GNU Linux-libre 5.4.274-gnu1
[releases.git] / arch / powerpc / platforms / pseries / hotplug-cpu.c
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * pseries CPU Hotplug infrastructure.
4  *
5  * Split out from arch/powerpc/platforms/pseries/setup.c
6  *  arch/powerpc/kernel/rtas.c, and arch/powerpc/platforms/pseries/smp.c
7  *
8  * Peter Bergner, IBM   March 2001.
9  * Copyright (C) 2001 IBM.
10  * Dave Engebretsen, Peter Bergner, and
11  * Mike Corrigan {engebret|bergner|mikec}@us.ibm.com
12  * Plus various changes from other IBM teams...
13  *
14  * Copyright (C) 2006 Michael Ellerman, IBM Corporation
15  */
16
17 #define pr_fmt(fmt)     "pseries-hotplug-cpu: " fmt
18
19 #include <linux/kernel.h>
20 #include <linux/interrupt.h>
21 #include <linux/delay.h>
22 #include <linux/sched.h>        /* for idle_task_exit */
23 #include <linux/sched/hotplug.h>
24 #include <linux/cpu.h>
25 #include <linux/of.h>
26 #include <linux/slab.h>
27 #include <asm/prom.h>
28 #include <asm/rtas.h>
29 #include <asm/firmware.h>
30 #include <asm/machdep.h>
31 #include <asm/vdso_datapage.h>
32 #include <asm/xics.h>
33 #include <asm/xive.h>
34 #include <asm/plpar_wrappers.h>
35 #include <asm/topology.h>
36
37 #include "pseries.h"
38 #include "offline_states.h"
39
40 /* This version can't take the spinlock, because it never returns */
41 static int rtas_stop_self_token = RTAS_UNKNOWN_SERVICE;
42
43 static DEFINE_PER_CPU(enum cpu_state_vals, preferred_offline_state) =
44                                                         CPU_STATE_OFFLINE;
45 static DEFINE_PER_CPU(enum cpu_state_vals, current_state) = CPU_STATE_OFFLINE;
46
47 static enum cpu_state_vals default_offline_state = CPU_STATE_OFFLINE;
48
49 static bool cede_offline_enabled __read_mostly = true;
50
51 /*
52  * Enable/disable cede_offline when available.
53  */
54 static int __init setup_cede_offline(char *str)
55 {
56         return (kstrtobool(str, &cede_offline_enabled) == 0);
57 }
58
59 __setup("cede_offline=", setup_cede_offline);
60
61 enum cpu_state_vals get_cpu_current_state(int cpu)
62 {
63         return per_cpu(current_state, cpu);
64 }
65
66 void set_cpu_current_state(int cpu, enum cpu_state_vals state)
67 {
68         per_cpu(current_state, cpu) = state;
69 }
70
71 enum cpu_state_vals get_preferred_offline_state(int cpu)
72 {
73         return per_cpu(preferred_offline_state, cpu);
74 }
75
76 void set_preferred_offline_state(int cpu, enum cpu_state_vals state)
77 {
78         per_cpu(preferred_offline_state, cpu) = state;
79 }
80
81 void set_default_offline_state(int cpu)
82 {
83         per_cpu(preferred_offline_state, cpu) = default_offline_state;
84 }
85
86 static void rtas_stop_self(void)
87 {
88         static struct rtas_args args;
89
90         local_irq_disable();
91
92         BUG_ON(rtas_stop_self_token == RTAS_UNKNOWN_SERVICE);
93
94         rtas_call_unlocked(&args, rtas_stop_self_token, 0, 1, NULL);
95
96         panic("Alas, I survived.\n");
97 }
98
99 static void pseries_mach_cpu_die(void)
100 {
101         unsigned int cpu = smp_processor_id();
102         unsigned int hwcpu = hard_smp_processor_id();
103         u8 cede_latency_hint = 0;
104
105         local_irq_disable();
106         idle_task_exit();
107         if (xive_enabled())
108                 xive_teardown_cpu();
109         else
110                 xics_teardown_cpu();
111
112         if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
113                 set_cpu_current_state(cpu, CPU_STATE_INACTIVE);
114                 if (ppc_md.suspend_disable_cpu)
115                         ppc_md.suspend_disable_cpu();
116
117                 cede_latency_hint = 2;
118
119                 get_lppaca()->idle = 1;
120                 if (!lppaca_shared_proc(get_lppaca()))
121                         get_lppaca()->donate_dedicated_cpu = 1;
122
123                 while (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
124                         while (!prep_irq_for_idle()) {
125                                 local_irq_enable();
126                                 local_irq_disable();
127                         }
128
129                         extended_cede_processor(cede_latency_hint);
130                 }
131
132                 local_irq_disable();
133
134                 if (!lppaca_shared_proc(get_lppaca()))
135                         get_lppaca()->donate_dedicated_cpu = 0;
136                 get_lppaca()->idle = 0;
137
138                 if (get_preferred_offline_state(cpu) == CPU_STATE_ONLINE) {
139                         unregister_slb_shadow(hwcpu);
140
141                         hard_irq_disable();
142                         /*
143                          * Call to start_secondary_resume() will not return.
