xref: /openbmc/linux/arch/x86/kernel/apic/x2apic_uv_x.c (revision f7777dcc)
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 /* BMC sets a bit this MMR non-zero before sending an NMI */
43 #define UVH_NMI_MMR				UVH_SCRATCH5
44 #define UVH_NMI_MMR_CLEAR			(UVH_NMI_MMR + 8)
45 #define UV_NMI_PENDING_MASK			(1UL << 63)
46 DEFINE_PER_CPU(unsigned long, cpu_last_nmi_count);
47 
48 DEFINE_PER_CPU(int, x2apic_extra_bits);
49 
50 #define PR_DEVEL(fmt, args...)	pr_devel("%s: " fmt, __func__, args)
51 
52 static enum uv_system_type uv_system_type;
53 static u64 gru_start_paddr, gru_end_paddr;
54 static u64 gru_dist_base, gru_first_node_paddr = -1LL, gru_last_node_paddr;
55 static u64 gru_dist_lmask, gru_dist_umask;
56 static union uvh_apicid uvh_apicid;
57 int uv_min_hub_revision_id;
58 EXPORT_SYMBOL_GPL(uv_min_hub_revision_id);
59 unsigned int uv_apicid_hibits;
60 EXPORT_SYMBOL_GPL(uv_apicid_hibits);
61 static DEFINE_SPINLOCK(uv_nmi_lock);
62 
63 static struct apic apic_x2apic_uv_x;
64 
65 static unsigned long __init uv_early_read_mmr(unsigned long addr)
66 {
67 	unsigned long val, *mmr;
68 
69 	mmr = early_ioremap(UV_LOCAL_MMR_BASE | addr, sizeof(*mmr));
70 	val = *mmr;
71 	early_iounmap(mmr, sizeof(*mmr));
72 	return val;
73 }
74 
75 static inline bool is_GRU_range(u64 start, u64 end)
76 {
77 	if (gru_dist_base) {
78 		u64 su = start & gru_dist_umask; /* upper (incl pnode) bits */
79 		u64 sl = start & gru_dist_lmask; /* base offset bits */
80 		u64 eu = end & gru_dist_umask;
81 		u64 el = end & gru_dist_lmask;
82 
83 		/* Must reside completely within a single GRU range */
84 		return (sl == gru_dist_base && el == gru_dist_base &&
85 			su >= gru_first_node_paddr &&
86 			su <= gru_last_node_paddr &&
87 			eu == su);
88 	} else {
89 		return start >= gru_start_paddr && end <= gru_end_paddr;
90 	}
91 }
92 
93 static bool uv_is_untracked_pat_range(u64 start, u64 end)
94 {
95 	return is_ISA_range(start, end) || is_GRU_range(start, end);
96 }
97 
98 static int __init early_get_pnodeid(void)
99 {
100 	union uvh_node_id_u node_id;
101 	union uvh_rh_gam_config_mmr_u  m_n_config;
102 	int pnode;
103 
104 	/* Currently, all blades have same revision number */
105 	node_id.v = uv_early_read_mmr(UVH_NODE_ID);
106 	m_n_config.v = uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR);
107 	uv_min_hub_revision_id = node_id.s.revision;
108 
109 	switch (node_id.s.part_number) {
110 	case UV2_HUB_PART_NUMBER:
111 	case UV2_HUB_PART_NUMBER_X:
112 		uv_min_hub_revision_id += UV2_HUB_REVISION_BASE - 1;
113 		break;
114 	case UV3_HUB_PART_NUMBER:
115 	case UV3_HUB_PART_NUMBER_X:
116 		uv_min_hub_revision_id += UV3_HUB_REVISION_BASE;
117 		break;
118 	}
119 
120 	uv_hub_info->hub_revision = uv_min_hub_revision_id;
121 	pnode = (node_id.s.node_id >> 1) & ((1 << m_n_config.s.n_skt) - 1);
122 	return pnode;
123 }
124 
125 static void __init early_get_apic_pnode_shift(void)
126 {
127 	uvh_apicid.v = uv_early_read_mmr(UVH_APICID);
128 	if (!uvh_apicid.v)
129 		/*
130 		 * Old bios, use default value
131 		 */
132 		uvh_apicid.s.pnode_shift = UV_APIC_PNODE_SHIFT;
133 }
134 
135 /*
136  * Add an extra bit as dictated by bios to the destination apicid of
137  * interrupts potentially passing through the UV HUB.  This prevents
138  * a deadlock between interrupts and IO port operations.
