Merge git://git.kernel.org/pub/scm/linux/kernel/git/pkl/squashfs-linus
[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 #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 struct apic apic_x2apic_uv_x;
62
63 static unsigned long __init uv_early_read_mmr(unsigned long addr)
64 {
65         unsigned long val, *mmr;
66
67         mmr = early_ioremap(UV_LOCAL_MMR_BASE | addr, sizeof(*mmr));
68         val = *mmr;
69         early_iounmap(mmr, sizeof(*mmr));
70         return val;
71 }
72
73 static inline bool is_GRU_range(u64 start, u64 end)
74 {
75         return start >= gru_start_paddr && end <= gru_end_paddr;
76 }
77
78 static bool uv_is_untracked_pat_range(u64 start, u64 end)
79 {
80         return is_ISA_range(start, end) || is_GRU_range(start, end);
81 }
82
83 static int __init early_get_pnodeid(void)
84 {
85         union uvh_node_id_u node_id;
86         union uvh_rh_gam_config_mmr_u  m_n_config;
87         int pnode;
88
89         /* Currently, all blades have same revision number */
90         node_id.v = uv_early_read_mmr(UVH_NODE_ID);
91         m_n_config.v = uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR);
92         uv_min_hub_revision_id = node_id.s.revision;
93
94         pnode = (node_id.s.node_id >> 1) & ((1 << m_n_config.s.n_skt) - 1);
95         return pnode;
96 }
97
98 static void __init early_get_apic_pnode_shift(void)
99 {
100         uvh_apicid.v = uv_early_read_mmr(UVH_APICID);
101         if (!uvh_apicid.v)
102                 /*
103                  * Old bios, use default value
104                  */
105                 uvh_apicid.s.pnode_shift = UV_APIC_PNODE_SHIFT;
106 }
107
108 /*
109  * Add an extra bit as dictated by bios to the destination apicid of
110  * interrupts potentially passing through the UV HUB.  This prevents
111  * a deadlock between interrupts and IO port operations.
112  */
113 static void __init uv_set_apicid_hibit(void)
114 {
115         union uvh_lb_target_physical_apic_id_mask_u apicid_mask;
116
117         apicid_mask.v = uv_early_read_mmr(UVH_LB_TARGET_PHYSICAL_APIC_ID_MASK);
118         uv_apicid_hibits = apicid_mask.s.bit_enables & UV_APICID_HIBIT_MASK;
119 }
120
121 static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
122 {
123         int pnodeid;
124
125         if (!strcmp(oem_id, "SGI")) {
126                 pnodeid = early_get_pnodeid();
127                 early_get_apic_pnode_shift();
128                 x86_platform.is_untracked_pat_range =  uv_is_untracked_pat_range;
129                 x86_platform.nmi_init = uv_nmi_init;
130                 if (!strcmp(oem_table_id, "UVL"))
131                         uv_system_type = UV_LEGACY_APIC;
132                 else if (!strcmp(oem_table_id, "UVX"))
133                         uv_system_type = UV_X2APIC;
134                 else if (!strcmp(oem_table_id, "UVH")) {
135                         __this_cpu_write(x2apic_extra_bits,
136                                 pnodeid << uvh_apicid.s.pnode_shift);
137                         uv_system_type = UV_NON_UNIQUE_APIC;
138                         uv_set_apicid_hibit();
139                         return 1;
140                 }
141         }
142         return 0;
143 }
144
145 enum uv_system_type get_uv_system_type(void)
146 {
147         return uv_system_type;
148 }
149
150 int is_uv_system(void)
151 {
152         return uv_system_type != UV_NONE;
153 }
154 EXPORT_SYMBOL_GPL(is_uv_system);
155
156 DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
157 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
158
159 struct uv_blade_info *uv_blade_info;
160 EXPORT_SYMBOL_GPL(uv_blade_info);
161
162 short *uv_node_to_blade;
163 EXPORT_SYMBOL_GPL(uv_node_to_blade);
164
165 short *uv_cpu_to_blade;
166 EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
167
168 short uv_possible_blades;
169 EXPORT_SYMBOL_GPL(uv_possible_blades);
170
171 unsigned long sn_rtc_cycles_per_second;
172 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
173
174 static const struct cpumask *uv_target_cpus(void)
175 {
176         return cpu_online_mask;
177 }
178
179 static void uv_vector_allocation_domain(int cpu, struct cpumask *retmask)
180 {
181         cpumask_clear(retmask);
182         cpumask_set_cpu(cpu, retmask);
183 }
184
185 static int __cpuinit uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
186 {
187 #ifdef CONFIG_SMP
188         unsigned long val;
189         int pnode;
190
191         pnode = uv_apicid_to_pnode(phys_apicid);
192         phys_apicid |= uv_apicid_hibits;
193         val = (1UL << UVH_IPI_INT_SEND_SHFT) |
194             (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
195             ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
196             APIC_DM_INIT;
197         uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
198         mdelay(10);
199
200         val = (1UL << UVH_IPI_INT_SEND_SHFT) |
201             (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
