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/stop_machine.h>
20 #include <linux/spinlock.h>
21 #include <linux/cpuhotplug.h>
22 #include <linux/workqueue.h>
23 #include <linux/proc_fs.h>
24 #include <linux/pgtable.h>
25 #include <linux/debugfs.h>
26 
27 #include <asm/processor.h>
28 #include <asm/mmu.h>
29 #include <asm/page.h>
30 #include <asm/machdep.h>
31 #include <asm/mmu_context.h>
32 #include <asm/iommu.h>
33 #include <asm/tlb.h>
34 #include <asm/prom.h>
35 #include <asm/cputable.h>
36 #include <asm/udbg.h>
37 #include <asm/smp.h>
38 #include <asm/trace.h>
39 #include <asm/firmware.h>
40 #include <asm/plpar_wrappers.h>
41 #include <asm/kexec.h>
42 #include <asm/fadump.h>
43 #include <asm/asm-prototypes.h>
44 #include <asm/dtl.h>
45 
46 #include "pseries.h"
47 
48 /* Flag bits for H_BULK_REMOVE */
49 #define HBR_REQUEST	0x4000000000000000UL
50 #define HBR_RESPONSE	0x8000000000000000UL
51 #define HBR_END		0xc000000000000000UL
52 #define HBR_AVPN	0x0200000000000000UL
53 #define HBR_ANDCOND	0x0100000000000000UL
54 
55 
56 /* in hvCall.S */
57 EXPORT_SYMBOL(plpar_hcall);
58 EXPORT_SYMBOL(plpar_hcall9);
59 EXPORT_SYMBOL(plpar_hcall_norets);
60 
61 /*
62  * H_BLOCK_REMOVE supported block size for this page size in segment who's base
63  * page size is that page size.
64  *
65  * The first index is the segment base page size, the second one is the actual
66  * page size.
67  */
68 static int hblkrm_size[MMU_PAGE_COUNT][MMU_PAGE_COUNT] __ro_after_init;
69 
70 /*
71  * Due to the involved complexity, and that the current hypervisor is only
72  * returning this value or 0, we are limiting the support of the H_BLOCK_REMOVE
73  * buffer size to 8 size block.
74  */
75 #define HBLKRM_SUPPORTED_BLOCK_SIZE 8
76 
77 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
78 static u8 dtl_mask = DTL_LOG_PREEMPT;
79 #else
80 static u8 dtl_mask;
81 #endif
82 
83 void alloc_dtl_buffers(unsigned long *time_limit)
84 {
85 	int cpu;
86 	struct paca_struct *pp;
87 	struct dtl_entry *dtl;
88 
89 	for_each_possible_cpu(cpu) {
90 		pp = paca_ptrs[cpu];
91 		if (pp->dispatch_log)
92 			continue;
93 		dtl = kmem_cache_alloc(dtl_cache, GFP_KERNEL);
94 		if (!dtl) {
95 			pr_warn("Failed to allocate dispatch trace log for cpu %d\n",
96 				cpu);
97 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
98 			pr_warn("Stolen time statistics will be unreliable\n");
99 #endif
100 			break;
101 		}
102 
103 		pp->dtl_ridx = 0;
104 		pp->dispatch_log = dtl;
105 		pp->dispatch_log_end = dtl + N_DISPATCH_LOG;
106 		pp->dtl_curr = dtl;
107 
108 		if (time_limit && time_after(jiffies, *time_limit)) {
109 			cond_resched();
110 			*time_limit = jiffies + HZ;
111 		}
112 	}
113 }
114 
115 void register_dtl_buffer(int cpu)
116 {
117 	long ret;
118 	struct paca_struct *pp;
119 	struct dtl_entry *dtl;
120 	int hwcpu = get_hard_smp_processor_id(cpu);
121 
122 	pp = paca_ptrs[cpu];
123 	dtl = pp->dispatch_log;
124 	if (dtl && dtl_mask) {
125 		pp->dtl_ridx = 0;
126 		pp->dtl_curr = dtl;
127 		lppaca_of(cpu).dtl_idx = 0;
128 
