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
6 * SGI UV APIC functions (note: not an Intel compatible APIC)
8 * Copyright (C) 2007-2008 Silicon Graphics, Inc. All rights reserved.
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>
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>
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;
43 mmr = early_ioremap(UV_LOCAL_MMR_BASE | UVH_NODE_ID, sizeof(*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;
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)
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)
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) |
119 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
122 val = (1UL << UVH_IPI_INT_SEND_SHFT) |
123 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
124 ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
126 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
128 atomic_set(&init_deasserted, 1);
133 static void uv_send_IPI_one(int cpu, int vector)
135 unsigned long apicid;
138 apicid = per_cpu(x86_cpu_to_apicid, cpu);
139 pnode = uv_apicid_to_pnode(apicid);
140 uv_hub_send_ipi(pnode, apicid, vector);
143 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
147 for_each_cpu(cpu, mask)
148 uv_send_IPI_one(cpu, vector);
151 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
153 unsigned int this_cpu = smp_processor_id();
156 for_each_cpu(cpu, mask) {
158 uv_send_IPI_one(cpu, vector);
162 static void uv_send_IPI_allbutself(int vector)
164 unsigned int this_cpu = smp_processor_id();
167 for_each_online_cpu(cpu) {
169 uv_send_IPI_one(cpu, vector);
173 static void uv_send_IPI_all(int vector)
175 uv_send_IPI_mask(cpu_online_mask, vector);
178 static int uv_apic_id_registered(void)
183 static void uv_init_apic_ldr(void)
187 static unsigned int uv_cpu_mask_to_apicid(const struct cpumask *cpumask)
190 * We're using fixed IRQ delivery, can only return one phys APIC ID.
191 * May as well be the first.
193 int cpu = cpumask_first(cpumask);
195 if ((unsigned)cpu < nr_cpu_ids)
196 return per_cpu(x86_cpu_to_apicid, cpu);
202 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
203 const struct cpumask *andmask)
208 * We're using fixed IRQ delivery, can only return one phys APIC ID.
209 * May as well be the first.
211 for_each_cpu_and(cpu, cpumask, andmask) {
212 if (cpumask_test_cpu(cpu, cpu_online_mask))
215 if (cpu < nr_cpu_ids)
216 return per_cpu(x86_cpu_to_apicid, cpu);
221 static unsigned int x2apic_get_apic_id(unsigned long x)
225 WARN_ON(preemptible() && num_online_cpus() > 1);
226 id = x | __get_cpu_var(x2apic_extra_bits);
231 static unsigned long set_apic_id(unsigned int id)
235 /* maskout x2apic_extra_bits ? */
240 static unsigned int uv_read_apic_id(void)
243 return x2apic_get_apic_id(apic_read(APIC_ID));
246 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
248 return uv_read_apic_id() >> index_msb;
251 static void uv_send_IPI_self(int vector)
253 apic_write(APIC_SELF_IPI, vector);
256 struct apic __refdata apic_x2apic_uv_x = {
258 .name = "UV large system",
260 .acpi_madt_oem_check = uv_acpi_madt_oem_check,
261 .apic_id_registered = uv_apic_id_registered,
263 .irq_delivery_mode = dest_Fixed,
264 .irq_dest_mode = 0, /* physical */
266 .target_cpus = uv_target_cpus,
268 .dest_logical = APIC_DEST_LOGICAL,
269 .check_apicid_used = NULL,
270 .check_apicid_present = NULL,
272 .vector_allocation_domain = uv_vector_allocation_domain,
273 .init_apic_ldr = uv_init_apic_ldr,
275 .ioapic_phys_id_map = NULL,
276 .setup_apic_routing = NULL,
277 .multi_timer_check = NULL,
278 .apicid_to_node = NULL,
