Commit | Line | Data |
---|---|---|
0332c2d4 ME |
1 | /* |
2 | * pseries CPU Hotplug infrastructure. | |
3 | * | |
413f7c40 ME |
4 | * Split out from arch/powerpc/platforms/pseries/setup.c |
5 | * arch/powerpc/kernel/rtas.c, and arch/powerpc/platforms/pseries/smp.c | |
0332c2d4 ME |
6 | * |
7 | * Peter Bergner, IBM March 2001. | |
8 | * Copyright (C) 2001 IBM. | |
413f7c40 ME |
9 | * Dave Engebretsen, Peter Bergner, and |
10 | * Mike Corrigan {engebret|bergner|mikec}@us.ibm.com | |
11 | * Plus various changes from other IBM teams... | |
0332c2d4 ME |
12 | * |
13 | * Copyright (C) 2006 Michael Ellerman, IBM Corporation | |
14 | * | |
15 | * This program is free software; you can redistribute it and/or | |
16 | * modify it under the terms of the GNU General Public License | |
17 | * as published by the Free Software Foundation; either version | |
18 | * 2 of the License, or (at your option) any later version. | |
19 | */ | |
20 | ||
e666ae0b NF |
21 | #define pr_fmt(fmt) "pseries-hotplug-cpu: " fmt |
22 | ||
0332c2d4 | 23 | #include <linux/kernel.h> |
0b05ac6e | 24 | #include <linux/interrupt.h> |
0332c2d4 | 25 | #include <linux/delay.h> |
62fe91bb | 26 | #include <linux/sched.h> /* for idle_task_exit */ |
0332c2d4 | 27 | #include <linux/cpu.h> |
1cf3d8b3 | 28 | #include <linux/of.h> |
ac713800 | 29 | #include <linux/slab.h> |
0332c2d4 ME |
30 | #include <asm/prom.h> |
31 | #include <asm/rtas.h> | |
32 | #include <asm/firmware.h> | |
33 | #include <asm/machdep.h> | |
34 | #include <asm/vdso_datapage.h> | |
0b05ac6e | 35 | #include <asm/xics.h> |
212bebb4 DD |
36 | #include <asm/plpar_wrappers.h> |
37 | ||
183deeea | 38 | #include "pseries.h" |
3aa565f5 | 39 | #include "offline_states.h" |
0332c2d4 ME |
40 | |
41 | /* This version can't take the spinlock, because it never returns */ | |
41dd03a9 | 42 | static int rtas_stop_self_token = RTAS_UNKNOWN_SERVICE; |
0332c2d4 | 43 | |
3aa565f5 GS |
44 | static DEFINE_PER_CPU(enum cpu_state_vals, preferred_offline_state) = |
45 | CPU_STATE_OFFLINE; | |
46 | static DEFINE_PER_CPU(enum cpu_state_vals, current_state) = CPU_STATE_OFFLINE; | |
47 | ||
48 | static enum cpu_state_vals default_offline_state = CPU_STATE_OFFLINE; | |
49 | ||
4cc7ecb7 | 50 | static bool cede_offline_enabled __read_mostly = true; |
3aa565f5 GS |
51 | |
52 | /* | |
53 | * Enable/disable cede_offline when available. | |
54 | */ | |
55 | static int __init setup_cede_offline(char *str) | |
56 | { | |
4cc7ecb7 | 57 | return (kstrtobool(str, &cede_offline_enabled) == 0); |
3aa565f5 GS |
58 | } |
59 | ||
60 | __setup("cede_offline=", setup_cede_offline); | |
61 | ||
62 | enum cpu_state_vals get_cpu_current_state(int cpu) | |
63 | { | |
64 | return per_cpu(current_state, cpu); | |
65 | } | |
66 | ||
67 | void set_cpu_current_state(int cpu, enum cpu_state_vals state) | |
68 | { | |
69 | per_cpu(current_state, cpu) = state; | |
70 | } | |
71 | ||
72 | enum cpu_state_vals get_preferred_offline_state(int cpu) | |
73 | { | |
74 | return per_cpu(preferred_offline_state, cpu); | |
75 | } | |
76 | ||
77 | void set_preferred_offline_state(int cpu, enum cpu_state_vals state) | |
78 | { | |
79 | per_cpu(preferred_offline_state, cpu) = state; | |
80 | } | |
81 | ||
82 | void set_default_offline_state(int cpu) | |
83 | { | |
84 | per_cpu(preferred_offline_state, cpu) = default_offline_state; | |
85 | } | |
86 | ||
04da6af9 | 87 | static void rtas_stop_self(void) |
0332c2d4 | 88 | { |
b2e8590f | 89 | static struct rtas_args args; |
4fb8d027 | 90 | |
0332c2d4 ME |
91 | local_irq_disable(); |
92 | ||
41dd03a9 | 93 | BUG_ON(rtas_stop_self_token == RTAS_UNKNOWN_SERVICE); |
0332c2d4 ME |
94 | |
95 | printk("cpu %u (hwid %u) Ready to die...\n", | |
96 | smp_processor_id(), hard_smp_processor_id()); | |
b2e8590f ME |
97 | |
98 | rtas_call_unlocked(&args, rtas_stop_self_token, 0, 1, NULL); | |
0332c2d4 ME |
99 | |
100 | panic("Alas, I survived.\n"); | |
101 | } | |
102 | ||
06ba30b6 | 103 | static void pseries_mach_cpu_die(void) |
04da6af9 | 104 | { |
3aa565f5 GS |
