Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs
[pandora-kernel.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
41 DEFINE_PER_CPU(int, x2apic_extra_bits);
42
43 #define PR_DEVEL(fmt, args...)  pr_devel("%s: " fmt, __func__, args)
44
45 static enum uv_system_type uv_system_type;
46 static u64 gru_start_paddr, gru_end_paddr;
47 static union uvh_apicid uvh_apicid;
48 int uv_min_hub_revision_id;
49 EXPORT_SYMBOL_GPL(uv_min_hub_revision_id);
50 unsigned int uv_apicid_hibits;
51 EXPORT_SYMBOL_GPL(uv_apicid_hibits);
52 static DEFINE_SPINLOCK(uv_nmi_lock);
53
54 static unsigned long __init uv_early_read_mmr(unsigned long addr)
55 {
56         unsigned long val, *mmr;
57
58         mmr = early_ioremap(UV_LOCAL_MMR_BASE | addr, sizeof(*mmr));
59         val = *mmr;
60         early_iounmap(mmr, sizeof(*mmr));
61         return val;
62 }
63
64 static inline bool is_GRU_range(u64 start, u64 end)
65 {
66         return start >= gru_start_paddr && end <= gru_end_paddr;
67 }
68
69 static bool uv_is_untracked_pat_range(u64 start, u64 end)
70 {
71         return is_ISA_range(start, end) || is_GRU_range(start, end);
72 }
73
74 static int __init early_get_pnodeid(void)
75 {
76         union uvh_node_id_u node_id;
77         union uvh_rh_gam_config_mmr_u  m_n_config;
78         int pnode;
79
80         /* Currently, all blades have same revision number */
81         node_id.v = uv_early_read_mmr(UVH_NODE_ID);
82         m_n_config.v = uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR);
83         uv_min_hub_revision_id = node_id.s.revision;
84
85         pnode = (node_id.s.node_id >> 1) & ((1 << m_n_config.s.n_skt) - 1);
86         return pnode;
87 }
88
89 static void __init early_get_apic_pnode_shift(void)
90 {
91         uvh_apicid.v = uv_early_read_mmr(UVH_APICID);
92         if (!uvh_apicid.v)
93                 /*
94                  * Old bios, use default value
95                  */
96                 uvh_apicid.s.pnode_shift = UV_APIC_PNODE_SHIFT;
97 }
98
99 /*
100  * Add an extra bit as dictated by bios to the destination apicid of
101  * interrupts potentially passing through the UV HUB.  This prevents
102  * a deadlock between interrupts and IO port operations.
103  */
104 static void __init uv_set_apicid_hibit(void)
105 {
106         union uvh_lb_target_physical_apic_id_mask_u apicid_mask;
107
108         apicid_mask.v = uv_early_read_mmr(UVH_LB_TARGET_PHYSICAL_APIC_ID_MASK);
109         uv_apicid_hibits = apicid_mask.s.bit_enables & UV_APICID_HIBIT_MASK;
110 }
111
112 static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
113 {
114         int pnodeid;
115
116         if (!strcmp(oem_id, "SGI")) {
117                 pnodeid = early_get_pnodeid();
118                 early_get_apic_pnode_shift();
119                 x86_platform.is_untracked_pat_range =  uv_is_untracked_pat_range;
120                 x86_platform.nmi_init = uv_nmi_init;
121                 if (!strcmp(oem_table_id, "UVL"))
122                         uv_system_type = UV_LEGACY_APIC;
123                 else if (!strcmp(oem_table_id, "UVX"))
124                         uv_system_type = UV_X2APIC;
125                 else if (!strcmp(oem_table_id, "UVH")) {
126                         __this_cpu_write(x2apic_extra_bits,
127                                 pnodeid << uvh_apicid.s.pnode_shift);
128                         uv_system_type = UV_NON_UNIQUE_APIC;
129                         uv_set_apicid_hibit();
130                         return 1;
131                 }
132         }
133         return 0;
134 }
135
136 enum uv_system_type get_uv_system_type(void)
137 {
138         return uv_system_type;
139 }
140
141 int is_uv_system(void)
142 {
143         return uv_system_type != UV_NONE;
144 }
145 EXPORT_SYMBOL_GPL(is_uv_system);
