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 */ |
ef8bd77f | 27 | #include <linux/sched/hotplug.h> |
0332c2d4 | 28 | #include <linux/cpu.h> |
1cf3d8b3 | 29 | #include <linux/of.h> |
ac713800 | 30 | #include <linux/slab.h> |
0332c2d4 ME |
31 | #include <asm/prom.h> |
32 | #include <asm/rtas.h> | |
33 | #include <asm/firmware.h> | |
34 | #include <asm/machdep.h> | |
35 | #include <asm/vdso_datapage.h> | |
0b05ac6e | 36 | #include <asm/xics.h> |
eac1e731 | 37 | #include <asm/xive.h> |
212bebb4 DD |
38 | #include <asm/plpar_wrappers.h> |
39 | ||
183deeea | 40 | #include "pseries.h" |
3aa565f5 | 41 | #include "offline_states.h" |
0332c2d4 ME |
42 | |
43 | /* This version can't take the spinlock, because it never returns */ | |
41dd03a9 | 44 | static int rtas_stop_self_token = RTAS_UNKNOWN_SERVICE; |
0332c2d4 | 45 | |
3aa565f5 GS |
46 | static DEFINE_PER_CPU(enum cpu_state_vals, preferred_offline_state) = |
47 | CPU_STATE_OFFLINE; | |
48 | static DEFINE_PER_CPU(enum cpu_state_vals, current_state) = CPU_STATE_OFFLINE; | |
49 | ||
50 | static enum cpu_state_vals default_offline_state = CPU_STATE_OFFLINE; | |
51 | ||
4cc7ecb7 | 52 | static bool cede_offline_enabled __read_mostly = true; |
3aa565f5 GS |
53 | |
54 | /* | |
55 | * Enable/disable cede_offline when available. | |
56 | */ | |
57 | static int __init setup_cede_offline(char *str) | |
58 | { | |
4cc7ecb7 | 59 | return (kstrtobool(str, &cede_offline_enabled) == 0); |
3aa565f5 GS |
60 | } |
61 | ||
62 | __setup("cede_offline=", setup_cede_offline); | |
63 | ||
64 | enum cpu_state_vals get_cpu_current_state(int cpu) | |
65 | { | |
66 | return per_cpu(current_state, cpu); | |
67 | } | |
68 | ||
69 | void set_cpu_current_state(int cpu, enum cpu_state_vals state) | |
70 | { | |
71 | per_cpu(current_state, cpu) = state; | |
72 | } | |
73 | ||
74 | enum cpu_state_vals get_preferred_offline_state(int cpu) | |
75 | { | |
76 | return per_cpu(preferred_offline_state, cpu); | |
77 | } | |
78 | ||
79 | void set_preferred_offline_state(int cpu, enum cpu_state_vals state) | |
80 | { | |
81 | per_cpu(preferred_offline_state, cpu) = state; | |
82 | } | |
83 | ||
84 | void set_default_offline_state(int cpu) | |
85 | { | |
86 | per_cpu(preferred_offline_state, cpu) = default_offline_state; | |
87 | } | |
88 | ||
04da6af9 | 89 | static void rtas_stop_self(void) |
0332c2d4 | 90 | { |
b2e8590f | 91 | static struct rtas_args args; |
4fb8d027 | 92 | |
0332c2d4 ME |
93 | local_irq_disable(); |
94 | ||
41dd03a9 | 95 | BUG_ON(rtas_stop_self_token == RTAS_UNKNOWN_SERVICE); |
0332c2d4 ME |
96 | |
97 | printk("cpu %u (hwid %u) Ready to die...\n", | |
98 | smp_processor_id(), hard_smp_processor_id()); | |
b2e8590f ME |
99 | |
100 | rtas_call_unlocked(&args, rtas_stop_self_token, 0, 1, NULL); | |
0332c2d4 ME |
101 | |
102 | panic("Alas, I survived.\n"); | |
103 | } | |
104 | ||
06ba30b6 | 105 | static void pseries_mach_cpu_die(void) |
04da6af9 | 106 | { |
3aa565f5 GS |
107 | unsigned int cpu = smp_processor_id(); |
108 | unsigned int hwcpu = hard_smp_processor_id(); | |
109 | u8 cede_latency_hint = 0; | |
110 | ||
04da6af9 ME |
111 | local_irq_disable(); |
112 | idle_task_exit(); | |
eac1e731 CLG |
113 | if (xive_enabled()) |
114 | xive_teardown_cpu(); | |
115 | else | |
116 | xics_teardown_cpu(); | |
3aa565f5 GS |
117 | |
118 | if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) { | |
119 | set_cpu_current_state(cpu, CPU_STATE_INACTIVE); | |
32d8ad4e BK |
120 | if (ppc_md.suspend_disable_cpu) |
121 | ppc_md.suspend_disable_cpu(); | |
122 | ||
3aa565f5 GS |
123 | cede_latency_hint = 2; |
124 | ||
125 | get_lppaca()->idle = 1; | |
f13c13a0 | 126 | if (!lppaca_shared_proc(get_lppaca())) |
3aa565f5 GS |
127 | get_lppaca()->donate_dedicated_cpu = 1; |
128 | ||
3aa565f5 | 129 | while (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) { |
fb912568 LZ |
130 | while (!prep_irq_for_idle()) { |
131 | local_irq_enable(); | |
132 | local_irq_disable(); | |
