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
;
51 int uv_min_hub_revision_id
;
52 EXPORT_SYMBOL_GPL(uv_min_hub_revision_id
);
53 unsigned int uv_apicid_hibits
;
54 EXPORT_SYMBOL_GPL(uv_apicid_hibits
);
56 static struct apic apic_x2apic_uv_x
;
58 /* Set this to use hardware error handler instead of kernel panic */
59 static int disable_uv_undefined_panic
= 1;
60 unsigned long uv_undefined(char *str
)
62 if (likely(!disable_uv_undefined_panic
))
63 panic("UV: error: undefined MMR: %s\n", str
);
65 pr_crit("UV: error: undefined MMR: %s\n", str
);
66 return ~0ul; /* cause a machine fault */
68 EXPORT_SYMBOL(uv_undefined
);
70 static unsigned long __init
uv_early_read_mmr(unsigned long addr
)
72 unsigned long val
, *mmr
;
74 mmr
= early_ioremap(UV_LOCAL_MMR_BASE
| addr
, sizeof(*mmr
));
76 early_iounmap(mmr
, sizeof(*mmr
));
80 static inline bool is_GRU_range(u64 start
, u64 end
)
83 u64 su
= start
& gru_dist_umask
; /* upper (incl pnode) bits */
84 u64 sl
= start
& gru_dist_lmask
; /* base offset bits */
85 u64 eu
= end
& gru_dist_umask
;
86 u64 el
= end
& gru_dist_lmask
;
88 /* Must reside completely within a single GRU range */
89 return (sl
== gru_dist_base
&& el
== gru_dist_base
&&
90 su
>= gru_first_node_paddr
&&
91 su
<= gru_last_node_paddr
&&
94 return start
>= gru_start_paddr
&& end
<= gru_end_paddr
;
98 static bool uv_is_untracked_pat_range(u64 start
, u64 end
)
100 return is_ISA_range(start
, end
) || is_GRU_range(start
, end
);
103 static int __init
early_get_pnodeid(void)
105 union uvh_node_id_u node_id
;
106 union uvh_rh_gam_config_mmr_u m_n_config
;
109 /* Currently, all blades have same revision number */
110 node_id
.v
= uv_early_read_mmr(UVH_NODE_ID
);
111 m_n_config
.v
= uv_early_read_mmr(UVH_RH_GAM_CONFIG_MMR
);
112 uv_min_hub_revision_id
= node_id
.s
.revision
;
114 switch (node_id
.s
.part_number
) {
115 case UV2_HUB_PART_NUMBER
:
116 case UV2_HUB_PART_NUMBER_X
:
117 uv_min_hub_revision_id
+= UV2_HUB_REVISION_BASE
- 1;
119 case UV3_HUB_PART_NUMBER
:
120 case UV3_HUB_PART_NUMBER_X
:
121 uv_min_hub_revision_id
+= UV3_HUB_REVISION_BASE
;
123 case UV4_HUB_PART_NUMBER
:
124 uv_min_hub_revision_id
+= UV4_HUB_REVISION_BASE
- 1;
128 uv_hub_info
->hub_revision
= uv_min_hub_revision_id
;
129 pnode
= (node_id
.s
.node_id
>> 1) & ((1 << m_n_config
.s
.n_skt
) - 1);
133 static void __init
early_get_apic_pnode_shift(void)
135 uvh_apicid
.v
= uv_early_read_mmr(UVH_APICID
);
138 * Old bios, use default value
140 uvh_apicid
.s
.pnode_shift
= UV_APIC_PNODE_SHIFT
;
144 * Add an extra bit as dictated by bios to the destination apicid of
145 * interrupts potentially passing through the UV HUB. This prevents
146 * a deadlock between interrupts and IO port operations.
148 static void __init
uv_set_apicid_hibit(void)
150 union uv1h_lb_target_physical_apic_id_mask_u apicid_mask
;
154 uv_early_read_mmr(UV1H_LB_TARGET_PHYSICAL_APIC_ID_MASK
);
156 apicid_mask
.s1
.bit_enables
& UV_APICID_HIBIT_MASK
;
160 static int __init
uv_acpi_madt_oem_check(char *oem_id
, char *oem_table_id
)
165 if (strncmp(oem_id
, "SGI", 3) != 0)
169 * Determine UV arch type.