202             ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
203             APIC_DM_STARTUP;
204         uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
205
206         atomic_set(&init_deasserted, 1);
207 #endif
208         return 0;
209 }
210
211 static void uv_send_IPI_one(int cpu, int vector)
212 {
213         unsigned long apicid;
214         int pnode;
215
216         apicid = per_cpu(x86_cpu_to_apicid, cpu);
217         pnode = uv_apicid_to_pnode(apicid);
218         uv_hub_send_ipi(pnode, apicid, vector);
219 }
220
221 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
222 {
223         unsigned int cpu;
224
225         for_each_cpu(cpu, mask)
226                 uv_send_IPI_one(cpu, vector);
227 }
228
229 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
230 {
231         unsigned int this_cpu = smp_processor_id();
232         unsigned int cpu;
233
234         for_each_cpu(cpu, mask) {
235                 if (cpu != this_cpu)
236                         uv_send_IPI_one(cpu, vector);
237         }
238 }
239
240 static void uv_send_IPI_allbutself(int vector)
241 {
242         unsigned int this_cpu = smp_processor_id();
243         unsigned int cpu;
244
245         for_each_online_cpu(cpu) {
246                 if (cpu != this_cpu)
247                         uv_send_IPI_one(cpu, vector);
248         }
249 }
250
251 static void uv_send_IPI_all(int vector)
252 {
253         uv_send_IPI_mask(cpu_online_mask, vector);
254 }
255
256 static int uv_apic_id_registered(void)
257 {
258         return 1;
259 }
260
261 static void uv_init_apic_ldr(void)
262 {
263 }
264
265 static unsigned int uv_cpu_mask_to_apicid(const struct cpumask *cpumask)
266 {
267         /*
268          * We're using fixed IRQ delivery, can only return one phys APIC ID.
269          * May as well be the first.
270          */
271         int cpu = cpumask_first(cpumask);
272
273         if ((unsigned)cpu < nr_cpu_ids)
274                 return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
275         else
276                 return BAD_APICID;
277 }
278
279 static unsigned int
280 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
281                           const struct cpumask *andmask)
282 {
283         int cpu;
284
285         /*
286          * We're using fixed IRQ delivery, can only return one phys APIC ID.
287          * May as well be the first.
288          */
289         for_each_cpu_and(cpu, cpumask, andmask) {
290                 if (cpumask_test_cpu(cpu, cpu_online_mask))
291                         break;
292         }
293         return per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
294 }
295
296 static unsigned int x2apic_get_apic_id(unsigned long x)
297 {
298         unsigned int id;
299
300         WARN_ON(preemptible() && num_online_cpus() > 1);
301         id = x | __this_cpu_read(x2apic_extra_bits);
302
303         return id;
304 }
305
306 static unsigned long set_apic_id(unsigned int id)
307 {
308         unsigned long x;
309
310         /* maskout x2apic_extra_bits ? */
311         x = id;
312         return x;
313 }
314
315 static unsigned int uv_read_apic_id(void)
316 {
317
318         return x2apic_get_apic_id(apic_read(APIC_ID));
319 }
320
321 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
322 {
323         return uv_read_apic_id() >> index_msb;
324 }
325
326 static void uv_send_IPI_self(int vector)
327 {
328         apic_write(APIC_SELF_IPI, vector);
329 }
330
331 static int uv_probe(void)
332 {
333         return apic == &apic_x2apic_uv_x;
334 }
335
336 static struct apic __refdata apic_x2apic_uv_x = {
337
338         .name                           = "UV large system",
339         .probe                          = uv_probe,
340         .acpi_madt_oem_check            = uv_acpi_madt_oem_check,
341         .apic_id_registered             = uv_apic_id_registered,
342
343         .irq_delivery_mode              = dest_Fixed,
344         .irq_dest_mode                  = 0, /* physical */
345
346         .target_cpus                    = uv_target_cpus,
347         .disable_esr                    = 0,
348         .dest_logical                   = APIC_DEST_LOGICAL,
349         .check_apicid_used              = NULL,
350         .check_apicid_present           = NULL,
351
352         .vector_allocation_domain       = uv_vector_allocation_domain,
353         .init_apic_ldr                  = uv_init_apic_ldr,
354
355         .ioapic_phys_id_map             = NULL,
356         .setup_apic_routing             = NULL,
357         .multi_timer_check              = NULL,
358         .cpu_present_to_apicid          = default_cpu_present_to_apicid,
359         .apicid_to_cpu_present          = NULL,
360         .setup_portio_remap             = NULL,
361         .check_phys_apicid_present      = default_check_phys_apicid_present,
362         .enable_apic_mode               = NULL,
363         .phys_pkg_id                    = uv_phys_pkg_id,