144                          * Kernel stack will be reset and start_secondary()
145                          * will be called to continue the online operation.
146                          */
147                         start_secondary_resume();
148                 }
149         }
150
151         /* Requested state is CPU_STATE_OFFLINE at this point */
152         WARN_ON(get_preferred_offline_state(cpu) != CPU_STATE_OFFLINE);
153
154         set_cpu_current_state(cpu, CPU_STATE_OFFLINE);
155         unregister_slb_shadow(hwcpu);
156         rtas_stop_self();
157
158         /* Should never get here... */
159         BUG();
160         for(;;);
161 }
162
163 static int pseries_cpu_disable(void)
164 {
165         int cpu = smp_processor_id();
166
167         set_cpu_online(cpu, false);
168         vdso_data->processorCount--;
169
170         /*fix boot_cpuid here*/
171         if (cpu == boot_cpuid)
172                 boot_cpuid = cpumask_any(cpu_online_mask);
173
174         /* FIXME: abstract this to not be platform specific later on */
175         if (xive_enabled())
176                 xive_smp_disable_cpu();
177         else
178                 xics_migrate_irqs_away();
179         return 0;
180 }
181
182 /*
183  * pseries_cpu_die: Wait for the cpu to die.
184  * @cpu: logical processor id of the CPU whose death we're awaiting.
185  *
186  * This function is called from the context of the thread which is performing
187  * the cpu-offline. Here we wait for long enough to allow the cpu in question
188  * to self-destroy so that the cpu-offline thread can send the CPU_DEAD
189  * notifications.
190  *
191  * OTOH, pseries_mach_cpu_die() is called by the @cpu when it wants to
192  * self-destruct.
193  */
194 static void pseries_cpu_die(unsigned int cpu)
195 {
196         int tries;
197         int cpu_status = 1;
198         unsigned int pcpu = get_hard_smp_processor_id(cpu);
199
200         if (get_preferred_offline_state(cpu) == CPU_STATE_INACTIVE) {
201                 cpu_status = 1;
202                 for (tries = 0; tries < 5000; tries++) {
203                         if (get_cpu_current_state(cpu) == CPU_STATE_INACTIVE) {
204                                 cpu_status = 0;
205                                 break;
206                         }
207                         msleep(1);
208                 }
209         } else if (get_preferred_offline_state(cpu) == CPU_STATE_OFFLINE) {
210
211                 for (tries = 0; tries < 25; tries++) {
212                         cpu_status = smp_query_cpu_stopped(pcpu);
213                         if (cpu_status == QCSS_STOPPED ||
214                             cpu_status == QCSS_HARDWARE_ERROR)
215                                 break;
216                         cpu_relax();
217                 }
218         }
219
220         if (cpu_status != 0) {
221                 printk("Querying DEAD? cpu %i (%i) shows %i\n",
222                        cpu, pcpu, cpu_status);
223         }
224
225         /* Isolation and deallocation are definitely done by
226          * drslot_chrp_cpu.  If they were not they would be
227          * done here.  Change isolate state to Isolate and
228          * change allocation-state to Unusable.