279 .cpu_to_logical_apicid = NULL,
280 .cpu_present_to_apicid = default_cpu_present_to_apicid,
281 .apicid_to_cpu_present = NULL,
282 .setup_portio_remap = NULL,
283 .check_phys_apicid_present = default_check_phys_apicid_present,
284 .enable_apic_mode = NULL,
285 .phys_pkg_id = uv_phys_pkg_id,
286 .mps_oem_check = NULL,
288 .get_apic_id = x2apic_get_apic_id,
289 .set_apic_id = set_apic_id,
290 .apic_id_mask = 0xFFFFFFFFu,
292 .cpu_mask_to_apicid = uv_cpu_mask_to_apicid,
293 .cpu_mask_to_apicid_and = uv_cpu_mask_to_apicid_and,
295 .send_IPI_mask = uv_send_IPI_mask,
296 .send_IPI_mask_allbutself = uv_send_IPI_mask_allbutself,
297 .send_IPI_allbutself = uv_send_IPI_allbutself,
298 .send_IPI_all = uv_send_IPI_all,
299 .send_IPI_self = uv_send_IPI_self,
301 .wakeup_secondary_cpu = uv_wakeup_secondary,
302 .trampoline_phys_low = DEFAULT_TRAMPOLINE_PHYS_LOW,
303 .trampoline_phys_high = DEFAULT_TRAMPOLINE_PHYS_HIGH,
304 .wait_for_init_deassert = NULL,
305 .smp_callin_clear_local_apic = NULL,
306 .inquire_remote_apic = NULL,
308 .read = native_apic_msr_read,
309 .write = native_apic_msr_write,
310 .icr_read = native_x2apic_icr_read,
311 .icr_write = native_x2apic_icr_write,
312 .wait_icr_idle = native_x2apic_wait_icr_idle,
313 .safe_wait_icr_idle = native_safe_x2apic_wait_icr_idle,
316 static __cpuinit void set_x2apic_extra_bits(int pnode)
318 __get_cpu_var(x2apic_extra_bits) = (pnode << 6);
322 * Called on boot cpu.
324 static __init int boot_pnode_to_blade(int pnode)
328 for (blade = 0; blade < uv_num_possible_blades(); blade++)
329 if (pnode == uv_blade_info[blade].pnode)
335 unsigned long redirect;
339 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
341 static __initdata struct redir_addr redir_addrs[] = {
342 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR, UVH_SI_ALIAS0_OVERLAY_CONFIG},
343 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR, UVH_SI_ALIAS1_OVERLAY_CONFIG},
344 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR, UVH_SI_ALIAS2_OVERLAY_CONFIG},
347 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
349 union uvh_si_alias0_overlay_config_u alias;
350 union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
353 for (i = 0; i < ARRAY_SIZE(redir_addrs); i++) {
354 alias.v = uv_read_local_mmr(redir_addrs[i].alias);
355 if (alias.s.base == 0) {
356 *size = (1UL << alias.s.m_alias);
357 redirect.v = uv_read_local_mmr(redir_addrs[i].redirect);
358 *base = (unsigned long)redirect.s.dest_base << DEST_SHIFT;
365 enum map_type {map_wb, map_uc};
367 static __init void map_high(char *id, unsigned long base, int shift,
368 int max_pnode, enum map_type map_type)
370 unsigned long bytes, paddr;
372 paddr = base << shift;
373 bytes = (1UL << shift) * (max_pnode + 1);
374 printk(KERN_INFO "UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr,
376 if (map_type == map_uc)
377 init_extra_mapping_uc(paddr, bytes);
379 init_extra_mapping_wb(paddr, bytes);
382 static __init void map_gru_high(int max_pnode)
384 union uvh_rh_gam_gru_overlay_config_mmr_u gru;
385 int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
387 gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
389 map_high("GRU", gru.s.base, shift, max_pnode, map_wb);
392 static __init void map_mmr_high(int max_pnode)
394 union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
395 int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
397 mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
399 map_high("MMR", mmr.s.base, shift, max_pnode, map_uc);
402 static __init void map_mmioh_high(int max_pnode)
404 union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
