init from v2.6.32.60
[mach-moxart.git] / arch / x86 / kernel / apic / x2apic_uv_x.c
blob95b0712e54c6cbbd9feaad5c243bdf88f7376cad
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.
6 * SGI UV APIC functions (note: not an Intel compatible APIC)
8 * Copyright (C) 2007-2008 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/cpu.h>
21 #include <linux/init.h>
22 #include <linux/io.h>
24 #include <asm/uv/uv_mmrs.h>
25 #include <asm/uv/uv_hub.h>
26 #include <asm/current.h>
27 #include <asm/pgtable.h>
28 #include <asm/uv/bios.h>
29 #include <asm/uv/uv.h>
30 #include <asm/apic.h>
31 #include <asm/ipi.h>
32 #include <asm/smp.h>
34 DEFINE_PER_CPU(int, x2apic_extra_bits);
36 static enum uv_system_type uv_system_type;
38 static int early_get_nodeid(void)
40 union uvh_node_id_u node_id;
41 unsigned long *mmr;
43 mmr = early_ioremap(UV_LOCAL_MMR_BASE | UVH_NODE_ID, sizeof(*mmr));
44 node_id.v = *mmr;
45 early_iounmap(mmr, sizeof(*mmr));
46 return node_id.s.node_id;
49 static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
51 if (!strcmp(oem_id, "SGI")) {
52 if (!strcmp(oem_table_id, "UVL"))
53 uv_system_type = UV_LEGACY_APIC;
54 else if (!strcmp(oem_table_id, "UVX"))
55 uv_system_type = UV_X2APIC;
56 else if (!strcmp(oem_table_id, "UVH")) {
57 __get_cpu_var(x2apic_extra_bits) =
58 early_get_nodeid() << (UV_APIC_PNODE_SHIFT - 1);
59 uv_system_type = UV_NON_UNIQUE_APIC;
60 return 1;
63 return 0;
66 enum uv_system_type get_uv_system_type(void)
68 return uv_system_type;
71 int is_uv_system(void)
73 return uv_system_type != UV_NONE;
75 EXPORT_SYMBOL_GPL(is_uv_system);
77 DEFINE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
78 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info);
80 struct uv_blade_info *uv_blade_info;
81 EXPORT_SYMBOL_GPL(uv_blade_info);
83 short *uv_node_to_blade;
84 EXPORT_SYMBOL_GPL(uv_node_to_blade);
86 short *uv_cpu_to_blade;
87 EXPORT_SYMBOL_GPL(uv_cpu_to_blade);
89 short uv_possible_blades;
90 EXPORT_SYMBOL_GPL(uv_possible_blades);
92 unsigned long sn_rtc_cycles_per_second;
93 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
95 /* Start with all IRQs pointing to boot CPU. IRQ balancing will shift them. */
97 static const struct cpumask *uv_target_cpus(void)
99 return cpumask_of(0);
102 static void uv_vector_allocation_domain(int cpu, struct cpumask *retmask)
104 cpumask_clear(retmask);
105 cpumask_set_cpu(cpu, retmask);
108 static int __cpuinit uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
110 #ifdef CONFIG_SMP
111 unsigned long val;
112 int pnode;
114 pnode = uv_apicid_to_pnode(phys_apicid);
115 val = (1UL << UVH_IPI_INT_SEND_SHFT) |
116 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
117 ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
118 APIC_DM_INIT;
119 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
121 val = (1UL << UVH_IPI_INT_SEND_SHFT) |
122 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
123 ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
124 APIC_DM_STARTUP;
125 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
127 atomic_set(&init_deasserted, 1);
128 #endif
129 return 0;
132 static void uv_send_IPI_one(int cpu, int vector)
134 unsigned long apicid;
135 int pnode;
137 apicid = per_cpu(x86_cpu_to_apicid, cpu);
138 pnode = uv_apicid_to_pnode(apicid);
139 uv_hub_send_ipi(pnode, apicid, vector);
142 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
144 unsigned int cpu;
146 for_each_cpu(cpu, mask)
147 uv_send_IPI_one(cpu, vector);
150 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
152 unsigned int this_cpu = smp_processor_id();
153 unsigned int cpu;
155 for_each_cpu(cpu, mask) {
156 if (cpu != this_cpu)
157 uv_send_IPI_one(cpu, vector);
161 static void uv_send_IPI_allbutself(int vector)
163 unsigned int this_cpu = smp_processor_id();
164 unsigned int cpu;
166 for_each_online_cpu(cpu) {
167 if (cpu != this_cpu)
168 uv_send_IPI_one(cpu, vector);
172 static void uv_send_IPI_all(int vector)
174 uv_send_IPI_mask(cpu_online_mask, vector);
177 static int uv_apic_id_registered(void)
179 return 1;
182 static void uv_init_apic_ldr(void)
186 static unsigned int uv_cpu_mask_to_apicid(const struct cpumask *cpumask)
189 * We're using fixed IRQ delivery, can only return one phys APIC ID.