172 * SGI3: UV300 (truncated to 4 chars because of different varieties)
173 * SGI4: UV400 (truncated to 4 chars because of different varieties)
175 uv_hub_info
->hub_revision
=
176 !strncmp(oem_id
, "SGI4", 4) ? UV4_HUB_REVISION_BASE
:
177 !strncmp(oem_id
, "SGI3", 4) ? UV3_HUB_REVISION_BASE
:
178 !strcmp(oem_id
, "SGI2") ? UV2_HUB_REVISION_BASE
:
179 !strcmp(oem_id
, "SGI") ? UV1_HUB_REVISION_BASE
: 0;
181 if (uv_hub_info
->hub_revision
== 0)
184 pnodeid
= early_get_pnodeid();
185 early_get_apic_pnode_shift();
186 x86_platform
.is_untracked_pat_range
= uv_is_untracked_pat_range
;
187 x86_platform
.nmi_init
= uv_nmi_init
;
189 if (!strcmp(oem_table_id
, "UVX")) { /* most common */
190 uv_system_type
= UV_X2APIC
;
193 } else if (!strcmp(oem_table_id
, "UVH")) { /* only UV1 systems */
194 uv_system_type
= UV_NON_UNIQUE_APIC
;
195 __this_cpu_write(x2apic_extra_bits
,
196 pnodeid
<< uvh_apicid
.s
.pnode_shift
);
197 uv_set_apicid_hibit();
200 } else if (!strcmp(oem_table_id
, "UVL")) { /* only used for */
201 uv_system_type
= UV_LEGACY_APIC
; /* very small systems */
208 pr_info("UV: OEM IDs %s/%s, System/HUB Types %d/%d, uv_apic %d\n",
209 oem_id
, oem_table_id
, uv_system_type
,
210 uv_min_hub_revision_id
, uv_apic
);
215 pr_err("UV: OEM_ID:%s OEM_TABLE_ID:%s\n", oem_id
, oem_table_id
);
216 pr_err("Current BIOS not supported, update kernel and/or BIOS\n");
220 enum uv_system_type
get_uv_system_type(void)
222 return uv_system_type
;
225 int is_uv_system(void)
227 return uv_system_type
!= UV_NONE
;
229 EXPORT_SYMBOL_GPL(is_uv_system
);
231 DEFINE_PER_CPU(struct uv_hub_info_s
, __uv_hub_info
);
232 EXPORT_PER_CPU_SYMBOL_GPL(__uv_hub_info
);
234 struct uv_blade_info
*uv_blade_info
;
235 EXPORT_SYMBOL_GPL(uv_blade_info
);
237 short *uv_node_to_blade
;
238 EXPORT_SYMBOL_GPL(uv_node_to_blade
);
240 short *uv_cpu_to_blade
;
241 EXPORT_SYMBOL_GPL(uv_cpu_to_blade
);
243 short uv_possible_blades
;
244 EXPORT_SYMBOL_GPL(uv_possible_blades
);
246 unsigned long sn_rtc_cycles_per_second
;
247 EXPORT_SYMBOL(sn_rtc_cycles_per_second
);
249 static int uv_wakeup_secondary(int phys_apicid
, unsigned long start_rip
)
254 pnode
= uv_apicid_to_pnode(phys_apicid
);
255 phys_apicid
|= uv_apicid_hibits
;
256 val
= (1UL << UVH_IPI_INT_SEND_SHFT
) |
257 (phys_apicid
<< UVH_IPI_INT_APIC_ID_SHFT
) |
258 ((start_rip
<< UVH_IPI_INT_VECTOR_SHFT
) >> 12) |
260 uv_write_global_mmr64(pnode
, UVH_IPI_INT
, val
);
262 val
= (1UL << UVH_IPI_INT_SEND_SHFT
) |
263 (phys_apicid
<< UVH_IPI_INT_APIC_ID_SHFT
) |
264 ((start_rip
<< UVH_IPI_INT_VECTOR_SHFT
) >> 12) |
266 uv_write_global_mmr64(pnode
, UVH_IPI_INT
, val
);
271 static void uv_send_IPI_one(int cpu
, int vector
)
273 unsigned long apicid
;
276 apicid
= per_cpu(x86_cpu_to_apicid
, cpu
);
277 pnode
= uv_apicid_to_pnode(apicid
);
278 uv_hub_send_ipi(pnode
, apicid
, vector
);
281 static void uv_send_IPI_mask(const struct cpumask
*mask
, int vector
)
285 for_each_cpu(cpu
, mask
)
286 uv_send_IPI_one(cpu
, vector
);
289 static void uv_send_IPI_mask_allbutself(const struct cpumask
*mask
, int vector
)
291 unsigned int this_cpu
= smp_processor_id();
294 for_each_cpu(cpu
, mask
) {
296 uv_send_IPI_one(cpu
, vector
);
300 static void uv_send_IPI_allbutself(int vector
)
302 unsigned int this_cpu
= smp_processor_id();
305 for_each_online_cpu(cpu
) {
307 uv_send_IPI_one(cpu
, vector
);
311 static void uv_send_IPI_all(int vector
)
313 uv_send_IPI_mask(cpu_online_mask
, vector
);
316 static int uv_apic_id_valid(int apicid
)
321 static int uv_apic_id_registered(void)
326 static void uv_init_apic_ldr(void)
331 uv_cpu_mask_to_apicid_and(const struct cpumask
*cpumask
,
332 const struct cpumask
*andmask
,
333 unsigned int *apicid
)
338 * We're using fixed IRQ delivery, can only return one phys APIC ID.