405 int shift = UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
407 mmioh.v = uv_read_local_mmr(UVH_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR);
409 map_high("MMIOH", mmioh.s.base, shift, max_pnode, map_uc);
412 static __init void map_low_mmrs(void)
414 init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
415 init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
418 static __init void uv_rtc_init(void)
423 status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
425 if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
427 "unable to determine platform RTC clock frequency, "
429 /* BIOS gives wrong value for clock freq. so guess */
430 sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
432 sn_rtc_cycles_per_second = ticks_per_sec;
436 * percpu heartbeat timer
438 static void uv_heartbeat(unsigned long ignored)
440 struct timer_list *timer = &uv_hub_info->scir.timer;
441 unsigned char bits = uv_hub_info->scir.state;
443 /* flip heartbeat bit */
444 bits ^= SCIR_CPU_HEARTBEAT;
446 /* is this cpu idle? */
447 if (idle_cpu(raw_smp_processor_id()))
448 bits &= ~SCIR_CPU_ACTIVITY;
450 bits |= SCIR_CPU_ACTIVITY;
452 /* update system controller interface reg */
453 uv_set_scir_bits(bits);
455 /* enable next timer period */
456 mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
459 static void __cpuinit uv_heartbeat_enable(int cpu)
461 if (!uv_cpu_hub_info(cpu)->scir.enabled) {
462 struct timer_list *timer = &uv_cpu_hub_info(cpu)->scir.timer;
464 uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
465 setup_timer(timer, uv_heartbeat, cpu);
466 timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
467 add_timer_on(timer, cpu);
468 uv_cpu_hub_info(cpu)->scir.enabled = 1;
472 if (!uv_cpu_hub_info(0)->scir.enabled)
473 uv_heartbeat_enable(0);
476 #ifdef CONFIG_HOTPLUG_CPU
477 static void __cpuinit uv_heartbeat_disable(int cpu)
479 if (uv_cpu_hub_info(cpu)->scir.enabled) {
480 uv_cpu_hub_info(cpu)->scir.enabled = 0;
481 del_timer(&uv_cpu_hub_info(cpu)->scir.timer);
483 uv_set_cpu_scir_bits(cpu, 0xff);
487 * cpu hotplug notifier
489 static __cpuinit int uv_scir_cpu_notify(struct notifier_block *self,
490 unsigned long action, void *hcpu)
492 long cpu = (long)hcpu;
496 uv_heartbeat_enable(cpu);
498 case CPU_DOWN_PREPARE:
499 uv_heartbeat_disable(cpu);
507 static __init void uv_scir_register_cpu_notifier(void)
509 hotcpu_notifier(uv_scir_cpu_notify, 0);
512 #else /* !CONFIG_HOTPLUG_CPU */
514 static __init void uv_scir_register_cpu_notifier(void)
518 static __init int uv_init_heartbeat(void)
523 for_each_online_cpu(cpu)
524 uv_heartbeat_enable(cpu);
528 late_initcall(uv_init_heartbeat);
530 #endif /* !CONFIG_HOTPLUG_CPU */
533 * Called on each cpu to initialize the per_cpu UV data area.
534 * FIXME: hotplug not supported yet
536 void __cpuinit uv_cpu_init(void)
538 /* CPU 0 initilization will be done via uv_system_init. */
542 uv_blade_info[uv_numa_blade_id()].nr_online_cpus++;
544 if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
545 set_x2apic_extra_bits(uv_hub_info->pnode);
549 void __init uv_system_init(void)
551 union uvh_si_addr_map_config_u m_n_config;
552 union uvh_node_id_u node_id;
553 unsigned long gnode_upper, lowmem_redir_base, lowmem_redir_size;
554 int bytes, nid, cpu, lcpu, pnode, blade, i, j, m_val, n_val;
555 int gnode_extra, max_pnode = 0;
556 unsigned long mmr_base, present, paddr;
557 unsigned short pnode_mask;
561 m_n_config.v = uv_read_local_mmr(UVH_SI_ADDR_MAP_CONFIG);
562 m_val = m_n_config.s.m_skt;
563 n_val = m_n_config.s.n_skt;
565 uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
