Merge branch 'master' into next
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / arch / x86 / kernel / apic / x2apic_uv_x.c
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) 2007-2010 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
29 #include <asm/uv/uv_mmrs.h>
30 #include <asm/uv/uv_hub.h>
31 #include <asm/current.h>
32 #include <asm/pgtable.h>
33 #include <asm/uv/bios.h>
34 #include <asm/uv/uv.h>
35 #include <asm/apic.h>
36 #include <asm/ipi.h>
37 #include <asm/smp.h>
38 #include <asm/x86_init.h>
39 #include <asm/emergency-restart.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 union uvh_apicid uvh_apicid;
55 int uv_min_hub_revision_id;
56 EXPORT_SYMBOL_GPL(uv_min_hub_revision_id);
57 unsigned int uv_apicid_hibits;
58 EXPORT_SYMBOL_GPL(uv_apicid_hibits);
59 static DEFINE_SPINLOCK(uv_nmi_lock);
60
61 static unsigned long __init uv_early_read_mmr(unsigned long addr)
62 {
63 unsigned long val, *mmr;
64
65 mmr = early_ioremap(UV_LOCAL_MMR_BASE | addr, sizeof(*mmr));
66 val = *mmr;
67 early_iounmap(mmr, sizeof(*mmr));
68 return val;
69 }
70
71 static inline bool is_GRU_range(u64 start, u64 end)
72 {
73 return start >= gru_start_paddr && end <= gru_end_paddr;
74 }
75
76 static bool uv_is_untracked_pat_range(u64 start, u64 end)
77 {
78 return is_ISA_range(start, end) || is_GRU_range(start, end);
79 }
80
81 static int __init early_get_pnodeid(void)
82 {
83 union uvh_node_id_u node_id;
84 union uvh_rh_gam_config_mmr_u m_n_config;
85 int pnode;
86
87 /* Currently, all blades have same revision number */
88 node_id.v = uv_early_read_mmr(UVH_NODE_ID);
89 m_n_config.v = uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR);
90 uv_min_hub_revision_id = node_id.s.revision;
91
92 pnode = (node_id.s.node_id >> 1) & ((1 << m_n_config.s.n_skt) - 1);
93 return pnode;
94 }
95
96 static void __init early_get_apic_pnode_shift(void)
97 {
98 uvh_apicid.v = uv_early_read_mmr(UVH_APICID);
99 if (!uvh_apicid.v)
100 /*
101 * Old bios, use default value
102 */
103 uvh_apicid.s.pnode_shift = UV_APIC_PNODE_SHIFT;
104 }
105
106 /*
107 * Add an extra bit as dictated by bios to the destination apicid of
108 * interrupts potentially passing through the UV HUB. This prevents
109 * a deadlock between interrupts and IO port operations.
110 */
111 static void __init uv_set_apicid_hibit(void)
112 {
113 union uvh_lb_target_physical_apic_id_mask_u apicid_mask;
114
115 apicid_mask.v = uv_early_read_mmr(UVH_LB_TARGET_PHYSICAL_APIC_ID_MASK);
116 uv_apicid_hibits = apicid_mask.s.bit_enables & UV_APICID_HIBIT_MASK;
117 }
118
119 static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
120 {
121 int pnodeid;
122
123 if (!strcmp(oem_id, "SGI")) {
124 pnodeid = early_get_pnodeid();
125 early_get_apic_pnode_shift();
126 x86_platform.is_untracked_pat_range = uv_is_untracked_pat_range;
127 x86_platform.nmi_init = uv_nmi_init;
128 if (!strcmp(oem_table_id, "UVL"))
129 uv_system_type = UV_LEGACY_APIC;
130 else if (!strcmp(oem_table_id, "UVX"))
131 uv_system_type = UV_X2APIC;
132 else if (!strcmp(oem_table_id, "UVH")) {
133 __this_cpu_write(x2apic_extra_bits,
134 pnodeid << uvh_apicid.s.pnode_shift);
135 uv_system_type = UV_NON_UNIQUE_APIC;
136 uv_set_apicid_hibit();
137 return 1;
138 }
139 }
140 return 0;
141 }
142
143 enum uv_system_type get_uv_system_type(void)
144 {
145 return uv_system_type;
146 }
147
148 int is_uv_system(void)
149 {
150 return uv_system_type != UV_NONE;
