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> |
212bebb4 DD |
37 | #include <asm/plpar_wrappers.h> |
38 | ||
183deeea | 39 | #include "pseries.h" |
3aa565f5 | 40 | #include "offline_states.h" |
0332c2d4 ME |
41 | |
42 | /* This version can't take the spinlock, because it never returns */ | |
41dd03a9 | 43 | static int rtas_stop_self_token = RTAS_UNKNOWN_SERVICE; |
0332c2d4 | 44 | |
3aa565f5 GS |
45 | static DEFINE_PER_CPU(enum cpu_state_vals, preferred_offline_state) = |
46 | CPU_STATE_OFFLINE; | |
47 | static DEFINE_PER_CPU(enum cpu_state_vals, current_state) = CPU_STATE_OFFLINE; | |
48 | ||
49 | static enum cpu_state_vals default_offline_state = CPU_STATE_OFFLINE; | |
50 | ||
4cc7ecb7 | 51 | static bool cede_offline_enabled __read_mostly = true; |
3aa565f5 GS |
52 | |
53 | /* | |
54 | * Enable/disable cede_offline when available. | |
55 | */ | |
56 | static int __init setup_cede_offline(char *str) | |
57 | { | |
4cc7ecb7 | 58 | return (kstrtobool(str, &cede_offline_enabled) == 0); |
3aa565f5 GS |
59 | } |
60 | ||
61 | __setup("cede_offline=", setup_cede_offline); | |
62 | ||
63 | enum cpu_state_vals get_cpu_current_state(int cpu) | |
64 | { | |
65 | return per_cpu(current_state, cpu); | |
66 | } | |
67 | ||
68 | void set_cpu_current_state(int cpu, enum cpu_state_vals state) | |
69 | { | |
70 | per_cpu(current_state, cpu) = state; | |
71 | } | |
72 | ||
73 | enum cpu_state_vals get_preferred_offline_state(int cpu) | |
74 | { | |
75 | return per_cpu(preferred_offline_state, cpu); | |
76 | } | |
77 | ||
78 | void set_preferred_offline_state(int cpu, enum cpu_state_vals state) | |
79 | { | |
80 | per_cpu(preferred_offline_state, cpu) = state; | |
81 | } | |
82 | ||
83 | void set_default_offline_state(int cpu) | |
84 | { | |
85 | per_cpu(preferred_offline_state, cpu) = default_offline_state; | |
86 | } | |
87 | ||
04da6af9 | 88 | static void rtas_stop_self(void) |
0332c2d4 | 89 | { |
b2e8590f | 90 | static struct rtas_args args; |
4fb8d027 | 91 | |
0332c2d4 ME |
92 | local_irq_disable(); |
93 | ||
41dd03a9 | 94 | BUG_ON(rtas_stop_self_token == RTAS_UNKNOWN_SERVICE); |
0332c2d4 ME |
95 | |
96 | printk("cpu %u (hwid %u) Ready to die...\n", | |
97 | smp_processor_id(), hard_smp_processor_id()); | |
b2e8590f ME |
98 | |
99 | rtas_call_unlocked(&args, rtas_stop_self_token, 0, 1, NULL); | |
0332c2d4 ME |
100 | |
101 | panic("Alas, I survived.\n"); | |
102 | } | |
103 | ||
06ba30b6 | 104 | static void pseries_mach_cpu_die(void) |
04da6af9 | 105 | { |
3aa565f5 GS |
106 | unsigned int cpu = smp_processor_id(); |
107 | unsigned int hwcpu = hard_smp_processor_id(); | |
108 | u8 cede_latency_hint = 0; | |
109 | ||
04da6af9 ME |
110 | local_irq_disable(); |
111 | idle_task_exit(); | |
c3e8506c | 112 | xics_teardown_cpu(); |
3aa565f5 GS |
113 | |
114 | if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) { | |
115 | set_cpu_current_state(cpu, CPU_STATE_INACTIVE); | |
32d8ad4e BK |
116 | if (ppc_md.suspend_disable_cpu) |
117 | ppc_md.suspend_disable_cpu(); | |
118 | ||
3aa565f5 GS |
119 | cede_latency_hint = 2; |
120 | ||
121 | get_lppaca()->idle = 1; | |
f13c13a0 | 122 | if (!lppaca_shared_proc(get_lppaca())) |
3aa565f5 GS |
123 | get_lppaca()->donate_dedicated_cpu = 1; |
124 | ||
3aa565f5 | 125 | while (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) { |
fb912568 LZ |
126 | while (!prep_irq_for_idle()) { |
127 | local_irq_enable(); | |
128 | local_irq_disable(); | |
129 | } | |
130 | ||
3aa565f5 | 131 | extended_cede_processor(cede_latency_hint); |
3aa565f5 GS |
132 | } |
133 | ||
fb912568 LZ |
134 | local_irq_disable(); |
135 | ||
f13c13a0 | 136 | if (!lppaca_shared_proc(get_lppaca())) |
3aa565f5 GS |
137 | get_lppaca()->donate_dedicated_cpu = 0; |
138 | get_lppaca()->idle = 0; | |
3aa565f5 | 139 | |
0212f260 | 140 | if (get_preferred_offline_state(cpu) == CPU_STATE_ONLINE) { |
