xref: /openbmc/linux/arch/x86/include/asm/uv/uv_hub.h (revision fd589a8f)
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.
5  *
6  * SGI UV architectural definitions
7  *
8  * Copyright (C) 2007-2008 Silicon Graphics, Inc. All rights reserved.
9  */
10 
11 #ifndef _ASM_X86_UV_UV_HUB_H
12 #define _ASM_X86_UV_UV_HUB_H
13 
14 #ifdef CONFIG_X86_64
15 #include <linux/numa.h>
16 #include <linux/percpu.h>
17 #include <linux/timer.h>
18 #include <asm/types.h>
19 #include <asm/percpu.h>
20 #include <asm/uv/uv_mmrs.h>
21 
22 
23 /*
24  * Addressing Terminology
25  *
26  *	M       - The low M bits of a physical address represent the offset
27  *		  into the blade local memory. RAM memory on a blade is physically
28  *		  contiguous (although various IO spaces may punch holes in
29  *		  it)..
30  *
31  * 	N	- Number of bits in the node portion of a socket physical
32  * 		  address.
33  *
34  * 	NASID   - network ID of a router, Mbrick or Cbrick. Nasid values of
35  * 	 	  routers always have low bit of 1, C/MBricks have low bit
36  * 		  equal to 0. Most addressing macros that target UV hub chips
37  * 		  right shift the NASID by 1 to exclude the always-zero bit.
38  * 		  NASIDs contain up to 15 bits.
39  *
40  *	GNODE   - NASID right shifted by 1 bit. Most mmrs contain gnodes instead
41  *		  of nasids.
42  *
43  * 	PNODE   - the low N bits of the GNODE. The PNODE is the most useful variant
44  * 		  of the nasid for socket usage.
45  *
46  *
47  *  NumaLink Global Physical Address Format:
48  *  +--------------------------------+---------------------+
49  *  |00..000|      GNODE             |      NodeOffset     |
50  *  +--------------------------------+---------------------+
51  *          |<-------53 - M bits --->|<--------M bits ----->
52  *
53  *	M - number of node offset bits (35 .. 40)
54  *
55  *
56  *  Memory/UV-HUB Processor Socket Address Format:
57  *  +----------------+---------------+---------------------+
58  *  |00..000000000000|   PNODE       |      NodeOffset     |
59  *  +----------------+---------------+---------------------+
60  *                   <--- N bits --->|<--------M bits ----->
61  *
62  *	M - number of node offset bits (35 .. 40)
63  *	N - number of PNODE bits (0 .. 10)
64  *
65  *		Note: M + N cannot currently exceed 44 (x86_64) or 46 (IA64).
66  *		The actual values are configuration dependent and are set at
67  *		boot time. M & N values are set by the hardware/BIOS at boot.
68  *
69  *
70  * APICID format
71  * 	NOTE!!!!!! This is the current format of the APICID. However, code
72  * 	should assume that this will change in the future. Use functions
73  * 	in this file for all APICID bit manipulations and conversion.
74  *
75  * 		1111110000000000
76  * 		5432109876543210
77  *		pppppppppplc0cch
78  *		sssssssssss
79  *
80  *			p  = pnode bits
81  *			l =  socket number on board
82  *			c  = core
83  *			h  = hyperthread
84  *			s  = bits that are in the SOCKET_ID CSR
85  *
86  *	Note: Processor only supports 12 bits in the APICID register. The ACPI
87  *	      tables hold all 16 bits. Software needs to be aware of this.
88  *
89  * 	      Unless otherwise specified, all references to APICID refer to
90  * 	      the FULL value contained in ACPI tables, not the subset in the
91  * 	      processor APICID register.
92  */
93 
94 
95 /*
96  * Maximum number of bricks in all partitions and in all coherency domains.
97  * This is the total number of bricks accessible in the numalink fabric. It
98  * includes all C & M bricks. Routers are NOT included.
99  *
100  * This value is also the value of the maximum number of non-router NASIDs
101  * in the numalink fabric.
102  *
103  * NOTE: a brick may contain 1 or 2 OS nodes. Don't get these confused.
104  */
105 #define UV_MAX_NUMALINK_BLADES	16384
106 
107 /*
108  * Maximum number of C/Mbricks within a software SSI (hardware may support
109  * more).
