xref: /openbmc/linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision 2596e07a)
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			= uv_send_IPI_one,
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 	.inquire_remote_apic		= NULL,
418 
419 	.read				= native_apic_msr_read,
420 	.write				= native_apic_msr_write,
421 	.eoi_write			= native_apic_msr_eoi_write,
422 	.icr_read			= native_x2apic_icr_read,
423 	.icr_write			= native_x2apic_icr_write,
424 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
425 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
426 };
427 
428 static void set_x2apic_extra_bits(int pnode)
429 {
430 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
431 }
432 
433 /*
434  * Called on boot cpu.
435  */
436 static __init int boot_pnode_to_blade(int pnode)
437 {
438 	int blade;
439 
440 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
441 		if (pnode == uv_blade_info[blade].pnode)
442 			return blade;
443 	BUG();
444 }
445 
446 struct redir_addr {
447 	unsigned long redirect;
448 	unsigned long alias;
449 };
450 
451 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
452 
453 static __initdata struct redir_addr redir_addrs[] = {
454 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
455 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
456 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
457 };
458 
459 static unsigned char get_n_lshift(int m_val)
460 {
461 	union uv3h_gr0_gam_gr_config_u m_gr_config;
462 
463 	if (is_uv1_hub())
464 		return m_val;
465 
466 	if (is_uv2_hub())
467 		return m_val == 40 ? 40 : 39;
468 
469 	m_gr_config.v = uv_read_local_mmr(UV3H_GR0_GAM_GR_CONFIG);
470 	return m_gr_config.s3.m_skt;
471 }
472 
473 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
474 {
475 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
476 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
477 	int i;
478 
479 	for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
480 		alias.v = uv_read_local_mmr(redir_addrs[i].alias);
481 		if (alias.s.enable && alias.s.base == 0) {
482 			*size = (1UL << alias.s.m_alias);
483 			redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
484 			*base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
485 			return;
486 		}
487 	}
488 	*base = *size = 0;
489 }
490 
491 enum map_type {map_wb, map_uc};
492 
493 static __init void map_high(char *id, unsigned long base, int pshift,
494 			int bshift, int max_pnode, enum map_type map_type)
495 {
496 	unsigned long bytes, paddr;
497 
498 	paddr = base << pshift;
499 	bytes = (1UL << bshift) * (max_pnode + 1);
500 	if (!paddr) {
501 		pr_info("UV: Map %s_HI base address NULL\n", id);
502 		return;
503 	}
504 	pr_debug("UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr, paddr + bytes);
505 	if (map_type == map_uc)
506 		init_extra_mapping_uc(paddr, bytes);
507 	else
508 		init_extra_mapping_wb(paddr, bytes);
509 }
510 
511 static __init void map_gru_distributed(unsigned long c)
512 {
513 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
514 	u64 paddr;
515 	unsigned long bytes;
516 	int nid;
517 
518 	gru.v = c;
519 	/* only base bits 42:28 relevant in dist mode */
520 	gru_dist_base = gru.v & 0x000007fff0000000UL;
521 	if (!gru_dist_base) {
522 		pr_info("UV: Map GRU_DIST base address NULL\n");
523 		return;
524 	}
525 	bytes = 1UL << UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
526 	gru_dist_lmask = ((1UL << uv_hub_info->m_val) - 1) & ~(bytes - 1);
527 	gru_dist_umask = ~((1UL << uv_hub_info->m_val) - 1);
528 	gru_dist_base &= gru_dist_lmask; /* Clear bits above M */
529 	for_each_online_node(nid) {
530 		paddr = ((u64)uv_node_to_pnode(nid) << uv_hub_info->m_val) |
531 				gru_dist_base;
532 		init_extra_mapping_wb(paddr, bytes);
533 		gru_first_node_paddr = min(paddr, gru_first_node_paddr);
534 		gru_last_node_paddr = max(paddr, gru_last_node_paddr);
535 	}
536 	/* Save upper (63:M) bits of address only for is_GRU_range */
537 	gru_first_node_paddr &= gru_dist_umask;
538 	gru_last_node_paddr &= gru_dist_umask;
539 	pr_debug("UV: Map GRU_DIST base 0x%016llx  0x%016llx - 0x%016llx\n",
540 		gru_dist_base, gru_first_node_paddr, gru_last_node_paddr);
541 }
542 
543 static __init void map_gru_high(int max_pnode)
544 {
545 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
546 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
547 
548 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
549 	if (!gru.s.enable) {
550 		pr_info("UV: GRU disabled\n");
551 		return;
552 	}
553 
554 	if (is_uv3_hub() && gru.s3.mode) {
555 		map_gru_distributed(gru.v);
556 		return;
557 	}
558 	map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
559 	gru_start_paddr = ((u64)gru.s.base << shift);
560 	gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
561 }
562 
563 static __init void map_mmr_high(int max_pnode)
564 {
565 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
566 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
567 
568 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
569 	if (mmr.s.enable)
570 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
571 	else
572 		pr_info("UV: MMR disabled\n");
573 }
574 
575 /*
576  * This commonality works because both 0 & 1 versions of the MMIOH OVERLAY
577  * and REDIRECT MMR regs are exactly the same on UV3.