229          */
230         paca_ptrs[cpu]->cpu_start = 0;
231 }
232
233 /*
234  * Update cpu_present_mask and paca(s) for a new cpu node.  The wrinkle
235  * here is that a cpu device node may represent up to two logical cpus
236  * in the SMT case.  We must honor the assumption in other code that
237  * the logical ids for sibling SMT threads x and y are adjacent, such
238  * that x^1 == y and y^1 == x.
239  */
240 static int pseries_add_processor(struct device_node *np)
241 {
242         unsigned int cpu;
243         cpumask_var_t candidate_mask, tmp;
244         int err = -ENOSPC, len, nthreads, i;
245         const __be32 *intserv;
246
247         intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
248         if (!intserv)
249                 return 0;
250
251         zalloc_cpumask_var(&candidate_mask, GFP_KERNEL);
252         zalloc_cpumask_var(&tmp, GFP_KERNEL);
253
254         nthreads = len / sizeof(u32);
255         for (i = 0; i < nthreads; i++)
256                 cpumask_set_cpu(i, tmp);
257
258         cpu_maps_update_begin();
259
260         BUG_ON(!cpumask_subset(cpu_present_mask, cpu_possible_mask));
261
262         /* Get a bitmap of unoccupied slots. */
263         cpumask_xor(candidate_mask, cpu_possible_mask, cpu_present_mask);
264         if (cpumask_empty(candidate_mask)) {
265                 /* If we get here, it most likely means that NR_CPUS is
266                  * less than the partition's max processors setting.
267                  */
268                 printk(KERN_ERR "Cannot add cpu %pOF; this system configuration"
269                        " supports %d logical cpus.\n", np,
270                        num_possible_cpus());
271                 goto out_unlock;
272         }
273
274         while (!cpumask_empty(tmp))
275                 if (cpumask_subset(tmp, candidate_mask))
276                         /* Found a range where we can insert the new cpu(s) */
277                         break;
278                 else
279                         cpumask_shift_left(tmp, tmp, nthreads);
280
281         if (cpumask_empty(tmp)) {
282                 printk(KERN_ERR "Unable to find space in cpu_present_mask for"
283                        " processor %pOFn with %d thread(s)\n", np,
284                        nthreads);
285                 goto out_unlock;
286         }
287
288         for_each_cpu(cpu, tmp) {
289                 BUG_ON(cpu_present(cpu));
290                 set_cpu_present(cpu, true);
291                 set_hard_smp_processor_id(cpu, be32_to_cpu(*intserv++));
292         }
293         err = 0;
294 out_unlock:
295         cpu_maps_update_done();
296         free_cpumask_var(candidate_mask);
297         free_cpumask_var(tmp);
298         return err;
299 }
300
301 /*
302  * Update the present map for a cpu node which is going away, and set
303  * the hard id in the paca(s) to -1 to be consistent with boot time
304  * convention for non-present cpus.
305  */
306 static void pseries_remove_processor(struct device_node *np)
307 {
308         unsigned int cpu;
309         int len, nthreads, i;
310         const __be32 *intserv;
311         u32 thread;
312
313         intserv = of_get_property(np, "ibm,ppc-interrupt-server#s", &len);
314         if (!intserv)
315                 return;
316
317         nthreads = len / sizeof(u32);
318
319         cpu_maps_update_begin();
320         for (i = 0; i < nthreads; i++) {
321                 thread = be32_to_cpu(intserv[i]);
322                 for_each_present_cpu(cpu) {
323                         if (get_hard_smp_processor_id(cpu) != thread)
324                                 continue;
325                         BUG_ON(cpu_online(cpu));
326                         set_cpu_present(cpu, false);
327                         set_hard_smp_processor_id(cpu, -1);
328                         update_numa_cpu_lookup_table(cpu, -1);
329                         break;
330                 }
331                 if (cpu >= nr_cpu_ids)
332                         printk(KERN_WARNING "Could not find cpu to remove "
333                                "with physical id 0x%x\n", thread);
334         }
335         cpu_maps_update_done();
336 }
337
338 static int dlpar_online_cpu(struct device_node *dn)
339 {
340         int rc = 0;
341         unsigned int cpu;
342         int len, nthreads, i;
343         const __be32 *intserv;
344         u32 thread;
345
346         intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
347         if (!intserv)
348                 return -EINVAL;
349
350         nthreads = len / sizeof(u32);
351
352         cpu_maps_update_begin();
353         for (i = 0; i < nthreads; i++) {
354                 thread = be32_to_cpu(intserv[i]);
355                 for_each_present_cpu(cpu) {
356                         if (get_hard_smp_processor_id(cpu) != thread)