190 * May as well be the first.
192 int cpu = cpumask_first(cpumask);
194 if ((unsigned)cpu < nr_cpu_ids)
195 return per_cpu(x86_cpu_to_apicid, cpu);
196 else
197 return BAD_APICID;
200 static unsigned int
201 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
202 const struct cpumask *andmask)
204 int cpu;
207 * We're using fixed IRQ delivery, can only return one phys APIC ID.
208 * May as well be the first.
210 for_each_cpu_and(cpu, cpumask, andmask) {
211 if (cpumask_test_cpu(cpu, cpu_online_mask))
212 break;
214 if (cpu < nr_cpu_ids)
215 return per_cpu(x86_cpu_to_apicid, cpu);
217 return BAD_APICID;
220 static unsigned int x2apic_get_apic_id(unsigned long x)
222 unsigned int id;
224 WARN_ON(preemptible() && num_online_cpus() > 1);
225 id = x | __get_cpu_var(x2apic_extra_bits);
227 return id;
230 static unsigned long set_apic_id(unsigned int id)
232 unsigned long x;
234 /* maskout x2apic_extra_bits ? */
235 x = id;
236 return x;
239 static unsigned int uv_read_apic_id(void)
242 return x2apic_get_apic_id(apic_read(APIC_ID));
245 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
247 return uv_read_apic_id() >> index_msb;
250 static void uv_send_IPI_self(int vector)
252 apic_write(APIC_SELF_IPI, vector);
255 struct apic __refdata apic_x2apic_uv_x = {
257 .name = "UV large system",
258 .probe = NULL,
259 .acpi_madt_oem_check = uv_acpi_madt_oem_check,
260 .apic_id_registered = uv_apic_id_registered,
262 .irq_delivery_mode = dest_Fixed,
263 .irq_dest_mode = 0, /* physical */
265 .target_cpus = uv_target_cpus,
266 .disable_esr = 0,
267 .dest_logical = APIC_DEST_LOGICAL,
268 .check_apicid_used = NULL,
269 .check_apicid_present = NULL,
271 .vector_allocation_domain = uv_vector_allocation_domain,
272 .init_apic_ldr = uv_init_apic_ldr,
274 .ioapic_phys_id_map = NULL,
275 .setup_apic_routing = NULL,
276 .multi_timer_check = NULL,
277 .apicid_to_node = NULL,
278 .cpu_to_logical_apicid = NULL,
279 .cpu_present_to_apicid = default_cpu_present_to_apicid,
280 .apicid_to_cpu_present = NULL,
281 .setup_portio_remap = NULL,
282 .check_phys_apicid_present = default_check_phys_apicid_present,
283 .enable_apic_mode = NULL,
284 .phys_pkg_id = uv_phys_pkg_id,
285 .mps_oem_check = NULL,
287 .get_apic_id = x2apic_get_apic_id,
288 .set_apic_id = set_apic_id,
289 .apic_id_mask = 0xFFFFFFFFu,
291 .cpu_mask_to_apicid = uv_cpu_mask_to_apicid,
292 .cpu_mask_to_apicid_and = uv_cpu_mask_to_apicid_and,
294 .send_IPI_mask = uv_send_IPI_mask,
295 .send_IPI_mask_allbutself = uv_send_IPI_mask_allbutself,
296 .send_IPI_allbutself = uv_send_IPI_allbutself,
297 .send_IPI_all = uv_send_IPI_all,
298 .send_IPI_self = uv_send_IPI_self,
300 .wakeup_secondary_cpu = uv_wakeup_secondary,
301 .trampoline_phys_low = DEFAULT_TRAMPOLINE_PHYS_LOW,
302 .trampoline_phys_high = DEFAULT_TRAMPOLINE_PHYS_HIGH,
303 .wait_for_init_deassert = NULL,
304 .smp_callin_clear_local_apic = NULL,
305 .inquire_remote_apic = NULL,
307 .read = native_apic_msr_read,
308 .write = native_apic_msr_write,
309 .icr_read = native_x2apic_icr_read,
310 .icr_write = native_x2apic_icr_write,
311 .wait_icr_idle = native_x2apic_wait_icr_idle,
312 .safe_wait_icr_idle = native_safe_x2apic_wait_icr_idle,
315 static __cpuinit void set_x2apic_extra_bits(int pnode)
317 __get_cpu_var(x2apic_extra_bits) = (pnode << 6);
321 * Called on boot cpu.
