]>
Commit | Line | Data |
---|---|---|
4fe29a85 GOC |
1 | #include <linux/kernel.h> |
2 | #include <linux/module.h> | |
3 | #include <linux/init.h> | |
4 | #include <linux/bootmem.h> | |
5 | #include <linux/percpu.h> | |
6 | #include <asm/smp.h> | |
7 | #include <asm/percpu.h> | |
8 | #include <asm/sections.h> | |
9 | #include <asm/processor.h> | |
10 | #include <asm/setup.h> | |
11 | #include <asm/topology.h> | |
0fc0906e | 12 | #include <asm/mpspec.h> |
76eb4131 AS |
13 | #include <asm/apicdef.h> |
14 | ||
f8955ebe | 15 | #ifdef CONFIG_X86_LOCAL_APIC |
2fe60147 AS |
16 | unsigned int num_processors; |
17 | unsigned disabled_cpus __cpuinitdata; | |
18 | /* Processor that is doing the boot up */ | |
19 | unsigned int boot_cpu_physical_apicid = -1U; | |
e0da3364 | 20 | unsigned int max_physical_apicid; |
2fe60147 AS |
21 | EXPORT_SYMBOL(boot_cpu_physical_apicid); |
22 | ||
0fc0906e AS |
23 | /* Bitmask of physically existing CPUs */ |
24 | physid_mask_t phys_cpu_present_map; | |
f8955ebe | 25 | #endif |
0fc0906e | 26 | |
23ca4bba MT |
27 | /* map cpu index to physical APIC ID */ |
28 | DEFINE_EARLY_PER_CPU(u16, x86_cpu_to_apicid, BAD_APICID); | |
29 | DEFINE_EARLY_PER_CPU(u16, x86_bios_cpu_apicid, BAD_APICID); | |
30 | EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid); | |
31 | EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid); | |
32 | ||
33 | #if defined(CONFIG_NUMA) && defined(CONFIG_X86_64) | |
34 | #define X86_64_NUMA 1 | |
35 | ||
7891a24e | 36 | /* map cpu index to node index */ |
23ca4bba MT |
37 | DEFINE_EARLY_PER_CPU(int, x86_cpu_to_node_map, NUMA_NO_NODE); |
38 | EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_node_map); | |
9f248bde MT |
39 | |
40 | /* which logical CPUs are on which nodes */ | |
41 | cpumask_t *node_to_cpumask_map; | |
42 | EXPORT_SYMBOL(node_to_cpumask_map); | |
43 | ||
44 | /* setup node_to_cpumask_map */ | |
45 | static void __init setup_node_to_cpumask_map(void); | |
46 | ||
47 | #else | |
48 | static inline void setup_node_to_cpumask_map(void) { } | |
23ca4bba MT |
49 | #endif |
50 | ||
f8955ebe | 51 | #if defined(CONFIG_HAVE_SETUP_PER_CPU_AREA) && defined(CONFIG_X86_SMP) |
4fe29a85 GOC |
52 | /* |
53 | * Copy data used in early init routines from the initial arrays to the | |
54 | * per cpu data areas. These arrays then become expendable and the | |
55 | * *_early_ptr's are zeroed indicating that the static arrays are gone. | |
56 | */ | |
57 | static void __init setup_per_cpu_maps(void) | |
58 | { | |
59 | int cpu; | |
60 | ||
61 | for_each_possible_cpu(cpu) { | |
23ca4bba MT |
62 | per_cpu(x86_cpu_to_apicid, cpu) = |
63 | early_per_cpu_map(x86_cpu_to_apicid, cpu); | |
b447a468 | 64 | per_cpu(x86_bios_cpu_apicid, cpu) = |
23ca4bba MT |
65 | early_per_cpu_map(x86_bios_cpu_apicid, cpu); |
66 | #ifdef X86_64_NUMA | |
