1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * pSeries_lpar.c
4 * Copyright (C) 2001 Todd Inglett, IBM Corporation
5 *
6 * pSeries LPAR support.
7 */
8
9/* Enables debugging of low-level hash table routines - careful! */
10#undef DEBUG
11#define pr_fmt(fmt) "lpar: " fmt
12
13#include <linux/kernel.h>
14#include <linux/dma-mapping.h>
15#include <linux/console.h>
16#include <linux/export.h>
17#include <linux/jump_label.h>
18#include <linux/delay.h>
19#include <linux/seq_file.h>
20#include <linux/stop_machine.h>
21#include <linux/spinlock.h>
22#include <linux/cpuhotplug.h>
23#include <linux/workqueue.h>
24#include <linux/proc_fs.h>
25#include <linux/pgtable.h>
26#include <linux/debugfs.h>
27
28#include <asm/processor.h>
29#include <asm/mmu.h>
30#include <asm/page.h>
31#include <asm/setup.h>
32#include <asm/mmu_context.h>
33#include <asm/iommu.h>
34#include <asm/tlb.h>
35#include <asm/cputable.h>
36#include <asm/papr-sysparm.h>
37#include <asm/udbg.h>
38#include <asm/smp.h>
39#include <asm/trace.h>
40#include <asm/firmware.h>
41#include <asm/plpar_wrappers.h>
42#include <asm/kexec.h>
43#include <asm/fadump.h>
44#include <asm/dtl.h>
45#include <asm/vphn.h>
46
47#include "pseries.h"
48
49/* Flag bits for H_BULK_REMOVE */
50#define HBR_REQUEST 0x4000000000000000UL
51#define HBR_RESPONSE 0x8000000000000000UL
52#define HBR_END 0xc000000000000000UL
53#define HBR_AVPN 0x0200000000000000UL
54#define HBR_ANDCOND 0x0100000000000000UL
55
56
57/* in hvCall.S */
58EXPORT_SYMBOL(plpar_hcall);
59EXPORT_SYMBOL(plpar_hcall9);
60EXPORT_SYMBOL(plpar_hcall_norets);
61
62#ifdef CONFIG_PPC_64S_HASH_MMU
63/*
64 * H_BLOCK_REMOVE supported block size for this page size in segment who's base
65 * page size is that page size.
66 *
67 * The first index is the segment base page size, the second one is the actual
68 * page size.
69 */
70static int hblkrm_size[MMU_PAGE_COUNT][MMU_PAGE_COUNT] __ro_after_init;
71#endif
72
73/*
74 * Due to the involved complexity, and that the current hypervisor is only
75 * returning this value or 0, we are limiting the support of the H_BLOCK_REMOVE
76 * buffer size to 8 size block.
77 */
78#define HBLKRM_SUPPORTED_BLOCK_SIZE 8
79
80#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
81static u8 dtl_mask = DTL_LOG_PREEMPT;
82#else
83static u8 dtl_mask;
84#endif
85
86void alloc_dtl_buffers(unsigned long *time_limit)
87{
88 int cpu;
89 struct paca_struct *pp;
90 struct dtl_entry *dtl;
91
92 for_each_possible_cpu(cpu) {
93 pp = paca_ptrs[cpu];
94 if (pp->dispatch_log)
95 continue;
96 dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
97 if (!dtl) {
98 pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
99 cpu);
100#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
101 pr_warn("Stolen time statistics will be unreliable\n");
102#endif
103 break;
104 }
105
106 pp->dtl_ridx = 0;
107 pp->dispatch_log = dtl;
108 pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
109 pp->dtl_curr = dtl;
110
111 if (time_limit && time_after(jiffies, *time_limit)) {
112 cond_resched();
113 *time_limit = jiffies + HZ;
114 }
115 }
116}
117
118void register_dtl_buffer(int cpu)
119{
120 long ret;
121 struct paca_struct *pp;
122 struct dtl_entry *dtl;
123 int hwcpu = get_hard_smp_processor_id(cpu);
124
125 pp = paca_ptrs[cpu];
126 dtl = pp->dispatch_log;
127 if (dtl && dtl_mask) {
128 pp->dtl_ridx = 0;
129 pp->dtl_curr = dtl;
130 lppaca_of(cpu).dtl_idx = 0;
131
132 /* hypervisor reads buffer length from this field */
133 dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
134 ret = register_dtl(hwcpu, __pa(dtl));
135 if (ret)
136 pr_err("WARNING: DTL registration of cpu %d (hw %d) failed with %ld\n",
137 cpu, hwcpu, ret);
138
139 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
140 }
141}
142
143#ifdef CONFIG_PPC_SPLPAR
144struct dtl_worker {
145 struct delayed_work work;
146 int cpu;
147};
148
149struct vcpu_dispatch_data {
150 int last_disp_cpu;
151
152 int total_disp;
153
154 int same_cpu_disp;
155 int same_chip_disp;
156 int diff_chip_disp;
157 int far_chip_disp;
158
159 int numa_home_disp;
160 int numa_remote_disp;
161 int numa_far_disp;
162};
163
164/*
165 * This represents the number of cpus in the hypervisor. Since there is no
166 * architected way to discover the number of processors in the host, we
167 * provision for dealing with NR_CPUS. This is currently 2048 by default, and
168 * is sufficient for our purposes. This will need to be tweaked if
169 * CONFIG_NR_CPUS is changed.
170 */
171#define NR_CPUS_H NR_CPUS
172
173DECLARE_RWSEM(dtl_access_lock);
174static DEFINE_PER_CPU(struct vcpu_dispatch_data, vcpu_disp_data);
175static DEFINE_PER_CPU(u64, dtl_entry_ridx);
176static DEFINE_PER_CPU(struct dtl_worker, dtl_workers);
177static enum cpuhp_state dtl_worker_state;
178static DEFINE_MUTEX(dtl_enable_mutex);
179static int vcpudispatch_stats_on __read_mostly;
180static int vcpudispatch_stats_freq = 50;
181static __be32 *vcpu_associativity, *pcpu_associativity;
182
183
184static void free_dtl_buffers(unsigned long *time_limit)
185{
186#ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
187 int cpu;
188 struct paca_struct *pp;
189
190 for_each_possible_cpu(cpu) {
191 pp = paca_ptrs[cpu];
192 if (!pp->dispatch_log)
193 continue;
194 kmem_cache_free(dtl_cache, pp->dispatch_log);
195 pp->dtl_ridx = 0;
196 pp->dispatch_log = NULL;
197 pp->dispatch_log_end = NULL;
198 pp->dtl_curr = NULL;
199
200 if (time_limit && time_after(jiffies, *time_limit)) {
201 cond_resched();
202 *time_limit = jiffies + HZ;
203 }
204 }
205#endif
206}
207
208static int init_cpu_associativity(void)
209{
210 vcpu_associativity = kcalloc(num_possible_cpus() / threads_per_core,
211 VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
212 pcpu_associativity = kcalloc(NR_CPUS_H / threads_per_core,
213 VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
214
215 if (!vcpu_associativity || !pcpu_associativity) {
216 pr_err("error allocating memory for associativity information\n");
217 return -ENOMEM;
218 }
219
220 return 0;
221}
222
223static void destroy_cpu_associativity(void)
224{
225 kfree(vcpu_associativity);
226 kfree(pcpu_associativity);
227 vcpu_associativity = pcpu_associativity = NULL;
228}
229
230static __be32 *__get_cpu_associativity(int cpu, __be32 *cpu_assoc, int flag)
231{
232 __be32 *assoc;
233 int rc = 0;
234
235 assoc = &cpu_assoc[(int)(cpu / threads_per_core) * VPHN_ASSOC_BUFSIZE];
236 if (!assoc[0]) {
237 rc = hcall_vphn(cpu, flag, &assoc[0]);
238 if (rc)
239 return NULL;
240 }
241
242 return assoc;
243}
244
245static __be32 *get_pcpu_associativity(int cpu)
246{
247 return __get_cpu_associativity(cpu, pcpu_associativity, VPHN_FLAG_PCPU);
248}
249
250static __be32 *get_vcpu_associativity(int cpu)
251{
252 return __get_cpu_associativity(cpu, vcpu_associativity, VPHN_FLAG_VCPU);
253}
254
255static int cpu_relative_dispatch_distance(int last_disp_cpu, int cur_disp_cpu)
256{
257 __be32 *last_disp_cpu_assoc, *cur_disp_cpu_assoc;
258
259 if (last_disp_cpu >= NR_CPUS_H || cur_disp_cpu >= NR_CPUS_H)
260 return -EINVAL;
261
262 last_disp_cpu_assoc = get_pcpu_associativity(last_disp_cpu);
263 cur_disp_cpu_assoc = get_pcpu_associativity(cur_disp_cpu);
264