339 * May as well be the first.
341 for_each_cpu_and(cpu
, cpumask
, andmask
) {
342 if (cpumask_test_cpu(cpu
, cpu_online_mask
))
346 if (likely(cpu
< nr_cpu_ids
)) {
347 *apicid
= per_cpu(x86_cpu_to_apicid
, cpu
) | uv_apicid_hibits
;
354 static unsigned int x2apic_get_apic_id(unsigned long x
)
358 WARN_ON(preemptible() && num_online_cpus() > 1);
359 id
= x
| __this_cpu_read(x2apic_extra_bits
);
364 static unsigned long set_apic_id(unsigned int id
)
368 /* maskout x2apic_extra_bits ? */
373 static unsigned int uv_read_apic_id(void)
376 return x2apic_get_apic_id(apic_read(APIC_ID
));
379 static int uv_phys_pkg_id(int initial_apicid
, int index_msb
)
381 return uv_read_apic_id() >> index_msb
;
384 static void uv_send_IPI_self(int vector
)
386 apic_write(APIC_SELF_IPI
, vector
);
389 static int uv_probe(void)
391 return apic
== &apic_x2apic_uv_x
;
394 static struct apic __refdata apic_x2apic_uv_x
= {
396 .name
= "UV large system",
398 .acpi_madt_oem_check
= uv_acpi_madt_oem_check
,
399 .apic_id_valid
= uv_apic_id_valid
,
400 .apic_id_registered
= uv_apic_id_registered
,
402 .irq_delivery_mode
= dest_Fixed
,
403 .irq_dest_mode
= 0, /* physical */
405 .target_cpus
= online_target_cpus
,
407 .dest_logical
= APIC_DEST_LOGICAL
,
408 .check_apicid_used
= NULL
,
410 .vector_allocation_domain
= default_vector_allocation_domain
,
411 .init_apic_ldr
= uv_init_apic_ldr
,
413 .ioapic_phys_id_map
= NULL
,
414 .setup_apic_routing
= NULL
,
415 .cpu_present_to_apicid
= default_cpu_present_to_apicid
,
416 .apicid_to_cpu_present
= NULL
,
417 .check_phys_apicid_present
= default_check_phys_apicid_present
,
418 .phys_pkg_id
= uv_phys_pkg_id
,
420 .get_apic_id
= x2apic_get_apic_id
,
421 .set_apic_id
= set_apic_id
,
422 .apic_id_mask
= 0xFFFFFFFFu
,
424 .cpu_mask_to_apicid_and
= uv_cpu_mask_to_apicid_and
,
426 .send_IPI
= uv_send_IPI_one
,
427 .send_IPI_mask
= uv_send_IPI_mask
,
428 .send_IPI_mask_allbutself
= uv_send_IPI_mask_allbutself
,
429 .send_IPI_allbutself
= uv_send_IPI_allbutself
,
430 .send_IPI_all
= uv_send_IPI_all
,
431 .send_IPI_self
= uv_send_IPI_self
,
433 .wakeup_secondary_cpu
= uv_wakeup_secondary
,
434 .inquire_remote_apic
= NULL
,
436 .read
= native_apic_msr_read
,
437 .write
= native_apic_msr_write
,
438 .eoi_write
= native_apic_msr_eoi_write
,
439 .icr_read
= native_x2apic_icr_read
,
440 .icr_write
= native_x2apic_icr_write
,
441 .wait_icr_idle
= native_x2apic_wait_icr_idle
,
442 .safe_wait_icr_idle
= native_safe_x2apic_wait_icr_idle
,
445 static void set_x2apic_extra_bits(int pnode
)
447 __this_cpu_write(x2apic_extra_bits
, pnode
<< uvh_apicid
.s
.pnode_shift
);
451 * Called on boot cpu.