357                                 continue;
358                         BUG_ON(get_cpu_current_state(cpu)
359                                         != CPU_STATE_OFFLINE);
360                         cpu_maps_update_done();
361                         timed_topology_update(1);
362                         find_and_online_cpu_nid(cpu);
363                         rc = device_online(get_cpu_device(cpu));
364                         if (rc)
365                                 goto out;
366                         cpu_maps_update_begin();
367
368                         break;
369                 }
370                 if (cpu == num_possible_cpus())
371                         printk(KERN_WARNING "Could not find cpu to online "
372                                "with physical id 0x%x\n", thread);
373         }
374         cpu_maps_update_done();
375
376 out:
377         return rc;
378
379 }
380
381 static bool dlpar_cpu_exists(struct device_node *parent, u32 drc_index)
382 {
383         struct device_node *child = NULL;
384         u32 my_drc_index;
385         bool found;
386         int rc;
387
388         /* Assume cpu doesn't exist */
389         found = false;
390
391         for_each_child_of_node(parent, child) {
392                 rc = of_property_read_u32(child, "ibm,my-drc-index",
393                                           &my_drc_index);
394                 if (rc)
395                         continue;
396
397                 if (my_drc_index == drc_index) {
398                         of_node_put(child);
399                         found = true;
400                         break;
401                 }
402         }
403
404         return found;
405 }
406
407 static bool valid_cpu_drc_index(struct device_node *parent, u32 drc_index)
408 {
409         bool found = false;
410         int rc, index;
411
412         index = 0;
413         while (!found) {
414                 u32 drc;
415
416                 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
417                                                 index++, &drc);
418                 if (rc)
419                         break;
420
421                 if (drc == drc_index)
422                         found = true;
423         }
424
425         return found;
426 }
427
428 static ssize_t dlpar_cpu_add(u32 drc_index)
429 {
430         struct device_node *dn, *parent;
431         int rc, saved_rc;
432
433         pr_debug("Attempting to add CPU, drc index: %x\n", drc_index);
434
435         parent = of_find_node_by_path("/cpus");
436         if (!parent) {
437                 pr_warn("Failed to find CPU root node \"/cpus\"\n");
438                 return -ENODEV;
439         }
440
441         if (dlpar_cpu_exists(parent, drc_index)) {
442                 of_node_put(parent);
443                 pr_warn("CPU with drc index %x already exists\n", drc_index);
444                 return -EINVAL;
445         }
446
447         if (!valid_cpu_drc_index(parent, drc_index)) {
448                 of_node_put(parent);
449                 pr_warn("Cannot find CPU (drc index %x) to add.\n", drc_index);
450                 return -EINVAL;
451         }
452
453         rc = dlpar_acquire_drc(drc_index);
454         if (rc) {
455                 pr_warn("Failed to acquire DRC, rc: %d, drc index: %x\n",
456                         rc, drc_index);
457                 of_node_put(parent);
458                 return -EINVAL;
459         }
460
461         dn = dlpar_configure_connector(cpu_to_be32(drc_index), parent);
462         if (!dn) {
463                 pr_warn("Failed call to configure-connector, drc index: %x\n",
464                         drc_index);
465                 dlpar_release_drc(drc_index);
466                 of_node_put(parent);
467                 return -EINVAL;
468         }
469
470         rc = dlpar_attach_node(dn, parent);
471
472         /* Regardless we are done with parent now */
473         of_node_put(parent);
474
475         if (rc) {
476                 saved_rc = rc;
477                 pr_warn("Failed to attach node %pOFn, rc: %d, drc index: %x\n",
478                         dn, rc, drc_index);
479
480                 rc = dlpar_release_drc(drc_index);
481                 if (!rc)
482                         dlpar_free_cc_nodes(dn);
483
484                 return saved_rc;
485         }
486
487         rc = dlpar_online_cpu(dn);
488         if (rc) {
489                 saved_rc = rc;
490                 pr_warn("Failed to online cpu %pOFn, rc: %d, drc index: %x\n",
491                         dn, rc, drc_index);
492
493                 rc = dlpar_detach_node(dn);
494                 if (!rc)
495                         dlpar_release_drc(drc_index);
496
497                 return saved_rc;
498         }
499
500         pr_debug("Successfully added CPU %pOFn, drc index: %x\n", dn,