578  */
579 struct mmioh_config {
580 	unsigned long overlay;
581 	unsigned long redirect;
582 	char *id;
583 };
584 
585 static __initdata struct mmioh_config mmiohs[] = {
586 	{
587 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR,
588 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR,
589 		"MMIOH0"
590 	},
591 	{
592 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG1_MMR,
593 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG1_MMR,
594 		"MMIOH1"
595 	},
596 };
597 
598 static __init void map_mmioh_high_uv3(int index, int min_pnode, int max_pnode)
599 {
600 	union uv3h_rh_gam_mmioh_overlay_config0_mmr_u overlay;
601 	unsigned long mmr;
602 	unsigned long base;
603 	int i, n, shift, m_io, max_io;
604 	int nasid, lnasid, fi, li;
605 	char *id;
606 
607 	id = mmiohs[index].id;
608 	overlay.v = uv_read_local_mmr(mmiohs[index].overlay);
609 	pr_info("UV: %s overlay 0x%lx base:0x%x m_io:%d\n",
610 		id, overlay.v, overlay.s3.base, overlay.s3.m_io);
611 	if (!overlay.s3.enable) {
612 		pr_info("UV: %s disabled\n", id);
613 		return;
614 	}
615 
616 	shift = UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR_BASE_SHFT;
617 	base = (unsigned long)overlay.s3.base;
618 	m_io = overlay.s3.m_io;
619 	mmr = mmiohs[index].redirect;
620 	n = UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR_DEPTH;
621 	min_pnode *= 2;				/* convert to NASID */
622 	max_pnode *= 2;
623 	max_io = lnasid = fi = li = -1;
624 
625 	for (i = 0; i < n; i++) {
626 		union uv3h_rh_gam_mmioh_redirect_config0_mmr_u redirect;
627 
628 		redirect.v = uv_read_local_mmr(mmr + i * 8);
629 		nasid = redirect.s3.nasid;
630 		if (nasid < min_pnode || max_pnode < nasid)
631 			nasid = -1;		/* invalid NASID */
632 
633 		if (nasid == lnasid) {
634 			li = i;
635 			if (i != n-1)		/* last entry check */
636 				continue;
637 		}
638 
639 		/* check if we have a cached (or last) redirect to print */
640 		if (lnasid != -1 || (i == n-1 && nasid != -1))  {
641 			unsigned long addr1, addr2;
642 			int f, l;
643 
644 			if (lnasid == -1) {
645 				f = l = i;
646 				lnasid = nasid;
647 			} else {
648 				f = fi;
649 				l = li;
650 			}
651 			addr1 = (base << shift) +
652 				f * (unsigned long)(1 << m_io);
653 			addr2 = (base << shift) +
654 				(l + 1) * (unsigned long)(1 << m_io);
655 			pr_info("UV: %s[%03d..%03d] NASID 0x%04x ADDR 0x%016lx - 0x%016lx\n",
656 				id, fi, li, lnasid, addr1, addr2);
657 			if (max_io < l)
658 				max_io = l;
659 		}
660 		fi = li = i;
661 		lnasid = nasid;
662 	}
663 
664 	pr_info("UV: %s base:0x%lx shift:%d M_IO:%d MAX_IO:%d\n",
665 		id, base, shift, m_io, max_io);
666 
667 	if (max_io >= 0)
668 		map_high(id, base, shift, m_io, max_io, map_uc);
669 }
670 
671 static __init void map_mmioh_high(int min_pnode, int max_pnode)
672 {
673 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
674 	unsigned long mmr, base;
675 	int shift, enable, m_io, n_io;
676 
677 	if (is_uv3_hub()) {
678 		/* Map both MMIOH Regions */
679 		map_mmioh_high_uv3(0, min_pnode, max_pnode);
680 		map_mmioh_high_uv3(1, min_pnode, max_pnode);
681 		return;
682 	}
683 
684 	if (is_uv1_hub()) {
685 		mmr = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
686 		shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
687 		mmioh.v = uv_read_local_mmr(mmr);
688 		enable = !!mmioh.s1.enable;
689 		base = mmioh.s1.base;
690 		m_io = mmioh.s1.m_io;
691 		n_io = mmioh.s1.n_io;
692 	} else if (is_uv2_hub()) {
693 		mmr = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
694 		shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
