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 architectural definitions
8 * Copyright (C) 2007-2010 Silicon Graphics, Inc. All rights reserved.
11 #ifndef _ASM_X86_UV_UV_HUB_H
12 #define _ASM_X86_UV_UV_HUB_H
15 #include <linux/numa.h>
16 #include <linux/percpu.h>
17 #include <linux/timer.h>
19 #include <asm/types.h>
20 #include <asm/percpu.h>
21 #include <asm/uv/uv_mmrs.h>
22 #include <asm/irq_vectors.h>
23 #include <asm/io_apic.h>
27 * Addressing Terminology
29 * M - The low M bits of a physical address represent the offset
30 * into the blade local memory. RAM memory on a blade is physically
31 * contiguous (although various IO spaces may punch holes in
34 * N - Number of bits in the node portion of a socket physical
37 * NASID - network ID of a router, Mbrick or Cbrick. Nasid values of
38 * routers always have low bit of 1, C/MBricks have low bit
39 * equal to 0. Most addressing macros that target UV hub chips
40 * right shift the NASID by 1 to exclude the always-zero bit.
41 * NASIDs contain up to 15 bits.
43 * GNODE - NASID right shifted by 1 bit. Most mmrs contain gnodes instead
46 * PNODE - the low N bits of the GNODE. The PNODE is the most useful variant
47 * of the nasid for socket usage.
49 * GPA - (global physical address) a socket physical address converted
50 * so that it can be used by the GRU as a global address. Socket
51 * physical addresses 1) need additional NASID (node) bits added
52 * to the high end of the address, and 2) unaliased if the
53 * partition does not have a physical address 0. In addition, on
54 * UV2 rev 1, GPAs need the gnode left shifted to bits 39 or 40.
57 * NumaLink Global Physical Address Format:
58 * +--------------------------------+---------------------+
59 * |00..000| GNODE | NodeOffset |
60 * +--------------------------------+---------------------+
61 * |<-------53 - M bits --->|<--------M bits ----->
63 * M - number of node offset bits (35 .. 40)
66 * Memory/UV-HUB Processor Socket Address Format:
67 * +----------------+---------------+---------------------+
68 * |00..000000000000| PNODE | NodeOffset |
69 * +----------------+---------------+---------------------+
70 * <--- N bits --->|<--------M bits ----->
72 * M - number of node offset bits (35 .. 40)
73 * N - number of PNODE bits (0 .. 10)
75 * Note: M + N cannot currently exceed 44 (x86_64) or 46 (IA64).
76 * The actual values are configuration dependent and are set at
77 * boot time. M & N values are set by the hardware/BIOS at boot.
81 * NOTE!!!!!! This is the current format of the APICID. However, code
82 * should assume that this will change in the future. Use functions
83 * in this file for all APICID bit manipulations and conversion.
87 * pppppppppplc0cch Nehalem-EX (12 bits in hdw reg)
88 * ppppppppplcc0cch Westmere-EX (12 bits in hdw reg)
89 * pppppppppppcccch SandyBridge (15 bits in hdw reg)
93 * l = socket number on board
96 * s = bits that are in the SOCKET_ID CSR
98 * Note: Processor may support fewer bits in the APICID register. The ACPI
99 * tables hold all 16 bits. Software needs to be aware of this.
101 * Unless otherwise specified, all references to APICID refer to
102 * the FULL value contained in ACPI tables, not the subset in the
103 * processor APICID register.
108 * Maximum number of bricks in all partitions and in all coherency domains.
109 * This is the total number of bricks accessible in the numalink fabric. It
110 * includes all C & M bricks. Routers are NOT included.
112 * This value is also the value of the maximum number of non-router NASIDs
113 * in the numalink fabric.
115 * NOTE: a brick may contain 1 or 2 OS nodes. Don't get these confused.
117 #define UV_MAX_NUMALINK_BLADES 16384
120 * Maximum number of C/Mbricks within a software SSI (hardware may support
123 #define UV_MAX_SSI_BLADES 256
126 * The largest possible NASID of a C or M brick (+ 2)
128 #define UV_MAX_NASID_VALUE (UV_MAX_NUMALINK_BLADES * 2)
131 struct timer_list timer
;
132 unsigned long offset
;
134 unsigned long idle_on
;
135 unsigned long idle_off
;
137 unsigned char enabled
;
141 * The following defines attributes of the HUB chip. These attributes are
142 * frequently referenced and are kept in the per-cpu data areas of each cpu.