501                  drc_index);
502         return rc;
503 }
504
505 static int dlpar_offline_cpu(struct device_node *dn)
506 {
507         int rc = 0;
508         unsigned int cpu;
509         int len, nthreads, i;
510         const __be32 *intserv;
511         u32 thread;
512
513         intserv = of_get_property(dn, "ibm,ppc-interrupt-server#s", &len);
514         if (!intserv)
515                 return -EINVAL;
516
517         nthreads = len / sizeof(u32);
518
519         cpu_maps_update_begin();
520         for (i = 0; i < nthreads; i++) {
521                 thread = be32_to_cpu(intserv[i]);
522                 for_each_present_cpu(cpu) {
523                         if (get_hard_smp_processor_id(cpu) != thread)
524                                 continue;
525
526                         if (get_cpu_current_state(cpu) == CPU_STATE_OFFLINE)
527                                 break;
528
529                         if (get_cpu_current_state(cpu) == CPU_STATE_ONLINE) {
530                                 set_preferred_offline_state(cpu,
531                                                             CPU_STATE_OFFLINE);
532                                 cpu_maps_update_done();
533                                 timed_topology_update(1);
534                                 rc = device_offline(get_cpu_device(cpu));
535                                 if (rc)
536                                         goto out;
537                                 cpu_maps_update_begin();
538                                 break;
539
540                         }
541
542                         /*
543                          * The cpu is in CPU_STATE_INACTIVE.
544                          * Upgrade it's state to CPU_STATE_OFFLINE.
545                          */
546                         set_preferred_offline_state(cpu, CPU_STATE_OFFLINE);
547                         BUG_ON(plpar_hcall_norets(H_PROD, thread)
548                                                                 != H_SUCCESS);
549                         __cpu_die(cpu);
550                         break;
551                 }
552                 if (cpu == num_possible_cpus())
553                         printk(KERN_WARNING "Could not find cpu to offline with physical id 0x%x\n", thread);
554         }
555         cpu_maps_update_done();
556
557 out:
558         return rc;
559
560 }
561
562 static ssize_t dlpar_cpu_remove(struct device_node *dn, u32 drc_index)
563 {
564         int rc;
565
566         pr_debug("Attempting to remove CPU %pOFn, drc index: %x\n",
567                  dn, drc_index);
568
569         rc = dlpar_offline_cpu(dn);
570         if (rc) {
571                 pr_warn("Failed to offline CPU %pOFn, rc: %d\n", dn, rc);
572                 return -EINVAL;
573         }
574
575         rc = dlpar_release_drc(drc_index);
576         if (rc) {
577                 pr_warn("Failed to release drc (%x) for CPU %pOFn, rc: %d\n",
578                         drc_index, dn, rc);
579                 dlpar_online_cpu(dn);
580                 return rc;
581         }
582
583         rc = dlpar_detach_node(dn);
584         if (rc) {
585                 int saved_rc = rc;
586
587                 pr_warn("Failed to detach CPU %pOFn, rc: %d", dn, rc);
588
589                 rc = dlpar_acquire_drc(drc_index);
590                 if (!rc)
591                         dlpar_online_cpu(dn);
592
593                 return saved_rc;
594         }
595
596         pr_debug("Successfully removed CPU, drc index: %x\n", drc_index);
597         return 0;
598 }
599
600 static struct device_node *cpu_drc_index_to_dn(u32 drc_index)
601 {
602         struct device_node *dn;
603         u32 my_index;
604         int rc;
605
606         for_each_node_by_type(dn, "cpu") {
607                 rc = of_property_read_u32(dn, "ibm,my-drc-index", &my_index);
608                 if (rc)
609                         continue;
610
611                 if (my_index == drc_index)
612                         break;
613         }
614
615         return dn;
616 }
617
618 static int dlpar_cpu_remove_by_index(u32 drc_index)
619 {
620         struct device_node *dn;
621         int rc;
622
623         dn = cpu_drc_index_to_dn(drc_index);
624         if (!dn) {
625                 pr_warn("Cannot find CPU (drc index %x) to remove\n",
626                         drc_index);
627                 return -ENODEV;
628         }
629
630         rc = dlpar_cpu_remove(dn, drc_index);
631         of_node_put(dn);
632         return rc;
633 }
634
635 static int find_dlpar_cpus_to_remove(u32 *cpu_drcs, int cpus_to_remove)
636 {
637         struct device_node *dn;
638         int cpus_found = 0;
639         int rc;
640
641         /* We want to find cpus_to_remove + 1 CPUs to ensure we do not
642          * remove the last CPU.
