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) 2007-2014 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/slab.h>
21 #include <linux/cpu.h>
22 #include <linux/init.h>
24 #include <linux/pci.h>
25 #include <linux/kdebug.h>
26 #include <linux/delay.h>
27 #include <linux/crash_dump.h>
28 #include <linux/reboot.h>
30 #include <asm/uv/uv_mmrs.h>
31 #include <asm/uv/uv_hub.h>
32 #include <asm/current.h>
33 #include <asm/pgtable.h>
34 #include <asm/uv/bios.h>
35 #include <asm/uv/uv.h>
39 #include <asm/x86_init.h>
42 DEFINE_PER_CPU(int, x2apic_extra_bits);
44 #define PR_DEVEL(fmt, args...) pr_devel("%s: " fmt, __func__, args)
46 static enum uv_system_type uv_system_type;
47 static u64 gru_start_paddr, gru_end_paddr;
48 static u64 gru_dist_base, gru_first_node_paddr = -1LL, gru_last_node_paddr;
49 static u64 gru_dist_lmask, gru_dist_umask;
50 static union uvh_apicid uvh_apicid;
52 /* info derived from CPUID */
54 unsigned int apicid_shift;
55 unsigned int apicid_mask;
56 unsigned int socketid_shift; /* aka pnode_shift for UV1/2/3 */
57 unsigned int pnode_mask;
58 unsigned int gpa_shift;
61 int uv_min_hub_revision_id;
62 EXPORT_SYMBOL_GPL(uv_min_hub_revision_id);
63 unsigned int uv_apicid_hibits;
64 EXPORT_SYMBOL_GPL(uv_apicid_hibits);
66 static struct apic apic_x2apic_uv_x;
67 static struct uv_hub_info_s uv_hub_info_node0;
69 /* Set this to use hardware error handler instead of kernel panic */
70 static int disable_uv_undefined_panic = 1;
71 unsigned long uv_undefined(char *str)
73 if (likely(!disable_uv_undefined_panic))
74 panic("UV: error: undefined MMR: %s\n", str);
76 pr_crit("UV: error: undefined MMR: %s\n", str);
77 return ~0ul; /* cause a machine fault */
79 EXPORT_SYMBOL(uv_undefined);
81 static unsigned long __init uv_early_read_mmr(unsigned long addr)
83 unsigned long val, *mmr;
85 mmr = early_ioremap(UV_LOCAL_MMR_BASE | addr, sizeof(*mmr));
87 early_iounmap(mmr, sizeof(*mmr));
91 static inline bool is_GRU_range(u64 start, u64 end)
94 u64 su = start & gru_dist_umask; /* upper (incl pnode) bits */
95 u64 sl = start & gru_dist_lmask; /* base offset bits */
96 u64 eu = end & gru_dist_umask;
97 u64 el = end & gru_dist_lmask;
99 /* Must reside completely within a single GRU range */
100 return (sl == gru_dist_base && el == gru_dist_base &&
101 su >= gru_first_node_paddr &&
102 su <= gru_last_node_paddr &&
105 return start >= gru_start_paddr && end <= gru_end_paddr;
109 static bool uv_is_untracked_pat_range(u64 start, u64 end)
111 return is_ISA_range(start, end) || is_GRU_range(start, end);
114 static int __init early_get_pnodeid(void)
116 union uvh_node_id_u node_id;
117 union uvh_rh_gam_config_mmr_u m_n_config;
120 /* Currently, all blades have same revision number */
121 node_id.v = uv_early_read_mmr(UVH_NODE_ID);
122 m_n_config.v = uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR);
123 uv_min_hub_revision_id = node_id.s.revision;
125 switch (node_id.s.part_number) {
126 case UV2_HUB_PART_NUMBER:
127 case UV2_HUB_PART_NUMBER_X:
128 uv_min_hub_revision_id += UV2_HUB_REVISION_BASE - 1;
130 case UV3_HUB_PART_NUMBER:
131 case UV3_HUB_PART_NUMBER_X:
132 uv_min_hub_revision_id += UV3_HUB_REVISION_BASE;
134 case UV4_HUB_PART_NUMBER:
135 uv_min_hub_revision_id += UV4_HUB_REVISION_BASE - 1;
139 uv_hub_info->hub_revision = uv_min_hub_revision_id;
140 uv_cpuid.pnode_mask = (1 << m_n_config.s.n_skt) - 1;
141 pnode = (node_id.s.node_id >> 1) & uv_cpuid.pnode_mask;
142 uv_cpuid.gpa_shift = 46; /* default unless changed */
144 pr_info("UV: rev:%d part#:%x nodeid:%04x n_skt:%d pnmsk:%x pn:%x\n",
145 node_id.s.revision, node_id.s.part_number, node_id.s.node_id,
146 m_n_config.s.n_skt, uv_cpuid.pnode_mask, pnode);
150 /* [copied from arch/x86/kernel/cpu/topology.c:detect_extended_topology()] */
151 #define SMT_LEVEL 0 /* leaf 0xb SMT level */
152 #define INVALID_TYPE 0 /* leaf 0xb sub-leaf types */
155 #define LEAFB_SUBTYPE(ecx) (((ecx) >> 8) & 0xff)
156 #define BITS_SHIFT_NEXT_LEVEL(eax) ((eax) & 0x1f)