151 }
152 EXPORT_SYMBOL_GPL(is_uv_system);
153
154 DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
155 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
156
157 struct uv_blade_info *uv_blade_info;
158 EXPORT_SYMBOL_GPL(uv_blade_info);
159
160 short *uv_node_to_blade;
161 EXPORT_SYMBOL_GPL(uv_node_to_blade);
162
163 short *uv_cpu_to_blade;
164 EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
165
166 short uv_possible_blades;
167 EXPORT_SYMBOL_GPL(uv_possible_blades);
168
169 unsigned long sn_rtc_cycles_per_second;
170 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
171
172 static const struct cpumask *uv_target_cpus(void)
173 {
174 return cpu_online_mask;
175 }
176
177 static void uv_vector_allocation_domain(int cpu, struct cpumask *retmask)
178 {
179 cpumask_clear(retmask);
180 cpumask_set_cpu(cpu, retmask);
181 }
182
183 static int __cpuinit uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
184 {
185 #ifdef CONFIG_SMP
186 unsigned long val;
187 int pnode;
188
189 pnode = uv_apicid_to_pnode(phys_apicid);
190 phys_apicid |= uv_apicid_hibits;
191 val = (1UL << UVH_IPI_INT_SEND_SHFT) |
192 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
193 ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
194 APIC_DM_INIT;
195 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
196 mdelay(10);
197
198 val = (1UL << UVH_IPI_INT_SEND_SHFT) |
199 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
200 ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
201 APIC_DM_STARTUP;
202 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
203
204 atomic_set(&init_deasserted, 1);
205 #endif
206 return 0;
207 }
208
209 static void uv_send_IPI_one(int cpu, int vector)
210 {
211 unsigned long apicid;
212 int pnode;
213
214 apicid = per_cpu(x86_cpu_to_apicid, cpu);
215 pnode = uv_apicid_to_pnode(apicid);
216 uv_hub_send_ipi(pnode, apicid, vector);
217 }
218
219 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
220 {
221 unsigned int cpu;
222
223 for_each_cpu(cpu, mask)
224 uv_send_IPI_one(cpu, vector);
225 }
226
227 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
228 {
229 unsigned int this_cpu = smp_processor_id();
230 unsigned int cpu;
231
232 for_each_cpu(cpu, mask) {
233 if (cpu != this_cpu)
234 uv_send_IPI_one(cpu, vector);
235 }
236 }
237
238 static void uv_send_IPI_allbutself(int vector)
239 {
240 unsigned int this_cpu = smp_processor_id();
241 unsigned int cpu;
242
243 for_each_online_cpu(cpu) {
244 if (cpu != this_cpu)
245 uv_send_IPI_one(cpu, vector);
246 }
247 }
248
249 static void uv_send_IPI_all(int vector)
250 {
251 uv_send_IPI_mask(cpu_online_mask, vector);
252 }
253
254 static int uv_apic_id_registered(void)
255 {
256 return 1;
257 }
258
259 static void uv_init_apic_ldr(void)
260 {
261 }
262
263 static unsigned int uv_cpu_mask_to_apicid(const struct cpumask *cpumask)
264 {
265 /*
266 * We're using fixed IRQ delivery, can only return one phys APIC ID.
267 * May as well be the first.
268 */
269 int cpu = cpumask_first(cpumask);
270
271 if ((unsigned)cpu < nr_cpu_ids)
272 return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
273 else
274 return BAD_APICID;
275 }
276
277 static unsigned int
278 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
279 const struct cpumask *andmask)
280 {
281 int cpu;
282
283 /*
284 * We're using fixed IRQ delivery, can only return one phys APIC ID.
285 * May as well be the first.
