• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * include/linux/topology.h
3  *
4  * Written by: Matthew Dobson, IBM Corporation
5  *
6  * Copyright (C) 2002, IBM Corp.
7  *
8  * All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or modify
11  * it under the terms of the GNU General Public License as published by
12  * the Free Software Foundation; either version 2 of the License, or
13  * (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful, but
16  * WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
18  * NON INFRINGEMENT.  See the GNU General Public License for more
19  * details.
20  *
21  * You should have received a copy of the GNU General Public License
22  * along with this program; if not, write to the Free Software
23  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
24  *
25  * Send feedback to <colpatch@us.ibm.com>
26  */
27 #ifndef _LINUX_TOPOLOGY_H
28 #define _LINUX_TOPOLOGY_H
29 
30 #include <linux/cpumask.h>
31 #include <linux/bitops.h>
32 #include <linux/mmzone.h>
33 #include <linux/smp.h>
34 #include <linux/percpu.h>
35 #include <asm/topology.h>
36 
37 #ifndef node_has_online_mem
38 #define node_has_online_mem(nid) (1)
39 #endif
40 
41 #ifndef nr_cpus_node
42 #define nr_cpus_node(node) cpumask_weight(cpumask_of_node(node))
43 #endif
44 
45 #define for_each_node_with_cpus(node)			\
46 	for_each_online_node(node)			\
47 		if (nr_cpus_node(node))
48 
49 int arch_update_cpu_topology(void);
50 
51 /* Conform to ACPI 2.0 SLIT distance definitions */
52 #define LOCAL_DISTANCE		10
53 #define REMOTE_DISTANCE		20
54 #ifndef node_distance
55 #define node_distance(from,to)	((from) == (to) ? LOCAL_DISTANCE : REMOTE_DISTANCE)
56 #endif
57 #ifndef RECLAIM_DISTANCE
58 /*
59  * If the distance between nodes in a system is larger than RECLAIM_DISTANCE
60  * (in whatever arch specific measurement units returned by node_distance())
61  * then switch on zone reclaim on boot.
62  */
63 #define RECLAIM_DISTANCE 30
64 #endif
65 #ifndef PENALTY_FOR_NODE_WITH_CPUS
66 #define PENALTY_FOR_NODE_WITH_CPUS	(1)
67 #endif
68 
69 /*
70  * Below are the 3 major initializers used in building sched_domains:
71  * SD_SIBLING_INIT, for SMT domains
72  * SD_CPU_INIT, for SMP domains
73  * SD_NODE_INIT, for NUMA domains
74  *
75  * Any architecture that cares to do any tuning to these values should do so
76  * by defining their own arch-specific initializer in include/asm/topology.h.
77  * A definition there will automagically override these default initializers
78  * and allow arch-specific performance tuning of sched_domains.
79  * (Only non-zero and non-null fields need be specified.)
80  */
81 
82 #ifdef CONFIG_SCHED_SMT
83 /* MCD - Do we really need this?  It is always on if CONFIG_SCHED_SMT is,
84  * so can't we drop this in favor of CONFIG_SCHED_SMT?
85  */
86 #define ARCH_HAS_SCHED_WAKE_IDLE
87 /* Common values for SMT siblings */
88 #ifndef SD_SIBLING_INIT
89 #define SD_SIBLING_INIT (struct sched_domain) {				\
90 	.min_interval		= 1,					\
91 	.max_interval		= 2,					\
92 	.busy_factor		= 64,					\
93 	.imbalance_pct		= 110,					\
94 									\
95 	.flags			= 1*SD_LOAD_BALANCE			\
96 				| 1*SD_BALANCE_NEWIDLE			\
97 				| 1*SD_BALANCE_EXEC			\
98 				| 1*SD_BALANCE_FORK			\
99 				| 0*SD_BALANCE_WAKE			\
100 				| 1*SD_WAKE_AFFINE			\
101 				| 1*SD_SHARE_CPUPOWER			\