139  */
140 static void __init uv_set_apicid_hibit(void)
141 {
142 	union uv1h_lb_target_physical_apic_id_mask_u apicid_mask;
143 
144 	if (is_uv1_hub()) {
145 		apicid_mask.v =
146 			uv_early_read_mmr(UV1H_LB_TARGET_PHYSICAL_APIC_ID_MASK);
147 		uv_apicid_hibits =
148 			apicid_mask.s1.bit_enables & UV_APICID_HIBIT_MASK;
149 	}
150 }
151 
152 static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
153 {
154 	int pnodeid, is_uv1, is_uv2, is_uv3;
155 
156 	is_uv1 = !strcmp(oem_id, "SGI");
157 	is_uv2 = !strcmp(oem_id, "SGI2");
158 	is_uv3 = !strncmp(oem_id, "SGI3", 4);	/* there are varieties of UV3 */
159 	if (is_uv1 || is_uv2 || is_uv3) {
160 		uv_hub_info->hub_revision =
161 			(is_uv1 ? UV1_HUB_REVISION_BASE :
162 			(is_uv2 ? UV2_HUB_REVISION_BASE :
163 				  UV3_HUB_REVISION_BASE));
164 		pnodeid = early_get_pnodeid();
165 		early_get_apic_pnode_shift();
166 		x86_platform.is_untracked_pat_range =  uv_is_untracked_pat_range;
167 		x86_platform.nmi_init = uv_nmi_init;
168 		if (!strcmp(oem_table_id, "UVL"))
169 			uv_system_type = UV_LEGACY_APIC;
170 		else if (!strcmp(oem_table_id, "UVX"))
171 			uv_system_type = UV_X2APIC;
172 		else if (!strcmp(oem_table_id, "UVH")) {
173 			__this_cpu_write(x2apic_extra_bits,
174 				pnodeid << uvh_apicid.s.pnode_shift);
175 			uv_system_type = UV_NON_UNIQUE_APIC;
176 			uv_set_apicid_hibit();
177 			return 1;
178 		}
179 	}
180 	return 0;
181 }
182 
183 enum uv_system_type get_uv_system_type(void)
184 {
185 	return uv_system_type;
186 }
187 
188 int is_uv_system(void)
189 {
190 	return uv_system_type != UV_NONE;
191 }
192 EXPORT_SYMBOL_GPL(is_uv_system);
193 
194 DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
195 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
196 
197 struct uv_blade_info *uv_blade_info;
198 EXPORT_SYMBOL_GPL(uv_blade_info);
199 
200 short *uv_node_to_blade;
201 EXPORT_SYMBOL_GPL(uv_node_to_blade);
202 
203 short *uv_cpu_to_blade;
204 EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
205 
206 short uv_possible_blades;
207 EXPORT_SYMBOL_GPL(uv_possible_blades);
208 
209 unsigned long sn_rtc_cycles_per_second;
210 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
211 
212 static int uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
213 {
214 #ifdef CONFIG_SMP
215 	unsigned long val;
216 	int pnode;
217 
218 	pnode = uv_apicid_to_pnode(phys_apicid);
219 	phys_apicid |= uv_apicid_hibits;
220 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
221 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
222 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
223 	    APIC_DM_INIT;
224 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
225 
226 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
227 	    (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
228 	    ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
229 	    APIC_DM_STARTUP;
230 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
231 
232 	atomic_set(&init_deasserted, 1);
233 #endif
234 	return 0;
235 }
236 
237 static void uv_send_IPI_one(int cpu, int vector)
238 {
239 	unsigned long apicid;
240 	int pnode;
241 
242 	apicid = per_cpu(x86_cpu_to_apicid, cpu);
243 	pnode = uv_apicid_to_pnode(apicid);
244 	uv_hub_send_ipi(pnode, apicid, vector);
245 }
246 
247 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
248 {
249 	unsigned int cpu;
250 
251 	for_each_cpu(cpu, mask)
252 		uv_send_IPI_one(cpu, vector);
253 }
254 
255 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
256 {
257 	unsigned int this_cpu = smp_processor_id();
258 	unsigned int cpu;
259 
260 	for_each_cpu(cpu, mask) {
261 		if (cpu != this_cpu)
262 			uv_send_IPI_one(cpu, vector);
263 	}
264 }
265 
266 static void uv_send_IPI_allbutself(int vector)
267 {
268 	unsigned int this_cpu = smp_processor_id();
269 	unsigned int cpu;
270 
271 	for_each_online_cpu(cpu) {
272 		if (cpu != this_cpu)
273 			uv_send_IPI_one(cpu, vector);
274 	}
275 }
276 
277 static void uv_send_IPI_all(int vector)
278 {
279 	uv_send_IPI_mask(cpu_online_mask, vector);
280 }
281 
282 static int uv_apic_id_valid(int apicid)
283 {
284 	return 1;
285 }
286 
287 static int uv_apic_id_registered(void)
288 {
289 	return 1;
290 }
291 
292 static void uv_init_apic_ldr(void)
293 {
294 }
295 
296 static int
297 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
298 			  const struct cpumask *andmask,
299 			  unsigned int *apicid)
300 {
301 	int unsigned cpu;
302 
303 	/*
304 	 * We're using fixed IRQ delivery, can only return one phys APIC ID.