286 */
287 for_each_cpu_and(cpu, cpumask, andmask) {
288 if (cpumask_test_cpu(cpu, cpu_online_mask))
289 break;
290 }
291 return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
292 }
293
294 static unsigned int x2apic_get_apic_id(unsigned long x)
295 {
296 unsigned int id;
297
298 WARN_ON(preemptible() && num_online_cpus() > 1);
299 id = x | __this_cpu_read(x2apic_extra_bits);
300
301 return id;
302 }
303
304 static unsigned long set_apic_id(unsigned int id)
305 {
306 unsigned long x;
307
308 /* maskout x2apic_extra_bits ? */
309 x = id;
310 return x;
311 }
312
313 static unsigned int uv_read_apic_id(void)
314 {
315
316 return x2apic_get_apic_id(apic_read(APIC_ID));
317 }
318
319 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
320 {
321 return uv_read_apic_id() >> index_msb;
322 }
323
324 static void uv_send_IPI_self(int vector)
325 {
326 apic_write(APIC_SELF_IPI, vector);
327 }
328
329 struct apic __refdata apic_x2apic_uv_x = {
330
331 .name = "UV large system",
332 .probe = NULL,
333 .acpi_madt_oem_check = uv_acpi_madt_oem_check,
334 .apic_id_registered = uv_apic_id_registered,
335
336 .irq_delivery_mode = dest_Fixed,
337 .irq_dest_mode = 0, /* physical */
338
339 .target_cpus = uv_target_cpus,
340 .disable_esr = 0,
341 .dest_logical = APIC_DEST_LOGICAL,
342 .check_apicid_used = NULL,
343 .check_apicid_present = NULL,
344
345 .vector_allocation_domain = uv_vector_allocation_domain,
346 .init_apic_ldr = uv_init_apic_ldr,
347
348 .ioapic_phys_id_map = NULL,
349 .setup_apic_routing = NULL,
350 .multi_timer_check = NULL,
351 .cpu_present_to_apicid = default_cpu_present_to_apicid,
352 .apicid_to_cpu_present = NULL,
353 .setup_portio_remap = NULL,
354 .check_phys_apicid_present = default_check_phys_apicid_present,
355 .enable_apic_mode = NULL,
356 .phys_pkg_id = uv_phys_pkg_id,
357 .mps_oem_check = NULL,
358
359 .get_apic_id = x2apic_get_apic_id,
360 .set_apic_id = set_apic_id,
361 .apic_id_mask = 0xFFFFFFFFu,
362
363 .cpu_mask_to_apicid = uv_cpu_mask_to_apicid,
364 .cpu_mask_to_apicid_and = uv_cpu_mask_to_apicid_and,
365
366 .send_IPI_mask = uv_send_IPI_mask,
367 .send_IPI_mask_allbutself = uv_send_IPI_mask_allbutself,
368 .send_IPI_allbutself = uv_send_IPI_allbutself,
369 .send_IPI_all = uv_send_IPI_all,
370 .send_IPI_self = uv_send_IPI_self,
371
372 .wakeup_secondary_cpu = uv_wakeup_secondary,
373 .trampoline_phys_low = DEFAULT_TRAMPOLINE_PHYS_LOW,
374 .trampoline_phys_high = DEFAULT_TRAMPOLINE_PHYS_HIGH,
375 .wait_for_init_deassert = NULL,
376 .smp_callin_clear_local_apic = NULL,
377 .inquire_remote_apic = NULL,
378
379 .read = native_apic_msr_read,
380 .write = native_apic_msr_write,
381 .icr_read = native_x2apic_icr_read,
382 .icr_write = native_x2apic_icr_write,
383 .wait_icr_idle = native_x2apic_wait_icr_idle,
384 .safe_wait_icr_idle = native_safe_x2apic_wait_icr_idle,
385 };
386
387 static __cpuinit void set_x2apic_extra_bits(int pnode)
388 {
389 __this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
390 }
391
392 /*
393 * Called on boot cpu.