133 | } | |
134 | ||
3aa565f5 | 135 | extended_cede_processor(cede_latency_hint); |
3aa565f5 GS |
136 | } |
137 | ||
fb912568 LZ |
138 | local_irq_disable(); |
139 | ||
f13c13a0 | 140 | if (!lppaca_shared_proc(get_lppaca())) |
3aa565f5 GS |
141 | get_lppaca()->donate_dedicated_cpu = 0; |
142 | get_lppaca()->idle = 0; | |
3aa565f5 | 143 | |
0212f260 | 144 | if (get_preferred_offline_state(cpu) == CPU_STATE_ONLINE) { |
598c8231 | 145 | unregister_slb_shadow(hwcpu); |
3aa565f5 | 146 | |
fb912568 | 147 | hard_irq_disable(); |
0212f260 VS |
148 | /* |
149 | * Call to start_secondary_resume() will not return. | |
150 | * Kernel stack will be reset and start_secondary() | |
151 | * will be called to continue the online operation. | |
152 | */ | |
153 | start_secondary_resume(); | |
154 | } | |
155 | } | |
3aa565f5 | 156 | |
0212f260 VS |
157 | /* Requested state is CPU_STATE_OFFLINE at this point */ |
158 | WARN_ON(get_preferred_offline_state(cpu) != CPU_STATE_OFFLINE); | |
3aa565f5 | 159 | |
0212f260 | 160 | set_cpu_current_state(cpu, CPU_STATE_OFFLINE); |
598c8231 | 161 | unregister_slb_shadow(hwcpu); |
0212f260 | 162 | rtas_stop_self(); |
3aa565f5 | 163 | |
04da6af9 ME |
164 | /* Should never get here... */ |
165 | BUG(); | |
166 | for(;;); | |
167 | } | |
168 | ||
06ba30b6 | 169 | static int pseries_cpu_disable(void) |
413f7c40 ME |
170 | { |
171 | int cpu = smp_processor_id(); | |
172 | ||
ea0f1cab | 173 | set_cpu_online(cpu, false); |
413f7c40 ME |
174 | vdso_data->processorCount--; |
175 | ||
176 | /*fix boot_cpuid here*/ | |
177 | if (cpu == boot_cpuid) | |
8729faaa | 178 | boot_cpuid = cpumask_any(cpu_online_mask); |
413f7c40 ME |
179 | |
180 | /* FIXME: abstract this to not be platform specific later on */ | |
eac1e731 CLG |
181 | if (xive_enabled()) |
182 | xive_smp_disable_cpu(); | |
183 | else | |
184 | xics_migrate_irqs_away(); | |
413f7c40 ME |
185 | return 0; |
186 | } | |
187 | ||
3aa565f5 GS |
188 | /* |
189 | * pseries_cpu_die: Wait for the cpu to die. | |
190 | * @cpu: logical processor id of the CPU whose death we're awaiting. | |
191 | * | |
192 | * This function is called from the context of the thread which is performing | |
193 | * the cpu-offline. Here we wait for long enough to allow the cpu in question | |
194 | * to self-destroy so that the cpu-offline thread can send the CPU_DEAD | |
195 | * notifications. | |
196 | * | |
197 | * OTOH, pseries_mach_cpu_die() is called by the @cpu when it wants to | |
198 | * self-destruct. | |
199 | */ | |
06ba30b6 | 200 | static void pseries_cpu_die(unsigned int cpu) |
413f7c40 ME |
201 | { |
202 | int tries; | |
3aa565f5 | 203 | int cpu_status = 1; |
413f7c40 ME |
204 | unsigned int pcpu = get_hard_smp_processor_id(cpu); |
205 | ||
3aa565f5 GS |
206 | if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) { |
207 | cpu_status = 1; | |
940ce422 | 208 | for (tries = 0; tries < 5000; tries++) { |
3aa565f5 GS |
209 | if (get_cpu_current_state(cpu) == CPU_STATE_INACTIVE) { |
210 | cpu_status = 0; | |
211 | break; | |
212 | } | |
940ce422 | 213 | msleep(1); |
3aa565f5 GS |
214 | } |
215 | } else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) { | |
216 | ||
217 | for (tries = 0; tries < 25; tries++) { | |
f8b67691 MN |
218 | cpu_status = smp_query_cpu_stopped(pcpu); |
219 | if (cpu_status == QCSS_STOPPED || | |
220 | cpu_status == QCSS_HARDWARE_ERROR) | |
3aa565f5 GS |
221 | break; |
222 | cpu_relax(); | |
223 | } | |
413f7c40 | 224 | } |
3aa565f5 | 225 | |
413f7c40 ME |
226 | if (cpu_status != 0) { |
227 | printk("Querying DEAD? cpu %i (%i) shows %i\n", | |
228 | cpu, pcpu, cpu_status); | |
229 | } | |
230 | ||
25985edc | 231 | /* Isolation and deallocation are definitely done by |
413f7c40 ME |
232 | * drslot_chrp_cpu. If they were not they would be |
233 | * done here. Change isolate state to Isolate and | |
234 | * change allocation-state to Unusable. | |
235 | */ | |
236 | paca[cpu].cpu_start = 0; | |