b447a468 | 67 | per_cpu(x86_cpu_to_node_map, cpu) = |
23ca4bba | 68 | early_per_cpu_map(x86_cpu_to_node_map, cpu); |
4fe29a85 GOC |
69 | #endif |
70 | } | |
71 | ||
72 | /* indicate the early static arrays will soon be gone */ | |
23ca4bba MT |
73 | early_per_cpu_ptr(x86_cpu_to_apicid) = NULL; |
74 | early_per_cpu_ptr(x86_bios_cpu_apicid) = NULL; | |
75 | #ifdef X86_64_NUMA | |
76 | early_per_cpu_ptr(x86_cpu_to_node_map) = NULL; | |
4fe29a85 GOC |
77 | #endif |
78 | } | |
79 | ||
9f0e8d04 MT |
80 | #ifdef CONFIG_HAVE_CPUMASK_OF_CPU_MAP |
81 | cpumask_t *cpumask_of_cpu_map __read_mostly; | |
82 | EXPORT_SYMBOL(cpumask_of_cpu_map); | |
83 | ||
84 | /* requires nr_cpu_ids to be initialized */ | |
85 | static void __init setup_cpumask_of_cpu(void) | |
86 | { | |
87 | int i; | |
88 | ||
89 | /* alloc_bootmem zeroes memory */ | |
90 | cpumask_of_cpu_map = alloc_bootmem_low(sizeof(cpumask_t) * nr_cpu_ids); | |
91 | for (i = 0; i < nr_cpu_ids; i++) | |
92 | cpu_set(i, cpumask_of_cpu_map[i]); | |
93 | } | |
94 | #else | |
95 | static inline void setup_cpumask_of_cpu(void) { } | |
96 | #endif | |
97 | ||
4fe29a85 GOC |
98 | #ifdef CONFIG_X86_32 |
99 | /* | |
100 | * Great future not-so-futuristic plan: make i386 and x86_64 do it | |
101 | * the same way | |
102 | */ | |
103 | unsigned long __per_cpu_offset[NR_CPUS] __read_mostly; | |
104 | EXPORT_SYMBOL(__per_cpu_offset); | |
3461b0af MT |
105 | static inline void setup_cpu_pda_map(void) { } |
106 | ||
107 | #elif !defined(CONFIG_SMP) | |
108 | static inline void setup_cpu_pda_map(void) { } | |
109 | ||
110 | #else /* CONFIG_SMP && CONFIG_X86_64 */ | |
111 | ||
112 | /* | |
113 | * Allocate cpu_pda pointer table and array via alloc_bootmem. | |
114 | */ | |
115 | static void __init setup_cpu_pda_map(void) | |
116 | { | |
117 | char *pda; | |
118 | struct x8664_pda **new_cpu_pda; | |
119 | unsigned long size; | |
120 | int cpu; | |
121 | ||
122 | size = roundup(sizeof(struct x8664_pda), cache_line_size()); | |
123 | ||
124 | /* allocate cpu_pda array and pointer table */ | |
125 | { | |
126 | unsigned long tsize = nr_cpu_ids * sizeof(void *); | |
127 | unsigned long asize = size * (nr_cpu_ids - 1); | |
128 | ||
129 | tsize = roundup(tsize, cache_line_size()); | |
130 | new_cpu_pda = alloc_bootmem(tsize + asize); | |
131 | pda = (char *)new_cpu_pda + tsize; | |
132 | } | |
133 | ||
134 | /* initialize pointer table to static pda's */ | |
135 | for_each_possible_cpu(cpu) { | |
136 | if (cpu == 0) { | |
137 | /* leave boot cpu pda in place */ | |
138 | new_cpu_pda[0] = cpu_pda(0); | |
139 | continue; | |
140 | } | |
141 | new_cpu_pda[cpu] = (struct x8664_pda *)pda; | |
142 | new_cpu_pda[cpu]->in_bootmem = 1; | |
143 | pda += size; | |
144 | } | |
145 | ||
146 | /* point to new pointer table */ | |
147 | _cpu_pda = new_cpu_pda; | |
148 | } | |
4fe29a85 GOC |
149 | #endif |
150 | ||