394 */
395 static __init int boot_pnode_to_blade(int pnode)
396 {
397 int blade;
398
399 for (blade = 0; blade < uv_num_possible_blades(); blade++)
400 if (pnode == uv_blade_info[blade].pnode)
401 return blade;
402 BUG();
403 }
404
405 struct redir_addr {
406 unsigned long redirect;
407 unsigned long alias;
408 };
409
410 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
411
412 static __initdata struct redir_addr redir_addrs[] = {
413 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
414 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
415 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
416 };
417
418 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
419 {
420 union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
421 union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
422 int i;
423
424 for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
425 alias.v = uv_read_local_mmr(redir_addrs[i].alias);
426 if (alias.s.enable && alias.s.base == 0) {
427 *size = (1UL << alias.s.m_alias);
428 redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
429 *base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
430 return;
431 }
432 }
433 *base = *size = 0;
434 }
435
436 enum map_type {map_wb, map_uc};
437
438 static __init void map_high(char *id, unsigned long base, int pshift,
439 int bshift, int max_pnode, enum map_type map_type)
440 {
441 unsigned long bytes, paddr;
442
443 paddr = base << pshift;
444 bytes = (1UL << bshift) * (max_pnode + 1);
445 printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
446 paddr + bytes);
447 if (map_type == map_uc)
448 init_extra_mapping_uc(paddr, bytes);
449 else
450 init_extra_mapping_wb(paddr, bytes);
451
452 }
453 static __init void map_gru_high(int max_pnode)
454 {
455 union uvh_rh_gam_gru_overlay_config_mmr_u gru;
456 int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
457
458 gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
459 if (gru.s.enable) {
460 map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
461 gru_start_paddr = ((u64)gru.s.base << shift);
462 gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
463
464 }
465 }
466
467 static __init void map_mmr_high(int max_pnode)
468 {
469 union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
470 int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
471
472 mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
473 if (mmr.s.enable)
474 map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
475 }
476
477 static __init void map_mmioh_high(int max_pnode)
478 {
479 union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
480 int shift = UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
481
482 mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
483 if (mmioh.s.enable)
484 map_high("MMIOH", mmioh.s.base, shift, mmioh.s.m_io,
485 max_pnode, map_uc);
486 }
487
488 static __init void map_low_mmrs(void)
489 {
490 init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
491 init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
492 }
493
494 static __init void uv_rtc_init(void)
495 {
496 long status;
497 u64 ticks_per_sec;
498
499 status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
500 &ticks_per_sec);
501 if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
502 printk(KERN_WARNING
503 "unable to determine platform RTC clock frequency, "
504 "guessing.\n");
505 /* BIOS gives wrong value for clock freq. so guess */
506 sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
507 } else
508 sn_rtc_cycles_per_second = ticks_per_sec;
509 }
510
511 /*
512 * percpu heartbeat timer
513 */
514 static void uv_heartbeat(unsigned long ignored)
515 {
516 struct timer_list *timer = &uv_hub_info->scir.timer;
517 unsigned char bits = uv_hub_info->scir.state;
518
519 /* flip heartbeat bit */
520 bits ^= SCIR_CPU_HEARTBEAT;
521
522 /* is this cpu idle? */
523 if (idle_cpu(raw_smp_processor_id()))
524 bits &= ~SCIR_CPU_ACTIVITY;
525 else
526 bits |= SCIR_CPU_ACTIVITY;
527
528 /* update system controller interface reg */
529 uv_set_scir_bits(bits);
530
531 /* enable next timer period */
532 mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
533 }
534
535 static void __cpuinit uv_heartbeat_enable(int cpu)
536 {
537 while (!uv_cpu_hub_info(cpu)->scir.enabled) {
538 struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
539
540 uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
541 setup_timer(timer, uv_heartbeat, cpu);
542 timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
543 add_timer_on(timer, cpu);
544 uv_cpu_hub_info(cpu)->scir.enabled = 1;
545
546 /* also ensure that boot cpu is enabled */
547 cpu = 0;
548 }
549 }
550
551 #ifdef CONFIG_HOTPLUG_CPU
552 static void __cpuinit uv_heartbeat_disable(int cpu)
553 {
554 if (uv_cpu_hub_info(cpu)->scir.enabled) {
555 uv_cpu_hub_info(cpu)->scir.enabled = 0;
556 del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
557 }
558 uv_set_cpu_scir_bits(cpu, 0xff);
559 }
560
561 /*
562 * cpu hotplug notifier
563 */
564 static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
565 unsigned long action, void *hcpu)
566 {
567 long cpu = (long)hcpu;
568
569 switch (action) {
570 case CPU_ONLINE:
571 uv_heartbeat_enable(cpu);
572 break;
573 case CPU_DOWN_PREPARE:
574 uv_heartbeat_disable(cpu);
575 break;
576 default:
577 break;
578 }
579 return NOTIFY_OK;
580 }
581
582 static __init void uv_scir_register_cpu_notifier(void)
583 {
584 hotcpu_notifier(uv_scir_cpu_notify, 0);
585 }
586
587 #else /* !CONFIG_HOTPLUG_CPU */
588
589 static __init void uv_scir_register_cpu_notifier(void)
590 {
591 }
592
593 static __init int uv_init_heartbeat(void)
594 {
595 int cpu;
596
597 if (is_uv_system())
598 for_each_online_cpu(cpu)
599 uv_heartbeat_enable(cpu);
600 return 0;
601 }
602
603 late_initcall(uv_init_heartbeat);
604
605 #endif /* !CONFIG_HOTPLUG_CPU */
606
607 /* Direct Legacy VGA I/O traffic to designated IOH */
608 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
609 unsigned int command_bits, bool change_bridge)
610 {
611 int domain, bus, rc;
612
613 PR_DEVEL("devfn %x decode %d cmd %x chg_brdg %d\n",
614 pdev->devfn, decode, command_bits, change_bridge);
615
616 if (!change_bridge)
617 return 0;
618
619 if ((command_bits & PCI_COMMAND_IO) == 0)
620 return 0;
621
622 domain = pci_domain_nr(pdev->bus);
623 bus = pdev->bus->number;
624
625 rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
626 PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
627
628 return rc;
629 }
630
631 /*
632 * Called on each cpu to initialize the per_cpu UV data area.