105 | unsigned int cpu = smp_processor_id(); |
106 | unsigned int hwcpu = hard_smp_processor_id(); | |
107 | u8 cede_latency_hint = 0; | |
108 | ||
04da6af9 ME |
109 | local_irq_disable(); |
110 | idle_task_exit(); | |
c3e8506c | 111 | xics_teardown_cpu(); |
3aa565f5 GS |
112 | |
113 | if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) { | |
114 | set_cpu_current_state(cpu, CPU_STATE_INACTIVE); | |
32d8ad4e BK |
115 | if (ppc_md.suspend_disable_cpu) |
116 | ppc_md.suspend_disable_cpu(); | |
117 | ||
3aa565f5 GS |
118 | cede_latency_hint = 2; |
119 | ||
120 | get_lppaca()->idle = 1; | |
f13c13a0 | 121 | if (!lppaca_shared_proc(get_lppaca())) |
3aa565f5 GS |
122 | get_lppaca()->donate_dedicated_cpu = 1; |
123 | ||
3aa565f5 | 124 | while (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) { |
fb912568 LZ |
125 | while (!prep_irq_for_idle()) { |
126 | local_irq_enable(); | |
127 | local_irq_disable(); | |
128 | } | |
129 | ||
3aa565f5 | 130 | extended_cede_processor(cede_latency_hint); |
3aa565f5 GS |
131 | } |
132 | ||
fb912568 LZ |
133 | local_irq_disable(); |
134 | ||
f13c13a0 | 135 | if (!lppaca_shared_proc(get_lppaca())) |
3aa565f5 GS |
136 | get_lppaca()->donate_dedicated_cpu = 0; |
137 | get_lppaca()->idle = 0; | |
3aa565f5 | 138 | |
0212f260 | 139 | if (get_preferred_offline_state(cpu) == CPU_STATE_ONLINE) { |
598c8231 | 140 | unregister_slb_shadow(hwcpu); |
3aa565f5 | 141 | |
fb912568 | 142 | hard_irq_disable(); |
0212f260 VS |
143 | /* |
144 | * Call to start_secondary_resume() will not return. | |
145 | * Kernel stack will be reset and start_secondary() | |
146 | * will be called to continue the online operation. | |
147 | */ | |
148 | start_secondary_resume(); | |
149 | } | |
150 | } | |
3aa565f5 | 151 | |
0212f260 VS |
152 | /* Requested state is CPU_STATE_OFFLINE at this point */ |
153 | WARN_ON(get_preferred_offline_state(cpu) != CPU_STATE_OFFLINE); | |
3aa565f5 | 154 | |
0212f260 | 155 | set_cpu_current_state(cpu, CPU_STATE_OFFLINE); |
598c8231 | 156 | unregister_slb_shadow(hwcpu); |
0212f260 | 157 | rtas_stop_self(); |
3aa565f5 | 158 | |
04da6af9 ME |
159 | /* Should never get here... */ |
160 | BUG(); | |
161 | for(;;); | |
162 | } | |
163 | ||
06ba30b6 | 164 | static int pseries_cpu_disable(void) |
413f7c40 ME |
165 | { |
166 | int cpu = smp_processor_id(); | |
167 | ||
ea0f1cab | 168 | set_cpu_online(cpu, false); |
413f7c40 ME |
169 | vdso_data->processorCount--; |
170 | ||
171 | /*fix boot_cpuid here*/ | |
172 | if (cpu == boot_cpuid) | |
8729faaa | 173 | boot_cpuid = cpumask_any(cpu_online_mask); |
413f7c40 ME |
174 | |
175 | /* FIXME: abstract this to not be platform specific later on */ | |
176 | xics_migrate_irqs_away(); | |
177 | return 0; | |
178 | } | |
179 | ||
3aa565f5 GS |
180 | /* |
181 | * pseries_cpu_die: Wait for the cpu to die. | |
182 | * @cpu: logical processor id of the CPU whose death we're awaiting. | |
183 | * | |
184 | * This function is called from the context of the thread which is performing | |
185 | * the cpu-offline. Here we wait for long enough to allow the cpu in question | |
186 | * to self-destroy so that the cpu-offline thread can send the CPU_DEAD | |
187 | * notifications. | |
188 | * | |
189 | * OTOH, pseries_mach_cpu_die() is called by the @cpu when it wants to | |
190 | * self-destruct. | |
191 | */ | |
06ba30b6 | 192 | static void pseries_cpu_die(unsigned int cpu) |
413f7c40 ME |
193 | { |
194 | int tries; | |
3aa565f5 | 195 | int cpu_status = 1; |
413f7c40 ME |
196 | unsigned int pcpu = get_hard_smp_processor_id(cpu); |
197 | ||
3aa565f5 GS |
198 | if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) { |
199 | cpu_status = 1; | |
940ce422 | 200 | for (tries = 0; tries < 5000; tries++) { |
3aa565f5 GS |
201 | if (get_cpu_current_state(cpu) == CPU_STATE_INACTIVE) { |
202 | cpu_status = 0; | |
203 | break; | |
204 | } | |
940ce422 | 205 | msleep(1); |
3aa565f5 GS |
206 | } |
207 | } else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) { | |
208 | ||