158 static void set_x2apic_bits(void)
160 unsigned int eax, ebx, ecx, edx, sub_index;
161 unsigned int sid_shift;
163 cpuid(0, &eax, &ebx, &ecx, &edx);
165 pr_info("UV: CPU does not have CPUID.11\n");
168 cpuid_count(0xb, SMT_LEVEL, &eax, &ebx, &ecx, &edx);
169 if (ebx == 0 || (LEAFB_SUBTYPE(ecx) != SMT_TYPE)) {
170 pr_info("UV: CPUID.11 not implemented\n");
173 sid_shift = BITS_SHIFT_NEXT_LEVEL(eax);
176 cpuid_count(0xb, sub_index, &eax, &ebx, &ecx, &edx);
177 if (LEAFB_SUBTYPE(ecx) == CORE_TYPE) {
178 sid_shift = BITS_SHIFT_NEXT_LEVEL(eax);
182 } while (LEAFB_SUBTYPE(ecx) != INVALID_TYPE);
183 uv_cpuid.apicid_shift = 0;
184 uv_cpuid.apicid_mask = (~(-1 << sid_shift));
185 uv_cpuid.socketid_shift = sid_shift;
188 static void __init early_get_apic_socketid_shift(void)
190 if (is_uv2_hub() || is_uv3_hub())
191 uvh_apicid.v = uv_early_read_mmr(UVH_APICID);
195 pr_info("UV: apicid_shift:%d apicid_mask:0x%x\n",
196 uv_cpuid.apicid_shift, uv_cpuid.apicid_mask);
197 pr_info("UV: socketid_shift:%d pnode_mask:0x%x\n",
198 uv_cpuid.socketid_shift, uv_cpuid.pnode_mask);
202 * Add an extra bit as dictated by bios to the destination apicid of
203 * interrupts potentially passing through the UV HUB. This prevents
204 * a deadlock between interrupts and IO port operations.
206 static void __init uv_set_apicid_hibit(void)
208 union uv1h_lb_target_physical_apic_id_mask_u apicid_mask;
212 uv_early_read_mmr(UV1H_LB_TARGET_PHYSICAL_APIC_ID_MASK);
214 apicid_mask.s1.bit_enables & UV_APICID_HIBIT_MASK;
218 static int __init uv_acpi_madt_oem_check(char *oem_id, char *oem_table_id)
223 if (strncmp(oem_id, "SGI", 3) != 0)
226 /* Setup early hub type field in uv_hub_info for Node 0 */
227 uv_cpu_info->p_uv_hub_info = &uv_hub_info_node0;
230 * Determine UV arch type.
233 * SGI3: UV300 (truncated to 4 chars because of different varieties)
234 * SGI4: UV400 (truncated to 4 chars because of different varieties)
236 uv_hub_info->hub_revision =
237 !strncmp(oem_id, "SGI4", 4) ? UV4_HUB_REVISION_BASE :
238 !strncmp(oem_id, "SGI3", 4) ? UV3_HUB_REVISION_BASE :
239 !strcmp(oem_id, "SGI2") ? UV2_HUB_REVISION_BASE :
240 !strcmp(oem_id, "SGI") ? UV1_HUB_REVISION_BASE : 0;
242 if (uv_hub_info->hub_revision == 0)
245 pnodeid = early_get_pnodeid();
246 early_get_apic_socketid_shift();
247 x86_platform.is_untracked_pat_range = uv_is_untracked_pat_range;
248 x86_platform.nmi_init = uv_nmi_init;
250 if (!strcmp(oem_table_id, "UVX")) { /* most common */
251 uv_system_type = UV_X2APIC;
254 } else if (!strcmp(oem_table_id, "UVH")) { /* only UV1 systems */
255 uv_system_type = UV_NON_UNIQUE_APIC;
256 __this_cpu_write(x2apic_extra_bits,
257 pnodeid << uvh_apicid.s.pnode_shift);
258 uv_set_apicid_hibit();
261 } else if (!strcmp(oem_table_id, "UVL")) { /* only used for */
262 uv_system_type = UV_LEGACY_APIC; /* very small systems */
269 pr_info("UV: OEM IDs %s/%s, System/HUB Types %d/%d, uv_apic %d\n",
270 oem_id, oem_table_id, uv_system_type,
271 uv_min_hub_revision_id, uv_apic);
276 pr_err("UV: OEM_ID:%s OEM_TABLE_ID:%s\n", oem_id, oem_table_id);
277 pr_err("Current BIOS not supported, update kernel and/or BIOS\n");
281 enum uv_system_type get_uv_system_type(void)
283 return uv_system_type;
286 int is_uv_system(void)
288 return uv_system_type != UV_NONE;
290 EXPORT_SYMBOL_GPL(is_uv_system);
292 void **__uv_hub_info_list;
293 EXPORT_SYMBOL_GPL(__uv_hub_info_list);
295 DEFINE_PER_CPU(struct uv_cpu_info_s, __uv_cpu_info);
296 EXPORT_PER_CPU_SYMBOL_GPL(__uv_cpu_info);
298 short uv_possible_blades;
299 EXPORT_SYMBOL_GPL(uv_possible_blades);
301 unsigned long sn_rtc_cycles_per_second;
302 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
304 static __initdata unsigned short *_node_to_pnode;
306 extern int uv_hub_info_version(void)
308 return UV_HUB_INFO_VERSION;
310 EXPORT_SYMBOL(uv_hub_info_version);
312 static int uv_wakeup_secondary(int phys_apicid, unsigned long start_rip)
317 pnode = uv_apicid_to_pnode(phys_apicid);
318 phys_apicid |= uv_apicid_hibits;