110  */
111 #define UV_MAX_SSI_BLADES	256
112 
113 /*
114  * The largest possible NASID of a C or M brick (+ 2)
115  */
116 #define UV_MAX_NASID_VALUE	(UV_MAX_NUMALINK_NODES * 2)
117 
118 struct uv_scir_s {
119 	struct timer_list timer;
120 	unsigned long	offset;
121 	unsigned long	last;
122 	unsigned long	idle_on;
123 	unsigned long	idle_off;
124 	unsigned char	state;
125 	unsigned char	enabled;
126 };
127 
128 /*
129  * The following defines attributes of the HUB chip. These attributes are
130  * frequently referenced and are kept in the per-cpu data areas of each cpu.
131  * They are kept together in a struct to minimize cache misses.
132  */
133 struct uv_hub_info_s {
134 	unsigned long		global_mmr_base;
135 	unsigned long		gpa_mask;
136 	unsigned int		gnode_extra;
137 	unsigned long		gnode_upper;
138 	unsigned long		lowmem_remap_top;
139 	unsigned long		lowmem_remap_base;
140 	unsigned short		pnode;
141 	unsigned short		pnode_mask;
142 	unsigned short		coherency_domain_number;
143 	unsigned short		numa_blade_id;
144 	unsigned char		blade_processor_id;
145 	unsigned char		m_val;
146 	unsigned char		n_val;
147 	struct uv_scir_s	scir;
148 };
149 
150 DECLARE_PER_CPU(struct uv_hub_info_s, __uv_hub_info);
151 #define uv_hub_info 		(&__get_cpu_var(__uv_hub_info))
152 #define uv_cpu_hub_info(cpu)	(&per_cpu(__uv_hub_info, cpu))
153 
154 /*
155  * Local & Global MMR space macros.
156  * 	Note: macros are intended to be used ONLY by inline functions
157  * 	in this file - not by other kernel code.
158  * 		n -  NASID (full 15-bit global nasid)
159  * 		g -  GNODE (full 15-bit global nasid, right shifted 1)
160  * 		p -  PNODE (local part of nsids, right shifted 1)
161  */
162 #define UV_NASID_TO_PNODE(n)		(((n) >> 1) & uv_hub_info->pnode_mask)
163 #define UV_PNODE_TO_GNODE(p)		((p) |uv_hub_info->gnode_extra)
164 #define UV_PNODE_TO_NASID(p)		(UV_PNODE_TO_GNODE(p) << 1)
165 
166 #define UV_LOCAL_MMR_BASE		0xf4000000UL
167 #define UV_GLOBAL_MMR32_BASE		0xf8000000UL
168 #define UV_GLOBAL_MMR64_BASE		(uv_hub_info->global_mmr_base)
169 #define UV_LOCAL_MMR_SIZE		(64UL * 1024 * 1024)
170 #define UV_GLOBAL_MMR32_SIZE		(64UL * 1024 * 1024)
171 
172 #define UV_GLOBAL_MMR32_PNODE_SHIFT	15
173 #define UV_GLOBAL_MMR64_PNODE_SHIFT	26
174 
175 #define UV_GLOBAL_MMR32_PNODE_BITS(p)	((p) << (UV_GLOBAL_MMR32_PNODE_SHIFT))
176 
177 #define UV_GLOBAL_MMR64_PNODE_BITS(p)					\
178 	(((unsigned long)(p)) << UV_GLOBAL_MMR64_PNODE_SHIFT)
179 
180 #define UV_APIC_PNODE_SHIFT	6
181 
182 /* Local Bus from cpu's perspective */
183 #define LOCAL_BUS_BASE		0x1c00000
184 #define LOCAL_BUS_SIZE		(4 * 1024 * 1024)
185 
186 /*
187  * System Controller Interface Reg
188  *
189  * Note there are NO leds on a UV system.  This register is only
190  * used by the system controller to monitor system-wide operation.
191  * There are 64 regs per node.  With Nahelem cpus (2 cores per node,
192  * 8 cpus per core, 2 threads per cpu) there are 32 cpu threads on
193  * a node.