695 		mmioh.v = uv_read_local_mmr(mmr);
696 		enable = !!mmioh.s2.enable;
697 		base = mmioh.s2.base;
698 		m_io = mmioh.s2.m_io;
699 		n_io = mmioh.s2.n_io;
700 	} else
701 		return;
702 
703 	if (enable) {
704 		max_pnode &= (1 << n_io) - 1;
705 		pr_info(
706 		    "UV: base:0x%lx shift:%d N_IO:%d M_IO:%d max_pnode:0x%x\n",
707 			base, shift, m_io, n_io, max_pnode);
708 		map_high("MMIOH", base, shift, m_io, max_pnode, map_uc);
709 	} else {
710 		pr_info("UV: MMIOH disabled\n");
711 	}
712 }
713 
714 static __init void map_low_mmrs(void)
715 {
716 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
717 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
718 }
719 
720 static __init void uv_rtc_init(void)
721 {
722 	long status;
723 	u64 ticks_per_sec;
724 
725 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
726 					&ticks_per_sec);
727 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
728 		printk(KERN_WARNING
729 			"unable to determine platform RTC clock frequency, "
730 			"guessing.\n");
731 		/* BIOS gives wrong value for clock freq. so guess */
732 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
733 	} else
734 		sn_rtc_cycles_per_second = ticks_per_sec;
735 }
736 
737 /*
738  * percpu heartbeat timer
739  */
740 static void uv_heartbeat(unsigned long ignored)
741 {
742 	struct timer_list *timer = &uv_hub_info->scir.timer;
743 	unsigned char bits = uv_hub_info->scir.state;
744 
745 	/* flip heartbeat bit */
746 	bits ^= SCIR_CPU_HEARTBEAT;
747 
748 	/* is this cpu idle? */
749 	if (idle_cpu(raw_smp_processor_id()))
750 		bits &= ~SCIR_CPU_ACTIVITY;
751 	else
752 		bits |= SCIR_CPU_ACTIVITY;
753 
754 	/* update system controller interface reg */
755 	uv_set_scir_bits(bits);
756 
757 	/* enable next timer period */
758 	mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
759 }
760 
761 static void uv_heartbeat_enable(int cpu)
762 {
763 	while (!uv_cpu_hub_info(cpu)->scir.enabled) {
764 		struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
765 
766 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
767 		setup_timer(timer, uv_heartbeat, cpu);
768 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
769 		add_timer_on(timer, cpu);
770 		uv_cpu_hub_info(cpu)->scir.enabled = 1;
771 
772 		/* also ensure that boot cpu is enabled */
773 		cpu = 0;
774 	}
775 }
776 
777 #ifdef CONFIG_HOTPLUG_CPU
778 static void uv_heartbeat_disable(int cpu)
779 {
780 	if (uv_cpu_hub_info(cpu)->scir.enabled) {
781 		uv_cpu_hub_info(cpu)->scir.enabled = 0;
782 		del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
783 	}
784 	uv_set_cpu_scir_bits(cpu, 0xff);
785 }
786 
787 /*
788  * cpu hotplug notifier
789  */
790 static int uv_scir_cpu_notify(struct notifier_block *self, unsigned long action,
791 			      void *hcpu)
792 {
793 	long cpu = (long)hcpu;
794 
795 	switch (action) {
796 	case CPU_ONLINE:
797 		uv_heartbeat_enable(cpu);
798 		break;
799 	case CPU_DOWN_PREPARE:
800 		uv_heartbeat_disable(cpu);
801 		break;
802 	default:
803 		break;
804 	}
805 	return NOTIFY_OK;
806 }
807 
808 static __init void uv_scir_register_cpu_notifier(void)
809 {
810 	hotcpu_notifier(uv_scir_cpu_notify, 0);
811 }
812 
813 #else /* !CONFIG_HOTPLUG_CPU */
814 
815 static __init void uv_scir_register_cpu_notifier(void)
816 {
817 }
818 
819 static __init int uv_init_heartbeat(void)
820 {
821 	int cpu;
822 
823 	if (is_uv_system())
824 		for_each_online_cpu(cpu)
825 			uv_heartbeat_enable(cpu);
826 	return 0;
827 }
828 
829 late_initcall(uv_init_heartbeat);