319 val = (1UL << UVH_IPI_INT_SEND_SHFT) |
320 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
321 ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
323 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
325 val = (1UL << UVH_IPI_INT_SEND_SHFT) |
326 (phys_apicid << UVH_IPI_INT_APIC_ID_SHFT) |
327 ((start_rip << UVH_IPI_INT_VECTOR_SHFT) >> 12) |
329 uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
334 static void uv_send_IPI_one(int cpu, int vector)
336 unsigned long apicid;
339 apicid = per_cpu(x86_cpu_to_apicid, cpu);
340 pnode = uv_apicid_to_pnode(apicid);
341 uv_hub_send_ipi(pnode, apicid, vector);
344 static void uv_send_IPI_mask(const struct cpumask *mask, int vector)
348 for_each_cpu(cpu, mask)
349 uv_send_IPI_one(cpu, vector);
352 static void uv_send_IPI_mask_allbutself(const struct cpumask *mask, int vector)
354 unsigned int this_cpu = smp_processor_id();
357 for_each_cpu(cpu, mask) {
359 uv_send_IPI_one(cpu, vector);
363 static void uv_send_IPI_allbutself(int vector)
365 unsigned int this_cpu = smp_processor_id();
368 for_each_online_cpu(cpu) {
370 uv_send_IPI_one(cpu, vector);
374 static void uv_send_IPI_all(int vector)
376 uv_send_IPI_mask(cpu_online_mask, vector);
379 static int uv_apic_id_valid(int apicid)
384 static int uv_apic_id_registered(void)
389 static void uv_init_apic_ldr(void)
394 uv_cpu_mask_to_apicid_and(const struct cpumask *cpumask,
395 const struct cpumask *andmask,
396 unsigned int *apicid)
401 * We're using fixed IRQ delivery, can only return one phys APIC ID.
402 * May as well be the first.
404 for_each_cpu_and(cpu, cpumask, andmask) {
405 if (cpumask_test_cpu(cpu, cpu_online_mask))
409 if (likely(cpu < nr_cpu_ids)) {
410 *apicid = per_cpu(x86_cpu_to_apicid, cpu) | uv_apicid_hibits;
417 static unsigned int x2apic_get_apic_id(unsigned long x)
421 WARN_ON(preemptible() && num_online_cpus() > 1);
422 id = x | __this_cpu_read(x2apic_extra_bits);
427 static unsigned long set_apic_id(unsigned int id)
431 /* maskout x2apic_extra_bits ? */
436 static unsigned int uv_read_apic_id(void)
438 return x2apic_get_apic_id(apic_read(APIC_ID));
441 static int uv_phys_pkg_id(int initial_apicid, int index_msb)
443 return uv_read_apic_id() >> index_msb;
446 static void uv_send_IPI_self(int vector)
448 apic_write(APIC_SELF_IPI, vector);
451 static int uv_probe(void)
453 return apic == &apic_x2apic_uv_x;
456 static struct apic __refdata apic_x2apic_uv_x = {
458 .name = "UV large system",
460 .acpi_madt_oem_check = uv_acpi_madt_oem_check,
461 .apic_id_valid = uv_apic_id_valid,
462 .apic_id_registered = uv_apic_id_registered,
464 .irq_delivery_mode = dest_Fixed,
465 .irq_dest_mode = 0, /* physical */
467 .target_cpus = online_target_cpus,
469 .dest_logical = APIC_DEST_LOGICAL,
470 .check_apicid_used = NULL,
472 .vector_allocation_domain = default_vector_allocation_domain,
473 .init_apic_ldr = uv_init_apic_ldr,
475 .ioapic_phys_id_map = NULL,
476 .setup_apic_routing = NULL,
477 .cpu_present_to_apicid = default_cpu_present_to_apicid,
478 .apicid_to_cpu_present = NULL,
479 .check_phys_apicid_present = default_check_phys_apicid_present,
480 .phys_pkg_id = uv_phys_pkg_id,
482 .get_apic_id = x2apic_get_apic_id,
483 .set_apic_id = set_apic_id,
484 .apic_id_mask = 0xFFFFFFFFu,
486 .cpu_mask_to_apicid_and = uv_cpu_mask_to_apicid_and,
488 .send_IPI = uv_send_IPI_one,
489 .send_IPI_mask = uv_send_IPI_mask,
490 .send_IPI_mask_allbutself = uv_send_IPI_mask_allbutself,
491 .send_IPI_allbutself = uv_send_IPI_allbutself,
492 .send_IPI_all = uv_send_IPI_all,
493 .send_IPI_self = uv_send_IPI_self,
495 .wakeup_secondary_cpu = uv_wakeup_secondary,
496 .inquire_remote_apic = NULL,
498 .read = native_apic_msr_read,
499 .write = native_apic_msr_write,
500 .eoi_write = native_apic_msr_eoi_write,
501 .icr_read = native_x2apic_icr_read,
502 .icr_write = native_x2apic_icr_write,
503 .wait_icr_idle = native_x2apic_wait_icr_idle,
504 .safe_wait_icr_idle = native_safe_x2apic_wait_icr_idle,
507 static void set_x2apic_extra_bits(int pnode)