194  *
195  * The window is located at top of ACPI MMR space
196  */
197 #define SCIR_WINDOW_COUNT	64
198 #define SCIR_LOCAL_MMR_BASE	(LOCAL_BUS_BASE + \
199 				 LOCAL_BUS_SIZE - \
200 				 SCIR_WINDOW_COUNT)
201 
202 #define SCIR_CPU_HEARTBEAT	0x01	/* timer interrupt */
203 #define SCIR_CPU_ACTIVITY	0x02	/* not idle */
204 #define SCIR_CPU_HB_INTERVAL	(HZ)	/* once per second */
205 
206 /* Loop through all installed blades */
207 #define for_each_possible_blade(bid)		\
208 	for ((bid) = 0; (bid) < uv_num_possible_blades(); (bid)++)
209 
210 /*
211  * Macros for converting between kernel virtual addresses, socket local physical
212  * addresses, and UV global physical addresses.
213  * 	Note: use the standard __pa() & __va() macros for converting
214  * 	      between socket virtual and socket physical addresses.
215  */
216 
217 /* socket phys RAM --> UV global physical address */
218 static inline unsigned long uv_soc_phys_ram_to_gpa(unsigned long paddr)
219 {
220 	if (paddr < uv_hub_info->lowmem_remap_top)
221 		paddr |= uv_hub_info->lowmem_remap_base;
222 	return paddr | uv_hub_info->gnode_upper;
223 }
224 
225 
226 /* socket virtual --> UV global physical address */
227 static inline unsigned long uv_gpa(void *v)
228 {
229 	return uv_soc_phys_ram_to_gpa(__pa(v));
230 }
231 
232 /* pnode, offset --> socket virtual */
233 static inline void *uv_pnode_offset_to_vaddr(int pnode, unsigned long offset)
234 {
235 	return __va(((unsigned long)pnode << uv_hub_info->m_val) | offset);
236 }
237 
238 
239 /*
240  * Extract a PNODE from an APICID (full apicid, not processor subset)
241  */
242 static inline int uv_apicid_to_pnode(int apicid)
243 {
244 	return (apicid >> UV_APIC_PNODE_SHIFT);
245 }
246 
247 /*
248  * Access global MMRs using the low memory MMR32 space. This region supports
249  * faster MMR access but not all MMRs are accessible in this space.
250  */
251 static inline unsigned long *uv_global_mmr32_address(int pnode,
252 				unsigned long offset)
253 {
254 	return __va(UV_GLOBAL_MMR32_BASE |
255 		       UV_GLOBAL_MMR32_PNODE_BITS(pnode) | offset);
256 }
257 
258 static inline void uv_write_global_mmr32(int pnode, unsigned long offset,
259 				 unsigned long val)
260 {
261 	*uv_global_mmr32_address(pnode, offset) = val;
262 }
263 
264 static inline unsigned long uv_read_global_mmr32(int pnode,
265 						 unsigned long offset)
266 {
267 	return *uv_global_mmr32_address(pnode, offset);
268 }
269 
270 /*
271  * Access Global MMR space using the MMR space located at the top of physical
272  * memory.
273  */
274 static inline unsigned long *uv_global_mmr64_address(int pnode,
275 				unsigned long offset)
276 {
277 	return __va(UV_GLOBAL_MMR64_BASE |
278 		    UV_GLOBAL_MMR64_PNODE_BITS(pnode) | offset);
279 }
280 
281 static inline void uv_write_global_mmr64(int pnode, unsigned long offset,
282 				unsigned long val)
283 {
284 	*uv_global_mmr64_address(pnode, offset) = val;
285 }
286 
287 static inline unsigned long uv_read_global_mmr64(int pnode,
288 						 unsigned long offset)
289 {
290 	return *uv_global_mmr64_address(pnode, offset);
291 }
292 
293 /*
294  * Access hub local MMRs. Faster than using global space but only local MMRs
295  * are accessible.
296  */
297 static inline unsigned long *uv_local_mmr_address(unsigned long offset)
298 {
299 	return __va(UV_LOCAL_MMR_BASE | offset);
300 }
301 
302 static inline unsigned long uv_read_local_mmr(unsigned long offset)
303 {
304 	return *uv_local_mmr_address(offset);
305 }
306 
307 static inline void uv_write_local_mmr(unsigned long offset, unsigned long val)
308 {
309 	*uv_local_mmr_address(offset) = val;
310 }
311 
312 static inline unsigned char uv_read_local_mmr8(unsigned long offset)
313 {
314 	return *((unsigned char *)uv_local_mmr_address(offset));
315 }
316 
317 static inline void uv_write_local_mmr8(unsigned long offset, unsigned char val)
318 {
319 	*((unsigned char *)uv_local_mmr_address(offset)) = val;
320 }
321 
322 /*
323  * Structures and definitions for converting between cpu, node, pnode, and blade
324  * numbers.