265 if (!last_disp_cpu_assoc || !cur_disp_cpu_assoc)
266 return -EIO;
267
268 return cpu_relative_distance(last_disp_cpu_assoc, cur_disp_cpu_assoc);
269}
270
271static int cpu_home_node_dispatch_distance(int disp_cpu)
272{
273 __be32 *disp_cpu_assoc, *vcpu_assoc;
274 int vcpu_id = smp_processor_id();
275
276 if (disp_cpu >= NR_CPUS_H) {
277 pr_debug_ratelimited("vcpu dispatch cpu %d > %d\n",
278 disp_cpu, NR_CPUS_H);
279 return -EINVAL;
280 }
281
282 disp_cpu_assoc = get_pcpu_associativity(disp_cpu);
283 vcpu_assoc = get_vcpu_associativity(vcpu_id);
284
285 if (!disp_cpu_assoc || !vcpu_assoc)
286 return -EIO;
287
288 return cpu_relative_distance(disp_cpu_assoc, vcpu_assoc);
289}
290
291static void update_vcpu_disp_stat(int disp_cpu)
292{
293 struct vcpu_dispatch_data *disp;
294 int distance;
295
296 disp = this_cpu_ptr(&vcpu_disp_data);
297 if (disp->last_disp_cpu == -1) {
298 disp->last_disp_cpu = disp_cpu;
299 return;
300 }
301
302 disp->total_disp++;
303
304 if (disp->last_disp_cpu == disp_cpu ||
305 (cpu_first_thread_sibling(disp->last_disp_cpu) ==
306 cpu_first_thread_sibling(disp_cpu)))
307 disp->same_cpu_disp++;
308 else {
309 distance = cpu_relative_dispatch_distance(disp->last_disp_cpu,
310 disp_cpu);
311 if (distance < 0)
312 pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
313 smp_processor_id());
314 else {
315 switch (distance) {
316 case 0:
317 disp->same_chip_disp++;
318 break;
319 case 1:
320 disp->diff_chip_disp++;
321 break;
322 case 2:
323 disp->far_chip_disp++;
324 break;
325 default:
326 pr_debug_ratelimited("vcpudispatch_stats: cpu %d (%d -> %d): unexpected relative dispatch distance %d\n",
327 smp_processor_id(),
328 disp->last_disp_cpu,
329 disp_cpu,
330 distance);
331 }
332 }
333 }
334
335 distance = cpu_home_node_dispatch_distance(disp_cpu);
336 if (distance < 0)
337 pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
338 smp_processor_id());
339 else {
340 switch (distance) {
341 case 0:
342 disp->numa_home_disp++;
343 break;
344 case 1:
345 disp->numa_remote_disp++;
346 break;
347 case 2:
348 disp->numa_far_disp++;
349 break;
350 default:
351 pr_debug_ratelimited("vcpudispatch_stats: cpu %d on %d: unexpected numa dispatch distance %d\n",
352 smp_processor_id(),
353 disp_cpu,
354 distance);
355 }
356 }
357
358 disp->last_disp_cpu = disp_cpu;
359}
360
361static void process_dtl_buffer(struct work_struct *work)
362{
363 struct dtl_entry dtle;
364 u64 i = __this_cpu_read(dtl_entry_ridx);
365 struct dtl_entry *dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
366 struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
367 struct lppaca *vpa = local_paca->lppaca_ptr;
368 struct dtl_worker *d = container_of(work, struct dtl_worker, work.work);
369
370 if (!local_paca->dispatch_log)
371 return;
372
373 /* if we have been migrated away, we cancel ourself */
374 if (d->cpu != smp_processor_id()) {
375 pr_debug("vcpudispatch_stats: cpu %d worker migrated -- canceling worker\n",
376 smp_processor_id());
377 return;
378 }
379
380 if (i == be64_to_cpu(vpa->dtl_idx))
381 goto out;
382
383 while (i < be64_to_cpu(vpa->dtl_idx)) {
384 dtle = *dtl;
385 barrier();
386 if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
387 /* buffer has overflowed */
388 pr_debug_ratelimited("vcpudispatch_stats: cpu %d lost %lld DTL samples\n",
389 d->cpu,
390 be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG - i);
391 i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
392 dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
393 continue;
394 }
395 update_vcpu_disp_stat(be16_to_cpu(dtle.processor_id));
396 ++i;
397 ++dtl;
398 if (dtl == dtl_end)
399 dtl = local_paca->dispatch_log;
400 }
401
402 __this_cpu_write(dtl_entry_ridx, i);
403
404out:
405 schedule_delayed_work_on(d->cpu, to_delayed_work(work),
406 HZ / vcpudispatch_stats_freq);
407}
408
409static int dtl_worker_online(unsigned int cpu)
410{
411 struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
412
413 memset(d, 0, sizeof(*d));
414 INIT_DELAYED_WORK(&d->work, process_dtl_buffer);
415 d->cpu = cpu;
416
417#ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
418 per_cpu(dtl_entry_ridx, cpu) = 0;
419 register_dtl_buffer(cpu);
420#else
421 per_cpu(dtl_entry_ridx, cpu) = be64_to_cpu(lppaca_of(cpu).dtl_idx);
422#endif
423
424 schedule_delayed_work_on(cpu, &d->work, HZ / vcpudispatch_stats_freq);
425 return 0;
426}
427
428static int dtl_worker_offline(unsigned int cpu)
429{
430 struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
431
432 cancel_delayed_work_sync(&d->work);
433
434#ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
435 unregister_dtl(get_hard_smp_processor_id(cpu));
436#endif
437
438 return 0;
439}
440
441static void set_global_dtl_mask(u8 mask)
442{
443 int cpu;
444
445 dtl_mask = mask;
446 for_each_present_cpu(cpu)
447 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
448}
449
450static void reset_global_dtl_mask(void)
451{
452 int cpu;
453
454#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
455 dtl_mask = DTL_LOG_PREEMPT;
456#else
457 dtl_mask = 0;
458#endif
459 for_each_present_cpu(cpu)
460 lppaca_of(cpu).dtl_enable_mask = dtl_mask;
461}
462
463static int dtl_worker_enable(unsigned long *time_limit)
464{
465 int rc = 0, state;
466
467 if (!down_write_trylock(&dtl_access_lock)) {
468 rc = -EBUSY;
469 goto out;
470 }
471
472 set_global_dtl_mask(DTL_LOG_ALL);
473
474 /* Setup dtl buffers and register those */
475 alloc_dtl_buffers(time_limit);
476
477 state = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "powerpc/dtl:online",
478 dtl_worker_online, dtl_worker_offline);
479 if (state < 0) {
480 pr_err("vcpudispatch_stats: unable to setup workqueue for DTL processing\n");
481 free_dtl_buffers(time_limit);
482 reset_global_dtl_mask();
483 up_write(&dtl_access_lock);
484 rc = -EINVAL;
485 goto out;
486 }
487 dtl_worker_state = state;
488
489out:
490 return rc;
491}
492
493static void dtl_worker_disable(unsigned long *time_limit)
494{
495 cpuhp_remove_state(dtl_worker_state);
496 free_dtl_buffers(time_limit);
497 reset_global_dtl_mask();
498 up_write(&dtl_access_lock);
499}
500
501static ssize_t vcpudispatch_stats_write(struct file *file, const char __user *p,
502 size_t count, loff_t *ppos)
503{
504 unsigned long time_limit = jiffies + HZ;
505 struct vcpu_dispatch_data *disp;
506 int rc, cmd, cpu;
507 char buf[16];
508
509 if (count > 15)
510 return -EINVAL;
511
512 if (copy_from_user(buf, p, count))
513 return -EFAULT;
514
515 buf[count] = 0;
516 rc = kstrtoint(buf, 0, &cmd);
517 if (rc || cmd < 0 || cmd > 1) {
518 pr_err("vcpudispatch_stats: please use 0 to disable or 1 to enable dispatch statistics\n");
519 return rc ? rc : -EINVAL;
520 }
521
522 mutex_lock(&dtl_enable_mutex);
523
524 if ((cmd == 0 && !vcpudispatch_stats_on) ||
525 (cmd == 1 && vcpudispatch_stats_on))
526 goto out;
527
528 if (cmd) {
529 rc = init_cpu_associativity();
530 if (rc) {
531 destroy_cpu_associativity();
532 goto out;
533 }
534
535 for_each_possible_cpu(cpu) {
536 disp = per_cpu_ptr(&vcpu_disp_data, cpu);
537 memset(disp, 0, sizeof(*disp));
538 disp->last_disp_cpu = -1;
539 }
540
541 rc = dtl_worker_enable(&time_limit);
542 if (rc) {
543 destroy_cpu_associativity();