209 | for (tries = 0; tries < 25; tries++) { | |
f8b67691 MN |
210 | cpu_status = smp_query_cpu_stopped(pcpu); |
211 | if (cpu_status == QCSS_STOPPED || | |
212 | cpu_status == QCSS_HARDWARE_ERROR) | |
3aa565f5 GS |
213 | break; |
214 | cpu_relax(); | |
215 | } | |
413f7c40 | 216 | } |
3aa565f5 | 217 | |
413f7c40 ME |
218 | if (cpu_status != 0) { |
219 | printk("Querying DEAD? cpu %i (%i) shows %i\n", | |
220 | cpu, pcpu, cpu_status); | |
221 | } | |
222 | ||
25985edc | 223 | /* Isolation and deallocation are definitely done by |
413f7c40 ME |
224 | * drslot_chrp_cpu. If they were not they would be |
225 | * done here. Change isolate state to Isolate and | |
226 | * change allocation-state to Unusable. | |
227 | */ | |
228 | paca[cpu].cpu_start = 0; | |
229 | } | |
230 | ||
231 | /* | |
828a6986 | 232 | * Update cpu_present_mask and paca(s) for a new cpu node. The wrinkle |
413f7c40 ME |
233 | * here is that a cpu device node may represent up to two logical cpus |
234 | * in the SMT case. We must honor the assumption in other code that | |
235 | * the logical ids for sibling SMT threads x and y are adjacent, such | |
236 | * that x^1 == y and y^1 == x. | |
237 | */ | |
06ba30b6 | 238 | static int pseries_add_processor(struct device_node *np) |
413f7c40 ME |
239 | { |
240 | unsigned int cpu; | |
8729faaa | 241 | cpumask_var_t candidate_mask, tmp; |
413f7c40 | 242 | int err = -ENOSPC, len, nthreads, i; |
d6f1e7ab | 243 | const __be32 *intserv; |
413f7c40 | 244 | |
e2eb6392 | 245 | intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len); |
413f7c40 ME |
246 | if (!intserv) |
247 | return 0; | |
248 | ||
8729faaa AB |
249 | zalloc_cpumask_var(&candidate_mask, GFP_KERNEL); |
250 | zalloc_cpumask_var(&tmp, GFP_KERNEL); | |
251 | ||
413f7c40 ME |
252 | nthreads = len / sizeof(u32); |
253 | for (i = 0; i < nthreads; i++) | |
8729faaa | 254 | cpumask_set_cpu(i, tmp); |
413f7c40 | 255 | |
86ef5c9a | 256 | cpu_maps_update_begin(); |
413f7c40 | 257 | |
8729faaa | 258 | BUG_ON(!cpumask_subset(cpu_present_mask, cpu_possible_mask)); |
413f7c40 ME |
259 | |
260 | /* Get a bitmap of unoccupied slots. */ | |
8729faaa AB |
261 | cpumask_xor(candidate_mask, cpu_possible_mask, cpu_present_mask); |
262 | if (cpumask_empty(candidate_mask)) { | |
413f7c40 ME |
263 | /* If we get here, it most likely means that NR_CPUS is |
264 | * less than the partition's max processors setting. | |
265 | */ | |
266 | printk(KERN_ERR "Cannot add cpu %s; this system configuration" | |
267 | " supports %d logical cpus.\n", np->full_name, | |
53a448c3 | 268 | num_possible_cpus()); |
413f7c40 ME |
269 | goto out_unlock; |
270 | } | |
271 | ||
8729faaa AB |
272 | while (!cpumask_empty(tmp)) |
273 | if (cpumask_subset(tmp, candidate_mask)) | |
413f7c40 ME |
274 | /* Found a range where we can insert the new cpu(s) */ |
275 | break; | |
276 | else | |
8729faaa | 277 | cpumask_shift_left(tmp, tmp, nthreads); |
413f7c40 | 278 | |
8729faaa | 279 | if (cpumask_empty(tmp)) { |
828a6986 | 280 | printk(KERN_ERR "Unable to find space in cpu_present_mask for" |
413f7c40 ME |
281 | " processor %s with %d thread(s)\n", np->name, |
282 | nthreads); | |
283 | goto out_unlock; | |
284 | } | |
285 | ||
8729faaa | 286 | for_each_cpu(cpu, tmp) { |
104699c0 | 287 | BUG_ON(cpu_present(cpu)); |
ea0f1cab | 288 | set_cpu_present(cpu, true); |
d6f1e7ab | 289 | set_hard_smp_processor_id(cpu, be32_to_cpu(*intserv++)); |
413f7c40 ME |
290 | } |
291 | err = 0; | |
292 | out_unlock: | |
86ef5c9a | 293 | cpu_maps_update_done(); |
8729faaa AB |
294 | free_cpumask_var(candidate_mask); |
295 | free_cpumask_var(tmp); | |
413f7c40 ME |
296 | return err; |
297 | } | |
298 | ||
299 | /* | |
300 | * Update the present map for a cpu node which is going away, and set | |
301 | * the hard id in the paca(s) to -1 to be consistent with boot time | |
302 | * convention for non-present cpus. | |
303 | */ | |
06ba30b6 | 304 | static void pseries_remove_processor(struct device_node *np) |
413f7c40 ME |
305 | { |
306 | unsigned int cpu; | |