325  */
326 struct uv_blade_info {
327 	unsigned short	nr_possible_cpus;
328 	unsigned short	nr_online_cpus;
329 	unsigned short	pnode;
330 	short		memory_nid;
331 };
332 extern struct uv_blade_info *uv_blade_info;
333 extern short *uv_node_to_blade;
334 extern short *uv_cpu_to_blade;
335 extern short uv_possible_blades;
336 
337 /* Blade-local cpu number of current cpu. Numbered 0 .. <# cpus on the blade> */
338 static inline int uv_blade_processor_id(void)
339 {
340 	return uv_hub_info->blade_processor_id;
341 }
342 
343 /* Blade number of current cpu. Numnbered 0 .. <#blades -1> */
344 static inline int uv_numa_blade_id(void)
345 {
346 	return uv_hub_info->numa_blade_id;
347 }
348 
349 /* Convert a cpu number to the the UV blade number */
350 static inline int uv_cpu_to_blade_id(int cpu)
351 {
352 	return uv_cpu_to_blade[cpu];
353 }
354 
355 /* Convert linux node number to the UV blade number */
356 static inline int uv_node_to_blade_id(int nid)
357 {
358 	return uv_node_to_blade[nid];
359 }
360 
361 /* Convert a blade id to the PNODE of the blade */
362 static inline int uv_blade_to_pnode(int bid)
363 {
364 	return uv_blade_info[bid].pnode;
365 }
366 
367 /* Nid of memory node on blade. -1 if no blade-local memory */
368 static inline int uv_blade_to_memory_nid(int bid)
369 {
370 	return uv_blade_info[bid].memory_nid;
371 }
372 
373 /* Determine the number of possible cpus on a blade */
374 static inline int uv_blade_nr_possible_cpus(int bid)
375 {
376 	return uv_blade_info[bid].nr_possible_cpus;
377 }
378 
379 /* Determine the number of online cpus on a blade */
380 static inline int uv_blade_nr_online_cpus(int bid)
381 {
382 	return uv_blade_info[bid].nr_online_cpus;
383 }
384 
385 /* Convert a cpu id to the PNODE of the blade containing the cpu */
386 static inline int uv_cpu_to_pnode(int cpu)
387 {
388 	return uv_blade_info[uv_cpu_to_blade_id(cpu)].pnode;
389 }
390 
391 /* Convert a linux node number to the PNODE of the blade */
392 static inline int uv_node_to_pnode(int nid)
393 {
394 	return uv_blade_info[uv_node_to_blade_id(nid)].pnode;
395 }
396 
397 /* Maximum possible number of blades */
398 static inline int uv_num_possible_blades(void)
399 {
400 	return uv_possible_blades;
401 }
402 
403 /* Update SCIR state */
404 static inline void uv_set_scir_bits(unsigned char value)
405 {
406 	if (uv_hub_info->scir.state != value) {
407 		uv_hub_info->scir.state = value;
408 		uv_write_local_mmr8(uv_hub_info->scir.offset, value);
409 	}
410 }
411 
412 static inline void uv_set_cpu_scir_bits(int cpu, unsigned char value)
413 {
414 	if (uv_cpu_hub_info(cpu)->scir.state != value) {
415 		uv_cpu_hub_info(cpu)->scir.state = value;
416 		uv_write_local_mmr8(uv_cpu_hub_info(cpu)->scir.offset, value);
417 	}
418 }
419 
420 static inline void uv_hub_send_ipi(int pnode, int apicid, int vector)
421 {
422 	unsigned long val;
423 
424 	val = (1UL << UVH_IPI_INT_SEND_SHFT) |
425 			((apicid & 0x3f) << UVH_IPI_INT_APIC_ID_SHFT) |
426 			(vector << UVH_IPI_INT_VECTOR_SHFT);
427 	uv_write_global_mmr64(pnode, UVH_IPI_INT, val);
428 }
429 
430 #endif /* CONFIG_X86_64 */
431 #endif /* _ASM_X86_UV_UV_HUB_H */
432