598c8231 | 141 | unregister_slb_shadow(hwcpu); |
3aa565f5 | 142 | |
fb912568 | 143 | hard_irq_disable(); |
0212f260 VS |
144 | /* |
145 | * Call to start_secondary_resume() will not return. | |
146 | * Kernel stack will be reset and start_secondary() | |
147 | * will be called to continue the online operation. | |
148 | */ | |
149 | start_secondary_resume(); | |
150 | } | |
151 | } | |
3aa565f5 | 152 | |
0212f260 VS |
153 | /* Requested state is CPU_STATE_OFFLINE at this point */ |
154 | WARN_ON(get_preferred_offline_state(cpu) != CPU_STATE_OFFLINE); | |
3aa565f5 | 155 | |
0212f260 | 156 | set_cpu_current_state(cpu, CPU_STATE_OFFLINE); |
598c8231 | 157 | unregister_slb_shadow(hwcpu); |
0212f260 | 158 | rtas_stop_self(); |
3aa565f5 | 159 | |
04da6af9 ME |
160 | /* Should never get here... */ |
161 | BUG(); | |
162 | for(;;); | |
163 | } | |
164 | ||
06ba30b6 | 165 | static int pseries_cpu_disable(void) |
413f7c40 ME |
166 | { |
167 | int cpu = smp_processor_id(); | |
168 | ||
ea0f1cab | 169 | set_cpu_online(cpu, false); |
413f7c40 ME |
170 | vdso_data->processorCount--; |
171 | ||
172 | /*fix boot_cpuid here*/ | |
173 | if (cpu == boot_cpuid) | |
8729faaa | 174 | boot_cpuid = cpumask_any(cpu_online_mask); |
413f7c40 ME |
175 | |
176 | /* FIXME: abstract this to not be platform specific later on */ | |
177 | xics_migrate_irqs_away(); | |
178 | return 0; | |
179 | } | |
180 | ||
3aa565f5 GS |
181 | /* |
182 | * pseries_cpu_die: Wait for the cpu to die. | |
183 | * @cpu: logical processor id of the CPU whose death we're awaiting. | |
184 | * | |
185 | * This function is called from the context of the thread which is performing | |
186 | * the cpu-offline. Here we wait for long enough to allow the cpu in question | |
187 | * to self-destroy so that the cpu-offline thread can send the CPU_DEAD | |
188 | * notifications. | |
189 | * | |
190 | * OTOH, pseries_mach_cpu_die() is called by the @cpu when it wants to | |
191 | * self-destruct. | |
192 | */ | |
06ba30b6 | 193 | static void pseries_cpu_die(unsigned int cpu) |
413f7c40 ME |
194 | { |
195 | int tries; | |
3aa565f5 | 196 | int cpu_status = 1; |
413f7c40 ME |
197 | unsigned int pcpu = get_hard_smp_processor_id(cpu); |
198 | ||
3aa565f5 GS |
199 | if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) { |
200 | cpu_status = 1; | |
940ce422 | 201 | for (tries = 0; tries < 5000; tries++) { |
3aa565f5 GS |
202 | if (get_cpu_current_state(cpu) == CPU_STATE_INACTIVE) { |
203 | cpu_status = 0; | |
204 | break; | |
205 | } | |
940ce422 | 206 | msleep(1); |
3aa565f5 GS |
207 | } |
208 | } else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) { | |
209 | ||
210 | for (tries = 0; tries < 25; tries++) { | |
f8b67691 MN |
211 | cpu_status = smp_query_cpu_stopped(pcpu); |
212 | if (cpu_status == QCSS_STOPPED || | |
213 | cpu_status == QCSS_HARDWARE_ERROR) | |
3aa565f5 GS |
214 | break; |
215 | cpu_relax(); | |
216 | } | |
413f7c40 | 217 | } |
3aa565f5 | 218 | |
413f7c40 ME |
219 | if (cpu_status != 0) { |
220 | printk("Querying DEAD? cpu %i (%i) shows %i\n", | |
221 | cpu, pcpu, cpu_status); | |
222 | } | |
223 | ||
25985edc | 224 | /* Isolation and deallocation are definitely done by |
413f7c40 ME |
225 | * drslot_chrp_cpu. If they were not they would be |
226 | * done here. Change isolate state to Isolate and | |
227 | * change allocation-state to Unusable. | |
228 | */ | |
229 | paca[cpu].cpu_start = 0; | |
230 | } | |
231 | ||
232 | /* | |
828a6986 | 233 | * Update cpu_present_mask and paca(s) for a new cpu node. The wrinkle |
413f7c40 ME |
234 | * here is that a cpu device node may represent up to two logical cpus |
235 | * in the SMT case. We must honor the assumption in other code that | |
236 | * the logical ids for sibling SMT threads x and y are adjacent, such | |
237 | * that x^1 == y and y^1 == x. | |