307 | int len, nthreads, i; | |
e36d1227 TF |
308 | const __be32 *intserv; |
309 | u32 thread; | |
413f7c40 | 310 | |
e2eb6392 | 311 | intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len); |
413f7c40 ME |
312 | if (!intserv) |
313 | return; | |
314 | ||
315 | nthreads = len / sizeof(u32); | |
316 | ||
86ef5c9a | 317 | cpu_maps_update_begin(); |
413f7c40 | 318 | for (i = 0; i < nthreads; i++) { |
e36d1227 | 319 | thread = be32_to_cpu(intserv[i]); |
413f7c40 | 320 | for_each_present_cpu(cpu) { |
e36d1227 | 321 | if (get_hard_smp_processor_id(cpu) != thread) |
413f7c40 ME |
322 | continue; |
323 | BUG_ON(cpu_online(cpu)); | |
ea0f1cab | 324 | set_cpu_present(cpu, false); |
413f7c40 ME |
325 | set_hard_smp_processor_id(cpu, -1); |
326 | break; | |
327 | } | |
8729faaa | 328 | if (cpu >= nr_cpu_ids) |
413f7c40 | 329 | printk(KERN_WARNING "Could not find cpu to remove " |
e36d1227 | 330 | "with physical id 0x%x\n", thread); |
413f7c40 | 331 | } |
86ef5c9a | 332 | cpu_maps_update_done(); |
413f7c40 ME |
333 | } |
334 | ||
183deeea NF |
335 | static int dlpar_online_cpu(struct device_node *dn) |
336 | { | |
337 | int rc = 0; | |
338 | unsigned int cpu; | |
339 | int len, nthreads, i; | |
340 | const __be32 *intserv; | |
341 | u32 thread; | |
342 | ||
343 | intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len); | |
344 | if (!intserv) | |
345 | return -EINVAL; | |
346 | ||
347 | nthreads = len / sizeof(u32); | |
348 | ||
349 | cpu_maps_update_begin(); | |
350 | for (i = 0; i < nthreads; i++) { | |
351 | thread = be32_to_cpu(intserv[i]); | |
352 | for_each_present_cpu(cpu) { | |
353 | if (get_hard_smp_processor_id(cpu) != thread) | |
354 | continue; | |
355 | BUG_ON(get_cpu_current_state(cpu) | |
356 | != CPU_STATE_OFFLINE); | |
357 | cpu_maps_update_done(); | |
358 | rc = device_online(get_cpu_device(cpu)); | |
359 | if (rc) | |
360 | goto out; | |
361 | cpu_maps_update_begin(); | |
362 | ||
363 | break; | |
364 | } | |
365 | if (cpu == num_possible_cpus()) | |
366 | printk(KERN_WARNING "Could not find cpu to online " | |
367 | "with physical id 0x%x\n", thread); | |
368 | } | |
369 | cpu_maps_update_done(); | |
370 | ||
371 | out: | |
372 | return rc; | |
373 | ||
374 | } | |
375 | ||
376 | static bool dlpar_cpu_exists(struct device_node *parent, u32 drc_index) | |
377 | { | |
378 | struct device_node *child = NULL; | |
379 | u32 my_drc_index; | |
380 | bool found; | |
381 | int rc; | |
382 | ||
383 | /* Assume cpu doesn't exist */ | |
384 | found = false; | |
385 | ||
386 | for_each_child_of_node(parent, child) { | |
387 | rc = of_property_read_u32(child, "ibm,my-drc-index", | |
388 | &my_drc_index); | |
389 | if (rc) | |
390 | continue; | |
391 | ||
392 | if (my_drc_index == drc_index) { | |
393 | of_node_put(child); | |
394 | found = true; | |
395 | break; | |
396 | } | |
397 | } | |
398 | ||
399 | return found; | |
400 | } | |
401 | ||
90edf184 NF |
402 | static bool valid_cpu_drc_index(struct device_node *parent, u32 drc_index) |
403 | { | |
404 | bool found = false; | |
405 | int rc, index; | |
406 | ||
407 | index = 0; | |
408 | while (!found) { | |
409 | u32 drc; | |
410 | ||
411 | rc = of_property_read_u32_index(parent, "ibm,drc-indexes", | |
412 | index++, &drc); | |
413 | if (rc) | |
414 | break; | |
415 | ||
416 | if (drc == drc_index) | |
417 | found = true; | |
418 | } | |
419 | ||
420 | return found; | |
421 | } | |
422 | ||
d98389f3 | 423 | static ssize_t dlpar_cpu_add(u32 drc_index) |
183deeea NF |
424 | { |
425 | struct device_node *dn, *parent; | |
e666ae0b NF |
426 | int rc, saved_rc; |
427 | ||
428 | pr_debug("Attempting to add CPU, drc index: %x\n", drc_index); | |
183deeea | 429 | |
183deeea | 430 | parent = of_find_node_by_path("/cpus"); |
e666ae0b NF |
431 | if (!parent) { |
432 | pr_warn("Failed to find CPU root node \"/cpus\"\n"); | |
183deeea | 433 | return -ENODEV; |
e666ae0b | 434 | } |
183deeea NF |
435 | |
436 | if (dlpar_cpu_exists(parent, drc_index)) { | |
437 | of_node_put(parent); | |