102 				| 0*SD_POWERSAVINGS_BALANCE		\
103 				| 1*SD_SHARE_PKG_RESOURCES		\
104 				| 0*SD_SERIALIZE			\
105 				| 0*SD_PREFER_SIBLING			\
106 				| arch_sd_sibling_asym_packing()	\
107 				,					\
108 	.last_balance		= jiffies,				\
109 	.balance_interval	= 1,					\
110 	.smt_gain		= 1178,	/* 15% */			\
111 }
112 #endif
113 #endif /* CONFIG_SCHED_SMT */
114 
115 #ifdef CONFIG_SCHED_MC
116 /* Common values for MC siblings. for now mostly derived from SD_CPU_INIT */
117 #ifndef SD_MC_INIT
118 #define SD_MC_INIT (struct sched_domain) {				\
119 	.min_interval		= 1,					\
120 	.max_interval		= 4,					\
121 	.busy_factor		= 64,					\
122 	.imbalance_pct		= 125,					\
123 	.cache_nice_tries	= 1,					\
124 	.busy_idx		= 2,					\
125 	.wake_idx		= 0,					\
126 	.forkexec_idx		= 0,					\
127 									\
128 	.flags			= 1*SD_LOAD_BALANCE			\
129 				| 1*SD_BALANCE_NEWIDLE			\
130 				| 1*SD_BALANCE_EXEC			\
131 				| 1*SD_BALANCE_FORK			\
132 				| 0*SD_BALANCE_WAKE			\
133 				| 1*SD_WAKE_AFFINE			\
134 				| 0*SD_PREFER_LOCAL			\
135 				| 0*SD_SHARE_CPUPOWER			\
136 				| 1*SD_SHARE_PKG_RESOURCES		\
137 				| 0*SD_SERIALIZE			\
138 				| sd_balance_for_mc_power()		\
139 				| sd_power_saving_flags()		\
140 				,					\
141 	.last_balance		= jiffies,				\
142 	.balance_interval	= 1,					\
143 }
144 #endif
145 #endif /* CONFIG_SCHED_MC */
146 
147 /* Common values for CPUs */
148 #ifndef SD_CPU_INIT
149 #define SD_CPU_INIT (struct sched_domain) {				\
150 	.min_interval		= 1,					\
151 	.max_interval		= 4,					\
152 	.busy_factor		= 64,					\
153 	.imbalance_pct		= 125,					\
154 	.cache_nice_tries	= 1,					\
155 	.busy_idx		= 2,					\
156 	.idle_idx		= 1,					\
157 	.newidle_idx		= 0,					\
158 	.wake_idx		= 0,					\
159 	.forkexec_idx		= 0,					\
160 									\
161 	.flags			= 1*SD_LOAD_BALANCE			\
162 				| 1*SD_BALANCE_NEWIDLE			\
163 				| 1*SD_BALANCE_EXEC			\
164 				| 1*SD_BALANCE_FORK			\
165 				| 0*SD_BALANCE_WAKE			\
166 				| 1*SD_WAKE_AFFINE			\
167 				| 0*SD_PREFER_LOCAL			\
168 				| 0*SD_SHARE_CPUPOWER			\
169 				| 0*SD_SHARE_PKG_RESOURCES		\
170 				| 0*SD_SERIALIZE			\
171 				| sd_balance_for_package_power()	\
172 				| sd_power_saving_flags()		\
173 				,					\
174 	.last_balance		= jiffies,				\
175 	.balance_interval	= 1,					\
176 }
177 #endif
178 
179 /* sched_domains SD_ALLNODES_INIT for NUMA machines */
180 #define SD_ALLNODES_INIT (struct sched_domain) {			\
181 	.min_interval		= 64,					\
182 	.max_interval		= 64*num_online_cpus(),			\
183 	.busy_factor		= 128,					\
184 	.imbalance_pct		= 133,					\
185 	.cache_nice_tries	= 1,					\
186 	.busy_idx		= 3,					\
187 	.idle_idx		= 3,					\
188 	.flags			= 1*SD_LOAD_BALANCE			\
189 				| 1*SD_BALANCE_NEWIDLE			\
190 				| 0*SD_BALANCE_EXEC			\
191 				| 0*SD_BALANCE_FORK			\
192 				| 0*SD_BALANCE_WAKE			\
193 				| 0*SD_WAKE_AFFINE			\
194 				| 0*SD_SHARE_CPUPOWER			\
195 				| 0*SD_POWERSAVINGS_BALANCE		\
196 				| 0*SD_SHARE_PKG_RESOURCES		\
197 				| 1*SD_SERIALIZE			\
198 				| 0*SD_PREFER_SIBLING			\
199 				,					\
200 	.last_balance		= jiffies,				\
201 	.balance_interval	= 64,					\
202 }
203 
204 #ifndef SD_NODES_PER_DOMAIN
205 #define SD_NODES_PER_DOMAIN 16
206 #endif
207 
208 #ifdef CONFIG_SCHED_BOOK
209 #ifndef SD_BOOK_INIT
210 #error Please define an appropriate SD_BOOK_INIT in include/asm/topology.h!!!