509 __this_cpu_write(x2apic_extra_bits, pnode << uvh_apicid.s.pnode_shift);
512 #define UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_LENGTH 3
513 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
515 static __init void get_lowmem_redirect(unsigned long *base, unsigned long *size)
517 union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias;
518 union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect;
519 unsigned long m_redirect;
520 unsigned long m_overlay;
523 for (i = 0; i < UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_LENGTH; i++) {
526 m_redirect = UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR;
527 m_overlay = UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR;
530 m_redirect = UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR;
531 m_overlay = UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR;
534 m_redirect = UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR;
535 m_overlay = UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR;
538 alias.v = uv_read_local_mmr(m_overlay);
539 if (alias.s.enable && alias.s.base == 0) {
540 *size = (1UL << alias.s.m_alias);
541 redirect.v = uv_read_local_mmr(m_redirect);
542 *base = (unsigned long)redirect.s.dest_base
550 enum map_type {map_wb, map_uc};
552 static __init void map_high(char *id, unsigned long base, int pshift,
553 int bshift, int max_pnode, enum map_type map_type)
555 unsigned long bytes, paddr;
557 paddr = base << pshift;
558 bytes = (1UL << bshift) * (max_pnode + 1);
560 pr_info("UV: Map %s_HI base address NULL\n", id);
563 pr_debug("UV: Map %s_HI 0x%lx - 0x%lx\n", id, paddr, paddr + bytes);
564 if (map_type == map_uc)
565 init_extra_mapping_uc(paddr, bytes);
567 init_extra_mapping_wb(paddr, bytes);
570 static __init void map_gru_distributed(unsigned long c)
572 union uvh_rh_gam_gru_overlay_config_mmr_u gru;
578 /* only base bits 42:28 relevant in dist mode */
579 gru_dist_base = gru.v & 0x000007fff0000000UL;
580 if (!gru_dist_base) {
581 pr_info("UV: Map GRU_DIST base address NULL\n");
584 bytes = 1UL << UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
585 gru_dist_lmask = ((1UL << uv_hub_info->m_val) - 1) & ~(bytes - 1);
586 gru_dist_umask = ~((1UL << uv_hub_info->m_val) - 1);
587 gru_dist_base &= gru_dist_lmask; /* Clear bits above M */
588 for_each_online_node(nid) {
589 paddr = ((u64)uv_node_to_pnode(nid) << uv_hub_info->m_val) |
591 init_extra_mapping_wb(paddr, bytes);
592 gru_first_node_paddr = min(paddr, gru_first_node_paddr);
593 gru_last_node_paddr = max(paddr, gru_last_node_paddr);
595 /* Save upper (63:M) bits of address only for is_GRU_range */
596 gru_first_node_paddr &= gru_dist_umask;
597 gru_last_node_paddr &= gru_dist_umask;
598 pr_debug("UV: Map GRU_DIST base 0x%016llx 0x%016llx - 0x%016llx\n",
599 gru_dist_base, gru_first_node_paddr, gru_last_node_paddr);
602 static __init void map_gru_high(int max_pnode)
604 union uvh_rh_gam_gru_overlay_config_mmr_u gru;
605 int shift = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT;
606 unsigned long mask = UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_MASK;
609 gru.v = uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR);
611 pr_info("UV: GRU disabled\n");
615 if (is_uv3_hub() && gru.s3.mode) {
616 map_gru_distributed(gru.v);
619 base = (gru.v & mask) >> shift;
620 map_high("GRU", base, shift, shift, max_pnode, map_wb);
621 gru_start_paddr = ((u64)base << shift);
622 gru_end_paddr = gru_start_paddr + (1UL << shift) * (max_pnode + 1);
625 static __init void map_mmr_high(int max_pnode)
627 union uvh_rh_gam_mmr_overlay_config_mmr_u mmr;
628 int shift = UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT;
630 mmr.v = uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR);
632 map_high("MMR", mmr.s.base, shift, shift, max_pnode, map_uc);
634 pr_info("UV: MMR disabled\n");
638 * This commonality works because both 0 & 1 versions of the MMIOH OVERLAY
639 * and REDIRECT MMR regs are exactly the same on UV3.