830 
831 #endif /* !CONFIG_HOTPLUG_CPU */
832 
833 /* Direct Legacy VGA I/O traffic to designated IOH */
834 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
835 		      unsigned int command_bits, u32 flags)
836 {
837 	int domain, bus, rc;
838 
839 	PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
840 			pdev->devfn, decode, command_bits, flags);
841 
842 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
843 		return 0;
844 
845 	if ((command_bits & PCI_COMMAND_IO) == 0)
846 		return 0;
847 
848 	domain = pci_domain_nr(pdev->bus);
849 	bus = pdev->bus->number;
850 
851 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
852 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
853 
854 	return rc;
855 }
856 
857 /*
858  * Called on each cpu to initialize the per_cpu UV data area.
859  * FIXME: hotplug not supported yet
860  */
861 void uv_cpu_init(void)
862 {
863 	/* CPU 0 initilization will be done via uv_system_init. */
864 	if (!uv_blade_info)
865 		return;
866 
867 	uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
868 
869 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
870 		set_x2apic_extra_bits(uv_hub_info->pnode);
871 }
872 
873 void __init uv_system_init(void)
874 {
875 	union uvh_rh_gam_config_mmr_u  m_n_config;
876 	union uvh_node_id_u node_id;
877 	unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
878 	int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val;
879 	int gnode_extra, min_pnode = 999999, max_pnode = -1;
880 	unsigned long mmr_base, present, paddr;
881 	unsigned short pnode_mask;
882 	unsigned char n_lshift;
883 	char *hub = (is_uv1_hub() ? "UV100/1000" :
884 		    (is_uv2_hub() ? "UV2000/3000" :
885 		    (is_uv3_hub() ? "UV300" : NULL)));
886 
887 	if (!hub) {
888 		pr_err("UV: Unknown/unsupported UV hub\n");
889 		return;
890 	}
891 	pr_info("UV: Found %s hub\n", hub);
892 
893 	/* We now only need to map the MMRs on UV1 */
894 	if (is_uv1_hub())
895 		map_low_mmrs();
896 
897 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
898 	m_val = m_n_config.s.m_skt;
899 	n_val = m_n_config.s.n_skt;
900 	pnode_mask = (1 << n_val) - 1;
901 	n_lshift = get_n_lshift(m_val);
902 	mmr_base =
903 	    uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
904 	    ~UV_MMR_ENABLE;
905 
906 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
907 	gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
908 	gnode_upper = ((unsigned long)gnode_extra  << m_val);
909 	pr_info("UV: N:%d M:%d pnode_mask:0x%x gnode_upper/extra:0x%lx/0x%x n_lshift 0x%x\n",
910 			n_val, m_val, pnode_mask, gnode_upper, gnode_extra,
911 			n_lshift);
912 
913 	pr_info("UV: global MMR base 0x%lx\n", mmr_base);
914 
915 	for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
916 		uv_possible_blades +=
917 		  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
918 
919 	/* uv_num_possible_blades() is really the hub count */
920 	pr_info("UV: Found %d blades, %d hubs\n",
921 			is_uv1_hub() ? uv_num_possible_blades() :
922 			(uv_num_possible_blades() + 1) / 2,
923 			uv_num_possible_blades());
924 
925 	bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
926 	uv_blade_info = kzalloc(bytes, GFP_KERNEL);
927 	BUG_ON(!uv_blade_info);
928 
929 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
930 		uv_blade_info[blade].memory_nid = -1;
931 
932 	get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
933 
934 	bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
935 	uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
936 	BUG_ON(!uv_node_to_blade);