129 		/* hypervisor reads buffer length from this field */
130 		dtl->enqueue_to_dispatch_time = cpu_to_be32(DISPATCH_LOG_BYTES);
131 		ret = register_dtl(hwcpu, __pa(dtl));
132 		if (ret)
133 			pr_err("WARNING: DTL registration of cpu %d (hw %d) failed with %ld\n",
134 			       cpu, hwcpu, ret);
135 
136 		lppaca_of(cpu).dtl_enable_mask = dtl_mask;
137 	}
138 }
139 
140 #ifdef CONFIG_PPC_SPLPAR
141 struct dtl_worker {
142 	struct delayed_work work;
143 	int cpu;
144 };
145 
146 struct vcpu_dispatch_data {
147 	int last_disp_cpu;
148 
149 	int total_disp;
150 
151 	int same_cpu_disp;
152 	int same_chip_disp;
153 	int diff_chip_disp;
154 	int far_chip_disp;
155 
156 	int numa_home_disp;
157 	int numa_remote_disp;
158 	int numa_far_disp;
159 };
160 
161 /*
162  * This represents the number of cpus in the hypervisor. Since there is no
163  * architected way to discover the number of processors in the host, we
164  * provision for dealing with NR_CPUS. This is currently 2048 by default, and
165  * is sufficient for our purposes. This will need to be tweaked if
166  * CONFIG_NR_CPUS is changed.
167  */
168 #define NR_CPUS_H	NR_CPUS
169 
170 DEFINE_RWLOCK(dtl_access_lock);
171 static DEFINE_PER_CPU(struct vcpu_dispatch_data, vcpu_disp_data);
172 static DEFINE_PER_CPU(u64, dtl_entry_ridx);
173 static DEFINE_PER_CPU(struct dtl_worker, dtl_workers);
174 static enum cpuhp_state dtl_worker_state;
175 static DEFINE_MUTEX(dtl_enable_mutex);
176 static int vcpudispatch_stats_on __read_mostly;
177 static int vcpudispatch_stats_freq = 50;
178 static __be32 *vcpu_associativity, *pcpu_associativity;
179 
180 
181 static void free_dtl_buffers(unsigned long *time_limit)
182 {
183 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
184 	int cpu;
185 	struct paca_struct *pp;
186 
187 	for_each_possible_cpu(cpu) {
188 		pp = paca_ptrs[cpu];
189 		if (!pp->dispatch_log)
190 			continue;
191 		kmem_cache_free(dtl_cache, pp->dispatch_log);
192 		pp->dtl_ridx = 0;
193 		pp->dispatch_log = 0;
194 		pp->dispatch_log_end = 0;
195 		pp->dtl_curr = 0;
196 
197 		if (time_limit && time_after(jiffies, *time_limit)) {
198 			cond_resched();
199 			*time_limit = jiffies + HZ;
200 		}
201 	}
202 #endif
203 }
204 
205 static int init_cpu_associativity(void)
206 {
207 	vcpu_associativity = kcalloc(num_possible_cpus() / threads_per_core,
208 			VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
209 	pcpu_associativity = kcalloc(NR_CPUS_H / threads_per_core,
210 			VPHN_ASSOC_BUFSIZE * sizeof(__be32), GFP_KERNEL);
211 
212 	if (!vcpu_associativity || !pcpu_associativity) {
213 		pr_err("error allocating memory for associativity information\n");
214 		return -ENOMEM;
215 	}
216 
217 	return 0;
218 }
219 
220 static void destroy_cpu_associativity(void)
221 {
222 	kfree(vcpu_associativity);
223 	kfree(pcpu_associativity);
224 	vcpu_associativity = pcpu_associativity = 0;
225 }
226 
227 static __be32 *__get_cpu_associativity(int cpu, __be32 *cpu_assoc, int flag)
228 {
229 	__be32 *assoc;
230 	int rc = 0;
231 
232 	assoc = &cpu_assoc[(int)(cpu / threads_per_core) * VPHN_ASSOC_BUFSIZE];