453 static __init
int boot_pnode_to_blade(int pnode
)
457 for (blade
= 0; blade
< uv_num_possible_blades(); blade
++)
458 if (pnode
== uv_blade_info
[blade
].pnode
)
464 unsigned long redirect
;
468 #define DEST_SHIFT UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR_DEST_BASE_SHFT
470 static __initdata
struct redir_addr redir_addrs
[] = {
471 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_0_MMR
, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_0_MMR
},
472 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_1_MMR
, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_1_MMR
},
473 {UVH_RH_GAM_ALIAS210_REDIRECT_CONFIG_2_MMR
, UVH_RH_GAM_ALIAS210_OVERLAY_CONFIG_2_MMR
},
476 static unsigned char get_n_lshift(int m_val
)
478 union uv3h_gr0_gam_gr_config_u m_gr_config
;
484 return m_val
== 40 ? 40 : 39;
486 m_gr_config
.v
= uv_read_local_mmr(UV3H_GR0_GAM_GR_CONFIG
);
487 return m_gr_config
.s3
.m_skt
;
490 static __init
void get_lowmem_redirect(unsigned long *base
, unsigned long *size
)
492 union uvh_rh_gam_alias210_overlay_config_2_mmr_u alias
;
493 union uvh_rh_gam_alias210_redirect_config_2_mmr_u redirect
;
496 for (i
= 0; i
< ARRAY_SIZE(redir_addrs
); i
++) {
497 alias
.v
= uv_read_local_mmr(redir_addrs
[i
].alias
);
498 if (alias
.s
.enable
&& alias
.s
.base
== 0) {
499 *size
= (1UL << alias
.s
.m_alias
);
500 redirect
.v
= uv_read_local_mmr(redir_addrs
[i
].redirect
);
501 *base
= (unsigned long)redirect
.s
.dest_base
<< DEST_SHIFT
;
508 enum map_type
{map_wb
, map_uc
};
510 static __init
void map_high(char *id
, unsigned long base
, int pshift
,
511 int bshift
, int max_pnode
, enum map_type map_type
)
513 unsigned long bytes
, paddr
;
515 paddr
= base
<< pshift
;
516 bytes
= (1UL << bshift
) * (max_pnode
+ 1);
518 pr_info("UV: Map %s_HI base address NULL\n", id
);
521 pr_debug("UV: Map %s_HI 0x%lx - 0x%lx\n", id
, paddr
, paddr
+ bytes
);
522 if (map_type
== map_uc
)
523 init_extra_mapping_uc(paddr
, bytes
);
525 init_extra_mapping_wb(paddr
, bytes
);
528 static __init
void map_gru_distributed(unsigned long c
)
530 union uvh_rh_gam_gru_overlay_config_mmr_u gru
;
536 /* only base bits 42:28 relevant in dist mode */
537 gru_dist_base
= gru
.v
& 0x000007fff0000000UL
;
538 if (!gru_dist_base
) {
539 pr_info("UV: Map GRU_DIST base address NULL\n");
542 bytes
= 1UL << UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT
;
543 gru_dist_lmask
= ((1UL << uv_hub_info
->m_val
) - 1) & ~(bytes
- 1);
544 gru_dist_umask
= ~((1UL << uv_hub_info
->m_val
) - 1);
545 gru_dist_base
&= gru_dist_lmask
; /* Clear bits above M */
546 for_each_online_node(nid
) {
547 paddr
= ((u64
)uv_node_to_pnode(nid
) << uv_hub_info
->m_val
) |
549 init_extra_mapping_wb(paddr
, bytes
);
550 gru_first_node_paddr
= min(paddr
, gru_first_node_paddr
);
551 gru_last_node_paddr
= max(paddr
, gru_last_node_paddr
);
553 /* Save upper (63:M) bits of address only for is_GRU_range */
554 gru_first_node_paddr
&= gru_dist_umask
;
555 gru_last_node_paddr
&= gru_dist_umask
;
556 pr_debug("UV: Map GRU_DIST base 0x%016llx 0x%016llx - 0x%016llx\n",
557 gru_dist_base
, gru_first_node_paddr
, gru_last_node_paddr
);
560 static __init
void map_gru_high(int max_pnode
)
562 union uvh_rh_gam_gru_overlay_config_mmr_u gru
;
563 int shift
= UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR_BASE_SHFT
;
565 gru
.v
= uv_read_local_mmr(UVH_RH_GAM_GRU_OVERLAY_CONFIG_MMR
);
567 pr_info("UV: GRU disabled\n");
571 if (is_uv3_hub() && gru
.s3
.mode
) {
572 map_gru_distributed(gru
.v
);
575 map_high("GRU", gru
.s
.base
, shift
, shift
, max_pnode
, map_wb
);
576 gru_start_paddr
= ((u64
)gru
.s
.base
<< shift
);
577 gru_end_paddr
= gru_start_paddr
+ (1UL << shift
) * (max_pnode
+ 1);
580 static __init
void map_mmr_high(int max_pnode
)
582 union uvh_rh_gam_mmr_overlay_config_mmr_u mmr
;
583 int shift
= UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR_BASE_SHFT
;
585 mmr
.v
= uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR
);
587 map_high("MMR", mmr
.s
.base
, shift
, shift
, max_pnode
, map_uc
);
589 pr_info("UV: MMR disabled\n");
593 * This commonality works because both 0 & 1 versions of the MMIOH OVERLAY
594 * and REDIRECT MMR regs are exactly the same on UV3.