143 * They are kept together in a struct to minimize cache misses.
145 struct uv_hub_info_s
{
146 unsigned long global_mmr_base
;
147 unsigned long gpa_mask
;
148 unsigned int gnode_extra
;
149 unsigned char hub_revision
;
150 unsigned char apic_pnode_shift
;
151 unsigned char m_shift
;
152 unsigned char n_lshift
;
153 unsigned long gnode_upper
;
154 unsigned long lowmem_remap_top
;
155 unsigned long lowmem_remap_base
;
156 unsigned short pnode
;
157 unsigned short pnode_mask
;
158 unsigned short coherency_domain_number
;
159 unsigned short numa_blade_id
;
160 unsigned char blade_processor_id
;
163 struct uv_scir_s scir
;
166 DECLARE_PER_CPU(struct uv_hub_info_s
, __uv_hub_info
);
167 #define uv_hub_info (&__get_cpu_var(__uv_hub_info))
168 #define uv_cpu_hub_info(cpu) (&per_cpu(__uv_hub_info, cpu))
171 * Hub revisions less than UV2_HUB_REVISION_BASE are UV1 hubs. All UV2
172 * hubs have revision numbers greater than or equal to UV2_HUB_REVISION_BASE.
173 * This is a software convention - NOT the hardware revision numbers in
176 #define UV1_HUB_REVISION_BASE 1
177 #define UV2_HUB_REVISION_BASE 3
179 static inline int is_uv1_hub(void)
181 return uv_hub_info
->hub_revision
< UV2_HUB_REVISION_BASE
;
184 static inline int is_uv2_hub(void)
186 return uv_hub_info
->hub_revision
>= UV2_HUB_REVISION_BASE
;
189 static inline int is_uv2_1_hub(void)
191 return uv_hub_info
->hub_revision
== UV2_HUB_REVISION_BASE
;
194 static inline int is_uv2_2_hub(void)
196 return uv_hub_info
->hub_revision
== UV2_HUB_REVISION_BASE
+ 1;
201 struct uvh_apicid_s
{
202 unsigned long local_apic_mask
: 24;
203 unsigned long local_apic_shift
: 5;
204 unsigned long unused1
: 3;
205 unsigned long pnode_mask
: 24;
206 unsigned long pnode_shift
: 5;
207 unsigned long unused2
: 3;
212 * Local & Global MMR space macros.
213 * Note: macros are intended to be used ONLY by inline functions
214 * in this file - not by other kernel code.
215 * n - NASID (full 15-bit global nasid)
216 * g - GNODE (full 15-bit global nasid, right shifted 1)
217 * p - PNODE (local part of nsids, right shifted 1)
219 #define UV_NASID_TO_PNODE(n) (((n) >> 1) & uv_hub_info->pnode_mask)
220 #define UV_PNODE_TO_GNODE(p) ((p) |uv_hub_info->gnode_extra)
221 #define UV_PNODE_TO_NASID(p) (UV_PNODE_TO_GNODE(p) << 1)
223 #define UV1_LOCAL_MMR_BASE 0xf4000000UL
224 #define UV1_GLOBAL_MMR32_BASE 0xf8000000UL
225 #define UV1_LOCAL_MMR_SIZE (64UL * 1024 * 1024)
226 #define UV1_GLOBAL_MMR32_SIZE (64UL * 1024 * 1024)
228 #define UV2_LOCAL_MMR_BASE 0xfa000000UL
229 #define UV2_GLOBAL_MMR32_BASE 0xfc000000UL
230 #define UV2_LOCAL_MMR_SIZE (32UL * 1024 * 1024)
231 #define UV2_GLOBAL_MMR32_SIZE (32UL * 1024 * 1024)
233 #define UV_LOCAL_MMR_BASE (is_uv1_hub() ? UV1_LOCAL_MMR_BASE \
234 : UV2_LOCAL_MMR_BASE)
235 #define UV_GLOBAL_MMR32_BASE (is_uv1_hub() ? UV1_GLOBAL_MMR32_BASE \
236 : UV2_GLOBAL_MMR32_BASE)
237 #define UV_LOCAL_MMR_SIZE (is_uv1_hub() ? UV1_LOCAL_MMR_SIZE : \
239 #define UV_GLOBAL_MMR32_SIZE (is_uv1_hub() ? UV1_GLOBAL_MMR32_SIZE :\
240 UV2_GLOBAL_MMR32_SIZE)
241 #define UV_GLOBAL_MMR64_BASE (uv_hub_info->global_mmr_base)
243 #define UV_GLOBAL_GRU_MMR_BASE 0x4000000
245 #define UV_GLOBAL_MMR32_PNODE_SHIFT 15
246 #define UV_GLOBAL_MMR64_PNODE_SHIFT 26
248 #define UV_GLOBAL_MMR32_PNODE_BITS(p) ((p) << (UV_GLOBAL_MMR32_PNODE_SHIFT))
250 #define UV_GLOBAL_MMR64_PNODE_BITS(p) \
251 (((unsigned long)(p)) << UV_GLOBAL_MMR64_PNODE_SHIFT)
253 #define UVH_APICID 0x002D0E00L
254 #define UV_APIC_PNODE_SHIFT 6
256 #define UV_APICID_HIBIT_MASK 0xffff0000