633 * FIXME: hotplug not supported yet
634 */
635 void __cpuinit uv_cpu_init(void)
636 {
637 /* CPU 0 initilization will be done via uv_system_init. */
638 if (!uv_blade_info)
639 return;
640
641 uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
642
643 if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
644 set_x2apic_extra_bits(uv_hub_info->pnode);
645 }
646
647 /*
648 * When NMI is received, print a stack trace.
649 */
650 int uv_handle_nmi(struct notifier_block *self, unsigned long reason, void *data)
651 {
652 unsigned long real_uv_nmi;
653 int bid;
654
655 if (reason != DIE_NMIUNKNOWN)
656 return NOTIFY_OK;
657
658 if (in_crash_kexec)
659 /* do nothing if entering the crash kernel */
660 return NOTIFY_OK;
661
662 /*
663 * Each blade has an MMR that indicates when an NMI has been sent
664 * to cpus on the blade. If an NMI is detected, atomically
665 * clear the MMR and update a per-blade NMI count used to
666 * cause each cpu on the blade to notice a new NMI.
667 */
668 bid = uv_numa_blade_id();
669 real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
670
671 if (unlikely(real_uv_nmi)) {
672 spin_lock(&uv_blade_info[bid].nmi_lock);
673 real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
674 if (real_uv_nmi) {
675 uv_blade_info[bid].nmi_count++;
676 uv_write_local_mmr(UVH_NMI_MMR_CLEAR, UV_NMI_PENDING_MASK);
677 }
678 spin_unlock(&uv_blade_info[bid].nmi_lock);
679 }
680
681 if (likely(__get_cpu_var(cpu_last_nmi_count) == uv_blade_info[bid].nmi_count))
682 return NOTIFY_DONE;
683
684 __get_cpu_var(cpu_last_nmi_count) = uv_blade_info[bid].nmi_count;
685
686 /*
687 * Use a lock so only one cpu prints at a time.
688 * This prevents intermixed output.
689 */
690 spin_lock(&uv_nmi_lock);
691 pr_info("UV NMI stack dump cpu %u:\n", smp_processor_id());
692 dump_stack();
693 spin_unlock(&uv_nmi_lock);
694
695 return NOTIFY_STOP;
696 }
697
698 static struct notifier_block uv_dump_stack_nmi_nb = {
699 .notifier_call = uv_handle_nmi,
700 .priority = NMI_LOCAL_LOW_PRIOR - 1,
701 };
702
703 void uv_register_nmi_notifier(void)
704 {
705 if (register_die_notifier(&uv_dump_stack_nmi_nb))
706 printk(KERN_WARNING "UV NMI handler failed to register\n");
707 }
708
709 void uv_nmi_init(void)
710 {
711 unsigned int value;
712
713 /*
714 * Unmask NMI on all cpus
715 */
716 value = apic_read(APIC_LVT1) | APIC_DM_NMI;
717 value &= ~APIC_LVT_MASKED;
718 apic_write(APIC_LVT1, value);
719 }
720
721 void __init uv_system_init(void)
722 {
723 union uvh_rh_gam_config_mmr_u m_n_config;
724 union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
725 union uvh_node_id_u node_id;
726 unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
727 int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val, n_io;
728 int gnode_extra, max_pnode = 0;
729 unsigned long mmr_base, present, paddr;
730 unsigned short pnode_mask, pnode_io_mask;
731
732 map_low_mmrs();
733
734 m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
735 m_val = m_n_config.s.m_skt;
736 n_val = m_n_config.s.n_skt;
737 mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
738 n_io = mmioh.s.n_io;
739 mmr_base =
740 uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
741 ~UV_MMR_ENABLE;
742 pnode_mask = (1 << n_val) - 1;
743 pnode_io_mask = (1 << n_io) - 1;
744
745 node_id.v = uv_read_local_mmr(UVH_NODE_ID);
746 gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
747 gnode_upper = ((unsigned long)gnode_extra << m_val);
748 printk(KERN_INFO "UV: N %d, M %d, N_IO: %d, gnode_upper 0x%lx, gnode_extra 0x%x, pnode_mask 0x%x, pnode_io_mask 0x%x\n",
749 n_val, m_val, n_io, gnode_upper, gnode_extra, pnode_mask, pnode_io_mask);
750
751 printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
752
753 for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