643          */
644         for_each_node_by_type(dn, "cpu") {
645                 cpus_found++;
646
647                 if (cpus_found > cpus_to_remove) {
648                         of_node_put(dn);
649                         break;
650                 }
651
652                 /* Note that cpus_found is always 1 ahead of the index
653                  * into the cpu_drcs array, so we use cpus_found - 1
654                  */
655                 rc = of_property_read_u32(dn, "ibm,my-drc-index",
656                                           &cpu_drcs[cpus_found - 1]);
657                 if (rc) {
658                         pr_warn("Error occurred getting drc-index for %pOFn\n",
659                                 dn);
660                         of_node_put(dn);
661                         return -1;
662                 }
663         }
664
665         if (cpus_found < cpus_to_remove) {
666                 pr_warn("Failed to find enough CPUs (%d of %d) to remove\n",
667                         cpus_found, cpus_to_remove);
668         } else if (cpus_found == cpus_to_remove) {
669                 pr_warn("Cannot remove all CPUs\n");
670         }
671
672         return cpus_found;
673 }
674
675 static int dlpar_cpu_remove_by_count(u32 cpus_to_remove)
676 {
677         u32 *cpu_drcs;
678         int cpus_found;
679         int cpus_removed = 0;
680         int i, rc;
681
682         pr_debug("Attempting to hot-remove %d CPUs\n", cpus_to_remove);
683
684         cpu_drcs = kcalloc(cpus_to_remove, sizeof(*cpu_drcs), GFP_KERNEL);
685         if (!cpu_drcs)
686                 return -EINVAL;
687
688         cpus_found = find_dlpar_cpus_to_remove(cpu_drcs, cpus_to_remove);
689         if (cpus_found <= cpus_to_remove) {
690                 kfree(cpu_drcs);
691                 return -EINVAL;
692         }
693
694         for (i = 0; i < cpus_to_remove; i++) {
695                 rc = dlpar_cpu_remove_by_index(cpu_drcs[i]);
696                 if (rc)
697                         break;
698
699                 cpus_removed++;
700         }
701
702         if (cpus_removed != cpus_to_remove) {
703                 pr_warn("CPU hot-remove failed, adding back removed CPUs\n");
704
705                 for (i = 0; i < cpus_removed; i++)
706                         dlpar_cpu_add(cpu_drcs[i]);
707
708                 rc = -EINVAL;
709         } else {
710                 rc = 0;
711         }
712
713         kfree(cpu_drcs);
714         return rc;
715 }
716
717 static int find_dlpar_cpus_to_add(u32 *cpu_drcs, u32 cpus_to_add)
718 {
719         struct device_node *parent;
720         int cpus_found = 0;
721         int index, rc;
722
723         parent = of_find_node_by_path("/cpus");
724         if (!parent) {
725                 pr_warn("Could not find CPU root node in device tree\n");
726                 kfree(cpu_drcs);
727                 return -1;
728         }
729
730         /* Search the ibm,drc-indexes array for possible CPU drcs to
731          * add. Note that the format of the ibm,drc-indexes array is
732          * the number of entries in the array followed by the array
733          * of drc values so we start looking at index = 1.