238 | */ | |
06ba30b6 | 239 | static int pseries_add_processor(struct device_node *np) |
413f7c40 ME |
240 | { |
241 | unsigned int cpu; | |
8729faaa | 242 | cpumask_var_t candidate_mask, tmp; |
413f7c40 | 243 | int err = -ENOSPC, len, nthreads, i; |
d6f1e7ab | 244 | const __be32 *intserv; |
413f7c40 | 245 | |
e2eb6392 | 246 | intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len); |
413f7c40 ME |
247 | if (!intserv) |
248 | return 0; | |
249 | ||
8729faaa AB |
250 | zalloc_cpumask_var(&candidate_mask, GFP_KERNEL); |
251 | zalloc_cpumask_var(&tmp, GFP_KERNEL); | |
252 | ||
413f7c40 ME |
253 | nthreads = len / sizeof(u32); |
254 | for (i = 0; i < nthreads; i++) | |
8729faaa | 255 | cpumask_set_cpu(i, tmp); |
413f7c40 | 256 | |
86ef5c9a | 257 | cpu_maps_update_begin(); |
413f7c40 | 258 | |
8729faaa | 259 | BUG_ON(!cpumask_subset(cpu_present_mask, cpu_possible_mask)); |
413f7c40 ME |
260 | |
261 | /* Get a bitmap of unoccupied slots. */ | |
8729faaa AB |
262 | cpumask_xor(candidate_mask, cpu_possible_mask, cpu_present_mask); |
263 | if (cpumask_empty(candidate_mask)) { | |
413f7c40 ME |
264 | /* If we get here, it most likely means that NR_CPUS is |
265 | * less than the partition's max processors setting. | |
266 | */ | |
b7c670d6 RH |
267 | printk(KERN_ERR "Cannot add cpu %pOF; this system configuration" |
268 | " supports %d logical cpus.\n", np, | |
53a448c3 | 269 | num_possible_cpus()); |
413f7c40 ME |
270 | goto out_unlock; |
271 | } | |
272 | ||
8729faaa AB |
273 | while (!cpumask_empty(tmp)) |
274 | if (cpumask_subset(tmp, candidate_mask)) | |
413f7c40 ME |
275 | /* Found a range where we can insert the new cpu(s) */ |
276 | break; | |
277 | else | |
8729faaa | 278 | cpumask_shift_left(tmp, tmp, nthreads); |
413f7c40 | 279 | |
8729faaa | 280 | if (cpumask_empty(tmp)) { |
828a6986 | 281 | printk(KERN_ERR "Unable to find space in cpu_present_mask for" |
413f7c40 ME |
282 | " processor %s with %d thread(s)\n", np->name, |
283 | nthreads); | |
284 | goto out_unlock; | |
285 | } | |
286 | ||
8729faaa | 287 | for_each_cpu(cpu, tmp) { |
104699c0 | 288 | BUG_ON(cpu_present(cpu)); |
ea0f1cab | 289 | set_cpu_present(cpu, true); |
d6f1e7ab | 290 | set_hard_smp_processor_id(cpu, be32_to_cpu(*intserv++)); |
413f7c40 ME |
291 | } |
292 | err = 0; | |
293 | out_unlock: | |
86ef5c9a | 294 | cpu_maps_update_done(); |
8729faaa AB |
295 | free_cpumask_var(candidate_mask); |
296 | free_cpumask_var(tmp); | |
413f7c40 ME |
297 | return err; |
298 | } | |
299 | ||
300 | /* | |
301 | * Update the present map for a cpu node which is going away, and set | |
302 | * the hard id in the paca(s) to -1 to be consistent with boot time | |
303 | * convention for non-present cpus. | |
304 | */ | |
06ba30b6 | 305 | static void pseries_remove_processor(struct device_node *np) |
413f7c40 ME |
306 | { |
307 | unsigned int cpu; | |
308 | int len, nthreads, i; | |
e36d1227 TF |
309 | const __be32 *intserv; |
310 | u32 thread; | |
413f7c40 | 311 | |
e2eb6392 | 312 | intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len); |
413f7c40 ME |
313 | if (!intserv) |
314 | return; | |
315 | ||
316 | nthreads = len / sizeof(u32); | |
317 | ||
86ef5c9a | 318 | cpu_maps_update_begin(); |
413f7c40 | 319 | for (i = 0; i < nthreads; i++) { |
e36d1227 | 320 | thread = be32_to_cpu(intserv[i]); |
413f7c40 | 321 | for_each_present_cpu(cpu) { |
e36d1227 | 322 | if (get_hard_smp_processor_id(cpu) != thread) |
413f7c40 ME |
323 | continue; |
324 | BUG_ON(cpu_online(cpu)); | |
ea0f1cab | 325 | set_cpu_present(cpu, false); |
413f7c40 ME |
326 | set_hard_smp_processor_id(cpu, -1); |
327 | break; | |
328 | } | |
8729faaa | 329 | if (cpu >= nr_cpu_ids) |
413f7c40 | 330 | printk(KERN_WARNING "Could not find cpu to remove " |
e36d1227 | 331 | "with physical id 0x%x\n", thread); |
413f7c40 | 332 | } |
