]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - arch/x86/kernel/setup_percpu.c
genirq: remove irq_to_desc_alloc
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / kernel / setup_percpu.c
CommitLineData
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>
1ecd2765 6#include <linux/kexec.h>
17b4cceb 7#include <linux/crash_dump.h>
4fe29a85
GOC
8#include <asm/smp.h>
9#include <asm/percpu.h>
10#include <asm/sections.h>
11#include <asm/processor.h>
12#include <asm/setup.h>
13#include <asm/topology.h>
0fc0906e 14#include <asm/mpspec.h>
76eb4131 15#include <asm/apicdef.h>
1ecd2765 16#include <asm/highmem.h>
76eb4131 17
f8955ebe 18#ifdef CONFIG_X86_LOCAL_APIC
2fe60147
AS
19unsigned int num_processors;
20unsigned disabled_cpus __cpuinitdata;
21/* Processor that is doing the boot up */
22unsigned int boot_cpu_physical_apicid = -1U;
e0da3364 23unsigned int max_physical_apicid;
2fe60147
AS
24EXPORT_SYMBOL(boot_cpu_physical_apicid);
25
0fc0906e
AS
26/* Bitmask of physically existing CPUs */
27physid_mask_t phys_cpu_present_map;
f8955ebe 28#endif
0fc0906e 29
23ca4bba
MT
30/* map cpu index to physical APIC ID */
31DEFINE_EARLY_PER_CPU(u16, x86_cpu_to_apicid, BAD_APICID);
32DEFINE_EARLY_PER_CPU(u16, x86_bios_cpu_apicid, BAD_APICID);
33EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_apicid);
34EXPORT_EARLY_PER_CPU_SYMBOL(x86_bios_cpu_apicid);
35
36#if defined(CONFIG_NUMA) && defined(CONFIG_X86_64)
37#define X86_64_NUMA 1
38
7891a24e 39/* map cpu index to node index */
23ca4bba
MT
40DEFINE_EARLY_PER_CPU(int, x86_cpu_to_node_map, NUMA_NO_NODE);
41EXPORT_EARLY_PER_CPU_SYMBOL(x86_cpu_to_node_map);
9f248bde
MT
42
43/* which logical CPUs are on which nodes */
44cpumask_t *node_to_cpumask_map;
45EXPORT_SYMBOL(node_to_cpumask_map);
46
47/* setup node_to_cpumask_map */
48static void __init setup_node_to_cpumask_map(void);
49
50#else
51static inline void setup_node_to_cpumask_map(void) { }
23ca4bba
MT
52#endif
53
f8955ebe 54#if defined(CONFIG_HAVE_SETUP_PER_CPU_AREA) && defined(CONFIG_X86_SMP)
4fe29a85
GOC
55/*
56 * Copy data used in early init routines from the initial arrays to the
57 * per cpu data areas. These arrays then become expendable and the
58 * *_early_ptr's are zeroed indicating that the static arrays are gone.
59 */
60static void __init setup_per_cpu_maps(void)
61{
62 int cpu;
63
64 for_each_possible_cpu(cpu) {
23ca4bba
MT
65 per_cpu(x86_cpu_to_apicid, cpu) =
66 early_per_cpu_map(x86_cpu_to_apicid, cpu);
b447a468 67 per_cpu(x86_bios_cpu_apicid, cpu) =
23ca4bba
MT
68 early_per_cpu_map(x86_bios_cpu_apicid, cpu);
69#ifdef X86_64_NUMA
b447a468 70 per_cpu(x86_cpu_to_node_map, cpu) =
23ca4bba 71 early_per_cpu_map(x86_cpu_to_node_map, cpu);
4fe29a85
GOC
72#endif
73 }
74
75 /* indicate the early static arrays will soon be gone */
23ca4bba
MT
76 early_per_cpu_ptr(x86_cpu_to_apicid) = NULL;
77 early_per_cpu_ptr(x86_bios_cpu_apicid) = NULL;
78#ifdef X86_64_NUMA
79 early_per_cpu_ptr(x86_cpu_to_node_map) = NULL;
4fe29a85
GOC
80#endif
81}
82
83#ifdef CONFIG_X86_32
84/*
85 * Great future not-so-futuristic plan: make i386 and x86_64 do it
86 * the same way
87 */
88unsigned long __per_cpu_offset[NR_CPUS] __read_mostly;
89EXPORT_SYMBOL(__per_cpu_offset);
3461b0af
MT
90static inline void setup_cpu_pda_map(void) { }
91
92#elif !defined(CONFIG_SMP)
93static inline void setup_cpu_pda_map(void) { }
94
95#else /* CONFIG_SMP && CONFIG_X86_64 */
96
97/*
98 * Allocate cpu_pda pointer table and array via alloc_bootmem.