734          */
735         index = 1;
736         while (cpus_found < cpus_to_add) {
737                 u32 drc;
738
739                 rc = of_property_read_u32_index(parent, "ibm,drc-indexes",
740                                                 index++, &drc);
741                 if (rc)
742                         break;
743
744                 if (dlpar_cpu_exists(parent, drc))
745                         continue;
746
747                 cpu_drcs[cpus_found++] = drc;
748         }
749
750         of_node_put(parent);
751         return cpus_found;
752 }
753
754 static int dlpar_cpu_add_by_count(u32 cpus_to_add)
755 {
756         u32 *cpu_drcs;
757         int cpus_added = 0;
758         int cpus_found;
759         int i, rc;
760
761         pr_debug("Attempting to hot-add %d CPUs\n", cpus_to_add);
762
763         cpu_drcs = kcalloc(cpus_to_add, sizeof(*cpu_drcs), GFP_KERNEL);
764         if (!cpu_drcs)
765                 return -EINVAL;
766
767         cpus_found = find_dlpar_cpus_to_add(cpu_drcs, cpus_to_add);
768         if (cpus_found < cpus_to_add) {
769                 pr_warn("Failed to find enough CPUs (%d of %d) to add\n",
770                         cpus_found, cpus_to_add);
771                 kfree(cpu_drcs);
772                 return -EINVAL;
773         }
774
775         for (i = 0; i < cpus_to_add; i++) {
776                 rc = dlpar_cpu_add(cpu_drcs[i]);
777                 if (rc)
778                         break;
779
780                 cpus_added++;
781         }
782
783         if (cpus_added < cpus_to_add) {
784                 pr_warn("CPU hot-add failed, removing any added CPUs\n");
785
786                 for (i = 0; i < cpus_added; i++)
787                         dlpar_cpu_remove_by_index(cpu_drcs[i]);
788
789                 rc = -EINVAL;
790         } else {
791                 rc = 0;
792         }
793
794         kfree(cpu_drcs);
795         return rc;
796 }
797
798 int dlpar_cpu_readd(int cpu)
799 {
800         struct device_node *dn;
801         struct device *dev;
802         u32 drc_index;
803         int rc;
804
805         dev = get_cpu_device(cpu);
806         dn = dev->of_node;
807
808         rc = of_property_read_u32(dn, "ibm,my-drc-index", &drc_index);
809
810         rc = dlpar_cpu_remove_by_index(drc_index);
811         if (!rc)
812                 rc = dlpar_cpu_add(drc_index);
813
814         return rc;
815 }
816
817 int dlpar_cpu(struct pseries_hp_errorlog *hp_elog)
818 {
819         u32 count, drc_index;
820         int rc;
821
822         count = hp_elog->_drc_u.drc_count;
823         drc_index = hp_elog->_drc_u.drc_index;
824
825         lock_device_hotplug();
826
827         switch (hp_elog->action) {
828         case PSERIES_HP_ELOG_ACTION_REMOVE:
829                 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT)
830                         rc = dlpar_cpu_remove_by_count(count);
831                 else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
832                         rc = dlpar_cpu_remove_by_index(drc_index);
833                 else
834                         rc = -EINVAL;
835                 break;
836         case PSERIES_HP_ELOG_ACTION_ADD:
837                 if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_COUNT)
838                         rc = dlpar_cpu_add_by_count(count);
839                 else if (hp_elog->id_type == PSERIES_HP_ELOG_ID_DRC_INDEX)
840                         rc = dlpar_cpu_add(drc_index);
841                 else
842                         rc = -EINVAL;
843                 break;
844         default:
845                 pr_err("Invalid action (%d) specified\n", hp_elog->action);
846                 rc = -EINVAL;
847                 break;
848         }
849
850         unlock_device_hotplug();
851         return rc;
852 }
853
854 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
855
856 static ssize_t dlpar_cpu_probe(const char *buf, size_t count)