237 | } | |
238 | ||
239 | /* | |
828a6986 | 240 | * Update cpu_present_mask and paca(s) for a new cpu node. The wrinkle |
413f7c40 ME |
241 | * here is that a cpu device node may represent up to two logical cpus |
242 | * in the SMT case. We must honor the assumption in other code that | |
243 | * the logical ids for sibling SMT threads x and y are adjacent, such | |
244 | * that x^1 == y and y^1 == x. | |
245 | */ | |
06ba30b6 | 246 | static int pseries_add_processor(struct device_node *np) |
413f7c40 ME |
247 | { |
248 | unsigned int cpu; | |
8729faaa | 249 | cpumask_var_t candidate_mask, tmp; |
413f7c40 | 250 | int err = -ENOSPC, len, nthreads, i; |
d6f1e7ab | 251 | const __be32 *intserv; |
413f7c40 | 252 | |
e2eb6392 | 253 | intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len); |
413f7c40 ME |
254 | if (!intserv) |
255 | return 0; | |
256 | ||
8729faaa AB |
257 | zalloc_cpumask_var(&candidate_mask, GFP_KERNEL); |
258 | zalloc_cpumask_var(&tmp, GFP_KERNEL); | |
259 | ||
413f7c40 ME |
260 | nthreads = len / sizeof(u32); |
261 | for (i = 0; i < nthreads; i++) | |
8729faaa | 262 | cpumask_set_cpu(i, tmp); |
413f7c40 | 263 | |
86ef5c9a | 264 | cpu_maps_update_begin(); |
413f7c40 | 265 | |
8729faaa | 266 | BUG_ON(!cpumask_subset(cpu_present_mask, cpu_possible_mask)); |
413f7c40 ME |
267 | |
268 | /* Get a bitmap of unoccupied slots. */ | |
8729faaa AB |
269 | cpumask_xor(candidate_mask, cpu_possible_mask, cpu_present_mask); |
270 | if (cpumask_empty(candidate_mask)) { | |
413f7c40 ME |
271 | /* If we get here, it most likely means that NR_CPUS is |
272 | * less than the partition's max processors setting. | |
273 | */ | |
b7c670d6 RH |
274 | printk(KERN_ERR "Cannot add cpu %pOF; this system configuration" |
275 | " supports %d logical cpus.\n", np, | |
53a448c3 | 276 | num_possible_cpus()); |
413f7c40 ME |
277 | goto out_unlock; |
278 | } | |
279 | ||
8729faaa AB |
280 | while (!cpumask_empty(tmp)) |
281 | if (cpumask_subset(tmp, candidate_mask)) | |
413f7c40 ME |
282 | /* Found a range where we can insert the new cpu(s) */ |
283 | break; | |
284 | else | |
8729faaa | 285 | cpumask_shift_left(tmp, tmp, nthreads); |
413f7c40 | 286 | |
8729faaa | 287 | if (cpumask_empty(tmp)) { |
828a6986 | 288 | printk(KERN_ERR "Unable to find space in cpu_present_mask for" |
413f7c40 ME |
289 | " processor %s with %d thread(s)\n", np->name, |
290 | nthreads); | |
291 | goto out_unlock; | |
292 | } | |
293 | ||
8729faaa | 294 | for_each_cpu(cpu, tmp) { |
104699c0 | 295 | BUG_ON(cpu_present(cpu)); |
ea0f1cab | 296 | set_cpu_present(cpu, true); |
d6f1e7ab | 297 | set_hard_smp_processor_id(cpu, be32_to_cpu(*intserv++)); |
413f7c40 ME |
298 | } |
299 | err = 0; | |
300 | out_unlock: | |
86ef5c9a | 301 | cpu_maps_update_done(); |
8729faaa AB |
302 | free_cpumask_var(candidate_mask); |
303 | free_cpumask_var(tmp); | |
413f7c40 ME |
304 | return err; |
305 | } | |
306 | ||
307 | /* | |
308 | * Update the present map for a cpu node which is going away, and set | |
309 | * the hard id in the paca(s) to -1 to be consistent with boot time | |
310 | * convention for non-present cpus. | |
311 | */ | |
06ba30b6 | 312 | static void pseries_remove_processor(struct device_node *np) |
413f7c40 ME |
313 | { |
314 | unsigned int cpu; | |
315 | int len, nthreads, i; | |
e36d1227 TF |
316 | const __be32 *intserv; |
317 | u32 thread; | |
413f7c40 | 318 | |
e2eb6392 | 319 | intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len); |
413f7c40 ME |
320 | if (!intserv) |
321 | return; | |
322 | ||
323 | nthreads = len / sizeof(u32); | |
324 | ||
86ef5c9a | 325 | cpu_maps_update_begin(); |
413f7c40 | 326 | for (i = 0; i < nthreads; i++) { |
e36d1227 | 327 | thread = be32_to_cpu(intserv[i]); |
413f7c40 | 328 | for_each_present_cpu(cpu) { |
e36d1227 | 329 | if (get_hard_smp_processor_id(cpu) != thread) |
413f7c40 ME |
330 | continue; |
331 | BUG_ON(cpu_online(cpu)); | |