641 struct mmioh_config {
642 unsigned long overlay;
643 unsigned long redirect;
647 static __initdata struct mmioh_config mmiohs[] = {
649 UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR,
650 UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR,
654 UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG1_MMR,
655 UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG1_MMR,
660 /* UV3 & UV4 have identical MMIOH overlay configs */
661 static __init void map_mmioh_high_uv3(int index, int min_pnode, int max_pnode)
663 union uv3h_rh_gam_mmioh_overlay_config0_mmr_u overlay;
666 int i, n, shift, m_io, max_io;
667 int nasid, lnasid, fi, li;
670 id = mmiohs[index].id;
671 overlay.v = uv_read_local_mmr(mmiohs[index].overlay);
672 pr_info("UV: %s overlay 0x%lx base:0x%x m_io:%d\n",
673 id, overlay.v, overlay.s3.base, overlay.s3.m_io);
674 if (!overlay.s3.enable) {
675 pr_info("UV: %s disabled\n", id);
679 shift = UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR_BASE_SHFT;
680 base = (unsigned long)overlay.s3.base;
681 m_io = overlay.s3.m_io;
682 mmr = mmiohs[index].redirect;
683 n = UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR_DEPTH;
684 min_pnode *= 2; /* convert to NASID */
686 max_io = lnasid = fi = li = -1;
688 for (i = 0; i < n; i++) {
689 union uv3h_rh_gam_mmioh_redirect_config0_mmr_u redirect;
691 redirect.v = uv_read_local_mmr(mmr + i * 8);
692 nasid = redirect.s3.nasid;
693 if (nasid < min_pnode || max_pnode < nasid)
694 nasid = -1; /* invalid NASID */
696 if (nasid == lnasid) {
698 if (i != n-1) /* last entry check */
702 /* check if we have a cached (or last) redirect to print */
703 if (lnasid != -1 || (i == n-1 && nasid != -1)) {
704 unsigned long addr1, addr2;
714 addr1 = (base << shift) +
715 f * (unsigned long)(1 << m_io);
716 addr2 = (base << shift) +
717 (l + 1) * (unsigned long)(1 << m_io);
718 pr_info("UV: %s[%03d..%03d] NASID 0x%04x ADDR 0x%016lx - 0x%016lx\n",
719 id, fi, li, lnasid, addr1, addr2);
727 pr_info("UV: %s base:0x%lx shift:%d M_IO:%d MAX_IO:%d\n",
728 id, base, shift, m_io, max_io);
731 map_high(id, base, shift, m_io, max_io, map_uc);
734 static __init void map_mmioh_high(int min_pnode, int max_pnode)
736 union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh;
737 unsigned long mmr, base;
738 int shift, enable, m_io, n_io;
740 if (is_uv3_hub() || is_uv4_hub()) {
741 /* Map both MMIOH Regions */
742 map_mmioh_high_uv3(0, min_pnode, max_pnode);
743 map_mmioh_high_uv3(1, min_pnode, max_pnode);
748 mmr = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
749 shift = UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
750 mmioh.v = uv_read_local_mmr(mmr);
751 enable = !!mmioh.s1.enable;
752 base = mmioh.s1.base;
753 m_io = mmioh.s1.m_io;
754 n_io = mmioh.s1.n_io;
755 } else if (is_uv2_hub()) {
756 mmr = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR;
757 shift = UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT;
758 mmioh.v = uv_read_local_mmr(mmr);
759 enable = !!mmioh.s2.enable;
760 base = mmioh.s2.base;
761 m_io = mmioh.s2.m_io;
762 n_io = mmioh.s2.n_io;
767 max_pnode &= (1 << n_io) - 1;
769 "UV: base:0x%lx shift:%d N_IO:%d M_IO:%d max_pnode:0x%x\n",
770 base, shift, m_io, n_io, max_pnode);