596 struct mmioh_config
{
597 unsigned long overlay
;
598 unsigned long redirect
;
602 static __initdata
struct mmioh_config mmiohs
[] = {
604 UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR
,
605 UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR
,
609 UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG1_MMR
,
610 UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG1_MMR
,
615 static __init
void map_mmioh_high_uv3(int index
, int min_pnode
, int max_pnode
)
617 union uv3h_rh_gam_mmioh_overlay_config0_mmr_u overlay
;
620 int i
, n
, shift
, m_io
, max_io
;
621 int nasid
, lnasid
, fi
, li
;
624 id
= mmiohs
[index
].id
;
625 overlay
.v
= uv_read_local_mmr(mmiohs
[index
].overlay
);
626 pr_info("UV: %s overlay 0x%lx base:0x%x m_io:%d\n",
627 id
, overlay
.v
, overlay
.s3
.base
, overlay
.s3
.m_io
);
628 if (!overlay
.s3
.enable
) {
629 pr_info("UV: %s disabled\n", id
);
633 shift
= UV3H_RH_GAM_MMIOH_OVERLAY_CONFIG0_MMR_BASE_SHFT
;
634 base
= (unsigned long)overlay
.s3
.base
;
635 m_io
= overlay
.s3
.m_io
;
636 mmr
= mmiohs
[index
].redirect
;
637 n
= UV3H_RH_GAM_MMIOH_REDIRECT_CONFIG0_MMR_DEPTH
;
638 min_pnode
*= 2; /* convert to NASID */
640 max_io
= lnasid
= fi
= li
= -1;
642 for (i
= 0; i
< n
; i
++) {
643 union uv3h_rh_gam_mmioh_redirect_config0_mmr_u redirect
;
645 redirect
.v
= uv_read_local_mmr(mmr
+ i
* 8);
646 nasid
= redirect
.s3
.nasid
;
647 if (nasid
< min_pnode
|| max_pnode
< nasid
)
648 nasid
= -1; /* invalid NASID */
650 if (nasid
== lnasid
) {
652 if (i
!= n
-1) /* last entry check */
656 /* check if we have a cached (or last) redirect to print */
657 if (lnasid
!= -1 || (i
== n
-1 && nasid
!= -1)) {
658 unsigned long addr1
, addr2
;
668 addr1
= (base
<< shift
) +
669 f
* (unsigned long)(1 << m_io
);
670 addr2
= (base
<< shift
) +
671 (l
+ 1) * (unsigned long)(1 << m_io
);
672 pr_info("UV: %s[%03d..%03d] NASID 0x%04x ADDR 0x%016lx - 0x%016lx\n",
673 id
, fi
, li
, lnasid
, addr1
, addr2
);
681 pr_info("UV: %s base:0x%lx shift:%d M_IO:%d MAX_IO:%d\n",
682 id
, base
, shift
, m_io
, max_io
);
685 map_high(id
, base
, shift
, m_io
, max_io
, map_uc
);
688 static __init
void map_mmioh_high(int min_pnode
, int max_pnode
)
690 union uvh_rh_gam_mmioh_overlay_config_mmr_u mmioh
;
691 unsigned long mmr
, base
;
692 int shift
, enable
, m_io
, n_io
;
695 /* Map both MMIOH Regions */
696 map_mmioh_high_uv3(0, min_pnode
, max_pnode
);
697 map_mmioh_high_uv3(1, min_pnode
, max_pnode
);
702 mmr
= UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR
;
703 shift
= UV1H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT
;
704 mmioh
.v
= uv_read_local_mmr(mmr
);
705 enable
= !!mmioh
.s1
.enable
;
706 base
= mmioh
.s1
.base
;
707 m_io
= mmioh
.s1
.m_io
;
708 n_io
= mmioh
.s1
.n_io
;
709 } else if (is_uv2_hub()) {
710 mmr
= UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR
;
711 shift
= UV2H_RH_GAM_MMIOH_OVERLAY_CONFIG_MMR_BASE_SHFT
;
712 mmioh
.v
= uv_read_local_mmr(mmr
);
713 enable
= !!mmioh
.s2
.enable
;
714 base
= mmioh
.s2
.base
;
715 m_io
= mmioh
.s2
.m_io
;
716 n_io
= mmioh
.s2
.n_io
;
721 max_pnode
&= (1 << n_io
) - 1;
723 "UV: base:0x%lx shift:%d N_IO:%d M_IO:%d max_pnode:0x%x\n",
724 base
, shift
, m_io
, n_io
, max_pnode
);
725 map_high("MMIOH", base
, shift
, m_io
