu idle? */
725 	if (idle_cpu(raw_smp_processor_id()))
726 		bits &= ~SCIR_CPU_ACTIVITY;
727 	else
728 		bits |= SCIR_CPU_ACTIVITY;
729 
730 	/* update system controller interface reg */
731 	uv_set_scir_bits(bits);
732 
733 	/* enable next timer period */
734 	mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
735 }
736 
737 static void uv_heartbeat_enable(int cpu)
738 {
739 	while (!uv_cpu_hub_info(cpu)->scir.enabled) {
740 		struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
741 
742 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
743 		setup_timer(timer, uv_heartbeat, cpu);
744 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
745 		add_timer_on(timer, cpu);
746 		uv_cpu_hub_info(cpu)->scir.enabled = 1;
747 
748 		/* also ensure that boot cpu is enabled */
749 		cpu = 0;
750 	}
751 }
752 
753 #ifdef CONFIG_HOTPLUG_CPU
754 static void uv_heartbeat_disable(int cpu)
755 {
756 	if (uv_cpu_hub_info(cpu)->scir.enabled) {
757 		uv_cpu_hub_info(cpu)->scir.enabled = 0;
758 		del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
759 	}
760 	uv_set_cpu_scir_bits(cpu, 0xff);
761 }
762 
763 /*
764  * cpu hotplug notifier
765  */
766 static int uv_scir_cpu_notify(struct notifier_block *self, unsigned long action,
767 			      void *hcpu)
768 {
769 	long cpu = (long)hcpu;
770 
771 	switch (action) {
772 	case CPU_ONLINE:
773 		uv_heartbeat_enable(cpu);
774 		break;
775 	case CPU_DOWN_PREPARE:
776 		uv_heartbeat_disable(cpu);
777 		break;
778 	default:
779 		break;
780 	}
781 	return NOTIFY_OK;
782 }
783 
784 static __init void uv_scir_register_cpu_notifier(void)
785 {
786 	hotcpu_notifier(uv_scir_cpu_notify, 0);
787 }
788 
789 #else /* !CONFIG_HOTPLUG_CPU */
790 
791 static __init void uv_scir_register_cpu_notifier(void)
792 {
793 }
794 
795 static __init int uv_init_heartbeat(void)
796 {
797 	int cpu;
798 
799 	if (is_uv_system())
800 		for_each_online_cpu(cpu)
801 			uv_heartbeat_enable(cpu);
802 	return 0;
803 }
804 
805 late_initcall(uv_init_heartbeat);
806 
807 #endif /* !CONFIG_HOTPLUG_CPU */
808 
809 /* Direct Legacy VGA I/O traffic to designated IOH */
810 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
811 		      unsigned int command_bits, u32 flags)
812 {
813 	int domain, bus, rc;
814 
815 	PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
816 			pdev->devfn, decode, command_bits, flags);
817 
818 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
819 		return 0;
820 
821 	if ((command_bits & PCI_COMMAND_IO) == 0)
822 		return 0;
823 
824 	domain = pci_domain_nr(pdev->bus);
825 	bus = pdev->bus->number;
826 
827 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
828 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
829 
830 	return rc;
831 }
832 
833 /*
834  * Called on each cpu to initialize the per_cpu UV data area.
835  * FIXME: hotplug not supported yet
836  */
837 void uv_cpu_init(void)
838 {
839 	/* CPU 0 initilization will be done via uv_system_init. */
840 	if (!uv_blade_info)
841 		return;
842 
843 	uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
844 
845 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
846 		set_x2apic_extra_bits(uv_hub_info->pnode);
847 }
848 
849 void __init uv_system_init(void)
850 {
851 	union uvh_rh_gam_config_mmr_u  m_n_config;
852 	union uvh_node_id_u node_id;
853 	unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
854 	int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val;
855 	int gnode_extra, min_pnode = 999999, max_pnode = -1;
856 	unsigned long mmr_base, present, paddr;
857 	unsigned short pnode_mask;
858 	unsigned char n_lshift;
859 	char *hub = (is_uv1_hub() ? "UV1" :
860 		    (is_uv2_hub() ? "UV2" :
861 				    "UV3"));
862 
863 	pr_info("UV: Found %s hub\n", hub);
864 	map_low_mmrs();
865 
866 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
867 	m_val = m_n_config.s.m_skt;
868 	n_val = m_n_config.s.n_skt;
869 	pnode_mask = (1 << n_val) - 1;
870 	n_lshift = get_n_lshift(m_val);
871 	mmr_base =
872 	    uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
873 	    ~UV_MMR_ENABLE;
874 
875 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
876 	gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
877 	gnode_upper = ((unsigned long)gnode_extra  << m_val);
878 	pr_info("UV: N:%d M:%d pnode_mask:0x%x gnode_upper/extra:0x%lx/0x%x n_lshift 0x%x\n",
879 			n_val, m_val, pnode_mask, gnode_upper, gnode_extra,
880 			n_lshift);
881 