771 map_high("MMIOH", base, shift, m_io, max_pnode, map_uc);
773 pr_info("UV: MMIOH disabled\n");
777 static __init void map_low_mmrs(void)
779 init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE, UV_GLOBAL_MMR32_SIZE);
780 init_extra_mapping_uc(UV_LOCAL_MMR_BASE, UV_LOCAL_MMR_SIZE);
783 static __init void uv_rtc_init(void)
788 status = uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK,
790 if (status != BIOS_STATUS_SUCCESS || ticks_per_sec < 100000) {
792 "unable to determine platform RTC clock frequency, "
794 /* BIOS gives wrong value for clock freq. so guess */
795 sn_rtc_cycles_per_second = 1000000000000UL / 30000UL;
797 sn_rtc_cycles_per_second = ticks_per_sec;
801 * percpu heartbeat timer
803 static void uv_heartbeat(unsigned long ignored)
805 struct timer_list *timer = &uv_scir_info->timer;
806 unsigned char bits = uv_scir_info->state;
808 /* flip heartbeat bit */
809 bits ^= SCIR_CPU_HEARTBEAT;
811 /* is this cpu idle? */
812 if (idle_cpu(raw_smp_processor_id()))
813 bits &= ~SCIR_CPU_ACTIVITY;
815 bits |= SCIR_CPU_ACTIVITY;
817 /* update system controller interface reg */
818 uv_set_scir_bits(bits);
820 /* enable next timer period */
821 mod_timer_pinned(timer, jiffies + SCIR_CPU_HB_INTERVAL);
824 static void uv_heartbeat_enable(int cpu)
826 while (!uv_cpu_scir_info(cpu)->enabled) {
827 struct timer_list *timer = &uv_cpu_scir_info(cpu)->timer;
829 uv_set_cpu_scir_bits(cpu, SCIR_CPU_HEARTBEAT|SCIR_CPU_ACTIVITY);
830 setup_timer(timer, uv_heartbeat, cpu);
831 timer->expires = jiffies + SCIR_CPU_HB_INTERVAL;
832 add_timer_on(timer, cpu);
833 uv_cpu_scir_info(cpu)->enabled = 1;
835 /* also ensure that boot cpu is enabled */
840 #ifdef CONFIG_HOTPLUG_CPU
841 static void uv_heartbeat_disable(int cpu)
843 if (uv_cpu_scir_info(cpu)->enabled) {
844 uv_cpu_scir_info(cpu)->enabled = 0;
845 del_timer(&uv_cpu_scir_info(cpu)->timer);
847 uv_set_cpu_scir_bits(cpu, 0xff);
851 * cpu hotplug notifier
853 static int uv_scir_cpu_notify(struct notifier_block *self, unsigned long action,
856 long cpu = (long)hcpu;
858 switch (action & ~CPU_TASKS_FROZEN) {
859 case CPU_DOWN_FAILED:
861 uv_heartbeat_enable(cpu);
863 case CPU_DOWN_PREPARE:
864 uv_heartbeat_disable(cpu);
872 static __init void uv_scir_register_cpu_notifier(void)
874 hotcpu_notifier(uv_scir_cpu_notify, 0);
877 #else /* !CONFIG_HOTPLUG_CPU */
879 static __init void uv_scir_register_cpu_notifier(void)
883 static __init int uv_init_heartbeat(void)
888 for_each_online_cpu(cpu)
889 uv_heartbeat_enable(cpu);
893 late_initcall(uv_init_heartbeat);
895 #endif /* !CONFIG_HOTPLUG_CPU */
897 /* Direct Legacy VGA I/O traffic to designated IOH */
898 int uv_set_vga_state(struct pci_dev *pdev, bool decode,
899 unsigned int command_bits, u32 flags)
903 PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
904 pdev->devfn, decode, command_bits, flags);
906 if (!(flags & PCI_VGA_STATE_CHANGE_BRIDGE))
909 if ((command_bits & PCI_COMMAND_IO) == 0)
912 domain = pci_domain_nr(pdev->bus);
913 bus = pdev->bus->number;
915 rc = uv_bios_set_legacy_vga_target(decode, domain, bus);
916 PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode, domain, bus, rc);
922 * Called on each cpu to initialize the per_cpu UV data area.