, max_pnode
, map_uc
);
727 pr_info("UV: MMIOH disabled\n");
731 static __init
void map_low_mmrs(void)
733 init_extra_mapping_uc(UV_GLOBAL_MMR32_BASE
, UV_GLOBAL_MMR32_SIZE
);
734 init_extra_mapping_uc(UV_LOCAL_MMR_BASE
, UV_LOCAL_MMR_SIZE
);
737 static __init
void uv_rtc_init(void)
742 status
= uv_bios_freq_base(BIOS_FREQ_BASE_REALTIME_CLOCK
,
744 if (status
!= BIOS_STATUS_SUCCESS
|| ticks_per_sec
< 100000) {
746 "unable to determine platform RTC clock frequency, "
748 /* BIOS gives wrong value for clock freq. so guess */
749 sn_rtc_cycles_per_second
= 1000000000000UL / 30000UL;
751 sn_rtc_cycles_per_second
= ticks_per_sec
;
755 * percpu heartbeat timer
757 static void uv_heartbeat(unsigned long ignored
)
759 struct timer_list
*timer
= &uv_hub_info
->scir
.timer
;
760 unsigned char bits
= uv_hub_info
->scir
.state
;
762 /* flip heartbeat bit */
763 bits
^= SCIR_CPU_HEARTBEAT
;
765 /* is this cpu idle? */
766 if (idle_cpu(raw_smp_processor_id()))
767 bits
&= ~SCIR_CPU_ACTIVITY
;
769 bits
|= SCIR_CPU_ACTIVITY
;
771 /* update system controller interface reg */
772 uv_set_scir_bits(bits
);
774 /* enable next timer period */
775 mod_timer_pinned(timer
, jiffies
+ SCIR_CPU_HB_INTERVAL
);
778 static void uv_heartbeat_enable(int cpu
)
780 while (!uv_cpu_hub_info(cpu
)->scir
.enabled
) {
781 struct timer_list
*timer
= &uv_cpu_hub_info(cpu
)->scir
.timer
;
783 uv_set_cpu_scir_bits(cpu
, SCIR_CPU_HEARTBEAT
|SCIR_CPU_ACTIVITY
);
784 setup_timer(timer
, uv_heartbeat
, cpu
);
785 timer
->expires
= jiffies
+ SCIR_CPU_HB_INTERVAL
;
786 add_timer_on(timer
, cpu
);
787 uv_cpu_hub_info(cpu
)->scir
.enabled
= 1;
789 /* also ensure that boot cpu is enabled */
794 #ifdef CONFIG_HOTPLUG_CPU
795 static void uv_heartbeat_disable(int cpu
)
797 if (uv_cpu_hub_info(cpu
)->scir
.enabled
) {
798 uv_cpu_hub_info(cpu
)->scir
.enabled
= 0;
799 del_timer(&uv_cpu_hub_info(cpu
)->scir
.timer
);
801 uv_set_cpu_scir_bits(cpu
, 0xff);
805 * cpu hotplug notifier
807 static int uv_scir_cpu_notify(struct notifier_block
*self
, unsigned long action
,
810 long cpu
= (long)hcpu
;
812 switch (action
& ~CPU_TASKS_FROZEN
) {
813 case CPU_DOWN_FAILED
:
815 uv_heartbeat_enable(cpu
);
817 case CPU_DOWN_PREPARE
:
818 uv_heartbeat_disable(cpu
);
826 static __init
void uv_scir_register_cpu_notifier(void)
828 hotcpu_notifier(uv_scir_cpu_notify
, 0);
831 #else /* !CONFIG_HOTPLUG_CPU */
833 static __init
void uv_scir_register_cpu_notifier(void)
837 static __init
int uv_init_heartbeat(void)
842 for_each_online_cpu(cpu
)
843 uv_heartbeat_enable(cpu
);
847 late_initcall(uv_init_heartbeat
);
849 #endif /* !CONFIG_HOTPLUG_CPU */
851 /* Direct Legacy VGA I/O traffic to designated IOH */
852 int uv_set_vga_state(struct pci_dev
*pdev
, bool decode
,
853 unsigned int command_bits
, u32 flags
)
857 PR_DEVEL("devfn %x decode %d cmd %x flags %d\n",
858 pdev
->devfn
, decode
, command_bits
, flags
);
860 if (!(flags
& PCI_VGA_STATE_CHANGE_BRIDGE
))
863 if ((command_bits
& PCI_COMMAND_IO
) == 0)
866 domain
= pci_domain_nr(pdev
->bus
);
867 bus
= pdev
->bus
->number
;
869 rc
= uv_bios_set_legacy_vga_target(decode
, domain
, bus
);
870 PR_DEVEL("vga decode %d %x:%x, rc: %d\n", decode
, domain
, bus
, rc
);
876 * Called on each cpu to initialize the per_cpu UV data area.