567 pnode_mask = (1 << n_val) - 1;
568 node_id.v = uv_read_local_mmr(UVH_NODE_ID);
569 gnode_extra = (node_id.s.node_id & ~((1 << n_val) - 1)) >> 1;
570 gnode_upper = ((unsigned long)gnode_extra << m_val);
571 printk(KERN_DEBUG "UV: N %d, M %d, gnode_upper 0x%lx, gnode_extra 0x%x\n",
572 n_val, m_val, gnode_upper, gnode_extra);
574 printk(KERN_DEBUG "UV: global MMR base 0x%lx\n", mmr_base);
576 for(i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++)
577 uv_possible_blades +=
578 hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE + i * 8));
579 printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
581 bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
582 uv_blade_info = kmalloc(bytes, GFP_KERNEL);
583 BUG_ON(!uv_blade_info);
584 for (blade = 0; blade < uv_num_possible_blades(); blade++)
585 uv_blade_info[blade].memory_nid = -1;
587 get_lowmem_redirect(&lowmem_redir_base, &lowmem_redir_size);
589 bytes = sizeof(uv_node_to_blade[0]) * num_possible_nodes();
590 uv_node_to_blade = kmalloc(bytes, GFP_KERNEL);
591 BUG_ON(!uv_node_to_blade);
592 memset(uv_node_to_blade, 255, bytes);
594 bytes = sizeof(uv_cpu_to_blade[0]) * num_possible_cpus();
595 uv_cpu_to_blade = kmalloc(bytes, GFP_KERNEL);
596 BUG_ON(!uv_cpu_to_blade);
597 memset(uv_cpu_to_blade, 255, bytes);
600 for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
601 present = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
602 for (j = 0; j < 64; j++) {
603 if (!test_bit(j, &present))
605 uv_blade_info[blade].pnode = (i * 64 + j);
606 uv_blade_info[blade].nr_possible_cpus = 0;
607 uv_blade_info[blade].nr_online_cpus = 0;
613 uv_bios_get_sn_info(0, &uv_type, &sn_partition_id,
614 &sn_coherency_id, &sn_region_size);
617 for_each_present_cpu(cpu) {
618 nid = cpu_to_node(cpu);
619 pnode = uv_apicid_to_pnode(per_cpu(x86_cpu_to_apicid, cpu));
620 blade = boot_pnode_to_blade(pnode);
621 lcpu = uv_blade_info[blade].nr_possible_cpus;
622 uv_blade_info[blade].nr_possible_cpus++;
624 /* Any node on the blade, else will contain -1. */
625 uv_blade_info[blade].memory_nid = nid;
627 uv_cpu_hub_info(cpu)->lowmem_remap_base = lowmem_redir_base;
628 uv_cpu_hub_info(cpu)->lowmem_remap_top = lowmem_redir_size;
629 uv_cpu_hub_info(cpu)->m_val = m_val;
630 uv_cpu_hub_info(cpu)->n_val = m_val;
631 uv_cpu_hub_info(cpu)->numa_blade_id = blade;
632 uv_cpu_hub_info(cpu)->blade_processor_id = lcpu;
633 uv_cpu_hub_info(cpu)->pnode = pnode;
634 uv_cpu_hub_info(cpu)->pnode_mask = pnode_mask;
635 uv_cpu_hub_info(cpu)->gpa_mask = (1 << (m_val + n_val)) - 1;
636 uv_cpu_hub_info(cpu)->gnode_upper = gnode_upper;
637 uv_cpu_hub_info(cpu)->gnode_extra = gnode_extra;
638 uv_cpu_hub_info(cpu)->global_mmr_base = mmr_base;
639 uv_cpu_hub_info(cpu)->coherency_domain_number = sn_coherency_id;
640 uv_cpu_hub_info(cpu)->scir.offset = SCIR_LOCAL_MMR_BASE + lcpu;
641 uv_node_to_blade[nid] = blade;
642 uv_cpu_to_blade[cpu] = blade;
643 max_pnode = max(pnode, max_pnode);
645 printk(KERN_DEBUG "UV: cpu %d, apicid 0x%x, pnode %d, nid %d, "
646 "lcpu %d, blade %d\n",
647 cpu, per_cpu(x86_cpu_to_apicid, cpu), pnode, nid,
651 /* Add blade/pnode info for nodes without cpus */
652 for_each_online_node(nid) {
653 if (uv_node_to_blade[nid] >= 0)
655 paddr = node_start_pfn(nid) << PAGE_SHIFT;
656 paddr = uv_soc_phys_ram_to_gpa(paddr);
657 pnode = (paddr >> m_val) & pnode_mask;
658 blade = boot_pnode_to_blade(pnode);
659 uv_node_to_blade[nid] = blade;
660 max_pnode = max(pnode, max_pnode);
663 map_gru_high(max_pnode);
664 map_mmr_high(max_pnode);
665 map_mmioh_high(max_pnode);
668 uv_scir_register_cpu_notifier();
669 proc_mkdir("sgi_uv", NULL);