86ef5c9a | 333 | cpu_maps_update_done(); |
413f7c40 ME |
334 | } |
335 | ||
183deeea NF |
336 | static int dlpar_online_cpu(struct device_node *dn) |
337 | { | |
338 | int rc = 0; | |
339 | unsigned int cpu; | |
340 | int len, nthreads, i; | |
341 | const __be32 *intserv; | |
342 | u32 thread; | |
343 | ||
344 | intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len); | |
345 | if (!intserv) | |
346 | return -EINVAL; | |
347 | ||
348 | nthreads = len / sizeof(u32); | |
349 | ||
350 | cpu_maps_update_begin(); | |
351 | for (i = 0; i < nthreads; i++) { | |
352 | thread = be32_to_cpu(intserv[i]); | |
353 | for_each_present_cpu(cpu) { | |
354 | if (get_hard_smp_processor_id(cpu) != thread) | |
355 | continue; | |
356 | BUG_ON(get_cpu_current_state(cpu) | |
357 | != CPU_STATE_OFFLINE); | |
358 | cpu_maps_update_done(); | |
359 | rc = device_online(get_cpu_device(cpu)); | |
360 | if (rc) | |
361 | goto out; | |
362 | cpu_maps_update_begin(); | |
363 | ||
364 | break; | |
365 | } | |
366 | if (cpu == num_possible_cpus()) | |
367 | printk(KERN_WARNING "Could not find cpu to online " | |
368 | "with physical id 0x%x\n", thread); | |
369 | } | |
370 | cpu_maps_update_done(); | |
371 | ||
372 | out: | |
373 | return rc; | |
374 | ||
375 | } | |
376 | ||
377 | static bool dlpar_cpu_exists(struct device_node *parent, u32 drc_index) | |
378 | { | |
379 | struct device_node *child = NULL; | |
380 | u32 my_drc_index; | |
381 | bool found; | |
382 | int rc; | |
383 | ||
384 | /* Assume cpu doesn't exist */ | |
385 | found = false; | |
386 | ||
387 | for_each_child_of_node(parent, child) { | |
388 | rc = of_property_read_u32(child, "ibm,my-drc-index", | |
389 | &my_drc_index); | |
390 | if (rc) | |
391 | continue; | |
392 | ||
393 | if (my_drc_index == drc_index) { | |
394 | of_node_put(child); | |
395 | found = true; | |
396 | break; | |
397 | } | |
398 | } | |
399 | ||
400 | return found; | |
401 | } | |
402 | ||
90edf184 NF |
403 | static bool valid_cpu_drc_index(struct device_node *parent, u32 drc_index) |
404 | { | |
405 | bool found = false; | |
406 | int rc, index; | |
407 | ||
408 | index = 0; | |
409 | while (!found) { | |
410 | u32 drc; | |
411 | ||
412 | rc = of_property_read_u32_index(parent, "ibm,drc-indexes", | |
413 | index++, &drc); | |
414 | if (rc) | |
415 | break; | |
416 | ||
417 | if (drc == drc_index) | |
418 | found = true; | |
419 | } | |
420 | ||
421 | return found; | |
422 | } | |
423 | ||
d98389f3 | 424 | static ssize_t dlpar_cpu_add(u32 drc_index) |
183deeea NF |
425 | { |
426 | struct device_node *dn, *parent; | |
e666ae0b NF |
427 | int rc, saved_rc; |
428 | ||
429 | pr_debug("Attempting to add CPU, drc index: %x\n", drc_index); | |
183deeea | 430 | |
183deeea | 431 | parent = of_find_node_by_path("/cpus"); |
e666ae0b NF |
432 | if (!parent) { |
433 | pr_warn("Failed to find CPU root node \"/cpus\"\n"); | |
183deeea | 434 | return -ENODEV; |
e666ae0b | 435 | } |
183deeea NF |
436 | |
437 | if (dlpar_cpu_exists(parent, drc_index)) { | |
438 | of_node_put(parent); | |
e666ae0b | 439 | pr_warn("CPU with drc index %x already exists\n", drc_index); |
183deeea NF |
440 | return -EINVAL; |
441 | } | |
442 | ||
90edf184 NF |
443 | if (!valid_cpu_drc_index(parent, drc_index)) { |
444 | of_node_put(parent); | |
445 | pr_warn("Cannot find CPU (drc index %x) to add.\n", drc_index); | |
446 | return -EINVAL; | |
447 | } | |
448 | ||
183deeea NF |
449 | rc = dlpar_acquire_drc(drc_index); |
450 | if (rc) { | |
e666ae0b NF |
451 | pr_warn("Failed to acquire DRC, rc: %d, drc index: %x\n", |
452 | rc, drc_index); | |
183deeea NF |
453 | of_node_put(parent); |
454 | return -EINVAL; | |
455 | } | |
456 | ||
457 | dn = dlpar_configure_connector(cpu_to_be32(drc_index), parent); | |
458 | of_node_put(parent); | |
e666ae0b NF |
459 | if (!dn) { |