877 * FIXME: hotplug not supported yet
879 void uv_cpu_init(void)
881 /* CPU 0 initialization will be done via uv_system_init. */
885 uv_blade_info
[uv_numa_blade_id()].nr_online_cpus
++;
887 if (get_uv_system_type() == UV_NON_UNIQUE_APIC
)
888 set_x2apic_extra_bits(uv_hub_info
->pnode
);
891 void __init
uv_system_init(void)
893 union uvh_rh_gam_config_mmr_u m_n_config
;
894 union uvh_node_id_u node_id
;
895 unsigned long gnode_upper
, lowmem_redir_base
, lowmem_redir_size
;
896 int bytes
, nid
, cpu
, lcpu
, pnode
, blade
, i
, j
, m_val
, n_val
;
897 int gnode_extra
, min_pnode
= 999999, max_pnode
= -1;
898 unsigned long mmr_base
, present
, paddr
;
899 unsigned short pnode_mask
;
900 unsigned char n_lshift
;
901 char *hub
= is_uv4_hub() ? "UV400" :
902 is_uv3_hub() ? "UV300" :
903 is_uv2_hub() ? "UV2000/3000" :
904 is_uv1_hub() ? "UV100/1000" : NULL
;
907 pr_err("UV: Unknown/unsupported UV hub\n");
910 pr_info("UV: Found %s hub\n", hub
);
912 /* We now only need to map the MMRs on UV1 */
916 m_n_config
.v
= uv_read_local_mmr(UVH_RH_GAM_CONFIG_MMR
);
917 m_val
= m_n_config
.s
.m_skt
;
918 n_val
= m_n_config
.s
.n_skt
;
919 pnode_mask
= (1 << n_val
) - 1;
920 n_lshift
= get_n_lshift(m_val
);
922 uv_read_local_mmr(UVH_RH_GAM_MMR_OVERLAY_CONFIG_MMR
) &
925 node_id
.v
= uv_read_local_mmr(UVH_NODE_ID
);
926 gnode_extra
= (node_id
.s
.node_id
& ~((1 << n_val
) - 1)) >> 1;
927 gnode_upper
= ((unsigned long)gnode_extra
<< m_val
);
928 pr_info("UV: N:%d M:%d pnode_mask:0x%x gnode_upper/extra:0x%lx/0x%x n_lshift 0x%x\n",
929 n_val
, m_val
, pnode_mask
, gnode_upper
, gnode_extra
,
932 pr_info("UV: global MMR base 0x%lx\n", mmr_base
);
934 for(i
= 0; i
< UVH_NODE_PRESENT_TABLE_DEPTH
; i
++)
935 uv_possible_blades
+=
936 hweight64(uv_read_local_mmr( UVH_NODE_PRESENT_TABLE
+ i
* 8));
938 /* uv_num_possible_blades() is really the hub count */
939 pr_info("UV: Found %d blades, %d hubs\n",
940 is_uv1_hub() ? uv_num_possible_blades() :
941 (uv_num_possible_blades() + 1) / 2,
942 uv_num_possible_blades());
944 bytes
= sizeof(struct uv_blade_info
) * uv_num_possible_blades();
945 uv_blade_info
= kzalloc(bytes
, GFP_KERNEL
);
946 BUG_ON(!uv_blade_info
);
948 for (blade
= 0; blade
< uv_num_possible_blades(); blade
++)
949 uv_blade_info
[blade
].memory_nid
= -1;
951 get_lowmem_redirect(&lowmem_redir_base
, &lowmem_redir_size
);
953 bytes
= sizeof(uv_node_to_blade
[0]) * num_possible_nodes();
954 uv_node_to_blade
= kmalloc(bytes
, GFP_KERNEL
);
955 BUG_ON(!uv_node_to_blade
);
956 memset(uv_node_to_blade
, 255, bytes
);
958 bytes
= sizeof(uv_cpu_to_blade
[0]) * num_possible_cpus();
959 uv_cpu_to_blade
= kmalloc(bytes
, GFP_KERNEL
);
960 BUG_ON(!uv_cpu_to_blade
);
961 memset(uv_cpu_to_blade
, 255, bytes
);
964 for (i
= 0; i
< UVH_NODE_PRESENT_TABLE_DEPTH
; i
++) {
965 present
= uv_read_local_mmr(UVH_NODE_PRESENT_TABLE
+ i
* 8);
966 for (j
= 0; j
< 64; j
++) {