ea0f1cab | 332 | set_cpu_present(cpu, false); |
413f7c40 ME |
333 | set_hard_smp_processor_id(cpu, -1); |
334 | break; | |
335 | } | |
8729faaa | 336 | if (cpu >= nr_cpu_ids) |
413f7c40 | 337 | printk(KERN_WARNING "Could not find cpu to remove " |
e36d1227 | 338 | "with physical id 0x%x\n", thread); |
413f7c40 | 339 | } |
86ef5c9a | 340 | cpu_maps_update_done(); |
413f7c40 ME |
341 | } |
342 | ||
183deeea NF |
343 | static int dlpar_online_cpu(struct device_node *dn) |
344 | { | |
345 | int rc = 0; | |
346 | unsigned int cpu; | |
347 | int len, nthreads, i; | |
348 | const __be32 *intserv; | |
349 | u32 thread; | |
350 | ||
351 | intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len); | |
352 | if (!intserv) | |
353 | return -EINVAL; | |
354 | ||
355 | nthreads = len / sizeof(u32); | |
356 | ||
357 | cpu_maps_update_begin(); | |
358 | for (i = 0; i < nthreads; i++) { | |
359 | thread = be32_to_cpu(intserv[i]); | |
360 | for_each_present_cpu(cpu) { | |
361 | if (get_hard_smp_processor_id(cpu) != thread) | |
362 | continue; | |
363 | BUG_ON(get_cpu_current_state(cpu) | |
364 | != CPU_STATE_OFFLINE); | |
365 | cpu_maps_update_done(); | |
cee5405d | 366 | timed_topology_update(1); |
183deeea NF |
367 | rc = device_online(get_cpu_device(cpu)); |
368 | if (rc) | |
369 | goto out; | |
370 | cpu_maps_update_begin(); | |
371 | ||
372 | break; | |
373 | } | |
374 | if (cpu == num_possible_cpus()) | |
375 | printk(KERN_WARNING "Could not find cpu to online " | |
376 | "with physical id 0x%x\n", thread); | |
377 | } | |
378 | cpu_maps_update_done(); | |
379 | ||
380 | out: | |
381 | return rc; | |
382 | ||
383 | } | |
384 | ||
385 | static bool dlpar_cpu_exists(struct device_node *parent, u32 drc_index) | |
386 | { | |
387 | struct device_node *child = NULL; | |
388 | u32 my_drc_index; | |
389 | bool found; | |
390 | int rc; | |
391 | ||
392 | /* Assume cpu doesn't exist */ | |
393 | found = false; | |
394 | ||
395 | for_each_child_of_node(parent, child) { | |
396 | rc = of_property_read_u32(child, "ibm,my-drc-index", | |
397 | &my_drc_index); | |
398 | if (rc) | |
399 | continue; | |
400 | ||
401 | if (my_drc_index == drc_index) { | |
402 | of_node_put(child); | |
403 | found = true; | |
404 | break; | |
405 | } | |
406 | } | |
407 | ||
408 | return found; | |
409 | } | |
410 | ||
90edf184 NF |
411 | static bool valid_cpu_drc_index(struct device_node *parent, u32 drc_index) |
412 | { | |
413 | bool found = false; | |
414 | int rc, index; | |
415 | ||
416 | index = 0; | |
417 | while (!found) { | |
418 | u32 drc; | |
419 | ||
420 | rc = of_property_read_u32_index(parent, "ibm,drc-indexes", | |
421 | index++, &drc); | |
422 | if (rc) | |
423 | break; | |
424 | ||
425 | if (drc == drc_index) | |
426 | found = true; | |
427 | } | |
428 | ||
429 | return found; | |
430 | } | |
431 | ||
d98389f3 | 432 | static ssize_t dlpar_cpu_add(u32 drc_index) |
183deeea NF |
433 | { |
434 | struct device_node *dn, *parent; | |
e666ae0b NF |
435 | int rc, saved_rc; |
436 | ||
437 | pr_debug("Attempting to add CPU, drc index: %x\n", drc_index); | |
183deeea | 438 | |
183deeea | 439 | parent = of_find_node_by_path("/cpus"); |
e666ae0b NF |
440 | if (!parent) { |
441 | pr_warn("Failed to find CPU root node \"/cpus\"\n"); | |
183deeea | 442 | return -ENODEV; |
e666ae0b | 443 | } |
183deeea NF |
444 | |
445 | if (dlpar_cpu_exists(parent, drc_index)) { | |
446 | of_node_put(parent); | |
e666ae0b | 447 | pr_warn("CPU with drc index %x already exists\n", drc_index); |
183deeea NF |
448 | return -EINVAL; |
449 | } | |
450 | ||
90edf184 NF |
451 | if (!valid_cpu_drc_index(parent, drc_index)) { |
452 | of_node_put(parent); | |
453 | pr_warn("Cannot find CPU (drc index %x) to add.\n", drc_index); | |
454 | return -EINVAL; | |
455 | } | |
456 | ||
183deeea NF |
457 | rc = dlpar_acquire_drc(drc_index); |
458 | if (rc) { | |
e666ae0b NF |
459 | pr_warn("Failed to acquire DRC, rc: %d, drc index: %x\n", |
460 | rc, drc_index); | |