305 	 * May as well be the first.
306 	 */
307 	for_each_cpu_and(cpu, cpumask, andmask) {
308 		if (cpumask_test_cpu(cpu, cpu_online_mask))
309 			break;
310 	}
311 
312 	if (likely(cpu < nr_cpu_ids)) {
313 		*apicid = per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
314 		return 0;
315 	}
316 
317 	return -EINVAL;
318 }
319 
320 static unsigned int x2apic_get_apic_id(unsigned long x)
321 {
322 	unsigned int id;
323 
324 	WARN_ON(preemptible() && num_online_cpus() > 1);
325 	id = x | __this_cpu_read(x2apic_extra_bits);
326 
327 	return id;
328 }
329 
330 static unsigned long set_apic_id(unsigned int id)
331 {
332 	unsigned long x;
333 
334 	/* maskout x2apic_extra_bits ? */
335 	x = id;
336 	return x;
337 }
338 
339 static unsigned int uv_read_apic_id(void)
340 {
341 
342 	return x2apic_get_apic_id(apic_read(APIC_ID));
343 }
344 
345 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
346 {
347 	return uv_read_apic_id() >> index_msb;
348 }
349 
350 static void uv_send_IPI_self(int vector)
351 {
352 	apic_write(APIC_SELF_IPI, vector);
353 }
354 
355 static int uv_probe(void)
356 {
357 	return apic == &apic_x2apic_uv_x;
358 }
359 
360 static struct apic __refdata apic_x2apic_uv_x = {
361 
362 	.name				= "UV large system",
363 	.probe				= uv_probe,
364 	.acpi_madt_oem_check		= uv_acpi_madt_oem_check,
365 	.apic_id_valid			= uv_apic_id_valid,
366 	.apic_id_registered		= uv_apic_id_registered,
367 
368 	.irq_delivery_mode		= dest_Fixed,
369 	.irq_dest_mode			= 0, /* physical */
370 
371 	.target_cpus			= online_target_cpus,
372 	.disable_esr			= 0,
373 	.dest_logical			= APIC_DEST_LOGICAL,
374 	.check_apicid_used		= NULL,
375 	.check_apicid_present		= NULL,
376 
377 	.vector_allocation_domain	= default_vector_allocation_domain,
378 	.init_apic_ldr			= uv_init_apic_ldr,
379 
380 	.ioapic_phys_id_map		= NULL,
381 	.setup_apic_routing		= NULL,
382 	.multi_timer_check		= NULL,
383 	.cpu_present_to_apicid		= default_cpu_present_to_apicid,
384 	.apicid_to_cpu_present		= NULL,
385 	.setup_portio_remap		= NULL,
386 	.check_phys_apicid_present	= default_check_phys_apicid_present,
387 	.enable_apic_mode		= NULL,
388 	.phys_pkg_id			= uv_phys_pkg_id,
389 	.mps_oem_check			= NULL,
390 
391 	.get_apic_id			= x2apic_get_apic_id,
392 	.set_apic_id			= set_apic_id,
393 	.apic_id_mask			= 0xFFFFFFFFu,
394 
395 	.cpu_mask_to_apicid_and		= uv_cpu_mask_to_apicid_and,
396 
397 	.send_IPI_mask			= uv_send_IPI_mask,
398 	.send_IPI_mask_allbutself	= uv_send_IPI_mask_allbutself,
399 	.send_IPI_allbutself		= uv_send_IPI_allbutself,
400 	.send_IPI_all			= uv_send_IPI_all,
401 	.send_IPI_self			= uv_send_IPI_self,
402 
403 	.wakeup_secondary_cpu		= uv_wakeup_secondary,
404 	.trampoline_phys_low		= DEFAULT_TRAMPOLINE_PHYS_LOW,
405 	.trampoline_phys_high		= DEFAULT_TRAMPOLINE_PHYS_HIGH,
406 	.wait_for_init_deassert		= NULL,
407 	.smp_callin_clear_local_apic	= NULL,
408 	.inquire_remote_apic		= NULL,
409 
410 	.read				= native_apic_msr_read,
411 	.write				= native_apic_msr_write,
412 	.eoi_write			= native_apic_msr_eoi_write,
413 	.icr_read			= native_x2apic_icr_read,
414 	.icr_write			= native_x2apic_icr_write,
415 	.wait_icr_idle			= native_x2apic_wait_icr_idle,
416 	.safe_wait_icr_idle		= native_safe_x2apic_wait_icr_idle,
417 };
418 
419 static void set_x2apic_extra_bits(int pnode)
420 {
421 	__this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
422 }
423 
424 /*
425  * Called on boot cpu.