151 | /* | |
152 | * Great future plan: | |
153 | * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data. | |
154 | * Always point %gs to its beginning | |
155 | */ | |
156 | void __init setup_per_cpu_areas(void) | |
157 | { | |
3461b0af MT |
158 | ssize_t size = PERCPU_ENOUGH_ROOM; |
159 | char *ptr; | |
160 | int cpu; | |
4fe29a85 GOC |
161 | |
162 | #ifdef CONFIG_HOTPLUG_CPU | |
163 | prefill_possible_map(); | |
3461b0af MT |
164 | #else |
165 | nr_cpu_ids = num_processors; | |
4fe29a85 GOC |
166 | #endif |
167 | ||
3461b0af MT |
168 | /* Setup cpu_pda map */ |
169 | setup_cpu_pda_map(); | |
170 | ||
4fe29a85 GOC |
171 | /* Copy section for each CPU (we discard the original) */ |
172 | size = PERCPU_ENOUGH_ROOM; | |
053713f5 | 173 | printk(KERN_INFO "PERCPU: Allocating %zd bytes of per cpu data\n", |
4fe29a85 | 174 | size); |
b447a468 | 175 | |
3461b0af | 176 | for_each_possible_cpu(cpu) { |
4fe29a85 GOC |
177 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
178 | ptr = alloc_bootmem_pages(size); | |
179 | #else | |
3461b0af | 180 | int node = early_cpu_to_node(cpu); |
b447a468 | 181 | if (!node_online(node) || !NODE_DATA(node)) { |
4fe29a85 | 182 | ptr = alloc_bootmem_pages(size); |
b447a468 | 183 | printk(KERN_INFO |
23ca4bba | 184 | "cpu %d has no node %d or node-local memory\n", |
3461b0af | 185 | cpu, node); |
b447a468 | 186 | } |
4fe29a85 GOC |
187 | else |
188 | ptr = alloc_bootmem_pages_node(NODE_DATA(node), size); | |
189 | #endif | |
3461b0af | 190 | per_cpu_offset(cpu) = ptr - __per_cpu_start; |
4fe29a85 | 191 | memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start); |
9f0e8d04 | 192 | |
4fe29a85 GOC |
193 | } |
194 | ||
9f248bde MT |
195 | printk(KERN_DEBUG "NR_CPUS: %d, nr_cpu_ids: %d, nr_node_ids %d\n", |
196 | NR_CPUS, nr_cpu_ids, nr_node_ids); | |
9f0e8d04 | 197 | |
b447a468 | 198 | /* Setup percpu data maps */ |
4fe29a85 | 199 | setup_per_cpu_maps(); |
9f0e8d04 | 200 | |
9f248bde MT |
201 | /* Setup node to cpumask map */ |
202 | setup_node_to_cpumask_map(); | |
203 | ||
9f0e8d04 MT |
204 | /* Setup cpumask_of_cpu map */ |
205 | setup_cpumask_of_cpu(); | |
4fe29a85 GOC |
206 | } |
207 | ||
208 | #endif | |
c45a707d HY |
209 | |
210 | void __init parse_setup_data(void) | |
211 | { | |
212 | struct setup_data *data; | |
213 | u64 pa_data; | |
214 | ||
215 | if (boot_params.hdr.version < 0x0209) | |
216 | return; | |
217 | pa_data = boot_params.hdr.setup_data; | |
218 | while (pa_data) { | |
219 | data = early_ioremap(pa_data, PAGE_SIZE); | |
220 | switch (data->type) { | |
8c5beb50 HY |
221 | case SETUP_E820_EXT: |
222 | parse_e820_ext(data, pa_data); | |
223 | break; | |
c45a707d HY |
224 | default: |
225 | break; | |
226 | } | |
227 | #ifndef CONFIG_DEBUG_BOOT_PARAMS | |
228 | free_early(pa_data, pa_data+sizeof(*data)+data->len); | |
229 | #endif | |