923 * FIXME: hotplug not supported yet
925 void uv_cpu_init(void)
927 /* CPU 0 initialization will be done via uv_system_init. */
928 if (smp_processor_id() == 0)
931 uv_hub_info->nr_online_cpus++;
933 if (get_uv_system_type() == UV_NON_UNIQUE_APIC)
934 set_x2apic_extra_bits(uv_hub_info->pnode);
940 unsigned char m_shift;
941 unsigned char n_lshift;
944 static void get_mn(struct mn *mnp)
946 union uvh_rh_gam_config_mmr_u m_n_config;
947 union uv3h_gr0_gam_gr_config_u m_gr_config;
949 m_n_config.v = uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR);
950 mnp->n_val = m_n_config.s.n_skt;
954 } else if (is_uv3_hub()) {
955 mnp->m_val = m_n_config.s3.m_skt;
956 m_gr_config.v = uv_read_local_mmr(UV3H_GR0_GAM_GR_CONFIG);
957 mnp->n_lshift = m_gr_config.s3.m_skt;
958 } else if (is_uv2_hub()) {
959 mnp->m_val = m_n_config.s2.m_skt;
960 mnp->n_lshift = mnp->m_val == 40 ? 40 : 39;
961 } else if (is_uv1_hub()) {
962 mnp->m_val = m_n_config.s1.m_skt;
963 mnp->n_lshift = mnp->m_val;
965 mnp->m_shift = mnp->m_val ? 64 - mnp->m_val : 0;
968 void __init uv_init_hub_info(struct uv_hub_info_s *hub_info)
970 struct mn mn = {0}; /* avoid unitialized warnings */
971 union uvh_node_id_u node_id;
974 hub_info->m_val = mn.m_val;
975 hub_info->n_val = mn.n_val;
976 hub_info->m_shift = mn.m_shift;
977 hub_info->n_lshift = mn.n_lshift ? mn.n_lshift : 0;
979 hub_info->hub_revision = uv_hub_info->hub_revision;
980 hub_info->pnode_mask = uv_cpuid.pnode_mask;
981 hub_info->gpa_mask = mn.m_val ?
982 (1UL << (mn.m_val + mn.n_val)) - 1 :
983 (1UL << uv_cpuid.gpa_shift) - 1;
985 node_id.v = uv_read_local_mmr(UVH_NODE_ID);
986 hub_info->gnode_extra =
987 (node_id.s.node_id & ~((1 << mn.n_val) - 1)) >> 1;
989 hub_info->gnode_upper =
990 ((unsigned long)hub_info->gnode_extra << mn.m_val);
992 hub_info->global_mmr_base =
993 uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR) &
997 &hub_info->lowmem_remap_base, &hub_info->lowmem_remap_top);
999 hub_info->apic_pnode_shift = uv_cpuid.socketid_shift;
1001 /* show system specific info */
1002 pr_info("UV: N:%d M:%d m_shift:%d n_lshift:%d\n",
1003 hub_info->n_val, hub_info->m_val,
1004 hub_info->m_shift, hub_info->n_lshift);
1006 pr_info("UV: pnode_mask:0x%x gpa_mask:0x%lx apic_pns:%d\n",
1007 hub_info->pnode_mask, hub_info->gpa_mask,
1008 hub_info->apic_pnode_shift);
1010 pr_info("UV: gnode_upper:0x%lx gnode_extra:0x%x\n",
1011 hub_info->gnode_upper, hub_info->gnode_extra);
1013 pr_info("UV: global MMR base 0x%lx\n", hub_info->global_mmr_base);
1018 * Setup physical blade translations from UVH_NODE_PRESENT_TABLE
1019 * .. NB: UVH_NODE_PRESENT_TABLE is going away,
1020 * .. being replaced by GAM Range Table
1022 static __init void boot_init_possible_blades(struct uv_hub_info_s *hub_info)
1025 int blade, i, j, uv_pb = 0, num_nodes = num_possible_nodes();
1027 pr_info("UV: NODE_PRESENT_DEPTH = %d\n", UVH_NODE_PRESENT_TABLE_DEPTH);
1028 for (i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
1031 np = uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
1033 pr_info("UV: NODE_PRESENT(%d) = 0x%016lx\n", i, np);
1035 uv_pb += hweight64(np);
1037 if (uv_possible_blades != uv_pb)
1038 uv_possible_blades = uv_pb;
1040 bytes = num_nodes * sizeof(_node_to_pnode[0]);
1041 _node_to_pnode = kmalloc(bytes, GFP_KERNEL);
1042 BUG_ON(!_node_to_pnode);
1044 for (blade = 0, i = 0; i < UVH_NODE_PRESENT_TABLE_DEPTH; i++) {
1045 unsigned short pnode;
1046 unsigned long present =