544 goto out;
545 }
546 } else {
547 dtl_worker_disable(&time_limit);
548 destroy_cpu_associativity();
549 }
550
551 vcpudispatch_stats_on = cmd;
552
553out:
554 mutex_unlock(&dtl_enable_mutex);
555 if (rc)
556 return rc;
557 return count;
558}
559
560static int vcpudispatch_stats_display(struct seq_file *p, void *v)
561{
562 int cpu;
563 struct vcpu_dispatch_data *disp;
564
565 if (!vcpudispatch_stats_on) {
566 seq_puts(p, "off\n");
567 return 0;
568 }
569
570 for_each_online_cpu(cpu) {
571 disp = per_cpu_ptr(&vcpu_disp_data, cpu);
572 seq_printf(p, "cpu%d", cpu);
573 seq_put_decimal_ull(p, " ", disp->total_disp);
574 seq_put_decimal_ull(p, " ", disp->same_cpu_disp);
575 seq_put_decimal_ull(p, " ", disp->same_chip_disp);
576 seq_put_decimal_ull(p, " ", disp->diff_chip_disp);
577 seq_put_decimal_ull(p, " ", disp->far_chip_disp);
578 seq_put_decimal_ull(p, " ", disp->numa_home_disp);
579 seq_put_decimal_ull(p, " ", disp->numa_remote_disp);
580 seq_put_decimal_ull(p, " ", disp->numa_far_disp);
581 seq_puts(p, "\n");
582 }
583
584 return 0;
585}
586
587static int vcpudispatch_stats_open(struct inode *inode, struct file *file)
588{
589 return single_open(file, vcpudispatch_stats_display, NULL);
590}
591
592static const struct proc_ops vcpudispatch_stats_proc_ops = {
593 .proc_open = vcpudispatch_stats_open,
594 .proc_read = seq_read,
595 .proc_write = vcpudispatch_stats_write,
596 .proc_lseek = seq_lseek,
597 .proc_release = single_release,
598};
599
600static ssize_t vcpudispatch_stats_freq_write(struct file *file,
601 const char __user *p, size_t count, loff_t *ppos)
602{
603 int rc, freq;
604 char buf[16];
605
606 if (count > 15)
607 return -EINVAL;
608
609 if (copy_from_user(buf, p, count))
610 return -EFAULT;
611
612 buf[count] = 0;
613 rc = kstrtoint(buf, 0, &freq);
614 if (rc || freq < 1 || freq > HZ) {
615 pr_err("vcpudispatch_stats_freq: please specify a frequency between 1 and %d\n",
616 HZ);
617 return rc ? rc : -EINVAL;
618 }
619
620 vcpudispatch_stats_freq = freq;
621
622 return count;
623}
624
625static int vcpudispatch_stats_freq_display(struct seq_file *p, void *v)
626{
627 seq_printf(p, "%d\n", vcpudispatch_stats_freq);
628 return 0;
629}
630
631static int vcpudispatch_stats_freq_open(struct inode *inode, struct file *file)
632{
633 return single_open(file, vcpudispatch_stats_freq_display, NULL);
634}
635
636static const struct proc_ops vcpudispatch_stats_freq_proc_ops = {
637 .proc_open = vcpudispatch_stats_freq_open,
638 .proc_read = seq_read,
639 .proc_write = vcpudispatch_stats_freq_write,
640 .proc_lseek = seq_lseek,
641 .proc_release = single_release,
642};
643
644static int __init vcpudispatch_stats_procfs_init(void)
645{
646 if (!lppaca_shared_proc())
647 return 0;
648
649 if (!proc_create("powerpc/vcpudispatch_stats", 0600, NULL,
650 &vcpudispatch_stats_proc_ops))
651 pr_err("vcpudispatch_stats: error creating procfs file\n");
652 else if (!proc_create("powerpc/vcpudispatch_stats_freq", 0600, NULL,
653 &vcpudispatch_stats_freq_proc_ops))
654 pr_err("vcpudispatch_stats_freq: error creating procfs file\n");
655
656 return 0;
657}
658
659machine_device_initcall(pseries, vcpudispatch_stats_procfs_init);
660
661#ifdef CONFIG_PARAVIRT_TIME_ACCOUNTING
662u64 pseries_paravirt_steal_clock(int cpu)
663{
664 struct lppaca *lppaca = &lppaca_of(cpu);
665
666 /*
667 * VPA steal time counters are reported at TB frequency. Hence do a
668 * conversion to ns before returning
669 */
670 return tb_to_ns(be64_to_cpu(READ_ONCE(lppaca->enqueue_dispatch_tb)) +
671 be64_to_cpu(READ_ONCE(lppaca->ready_enqueue_tb)));
672}
673#endif
674
675#endif /* CONFIG_PPC_SPLPAR */
676
677void vpa_init(int cpu)
678{
679 int hwcpu = get_hard_smp_processor_id(cpu);
680 unsigned long addr;
681 long ret;
682
683 /*
684 * The spec says it "may be problematic" if CPU x registers the VPA of
685 * CPU y. We should never do that, but wail if we ever do.
686 */
687 WARN_ON(cpu != smp_processor_id());
688
689 if (cpu_has_feature(CPU_FTR_ALTIVEC))
690 lppaca_of(cpu).vmxregs_in_use = 1;
691
692 if (cpu_has_feature(CPU_FTR_ARCH_207S))
693 lppaca_of(cpu).ebb_regs_in_use = 1;
694
695 addr = __pa(&lppaca_of(cpu));
696 ret = register_vpa(hwcpu, addr);
697
698 if (ret) {
699 pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
700 "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
701 return;
702 }
703
704#ifdef CONFIG_PPC_64S_HASH_MMU
705 /*
706 * PAPR says this feature is SLB-Buffer but firmware never
707 * reports that. All SPLPAR support SLB shadow buffer.
708 */
709 if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
710 addr = __pa(paca_ptrs[cpu]->slb_shadow_ptr);
711 ret = register_slb_shadow(hwcpu, addr);
712 if (ret)
713 pr_err("WARNING: SLB shadow buffer registration for "
714 "cpu %d (hw %d) of area %lx failed with %ld\n",
715 cpu, hwcpu, addr, ret);
716 }
717#endif /* CONFIG_PPC_64S_HASH_MMU */
718
719 /*
720 * Register dispatch trace log, if one has been allocated.
721 */
722 register_dtl_buffer(cpu);
723}
724
725#ifdef CONFIG_PPC_BOOK3S_64
726
727static int __init pseries_lpar_register_process_table(unsigned long base,
728 unsigned long page_size, unsigned long table_size)
729{
730 long rc;
731 unsigned long flags = 0;
732
733 if (table_size)
734 flags |= PROC_TABLE_NEW;
735 if (radix_enabled()) {
736 flags |= PROC_TABLE_RADIX;
737 if (mmu_has_feature(MMU_FTR_GTSE))
738 flags |= PROC_TABLE_GTSE;
739 } else
740 flags |= PROC_TABLE_HPT_SLB;
741 for (;;) {
742 rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
743 page_size, table_size);
744 if (!H_IS_LONG_BUSY(rc))
745 break;
746 mdelay(get_longbusy_msecs(rc));
747 }
748 if (rc != H_SUCCESS) {
749 pr_err("Failed to register process table (rc=%ld)\n", rc);
750 BUG();
751 }
752 return rc;
753}
754
755#ifdef CONFIG_PPC_64S_HASH_MMU
756
757static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
758 unsigned long vpn, unsigned long pa,
759 unsigned long rflags, unsigned long vflags,
760 int psize, int apsize, int ssize)
761{
762 unsigned long lpar_rc;
763 unsigned long flags;
764 unsigned long slot;
765 unsigned long hpte_v, hpte_r;
766
767 if (!(vflags & HPTE_V_BOLTED))
768 pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
769 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
770 hpte_group, vpn, pa, rflags, vflags, psize);
771
772 hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
773 hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
774
775 if (!(vflags & HPTE_V_BOLTED))
776 pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
777
778 /* Now fill in the actual HPTE */
779 /* Set CEC cookie to 0 */
780 /* Zero page = 0 */
781 /* I-cache Invalidate = 0 */
782 /* I-cache synchronize = 0 */
783 /* Exact = 0 */
784 flags = 0;
785
786 if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
787 flags |= H_COALESCE_CAND;
788
789 lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
790 if (unlikely(lpar_rc == H_PTEG_FULL)) {
791 pr_devel("Hash table group is full\n");
792 return -1;
793 }
794
795 /*
796 * Since we try and ioremap PHBs we don't own, the pte insert
797 * will fail. However we must catch the failure in hash_page
798 * or we will loop forever, so return -2 in this case.