754 uv_possible_blades +=
755 hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
756 printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
757
758 bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
759 uv_blade_info = kzalloc(bytes, GFP_KERNEL);
760 BUG_ON(!uv_blade_info);
761
762 for (blade = 0; blade < uv_num_possible_blades(); blade++)
763 uv_blade_info[blade].memory_nid = -1;
764
765 get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
766
767 bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
768 uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
769 BUG_ON(!uv_node_to_blade);
770 memset(uv_node_to_blade, 255, bytes);
771
772 bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
773 uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
774 BUG_ON(!uv_cpu_to_blade);
775 memset(uv_cpu_to_blade, 255, bytes);
776
777 blade = 0;
778 for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
779 present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
780 for (j = 0; j < 64; j++) {
781 if (!test_bit(j, &present))
782 continue;
783 pnode = (i * 64 + j) & pnode_mask;
784 uv_blade_info[blade].pnode = pnode;
785 uv_blade_info[blade].nr_possible_cpus = 0;
786 uv_blade_info[blade].nr_online_cpus = 0;
787 spin_lock_init(&uv_blade_info[blade].nmi_lock);
788 max_pnode = max(pnode, max_pnode);
789 blade++;
790 }
791 }
792
793 uv_bios_init();
794 uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
795 &sn_region_size, &system_serial_number);
796 uv_rtc_init();
797
798 for_each_present_cpu(cpu) {
799 int apicid = per_cpu(x86_cpu_to_apicid, cpu);
800
801 nid = cpu_to_node(cpu);
802 /*
803 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
804 */
805 uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
806 uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
807 pnode = uv_apicid_to_pnode(apicid);
808 blade = boot_pnode_to_blade(pnode);
809 lcpu = uv_blade_info[blade].nr_possible_cpus;
810 uv_blade_info[blade].nr_possible_cpus++;
811
812 /* Any node on the blade, else will contain -1. */
813 uv_blade_info[blade].memory_nid = nid;
814
815 uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
816 uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
817 uv_cpu_hub_info(cpu)->m_val = m_val;
818 uv_cpu_hub_info(cpu)->n_val = n_val;
819 uv_cpu_hub_info(cpu)->numa_blade_id = blade;
820 uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
821 uv_cpu_hub_info(cpu)->pnode = pnode;
822 uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
823 uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
824 uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
825 uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
826 uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
827 uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
828 uv_node_to_blade[nid] = blade;
829 uv_cpu_to_blade[cpu] = blade;
830 }
831
832 /* Add blade/pnode info for nodes without cpus */
833 for_each_online_node(nid) {
834 if (uv_node_to_blade[nid] >= 0)
835 continue;
836 paddr = node_start_pfn(nid) << PAGE_SHIFT;
837 paddr = uv_soc_phys_ram_to_gpa(paddr);
838 pnode = (paddr >> m_val) & pnode_mask;
839 blade = boot_pnode_to_blade(pnode);
840 uv_node_to_blade[nid] = blade;
841 }
842
843 map_gru_high(max_pnode);
844 map_mmr_high(max_pnode);
845 map_mmioh_high(max_pnode & pnode_io_mask);
846
847 uv_cpu_init();
848 uv_scir_register_cpu_notifier();
849 uv_register_nmi_notifier();
850 proc_mkdir("sgi_uv", NULL);
851
852 /* register Legacy VGA I/O redirection handler */
853 pci_register_set_vga_state(uv_set_vga_state);
854
855 /*
856 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
857 * EFI is not enabled in the kdump kernel.
858 */
859 if (is_kdump_kernel())
860 reboot_type = BOOT_ACPI;
861 }