146
147 DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
148 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
149
150 struct uv_blade_info *uv_blade_info;
151 EXPORT_SYMBOL_GPL(uv_blade_info);
152
153 short *uv_node_to_blade;
154 EXPORT_SYMBOL_GPL(uv_node_to_blade);
155
156 short *uv_cpu_to_blade;
157 EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
158
159 short uv_possible_blades;
160 EXPORT_SYMBOL_GPL(uv_possible_blades);
161
162 unsigned long sn_rtc_cycles_per_second;
163 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
164
165 static const struct cpumask *uv_target_cpus(void)
166 {
167         return cpu_online_mask;
168 }
169
170 static void uv_vector_allocation_domain(int cpu, struct cpumask *retmask)
171 {
172         cpumask_clear(retmask);
173         cpumask_set_cpu(cpu, retmask);
174 }
175
176 static int __cpuinit uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
177 {
178 #ifdef CONFIG_SMP
179         unsigned long val;
180         int pnode;
181
182         pnode = uv_apicid_to_pnode(phys_apicid);
183         phys_apicid |= uv_apicid_hibits;
184         val = (1UL << UVH_IPI_INT_SEND_SHFT) |
185             (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
186             ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
187             APIC_DM_INIT;
188         uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
189         mdelay(10);
190
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_STARTUP;
195         uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
196
197         atomic_set(&init_deasserted, 1);
198 #endif
199         return 0;
200 }
201
202 static void uv_send_IPI_one(int cpu, int vector)
203 {
204         unsigned long apicid;
205         int pnode;
206
207         apicid = per_cpu(x86_cpu_to_apicid, cpu);
208         pnode = uv_apicid_to_pnode(apicid);
209         uv_hub_send_ipi(pnode, apicid, vector);
210 }
211
212 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
213 {
214         unsigned int cpu;
215
216         for_each_cpu(cpu, mask)
217                 uv_send_IPI_one(cpu, vector);
218 }
219
220 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
221 {
222         unsigned int this_cpu = smp_processor_id();
223         unsigned int cpu;
224
225         for_each_cpu(cpu, mask) {
226                 if (cpu != this_cpu)
227                         uv_send_IPI_one(cpu, vector);
228         }
229 }
230
231 static void uv_send_IPI_allbutself(int vector)
232 {
233         unsigned int this_cpu = smp_processor_id();
234         unsigned int cpu;
235
236         for_each_online_cpu(cpu) {
237                 if (cpu != this_cpu)
238                         uv_send_IPI_one(cpu, vector);
239         }
240 }
241
242 static void uv_send_IPI_all(int vector)
243 {
244         uv_send_IPI_mask(cpu_online_mask, vector);
245 }
246
247 static int uv_apic_id_registered(void)
248 {
249         return 1;
250 }
251
252 static void uv_init_apic_ldr(void)
253 {
254 }
255
256 static unsigned int uv_cpu_mask_to_apicid(const struct cpumask *cpumask)
257 {
258         /*
259          * We're using fixed IRQ delivery, can only return one phys APIC ID.
260          * May as well be the first.
261          */
262         int cpu = cpumask_first(cpumask);
263
264         if ((unsigned)cpu < nr_cpu_ids)
265                 return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
266         else
267                 return BAD_APICID;
268 }
269
270 static unsigned int
271 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
272                           const struct cpumask *andmask)
273 {
274         int cpu;
275
276         /*
277          * We're using fixed IRQ delivery, can only return one phys APIC ID.
278          * May as well be the first.