857 {
858         u32 drc_index;
859         int rc;
860
861         rc = kstrtou32(buf, 0, &drc_index);
862         if (rc)
863                 return -EINVAL;
864
865         rc = dlpar_cpu_add(drc_index);
866
867         return rc ? rc : count;
868 }
869
870 static ssize_t dlpar_cpu_release(const char *buf, size_t count)
871 {
872         struct device_node *dn;
873         u32 drc_index;
874         int rc;
875
876         dn = of_find_node_by_path(buf);
877         if (!dn)
878                 return -EINVAL;
879
880         rc = of_property_read_u32(dn, "ibm,my-drc-index", &drc_index);
881         if (rc) {
882                 of_node_put(dn);
883                 return -EINVAL;
884         }
885
886         rc = dlpar_cpu_remove(dn, drc_index);
887         of_node_put(dn);
888
889         return rc ? rc : count;
890 }
891
892 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
893
894 static int pseries_smp_notifier(struct notifier_block *nb,
895                                 unsigned long action, void *data)
896 {
897         struct of_reconfig_data *rd = data;
898         int err = 0;
899
900         switch (action) {
901         case OF_RECONFIG_ATTACH_NODE:
902                 err = pseries_add_processor(rd->dn);
903                 break;
904         case OF_RECONFIG_DETACH_NODE:
905                 pseries_remove_processor(rd->dn);
906                 break;
907         }
908         return notifier_from_errno(err);
909 }
910
911 static struct notifier_block pseries_smp_nb = {
912         .notifier_call = pseries_smp_notifier,
913 };
914
915 #define MAX_CEDE_LATENCY_LEVELS         4
916 #define CEDE_LATENCY_PARAM_LENGTH       10
917 #define CEDE_LATENCY_PARAM_MAX_LENGTH   \
918         (MAX_CEDE_LATENCY_LEVELS * CEDE_LATENCY_PARAM_LENGTH * sizeof(char))
919 #define CEDE_LATENCY_TOKEN              45
920
921 static char cede_parameters[CEDE_LATENCY_PARAM_MAX_LENGTH];
922
923 static int parse_cede_parameters(void)
924 {
925         memset(cede_parameters, 0, CEDE_LATENCY_PARAM_MAX_LENGTH);
926         return rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
927                          NULL,
928                          CEDE_LATENCY_TOKEN,
929                          __pa(cede_parameters),
930                          CEDE_LATENCY_PARAM_MAX_LENGTH);
931 }
932
933 static int __init pseries_cpu_hotplug_init(void)
934 {
935         int cpu;
936         int qcss_tok;
937
938 #ifdef CONFIG_ARCH_CPU_PROBE_RELEASE
939         ppc_md.cpu_probe = dlpar_cpu_probe;
940         ppc_md.cpu_release = dlpar_cpu_release;
941 #endif /* CONFIG_ARCH_CPU_PROBE_RELEASE */
942
943         rtas_stop_self_token = rtas_token("stop-self");
944         qcss_tok = rtas_token("query-cpu-stopped-state");
945
946         if (rtas_stop_self_token == RTAS_UNKNOWN_SERVICE ||
947                         qcss_tok == RTAS_UNKNOWN_SERVICE) {
948                 printk(KERN_INFO "CPU Hotplug not supported by firmware "
949                                 "- disabling.\n");
950                 return 0;
951         }
952
953         ppc_md.cpu_die = pseries_mach_cpu_die;
954         smp_ops->cpu_disable = pseries_cpu_disable;
955         smp_ops->cpu_die = pseries_cpu_die;
956
957         /* Processors can be added/removed only on LPAR */
958         if (firmware_has_feature(FW_FEATURE_LPAR)) {
959                 of_reconfig_notifier_register(&pseries_smp_nb);
960                 cpu_maps_update_begin();
961                 if (cede_offline_enabled && parse_cede_parameters() == 0) {
962                         default_offline_state = CPU_STATE_INACTIVE;
963                         for_each_online_cpu(cpu)
964                                 set_default_offline_state(cpu);
965                 }
966                 cpu_maps_update_done();
967         }
968
969         return 0;
970 }
971 machine_arch_initcall(pseries, pseries_cpu_hotplug_init);