364         .mps_oem_check                  = NULL,
365
366         .get_apic_id                    = x2apic_get_apic_id,
367         .set_apic_id                    = set_apic_id,
368         .apic_id_mask                   = 0xFFFFFFFFu,
369
370         .cpu_mask_to_apicid             = uv_cpu_mask_to_apicid,
371         .cpu_mask_to_apicid_and         = uv_cpu_mask_to_apicid_and,
372
373         .send_IPI_mask                  = uv_send_IPI_mask,
374         .send_IPI_mask_allbutself       = uv_send_IPI_mask_allbutself,
375         .send_IPI_allbutself            = uv_send_IPI_allbutself,
376         .send_IPI_all                   = uv_send_IPI_all,
377         .send_IPI_self                  = uv_send_IPI_self,
378
379         .wakeup_secondary_cpu           = uv_wakeup_secondary,
380         .trampoline_phys_low            = DEFAULT_TRAMPOLINE_PHYS_LOW,
381         .trampoline_phys_high           = DEFAULT_TRAMPOLINE_PHYS_HIGH,
382         .wait_for_init_deassert         = NULL,
383         .smp_callin_clear_local_apic    = NULL,
384         .inquire_remote_apic            = NULL,
385
386         .read                           = native_apic_msr_read,
387         .write                          = native_apic_msr_write,
388         .icr_read                       = native_x2apic_icr_read,
389         .icr_write                      = native_x2apic_icr_write,
390         .wait_icr_idle                  = native_x2apic_wait_icr_idle,
391         .safe_wait_icr_idle             = native_safe_x2apic_wait_icr_idle,
392 };
393
394 static __cpuinit void set_x2apic_extra_bits(int pnode)
395 {
396         __this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
397 }
398
399 /*
400  * Called on boot cpu.
401  */
402 static __init int boot_pnode_to_blade(int pnode)
403 {
404         int blade;
405
406         for (blade = 0; blade < uv_num_possible_blades(); blade++)
407                 if (pnode == uv_blade_info[blade].pnode)
408                         return blade;
409         BUG();
410 }
411
412 struct redir_addr {
413         unsigned long redirect;
414         unsigned long alias;
415 };
416
417 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
418
419 static __initdata struct redir_addr redir_addrs[] = {
420         {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR},
421         {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR},
422         {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR},
423 };
424
425 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
426 {
427         union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
428         union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
429         int i;
430
431         for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
432                 alias.v = uv_read_local_mmr(redir_addrs[i].alias);
433                 if (alias.s.enable && alias.s.base == 0) {
434                         *size = (1UL << alias.s.m_alias);
435                         redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
436                         *base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
437                         return;
438                 }
439         }
440         *base = *size = 0;
441 }
442
443 enum map_type {map_wb, map_uc};
444
445 static __init void map_high(char *id, unsigned long base, int pshift,
446                         int bshift, int max_pnode, enum map_type map_type)
447 {
448         unsigned long bytes, paddr;
449
450         paddr = base << pshift;
451         bytes = (1UL << bshift) * (max_pnode + 1);
452         printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
453                                                 paddr + bytes);
454         if (map_type == map_uc)
455                 init_extra_mapping_uc(paddr, bytes);
456         else
457                 init_extra_mapping_wb(paddr, bytes);
458
459 }
460 static __init void map_gru_high(int max_pnode)
461 {
462         union uvh_rh_gam_gru_overlay_config_mmr_u gru;
463         int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
464
465         gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
466         if (gru.s.enable) {
467                 map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
468                 gru_start_paddr = ((u64)gru.s.base << shift);
469                 gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
470
471         }
472 }
473
474 static __init void map_mmr_high(int max_pnode)
475 {
476         union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
477         int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
478
479         mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
480         if (mmr.s.enable)
481                 map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
482 }
483
484 static __init void map_mmioh_high(int max_pnode)
485 {
486         union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