183deeea NF |
461 | of_node_put(parent); |
462 | return -EINVAL; | |
463 | } | |
464 | ||
465 | dn = dlpar_configure_connector(cpu_to_be32(drc_index), parent); | |
e666ae0b NF |
466 | if (!dn) { |
467 | pr_warn("Failed call to configure-connector, drc index: %x\n", | |
468 | drc_index); | |
469 | dlpar_release_drc(drc_index); | |
087ff6a5 | 470 | of_node_put(parent); |
183deeea | 471 | return -EINVAL; |
e666ae0b | 472 | } |
183deeea | 473 | |
215ee763 | 474 | rc = dlpar_attach_node(dn, parent); |
087ff6a5 TD |
475 | |
476 | /* Regardless we are done with parent now */ | |
477 | of_node_put(parent); | |
478 | ||
183deeea | 479 | if (rc) { |
e666ae0b NF |
480 | saved_rc = rc; |
481 | pr_warn("Failed to attach node %s, rc: %d, drc index: %x\n", | |
482 | dn->name, rc, drc_index); | |
483 | ||
484 | rc = dlpar_release_drc(drc_index); | |
485 | if (!rc) | |
486 | dlpar_free_cc_nodes(dn); | |
487 | ||
488 | return saved_rc; | |
183deeea NF |
489 | } |
490 | ||
491 | rc = dlpar_online_cpu(dn); | |
e666ae0b NF |
492 | if (rc) { |
493 | saved_rc = rc; | |
494 | pr_warn("Failed to online cpu %s, rc: %d, drc index: %x\n", | |
495 | dn->name, rc, drc_index); | |
496 | ||
497 | rc = dlpar_detach_node(dn); | |
498 | if (!rc) | |
499 | dlpar_release_drc(drc_index); | |
500 | ||
501 | return saved_rc; | |
502 | } | |
503 | ||
504 | pr_debug("Successfully added CPU %s, drc index: %x\n", dn->name, | |
505 | drc_index); | |
d98389f3 | 506 | return rc; |
183deeea NF |
507 | } |
508 | ||
509 | static int dlpar_offline_cpu(struct device_node *dn) | |
510 | { | |
511 | int rc = 0; | |
512 | unsigned int cpu; | |
513 | int len, nthreads, i; | |
514 | const __be32 *intserv; | |
515 | u32 thread; | |
516 | ||
517 | intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len); | |
518 | if (!intserv) | |
519 | return -EINVAL; | |
520 | ||
521 | nthreads = len / sizeof(u32); | |
522 | ||
523 | cpu_maps_update_begin(); | |
524 | for (i = 0; i < nthreads; i++) { | |
525 | thread = be32_to_cpu(intserv[i]); | |
526 | for_each_present_cpu(cpu) { | |
527 | if (get_hard_smp_processor_id(cpu) != thread) | |
528 | continue; | |
529 | ||
530 | if (get_cpu_current_state(cpu) == CPU_STATE_OFFLINE) | |
531 | break; | |
532 | ||
533 | if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) { | |
534 | set_preferred_offline_state(cpu, | |
535 | CPU_STATE_OFFLINE); | |
536 | cpu_maps_update_done(); | |
cee5405d | 537 | timed_topology_update(1); |
183deeea NF |
538 | rc = device_offline(get_cpu_device(cpu)); |
539 | if (rc) | |
540 | goto out; | |
541 | cpu_maps_update_begin(); | |
542 | break; | |
543 | ||
544 | } | |
545 | ||
546 | /* | |
547 | * The cpu is in CPU_STATE_INACTIVE. | |
548 | * Upgrade it's state to CPU_STATE_OFFLINE. | |
549 | */ | |
550 | set_preferred_offline_state(cpu, CPU_STATE_OFFLINE); | |
551 | BUG_ON(plpar_hcall_norets(H_PROD, thread) | |
552 | != H_SUCCESS); | |
553 | __cpu_die(cpu); | |
554 | break; | |
555 | } | |
556 | if (cpu == num_possible_cpus()) | |
557 | printk(KERN_WARNING "Could not find cpu to offline with physical id 0x%x\n", thread); | |
558 | } | |
559 | cpu_maps_update_done(); | |
560 | ||
561 | out: | |
562 | return rc; | |
563 | ||
564 | } | |
565 | ||
d98389f3 NF |
566 | static ssize_t dlpar_cpu_remove(struct device_node *dn, u32 drc_index) |
567 | { | |
568 | int rc; | |
569 | ||
b802ab46 | 570 | pr_debug("Attempting to remove CPU %s, drc index: %x\n", |
e666ae0b NF |
571 | dn->name, drc_index); |
572 | ||
d98389f3 | 573 | rc = dlpar_offline_cpu(dn); |
e666ae0b NF |
574 | if (rc) { |
575 | pr_warn("Failed to offline CPU %s, rc: %d\n", dn->name, rc); | |
d98389f3 | 576 | return -EINVAL; |
e666ae0b | 577 | } |
d98389f3 NF |
578 | |
579 | rc = dlpar_release_drc(drc_index); | |
e666ae0b NF |
580 | if (rc) { |
581 | pr_warn("Failed to release drc (%x) for CPU %s, rc: %d\n", | |
582 | drc_index, dn->name, rc); | |
583 | dlpar_online_cpu(dn); | |
d98389f3 | 584 | return rc; |