258 /* Local Bus from cpu's perspective */
259 #define LOCAL_BUS_BASE 0x1c00000
260 #define LOCAL_BUS_SIZE (4 * 1024 * 1024)
263 * System Controller Interface Reg
265 * Note there are NO leds on a UV system. This register is only
266 * used by the system controller to monitor system-wide operation.
267 * There are 64 regs per node. With Nahelem cpus (2 cores per node,
268 * 8 cpus per core, 2 threads per cpu) there are 32 cpu threads on
271 * The window is located at top of ACPI MMR space
273 #define SCIR_WINDOW_COUNT 64
274 #define SCIR_LOCAL_MMR_BASE (LOCAL_BUS_BASE + \
278 #define SCIR_CPU_HEARTBEAT 0x01 /* timer interrupt */
279 #define SCIR_CPU_ACTIVITY 0x02 /* not idle */
280 #define SCIR_CPU_HB_INTERVAL (HZ) /* once per second */
282 /* Loop through all installed blades */
283 #define for_each_possible_blade(bid) \
284 for ((bid) = 0; (bid) < uv_num_possible_blades(); (bid)++)
287 * Macros for converting between kernel virtual addresses, socket local physical
288 * addresses, and UV global physical addresses.
289 * Note: use the standard __pa() & __va() macros for converting
290 * between socket virtual and socket physical addresses.
293 /* socket phys RAM --> UV global physical address */
294 static inline unsigned long uv_soc_phys_ram_to_gpa(unsigned long paddr
)
296 if (paddr
< uv_hub_info
->lowmem_remap_top
)
297 paddr
|= uv_hub_info
->lowmem_remap_base
;
298 paddr
|= uv_hub_info
->gnode_upper
;
299 paddr
= ((paddr
<< uv_hub_info
->m_shift
) >> uv_hub_info
->m_shift
) |
300 ((paddr
>> uv_hub_info
->m_val
) << uv_hub_info
->n_lshift
);
305 /* socket virtual --> UV global physical address */
306 static inline unsigned long uv_gpa(void *v
)
308 return uv_soc_phys_ram_to_gpa(__pa(v
));
311 /* Top two bits indicate the requested address is in MMR space. */
313 uv_gpa_in_mmr_space(unsigned long gpa
)
315 return (gpa
>> 62) == 0x3UL
;
318 /* UV global physical address --> socket phys RAM */
319 static inline unsigned long uv_gpa_to_soc_phys_ram(unsigned long gpa
)
322 unsigned long remap_base
= uv_hub_info
->lowmem_remap_base
;
323 unsigned long remap_top
= uv_hub_info
->lowmem_remap_top
;
325 gpa
= ((gpa
<< uv_hub_info
->m_shift
) >> uv_hub_info
->m_shift
) |
326 ((gpa
>> uv_hub_info
->n_lshift
) << uv_hub_info
->m_val
);
327 paddr
= gpa
& uv_hub_info
->gpa_mask
;
328 if (paddr
>= remap_base
&& paddr
< remap_base
+ remap_top
)
335 static inline unsigned long uv_gpa_to_gnode(unsigned long gpa
)
337 return gpa
>> uv_hub_info
->n_lshift
;
341 static inline int uv_gpa_to_pnode(unsigned long gpa
)
343 unsigned long n_mask
= (1UL << uv_hub_info
->n_val
) - 1;
345 return uv_gpa_to_gnode(gpa
) & n_mask
;
348 /* gpa -> node offset*/
349 static inline unsigned long uv_gpa_to_offset(unsigned long gpa
)
351 return (gpa
<< uv_hub_info
->m_shift
) >> uv_hub_info
->m_shift
;
354 /* pnode, offset --> socket virtual */
355 static inline void *uv_pnode_offset_to_vaddr(int pnode
, unsigned long offset
)
357 return __va(((unsigned long)pnode
<< uv_hub_info
->m_val
) | offset
);
362 * Extract a PNODE from an APICID (full apicid, not processor subset)
364 static inline int uv_apicid_to_pnode(int apicid
)
366 return (apicid
>> uv_hub_info
->apic_pnode_shift
);
370 * Convert an apicid to the socket number on the blade
372 static inline int uv_apicid_to_socket(int apicid
)
375 return (apicid
>> (uv_hub_info
->apic_pnode_shift
- 1)) & 1;
381 * Access global MMRs using the low memory MMR32 space. This region supports
382 * faster MMR access but not all MMRs are accessible in this space.