279          */
280         for_each_cpu_and(cpu, cpumask, andmask) {
281                 if (cpumask_test_cpu(cpu, cpu_online_mask))
282                         break;
283         }
284         return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
285 }
286
287 static unsigned int x2apic_get_apic_id(unsigned long x)
288 {
289         unsigned int id;
290
291         WARN_ON(preemptible() && num_online_cpus() > 1);
292         id = x | __this_cpu_read(x2apic_extra_bits);
293
294         return id;
295 }
296
297 static unsigned long set_apic_id(unsigned int id)
298 {
299         unsigned long x;
300
301         /* maskout x2apic_extra_bits ? */
302         x = id;
303         return x;
304 }
305
306 static unsigned int uv_read_apic_id(void)
307 {
308
309         return x2apic_get_apic_id(apic_read(APIC_ID));
310 }
311
312 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
313 {
314         return uv_read_apic_id() >> index_msb;
315 }
316
317 static void uv_send_IPI_self(int vector)
318 {
319         apic_write(APIC_SELF_IPI, vector);
320 }
321
322 struct apic __refdata apic_x2apic_uv_x = {
323
324         .name                           = "UV large system",
325         .probe                          = NULL,
326         .acpi_madt_oem_check            = uv_acpi_madt_oem_check,
327         .apic_id_registered             = uv_apic_id_registered,
328
329         .irq_delivery_mode              = dest_Fixed,
330         .irq_dest_mode                  = 0, /* physical */
331
332         .target_cpus                    = uv_target_cpus,
333         .disable_esr                    = 0,
334         .dest_logical                   = APIC_DEST_LOGICAL,
335         .check_apicid_used              = NULL,
336         .check_apicid_present           = NULL,
337
338         .vector_allocation_domain       = uv_vector_allocation_domain,
339         .init_apic_ldr                  = uv_init_apic_ldr,
340
341         .ioapic_phys_id_map             = NULL,
342         .setup_apic_routing             = NULL,
343         .multi_timer_check              = NULL,
344         .cpu_present_to_apicid          = default_cpu_present_to_apicid,
345         .apicid_to_cpu_present          = NULL,
346         .setup_portio_remap             = NULL,
347         .check_phys_apicid_present      = default_check_phys_apicid_present,
348         .enable_apic_mode               = NULL,
349         .phys_pkg_id                    = uv_phys_pkg_id,
350         .mps_oem_check                  = NULL,
351
352         .get_apic_id                    = x2apic_get_apic_id,
353         .set_apic_id                    = set_apic_id,
354         .apic_id_mask                   = 0xFFFFFFFFu,
355
356         .cpu_mask_to_apicid             = uv_cpu_mask_to_apicid,
357         .cpu_mask_to_apicid_and         = uv_cpu_mask_to_apicid_and,
358
359         .send_IPI_mask                  = uv_send_IPI_mask,
360         .send_IPI_mask_allbutself       = uv_send_IPI_mask_allbutself,
361         .send_IPI_allbutself            = uv_send_IPI_allbutself,
362         .send_IPI_all                   = uv_send_IPI_all,
363         .send_IPI_self                  = uv_send_IPI_self,
364
365         .wakeup_secondary_cpu           = uv_wakeup_secondary,
366         .trampoline_phys_low            = DEFAULT_TRAMPOLINE_PHYS_LOW,
367         .trampoline_phys_high           = DEFAULT_TRAMPOLINE_PHYS_HIGH,
368         .wait_for_init_deassert         = NULL,
369         .smp_callin_clear_local_apic    = NULL,
370         .inquire_remote_apic            = NULL,
371
372         .read                           = native_apic_msr_read,
373         .write                          = native_apic_msr_write,
374         .icr_read                       = native_x2apic_icr_read,
375         .icr_write                      = native_x2apic_icr_write,
376         .wait_icr_idle                  = native_x2apic_wait_icr_idle,
377         .safe_wait_icr_idle             = native_safe_x2apic_wait_icr_idle,
378 };
379
380 static __cpuinit void set_x2apic_extra_bits(int pnode)
381 {
382         __this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
383 }
384
385 /*
386  * Called on boot cpu.