99 */
100static void __init setup_cpu_pda_map(void)
101{
102 char *pda;
103 struct x8664_pda **new_cpu_pda;
104 unsigned long size;
105 int cpu;
106
107 size = roundup(sizeof(struct x8664_pda), cache_line_size());
108
109 /* allocate cpu_pda array and pointer table */
110 {
111 unsigned long tsize = nr_cpu_ids * sizeof(void *);
112 unsigned long asize = size * (nr_cpu_ids - 1);
113
114 tsize = roundup(tsize, cache_line_size());
115 new_cpu_pda = alloc_bootmem(tsize + asize);
116 pda = (char *)new_cpu_pda + tsize;
117 }
118
119 /* initialize pointer table to static pda's */
120 for_each_possible_cpu(cpu) {
121 if (cpu == 0) {
122 /* leave boot cpu pda in place */
123 new_cpu_pda[0] = cpu_pda(0);
124 continue;
125 }
126 new_cpu_pda[cpu] = (struct x8664_pda *)pda;
127 new_cpu_pda[cpu]->in_bootmem = 1;
128 pda += size;
129 }
130
131 /* point to new pointer table */
132 _cpu_pda = new_cpu_pda;
133}
4fe29a85
GOC
134#endif
135
136/*
137 * Great future plan:
138 * Declare PDA itself and support (irqstack,tss,pgd) as per cpu data.
139 * Always point %gs to its beginning
140 */
141void __init setup_per_cpu_areas(void)
142{
1f8ff037 143 ssize_t size, old_size, da_size;
3461b0af
MT
144 char *ptr;
145 int cpu;
1f8ff037 146 unsigned long align = 1;
4fe29a85 147
3461b0af
MT
148 /* Setup cpu_pda map */
149 setup_cpu_pda_map();
150
4fe29a85 151 /* Copy section for each CPU (we discard the original) */
1f3fcd4b 152 old_size = PERCPU_ENOUGH_ROOM;
1f8ff037
YL
153 da_size = per_cpu_dyn_array_size(&align);
154 align = max_t(unsigned long, PAGE_SIZE, align);
155 size = roundup(old_size + da_size, align);
053713f5 156 printk(KERN_INFO "PERCPU: Allocating %zd bytes of per cpu data\n",
4fe29a85 157 size);
b447a468 158
3461b0af 159 for_each_possible_cpu(cpu) {
4fe29a85 160#ifndef CONFIG_NEED_MULTIPLE_NODES
1f8ff037
YL
161 ptr = __alloc_bootmem(size, align,
162 __pa(MAX_DMA_ADDRESS));
4fe29a85 163#else
3461b0af 164 int node = early_cpu_to_node(cpu);
b447a468 165 if (!node_online(node) || !NODE_DATA(node)) {
1f8ff037
YL
166 ptr = __alloc_bootmem(size, align,
167 __pa(MAX_DMA_ADDRESS));
b447a468 168 printk(KERN_INFO
23ca4bba 169 "cpu %d has no node %d or node-local memory\n",
3461b0af 170 cpu, node);
a677f58a
YL
171 if (ptr)
172 printk(KERN_DEBUG "per cpu data for cpu%d at %016lx\n",
173 cpu, __pa(ptr));
b447a468 174 }
a677f58a 175 else {
1f8ff037
YL
176 ptr = __alloc_bootmem_node(NODE_DATA(node), size, align,
177 __pa(MAX_DMA_ADDRESS));
a677f58a
YL
178 if (ptr)
179 printk(KERN_DEBUG "per cpu data for cpu%d on node%d at %016lx\n",
180 cpu, node, __pa(ptr));
181 }
4fe29a85 182#endif
3461b0af 183 per_cpu_offset(cpu) = ptr - __per_cpu_start;
4fe29a85 184 memcpy(ptr, __per_cpu_start, __per_cpu_end - __per_cpu_start);
9f0e8d04 185
1f3fcd4b
YL
186 per_cpu_alloc_dyn_array(cpu, ptr + old_size);
187
4fe29a85
GOC
188 }
189
9f248bde
MT
190 printk(KERN_DEBUG "NR_CPUS: %d, nr_cpu_ids: %d, nr_node_ids %d\n",
191 NR_CPUS, nr_cpu_ids, nr_node_ids);
9f0e8d04 192
b447a468 193 /* Setup percpu data maps */
4fe29a85 194 setup_per_cpu_maps();
9f0e8d04 195
9f248bde
MT
196 /* Setup node to cpumask map */
197 setup_node_to_cpumask_map();
4fe29a85
GOC
198}
199
200#endif
c45a707d 201
23ca4bba 202#ifdef X86_64_NUMA
9f248bde
MT
203
204/*
205 * Allocate node_to_cpumask_map based on number of available nodes
206 * Requires node_possible_map to be valid.