e666ae0b | 438 | pr_warn("CPU with drc index %x already exists\n", drc_index); |
183deeea NF |
439 | return -EINVAL; |
440 | } | |
441 | ||
90edf184 NF |
442 | if (!valid_cpu_drc_index(parent, drc_index)) { |
443 | of_node_put(parent); | |
444 | pr_warn("Cannot find CPU (drc index %x) to add.\n", drc_index); | |
445 | return -EINVAL; | |
446 | } | |
447 | ||
183deeea NF |
448 | rc = dlpar_acquire_drc(drc_index); |
449 | if (rc) { | |
e666ae0b NF |
450 | pr_warn("Failed to acquire DRC, rc: %d, drc index: %x\n", |
451 | rc, drc_index); | |
183deeea NF |
452 | of_node_put(parent); |
453 | return -EINVAL; | |
454 | } | |
455 | ||
456 | dn = dlpar_configure_connector(cpu_to_be32(drc_index), parent); | |
457 | of_node_put(parent); | |
e666ae0b NF |
458 | if (!dn) { |
459 | pr_warn("Failed call to configure-connector, drc index: %x\n", | |
460 | drc_index); | |
461 | dlpar_release_drc(drc_index); | |
183deeea | 462 | return -EINVAL; |
e666ae0b | 463 | } |
183deeea NF |
464 | |
465 | rc = dlpar_attach_node(dn); | |
466 | if (rc) { | |
e666ae0b NF |
467 | saved_rc = rc; |
468 | pr_warn("Failed to attach node %s, rc: %d, drc index: %x\n", | |
469 | dn->name, rc, drc_index); | |
470 | ||
471 | rc = dlpar_release_drc(drc_index); | |
472 | if (!rc) | |
473 | dlpar_free_cc_nodes(dn); | |
474 | ||
475 | return saved_rc; | |
183deeea NF |
476 | } |
477 | ||
478 | rc = dlpar_online_cpu(dn); | |
e666ae0b NF |
479 | if (rc) { |
480 | saved_rc = rc; | |
481 | pr_warn("Failed to online cpu %s, rc: %d, drc index: %x\n", | |
482 | dn->name, rc, drc_index); | |
483 | ||
484 | rc = dlpar_detach_node(dn); | |
485 | if (!rc) | |
486 | dlpar_release_drc(drc_index); | |
487 | ||
488 | return saved_rc; | |
489 | } | |
490 | ||
491 | pr_debug("Successfully added CPU %s, drc index: %x\n", dn->name, | |
492 | drc_index); | |
d98389f3 | 493 | return rc; |
183deeea NF |
494 | } |
495 | ||
496 | static int dlpar_offline_cpu(struct device_node *dn) | |
497 | { | |
498 | int rc = 0; | |
499 | unsigned int cpu; | |
500 | int len, nthreads, i; | |
501 | const __be32 *intserv; | |
502 | u32 thread; | |
503 | ||
504 | intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len); | |
505 | if (!intserv) | |
506 | return -EINVAL; | |
507 | ||
508 | nthreads = len / sizeof(u32); | |
509 | ||
510 | cpu_maps_update_begin(); | |
511 | for (i = 0; i < nthreads; i++) { | |
512 | thread = be32_to_cpu(intserv[i]); | |
513 | for_each_present_cpu(cpu) { | |
514 | if (get_hard_smp_processor_id(cpu) != thread) | |
515 | continue; | |
516 | ||
517 | if (get_cpu_current_state(cpu) == CPU_STATE_OFFLINE) | |
518 | break; | |
519 | ||
520 | if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) { | |
521 | set_preferred_offline_state(cpu, | |
522 | CPU_STATE_OFFLINE); | |
523 | cpu_maps_update_done(); | |
524 | rc = device_offline(get_cpu_device(cpu)); | |
525 | if (rc) | |
526 | goto out; | |
527 | cpu_maps_update_begin(); | |
528 | break; | |
529 | ||
530 | } | |
531 | ||
532 | /* | |
533 | * The cpu is in CPU_STATE_INACTIVE. | |
534 | * Upgrade it's state to CPU_STATE_OFFLINE. | |
535 | */ | |
536 | set_preferred_offline_state(cpu, CPU_STATE_OFFLINE); | |
537 | BUG_ON(plpar_hcall_norets(H_PROD, thread) | |
538 | != H_SUCCESS); | |
539 | __cpu_die(cpu); | |
540 | break; | |
541 | } | |
542 | if (cpu == num_possible_cpus()) | |
543 | printk(KERN_WARNING "Could not find cpu to offline with physical id 0x%x\n", thread); | |
544 | } | |
545 | cpu_maps_update_done(); | |
546 | ||
547 | out: | |
548 | return rc; | |
549 | ||
550 | } | |
551 | ||
d98389f3 NF |
552 | static ssize_t dlpar_cpu_remove(struct device_node *dn, u32 drc_index) |
553 | { | |
554 | int rc; | |
555 | ||
e666ae0b NF |
556 | pr_debug("Attemping to remove CPU %s, drc index: %x\n", |
557 | dn->name, drc_index); | |
558 | ||
d98389f3 | 559 | rc = dlpar_offline_cpu(dn); |
e666ae0b NF |
560 | if (rc) { |
561 | pr_warn("Failed to offline CPU %s, rc: %d\n", dn->name, rc); | |
d98389f3 | 562 | return -EINVAL; |
e666ae0b | 563 | } |
d98389f3 NF |
564 | |