487         int shift = UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
488
489         mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
490         if (mmioh.s.enable)
491                 map_high("MMIOH", mmioh.s.base, shift, mmioh.s.m_io,
492                         max_pnode, map_uc);
493 }
494
495 static __init void map_low_mmrs(void)
496 {
497         init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
498         init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
499 }
500
501 static __init void uv_rtc_init(void)
502 {
503         long status;
504         u64 ticks_per_sec;
505
506         status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
507                                         &ticks_per_sec);
508         if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
509                 printk(KERN_WARNING
510                         "unable to determine platform RTC clock frequency, "
511                         "guessing.\n");
512                 /* BIOS gives wrong value for clock freq. so guess */
513                 sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
514         } else
515                 sn_rtc_cycles_per_second = ticks_per_sec;
516 }
517
518 /*
519  * percpu heartbeat timer
520  */
521 static void uv_heartbeat(unsigned long ignored)
522 {
523         struct timer_list *timer = &uv_hub_info->scir.timer;
524         unsigned char bits = uv_hub_info->scir.state;
525
526         /* flip heartbeat bit */
527         bits ^= SCIR_CPU_HEARTBEAT;
528
529         /* is this cpu idle? */
530         if (idle_cpu(raw_smp_processor_id()))
531                 bits &= ~SCIR_CPU_ACTIVITY;
532         else
533                 bits |= SCIR_CPU_ACTIVITY;
534
535         /* update system controller interface reg */
536         uv_set_scir_bits(bits);
537
538         /* enable next timer period */
539         mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
540 }
541
542 static void __cpuinit uv_heartbeat_enable(int cpu)
543 {
544         while (!uv_cpu_hub_info(cpu)->scir.enabled) {
545                 struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
546
547                 uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
548                 setup_timer(timer, uv_heartbeat, cpu);
549                 timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
550                 add_timer_on(timer, cpu);
551                 uv_cpu_hub_info(cpu)->scir.enabled = 1;
552
553                 /* also ensure that boot cpu is enabled */
554                 cpu = 0;
555         }
556 }
557
558 #ifdef CONFIG_HOTPLUG_CPU
559 static void __cpuinit uv_heartbeat_disable(int cpu)
560 {
561         if (uv_cpu_hub_info(cpu)->scir.enabled) {
562                 uv_cpu_hub_info(cpu)->scir.enabled = 0;
563                 del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
564         }
565         uv_set_cpu_scir_bits(cpu, 0xff);
566 }
567
568 /*
569  * cpu hotplug notifier
570  */
571 static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
572                                        unsigned long action, void *hcpu)
573 {
574         long cpu = (long)hcpu;
575
576         switch (action) {
577         case CPU_ONLINE:
578                 uv_heartbeat_enable(cpu);
579                 break;
580         case CPU_DOWN_PREPARE:
581                 uv_heartbeat_disable(cpu);
582                 break;
583         default:
584                 break;
585         }
586         return NOTIFY_OK;
587 }
588
589 static __init void uv_scir_register_cpu_notifier(void)
590 {
591         hotcpu_notifier(uv_scir_cpu_notify, 0);
592 }
593
594 #else /* !CONFIG_HOTPLUG_CPU */
595
596 static __init void uv_scir_register_cpu_notifier(void)
597 {
598 }
599
600 static __init int uv_init_heartbeat(void)
601 {
602         int cpu;
603
604         if (is_uv_system())
605                 for_each_online_cpu(cpu)
606                         uv_heartbeat_enable(cpu);
607         return 0;
608 }
609
610 late_initcall(uv_init_heartbeat);
611
612 #endif /* !CONFIG_HOTPLUG_CPU */
613
614 /* Direct Legacy VGA I/O traffic to designated IOH */
615 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
616                       unsigned int command_bits, bool change_bridge)
617 {
618         int domain, bus, rc;
619
620         PR_DEVEL("devfn %x decode %d cmd %x chg_brdg %d\n",
621                         pdev->devfn, decode, command_bits, change_bridge);
622
623         if (!change_bridge)
624                 return 0;
625
626         if ((command_bits & PCI_COMMAND_IO) == 0)
627                 return 0;
628
629         domain = pci_domain_nr(pdev->bus);
630         bus = pdev->bus->number;
631
632         rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
633         PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
634
635         return rc;
636 }
637
638 /*
639  * Called on each cpu to initialize the per_cpu UV data area.