207 *
208 * Note: node_to_cpumask() is not valid until after this is done.
209 */
210static void __init setup_node_to_cpumask_map(void)
211{
212 unsigned int node, num = 0;
213 cpumask_t *map;
214
215 /* setup nr_node_ids if not done yet */
216 if (nr_node_ids == MAX_NUMNODES) {
217 for_each_node_mask(node, node_possible_map)
218 num = node;
219 nr_node_ids = num + 1;
220 }
221
222 /* allocate the map */
223 map = alloc_bootmem_low(nr_node_ids * sizeof(cpumask_t));
224
cfc1b9a6
TG
225 pr_debug(KERN_DEBUG "Node to cpumask map at %p for %d nodes\n",
226 map, nr_node_ids);
9f248bde
MT
227
228 /* node_to_cpumask() will now work */
229 node_to_cpumask_map = map;
230}
231
23ca4bba
MT
232void __cpuinit numa_set_node(int cpu, int node)
233{
234 int *cpu_to_node_map = early_per_cpu_ptr(x86_cpu_to_node_map);
235
3461b0af 236 if (cpu_pda(cpu) && node != NUMA_NO_NODE)
7891a24e
MT
237 cpu_pda(cpu)->nodenumber = node;
238
23ca4bba
MT
239 if (cpu_to_node_map)
240 cpu_to_node_map[cpu] = node;
241
242 else if (per_cpu_offset(cpu))
243 per_cpu(x86_cpu_to_node_map, cpu) = node;
244
245 else
cfc1b9a6 246 pr_debug("Setting node for non-present cpu %d\n", cpu);
23ca4bba
MT
247}
248
249void __cpuinit numa_clear_node(int cpu)
250{
251 numa_set_node(cpu, NUMA_NO_NODE);
252}
253
9f248bde
MT
254#ifndef CONFIG_DEBUG_PER_CPU_MAPS
255
23ca4bba
MT
256void __cpuinit numa_add_cpu(int cpu)
257{
258 cpu_set(cpu, node_to_cpumask_map[early_cpu_to_node(cpu)]);
259}
260
261void __cpuinit numa_remove_cpu(int cpu)
262{
263 cpu_clear(cpu, node_to_cpumask_map[cpu_to_node(cpu)]);
264}
23ca4bba 265
9f248bde
MT
266#else /* CONFIG_DEBUG_PER_CPU_MAPS */
267
268/*
269 * --------- debug versions of the numa functions ---------
270 */
271static void __cpuinit numa_set_cpumask(int cpu, int enable)
272{
273 int node = cpu_to_node(cpu);
274 cpumask_t *mask;
275 char buf[64];
276
277 if (node_to_cpumask_map == NULL) {
278 printk(KERN_ERR "node_to_cpumask_map NULL\n");
279 dump_stack();
280 return;
281 }
282
283 mask = &node_to_cpumask_map[node];
284 if (enable)
285 cpu_set(cpu, *mask);
286 else
287 cpu_clear(cpu, *mask);
288
289 cpulist_scnprintf(buf, sizeof(buf), *mask);
290 printk(KERN_DEBUG "%s cpu %d node %d: mask now %s\n",
291 enable? "numa_add_cpu":"numa_remove_cpu", cpu, node, buf);
292 }
293
294void __cpuinit numa_add_cpu(int cpu)
295{
296 numa_set_cpumask(cpu, 1);
297}
298
299void __cpuinit numa_remove_cpu(int cpu)
300{
301 numa_set_cpumask(cpu, 0);
302}
23ca4bba
MT
303
304int cpu_to_node(int cpu)
305{
306 if (early_per_cpu_ptr(x86_cpu_to_node_map)) {
307 printk(KERN_WARNING
308 "cpu_to_node(%d): usage too early!\n", cpu);
309 dump_stack();
310 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
311 }
312 return per_cpu(x86_cpu_to_node_map, cpu);
313}
314EXPORT_SYMBOL(cpu_to_node);
315
9f248bde
MT
316/*
317 * Same function as cpu_to_node() but used if called before the
318 * per_cpu areas are setup.