460 | pr_warn("Failed call to configure-connector, drc index: %x\n", | |
461 | drc_index); | |
462 | dlpar_release_drc(drc_index); | |
183deeea | 463 | return -EINVAL; |
e666ae0b | 464 | } |
183deeea NF |
465 | |
466 | rc = dlpar_attach_node(dn); | |
467 | if (rc) { | |
e666ae0b NF |
468 | saved_rc = rc; |
469 | pr_warn("Failed to attach node %s, rc: %d, drc index: %x\n", | |
470 | dn->name, rc, drc_index); | |
471 | ||
472 | rc = dlpar_release_drc(drc_index); | |
473 | if (!rc) | |
474 | dlpar_free_cc_nodes(dn); | |
475 | ||
476 | return saved_rc; | |
183deeea NF |
477 | } |
478 | ||
479 | rc = dlpar_online_cpu(dn); | |
e666ae0b NF |
480 | if (rc) { |
481 | saved_rc = rc; | |
482 | pr_warn("Failed to online cpu %s, rc: %d, drc index: %x\n", | |
483 | dn->name, rc, drc_index); | |
484 | ||
485 | rc = dlpar_detach_node(dn); | |
486 | if (!rc) | |
487 | dlpar_release_drc(drc_index); | |
488 | ||
489 | return saved_rc; | |
490 | } | |
491 | ||
492 | pr_debug("Successfully added CPU %s, drc index: %x\n", dn->name, | |
493 | drc_index); | |
d98389f3 | 494 | return rc; |
183deeea NF |
495 | } |
496 | ||
497 | static int dlpar_offline_cpu(struct device_node *dn) | |
498 | { | |
499 | int rc = 0; | |
500 | unsigned int cpu; | |
501 | int len, nthreads, i; | |
502 | const __be32 *intserv; | |
503 | u32 thread; | |
504 | ||
505 | intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len); | |
506 | if (!intserv) | |
507 | return -EINVAL; | |
508 | ||
509 | nthreads = len / sizeof(u32); | |
510 | ||
511 | cpu_maps_update_begin(); | |
512 | for (i = 0; i < nthreads; i++) { | |
513 | thread = be32_to_cpu(intserv[i]); | |
514 | for_each_present_cpu(cpu) { | |
515 | if (get_hard_smp_processor_id(cpu) != thread) | |
516 | continue; | |
517 | ||
518 | if (get_cpu_current_state(cpu) == CPU_STATE_OFFLINE) | |
519 | break; | |
520 | ||
521 | if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) { | |
522 | set_preferred_offline_state(cpu, | |
523 | CPU_STATE_OFFLINE); | |
524 | cpu_maps_update_done(); | |
525 | rc = device_offline(get_cpu_device(cpu)); | |
526 | if (rc) | |
527 | goto out; | |
528 | cpu_maps_update_begin(); | |
529 | break; | |
530 | ||
531 | } | |
532 | ||
533 | /* | |
534 | * The cpu is in CPU_STATE_INACTIVE. | |
535 | * Upgrade it's state to CPU_STATE_OFFLINE. | |
536 | */ | |
537 | set_preferred_offline_state(cpu, CPU_STATE_OFFLINE); | |
538 | BUG_ON(plpar_hcall_norets(H_PROD, thread) | |
539 | != H_SUCCESS); | |
540 | __cpu_die(cpu); | |
541 | break; | |
542 | } | |
543 | if (cpu == num_possible_cpus()) | |
544 | printk(KERN_WARNING "Could not find cpu to offline with physical id 0x%x\n", thread); | |
545 | } | |
546 | cpu_maps_update_done(); | |
547 | ||
548 | out: | |
549 | return rc; | |
550 | ||
551 | } | |
552 | ||
d98389f3 NF |
553 | static ssize_t dlpar_cpu_remove(struct device_node *dn, u32 drc_index) |
554 | { | |
555 | int rc; | |
556 | ||
b802ab46 | 557 | pr_debug("Attempting to remove CPU %s, drc index: %x\n", |
e666ae0b NF |
558 | dn->name, drc_index); |
559 | ||
d98389f3 | 560 | rc = dlpar_offline_cpu(dn); |
e666ae0b NF |
561 | if (rc) { |
562 | pr_warn("Failed to offline CPU %s, rc: %d\n", dn->name, rc); | |
d98389f3 | 563 | return -EINVAL; |
e666ae0b | 564 | } |
d98389f3 NF |
565 | |
566 | rc = dlpar_release_drc(drc_index); | |
e666ae0b NF |
567 | if (rc) { |
568 | pr_warn("Failed to release drc (%x) for CPU %s, rc: %d\n", | |
569 | drc_index, dn->name, rc); | |
570 | dlpar_online_cpu(dn); | |
d98389f3 | 571 | return rc; |
e666ae0b | 572 | } |
d98389f3 NF |
573 | |
574 | rc = dlpar_detach_node(dn); | |
e666ae0b NF |
575 | if (rc) { |
576 | int saved_rc = rc; | |
d98389f3 | 577 | |
e666ae0b NF |
578 | pr_warn("Failed to detach CPU %s, rc: %d", dn->name, rc); |
579 | ||
580 | rc = dlpar_acquire_drc(drc_index); | |
581 | if (!rc) | |
582 | dlpar_online_cpu(dn); | |