1047 uv_read_local_mmr(UVH_NODE_PRESENT_TABLE + i * 8);
1049 for (j = 0; j < 64; j++) {
1050 if (!test_bit(j, &present))
1052 pnode = (i * 64 + j) & hub_info->pnode_mask;
1053 _node_to_pnode[blade++] = pnode;
1055 if (blade > num_nodes) {
1056 pr_err("UV: blade count(%d) exceeds node count(%d)!\n",
1063 void __init uv_system_init(void)
1065 struct uv_hub_info_s hub_info = {0};
1066 int bytes, cpu, nodeid;
1067 unsigned short min_pnode = 9999, max_pnode = 0;
1068 char *hub = is_uv4_hub() ? "UV400" :
1069 is_uv3_hub() ? "UV300" :
1070 is_uv2_hub() ? "UV2000/3000" :
1071 is_uv1_hub() ? "UV100/1000" : NULL;
1074 pr_err("UV: Unknown/unsupported UV hub\n");
1077 pr_info("UV: Found %s hub\n", hub);
1079 /* We now only need to map the MMRs on UV1 */
1083 uv_init_hub_info(&hub_info);
1084 uv_possible_blades = num_possible_nodes();
1085 if (!_node_to_pnode)
1086 boot_init_possible_blades(&hub_info);
1088 /* uv_num_possible_blades() is really the hub count */
1089 pr_info("UV: Found %d hubs, %d nodes, %d cpus\n",
1090 uv_num_possible_blades(),
1091 num_possible_nodes(),
1092 num_possible_cpus());
1095 uv_bios_get_sn_info(0, &uv_type, &sn_partition_id, &sn_coherency_id,
1096 &sn_region_size, &system_serial_number);
1097 hub_info.coherency_domain_number = sn_coherency_id;
1100 bytes = sizeof(void *) * uv_num_possible_blades();
1101 __uv_hub_info_list = kzalloc(bytes, GFP_KERNEL);
1102 BUG_ON(!__uv_hub_info_list);
1104 bytes = sizeof(struct uv_hub_info_s);
1105 for_each_node(nodeid) {
1106 struct uv_hub_info_s *new_hub;
1107 unsigned short pnode;
1109 if (__uv_hub_info_list[nodeid]) {
1110 pr_err("UV: Node %d UV HUB already initialized!?\n",
1115 /* Allocate new per hub info list */
1116 new_hub = (nodeid == 0) ?
1117 &uv_hub_info_node0 :
1118 kzalloc_node(bytes, GFP_KERNEL, nodeid);
1120 __uv_hub_info_list[nodeid] = new_hub;
1121 new_hub = uv_hub_info_list(nodeid);
1123 *new_hub = hub_info;
1125 pnode = _node_to_pnode[nodeid];
1126 min_pnode = min(pnode, min_pnode);
1127 max_pnode = max(pnode, max_pnode);
1128 new_hub->pnode = pnode;
1129 new_hub->numa_blade_id = uv_node_to_blade_id(nodeid);
1130 new_hub->memory_nid = -1;
1131 new_hub->nr_possible_cpus = 0;
1132 new_hub->nr_online_cpus = 0;
1135 /* Initialize per cpu info */
1136 for_each_possible_cpu(cpu) {
1137 int apicid = per_cpu(x86_cpu_to_apicid, cpu);
1139 nodeid = cpu_to_node(cpu);
1140 uv_cpu_info_per(cpu)->p_uv_hub_info = uv_hub_info_list(nodeid);
1141 uv_cpu_info_per(cpu)->blade_cpu_id =
1142 uv_cpu_hub_info(cpu)->nr_possible_cpus++;
1143 if (uv_cpu_hub_info(cpu)->memory_nid == -1)
1144 uv_cpu_hub_info(cpu)->memory_nid = cpu_to_node(cpu);
1145 uv_cpu_scir_info(cpu)->offset = uv_scir_offset(apicid);
1148 /* Display per node info */
1149 for_each_node(nodeid) {
1150 pr_info("UV: UVHUB node:%2d pn:%02x nrcpus:%d\n",
1152 uv_hub_info_list(nodeid)->pnode,
1153 uv_hub_info_list(nodeid)->nr_possible_cpus);
1156 pr_info("UV: min_pnode:%02x max_pnode:%02x\n", min_pnode, max_pnode);
1157 map_gru_high(max_pnode);
1158 map_mmr_high(max_pnode);
1159 map_mmioh_high(min_pnode, max_pnode);
1163 uv_scir_register_cpu_notifier();
1164 proc_mkdir("sgi_uv", NULL);
1166 /* register Legacy VGA I/O redirection handler */
1167 pci_register_set_vga_state(uv_set_vga_state);
1170 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
1171 * EFI is not enabled in the kdump kernel.
1173 if (is_kdump_kernel())
1174 reboot_type = BOOT_ACPI;
1177 apic_driver(apic_x2apic_uv_x);