640  * FIXME: hotplug not supported yet
641  */
642 void __cpuinit uv_cpu_init(void)
643 {
644         /* CPU 0 initilization will be done via uv_system_init. */
645         if (!uv_blade_info)
646                 return;
647
648         uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
649
650         if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
651                 set_x2apic_extra_bits(uv_hub_info->pnode);
652 }
653
654 /*
655  * When NMI is received, print a stack trace.
656  */
657 int uv_handle_nmi(struct notifier_block *self, unsigned long reason, void *data)
658 {
659         unsigned long real_uv_nmi;
660         int bid;
661
662         if (reason != DIE_NMIUNKNOWN)
663                 return NOTIFY_OK;
664
665         if (in_crash_kexec)
666                 /* do nothing if entering the crash kernel */
667                 return NOTIFY_OK;
668
669         /*
670          * Each blade has an MMR that indicates when an NMI has been sent
671          * to cpus on the blade. If an NMI is detected, atomically
672          * clear the MMR and update a per-blade NMI count used to
673          * cause each cpu on the blade to notice a new NMI.
674          */
675         bid = uv_numa_blade_id();
676         real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
677
678         if (unlikely(real_uv_nmi)) {
679                 spin_lock(&uv_blade_info[bid].nmi_lock);
680                 real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
681                 if (real_uv_nmi) {
682                         uv_blade_info[bid].nmi_count++;
683                         uv_write_local_mmr(UVH_NMI_MMR_CLEAR, UV_NMI_PENDING_MASK);
684                 }
685                 spin_unlock(&uv_blade_info[bid].nmi_lock);
686         }
687
688         if (likely(__get_cpu_var(cpu_last_nmi_count) == uv_blade_info[bid].nmi_count))
689                 return NOTIFY_DONE;
690
691         __get_cpu_var(cpu_last_nmi_count) = uv_blade_info[bid].nmi_count;
692
693         /*
694          * Use a lock so only one cpu prints at a time.
695          * This prevents intermixed output.
696          */
697         spin_lock(&uv_nmi_lock);
698         pr_info("UV NMI stack dump cpu %u:\n", smp_processor_id());
699         dump_stack();
700         spin_unlock(&uv_nmi_lock);
701
702         return NOTIFY_STOP;
703 }
704
705 static struct notifier_block uv_dump_stack_nmi_nb = {
706         .notifier_call  = uv_handle_nmi,
707         .priority = NMI_LOCAL_LOW_PRIOR - 1,
708 };
709
710 void uv_register_nmi_notifier(void)
711 {
712         if (register_die_notifier(&uv_dump_stack_nmi_nb))
713                 printk(KERN_WARNING "UV NMI handler failed to register\n");
714 }
715
716 void uv_nmi_init(void)
717 {
718         unsigned int value;
719
720         /*
721          * Unmask NMI on all cpus
722          */
723         value = apic_read(APIC_LVT1) | APIC_DM_NMI;
724         value &= ~APIC_LVT_MASKED;
725         apic_write(APIC_LVT1, value);
726 }
727
728 void __init uv_system_init(void)
729 {
730         union uvh_rh_gam_config_mmr_u  m_n_config;
731         union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
732         union uvh_node_id_u node_id;
733         unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
734         int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val, n_io;
735         int gnode_extra, max_pnode = 0;
736         unsigned long mmr_base, present, paddr;
737         unsigned short pnode_mask, pnode_io_mask;
738
739         map_low_mmrs();
740
741         m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR );
742         m_val = m_n_config.s.m_skt;
743         n_val = m_n_config.s.n_skt;
744         mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
745         n_io = mmioh.s.n_io;
746         mmr_base =
747             uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
748             ~UV_MMR_ENABLE;
749         pnode_mask = (1 << n_val) - 1;
750         pnode_io_mask = (1 << n_io) - 1;
751
752         node_id.v = uv_read_local_mmr(UVH_NODE_ID);
753         gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
754         gnode_upper = ((unsigned long)gnode_extra  << m_val);
755         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",
756                         n_val, m_val, n_io, gnode_upper, gnode_extra, pnode_mask, pnode_io_mask);
757
758         printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
759
760         for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
761                 uv_possible_blades +=
762                   hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
763         printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
764
765         bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