387  */
388 static __init int boot_pnode_to_blade(int pnode)
389 {
390         int blade;
391
392         for (blade = 0; blade < uv_num_possible_blades(); blade++)
393                 if (pnode == uv_blade_info[blade].pnode)
394                         return blade;
395         BUG();
396 }
397
398 struct redir_addr {
399         unsigned long redirect;
400         unsigned long alias;
401 };
402
403 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
404
405 static __initdata struct redir_addr redir_addrs[] = {
406         {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
407         {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
408         {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
409 };
410
411 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
412 {
413         union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
414         union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
415         int i;
416
417         for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
418                 alias.v = uv_read_local_mmr(redir_addrs[i].alias);
419                 if (alias.s.enable && alias.s.base == 0) {
420                         *size = (1UL << alias.s.m_alias);
421                         redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
422                         *base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
423                         return;
424                 }
425         }
426         *base = *size = 0;
427 }
428
429 enum map_type {map_wb, map_uc};
430
431 static __init void map_high(char *id, unsigned long base, int pshift,
432                         int bshift, int max_pnode, enum map_type map_type)
433 {
434         unsigned long bytes, paddr;
435
436         paddr = base << pshift;
437         bytes = (1UL << bshift) * (max_pnode + 1);
438         printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
439                                                 paddr + bytes);
440         if (map_type == map_uc)
441                 init_extra_mapping_uc(paddr, bytes);
442         else
443                 init_extra_mapping_wb(paddr, bytes);
444
445 }
446 static __init void map_gru_high(int max_pnode)
447 {
448         union uvh_rh_gam_gru_overlay_config_mmr_u gru;
449         int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
450
451         gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
452         if (gru.s.enable) {
453                 map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
454                 gru_start_paddr = ((u64)gru.s.base << shift);
455                 gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
456
457         }
458 }
459
460 static __init void map_mmr_high(int max_pnode)
461 {
462         union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
463         int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
464
465         mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
466         if (mmr.s.enable)
467                 map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
468 }
469
470 static __init void map_mmioh_high(int max_pnode)
471 {
472         union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
473         int shift = UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
474
475         mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
476         if (mmioh.s.enable)
477                 map_high("MMIOH", mmioh.s.base, shift, mmioh.s.m_io,
478                         max_pnode, map_uc);
479 }
480
481 static __init void map_low_mmrs(void)
482 {
483         init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
484         init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
485 }
486
487 static __init void uv_rtc_init(void)
488 {
489         long status;
490         u64 ticks_per_sec;
491
492         status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
493                                         &ticks_per_sec);
494         if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
495                 printk(KERN_WARNING
496                         "unable to determine platform RTC clock frequency, "
497                         "guessing.\n");
498                 /* BIOS gives wrong value for clock freq. so guess */
499                 sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
500         } else
501                 sn_rtc_cycles_per_second = ticks_per_sec;
502 }
503
504 /*
505  * percpu heartbeat timer
506  */
507 static void uv_heartbeat(unsigned long ignored)
508 {
509         struct timer_list *timer = &uv_hub_info->scir.timer;
510         unsigned char bits = uv_hub_info->scir.state;
511
512         /* flip heartbeat bit */
513         bits ^= SCIR_CPU_HEARTBEAT;
514
515         /* is this cpu idle? */
516         if (idle_cpu(raw_smp_processor_id()))
517                 bits &= ~SCIR_CPU_ACTIVITY;
518         else
519                 bits |= SCIR_CPU_ACTIVITY;
520
521         /* update system controller interface reg */
522         uv_set_scir_bits(bits);
523
524         /* enable next timer period */
525         mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
526 }
527
528 static void __cpuinit uv_heartbeat_enable(int cpu)
529 {
530         while (!uv_cpu_hub_info(cpu)->scir.enabled) {
531                 struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
532
533                 uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