565 | rc = dlpar_release_drc(drc_index); | |
e666ae0b NF |
566 | if (rc) { |
567 | pr_warn("Failed to release drc (%x) for CPU %s, rc: %d\n", | |
568 | drc_index, dn->name, rc); | |
569 | dlpar_online_cpu(dn); | |
d98389f3 | 570 | return rc; |
e666ae0b | 571 | } |
d98389f3 NF |
572 | |
573 | rc = dlpar_detach_node(dn); | |
e666ae0b NF |
574 | if (rc) { |
575 | int saved_rc = rc; | |
d98389f3 | 576 | |
e666ae0b NF |
577 | pr_warn("Failed to detach CPU %s, rc: %d", dn->name, rc); |
578 | ||
579 | rc = dlpar_acquire_drc(drc_index); | |
580 | if (!rc) | |
581 | dlpar_online_cpu(dn); | |
582 | ||
583 | return saved_rc; | |
584 | } | |
585 | ||
586 | pr_debug("Successfully removed CPU, drc index: %x\n", drc_index); | |
587 | return 0; | |
d98389f3 NF |
588 | } |
589 | ||
ac713800 NF |
590 | static struct device_node *cpu_drc_index_to_dn(u32 drc_index) |
591 | { | |
592 | struct device_node *dn; | |
593 | u32 my_index; | |
594 | int rc; | |
595 | ||
596 | for_each_node_by_type(dn, "cpu") { | |
597 | rc = of_property_read_u32(dn, "ibm,my-drc-index", &my_index); | |
598 | if (rc) | |
599 | continue; | |
600 | ||
601 | if (my_index == drc_index) | |
602 | break; | |
603 | } | |
604 | ||
605 | return dn; | |
606 | } | |
607 | ||
608 | static int dlpar_cpu_remove_by_index(u32 drc_index) | |
609 | { | |
610 | struct device_node *dn; | |
611 | int rc; | |
612 | ||
613 | dn = cpu_drc_index_to_dn(drc_index); | |
614 | if (!dn) { | |
615 | pr_warn("Cannot find CPU (drc index %x) to remove\n", | |
616 | drc_index); | |
617 | return -ENODEV; | |
618 | } | |
619 | ||
620 | rc = dlpar_cpu_remove(dn, drc_index); | |
621 | of_node_put(dn); | |
622 | return rc; | |
623 | } | |
624 | ||
625 | static int find_dlpar_cpus_to_remove(u32 *cpu_drcs, int cpus_to_remove) | |
626 | { | |
627 | struct device_node *dn; | |
628 | int cpus_found = 0; | |
629 | int rc; | |
630 | ||
631 | /* We want to find cpus_to_remove + 1 CPUs to ensure we do not | |
632 | * remove the last CPU. | |
633 | */ | |
634 | for_each_node_by_type(dn, "cpu") { | |
635 | cpus_found++; | |
636 | ||
637 | if (cpus_found > cpus_to_remove) { | |
638 | of_node_put(dn); | |
639 | break; | |
640 | } | |
641 | ||
642 | /* Note that cpus_found is always 1 ahead of the index | |
643 | * into the cpu_drcs array, so we use cpus_found - 1 | |
644 | */ | |
645 | rc = of_property_read_u32(dn, "ibm,my-drc-index", | |
646 | &cpu_drcs[cpus_found - 1]); | |
647 | if (rc) { | |
648 | pr_warn("Error occurred getting drc-index for %s\n", | |
649 | dn->name); | |
650 | of_node_put(dn); | |
651 | return -1; | |
652 | } | |
653 | } | |
654 | ||
655 | if (cpus_found < cpus_to_remove) { | |
656 | pr_warn("Failed to find enough CPUs (%d of %d) to remove\n", | |
657 | cpus_found, cpus_to_remove); | |
658 | } else if (cpus_found == cpus_to_remove) { | |
659 | pr_warn("Cannot remove all CPUs\n"); | |
660 | } | |
661 | ||
662 | return cpus_found; | |
663 | } | |
664 | ||
665 | static int dlpar_cpu_remove_by_count(u32 cpus_to_remove) | |
666 | { | |
667 | u32 *cpu_drcs; | |
668 | int cpus_found; | |
669 | int cpus_removed = 0; | |
670 | int i, rc; | |
671 | ||
672 | pr_debug("Attempting to hot-remove %d CPUs\n", cpus_to_remove); | |
673 | ||
674 | cpu_drcs = kcalloc(cpus_to_remove, sizeof(*cpu_drcs), GFP_KERNEL); | |
675 | if (!cpu_drcs) | |
676 | return -EINVAL; | |
677 | ||
678 | cpus_found = find_dlpar_cpus_to_remove(cpu_drcs, cpus_to_remove); | |
679 | if (cpus_found <= cpus_to_remove) { | |
680 | kfree(cpu_drcs); | |
681 | return -EINVAL; | |
682 | } | |
683 | ||
684 | for (i = 0; i < cpus_to_remove; i++) { | |
685 | rc = dlpar_cpu_remove_by_index(cpu_drcs[i]); | |
686 | if (rc) | |
687 | break; | |
688 | ||
689 | cpus_removed++; | |
690 | } | |
691 | ||
692 | if (cpus_removed != cpus_to_remove) { | |
693 | pr_warn("CPU hot-remove failed, adding back removed CPUs\n"); | |
694 | ||
695 | for (i = 0; i < cpus_removed; i++) | |
696 | dlpar_cpu_add(cpu_drcs[i]); | |
697 | ||
698 | rc = -EINVAL; | |
699 | } else { | |
700 | rc = 0; | |
701 | } | |