426  */
427 static __init int boot_pnode_to_blade(int pnode)
428 {
429 	int blade;
430 
431 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
432 		if (pnode == uv_blade_info[blade].pnode)
433 			return blade;
434 	BUG();
435 }
436 
437 struct redir_addr {
438 	unsigned long redirect;
439 	unsigned long alias;
440 };
441 
442 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
443 
444 static __initdata struct redir_addr redir_addrs[] = {
445 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
446 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
447 	{UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
448 };
449 
450 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
451 {
452 	union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
453 	union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
454 	int i;
455 
456 	for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
457 		alias.v = uv_read_local_mmr(redir_addrs[i].alias);
458 		if (alias.s.enable && alias.s.base == 0) {
459 			*size = (1UL << alias.s.m_alias);
460 			redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
461 			*base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
462 			return;
463 		}
464 	}
465 	*base = *size = 0;
466 }
467 
468 enum map_type {map_wb, map_uc};
469 
470 static __init void map_high(char *id, unsigned long base, int pshift,
471 			int bshift, int max_pnode, enum map_type map_type)
472 {
473 	unsigned long bytes, paddr;
474 
475 	paddr = base << pshift;
476 	bytes = (1UL << bshift) * (max_pnode + 1);
477 	if (!paddr) {
478 		pr_info("UV: Map %s_HI base address NULL\n", id);
479 		return;
480 	}
481 	pr_debug("UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr, paddr + bytes);
482 	if (map_type == map_uc)
483 		init_extra_mapping_uc(paddr, bytes);
484 	else
485 		init_extra_mapping_wb(paddr, bytes);
486 }
487 
488 static __init void map_gru_distributed(unsigned long c)
489 {
490 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
491 	u64 paddr;
492 	unsigned long bytes;
493 	int nid;
494 
495 	gru.v = c;
496 	/* only base bits 42:28 relevant in dist mode */
497 	gru_dist_base = gru.v & 0x000007fff0000000UL;
498 	if (!gru_dist_base) {
499 		pr_info("UV: Map GRU_DIST base address NULL\n");
500 		return;
501 	}
502 	bytes = 1UL << UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
503 	gru_dist_lmask = ((1UL << uv_hub_info->m_val) - 1) & ~(bytes - 1);
504 	gru_dist_umask = ~((1UL << uv_hub_info->m_val) - 1);
505 	gru_dist_base &= gru_dist_lmask; /* Clear bits above M */
506 	for_each_online_node(nid) {
507 		paddr = ((u64)uv_node_to_pnode(nid) << uv_hub_info->m_val) |
508 				gru_dist_base;
509 		init_extra_mapping_wb(paddr, bytes);
510 		gru_first_node_paddr = min(paddr, gru_first_node_paddr);
511 		gru_last_node_paddr = max(paddr, gru_last_node_paddr);
512 	}
513 	/* Save upper (63:M) bits of address only for is_GRU_range */
514 	gru_first_node_paddr &= gru_dist_umask;
515 	gru_last_node_paddr &= gru_dist_umask;
516 	pr_debug("UV: Map GRU_DIST base 0x%016llx  0x%016llx - 0x%016llx\n",
517 		gru_dist_base, gru_first_node_paddr, gru_last_node_paddr);
518 }
519 
520 static __init void map_gru_high(int max_pnode)
521 {
522 	union uvh_rh_gam_gru_overlay_config_mmr_u gru;
523 	int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
524 
525 	gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
526 	if (!gru.s.enable) {
527 		pr_info("UV: GRU disabled\n");
528 		return;
529 	}
530 
531 	if (is_uv3_hub() && gru.s3.mode) {
532 		map_gru_distributed(gru.v);
533 		return;
534 	}
535 	map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
536 	gru_start_paddr = ((u64)gru.s.base << shift);
537 	gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
538 }
539 
540 static __init void map_mmr_high(int max_pnode)
541 {
542 	union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
543 	int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
544 
545 	mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
546 	if (mmr.s.enable)
547 		map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
548 	else
549 		pr_info("UV: MMR disabled\n");
550 }
551 
552 /*
553  * This commonality works because both 0 & 1 versions of the MMIOH OVERLAY
554  * and REDIRECT MMR regs are exactly the same on UV3.