583 | ||
584 | return saved_rc; | |
585 | } | |
586 | ||
587 | pr_debug("Successfully removed CPU, drc index: %x\n", drc_index); | |
588 | return 0; | |
d98389f3 NF |
589 | } |
590 | ||
ac713800 NF |
591 | static struct device_node *cpu_drc_index_to_dn(u32 drc_index) |
592 | { | |
593 | struct device_node *dn; | |
594 | u32 my_index; | |
595 | int rc; | |
596 | ||
597 | for_each_node_by_type(dn, "cpu") { | |
598 | rc = of_property_read_u32(dn, "ibm,my-drc-index", &my_index); | |
599 | if (rc) | |
600 | continue; | |
601 | ||
602 | if (my_index == drc_index) | |
603 | break; | |
604 | } | |
605 | ||
606 | return dn; | |
607 | } | |
608 | ||
609 | static int dlpar_cpu_remove_by_index(u32 drc_index) | |
610 | { | |
611 | struct device_node *dn; | |
612 | int rc; | |
613 | ||
614 | dn = cpu_drc_index_to_dn(drc_index); | |
615 | if (!dn) { | |
616 | pr_warn("Cannot find CPU (drc index %x) to remove\n", | |
617 | drc_index); | |
618 | return -ENODEV; | |
619 | } | |
620 | ||
621 | rc = dlpar_cpu_remove(dn, drc_index); | |
622 | of_node_put(dn); | |
623 | return rc; | |
624 | } | |
625 | ||
626 | static int find_dlpar_cpus_to_remove(u32 *cpu_drcs, int cpus_to_remove) | |
627 | { | |
628 | struct device_node *dn; | |
629 | int cpus_found = 0; | |
630 | int rc; | |
631 | ||
632 | /* We want to find cpus_to_remove + 1 CPUs to ensure we do not | |
633 | * remove the last CPU. | |
634 | */ | |
635 | for_each_node_by_type(dn, "cpu") { | |
636 | cpus_found++; | |
637 | ||
638 | if (cpus_found > cpus_to_remove) { | |
639 | of_node_put(dn); | |
640 | break; | |
641 | } | |
642 | ||
643 | /* Note that cpus_found is always 1 ahead of the index | |
644 | * into the cpu_drcs array, so we use cpus_found - 1 | |
645 | */ | |
646 | rc = of_property_read_u32(dn, "ibm,my-drc-index", | |
647 | &cpu_drcs[cpus_found - 1]); | |
648 | if (rc) { | |
649 | pr_warn("Error occurred getting drc-index for %s\n", | |
650 | dn->name); | |
651 | of_node_put(dn); | |
652 | return -1; | |
653 | } | |
654 | } | |
655 | ||
656 | if (cpus_found < cpus_to_remove) { | |
657 | pr_warn("Failed to find enough CPUs (%d of %d) to remove\n", | |
658 | cpus_found, cpus_to_remove); | |
659 | } else if (cpus_found == cpus_to_remove) { | |
660 | pr_warn("Cannot remove all CPUs\n"); | |
661 | } | |
662 | ||
663 | return cpus_found; | |
664 | } | |
665 | ||
666 | static int dlpar_cpu_remove_by_count(u32 cpus_to_remove) | |
667 | { | |
668 | u32 *cpu_drcs; | |
669 | int cpus_found; | |
670 | int cpus_removed = 0; | |
671 | int i, rc; | |
672 | ||
673 | pr_debug("Attempting to hot-remove %d CPUs\n", cpus_to_remove); | |
674 | ||
675 | cpu_drcs = kcalloc(cpus_to_remove, sizeof(*cpu_drcs), GFP_KERNEL); | |
676 | if (!cpu_drcs) | |
677 | return -EINVAL; | |
678 | ||
679 | cpus_found = find_dlpar_cpus_to_remove(cpu_drcs, cpus_to_remove); | |
680 | if (cpus_found <= cpus_to_remove) { | |
681 | kfree(cpu_drcs); | |
682 | return -EINVAL; | |
683 | } | |
684 | ||
685 | for (i = 0; i < cpus_to_remove; i++) { | |
686 | rc = dlpar_cpu_remove_by_index(cpu_drcs[i]); | |
687 | if (rc) | |
688 | break; | |
689 | ||
690 | cpus_removed++; | |
691 | } | |
692 | ||
693 | if (cpus_removed != cpus_to_remove) { | |
694 | pr_warn("CPU hot-remove failed, adding back removed CPUs\n"); | |
695 | ||
696 | for (i = 0; i < cpus_removed; i++) | |
697 | dlpar_cpu_add(cpu_drcs[i]); | |
698 | ||
699 | rc = -EINVAL; | |
700 | } else { | |
701 | rc = 0; | |
702 | } | |
703 | ||
704 | kfree(cpu_drcs); | |
705 | return rc; | |
706 | } | |
707 | ||
90edf184 NF |
708 | static int find_dlpar_cpus_to_add(u32 *cpu_drcs, u32 cpus_to_add) |
709 | { | |
710 | struct device_node *parent; | |
711 | int cpus_found = 0; | |
712 | int index, rc; | |
713 | ||
714 | parent = of_find_node_by_path("/cpus"); | |
715 | if (!parent) { | |