211 #endif
212 #endif /* CONFIG_SCHED_BOOK */
213 
214 #ifdef CONFIG_NUMA
215 #ifndef SD_NODE_INIT
216 #error Please define an appropriate SD_NODE_INIT in include/asm/topology.h!!!
217 #endif
218 
219 #endif /* CONFIG_NUMA */
220 
221 #ifdef CONFIG_USE_PERCPU_NUMA_NODE_ID
222 DECLARE_PER_CPU(int, numa_node);
223 
224 #ifndef numa_node_id
225 /* Returns the number of the current Node. */
numa_node_id(void)226 static inline int numa_node_id(void)
227 {
228 	return __this_cpu_read(numa_node);
229 }
230 #endif
231 
232 #ifndef cpu_to_node
cpu_to_node(int cpu)233 static inline int cpu_to_node(int cpu)
234 {
235 	return per_cpu(numa_node, cpu);
236 }
237 #endif
238 
239 #ifndef set_numa_node
set_numa_node(int node)240 static inline void set_numa_node(int node)
241 {
242 	percpu_write(numa_node, node);
243 }
244 #endif
245 
246 #ifndef set_cpu_numa_node
set_cpu_numa_node(int cpu,int node)247 static inline void set_cpu_numa_node(int cpu, int node)
248 {
249 	per_cpu(numa_node, cpu) = node;
250 }
251 #endif
252 
253 #else	/* !CONFIG_USE_PERCPU_NUMA_NODE_ID */
254 
255 /* Returns the number of the current Node. */
256 #ifndef numa_node_id
numa_node_id(void)257 static inline int numa_node_id(void)
258 {
259 	return cpu_to_node(raw_smp_processor_id());
260 }
261 #endif
262 
263 #endif	/* [!]CONFIG_USE_PERCPU_NUMA_NODE_ID */
264 
265 #ifdef CONFIG_HAVE_MEMORYLESS_NODES
266 
267 /*
268  * N.B., Do NOT reference the '_numa_mem_' per cpu variable directly.
269  * It will not be defined when CONFIG_HAVE_MEMORYLESS_NODES is not defined.
270  * Use the accessor functions set_numa_mem(), numa_mem_id() and cpu_to_mem().
271  */
272 DECLARE_PER_CPU(int, _numa_mem_);
273 
274 #ifndef set_numa_mem
set_numa_mem(int node)275 static inline void set_numa_mem(int node)
276 {
277 	percpu_write(_numa_mem_, node);
278 }
279 #endif
280 
281 #ifndef numa_mem_id
282 /* Returns the number of the nearest Node with memory */
numa_mem_id(void)283 static inline int numa_mem_id(void)
284 {
285 	return __this_cpu_read(_numa_mem_);
286 }
287 #endif
288 
289 #ifndef cpu_to_mem
cpu_to_mem(int cpu)290 static inline int cpu_to_mem(int cpu)
291 {
292 	return per_cpu(_numa_mem_, cpu);
293 }
294 #endif
295 
296 #ifndef set_cpu_numa_mem
set_cpu_numa_mem(int cpu,int node)297 static inline void set_cpu_numa_mem(int cpu, int node)
298 {
299 	per_cpu(_numa_mem_, cpu) = node;
300 }
301 #endif
302 
303 #else	/* !CONFIG_HAVE_MEMORYLESS_NODES */
304 
305 #ifndef numa_mem_id
306 /* Returns the number of the nearest Node with memory */
numa_mem_id(void)307 static inline int numa_mem_id(void)
308 {
309 	return numa_node_id();
310 }
311 #endif
312 
313 #ifndef cpu_to_mem
cpu_to_mem(int cpu)314 static inline int cpu_to_mem(int cpu)
315 {
316 	return cpu_to_node(cpu);
317 }
318 #endif
319 
320 #endif	/* [!]CONFIG_HAVE_MEMORYLESS_NODES */
321 
322 #ifndef topology_physical_package_id
323 #define topology_physical_package_id(cpu)	((void)(cpu), -1)
324 #endif
325 #ifndef topology_core_id
326 #define topology_core_id(cpu)			((void)(cpu), 0)
327 #endif
328 #ifndef topology_thread_cpumask
329 #define topology_thread_cpumask(cpu)		cpumask_of(cpu)
330 #endif
331 #ifndef topology_core_cpumask
332 #define topology_core_cpumask(cpu)		cpumask_of(cpu)
333 #endif
334 
335 #endif /* _LINUX_TOPOLOGY_H */
336