319 */
23ca4bba
MT
320int early_cpu_to_node(int cpu)
321{
322 if (early_per_cpu_ptr(x86_cpu_to_node_map))
323 return early_per_cpu_ptr(x86_cpu_to_node_map)[cpu];
324
325 if (!per_cpu_offset(cpu)) {
326 printk(KERN_WARNING
327 "early_cpu_to_node(%d): no per_cpu area!\n", cpu);
9f248bde 328 dump_stack();
23ca4bba
MT
329 return NUMA_NO_NODE;
330 }
331 return per_cpu(x86_cpu_to_node_map, cpu);
332}
9f248bde 333
6a2f47ca
MT
334
335/* empty cpumask */
336static const cpumask_t cpu_mask_none;
337
9f248bde
MT
338/*
339 * Returns a pointer to the bitmask of CPUs on Node 'node'.
340 */
11369f35 341const cpumask_t *_node_to_cpumask_ptr(int node)
9f248bde
MT
342{
343 if (node_to_cpumask_map == NULL) {
344 printk(KERN_WARNING
345 "_node_to_cpumask_ptr(%d): no node_to_cpumask_map!\n",
346 node);
347 dump_stack();
11369f35 348 return (const cpumask_t *)&cpu_online_map;
9f248bde 349 }
6a2f47ca
MT
350 if (node >= nr_node_ids) {
351 printk(KERN_WARNING
352 "_node_to_cpumask_ptr(%d): node > nr_node_ids(%d)\n",
353 node, nr_node_ids);
354 dump_stack();
11369f35 355 return &cpu_mask_none;
6a2f47ca 356 }
11369f35 357 return &node_to_cpumask_map[node];
9f248bde
MT
358}
359EXPORT_SYMBOL(_node_to_cpumask_ptr);
360
361/*
362 * Returns a bitmask of CPUs on Node 'node'.
6a2f47ca
MT
363 *
364 * Side note: this function creates the returned cpumask on the stack
365 * so with a high NR_CPUS count, excessive stack space is used. The
366 * node_to_cpumask_ptr function should be used whenever possible.
9f248bde
MT
367 */
368cpumask_t node_to_cpumask(int node)
369{
370 if (node_to_cpumask_map == NULL) {
371 printk(KERN_WARNING
372 "node_to_cpumask(%d): no node_to_cpumask_map!\n", node);
373 dump_stack();
374 return cpu_online_map;
375 }
6a2f47ca
MT
376 if (node >= nr_node_ids) {
377 printk(KERN_WARNING
378 "node_to_cpumask(%d): node > nr_node_ids(%d)\n",
379 node, nr_node_ids);
380 dump_stack();
381 return cpu_mask_none;
382 }
9f248bde
MT
383 return node_to_cpumask_map[node];
384}
385EXPORT_SYMBOL(node_to_cpumask);
386
387/*
388 * --------- end of debug versions of the numa functions ---------
389 */
390
391#endif /* CONFIG_DEBUG_PER_CPU_MAPS */
392
393#endif /* X86_64_NUMA */
1ecd2765 394