230 | pa_data = data->next; | |
231 | early_iounmap(data, PAGE_SIZE); | |
232 | } | |
233 | } | |
2b4fa851 | 234 | |
23ca4bba | 235 | #ifdef X86_64_NUMA |
9f248bde MT |
236 | |
237 | /* | |
238 | * Allocate node_to_cpumask_map based on number of available nodes | |
239 | * Requires node_possible_map to be valid. | |
240 | * | |
241 | * Note: node_to_cpumask() is not valid until after this is done. | |
242 | */ | |
243 | static void __init setup_node_to_cpumask_map(void) | |
244 | { | |
245 | unsigned int node, num = 0; | |
246 | cpumask_t *map; | |
247 | ||
248 | /* setup nr_node_ids if not done yet */ | |
249 | if (nr_node_ids == MAX_NUMNODES) { | |
250 | for_each_node_mask(node, node_possible_map) | |
251 | num = node; | |
252 | nr_node_ids = num + 1; | |
253 | } | |
254 | ||
255 | /* allocate the map */ | |
256 | map = alloc_bootmem_low(nr_node_ids * sizeof(cpumask_t)); | |
257 | ||
258 | Dprintk(KERN_DEBUG "Node to cpumask map at %p for %d nodes\n", | |
259 | map, nr_node_ids); | |
260 | ||
261 | /* node_to_cpumask() will now work */ | |
262 | node_to_cpumask_map = map; | |
263 | } | |
264 | ||
23ca4bba MT |
265 | void __cpuinit numa_set_node(int cpu, int node) |
266 | { | |
267 | int *cpu_to_node_map = early_per_cpu_ptr(x86_cpu_to_node_map); | |
268 | ||
3461b0af | 269 | if (cpu_pda(cpu) && node != NUMA_NO_NODE) |
7891a24e MT |
270 | cpu_pda(cpu)->nodenumber = node; |
271 | ||
23ca4bba MT |
272 | if (cpu_to_node_map) |
273 | cpu_to_node_map[cpu] = node; | |
274 | ||
275 | else if (per_cpu_offset(cpu)) | |
276 | per_cpu(x86_cpu_to_node_map, cpu) = node; | |
277 | ||
278 | else | |
279 | Dprintk(KERN_INFO "Setting node for non-present cpu %d\n", cpu); | |
280 | } | |
281 | ||
282 | void __cpuinit numa_clear_node(int cpu) | |
283 | { | |
284 | numa_set_node(cpu, NUMA_NO_NODE); | |
285 | } | |
286 | ||
9f248bde MT |
287 | #ifndef CONFIG_DEBUG_PER_CPU_MAPS |
288 | ||
23ca4bba MT |
289 | void __cpuinit numa_add_cpu(int cpu) |
290 | { | |
291 | cpu_set(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]); | |
292 | } | |
293 | ||
294 | void __cpuinit numa_remove_cpu(int cpu) | |
295 | { | |
296 | cpu_clear(cpu, node_to_cpumask_map[cpu_to_node(cpu)]); | |
297 | } | |
23ca4bba | 298 | |
9f248bde MT |
299 | #else /* CONFIG_DEBUG_PER_CPU_MAPS */ |
300 | ||
301 | /* | |
302 | * --------- debug versions of the numa functions --------- | |
303 | */ | |
304 | static void __cpuinit numa_set_cpumask(int cpu, int enable) | |
305 | { | |
306 | int node = cpu_to_node(cpu); | |
307 | cpumask_t *mask; | |
308 | char buf[64]; | |
309 | ||
310 | if (node_to_cpumask_map == NULL) { | |
311 | printk(KERN_ERR "node_to_cpumask_map NULL\n"); | |
312 | dump_stack(); | |
313 | return; | |
314 | } | |
315 | ||
316 | mask = &node_to_cpumask_map[node]; | |
317 | if (enable) | |
318 | cpu_set(cpu, *mask); | |
319 | else | |