702 | ||
703 | kfree(cpu_drcs); | |
704 | return rc; | |
705 | } | |
706 | ||
90edf184 NF |
707 | static int find_dlpar_cpus_to_add(u32 *cpu_drcs, u32 cpus_to_add) |
708 | { | |
709 | struct device_node *parent; | |
710 | int cpus_found = 0; | |
711 | int index, rc; | |
712 | ||
713 | parent = of_find_node_by_path("/cpus"); | |
714 | if (!parent) { | |
715 | pr_warn("Could not find CPU root node in device tree\n"); | |
716 | kfree(cpu_drcs); | |
717 | return -1; | |
718 | } | |
719 | ||
720 | /* Search the ibm,drc-indexes array for possible CPU drcs to | |
721 | * add. Note that the format of the ibm,drc-indexes array is | |
722 | * the number of entries in the array followed by the array | |
723 | * of drc values so we start looking at index = 1. | |
724 | */ | |
725 | index = 1; | |
726 | while (cpus_found < cpus_to_add) { | |
727 | u32 drc; | |
728 | ||
729 | rc = of_property_read_u32_index(parent, "ibm,drc-indexes", | |
730 | index++, &drc); | |
731 | if (rc) | |
732 | break; | |
733 | ||
734 | if (dlpar_cpu_exists(parent, drc)) | |
735 | continue; | |
736 | ||
737 | cpu_drcs[cpus_found++] = drc; | |
738 | } | |
739 | ||
740 | of_node_put(parent); | |
741 | return cpus_found; | |
742 | } | |
743 | ||
744 | static int dlpar_cpu_add_by_count(u32 cpus_to_add) | |
745 | { | |
746 | u32 *cpu_drcs; | |
747 | int cpus_added = 0; | |
748 | int cpus_found; | |
749 | int i, rc; | |
750 | ||
751 | pr_debug("Attempting to hot-add %d CPUs\n", cpus_to_add); | |
752 | ||
753 | cpu_drcs = kcalloc(cpus_to_add, sizeof(*cpu_drcs), GFP_KERNEL); | |
754 | if (!cpu_drcs) | |
755 | return -EINVAL; | |
756 | ||
757 | cpus_found = find_dlpar_cpus_to_add(cpu_drcs, cpus_to_add); | |
758 | if (cpus_found < cpus_to_add) { | |
759 | pr_warn("Failed to find enough CPUs (%d of %d) to add\n", | |
760 | cpus_found, cpus_to_add); | |
761 | kfree(cpu_drcs); | |
762 | return -EINVAL; | |
763 | } | |
764 | ||
765 | for (i = 0; i < cpus_to_add; i++) { | |
766 | rc = dlpar_cpu_add(cpu_drcs[i]); | |
767 | if (rc) | |
768 | break; | |
769 | ||
770 | cpus_added++; | |
771 | } | |
772 | ||
773 | if (cpus_added < cpus_to_add) { | |
774 | pr_warn("CPU hot-add failed, removing any added CPUs\n"); | |
775 | ||
776 | for (i = 0; i < cpus_added; i++) | |
777 | dlpar_cpu_remove_by_index(cpu_drcs[i]); | |
778 | ||
779 | rc = -EINVAL; | |
780 | } else { | |
781 | rc = 0; | |
782 | } | |
783 | ||
784 | kfree(cpu_drcs); | |
785 | return rc; | |
786 | } | |
787 | ||
ac713800 NF |
788 | int dlpar_cpu(struct pseries_hp_errorlog *hp_elog) |
789 | { | |
790 | u32 count, drc_index; | |
791 | int rc; | |
792 | ||
793 | count = hp_elog->_drc_u.drc_count; | |
794 | drc_index = hp_elog->_drc_u.drc_index; | |
795 | ||
796 | lock_device_hotplug(); | |
797 | ||
798 | switch (hp_elog->action) { | |
799 | case PSERIES_HP_ELOG_ACTION_REMOVE: | |
800 | if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT) | |
801 | rc = dlpar_cpu_remove_by_count(count); | |
802 | else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX) | |
803 | rc = dlpar_cpu_remove_by_index(drc_index); | |
804 | else | |
805 | rc = -EINVAL; | |
806 | break; | |
90edf184 NF |
807 | case PSERIES_HP_ELOG_ACTION_ADD: |
808 | if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT) | |
809 | rc = dlpar_cpu_add_by_count(count); | |
810 | else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX) | |
811 | rc = dlpar_cpu_add(drc_index); | |
812 | else | |
813 | rc = -EINVAL; | |
814 | break; | |
ac713800 NF |
815 | default: |
816 | pr_err("Invalid action (%d) specified\n", hp_elog->action); | |
817 | rc = -EINVAL; | |
818 | break; | |
819 | } | |
820 | ||
821 | unlock_device_hotplug(); | |
822 | return rc; | |
823 | } | |
824 | ||
d98389f3 NF |
825 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE |
826 | ||
827 | static ssize_t dlpar_cpu_probe(const char *buf, size_t count) | |
828 | { | |
829 | u32 drc_index; | |
830 | int rc; | |
831 | ||
832 | rc = kstrtou32(buf, 0, &drc_index); | |
833 | if (rc) | |
834 | return -EINVAL; | |