716 | pr_warn("Could not find CPU root node in device tree\n"); | |
717 | kfree(cpu_drcs); | |
718 | return -1; | |
719 | } | |
720 | ||
721 | /* Search the ibm,drc-indexes array for possible CPU drcs to | |
722 | * add. Note that the format of the ibm,drc-indexes array is | |
723 | * the number of entries in the array followed by the array | |
724 | * of drc values so we start looking at index = 1. | |
725 | */ | |
726 | index = 1; | |
727 | while (cpus_found < cpus_to_add) { | |
728 | u32 drc; | |
729 | ||
730 | rc = of_property_read_u32_index(parent, "ibm,drc-indexes", | |
731 | index++, &drc); | |
732 | if (rc) | |
733 | break; | |
734 | ||
735 | if (dlpar_cpu_exists(parent, drc)) | |
736 | continue; | |
737 | ||
738 | cpu_drcs[cpus_found++] = drc; | |
739 | } | |
740 | ||
741 | of_node_put(parent); | |
742 | return cpus_found; | |
743 | } | |
744 | ||
745 | static int dlpar_cpu_add_by_count(u32 cpus_to_add) | |
746 | { | |
747 | u32 *cpu_drcs; | |
748 | int cpus_added = 0; | |
749 | int cpus_found; | |
750 | int i, rc; | |
751 | ||
752 | pr_debug("Attempting to hot-add %d CPUs\n", cpus_to_add); | |
753 | ||
754 | cpu_drcs = kcalloc(cpus_to_add, sizeof(*cpu_drcs), GFP_KERNEL); | |
755 | if (!cpu_drcs) | |
756 | return -EINVAL; | |
757 | ||
758 | cpus_found = find_dlpar_cpus_to_add(cpu_drcs, cpus_to_add); | |
759 | if (cpus_found < cpus_to_add) { | |
760 | pr_warn("Failed to find enough CPUs (%d of %d) to add\n", | |
761 | cpus_found, cpus_to_add); | |
762 | kfree(cpu_drcs); | |
763 | return -EINVAL; | |
764 | } | |
765 | ||
766 | for (i = 0; i < cpus_to_add; i++) { | |
767 | rc = dlpar_cpu_add(cpu_drcs[i]); | |
768 | if (rc) | |
769 | break; | |
770 | ||
771 | cpus_added++; | |
772 | } | |
773 | ||
774 | if (cpus_added < cpus_to_add) { | |
775 | pr_warn("CPU hot-add failed, removing any added CPUs\n"); | |
776 | ||
777 | for (i = 0; i < cpus_added; i++) | |
778 | dlpar_cpu_remove_by_index(cpu_drcs[i]); | |
779 | ||
780 | rc = -EINVAL; | |
781 | } else { | |
782 | rc = 0; | |
783 | } | |
784 | ||
785 | kfree(cpu_drcs); | |
786 | return rc; | |
787 | } | |
788 | ||
ac713800 NF |
789 | int dlpar_cpu(struct pseries_hp_errorlog *hp_elog) |
790 | { | |
791 | u32 count, drc_index; | |
792 | int rc; | |
793 | ||
794 | count = hp_elog->_drc_u.drc_count; | |
795 | drc_index = hp_elog->_drc_u.drc_index; | |
796 | ||
797 | lock_device_hotplug(); | |
798 | ||
799 | switch (hp_elog->action) { | |
800 | case PSERIES_HP_ELOG_ACTION_REMOVE: | |
801 | if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT) | |
802 | rc = dlpar_cpu_remove_by_count(count); | |
803 | else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX) | |
804 | rc = dlpar_cpu_remove_by_index(drc_index); | |
805 | else | |
806 | rc = -EINVAL; | |
807 | break; | |
90edf184 NF |
808 | case PSERIES_HP_ELOG_ACTION_ADD: |
809 | if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT) | |
810 | rc = dlpar_cpu_add_by_count(count); | |
811 | else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX) | |
812 | rc = dlpar_cpu_add(drc_index); | |
813 | else | |
814 | rc = -EINVAL; | |
815 | break; | |
ac713800 NF |
816 | default: |
817 | pr_err("Invalid action (%d) specified\n", hp_elog->action); | |
818 | rc = -EINVAL; | |
819 | break; | |
820 | } | |
821 | ||
822 | unlock_device_hotplug(); | |
823 | return rc; | |
824 | } | |
825 | ||
d98389f3 NF |
826 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE |
827 | ||
828 | static ssize_t dlpar_cpu_probe(const char *buf, size_t count) | |
829 | { | |
830 | u32 drc_index; | |
831 | int rc; | |
832 | ||
833 | rc = kstrtou32(buf, 0, &drc_index); | |
834 | if (rc) | |
835 | return -EINVAL; | |
836 | ||
837 | rc = dlpar_cpu_add(drc_index); | |
838 | ||
839 | return rc ? rc : count; | |
840 | } | |
841 | ||
183deeea NF |
842 | static ssize_t dlpar_cpu_release(const char *buf, size_t count) |