799 */
800 if (unlikely(lpar_rc != H_SUCCESS)) {
801 pr_err("Failed hash pte insert with error %ld\n", lpar_rc);
802 return -2;
803 }
804 if (!(vflags & HPTE_V_BOLTED))
805 pr_devel(" -> slot: %lu\n", slot & 7);
806
807 /* Because of iSeries, we have to pass down the secondary
808 * bucket bit here as well
809 */
810 return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
811}
812
813static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
814
815static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
816{
817 unsigned long slot_offset;
818 unsigned long lpar_rc;
819 int i;
820 unsigned long dummy1, dummy2;
821
822 /* pick a random slot to start at */
823 slot_offset = mftb() & 0x7;
824
825 for (i = 0; i < HPTES_PER_GROUP; i++) {
826
827 /* don't remove a bolted entry */
828 lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
829 HPTE_V_BOLTED, &dummy1, &dummy2);
830 if (lpar_rc == H_SUCCESS)
831 return i;
832
833 /*
834 * The test for adjunct partition is performed before the
835 * ANDCOND test. H_RESOURCE may be returned, so we need to
836 * check for that as well.
837 */
838 BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
839
840 slot_offset++;
841 slot_offset &= 0x7;
842 }
843
844 return -1;
845}
846
847/* Called during kexec sequence with MMU off */
848static notrace void manual_hpte_clear_all(void)
849{
850 unsigned long size_bytes = 1UL << ppc64_pft_size;
851 unsigned long hpte_count = size_bytes >> 4;
852 struct {
853 unsigned long pteh;
854 unsigned long ptel;
855 } ptes[4];
856 long lpar_rc;
857 unsigned long i, j;
858
859 /* Read in batches of 4,
860 * invalidate only valid entries not in the VRMA
861 * hpte_count will be a multiple of 4
862 */
863 for (i = 0; i < hpte_count; i += 4) {
864 lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
865 if (lpar_rc != H_SUCCESS) {
866 pr_info("Failed to read hash page table at %ld err %ld\n",
867 i, lpar_rc);
868 continue;
869 }
870 for (j = 0; j < 4; j++){
871 if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
872 HPTE_V_VRMA_MASK)
873 continue;
874 if (ptes[j].pteh & HPTE_V_VALID)
875 plpar_pte_remove_raw(0, i + j, 0,
876 &(ptes[j].pteh), &(ptes[j].ptel));
877 }
878 }
879}
880
881/* Called during kexec sequence with MMU off */
882static notrace int hcall_hpte_clear_all(void)
883{
884 int rc;
885
886 do {
887 rc = plpar_hcall_norets(H_CLEAR_HPT);
888 } while (rc == H_CONTINUE);
889
890 return rc;
891}
892
893/* Called during kexec sequence with MMU off */
894static notrace void pseries_hpte_clear_all(void)
895{
896 int rc;
897
898 rc = hcall_hpte_clear_all();
899 if (rc != H_SUCCESS)
900 manual_hpte_clear_all();
901
902#ifdef __LITTLE_ENDIAN__
903 /*
904 * Reset exceptions to big endian.
905 *
906 * FIXME this is a hack for kexec, we need to reset the exception
907 * endian before starting the new kernel and this is a convenient place
908 * to do it.
909 *
910 * This is also called on boot when a fadump happens. In that case we
911 * must not change the exception endian mode.
912 */
913 if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
914 pseries_big_endian_exceptions();
915#endif
916}
917
918/*
919 * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
920 * the low 3 bits of flags happen to line up. So no transform is needed.
921 * We can probably optimize here and assume the high bits of newpp are
922 * already zero. For now I am paranoid.
923 */
924static long pSeries_lpar_hpte_updatepp(unsigned long slot,
925 unsigned long newpp,
926 unsigned long vpn,
927 int psize, int apsize,
928 int ssize, unsigned long inv_flags)
929{
930 unsigned long lpar_rc;
931 unsigned long flags;
932 unsigned long want_v;
933
934 want_v = hpte_encode_avpn(vpn, psize, ssize);
935
936 flags = (newpp & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO)) | H_AVPN;
937 flags |= (newpp & HPTE_R_KEY_HI) >> 48;
938 if (mmu_has_feature(MMU_FTR_KERNEL_RO))
939 /* Move pp0 into bit 8 (IBM 55) */
940 flags |= (newpp & HPTE_R_PP0) >> 55;
941
942 pr_devel(" update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
943 want_v, slot, flags, psize);
944
945 lpar_rc = plpar_pte_protect(flags, slot, want_v);
946
947 if (lpar_rc == H_NOT_FOUND) {
948 pr_devel("not found !\n");
949 return -1;
950 }
951
952 pr_devel("ok\n");
953
954 BUG_ON(lpar_rc != H_SUCCESS);
955
956 return 0;
957}
958
959static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
960{
961 long lpar_rc;
962 unsigned long i, j;
963 struct {
964 unsigned long pteh;
965 unsigned long ptel;
966 } ptes[4];
967
968 for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
969
970 lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
971 if (lpar_rc != H_SUCCESS) {
972 pr_info("Failed to read hash page table at %ld err %ld\n",
973 hpte_group, lpar_rc);
974 continue;
975 }
976
977 for (j = 0; j < 4; j++) {
978 if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
979 (ptes[j].pteh & HPTE_V_VALID))
980 return i + j;
981 }
982 }
983
984 return -1;
985}
986
987static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
988{
989 long slot;
990 unsigned long hash;
991 unsigned long want_v;
992 unsigned long hpte_group;
993
994 hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
995 want_v = hpte_encode_avpn(vpn, psize, ssize);
996
997 /*
998 * We try to keep bolted entries always in primary hash
999 * But in some case we can find them in secondary too.
1000 */
1001 hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1002 slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
1003 if (slot < 0) {
1004 /* Try in secondary */
1005 hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
1006 slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
1007 if (slot < 0)
1008 return -1;
1009 }
1010 return hpte_group + slot;
1011}
1012
1013static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
1014 unsigned long ea,
1015 int psize, int ssize)
1016{
1017 unsigned long vpn;
1018 unsigned long lpar_rc, slot, vsid, flags;
1019
1020 vsid = get_kernel_vsid(ea, ssize);
1021 vpn = hpt_vpn(ea, vsid, ssize);
1022
1023 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
1024 BUG_ON(slot == -1);
1025
1026 flags = newpp & (HPTE_R_PP | HPTE_R_N);
1027 if (mmu_has_feature(MMU_FTR_KERNEL_RO))
1028 /* Move pp0 into bit 8 (IBM 55) */
1029 flags |= (newpp & HPTE_R_PP0) >> 55;
1030
1031 flags |= ((newpp & HPTE_R_KEY_HI) >> 48) | (newpp & HPTE_R_KEY_LO);
1032
1033 lpar_rc = plpar_pte_protect(flags, slot, 0);
1034
1035 BUG_ON(lpar_rc != H_SUCCESS);
1036}
1037
1038static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
1039 int psize, int apsize,
1040 int ssize, int local)
1041{
1042 unsigned long want_v;
1043 unsigned long lpar_rc;
1044 unsigned long dummy1, dummy2;
1045
1046 pr_devel(" inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
1047 slot, vpn, psize, local);
1048
1049 want_v = hpte_encode_avpn(vpn, psize, ssize);
1050 lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
1051 if (lpar_rc == H_NOT_FOUND)
1052 return;
1053
1054 BUG_ON(lpar_rc != H_SUCCESS);
1055}
1056
1057
1058/*
1059 * As defined in the PAPR's section 14.5.4.1.8
1060 * The control mask doesn't include the returned reference and change bit from
1061 * the processed PTE.