555  */
556 struct mmioh_config {
557 	unsigned long overlay;
558 	unsigned long redirect;
559 	char *id;
560 };
561 
562 static __initdata struct mmioh_config mmiohs[] = {
563 	{
564 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR,
565 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR,
566 		"MMIOH0"
567 	},
568 	{
569 		UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG1_MMR,
570 		UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG1_MMR,
571 		"MMIOH1"
572 	},
573 };
574 
575 static __init void map_mmioh_high_uv3(int index, int min_pnode, int max_pnode)
576 {
577 	union uv3h_rh_gam_mmioh_overlay_config0_mmr_u overlay;
578 	unsigned long mmr;
579 	unsigned long base;
580 	int i, n, shift, m_io, max_io;
581 	int nasid, lnasid, fi, li;
582 	char *id;
583 
584 	id = mmiohs[index].id;
585 	overlay.v = uv_read_local_mmr(mmiohs[index].overlay);
586 	pr_info("UV: %s overlay 0x%lx base:0x%x m_io:%d\n",
587 		id, overlay.v, overlay.s3.base, overlay.s3.m_io);
588 	if (!overlay.s3.enable) {
589 		pr_info("UV: %s disabled\n", id);
590 		return;
591 	}
592 
593 	shift = UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR_BASE_SHFT;
594 	base = (unsigned long)overlay.s3.base;
595 	m_io = overlay.s3.m_io;
596 	mmr = mmiohs[index].redirect;
597 	n = UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR_DEPTH;
598 	min_pnode *= 2;				/* convert to NASID */
599 	max_pnode *= 2;
600 	max_io = lnasid = fi = li = -1;
601 
602 	for (i = 0; i < n; i++) {
603 		union uv3h_rh_gam_mmioh_redirect_config0_mmr_u redirect;
604 
605 		redirect.v = uv_read_local_mmr(mmr + i * 8);
606 		nasid = redirect.s3.nasid;
607 		if (nasid < min_pnode || max_pnode < nasid)
608 			nasid = -1;		/* invalid NASID */
609 
610 		if (nasid == lnasid) {
611 			li = i;
612 			if (i != n-1)		/* last entry check */
613 				continue;
614 		}
615 
616 		/* check if we have a cached (or last) redirect to print */
617 		if (lnasid != -1 || (i == n-1 && nasid != -1))  {
618 			unsigned long addr1, addr2;
619 			int f, l;
620 
621 			if (lnasid == -1) {
622 				f = l = i;
623 				lnasid = nasid;
624 			} else {
625 				f = fi;
626 				l = li;
627 			}
628 			addr1 = (base << shift) +
629 				f * (unsigned long)(1 << m_io);
630 			addr2 = (base << shift) +
631 				(l + 1) * (unsigned long)(1 << m_io);
632 			pr_info("UV: %s[%03d..%03d] NASID 0x%04x ADDR 0x%016lx - 0x%016lx\n",
633 				id, fi, li, lnasid, addr1, addr2);
634 			if (max_io < l)
635 				max_io = l;
636 		}
637 		fi = li = i;
638 		lnasid = nasid;
639 	}
640 
641 	pr_info("UV: %s base:0x%lx shift:%d M_IO:%d MAX_IO:%d\n",
642 		id, base, shift, m_io, max_io);
643 
644 	if (max_io >= 0)
645 		map_high(id, base, shift, m_io, max_io, map_uc);
646 }
647 
648 static __init void map_mmioh_high(int min_pnode, int max_pnode)
649 {
650 	union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
651 	unsigned long mmr, base;
652 	int shift, enable, m_io, n_io;
653 
654 	if (is_uv3_hub()) {
655 		/* Map both MMIOH Regions */
656 		map_mmioh_high_uv3(0, min_pnode, max_pnode);
657 		map_mmioh_high_uv3(1, min_pnode, max_pnode);
658 		return;
659 	}
660 
661 	if (is_uv1_hub()) {
662 		mmr = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
663 		shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
664 		mmioh.v = uv_read_local_mmr(mmr);
665 		enable = !!mmioh.s1.enable;
666 		base = mmioh.s1.base;
667 		m_io = mmioh.s1.m_io;
668 		n_io = mmioh.s1.n_io;
669 	} else if (is_uv2_hub()) {
670 		mmr = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
671 		shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
672 		mmioh.v = uv_read_local_mmr(mmr);
673 		enable = !!mmioh.s2.enable;
674 		base = mmioh.s2.base;
675 		m_io = mmioh.s2.m_io;
676 		n_io = mmioh.s2.n_io;
677 	} else
678 		return;
679 
680 	if (enable) {
681 		max_pnode &= (1 << n_io) - 1;
682 		pr_info(
683 		    "UV: base:0x%lx shift:%d N_IO:%d M_IO:%d max_pnode:0x%x\n",
684 			base, shift, m_io, n_io, max_pnode);