937 	memset(uv_node_to_blade, 255, bytes);
938 
939 	bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
940 	uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
941 	BUG_ON(!uv_cpu_to_blade);
942 	memset(uv_cpu_to_blade, 255, bytes);
943 
944 	blade = 0;
945 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
946 		present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
947 		for (j = 0; j < 64; j++) {
948 			if (!test_bit(j, &present))
949 				continue;
950 			pnode = (i * 64 + j) & pnode_mask;
951 			uv_blade_info[blade].pnode = pnode;
952 			uv_blade_info[blade].nr_possible_cpus = 0;
953 			uv_blade_info[blade].nr_online_cpus = 0;
954 			spin_lock_init(&uv_blade_info[blade].nmi_lock);
955 			min_pnode = min(pnode, min_pnode);
956 			max_pnode = max(pnode, max_pnode);
957 			blade++;
958 		}
959 	}
960 
961 	uv_bios_init();
962 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
963 			    &sn_region_size, &system_serial_number);
964 	uv_rtc_init();
965 
966 	for_each_present_cpu(cpu) {
967 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
968 
969 		nid = cpu_to_node(cpu);
970 		/*
971 		 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
972 		 */
973 		uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
974 		uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
975 		uv_cpu_hub_info(cpu)->hub_revision = uv_hub_info->hub_revision;
976 
977 		uv_cpu_hub_info(cpu)->m_shift = 64 - m_val;
978 		uv_cpu_hub_info(cpu)->n_lshift = n_lshift;
979 
980 		pnode = uv_apicid_to_pnode(apicid);
981 		blade = boot_pnode_to_blade(pnode);
982 		lcpu = uv_blade_info[blade].nr_possible_cpus;
983 		uv_blade_info[blade].nr_possible_cpus++;
984 
985 		/* Any node on the blade, else will contain -1. */
986 		uv_blade_info[blade].memory_nid = nid;
987 
988 		uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
989 		uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
990 		uv_cpu_hub_info(cpu)->m_val = m_val;
991 		uv_cpu_hub_info(cpu)->n_val = n_val;
992 		uv_cpu_hub_info(cpu)->numa_blade_id = blade;
993 		uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
994 		uv_cpu_hub_info(cpu)->pnode = pnode;
995 		uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
996 		uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
997 		uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
998 		uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
999 		uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
1000 		uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
1001 		uv_node_to_blade[nid] = blade;
1002 		uv_cpu_to_blade[cpu] = blade;
1003 	}
1004 
1005 	/* Add blade/pnode info for nodes without cpus */
1006 	for_each_online_node(nid) {
1007 		if (uv_node_to_blade[nid] >= 0)
1008 			continue;
1009 		paddr = node_start_pfn(nid) << PAGE_SHIFT;
1010 		pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
1011 		blade = boot_pnode_to_blade(pnode);
1012 		uv_node_to_blade[nid] = blade;
1013 	}
1014 
1015 	map_gru_high(max_pnode);
1016 	map_mmr_high(max_pnode);
1017 	map_mmioh_high(min_pnode, max_pnode);
1018 
1019 	uv_nmi_setup();
1020 	uv_cpu_init();
1021 	uv_scir_register_cpu_notifier();
1022 	proc_mkdir("sgi_uv", NULL);
1023 
1024 	/* register Legacy VGA I/O redirection handler */
1025 	pci_register_set_vga_state(uv_set_vga_state);
1026 
1027 	/*
1028 	 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
1029 	 * EFI is not enabled in the kdump kernel.
1030 	 */
1031 	if (is_kdump_kernel())
1032 		reboot_type = BOOT_ACPI;
1033 }
1034 
1035 apic_driver(apic_x2apic_uv_x);
1036