534                 setup_timer(timer, uv_heartbeat, cpu);
535                 timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
536                 add_timer_on(timer, cpu);
537                 uv_cpu_hub_info(cpu)->scir.enabled = 1;
538
539                 /* also ensure that boot cpu is enabled */
540                 cpu = 0;
541         }
542 }
543
544 #ifdef CONFIG_HOTPLUG_CPU
545 static void __cpuinit uv_heartbeat_disable(int cpu)
546 {
547         if (uv_cpu_hub_info(cpu)->scir.enabled) {
548                 uv_cpu_hub_info(cpu)->scir.enabled = 0;
549                 del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
550         }
551         uv_set_cpu_scir_bits(cpu, 0xff);
552 }
553
554 /*
555  * cpu hotplug notifier
556  */
557 static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
558                                        unsigned long action, void *hcpu)
559 {
560         long cpu = (long)hcpu;
561
562         switch (action) {
563         case CPU_ONLINE:
564                 uv_heartbeat_enable(cpu);
565                 break;
566         case CPU_DOWN_PREPARE:
567                 uv_heartbeat_disable(cpu);
568                 break;
569         default:
570                 break;
571         }
572         return NOTIFY_OK;
573 }
574
575 static __init void uv_scir_register_cpu_notifier(void)
576 {
577         hotcpu_notifier(uv_scir_cpu_notify, 0);
578 }
579
580 #else /* !CONFIG_HOTPLUG_CPU */
581
582 static __init void uv_scir_register_cpu_notifier(void)
583 {
584 }
585
586 static __init int uv_init_heartbeat(void)
587 {
588         int cpu;
589
590         if (is_uv_system())
591                 for_each_online_cpu(cpu)
592                         uv_heartbeat_enable(cpu);
593         return 0;
594 }
595
596 late_initcall(uv_init_heartbeat);
597
598 #endif /* !CONFIG_HOTPLUG_CPU */
599
600 /* Direct Legacy VGA I/O traffic to designated IOH */
601 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
602                       unsigned int command_bits, bool change_bridge)
603 {
604         int domain, bus, rc;
605
606         PR_DEVEL("devfn %x decode %d cmd %x chg_brdg %d\n",
607                         pdev->devfn, decode, command_bits, change_bridge);
608
609         if (!change_bridge)
610                 return 0;
611
612         if ((command_bits & PCI_COMMAND_IO) == 0)
613                 return 0;
614
615         domain = pci_domain_nr(pdev->bus);
616         bus = pdev->bus->number;
617
618         rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
619         PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
620
621         return rc;
622 }
623
624 /*
625  * Called on each cpu to initialize the per_cpu UV data area.
626  * FIXME: hotplug not supported yet
627  */
628 void __cpuinit uv_cpu_init(void)
629 {
630         /* CPU 0 initilization will be done via uv_system_init. */
631         if (!uv_blade_info)
632                 return;
633
634         uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
635
636         if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
637                 set_x2apic_extra_bits(uv_hub_info->pnode);
638 }
639
640 /*
641  * When NMI is received, print a stack trace.
642  */
643 int uv_handle_nmi(struct notifier_block *self, unsigned long reason, void *data)
644 {
645         if (reason != DIE_NMIUNKNOWN)
646                 return NOTIFY_OK;
647
648         if (in_crash_kexec)
649                 /* do nothing if entering the crash kernel */
650                 return NOTIFY_OK;
651         /*
652          * Use a lock so only one cpu prints at a time
653          * to prevent intermixed output.
654          */
655         spin_lock(&uv_nmi_lock);
656         pr_info("NMI stack dump cpu %u:\n", smp_processor_id());
657         dump_stack();
658         spin_unlock(&uv_nmi_lock);
659
660         return NOTIFY_STOP;
661 }
662
663 static struct notifier_block uv_dump_stack_nmi_nb = {
664         .notifier_call  = uv_handle_nmi
665 };
666
667 void uv_register_nmi_notifier(void)
668 {
669         if (register_die_notifier(&uv_dump_stack_nmi_nb))
670                 printk(KERN_WARNING "UV NMI handler failed to register\n");
671 }
672
673 void uv_nmi_init(void)
674 {
675         unsigned int value;
676
677         /*
678          * Unmask NMI on all cpus
679          */
680         value = apic_read(APIC_LVT1) | APIC_DM_NMI;
681         value &= ~APIC_LVT_MASKED;
682         apic_write(APIC_LVT1, value);
683 }
684
685 void __init uv_system_init(void)
686 {
687         union uvh_rh_gam_config_mmr_u  m_n_config;
688         union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
689         union uvh_node_id_u node_id;
690         unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
691         int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val, n_io;
692         int gnode_extra, max_pnode = 0;
693         unsigned long mmr_base, present, paddr;
694         unsigned short pnode_mask, pnode_io_mask;
695
696         map_low_mmrs();
697
698         m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
699         m_val = m_n_config.s.m_skt;
700         n_val = m_n_config.s.n_skt;
701         mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
702         n_io = mmioh.s.n_io;
703         mmr_base =
704             uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
705             ~UV_MMR_ENABLE;
706         pnode_mask = (1 << n_val) - 1;
707         pnode_io_mask = (1 << n_io) - 1;