766         uv_blade_info = kzalloc(bytes, GFP_KERNEL);
767         BUG_ON(!uv_blade_info);
768
769         for (blade = 0; blade < uv_num_possible_blades(); blade++)
770                 uv_blade_info[blade].memory_nid = -1;
771
772         get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
773
774         bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
775         uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
776         BUG_ON(!uv_node_to_blade);
777         memset(uv_node_to_blade, 255, bytes);
778
779         bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
780         uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
781         BUG_ON(!uv_cpu_to_blade);
782         memset(uv_cpu_to_blade, 255, bytes);
783
784         blade = 0;
785         for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
786                 present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
787                 for (j = 0; j < 64; j++) {
788                         if (!test_bit(j, &present))
789                                 continue;
790                         pnode = (i * 64 + j) & pnode_mask;
791                         uv_blade_info[blade].pnode = pnode;
792                         uv_blade_info[blade].nr_possible_cpus = 0;
793                         uv_blade_info[blade].nr_online_cpus = 0;
794                         spin_lock_init(&uv_blade_info[blade].nmi_lock);
795                         max_pnode = max(pnode, max_pnode);
796                         blade++;
797                 }
798         }
799
800         uv_bios_init();
801         uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
802                             &sn_region_size, &system_serial_number);
803         uv_rtc_init();
804
805         for_each_present_cpu(cpu) {
806                 int apicid = per_cpu(x86_cpu_to_apicid, cpu);
807
808                 nid = cpu_to_node(cpu);
809                 /*
810                  * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
811                  */
812                 uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
813                 uv_cpu_hub_info(cpu)->apic_pnode_shift = uvh_apicid.s.pnode_shift;
814                 pnode = uv_apicid_to_pnode(apicid);
815                 blade = boot_pnode_to_blade(pnode);
816                 lcpu = uv_blade_info[blade].nr_possible_cpus;
817                 uv_blade_info[blade].nr_possible_cpus++;
818
819                 /* Any node on the blade, else will contain -1. */
820                 uv_blade_info[blade].memory_nid = nid;
821
822                 uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
823                 uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
824                 uv_cpu_hub_info(cpu)->m_val = m_val;
825                 uv_cpu_hub_info(cpu)->n_val = n_val;
826                 uv_cpu_hub_info(cpu)->numa_blade_id = blade;
827                 uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
828                 uv_cpu_hub_info(cpu)->pnode = pnode;
829                 uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
830                 uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
831                 uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
832                 uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
833                 uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
834                 uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
835                 uv_node_to_blade[nid] = blade;
836                 uv_cpu_to_blade[cpu] = blade;
837         }
838
839         /* Add blade/pnode info for nodes without cpus */
840         for_each_online_node(nid) {
841                 if (uv_node_to_blade[nid] >= 0)
842                         continue;
843                 paddr = node_start_pfn(nid) << PAGE_SHIFT;
844                 paddr = uv_soc_phys_ram_to_gpa(paddr);
845                 pnode = (paddr >> m_val) & pnode_mask;
846                 blade = boot_pnode_to_blade(pnode);
847                 uv_node_to_blade[nid] = blade;
848         }
849
850         map_gru_high(max_pnode);
851         map_mmr_high(max_pnode);
852         map_mmioh_high(max_pnode & pnode_io_mask);
853
854         uv_cpu_init();
855         uv_scir_register_cpu_notifier();
856         uv_register_nmi_notifier();
857         proc_mkdir("sgi_uv", NULL);
858
859         /* register Legacy VGA I/O redirection handler */
860         pci_register_set_vga_state(uv_set_vga_state);
861
862         /*
863          * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
864          * EFI is not enabled in the kdump kernel.
865          */
866         if (is_kdump_kernel())
867                 reboot_type = BOOT_ACPI;
868 }
869
870 apic_driver(apic_x2apic_uv_x);