323 static __init int boot_pnode_to_blade(int pnode)
325 int blade;
327 for (blade = 0; blade < uv_num_possible_blades(); blade++)
328 if (pnode == uv_blade_info[blade].pnode)
329 return blade;
330 BUG();
333 struct redir_addr {
334 unsigned long redirect;
335 unsigned long alias;
338 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
340 static __initdata struct redir_addr redir_addrs[] = {
341 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_SI_ALIAS0_OVERLAY_CONFIG},
342 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_SI_ALIAS1_OVERLAY_CONFIG},
343 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_SI_ALIAS2_OVERLAY_CONFIG},
346 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
348 union uvh_si_alias0_overlay_config_u alias;
349 union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
350 int i;
352 for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
353 alias.v = uv_read_local_mmr(redir_addrs[i].alias);
354 if (alias.s.enable && alias.s.base == 0) {
355 *size = (1UL << alias.s.m_alias);
356 redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
357 *base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
358 return;
361 *base = *size = 0;
364 enum map_type {map_wb, map_uc};
366 static __init void map_high(char *id, unsigned long base, int pshift,
367 int bshift, int max_pnode, enum map_type map_type)
369 unsigned long bytes, paddr;
371 paddr = base << pshift;
372 bytes = (1UL << bshift) * (max_pnode + 1);
373 printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
374 paddr + bytes);
375 if (map_type == map_uc)
376 init_extra_mapping_uc(paddr, bytes);
377 else
378 init_extra_mapping_wb(paddr, bytes);
381 static __init void map_gru_high(int max_pnode)
383 union uvh_rh_gam_gru_overlay_config_mmr_u gru;
384 int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
386 gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
387 if (gru.s.enable)
388 map_high("GRU", gru.s.base, shift, shift, max_pnode, map_wb);
391 static __init void map_mmr_high(int max_pnode)
393 union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
394 int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
396 mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
397 if (mmr.s.enable)
398 map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
401 static __init void map_mmioh_high(int max_pnode)
403 union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
404 int shift = UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
406 mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
407 if (mmioh.s.enable)
408 map_high("MMIOH", mmioh.s.base, shift, mmioh.s.m_io,
409 max_pnode, map_uc);
412 static __init void uv_rtc_init(void)
414 long status;
415 u64 ticks_per_sec;
417 status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
418 &ticks_per_sec);
419 if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
420 printk(KERN_WARNING
421 "unable to determine platform RTC clock frequency, "
422 "guessing.\n");
423 /* BIOS gives wrong value for clock freq. so guess */
424 sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
425 } else
426 sn_rtc_cycles_per_second = ticks_per_sec;
430 * percpu heartbeat timer
432 static void uv_heartbeat(unsigned long ignored)
434 struct timer_list *timer = &uv_hub_info->scir.timer;
435 unsigned char bits = uv_hub_info->scir.state;
437 /* flip heartbeat bit */
438 bits ^= SCIR_CPU_HEARTBEAT;
440 /* is this cpu idle? */
441 if (idle_cpu(raw_smp_processor_id()))
442 bits &= ~SCIR_CPU_ACTIVITY;
443 else
444 bits |= SCIR_CPU_ACTIVITY;
446 /* update system controller interface reg */
447 uv_set_scir_bits(bits);
449 /* enable next timer period */
450 mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
453 static void __cpuinit uv_heartbeat_enable(int cpu)
455 if (!uv_cpu_hub_info(cpu)->scir.enabled) {
456 struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
458 uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
459 setup_timer(timer, uv_heartbeat, cpu);
460 timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
461 add_timer_on(timer, cpu);
462 uv_cpu_hub_info(cpu)->scir.enabled = 1;
465 /* check boot cpu */
466 if (!uv_cpu_hub_info(0)->scir.enabled)
467 uv_heartbeat_enable(0);
470 #ifdef CONFIG_HOTPLUG_CPU
471 static void __cpuinit uv_heartbeat_disable(int cpu)
473 if (uv_cpu_hub_info(cpu)->scir.enabled) {
474 uv_cpu_hub_info(cpu)->scir.enabled = 0;
475 del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
477 uv_set_cpu_scir_bits(cpu, 0xff);
481 * cpu hotplug notifier
483 static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
484 unsigned long action, void *hcpu)
486 long cpu = (long)hcpu;
488 switch (action) {
489 case CPU_ONLINE:
490 uv_heartbeat_enable(cpu);
491 break;
492 case CPU_DOWN_PREPARE:
493 uv_heartbeat_disable(cpu);
494 break;
495 default:
496 break;
498 return NOTIFY_OK;
501 static __init void uv_scir_register_cpu_notifier(void)
503 hotcpu_notifier(uv_scir_cpu_notify, 0);
506 #else /* !CONFIG_HOTPLUG_CPU */
508 static __init void uv_scir_register_cpu_notifier(void)
512 static __init int uv_init_heartbeat(void)
514 int cpu;
516 if (is_uv_system())
517 for_each_online_cpu(cpu)
518 uv_heartbeat_enable(cpu);
519 return 0;
522 late_initcall(uv_init_heartbeat);
524 #endif /* !CONFIG_HOTPLUG_CPU */
527 * Called on each cpu to initialize the per_cpu UV data area.