685 		map_high("MMIOH", base, shift, m_io, max_pnode, map_uc);
686 	} else {
687 		pr_info("UV: MMIOH disabled\n");
688 	}
689 }
690 
691 static __init void map_low_mmrs(void)
692 {
693 	init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
694 	init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
695 }
696 
697 static __init void uv_rtc_init(void)
698 {
699 	long status;
700 	u64 ticks_per_sec;
701 
702 	status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
703 					&ticks_per_sec);
704 	if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
705 		printk(KERN_WARNING
706 			"unable to determine platform RTC clock frequency, "
707 			"guessing.\n");
708 		/* BIOS gives wrong value for clock freq. so guess */
709 		sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
710 	} else
711 		sn_rtc_cycles_per_second = ticks_per_sec;
712 }
713 
714 /*
715  * percpu heartbeat timer
716  */
717 static void uv_heartbeat(unsigned long ignored)
718 {
719 	struct timer_list *timer = &uv_hub_info->scir.timer;
720 	unsigned char bits = uv_hub_info->scir.state;
721 
722 	/* flip heartbeat bit */
723 	bits ^= SCIR_CPU_HEARTBEAT;
724 
725 	/* is this cpu idle? */
726 	if (idle_cpu(raw_smp_processor_id()))
727 		bits &= ~SCIR_CPU_ACTIVITY;
728 	else
729 		bits |= SCIR_CPU_ACTIVITY;
730 
731 	/* update system controller interface reg */
732 	uv_set_scir_bits(bits);
733 
734 	/* enable next timer period */
735 	mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
736 }
737 
738 static void uv_heartbeat_enable(int cpu)
739 {
740 	while (!uv_cpu_hub_info(cpu)->scir.enabled) {
741 		struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
742 
743 		uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
744 		setup_timer(timer, uv_heartbeat, cpu);
745 		timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
746 		add_timer_on(timer, cpu);
747 		uv_cpu_hub_info(cpu)->scir.enabled = 1;
748 
749 		/* also ensure that boot cpu is enabled */
750 		cpu = 0;
751 	}
752 }
753 
754 #ifdef CONFIG_HOTPLUG_CPU
755 static void uv_heartbeat_disable(int cpu)
756 {
757 	if (uv_cpu_hub_info(cpu)->scir.enabled) {
758 		uv_cpu_hub_info(cpu)->scir.enabled = 0;
759 		del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
760 	}
761 	uv_set_cpu_scir_bits(cpu, 0xff);
762 }
763 
764 /*
765  * cpu hotplug notifier
766  */
767 static int uv_scir_cpu_notify(struct notifier_block *self, unsigned long action,
768 			      void *hcpu)
769 {
770 	long cpu = (long)hcpu;
771 
772 	switch (action) {
773 	case CPU_ONLINE:
774 		uv_heartbeat_enable(cpu);
775 		break;
776 	case CPU_DOWN_PREPARE:
777 		uv_heartbeat_disable(cpu);
778 		break;
779 	default:
780 		break;
781 	}
782 	return NOTIFY_OK;
783 }
784 
785 static __init void uv_scir_register_cpu_notifier(void)
786 {
787 	hotcpu_notifier(uv_scir_cpu_notify, 0);
788 }
789 
790 #else /* !CONFIG_HOTPLUG_CPU */
791 
792 static __init void uv_scir_register_cpu_notifier(void)
793 {
794 }
795 
796 static __init int uv_init_heartbeat(void)
797 {
798 	int cpu;
799 
800 	if (is_uv_system())
801 		for_each_online_cpu(cpu)
802 			uv_heartbeat_enable(cpu);
803 	return 0;
804 }
805 
806 late_initcall(uv_init_heartbeat);
807 
808 #endif /* !CONFIG_HOTPLUG_CPU */
809 
810 /* Direct Legacy VGA I/O traffic to designated IOH */
811 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
812 		      unsigned int command_bits, u32 flags)
813 {
814 	int domain, bus, rc;
815 
816 	PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
817 			pdev->devfn, decode, command_bits, flags);
818 
819 	if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
820 		return 0;
821 
822 	if ((command_bits & PCI_COMMAND_IO) == 0)
823 		return 0;
824 
825 	domain = pci_domain_nr(pdev->bus);
826 	bus = pdev->bus->number;
827 
828 	rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
829 	PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
830 
831 	return rc;
832 }
833 
834 /*
835  * Called on each cpu to initialize the per_cpu UV data area.