967 if (!test_bit(j
, &present
))
969 pnode
= (i
* 64 + j
) & pnode_mask
;
970 uv_blade_info
[blade
].pnode
= pnode
;
971 uv_blade_info
[blade
].nr_possible_cpus
= 0;
972 uv_blade_info
[blade
].nr_online_cpus
= 0;
973 spin_lock_init(&uv_blade_info
[blade
].nmi_lock
);
974 min_pnode
= min(pnode
, min_pnode
);
975 max_pnode
= max(pnode
, max_pnode
);
981 uv_bios_get_sn_info(0, &uv_type
, &sn_partition_id
, &sn_coherency_id
,
982 &sn_region_size
, &system_serial_number
);
985 for_each_present_cpu(cpu
) {
986 int apicid
= per_cpu(x86_cpu_to_apicid
, cpu
);
988 nid
= cpu_to_node(cpu
);
990 * apic_pnode_shift must be set before calling uv_apicid_to_pnode();
992 uv_cpu_hub_info(cpu
)->pnode_mask
= pnode_mask
;
993 uv_cpu_hub_info(cpu
)->apic_pnode_shift
= uvh_apicid
.s
.pnode_shift
;
994 uv_cpu_hub_info(cpu
)->hub_revision
= uv_hub_info
->hub_revision
;
996 uv_cpu_hub_info(cpu
)->m_shift
= 64 - m_val
;
997 uv_cpu_hub_info(cpu
)->n_lshift
= n_lshift
;
999 pnode
= uv_apicid_to_pnode(apicid
);
1000 blade
= boot_pnode_to_blade(pnode
);
1001 lcpu
= uv_blade_info
[blade
].nr_possible_cpus
;
1002 uv_blade_info
[blade
].nr_possible_cpus
++;
1004 /* Any node on the blade, else will contain -1. */
1005 uv_blade_info
[blade
].memory_nid
= nid
;
1007 uv_cpu_hub_info(cpu
)->lowmem_remap_base
= lowmem_redir_base
;
1008 uv_cpu_hub_info(cpu
)->lowmem_remap_top
= lowmem_redir_size
;
1009 uv_cpu_hub_info(cpu
)->m_val
= m_val
;
1010 uv_cpu_hub_info(cpu
)->n_val
= n_val
;
1011 uv_cpu_hub_info(cpu
)->numa_blade_id
= blade
;
1012 uv_cpu_hub_info(cpu
)->blade_processor_id
= lcpu
;
1013 uv_cpu_hub_info(cpu
)->pnode
= pnode
;
1014 uv_cpu_hub_info(cpu
)->gpa_mask
= (1UL << (m_val
+ n_val
)) - 1;
1015 uv_cpu_hub_info(cpu
)->gnode_upper
= gnode_upper
;
1016 uv_cpu_hub_info(cpu
)->gnode_extra
= gnode_extra
;
1017 uv_cpu_hub_info(cpu
)->global_mmr_base
= mmr_base
;
1018 uv_cpu_hub_info(cpu
)->coherency_domain_number
= sn_coherency_id
;
1019 uv_cpu_hub_info(cpu
)->scir
.offset
= uv_scir_offset(apicid
);
1020 uv_node_to_blade
[nid
] = blade
;
1021 uv_cpu_to_blade
[cpu
] = blade
;
1024 /* Add blade/pnode info for nodes without cpus */
1025 for_each_online_node(nid
) {
1026 if (uv_node_to_blade
[nid
] >= 0)
1028 paddr
= node_start_pfn(nid
) << PAGE_SHIFT
;
1029 pnode
= uv_gpa_to_pnode(uv_soc_phys_ram_to_gpa(paddr
));
1030 blade
= boot_pnode_to_blade(pnode
);
1031 uv_node_to_blade
[nid
] = blade
;
1034 map_gru_high(max_pnode
);
1035 map_mmr_high(max_pnode
);
1036 map_mmioh_high(min_pnode
, max_pnode
);
1040 uv_scir_register_cpu_notifier();
1041 proc_mkdir("sgi_uv", NULL
);
1043 /* register Legacy VGA I/O redirection handler */
1044 pci_register_set_vga_state(uv_set_vga_state
);
1047 * For a kdump kernel the reset must be BOOT_ACPI, not BOOT_EFI, as
1048 * EFI is not enabled in the kdump kernel.
1050 if (is_kdump_kernel())
1051 reboot_type
= BOOT_ACPI
;
1054 apic_driver(apic_x2apic_uv_x
);