320 | cpu_clear(cpu, *mask); | |
321 | ||
322 | cpulist_scnprintf(buf, sizeof(buf), *mask); | |
323 | printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n", | |
324 | enable? "numa_add_cpu":"numa_remove_cpu", cpu, node, buf); | |
325 | } | |
326 | ||
327 | void __cpuinit numa_add_cpu(int cpu) | |
328 | { | |
329 | numa_set_cpumask(cpu, 1); | |
330 | } | |
331 | ||
332 | void __cpuinit numa_remove_cpu(int cpu) | |
333 | { | |
334 | numa_set_cpumask(cpu, 0); | |
335 | } | |
23ca4bba MT |
336 | |
337 | int cpu_to_node(int cpu) | |
338 | { | |
339 | if (early_per_cpu_ptr(x86_cpu_to_node_map)) { | |
340 | printk(KERN_WARNING | |
341 | "cpu_to_node(%d): usage too early!\n", cpu); | |
342 | dump_stack(); | |
343 | return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu]; | |
344 | } | |
345 | return per_cpu(x86_cpu_to_node_map, cpu); | |
346 | } | |
347 | EXPORT_SYMBOL(cpu_to_node); | |
348 | ||
9f248bde MT |
349 | /* |
350 | * Same function as cpu_to_node() but used if called before the | |
351 | * per_cpu areas are setup. | |
352 | */ | |
23ca4bba MT |
353 | int early_cpu_to_node(int cpu) |
354 | { | |
355 | if (early_per_cpu_ptr(x86_cpu_to_node_map)) | |
356 | return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu]; | |
357 | ||
358 | if (!per_cpu_offset(cpu)) { | |
359 | printk(KERN_WARNING | |
360 | "early_cpu_to_node(%d): no per_cpu area!\n", cpu); | |
9f248bde | 361 | dump_stack(); |
23ca4bba MT |
362 | return NUMA_NO_NODE; |
363 | } | |
364 | return per_cpu(x86_cpu_to_node_map, cpu); | |
365 | } | |
9f248bde MT |
366 | |
367 | /* | |
368 | * Returns a pointer to the bitmask of CPUs on Node 'node'. | |
369 | */ | |
370 | cpumask_t *_node_to_cpumask_ptr(int node) | |
371 | { | |
372 | if (node_to_cpumask_map == NULL) { | |
373 | printk(KERN_WARNING | |
374 | "_node_to_cpumask_ptr(%d): no node_to_cpumask_map!\n", | |
375 | node); | |
376 | dump_stack(); | |
377 | return &cpu_online_map; | |
378 | } | |
03db1f74 | 379 | BUG_ON(node >= nr_node_ids); |
9f248bde MT |
380 | return &node_to_cpumask_map[node]; |
381 | } | |
382 | EXPORT_SYMBOL(_node_to_cpumask_ptr); | |
383 | ||
384 | /* | |
385 | * Returns a bitmask of CPUs on Node 'node'. | |
386 | */ | |
387 | cpumask_t node_to_cpumask(int node) | |
388 | { | |
389 | if (node_to_cpumask_map == NULL) { | |
390 | printk(KERN_WARNING | |
391 | "node_to_cpumask(%d): no node_to_cpumask_map!\n", node); | |
392 | dump_stack(); | |
393 | return cpu_online_map; | |
394 | } | |
03db1f74 | 395 | BUG_ON(node >= nr_node_ids); |
9f248bde MT |
396 | return node_to_cpumask_map[node]; |
397 | } | |
398 | EXPORT_SYMBOL(node_to_cpumask); | |
399 | ||
400 | /* | |
401 | * --------- end of debug versions of the numa functions --------- | |
402 | */ | |
403 | ||
404 | #endif /* CONFIG_DEBUG_PER_CPU_MAPS */ | |
405 | ||
406 | #endif /* X86_64_NUMA */ |