836  * FIXME: hotplug not supported yet
837  */
838 void uv_cpu_init(void)
839 {
840 	/* CPU 0 initilization will be done via uv_system_init. */
841 	if (!uv_blade_info)
842 		return;
843 
844 	uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
845 
846 	if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
847 		set_x2apic_extra_bits(uv_hub_info->pnode);
848 }
849 
850 /*
851  * When NMI is received, print a stack trace.
852  */
853 int uv_handle_nmi(unsigned int reason, struct pt_regs *regs)
854 {
855 	unsigned long real_uv_nmi;
856 	int bid;
857 
858 	/*
859 	 * Each blade has an MMR that indicates when an NMI has been sent
860 	 * to cpus on the blade. If an NMI is detected, atomically
861 	 * clear the MMR and update a per-blade NMI count used to
862 	 * cause each cpu on the blade to notice a new NMI.
863 	 */
864 	bid = uv_numa_blade_id();
865 	real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
866 
867 	if (unlikely(real_uv_nmi)) {
868 		spin_lock(&uv_blade_info[bid].nmi_lock);
869 		real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
870 		if (real_uv_nmi) {
871 			uv_blade_info[bid].nmi_count++;
872 			uv_write_local_mmr(UVH_NMI_MMR_CLEAR, UV_NMI_PENDING_MASK);
873 		}
874 		spin_unlock(&uv_blade_info[bid].nmi_lock);
875 	}
876 
877 	if (likely(__get_cpu_var(cpu_last_nmi_count) == uv_blade_info[bid].nmi_count))
878 		return NMI_DONE;
879 
880 	__get_cpu_var(cpu_last_nmi_count) = uv_blade_info[bid].nmi_count;
881 
882 	/*
883 	 * Use a lock so only one cpu prints at a time.
884 	 * This prevents intermixed output.
885 	 */
886 	spin_lock(&uv_nmi_lock);
887 	pr_info("UV NMI stack dump cpu %u:\n", smp_processor_id());
888 	dump_stack();
889 	spin_unlock(&uv_nmi_lock);
890 
891 	return NMI_HANDLED;
892 }
893 
894 void uv_register_nmi_notifier(void)
895 {
896 	if (register_nmi_handler(NMI_UNKNOWN, uv_handle_nmi, 0, "uv"))
897 		printk(KERN_WARNING "UV NMI handler failed to register\n");
898 }
899 
900 void uv_nmi_init(void)
901 {
902 	unsigned int value;
903 
904 	/*
905 	 * Unmask NMI on all cpus
906 	 */
907 	value = apic_read(APIC_LVT1) | APIC_DM_NMI;
908 	value &= ~APIC_LVT_MASKED;
909 	apic_write(APIC_LVT1, value);
910 }
911 
912 void __init uv_system_init(void)
913 {
914 	union uvh_rh_gam_config_mmr_u  m_n_config;
915 	union uvh_node_id_u node_id;
916 	unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
917 	int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val;
918 	int gnode_extra, min_pnode = 999999, max_pnode = -1;
919 	unsigned long mmr_base, present, paddr;
920 	unsigned short pnode_mask;
921 	char *hub = (is_uv1_hub() ? "UV1" :
922 		    (is_uv2_hub() ? "UV2" :
923 				    "UV3"));
924 
925 	pr_info("UV: Found %s hub\n", hub);
926 	map_low_mmrs();
927 
928 	m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
929 	m_val = m_n_config.s.m_skt;
930 	n_val = m_n_config.s.n_skt;
931 	pnode_mask = (1 << n_val) - 1;
932 	mmr_base =
933 	    uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
934 	    ~UV_MMR_ENABLE;
935 
936 	node_id.v = uv_read_local_mmr(UVH_NODE_ID);
937 	gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
938 	gnode_upper = ((unsigned long)gnode_extra  << m_val);
939 	pr_info("UV: N:%d M:%d pnode_mask:0x%x gnode_upper/extra:0x%lx/0x%x\n",
940 			n_val, m_val, pnode_mask, gnode_upper, gnode_extra);
941 
942 	pr_info("UV: global MMR base 0x%lx\n", mmr_base);
943 
944 	for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
945 		uv_possible_blades +=
946 		  hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
947 
948 	/* uv_num_possible_blades() is really the hub count */
949 	pr_info("UV: Found %d blades, %d hubs\n",
950 			is_uv1_hub() ? uv_num_possible_blades() :
951 			(uv_num_possible_blades() + 1) / 2,
952 			uv_num_possible_blades());
953 
954 	bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
955 	uv_blade_info = kzalloc(bytes, GFP_KERNEL);
956 	BUG_ON(!uv_blade_info);
957 