882 	pr_info("UV: global MMR base 0x%lx\n", mmr_base);
883 
884 	for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
885 		uv_possible_blades +=
886 		  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
887 
888 	/* uv_num_possible_blades() is really the hub count */
889 	pr_info("UV: Found %d blades, %d hubs\n",
890 			is_uv1_hub() ? uv_num_possible_blades() :
891 			(uv_num_possible_blades() + 1) / 2,
892 			uv_num_possible_blades());
893 
894 	bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
895 	uv_blade_info = kzalloc(bytes, GFP_KERNEL);
896 	BUG_ON(!uv_blade_info);
897 
898 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
899 		uv_blade_info[blade].memory_nid = -1;
900 
901 	get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
902 
903 	bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
904 	uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
905 	BUG_ON(!uv_node_to_blade);
906 	memset(uv_node_to_blade, 255, bytes);
907 
908 	bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
909 	uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
910 	BUG_ON(!uv_cpu_to_blade);
911 	memset(uv_cpu_to_blade, 255, bytes);
912 
913 	blade = 0;
914 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
915 		present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
916 		for (j = 0; j < 64; j++) {
917 			if (!test_bit(j, &present))
918 				continue;
919 			pnode = (i * 64 + j) & pnode_mask;
920 			uv_blade_info[blade].pnode = pnode;
921 			uv_blade_info[blade].nr_possible_cpus = 0;
922 			uv_blade_info[blade].nr_online_cpus = 0;
923 			spin_lock_init(&uv_blade_info[blade].nmi_lock);
924 			min_pnode = min(pnode, min_pnode);
925 			max_pnode = max(pnode, max_pnode);
926 			blade++;
927 		}
928 	}
929 
930 	uv_bios_init();
931 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
932 			    &sn_region_size, &system_serial_number);
933 	uv_rtc_init();
934 
935 	for_each_present_cpu(cpu) {
936 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
937 
938 		nid = cpu_to_node(cpu);
939 		/*
940 		 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
941 		 */
942 		uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
943 		uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
944 		uv_cpu_hub_info(cpu)->hub_revision = uv_hub_info->hub_revision;
945 
946 		uv_cpu_hub_info(cpu)->m_shift = 64 - m_val;
947 		uv_cpu_hub_info(cpu)->n_lshift = n_lshift;
948 
949 		pnode = uv_apicid_to_pnode(apicid);
950 		blade = boot_pnode_to_blade(pnode);
951 		lcpu = uv_blade_info[blade].nr_possible_cpus;
952 		uv_blade_info[blade].nr_possible_cpus++;
953 
954 		/* Any node on the blade, else will contain -1. */
955 		uv_blade_info[blade].memory_nid = nid;
956 
957 		uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
958 		uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
959 		uv_cpu_hub_info(cpu)->m_val = m_val;
960 		uv_cpu_hub_info(cpu)->n_val = n_val;
961 		uv_cpu_hub_info(cpu)->numa_blade_id = blade;
962 		uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
963 		uv_cpu_hub_info(cpu)->pnode = pnode;
964 		uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
965 		uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
966 		uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
967 		uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
968 		uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
969 		uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
970 		uv_node_to_blade[nid] = blade;
971 		uv_cpu_to_blade[cpu] = blade;
972 	}
973 
974 	/* Add blade/pnode info for nodes without cpus */
975 	for_each_online_node(nid) {
976 		if (uv_node_to_blade[nid] >= 0)
977 			continue;
978 		paddr = node_start_pfn(nid) << PAGE_SHIFT;
979 		pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
980 		blade = boot_pnode_to_blade(pnode);
981 		uv_node_to_blade[nid] = blade;
982 	}
983 
984 	map_gru_high(max_pnode);
985 	map_mmr_high(max_pnode);
986 	map_mmioh_high(min_pnode, max_pnode);
987 
988 	uv_nmi_setup();
989 	uv_cpu_init();
990 	uv_scir_register_cpu_notifier();
991 	proc_mkdir("sgi_uv", NULL);
992 
993 	/* register Legacy VGA I/O redirection handler */
994 	pci_register_set_vga_state(uv_set_vga_state);
995 
996 	/*
997 	 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
998 	 * EFI is not enabled in the kdump kernel.
999 	 */
1000 	if (is_kdump_kernel())
1001 		reboot_type = BOOT_ACPI;
1002 }
1003 
1004 apic_driver(apic_x2apic_uv_x);
1005