1062 */
1063#define HBLKR_AVPN 0x0100000000000000UL
1064#define HBLKR_CTRL_MASK 0xf800000000000000UL
1065#define HBLKR_CTRL_SUCCESS 0x8000000000000000UL
1066#define HBLKR_CTRL_ERRNOTFOUND 0x8800000000000000UL
1067#define HBLKR_CTRL_ERRBUSY 0xa000000000000000UL
1068
1069/*
1070 * Returned true if we are supporting this block size for the specified segment
1071 * base page size and actual page size.
1072 *
1073 * Currently, we only support 8 size block.
1074 */
1075static inline bool is_supported_hlbkrm(int bpsize, int psize)
1076{
1077 return (hblkrm_size[bpsize][psize] == HBLKRM_SUPPORTED_BLOCK_SIZE);
1078}
1079
1080/**
1081 * H_BLOCK_REMOVE caller.
1082 * @idx should point to the latest @param entry set with a PTEX.
1083 * If PTE cannot be processed because another CPUs has already locked that
1084 * group, those entries are put back in @param starting at index 1.
1085 * If entries has to be retried and @retry_busy is set to true, these entries
1086 * are retried until success. If @retry_busy is set to false, the returned
1087 * is the number of entries yet to process.
1088 */
1089static unsigned long call_block_remove(unsigned long idx, unsigned long *param,
1090 bool retry_busy)
1091{
1092 unsigned long i, rc, new_idx;
1093 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1094
1095 if (idx < 2) {
1096 pr_warn("Unexpected empty call to H_BLOCK_REMOVE");
1097 return 0;
1098 }
1099again:
1100 new_idx = 0;
1101 if (idx > PLPAR_HCALL9_BUFSIZE) {
1102 pr_err("Too many PTEs (%lu) for H_BLOCK_REMOVE", idx);
1103 idx = PLPAR_HCALL9_BUFSIZE;
1104 } else if (idx < PLPAR_HCALL9_BUFSIZE)
1105 param[idx] = HBR_END;
1106
1107 rc = plpar_hcall9(H_BLOCK_REMOVE, retbuf,
1108 param[0], /* AVA */
1109 param[1], param[2], param[3], param[4], /* TS0-7 */
1110 param[5], param[6], param[7], param[8]);
1111 if (rc == H_SUCCESS)
1112 return 0;
1113
1114 BUG_ON(rc != H_PARTIAL);
1115
1116 /* Check that the unprocessed entries were 'not found' or 'busy' */
1117 for (i = 0; i < idx-1; i++) {
1118 unsigned long ctrl = retbuf[i] & HBLKR_CTRL_MASK;
1119
1120 if (ctrl == HBLKR_CTRL_ERRBUSY) {
1121 param[++new_idx] = param[i+1];
1122 continue;
1123 }
1124
1125 BUG_ON(ctrl != HBLKR_CTRL_SUCCESS
1126 && ctrl != HBLKR_CTRL_ERRNOTFOUND);
1127 }
1128
1129 /*
1130 * If there were entries found busy, retry these entries if requested,
1131 * of if all the entries have to be retried.
1132 */
1133 if (new_idx && (retry_busy || new_idx == (PLPAR_HCALL9_BUFSIZE-1))) {
1134 idx = new_idx + 1;
1135 goto again;
1136 }
1137
1138 return new_idx;
1139}
1140
1141#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1142/*
1143 * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
1144 * to make sure that we avoid bouncing the hypervisor tlbie lock.
1145 */
1146#define PPC64_HUGE_HPTE_BATCH 12
1147
1148static void hugepage_block_invalidate(unsigned long *slot, unsigned long *vpn,
1149 int count, int psize, int ssize)
1150{
1151 unsigned long param[PLPAR_HCALL9_BUFSIZE];
1152 unsigned long shift, current_vpgb, vpgb;
1153 int i, pix = 0;
1154
1155 shift = mmu_psize_defs[psize].shift;
1156
1157 for (i = 0; i < count; i++) {
1158 /*
1159 * Shifting 3 bits more on the right to get a
1160 * 8 pages aligned virtual addresse.
1161 */
1162 vpgb = (vpn[i] >> (shift - VPN_SHIFT + 3));
1163 if (!pix || vpgb != current_vpgb) {
1164 /*
1165 * Need to start a new 8 pages block, flush
1166 * the current one if needed.
1167 */
1168 if (pix)
1169 (void)call_block_remove(pix, param, true);
1170 current_vpgb = vpgb;
1171 param[0] = hpte_encode_avpn(vpn[i], psize, ssize);
1172 pix = 1;
1173 }
1174
1175 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot[i];
1176 if (pix == PLPAR_HCALL9_BUFSIZE) {
1177 pix = call_block_remove(pix, param, false);
1178 /*
1179 * pix = 0 means that all the entries were
1180 * removed, we can start a new block.
1181 * Otherwise, this means that there are entries
1182 * to retry, and pix points to latest one, so
1183 * we should increment it and try to continue
1184 * the same block.
1185 */
1186 if (pix)
1187 pix++;
1188 }
1189 }
1190 if (pix)
1191 (void)call_block_remove(pix, param, true);
1192}
1193
1194static void hugepage_bulk_invalidate(unsigned long *slot, unsigned long *vpn,
1195 int count, int psize, int ssize)
1196{
1197 unsigned long param[PLPAR_HCALL9_BUFSIZE];
1198 int i = 0, pix = 0, rc;
1199
1200 for (i = 0; i < count; i++) {
1201
1202 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1203 pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
1204 ssize, 0);
1205 } else {
1206 param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
1207 param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
1208 pix += 2;
1209 if (pix == 8) {
1210 rc = plpar_hcall9(H_BULK_REMOVE, param,
1211 param[0], param[1], param[2],
1212 param[3], param[4], param[5],
1213 param[6], param[7]);
1214 BUG_ON(rc != H_SUCCESS);
1215 pix = 0;
1216 }
1217 }
1218 }
1219 if (pix) {
1220 param[pix] = HBR_END;
1221 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1222 param[2], param[3], param[4], param[5],
1223 param[6], param[7]);
1224 BUG_ON(rc != H_SUCCESS);
1225 }
1226}
1227
1228static inline void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
1229 unsigned long *vpn,
1230 int count, int psize,
1231 int ssize)
1232{
1233 unsigned long flags = 0;
1234 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1235
1236 if (lock_tlbie)
1237 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1238
1239 /* Assuming THP size is 16M */
1240 if (is_supported_hlbkrm(psize, MMU_PAGE_16M))
1241 hugepage_block_invalidate(slot, vpn, count, psize, ssize);
1242 else
1243 hugepage_bulk_invalidate(slot, vpn, count, psize, ssize);
1244
1245 if (lock_tlbie)
1246 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1247}
1248
1249static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1250 unsigned long addr,
1251 unsigned char *hpte_slot_array,
1252 int psize, int ssize, int local)
1253{
1254 int i, index = 0;
1255 unsigned long s_addr = addr;
1256 unsigned int max_hpte_count, valid;
1257 unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
1258 unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
1259 unsigned long shift, hidx, vpn = 0, hash, slot;
1260
1261 shift = mmu_psize_defs[psize].shift;
1262 max_hpte_count = 1U << (PMD_SHIFT - shift);
1263
1264 for (i = 0; i < max_hpte_count; i++) {
1265 valid = hpte_valid(hpte_slot_array, i);
1266 if (!valid)
1267 continue;
1268 hidx = hpte_hash_index(hpte_slot_array, i);
1269
1270 /* get the vpn */
1271 addr = s_addr + (i * (1ul << shift));
1272 vpn = hpt_vpn(addr, vsid, ssize);
1273 hash = hpt_hash(vpn, shift, ssize);
1274 if (hidx & _PTEIDX_SECONDARY)
1275 hash = ~hash;
1276
1277 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1278 slot += hidx & _PTEIDX_GROUP_IX;
1279
1280 slot_array[index] = slot;
1281 vpn_array[index] = vpn;
1282 if (index == PPC64_HUGE_HPTE_BATCH - 1) {
1283 /*
1284 * Now do a bluk invalidate
1285 */
1286 __pSeries_lpar_hugepage_invalidate(slot_array,
1287 vpn_array,
1288 PPC64_HUGE_HPTE_BATCH,
1289 psize, ssize);
1290 index = 0;
1291 } else
1292 index++;
1293 }
1294 if (index)
1295 __pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
1296 index, psize, ssize);
1297}
1298#else
1299static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1300 unsigned long addr,
1301 unsigned char *hpte_slot_array,
1302 int psize, int ssize, int local)
1303{
1304 WARN(1, "%s called without THP support\n", __func__);
1305}
1306#endif
1307
1308static int pSeries_lpar_hpte_removebolted(unsigned long ea,
1309 int psize, int ssize)
1310{
1311 unsigned long vpn;
1312 unsigned long slot, vsid;
1313
1314 vsid = get_kernel_vsid(ea, ssize);
1315 vpn = hpt_vpn(ea, vsid, ssize);
1316
1317 slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
1318 if (slot == -1)
1319 return -ENOENT;
1320
1321 /*
1322 * lpar doesn't use the passed actual page size
1323 */
1324 pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
1325 return 0;
1326}
1327
1328
1329static inline unsigned long compute_slot(real_pte_t pte,
1330 unsigned long vpn,
1331 unsigned long index,
1332 unsigned long shift,
1333 int ssize)
1334{
1335 unsigned long slot, hash, hidx;
1336
1337 hash = hpt_hash(vpn, shift, ssize);
1338 hidx = __rpte_to_hidx(pte, index);
1339 if (hidx & _PTEIDX_SECONDARY)
1340 hash = ~hash;
1341 slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1342 slot += hidx & _PTEIDX_GROUP_IX;
1343 return slot;
1344}
1345
1346/**
1347 * The hcall H_BLOCK_REMOVE implies that the virtual pages to processed are
1348 * "all within the same naturally aligned 8 page virtual address block".