384 static inline unsigned long *uv_global_mmr32_address(int pnode
, unsigned long offset
)
386 return __va(UV_GLOBAL_MMR32_BASE
|
387 UV_GLOBAL_MMR32_PNODE_BITS(pnode
) | offset
);
390 static inline void uv_write_global_mmr32(int pnode
, unsigned long offset
, unsigned long val
)
392 writeq(val
, uv_global_mmr32_address(pnode
, offset
));
395 static inline unsigned long uv_read_global_mmr32(int pnode
, unsigned long offset
)
397 return readq(uv_global_mmr32_address(pnode
, offset
));
401 * Access Global MMR space using the MMR space located at the top of physical
404 static inline volatile void __iomem
*uv_global_mmr64_address(int pnode
, unsigned long offset
)
406 return __va(UV_GLOBAL_MMR64_BASE
|
407 UV_GLOBAL_MMR64_PNODE_BITS(pnode
) | offset
);
410 static inline void uv_write_global_mmr64(int pnode
, unsigned long offset
, unsigned long val
)
412 writeq(val
, uv_global_mmr64_address(pnode
, offset
));
415 static inline unsigned long uv_read_global_mmr64(int pnode
, unsigned long offset
)
417 return readq(uv_global_mmr64_address(pnode
, offset
));
421 * Global MMR space addresses when referenced by the GRU. (GRU does
422 * NOT use socket addressing).
424 static inline unsigned long uv_global_gru_mmr_address(int pnode
, unsigned long offset
)
426 return UV_GLOBAL_GRU_MMR_BASE
| offset
|
427 ((unsigned long)pnode
<< uv_hub_info
->m_val
);
430 static inline void uv_write_global_mmr8(int pnode
, unsigned long offset
, unsigned char val
)
432 writeb(val
, uv_global_mmr64_address(pnode
, offset
));
435 static inline unsigned char uv_read_global_mmr8(int pnode
, unsigned long offset
)
437 return readb(uv_global_mmr64_address(pnode
, offset
));
441 * Access hub local MMRs. Faster than using global space but only local MMRs
444 static inline unsigned long *uv_local_mmr_address(unsigned long offset
)
446 return __va(UV_LOCAL_MMR_BASE
| offset
);
449 static inline unsigned long uv_read_local_mmr(unsigned long offset
)
451 return readq(uv_local_mmr_address(offset
));
454 static inline void uv_write_local_mmr(unsigned long offset
, unsigned long val
)
456 writeq(val
, uv_local_mmr_address(offset
));
459 static inline unsigned char uv_read_local_mmr8(unsigned long offset
)
461 return readb(uv_local_mmr_address(offset
));
464 static inline void uv_write_local_mmr8(unsigned long offset
, unsigned char val
)
466 writeb(val
, uv_local_mmr_address(offset
));
470 * Structures and definitions for converting between cpu, node, pnode, and blade
473 struct uv_blade_info
{
474 unsigned short nr_possible_cpus
;
475 unsigned short nr_online_cpus
;
476 unsigned short pnode
;
479 unsigned long nmi_count
;
481 extern struct uv_blade_info
*uv_blade_info
;
482 extern short *uv_node_to_blade
;
483 extern short *uv_cpu_to_blade
;
484 extern short uv_possible_blades
;
486 /* Blade-local cpu number of current cpu. Numbered 0 .. <# cpus on the blade> */
487 static inline int uv_blade_processor_id(void)
489 return uv_hub_info
->blade_processor_id
;
492 /* Blade number of current cpu. Numnbered 0 .. <#blades -1> */