958 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
959 		uv_blade_info[blade].memory_nid = -1;
960 
961 	get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
962 
963 	bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
964 	uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
965 	BUG_ON(!uv_node_to_blade);
966 	memset(uv_node_to_blade, 255, bytes);
967 
968 	bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
969 	uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
970 	BUG_ON(!uv_cpu_to_blade);
971 	memset(uv_cpu_to_blade, 255, bytes);
972 
973 	blade = 0;
974 	for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
975 		present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
976 		for (j = 0; j < 64; j++) {
977 			if (!test_bit(j, &present))
978 				continue;
979 			pnode = (i * 64 + j) & pnode_mask;
980 			uv_blade_info[blade].pnode = pnode;
981 			uv_blade_info[blade].nr_possible_cpus = 0;
982 			uv_blade_info[blade].nr_online_cpus = 0;
983 			spin_lock_init(&uv_blade_info[blade].nmi_lock);
984 			min_pnode = min(pnode, min_pnode);
985 			max_pnode = max(pnode, max_pnode);
986 			blade++;
987 		}
988 	}
989 
990 	uv_bios_init();
991 	uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
992 			    &sn_region_size, &system_serial_number);
993 	uv_rtc_init();
994 
995 	for_each_present_cpu(cpu) {
996 		int apicid = per_cpu(x86_cpu_to_apicid, cpu);
997 
998 		nid = cpu_to_node(cpu);
999 		/*
1000 		 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
1001 		 */
1002 		uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
1003 		uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
1004 		uv_cpu_hub_info(cpu)->hub_revision = uv_hub_info->hub_revision;
1005 
1006 		uv_cpu_hub_info(cpu)->m_shift = 64 - m_val;
1007 		uv_cpu_hub_info(cpu)->n_lshift = is_uv2_1_hub() ?
1008 				(m_val == 40 ? 40 : 39) : m_val;
1009 
1010 		pnode = uv_apicid_to_pnode(apicid);
1011 		blade = boot_pnode_to_blade(pnode);
1012 		lcpu = uv_blade_info[blade].nr_possible_cpus;
1013 		uv_blade_info[blade].nr_possible_cpus++;
1014 
1015 		/* Any node on the blade, else will contain -1. */
1016 		uv_blade_info[blade].memory_nid = nid;
1017 
1018 		uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
1019 		uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
1020 		uv_cpu_hub_info(cpu)->m_val = m_val;
1021 		uv_cpu_hub_info(cpu)->n_val = n_val;
1022 		uv_cpu_hub_info(cpu)->numa_blade_id = blade;
1023 		uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
1024 		uv_cpu_hub_info(cpu)->pnode = pnode;
1025 		uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
1026 		uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
1027 		uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
1028 		uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
1029 		uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
1030 		uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
1031 		uv_node_to_blade[nid] = blade;
1032 		uv_cpu_to_blade[cpu] = blade;
1033 	}
1034 
1035 	/* Add blade/pnode info for nodes without cpus */
1036 	for_each_online_node(nid) {
1037 		if (uv_node_to_blade[nid] >= 0)
1038 			continue;
1039 		paddr = node_start_pfn(nid) << PAGE_SHIFT;
1040 		pnode = uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr));
1041 		blade = boot_pnode_to_blade(pnode);
1042 		uv_node_to_blade[nid] = blade;
1043 	}
1044 
1045 	map_gru_high(max_pnode);
1046 	map_mmr_high(max_pnode);
1047 	map_mmioh_high(min_pnode, max_pnode);
1048 
1049 	uv_cpu_init();
1050 	uv_scir_register_cpu_notifier();
1051 	uv_register_nmi_notifier();
1052 	proc_mkdir("sgi_uv", NULL);
1053 
1054 	/* register Legacy VGA I/O redirection handler */
1055 	pci_register_set_vga_state(uv_set_vga_state);
1056 
1057 	/*
1058 	 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
1059 	 * EFI is not enabled in the kdump kernel.
1060 	 */
1061 	if (is_kdump_kernel())
1062 		reboot_type = BOOT_ACPI;
1063 }
1064 
1065 apic_driver(apic_x2apic_uv_x);
1066