1349 */
1350static void do_block_remove(unsigned long number, struct ppc64_tlb_batch *batch,
1351 unsigned long *param)
1352{
1353 unsigned long vpn;
1354 unsigned long i, pix = 0;
1355 unsigned long index, shift, slot, current_vpgb, vpgb;
1356 real_pte_t pte;
1357 int psize, ssize;
1358
1359 psize = batch->psize;
1360 ssize = batch->ssize;
1361
1362 for (i = 0; i < number; i++) {
1363 vpn = batch->vpn[i];
1364 pte = batch->pte[i];
1365 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1366 /*
1367 * Shifting 3 bits more on the right to get a
1368 * 8 pages aligned virtual addresse.
1369 */
1370 vpgb = (vpn >> (shift - VPN_SHIFT + 3));
1371 if (!pix || vpgb != current_vpgb) {
1372 /*
1373 * Need to start a new 8 pages block, flush
1374 * the current one if needed.
1375 */
1376 if (pix)
1377 (void)call_block_remove(pix, param,
1378 true);
1379 current_vpgb = vpgb;
1380 param[0] = hpte_encode_avpn(vpn, psize,
1381 ssize);
1382 pix = 1;
1383 }
1384
1385 slot = compute_slot(pte, vpn, index, shift, ssize);
1386 param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot;
1387
1388 if (pix == PLPAR_HCALL9_BUFSIZE) {
1389 pix = call_block_remove(pix, param, false);
1390 /*
1391 * pix = 0 means that all the entries were
1392 * removed, we can start a new block.
1393 * Otherwise, this means that there are entries
1394 * to retry, and pix points to latest one, so
1395 * we should increment it and try to continue
1396 * the same block.
1397 */
1398 if (pix)
1399 pix++;
1400 }
1401 } pte_iterate_hashed_end();
1402 }
1403
1404 if (pix)
1405 (void)call_block_remove(pix, param, true);
1406}
1407
1408/*
1409 * TLB Block Invalidate Characteristics
1410 *
1411 * These characteristics define the size of the block the hcall H_BLOCK_REMOVE
1412 * is able to process for each couple segment base page size, actual page size.
1413 *
1414 * The ibm,get-system-parameter properties is returning a buffer with the
1415 * following layout:
1416 *
1417 * [ 2 bytes size of the RTAS buffer (excluding these 2 bytes) ]
1418 * -----------------
1419 * TLB Block Invalidate Specifiers:
1420 * [ 1 byte LOG base 2 of the TLB invalidate block size being specified ]
1421 * [ 1 byte Number of page sizes (N) that are supported for the specified
1422 * TLB invalidate block size ]
1423 * [ 1 byte Encoded segment base page size and actual page size
1424 * MSB=0 means 4k segment base page size and actual page size
1425 * MSB=1 the penc value in mmu_psize_def ]
1426 * ...
1427 * -----------------
1428 * Next TLB Block Invalidate Specifiers...
1429 * -----------------
1430 * [ 0 ]
1431 */
1432static inline void set_hblkrm_bloc_size(int bpsize, int psize,
1433 unsigned int block_size)
1434{
1435 if (block_size > hblkrm_size[bpsize][psize])
1436 hblkrm_size[bpsize][psize] = block_size;
1437}
1438
1439/*
1440 * Decode the Encoded segment base page size and actual page size.
1441 * PAPR specifies:
1442 * - bit 7 is the L bit
1443 * - bits 0-5 are the penc value
1444 * If the L bit is 0, this means 4K segment base page size and actual page size
1445 * otherwise the penc value should be read.
1446 */
1447#define HBLKRM_L_MASK 0x80
1448#define HBLKRM_PENC_MASK 0x3f
1449static inline void __init check_lp_set_hblkrm(unsigned int lp,
1450 unsigned int block_size)
1451{
1452 unsigned int bpsize, psize;
1453
1454 /* First, check the L bit, if not set, this means 4K */
1455 if ((lp & HBLKRM_L_MASK) == 0) {
1456 set_hblkrm_bloc_size(MMU_PAGE_4K, MMU_PAGE_4K, block_size);
1457 return;
1458 }
1459
1460 lp &= HBLKRM_PENC_MASK;
1461 for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++) {
1462 struct mmu_psize_def *def = &mmu_psize_defs[bpsize];
1463
1464 for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1465 if (def->penc[psize] == lp) {
1466 set_hblkrm_bloc_size(bpsize, psize, block_size);
1467 return;
1468 }
1469 }
1470 }
1471}
1472
1473/*
1474 * The size of the TLB Block Invalidate Characteristics is variable. But at the
1475 * maximum it will be the number of possible page sizes *2 + 10 bytes.
1476 * Currently MMU_PAGE_COUNT is 16, which means 42 bytes. Use a cache line size
1477 * (128 bytes) for the buffer to get plenty of space.
1478 */
1479#define SPLPAR_TLB_BIC_MAXLENGTH 128
1480
1481void __init pseries_lpar_read_hblkrm_characteristics(void)
1482{
1483 static struct papr_sysparm_buf buf __initdata;
1484 int len, idx, bpsize;
1485
1486 if (!firmware_has_feature(FW_FEATURE_BLOCK_REMOVE))
1487 return;
1488
1489 if (papr_sysparm_get(PAPR_SYSPARM_TLB_BLOCK_INVALIDATE_ATTRS, &buf))
1490 return;
1491
1492 len = be16_to_cpu(buf.len);
1493 if (len > SPLPAR_TLB_BIC_MAXLENGTH) {
1494 pr_warn("%s too large returned buffer %d", __func__, len);
1495 return;
1496 }
1497
1498 idx = 0;
1499 while (idx < len) {
1500 u8 block_shift = buf.val[idx++];
1501 u32 block_size;
1502 unsigned int npsize;
1503
1504 if (!block_shift)
1505 break;
1506
1507 block_size = 1 << block_shift;
1508
1509 for (npsize = buf.val[idx++];
1510 npsize > 0 && idx < len; npsize--)
1511 check_lp_set_hblkrm((unsigned int)buf.val[idx++],
1512 block_size);
1513 }
1514
1515 for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
1516 for (idx = 0; idx < MMU_PAGE_COUNT; idx++)
1517 if (hblkrm_size[bpsize][idx])
1518 pr_info("H_BLOCK_REMOVE supports base psize:%d psize:%d block size:%d",
1519 bpsize, idx, hblkrm_size[bpsize][idx]);
1520}
1521
1522/*
1523 * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
1524 * lock.