843 | { | |
844 | struct device_node *dn; | |
845 | u32 drc_index; | |
846 | int rc; | |
847 | ||
848 | dn = of_find_node_by_path(buf); | |
849 | if (!dn) | |
850 | return -EINVAL; | |
851 | ||
852 | rc = of_property_read_u32(dn, "ibm,my-drc-index", &drc_index); | |
853 | if (rc) { | |
854 | of_node_put(dn); | |
855 | return -EINVAL; | |
856 | } | |
857 | ||
d98389f3 | 858 | rc = dlpar_cpu_remove(dn, drc_index); |
183deeea NF |
859 | of_node_put(dn); |
860 | ||
d98389f3 | 861 | return rc ? rc : count; |
183deeea NF |
862 | } |
863 | ||
864 | #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ | |
865 | ||
06ba30b6 | 866 | static int pseries_smp_notifier(struct notifier_block *nb, |
f5242e5a | 867 | unsigned long action, void *data) |
413f7c40 | 868 | { |
f5242e5a | 869 | struct of_reconfig_data *rd = data; |
de2780a3 | 870 | int err = 0; |
413f7c40 ME |
871 | |
872 | switch (action) { | |
1cf3d8b3 | 873 | case OF_RECONFIG_ATTACH_NODE: |
f5242e5a | 874 | err = pseries_add_processor(rd->dn); |
413f7c40 | 875 | break; |
1cf3d8b3 | 876 | case OF_RECONFIG_DETACH_NODE: |
f5242e5a | 877 | pseries_remove_processor(rd->dn); |
413f7c40 | 878 | break; |
413f7c40 | 879 | } |
de2780a3 | 880 | return notifier_from_errno(err); |
413f7c40 ME |
881 | } |
882 | ||
06ba30b6 ME |
883 | static struct notifier_block pseries_smp_nb = { |
884 | .notifier_call = pseries_smp_notifier, | |
413f7c40 ME |
885 | }; |
886 | ||
3aa565f5 GS |
887 | #define MAX_CEDE_LATENCY_LEVELS 4 |
888 | #define CEDE_LATENCY_PARAM_LENGTH 10 | |
889 | #define CEDE_LATENCY_PARAM_MAX_LENGTH \ | |
890 | (MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char)) | |
891 | #define CEDE_LATENCY_TOKEN 45 | |
892 | ||
893 | static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH]; | |
894 | ||
895 | static int parse_cede_parameters(void) | |
896 | { | |
3aa565f5 | 897 | memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH); |
20a8ab97 AB |
898 | return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1, |
899 | NULL, | |
900 | CEDE_LATENCY_TOKEN, | |
901 | __pa(cede_parameters), | |
902 | CEDE_LATENCY_PARAM_MAX_LENGTH); | |
3aa565f5 GS |
903 | } |
904 | ||
0332c2d4 ME |
905 | static int __init pseries_cpu_hotplug_init(void) |
906 | { | |
3aa565f5 | 907 | int cpu; |
f8b67691 | 908 | int qcss_tok; |
64f27585 | 909 | |
183deeea NF |
910 | #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE |
911 | ppc_md.cpu_probe = dlpar_cpu_probe; | |
912 | ppc_md.cpu_release = dlpar_cpu_release; | |
913 | #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */ | |
914 | ||
41dd03a9 | 915 | rtas_stop_self_token = rtas_token("stop-self"); |
674fa677 | 916 | qcss_tok = rtas_token("query-cpu-stopped-state"); |
0332c2d4 | 917 | |
41dd03a9 | 918 | if (rtas_stop_self_token == RTAS_UNKNOWN_SERVICE || |
674fa677 ME |
919 | qcss_tok == RTAS_UNKNOWN_SERVICE) { |
920 | printk(KERN_INFO "CPU Hotplug not supported by firmware " | |
921 | "- disabling.\n"); | |
922 | return 0; | |
923 | } | |
04da6af9 | 924 | |
06ba30b6 ME |
925 | ppc_md.cpu_die = pseries_mach_cpu_die; |
926 | smp_ops->cpu_disable = pseries_cpu_disable; | |
927 | smp_ops->cpu_die = pseries_cpu_die; | |
413f7c40 ME |
928 | |
929 | /* Processors can be added/removed only on LPAR */ | |
3aa565f5 | 930 | if (firmware_has_feature(FW_FEATURE_LPAR)) { |
1cf3d8b3 | 931 | of_reconfig_notifier_register(&pseries_smp_nb); |
3aa565f5 GS |
932 | cpu_maps_update_begin(); |
933 | if (cede_offline_enabled && parse_cede_parameters() == 0) { | |
934 | default_offline_state = CPU_STATE_INACTIVE; | |
935 | for_each_online_cpu(cpu) | |
936 | set_default_offline_state(cpu); | |
937 | } | |
938 | cpu_maps_update_done(); | |
939 | } | |
413f7c40 | 940 | |
0332c2d4 ME |
941 | return 0; |
942 | } | |
d2a36071 | 943 | machine_arch_initcall(pseries, pseries_cpu_hotplug_init); |