835 | ||
836 | rc = dlpar_cpu_add(drc_index); | |
837 | ||
838 | return rc ? rc : count; | |
839 | } | |
840 | ||
183deeea NF |
841 | static ssize_t dlpar_cpu_release(const char *buf, size_t count) |
842 | { | |
843 | struct device_node *dn; | |
844 | u32 drc_index; | |
845 | int rc; | |
846 | ||
847 | dn = of_find_node_by_path(buf); | |
848 | if (!dn) | |
849 | return -EINVAL; | |
850 | ||
851 | rc = of_property_read_u32(dn, "ibm,my-drc-index", &drc_index); | |
852 | if (rc) { | |
853 | of_node_put(dn); | |
854 | return -EINVAL; | |
855 | } | |
856 | ||
d98389f3 | 857 | rc = dlpar_cpu_remove(dn, drc_index); |
183deeea NF |
858 | of_node_put(dn); |
859 | ||
d98389f3 | 860 | return rc ? rc : count; |
183deeea NF |
861 | } |
862 | ||
863 | #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ | |
864 | ||
06ba30b6 | 865 | static int pseries_smp_notifier(struct notifier_block *nb, |
f5242e5a | 866 | unsigned long action, void *data) |
413f7c40 | 867 | { |
f5242e5a | 868 | struct of_reconfig_data *rd = data; |
de2780a3 | 869 | int err = 0; |
413f7c40 ME |
870 | |
871 | switch (action) { | |
1cf3d8b3 | 872 | case OF_RECONFIG_ATTACH_NODE: |
f5242e5a | 873 | err = pseries_add_processor(rd->dn); |
413f7c40 | 874 | break; |
1cf3d8b3 | 875 | case OF_RECONFIG_DETACH_NODE: |
f5242e5a | 876 | pseries_remove_processor(rd->dn); |
413f7c40 | 877 | break; |
413f7c40 | 878 | } |
de2780a3 | 879 | return notifier_from_errno(err); |
413f7c40 ME |
880 | } |
881 | ||
06ba30b6 ME |
882 | static struct notifier_block pseries_smp_nb = { |
883 | .notifier_call = pseries_smp_notifier, | |
413f7c40 ME |
884 | }; |
885 | ||
3aa565f5 GS |
886 | #define MAX_CEDE_LATENCY_LEVELS 4 |
887 | #define CEDE_LATENCY_PARAM_LENGTH 10 | |
888 | #define CEDE_LATENCY_PARAM_MAX_LENGTH \ | |
889 | (MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char)) | |
890 | #define CEDE_LATENCY_TOKEN 45 | |
891 | ||
892 | static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH]; | |
893 | ||
894 | static int parse_cede_parameters(void) | |
895 | { | |
3aa565f5 | 896 | memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH); |
20a8ab97 AB |
897 | return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1, |
898 | NULL, | |
899 | CEDE_LATENCY_TOKEN, | |
900 | __pa(cede_parameters), | |
901 | CEDE_LATENCY_PARAM_MAX_LENGTH); | |
3aa565f5 GS |
902 | } |
903 | ||
0332c2d4 ME |
904 | static int __init pseries_cpu_hotplug_init(void) |
905 | { | |
3aa565f5 | 906 | int cpu; |
f8b67691 | 907 | int qcss_tok; |
64f27585 | 908 | |
183deeea NF |
909 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE |
910 | ppc_md.cpu_probe = dlpar_cpu_probe; | |
911 | ppc_md.cpu_release = dlpar_cpu_release; | |
912 | #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ | |
913 | ||
41dd03a9 | 914 | rtas_stop_self_token = rtas_token("stop-self"); |
674fa677 | 915 | qcss_tok = rtas_token("query-cpu-stopped-state"); |
0332c2d4 | 916 | |
41dd03a9 | 917 | if (rtas_stop_self_token == RTAS_UNKNOWN_SERVICE || |
674fa677 ME |
918 | qcss_tok == RTAS_UNKNOWN_SERVICE) { |
919 | printk(KERN_INFO "CPU Hotplug not supported by firmware " | |
920 | "- disabling.\n"); | |
921 | return 0; | |
922 | } | |
04da6af9 | 923 | |
06ba30b6 ME |
924 | ppc_md.cpu_die = pseries_mach_cpu_die; |
925 | smp_ops->cpu_disable = pseries_cpu_disable; | |
926 | smp_ops->cpu_die = pseries_cpu_die; | |
413f7c40 ME |
927 | |
928 | /* Processors can be added/removed only on LPAR */ | |
3aa565f5 | 929 | if (firmware_has_feature(FW_FEATURE_LPAR)) { |
1cf3d8b3 | 930 | of_reconfig_notifier_register(&pseries_smp_nb); |
3aa565f5 GS |
931 | cpu_maps_update_begin(); |
932 | if (cede_offline_enabled && parse_cede_parameters() == 0) { | |
933 | default_offline_state = CPU_STATE_INACTIVE; | |
934 | for_each_online_cpu(cpu) | |
935 | set_default_offline_state(cpu); | |
936 | } | |
937 | cpu_maps_update_done(); | |
938 | } | |
413f7c40 | 939 | |
0332c2d4 ME |
940 | return 0; |
941 | } | |
d2a36071 | 942 | machine_arch_initcall(pseries, pseries_cpu_hotplug_init); |