1525 */
1526static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
1527{
1528 unsigned long vpn;
1529 unsigned long i, pix, rc;
1530 unsigned long flags = 0;
1531 struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
1532 int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1533 unsigned long param[PLPAR_HCALL9_BUFSIZE];
1534 unsigned long index, shift, slot;
1535 real_pte_t pte;
1536 int psize, ssize;
1537
1538 if (lock_tlbie)
1539 spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1540
1541 if (is_supported_hlbkrm(batch->psize, batch->psize)) {
1542 do_block_remove(number, batch, param);
1543 goto out;
1544 }
1545
1546 psize = batch->psize;
1547 ssize = batch->ssize;
1548 pix = 0;
1549 for (i = 0; i < number; i++) {
1550 vpn = batch->vpn[i];
1551 pte = batch->pte[i];
1552 pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1553 slot = compute_slot(pte, vpn, index, shift, ssize);
1554 if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1555 /*
1556 * lpar doesn't use the passed actual page size
1557 */
1558 pSeries_lpar_hpte_invalidate(slot, vpn, psize,
1559 0, ssize, local);
1560 } else {
1561 param[pix] = HBR_REQUEST | HBR_AVPN | slot;
1562 param[pix+1] = hpte_encode_avpn(vpn, psize,
1563 ssize);
1564 pix += 2;
1565 if (pix == 8) {
1566 rc = plpar_hcall9(H_BULK_REMOVE, param,
1567 param[0], param[1], param[2],
1568 param[3], param[4], param[5],
1569 param[6], param[7]);
1570 BUG_ON(rc != H_SUCCESS);
1571 pix = 0;
1572 }
1573 }
1574 } pte_iterate_hashed_end();
1575 }
1576 if (pix) {
1577 param[pix] = HBR_END;
1578 rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1579 param[2], param[3], param[4], param[5],
1580 param[6], param[7]);
1581 BUG_ON(rc != H_SUCCESS);
1582 }
1583
1584out:
1585 if (lock_tlbie)
1586 spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1587}
1588
1589static int __init disable_bulk_remove(char *str)
1590{
1591 if (strcmp(str, "off") == 0 &&
1592 firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1593 pr_info("Disabling BULK_REMOVE firmware feature");
1594 powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
1595 }
1596 return 1;
1597}
1598
1599__setup("bulk_remove=", disable_bulk_remove);
1600
1601#define HPT_RESIZE_TIMEOUT 10000 /* ms */
1602
1603struct hpt_resize_state {
1604 unsigned long shift;
1605 int commit_rc;
1606};
1607
1608static int pseries_lpar_resize_hpt_commit(void *data)
1609{
1610 struct hpt_resize_state *state = data;
1611
1612 state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
1613 if (state->commit_rc != H_SUCCESS)
1614 return -EIO;
1615
1616 /* Hypervisor has transitioned the HTAB, update our globals */
1617 ppc64_pft_size = state->shift;
1618 htab_size_bytes = 1UL << ppc64_pft_size;
1619 htab_hash_mask = (htab_size_bytes >> 7) - 1;
1620
1621 return 0;
1622}
1623
1624/*
1625 * Must be called in process context. The caller must hold the
1626 * cpus_lock.
1627 */
1628static int pseries_lpar_resize_hpt(unsigned long shift)
1629{
1630 struct hpt_resize_state state = {
1631 .shift = shift,
1632 .commit_rc = H_FUNCTION,
1633 };
1634 unsigned int delay, total_delay = 0;
1635 int rc;
1636 ktime_t t0, t1, t2;
1637
1638 might_sleep();
1639
1640 if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1641 return -ENODEV;
1642
1643 pr_info("Attempting to resize HPT to shift %lu\n", shift);
1644
1645 t0 = ktime_get();
1646
1647 rc = plpar_resize_hpt_prepare(0, shift);
1648 while (H_IS_LONG_BUSY(rc)) {
1649 delay = get_longbusy_msecs(rc);
1650 total_delay += delay;
1651 if (total_delay > HPT_RESIZE_TIMEOUT) {
1652 /* prepare with shift==0 cancels an in-progress resize */
1653 rc = plpar_resize_hpt_prepare(0, 0);
1654 if (rc != H_SUCCESS)
1655 pr_warn("Unexpected error %d cancelling timed out HPT resize\n",
1656 rc);
1657 return -ETIMEDOUT;
1658 }
1659 msleep(delay);
1660 rc = plpar_resize_hpt_prepare(0, shift);
1661 }
1662
1663 switch (rc) {
1664 case H_SUCCESS:
1665 /* Continue on */
1666 break;
1667
1668 case H_PARAMETER:
1669 pr_warn("Invalid argument from H_RESIZE_HPT_PREPARE\n");
1670 return -EINVAL;
1671 case H_RESOURCE:
1672 pr_warn("Operation not permitted from H_RESIZE_HPT_PREPARE\n");
1673 return -EPERM;
1674 default:
1675 pr_warn("Unexpected error %d from H_RESIZE_HPT_PREPARE\n", rc);
1676 return -EIO;
1677 }
1678
1679 t1 = ktime_get();
1680
1681 rc = stop_machine_cpuslocked(pseries_lpar_resize_hpt_commit,
1682 &state, NULL);
1683
1684 t2 = ktime_get();
1685
1686 if (rc != 0) {
1687 switch (state.commit_rc) {
1688 case H_PTEG_FULL:
1689 return -ENOSPC;
1690
1691 default:
1692 pr_warn("Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
1693 state.commit_rc);
1694 return -EIO;
1695 };
1696 }
1697
1698 pr_info("HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
1699 shift, (long long) ktime_ms_delta(t1, t0),
1700 (long long) ktime_ms_delta(t2, t1));
1701
1702 return 0;
1703}
1704
1705void __init hpte_init_pseries(void)
1706{
1707 mmu_hash_ops.hpte_invalidate = pSeries_lpar_hpte_invalidate;
1708 mmu_hash_ops.hpte_updatepp = pSeries_lpar_hpte_updatepp;
1709 mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
1710 mmu_hash_ops.hpte_insert = pSeries_lpar_hpte_insert;
1711 mmu_hash_ops.hpte_remove = pSeries_lpar_hpte_remove;
1712 mmu_hash_ops.hpte_removebolted = pSeries_lpar_hpte_removebolted;
1713 mmu_hash_ops.flush_hash_range = pSeries_lpar_flush_hash_range;
1714 mmu_hash_ops.hpte_clear_all = pseries_hpte_clear_all;
1715 mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
1716
1717 if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1718 mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
1719
1720 /*
1721 * On POWER9, we need to do a H_REGISTER_PROC_TBL hcall
1722 * to inform the hypervisor that we wish to use the HPT.
1723 */
1724 if (cpu_has_feature(CPU_FTR_ARCH_300))
1725 pseries_lpar_register_process_table(0, 0, 0);
1726}
1727#endif /* CONFIG_PPC_64S_HASH_MMU */
1728
1729#ifdef CONFIG_PPC_RADIX_MMU
1730void __init radix_init_pseries(void)
1731{
1732 pr_info("Using radix MMU under hypervisor\n");
1733
1734 pseries_lpar_register_process_table(__pa(process_tb),
1735 0, PRTB_SIZE_SHIFT - 12);
1736}
1737#endif
1738
1739#ifdef CONFIG_PPC_SMLPAR
1740#define CMO_FREE_HINT_DEFAULT 1
1741static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
1742
1743static int __init cmo_free_hint(char *str)
1744{
1745 char *parm;
1746 parm = strstrip(str);
1747
1748 if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
1749 pr_info("%s: CMO free page hinting is not active.\n", __func__);
1750 cmo_free_hint_flag = 0;
1751 return 1;
1752 }
1753
1754 cmo_free_hint_flag = 1;
1755 pr_info("%s: CMO free page hinting is active.\n", __func__);
1756
1757 if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
1758 return 1;
1759
1760 return 0;
1761}
1762
1763__setup("cmo_free_hint=", cmo_free_hint);
1764
1765static void pSeries_set_page_state(struct page *page, int order,
1766 unsigned long state)
1767{
1768 int i, j;
1769 unsigned long cmo_page_sz, addr;
1770
1771 cmo_page_sz = cmo_get_page_size();
1772 addr = __pa((unsigned long)page_address(page));
1773
1774 for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
1775 for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
1776 plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
1777 }
1778}
1779
1780void arch_free_page(struct page *page, int order)
1781{
1782 if (radix_enabled())
1783 return;
1784 if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
1785 return;
1786
1787 pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
1788}
1789EXPORT_SYMBOL(arch_free_page);
1790
1791#endif /* CONFIG_PPC_SMLPAR */
1792#endif /* CONFIG_PPC_BOOK3S_64 */
1793
1794#ifdef CONFIG_TRACEPOINTS
1795#ifdef CONFIG_JUMP_LABEL
1796struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
1797
1798int hcall_tracepoint_regfunc(void)
1799{
1800 static_key_slow_inc(key: &hcall_tracepoint_key);
1801 return 0;
1802}
1803
1804void hcall_tracepoint_unregfunc(void)
1805{
1806 static_key_slow_dec(key: &hcall_tracepoint_key);
1807}
1808#else
1809/*
1810 * We optimise our hcall path by placing hcall_tracepoint_refcount
1811 * directly in the TOC so we can check if the hcall tracepoints are
1812 * enabled via a single load.