493 static inline int uv_numa_blade_id(void)
495 return uv_hub_info
->numa_blade_id
;
498 /* Convert a cpu number to the the UV blade number */
499 static inline int uv_cpu_to_blade_id(int cpu
)
501 return uv_cpu_to_blade
[cpu
];
504 /* Convert linux node number to the UV blade number */
505 static inline int uv_node_to_blade_id(int nid
)
507 return uv_node_to_blade
[nid
];
510 /* Convert a blade id to the PNODE of the blade */
511 static inline int uv_blade_to_pnode(int bid
)
513 return uv_blade_info
[bid
].pnode
;
516 /* Nid of memory node on blade. -1 if no blade-local memory */
517 static inline int uv_blade_to_memory_nid(int bid
)
519 return uv_blade_info
[bid
].memory_nid
;
522 /* Determine the number of possible cpus on a blade */
523 static inline int uv_blade_nr_possible_cpus(int bid
)
525 return uv_blade_info
[bid
].nr_possible_cpus
;
528 /* Determine the number of online cpus on a blade */
529 static inline int uv_blade_nr_online_cpus(int bid
)
531 return uv_blade_info
[bid
].nr_online_cpus
;
534 /* Convert a cpu id to the PNODE of the blade containing the cpu */
535 static inline int uv_cpu_to_pnode(int cpu
)
537 return uv_blade_info
[uv_cpu_to_blade_id(cpu
)].pnode
;
540 /* Convert a linux node number to the PNODE of the blade */
541 static inline int uv_node_to_pnode(int nid
)
543 return uv_blade_info
[uv_node_to_blade_id(nid
)].pnode
;
546 /* Maximum possible number of blades */
547 static inline int uv_num_possible_blades(void)
549 return uv_possible_blades
;
552 /* Update SCIR state */
553 static inline void uv_set_scir_bits(unsigned char value
)
555 if (uv_hub_info
->scir
.state
!= value
) {
556 uv_hub_info
->scir
.state
= value
;
557 uv_write_local_mmr8(uv_hub_info
->scir
.offset
, value
);
561 static inline unsigned long uv_scir_offset(int apicid
)
563 return SCIR_LOCAL_MMR_BASE
| (apicid
& 0x3f);
566 static inline void uv_set_cpu_scir_bits(int cpu
, unsigned char value
)
568 if (uv_cpu_hub_info(cpu
)->scir
.state
!= value
) {
569 uv_write_global_mmr8(uv_cpu_to_pnode(cpu
),
570 uv_cpu_hub_info(cpu
)->scir
.offset
, value
);
571 uv_cpu_hub_info(cpu
)->scir
.state
= value
;
575 extern unsigned int uv_apicid_hibits
;
576 static unsigned long uv_hub_ipi_value(int apicid
, int vector
, int mode
)
578 apicid
|= uv_apicid_hibits
;
579 return (1UL << UVH_IPI_INT_SEND_SHFT
) |
580 ((apicid
) << UVH_IPI_INT_APIC_ID_SHFT
) |
581 (mode
<< UVH_IPI_INT_DELIVERY_MODE_SHFT
) |
582 (vector
<< UVH_IPI_INT_VECTOR_SHFT
);
585 static inline void uv_hub_send_ipi(int pnode
, int apicid
, int vector
)
588 unsigned long dmode
= dest_Fixed
;
590 if (vector
== NMI_VECTOR
)
593 val
= uv_hub_ipi_value(apicid
, vector
, dmode
);
594 uv_write_global_mmr64(pnode
, UVH_IPI_INT
, val
);
598 * Get the minimum revision number of the hub chips within the partition.
599 * 1 - UV1 rev 1.0 initial silicon
600 * 2 - UV1 rev 2.0 production silicon
601 * 3 - UV2 rev 1.0 initial silicon
603 static inline int uv_get_min_hub_revision_id(void)
605 return uv_hub_info
->hub_revision
;
608 #endif /* CONFIG_X86_64 */
609 #endif /* _ASM_X86_UV_UV_HUB_H */