528 * FIXME: hotplug not supported yet
530 void __cpuinit uv_cpu_init(void)
532 /* CPU 0 initilization will be done via uv_system_init. */
533 if (!uv_blade_info)
534 return;
536 uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
538 if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
539 set_x2apic_extra_bits(uv_hub_info->pnode);
543 void __init uv_system_init(void)
545 union uvh_si_addr_map_config_u m_n_config;
546 union uvh_node_id_u node_id;
547 unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
548 int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val;
549 int gnode_extra, max_pnode = 0;
550 unsigned long mmr_base, present, paddr;
551 unsigned short pnode_mask;
553 m_n_config.v = uv_read_local_mmr(UVH_SI_ADDR_MAP_CONFIG);
554 m_val = m_n_config.s.m_skt;
555 n_val = m_n_config.s.n_skt;
556 mmr_base =
557 uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
558 ~UV_MMR_ENABLE;
559 pnode_mask = (1 << n_val) - 1;
560 node_id.v = uv_read_local_mmr(UVH_NODE_ID);
561 gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
562 gnode_upper = ((unsigned long)gnode_extra << m_val);
563 printk(KERN_DEBUG "UV: N %d, M %d, gnode_upper 0x%lx, gnode_extra 0x%x\n",
564 n_val, m_val, gnode_upper, gnode_extra);
566 printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
568 for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
569 uv_possible_blades +=
570 hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
571 printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
573 bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
574 uv_blade_info = kmalloc(bytes, GFP_KERNEL);
575 BUG_ON(!uv_blade_info);
576 for (blade = 0; blade < uv_num_possible_blades(); blade++)
577 uv_blade_info[blade].memory_nid = -1;
579 get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
581 bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
582 uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
583 BUG_ON(!uv_node_to_blade);
584 memset(uv_node_to_blade, 255, bytes);
586 bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
587 uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
588 BUG_ON(!uv_cpu_to_blade);
589 memset(uv_cpu_to_blade, 255, bytes);
591 blade = 0;
592 for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
593 present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
594 for (j = 0; j < 64; j++) {
595 if (!test_bit(j, &present))
596 continue;
597 pnode = (i * 64 + j);
598 uv_blade_info[blade].pnode = pnode;
599 uv_blade_info[blade].nr_possible_cpus = 0;
600 uv_blade_info[blade].nr_online_cpus = 0;
601 max_pnode = max(pnode, max_pnode);
602 blade++;
606 uv_bios_init();
607 uv_bios_get_sn_info(0, &uv_type, &sn_partition_id,
608 &sn_coherency_id, &sn_region_size);
609 uv_rtc_init();
611 for_each_present_cpu(cpu) {
612 int apicid = per_cpu(x86_cpu_to_apicid, cpu);
614 nid = cpu_to_node(cpu);
615 pnode = uv_apicid_to_pnode(apicid);
616 blade = boot_pnode_to_blade(pnode);
617 lcpu = uv_blade_info[blade].nr_possible_cpus;
618 uv_blade_info[blade].nr_possible_cpus++;
620 /* Any node on the blade, else will contain -1. */
621 uv_blade_info[blade].memory_nid = nid;
623 uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
624 uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
625 uv_cpu_hub_info(cpu)->m_val = m_val;
626 uv_cpu_hub_info(cpu)->n_val = n_val;
627 uv_cpu_hub_info(cpu)->numa_blade_id = blade;
628 uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
629 uv_cpu_hub_info(cpu)->pnode = pnode;
630 uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
631 uv_cpu_hub_info(cpu)->gpa_mask = (1UL << (m_val + n_val)) - 1;
632 uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
633 uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
634 uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
635 uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
636 uv_cpu_hub_info(cpu)->scir.offset = uv_scir_offset(apicid);
637 uv_node_to_blade[nid] = blade;
638 uv_cpu_to_blade[cpu] = blade;
641 /* Add blade/pnode info for nodes without cpus */
642 for_each_online_node(nid) {
643 if (uv_node_to_blade[nid] >= 0)
644 continue;
645 paddr = node_start_pfn(nid) << PAGE_SHIFT;
646 paddr = uv_soc_phys_ram_to_gpa(paddr);
647 pnode = (paddr >> m_val) & pnode_mask;
648 blade = boot_pnode_to_blade(pnode);
649 uv_node_to_blade[nid] = blade;
652 map_gru_high(max_pnode);
653 map_mmr_high(max_pnode);
654 map_mmioh_high(max_pnode);
656 uv_cpu_init();
657 uv_scir_register_cpu_notifier();
658 proc_mkdir("sgi_uv", NULL);