1813 */
1814
1815/* NB: reg/unreg are called while guarded with the tracepoints_mutex */
1816extern long hcall_tracepoint_refcount;
1817
1818int hcall_tracepoint_regfunc(void)
1819{
1820 hcall_tracepoint_refcount++;
1821 return 0;
1822}
1823
1824void hcall_tracepoint_unregfunc(void)
1825{
1826 hcall_tracepoint_refcount--;
1827}
1828#endif
1829
1830/*
1831 * Keep track of hcall tracing depth and prevent recursion. Warn if any is
1832 * detected because it may indicate a problem. This will not catch all
1833 * problems with tracing code making hcalls, because the tracing might have
1834 * been invoked from a non-hcall, so the first hcall could recurse into it
1835 * without warning here, but this better than nothing.
1836 *
1837 * Hcalls with specific problems being traced should use the _notrace
1838 * plpar_hcall variants.
1839 */
1840static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
1841
1842
1843notrace void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
1844{
1845 unsigned long flags;
1846 unsigned int *depth;
1847
1848 local_irq_save(flags);
1849
1850 depth = this_cpu_ptr(&hcall_trace_depth);
1851
1852 if (WARN_ON_ONCE(*depth))
1853 goto out;
1854
1855 (*depth)++;
1856 preempt_disable();
1857 trace_hcall_entry(opcode, args);
1858 (*depth)--;
1859
1860out:
1861 local_irq_restore(flags);
1862}
1863
1864notrace void __trace_hcall_exit(long opcode, long retval, unsigned long *retbuf)
1865{
1866 unsigned long flags;
1867 unsigned int *depth;
1868
1869 local_irq_save(flags);
1870
1871 depth = this_cpu_ptr(&hcall_trace_depth);
1872
1873 if (*depth) /* Don't warn again on the way out */
1874 goto out;
1875
1876 (*depth)++;
1877 trace_hcall_exit(opcode, retval, retbuf);
1878 preempt_enable();
1879 (*depth)--;
1880
1881out:
1882 local_irq_restore(flags);
1883}
1884#endif
1885
1886/**
1887 * h_get_mpp
1888 * H_GET_MPP hcall returns info in 7 parms
1889 */
1890long h_get_mpp(struct hvcall_mpp_data *mpp_data)
1891{
1892 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = {0};
1893 long rc;
1894
1895 rc = plpar_hcall9(H_GET_MPP, retbuf);
1896
1897 mpp_data->entitled_mem = retbuf[0];
1898 mpp_data->mapped_mem = retbuf[1];
1899
1900 mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
1901 mpp_data->pool_num = retbuf[2] & 0xffff;
1902
1903 mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
1904 mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
1905 mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
1906
1907 mpp_data->pool_size = retbuf[4];
1908 mpp_data->loan_request = retbuf[5];
1909 mpp_data->backing_mem = retbuf[6];
1910
1911 return rc;
1912}
1913EXPORT_SYMBOL(h_get_mpp);
1914
1915int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
1916{
1917 int rc;
1918 unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
1919
1920 rc = plpar_hcall9(H_GET_MPP_X, retbuf);
1921
1922 mpp_x_data->coalesced_bytes = retbuf[0];
1923 mpp_x_data->pool_coalesced_bytes = retbuf[1];
1924 mpp_x_data->pool_purr_cycles = retbuf[2];
1925 mpp_x_data->pool_spurr_cycles = retbuf[3];
1926
1927 return rc;
1928}
1929
1930#ifdef CONFIG_PPC_64S_HASH_MMU
1931static unsigned long __init vsid_unscramble(unsigned long vsid, int ssize)
1932{
1933 unsigned long protovsid;
1934 unsigned long va_bits = VA_BITS;
1935 unsigned long modinv, vsid_modulus;
1936 unsigned long max_mod_inv, tmp_modinv;
1937
1938 if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
1939 va_bits = 65;
1940
1941 if (ssize == MMU_SEGSIZE_256M) {
1942 modinv = VSID_MULINV_256M;
1943 vsid_modulus = ((1UL << (va_bits - SID_SHIFT)) - 1);
1944 } else {
1945 modinv = VSID_MULINV_1T;
1946 vsid_modulus = ((1UL << (va_bits - SID_SHIFT_1T)) - 1);
1947 }
1948
1949 /*
1950 * vsid outside our range.
1951 */
1952 if (vsid >= vsid_modulus)
1953 return 0;
1954
1955 /*
1956 * If modinv is the modular multiplicate inverse of (x % vsid_modulus)
1957 * and vsid = (protovsid * x) % vsid_modulus, then we say:
1958 * protovsid = (vsid * modinv) % vsid_modulus
1959 */
1960
1961 /* Check if (vsid * modinv) overflow (63 bits) */
1962 max_mod_inv = 0x7fffffffffffffffull / vsid;
1963 if (modinv < max_mod_inv)
1964 return (vsid * modinv) % vsid_modulus;
1965
1966 tmp_modinv = modinv/max_mod_inv;
1967 modinv %= max_mod_inv;
1968
1969 protovsid = (((vsid * max_mod_inv) % vsid_modulus) * tmp_modinv) % vsid_modulus;
1970 protovsid = (protovsid + vsid * modinv) % vsid_modulus;
1971
1972 return protovsid;
1973}
1974
1975static int __init reserve_vrma_context_id(void)
1976{
1977 unsigned long protovsid;
1978
1979 /*
1980 * Reserve context ids which map to reserved virtual addresses. For now
1981 * we only reserve the context id which maps to the VRMA VSID. We ignore
1982 * the addresses in "ibm,adjunct-virtual-addresses" because we don't
1983 * enable adjunct support via the "ibm,client-architecture-support"
1984 * interface.
1985 */
1986 protovsid = vsid_unscramble(VRMA_VSID, MMU_SEGSIZE_1T);
1987 hash__reserve_context_id(protovsid >> ESID_BITS_1T);
1988 return 0;
1989}
1990machine_device_initcall(pseries, reserve_vrma_context_id);
1991#endif
1992
1993#ifdef CONFIG_DEBUG_FS
1994/* debugfs file interface for vpa data */
1995static ssize_t vpa_file_read(struct file *filp, char __user *buf, size_t len,
1996 loff_t *pos)
1997{
1998 int cpu = (long)filp->private_data;
1999 struct lppaca *lppaca = &lppaca_of(cpu);
2000
2001 return simple_read_from_buffer(buf, len, pos, lppaca,
2002 sizeof(struct lppaca));
2003}
2004
2005static const struct file_operations vpa_fops = {
2006 .open = simple_open,
2007 .read = vpa_file_read,
2008 .llseek = default_llseek,
2009};
2010
2011static int __init vpa_debugfs_init(void)
2012{
2013 char name[16];
2014 long i;
2015 struct dentry *vpa_dir;
2016
2017 if (!firmware_has_feature(FW_FEATURE_SPLPAR))
2018 return 0;
2019
2020 vpa_dir = debugfs_create_dir(name: "vpa", parent: arch_debugfs_dir);
2021
2022 /* set up the per-cpu vpa file*/
2023 for_each_possible_cpu(i) {
2024 sprintf(buf: name, fmt: "cpu-%ld", i);
2025 debugfs_create_file(name, 0400, vpa_dir, (void *)i, &vpa_fops);
2026 }
2027
2028 return 0;
2029}
2030machine_arch_initcall(pseries, vpa_debugfs_init);
2031#endif /* CONFIG_DEBUG_FS */
2032

source code of linux/arch/powerpc/platforms/pseries/lpar.c