708
709         node_id.v = uv_read_local_mmr(UVH_NODE_ID);
710         gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
711         gnode_upper = ((unsigned long)gnode_extra  << m_val);
712         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",
713                         n_val, m_val, n_io, gnode_upper, gnode_extra, pnode_mask, pnode_io_mask);
714
715         printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
716
717         for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
718                 uv_possible_blades +=
719                   hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
720         printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
721
722         bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
723         uv_blade_info = kmalloc(bytes, GFP_KERNEL);
724         BUG_ON(!uv_blade_info);
725         for (blade = 0; blade < uv_num_possible_blades(); blade++)
726                 uv_blade_info[blade].memory_nid = -1;
727
728         get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
729
730         bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
731         uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
732         BUG_ON(!uv_node_to_blade);
733         memset(uv_node_to_blade, 255, bytes);
734
735         bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
736         uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
737         BUG_ON(!uv_cpu_to_blade);
738         memset(uv_cpu_to_blade, 255, bytes);
739
740         blade = 0;
741         for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
742                 present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
743                 for (j = 0; j < 64; j++) {
744                         if (!test_bit(j, &present))
745                                 continue;
746                         pnode = (i * 64 + j) & pnode_mask;
747                         uv_blade_info[blade].pnode = pnode;
748                         uv_blade_info[blade].nr_possible_cpus = 0;
749                         uv_blade_info[blade].nr_online_cpus = 0;
750                         max_pnode = max(pnode, max_pnode);
751                         blade++;
752                 }
753         }
754
755         uv_bios_init();
756         uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
757                             &sn_region_size, &system_serial_number);
758         uv_rtc_init();
759
760         for_each_present_cpu(cpu) {
761                 int apicid = per_cpu(x86_cpu_to_apicid, cpu);
762
763                 nid = cpu_to_node(cpu);
764                 /*
765                  * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
766                  */
767                 uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
768                 uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
769                 pnode = uv_apicid_to_pnode(apicid);
770                 blade = boot_pnode_to_blade(pnode);
771                 lcpu = uv_blade_info[blade].nr_possible_cpus;
772                 uv_blade_info[blade].nr_possible_cpus++;
773
774                 /* Any node on the blade, else will contain -1. */
775                 uv_blade_info[blade].memory_nid = nid;
776
777                 uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
778                 uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
779                 uv_cpu_hub_info(cpu)->m_val = m_val;
780                 uv_cpu_hub_info(cpu)->n_val = n_val;
781                 uv_cpu_hub_info(cpu)->numa_blade_id = blade;
782                 uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
783                 uv_cpu_hub_info(cpu)->pnode = pnode;
784                 uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
785                 uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
786                 uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
787                 uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
788                 uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
789                 uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
790                 uv_node_to_blade[nid] = blade;
791                 uv_cpu_to_blade[cpu] = blade;
792         }
793
794         /* Add blade/pnode info for nodes without cpus */
795         for_each_online_node(nid) {
796                 if (uv_node_to_blade[nid] >= 0)
797                         continue;
798                 paddr = node_start_pfn(nid) << PAGE_SHIFT;
799                 paddr = uv_soc_phys_ram_to_gpa(paddr);
800                 pnode = (paddr >> m_val) & pnode_mask;
801                 blade = boot_pnode_to_blade(pnode);
802                 uv_node_to_blade[nid] = blade;
803         }
804
805         map_gru_high(max_pnode);
806         map_mmr_high(max_pnode);
807         map_mmioh_high(max_pnode & pnode_io_mask);
808
809         uv_cpu_init();
810         uv_scir_register_cpu_notifier();
811         uv_register_nmi_notifier();
812         proc_mkdir("sgi_uv", NULL);
813
814         /* register Legacy VGA I/O redirection handler */
815         pci_register_set_vga_state(uv_set_vga_state);
816
817         /*
818          * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
819          * EFI is not enabled in the kdump kernel.
820          */
821         if (is_kdump_kernel())
822                 reboot_type = BOOT_ACPI;
823 }