e666ae0b | 585 | } |
d98389f3 NF |
586 | |
587 | rc = dlpar_detach_node(dn); | |
e666ae0b NF |
588 | if (rc) { |
589 | int saved_rc = rc; | |
d98389f3 | 590 | |
e666ae0b NF |
591 | pr_warn("Failed to detach CPU %s, rc: %d", dn->name, rc); |
592 | ||
593 | rc = dlpar_acquire_drc(drc_index); | |
594 | if (!rc) | |
595 | dlpar_online_cpu(dn); | |
596 | ||
597 | return saved_rc; | |
598 | } | |
599 | ||
600 | pr_debug("Successfully removed CPU, drc index: %x\n", drc_index); | |
601 | return 0; | |
d98389f3 NF |
602 | } |
603 | ||
ac713800 NF |
604 | static struct device_node *cpu_drc_index_to_dn(u32 drc_index) |
605 | { | |
606 | struct device_node *dn; | |
607 | u32 my_index; | |
608 | int rc; | |
609 | ||
610 | for_each_node_by_type(dn, "cpu") { | |
611 | rc = of_property_read_u32(dn, "ibm,my-drc-index", &my_index); | |
612 | if (rc) | |
613 | continue; | |
614 | ||
615 | if (my_index == drc_index) | |
616 | break; | |
617 | } | |
618 | ||
619 | return dn; | |
620 | } | |
621 | ||
622 | static int dlpar_cpu_remove_by_index(u32 drc_index) | |
623 | { | |
624 | struct device_node *dn; | |
625 | int rc; | |
626 | ||
627 | dn = cpu_drc_index_to_dn(drc_index); | |
628 | if (!dn) { | |
629 | pr_warn("Cannot find CPU (drc index %x) to remove\n", | |
630 | drc_index); | |
631 | return -ENODEV; | |
632 | } | |
633 | ||
634 | rc = dlpar_cpu_remove(dn, drc_index); | |
635 | of_node_put(dn); | |
636 | return rc; | |
637 | } | |
638 | ||
639 | static int find_dlpar_cpus_to_remove(u32 *cpu_drcs, int cpus_to_remove) | |
640 | { | |
641 | struct device_node *dn; | |
642 | int cpus_found = 0; | |
643 | int rc; | |
644 | ||
645 | /* We want to find cpus_to_remove + 1 CPUs to ensure we do not | |
646 | * remove the last CPU. | |
647 | */ | |
648 | for_each_node_by_type(dn, "cpu") { | |
649 | cpus_found++; | |
650 | ||
651 | if (cpus_found > cpus_to_remove) { | |
652 | of_node_put(dn); | |
653 | break; | |
654 | } | |
655 | ||
656 | /* Note that cpus_found is always 1 ahead of the index | |
657 | * into the cpu_drcs array, so we use cpus_found - 1 | |
658 | */ | |
659 | rc = of_property_read_u32(dn, "ibm,my-drc-index", | |
660 | &cpu_drcs[cpus_found - 1]); | |
661 | if (rc) { | |
662 | pr_warn("Error occurred getting drc-index for %s\n", | |
663 | dn->name); | |
664 | of_node_put(dn); | |
665 | return -1; | |
666 | } | |
667 | } | |
668 | ||
669 | if (cpus_found < cpus_to_remove) { | |
670 | pr_warn("Failed to find enough CPUs (%d of %d) to remove\n", | |
671 | cpus_found, cpus_to_remove); | |
672 | } else if (cpus_found == cpus_to_remove) { | |
673 | pr_warn("Cannot remove all CPUs\n"); | |
674 | } | |
675 | ||
676 | return cpus_found; | |
677 | } | |
678 | ||
679 | static int dlpar_cpu_remove_by_count(u32 cpus_to_remove) | |
680 | { | |
681 | u32 *cpu_drcs; | |
682 | int cpus_found; | |
683 | int cpus_removed = 0; | |
684 | int i, rc; | |
685 | ||
686 | pr_debug("Attempting to hot-remove %d CPUs\n", cpus_to_remove); | |
687 | ||
688 | cpu_drcs = kcalloc(cpus_to_remove, sizeof(*cpu_drcs), GFP_KERNEL); | |
689 | if (!cpu_drcs) | |
690 | return -EINVAL; | |
691 | ||
692 | cpus_found = find_dlpar_cpus_to_remove(cpu_drcs, cpus_to_remove); | |
693 | if (cpus_found <= cpus_to_remove) { | |
694 | kfree(cpu_drcs); | |
695 | return -EINVAL; | |
696 | } | |
697 | ||
698 | for (i = 0; i < cpus_to_remove; i++) { | |
699 | rc = dlpar_cpu_remove_by_index(cpu_drcs[i]); | |
700 | if (rc) | |
701 | break; | |
702 | ||
703 | cpus_removed++; | |
704 | } | |
705 | ||
706 | if (cpus_removed != cpus_to_remove) { | |
707 | pr_warn("CPU hot-remove failed, adding back removed CPUs\n"); | |
708 | ||
709 | for (i = 0; i < cpus_removed; i++) | |
710 | dlpar_cpu_add(cpu_drcs[i]); | |
711 | ||
712 | rc = -EINVAL; | |
713 | } else { | |
714 | rc = 0; | |
715 | } | |
716 | ||
717 | kfree(cpu_drcs); | |
718 | return rc; | |
719 | } | |
720 | ||
90edf184 NF |
721 | static int find_dlpar_cpus_to_add(u32 *cpu_drcs, u32 cpus_to_add) |
722 | { | |
723 | struct device_node *parent; | |
724 | int cpus_found = 0; | |
725 | int index, rc; | |
726 | ||
727 | parent = of_find_node_by_path("/cpus"); | |
728 | if (!parent) { | |
729 | pr_warn("Could not find CPU root node in device tree\n"); | |
730 | kfree(cpu_drcs); | |
731 | return -1; | |
732 | } | |
733 | ||
734 | /* Search the ibm,drc-indexes array for possible CPU drcs to | |
735 | * add. Note that the format of the ibm,drc-indexes array is | |
736 | * the number of entries in the array followed by the array | |
737 | * of drc values so we start looking at index = 1. | |
738 | */ | |
739 | index = 1; | |
740 | while (cpus_found < cpus_to_add) { | |
741 | u32 drc; | |
742 | ||
743 | rc = of_property_read_u32_index(parent, "ibm,drc-indexes", | |
744 | index++, &drc); | |
745 | if (rc) | |
746 | break; | |
747 | ||
748 | if (dlpar_cpu_exists(parent, drc)) | |
749 | continue; | |
750 | ||
751 | cpu_drcs[cpus_found++] = drc; | |
752 | } | |
753 | ||
754 | of_node_put(parent); | |
755 | return cpus_found; | |
756 | } | |
757 | ||
758 | static int dlpar_cpu_add_by_count(u32 cpus_to_add) | |
759 | { | |
760 | u32 *cpu_drcs; | |
761 | int cpus_added = 0; | |
762 | int cpus_found; | |
763 | int i, rc; | |
764 | ||
765 | pr_debug("Attempting to hot-add %d CPUs\n", cpus_to_add); | |
766 | ||
767 | cpu_drcs = kcalloc(cpus_to_add, sizeof(*cpu_drcs), GFP_KERNEL); | |
768 | if (!cpu_drcs) | |
769 | return -EINVAL; | |
770 | ||
771 | cpus_found = find_dlpar_cpus_to_add(cpu_drcs, cpus_to_add); | |
772 | if (cpus_found < cpus_to_add) { | |
773 | pr_warn("Failed to find enough CPUs (%d of %d) to add\n", | |
774 | cpus_found, cpus_to_add); | |
775 | kfree(cpu_drcs); | |
776 | return -EINVAL; | |
777 | } | |
778 | ||
779 | for (i = 0; i < cpus_to_add; i++) { | |
780 | rc = dlpar_cpu_add(cpu_drcs[i]); | |
781 | if (rc) | |
782 | break; | |
783 | ||
784 | cpus_added++; | |
785 | } | |
786 | ||
787 | if (cpus_added < cpus_to_add) { | |
788 | pr_warn("CPU hot-add failed, removing any added CPUs\n"); | |
789 | ||
790 | for (i = 0; i < cpus_added; i++) | |
791 | dlpar_cpu_remove_by_index(cpu_drcs[i]); | |
792 | ||
793 | rc = -EINVAL; | |
794 | } else { | |
795 | rc = 0; | |
796 | } | |
797 | ||
798 | kfree(cpu_drcs); | |
799 | return rc; | |
800 | } | |
801 | ||
ac713800 NF |
802 | int dlpar_cpu(struct pseries_hp_errorlog *hp_elog) |
803 | { | |
804 | u32 count, drc_index; | |
805 | int rc; | |
806 | ||
807 | count = hp_elog->_drc_u.drc_count; | |
808 | drc_index = hp_elog->_drc_u.drc_index; | |
809 | ||
810 | lock_device_hotplug(); | |
811 | ||
812 | switch (hp_elog->action) { | |
813 | case PSERIES_HP_ELOG_ACTION_REMOVE: | |
814 | if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT) | |
815 | rc = dlpar_cpu_remove_by_count(count); | |
816 | else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX) | |
817 | rc = dlpar_cpu_remove_by_index(drc_index); | |
818 | else | |
819 | rc = -EINVAL; | |
820 | break; | |
90edf184 NF |
821 | case PSERIES_HP_ELOG_ACTION_ADD: |
822 | if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT) | |
823 | rc = dlpar_cpu_add_by_count(count); | |
824 | else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX) | |
825 | rc = dlpar_cpu_add(drc_index); | |
826 | else | |
827 | rc = -EINVAL; | |
828 | break; | |
ac713800 NF |
829 | default: |
830 | pr_err("Invalid action (%d) specified\n", hp_elog->action); | |
831 | rc = -EINVAL; | |
832 | break; | |
833 | } | |
834 | ||
835 | unlock_device_hotplug(); | |
836 | return rc; | |
837 | } | |
838 | ||
d98389f3 NF |
839 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE |
840 | ||
841 | static ssize_t dlpar_cpu_probe(const char *buf, size_t count) | |
842 | { | |
843 | u32 drc_index; | |
844 | int rc; | |
845 | ||
846 | rc = kstrtou32(buf, 0, &drc_index); | |
847 | if (rc) | |
848 | return -EINVAL; | |
849 | ||
850 | rc = dlpar_cpu_add(drc_index); | |
851 | ||
852 | return rc ? rc : count; | |
853 | } | |
854 | ||
183deeea NF |
855 | static ssize_t dlpar_cpu_release(const char *buf, size_t count) |
856 | { | |
857 | struct device_node *dn; | |
858 | u32 drc_index; | |
859 | int rc; | |
860 | ||
861 | dn = of_find_node_by_path(buf); | |
862 | if (!dn) | |
863 | return -EINVAL; | |
864 | ||
865 | rc = of_property_read_u32(dn, "ibm,my-drc-index", &drc_index); | |
866 | if (rc) { | |
867 | of_node_put(dn); | |
868 | return -EINVAL; | |
869 | } | |
870 | ||
d98389f3 | 871 | rc = dlpar_cpu_remove(dn, drc_index); |
183deeea NF |
872 | of_node_put(dn); |
873 | ||
d98389f3 | 874 | return rc ? rc : count; |
183deeea NF |
875 | } |
876 | ||
877 | #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ | |
878 | ||
06ba30b6 | 879 | static int pseries_smp_notifier(struct notifier_block *nb, |
f5242e5a | 880 | unsigned long action, void *data) |
413f7c40 | 881 | { |
f5242e5a | 882 | struct of_reconfig_data *rd = data; |
de2780a3 | 883 | int err = 0; |
413f7c40 ME |
884 | |
885 | switch (action) { | |
1cf3d8b3 | 886 | case OF_RECONFIG_ATTACH_NODE: |
f5242e5a | 887 | err = pseries_add_processor(rd->dn); |
413f7c40 | 888 | break; |
1cf3d8b3 | 889 | case OF_RECONFIG_DETACH_NODE: |
f5242e5a | 890 | pseries_remove_processor(rd->dn); |
413f7c40 | 891 | break; |
413f7c40 | 892 | } |
de2780a3 | 893 | return notifier_from_errno(err); |
413f7c40 ME |
894 | } |
895 | ||
06ba30b6 ME |
896 | static struct notifier_block pseries_smp_nb = { |
897 | .notifier_call = pseries_smp_notifier, | |
413f7c40 ME |
898 | }; |
899 | ||
3aa565f5 GS |
900 | #define MAX_CEDE_LATENCY_LEVELS 4 |
901 | #define CEDE_LATENCY_PARAM_LENGTH 10 | |
902 | #define CEDE_LATENCY_PARAM_MAX_LENGTH \ | |
903 | (MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char)) | |
904 | #define CEDE_LATENCY_TOKEN 45 | |
905 | ||
906 | static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH]; | |
907 | ||
908 | static int parse_cede_parameters(void) | |
909 | { | |
3aa565f5 | 910 | memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH); |
20a8ab97 AB |
911 | return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1, |
912 | NULL, | |
913 | CEDE_LATENCY_TOKEN, | |
914 | __pa(cede_parameters), | |
915 | CEDE_LATENCY_PARAM_MAX_LENGTH); | |
3aa565f5 GS |
916 | } |
917 | ||
0332c2d4 ME |
918 | static int __init pseries_cpu_hotplug_init(void) |
919 | { | |
3aa565f5 | 920 | int cpu; |
f8b67691 | 921 | int qcss_tok; |
64f27585 | 922 | |
183deeea NF |
923 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE |
924 | ppc_md.cpu_probe = dlpar_cpu_probe; | |
925 | ppc_md.cpu_release = dlpar_cpu_release; | |
926 | #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ | |
927 | ||
41dd03a9 | 928 | rtas_stop_self_token = rtas_token("stop-self"); |
674fa677 | 929 | qcss_tok = rtas_token("query-cpu-stopped-state"); |
0332c2d4 | 930 | |
41dd03a9 | 931 | if (rtas_stop_self_token == RTAS_UNKNOWN_SERVICE || |
674fa677 ME |
932 | qcss_tok == RTAS_UNKNOWN_SERVICE) { |
933 | printk(KERN_INFO "CPU Hotplug not supported by firmware " | |
934 | "- disabling.\n"); | |
935 | return 0; | |
936 | } | |
04da6af9 | 937 | |
06ba30b6 ME |
938 | ppc_md.cpu_die = pseries_mach_cpu_die; |
939 | smp_ops->cpu_disable = pseries_cpu_disable; | |
940 | smp_ops->cpu_die = pseries_cpu_die; | |
413f7c40 ME |
941 | |
942 | /* Processors can be added/removed only on LPAR */ | |
3aa565f5 | 943 | if (firmware_has_feature(FW_FEATURE_LPAR)) { |
1cf3d8b3 | 944 | of_reconfig_notifier_register(&pseries_smp_nb); |
3aa565f5 GS |
945 | cpu_maps_update_begin(); |
946 | if (cede_offline_enabled && parse_cede_parameters() == 0) { | |
947 | default_offline_state = CPU_STATE_INACTIVE; | |
948 | for_each_online_cpu(cpu) | |
949 | set_default_offline_state(cpu); | |
950 | } | |
951 | cpu_maps_update_done(); | |
952 | } | |
413f7c40 | 953 | |
0332c2d4 ME |
954 | return 0; |
955 | } | |
d2a36071 | 956 | machine_arch_initcall(pseries, pseries_cpu_hotplug_init); |