233 	if (!assoc[0]) {
234 		rc = hcall_vphn(cpu, flag, &assoc[0]);
235 		if (rc)
236 			return NULL;
237 	}
238 
239 	return assoc;
240 }
241 
242 static __be32 *get_pcpu_associativity(int cpu)
243 {
244 	return __get_cpu_associativity(cpu, pcpu_associativity, VPHN_FLAG_PCPU);
245 }
246 
247 static __be32 *get_vcpu_associativity(int cpu)
248 {
249 	return __get_cpu_associativity(cpu, vcpu_associativity, VPHN_FLAG_VCPU);
250 }
251 
252 static int cpu_relative_dispatch_distance(int last_disp_cpu, int cur_disp_cpu)
253 {
254 	__be32 *last_disp_cpu_assoc, *cur_disp_cpu_assoc;
255 
256 	if (last_disp_cpu >= NR_CPUS_H || cur_disp_cpu >= NR_CPUS_H)
257 		return -EINVAL;
258 
259 	last_disp_cpu_assoc = get_pcpu_associativity(last_disp_cpu);
260 	cur_disp_cpu_assoc = get_pcpu_associativity(cur_disp_cpu);
261 
262 	if (!last_disp_cpu_assoc || !cur_disp_cpu_assoc)
263 		return -EIO;
264 
265 	return cpu_relative_distance(last_disp_cpu_assoc, cur_disp_cpu_assoc);
266 }
267 
268 static int cpu_home_node_dispatch_distance(int disp_cpu)
269 {
270 	__be32 *disp_cpu_assoc, *vcpu_assoc;
271 	int vcpu_id = smp_processor_id();
272 
273 	if (disp_cpu >= NR_CPUS_H) {
274 		pr_debug_ratelimited("vcpu dispatch cpu %d > %d\n",
275 						disp_cpu, NR_CPUS_H);
276 		return -EINVAL;
277 	}
278 
279 	disp_cpu_assoc = get_pcpu_associativity(disp_cpu);
280 	vcpu_assoc = get_vcpu_associativity(vcpu_id);
281 
282 	if (!disp_cpu_assoc || !vcpu_assoc)
283 		return -EIO;
284 
285 	return cpu_relative_distance(disp_cpu_assoc, vcpu_assoc);
286 }
287 
288 static void update_vcpu_disp_stat(int disp_cpu)
289 {
290 	struct vcpu_dispatch_data *disp;
291 	int distance;
292 
293 	disp = this_cpu_ptr(&vcpu_disp_data);
294 	if (disp->last_disp_cpu == -1) {
295 		disp->last_disp_cpu = disp_cpu;
296 		return;
297 	}
298 
299 	disp->total_disp++;
300 
301 	if (disp->last_disp_cpu == disp_cpu ||
302 		(cpu_first_thread_sibling(disp->last_disp_cpu) ==
303 					cpu_first_thread_sibling(disp_cpu)))
304 		disp->same_cpu_disp++;
305 	else {
306 		distance = cpu_relative_dispatch_distance(disp->last_disp_cpu,
307 								disp_cpu);
308 		if (distance < 0)
309 			pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
310 					smp_processor_id());
311 		else {
312 			switch (distance) {
313 			case 0:
314 				disp->same_chip_disp++;
315 				break;
316 			case 1:
317 				disp->diff_chip_disp++;
318 				break;
319 			case 2:
320 				disp->far_chip_disp++;
321 				break;
322 			default:
323 				pr_debug_ratelimited("vcpudispatch_stats: cpu %d (%d -> %d): unexpected relative dispatch distance %d\n",
324 						 smp_processor_id(),
325 						 disp->last_disp_cpu,
326 						 disp_cpu,
327 						 distance);
328 			}
329 		}
330 	}
331 
332 	distance = cpu_home_node_dispatch_distance(disp_cpu);
333 	if (distance < 0)
334 		pr_debug_ratelimited("vcpudispatch_stats: cpu %d: error determining associativity\n",
335 				smp_processor_id());
336 	else {
337 		switch (distance) {
338 		case 0:
339 			disp->numa_home_disp++;
340 			break;
341 		case 1:
342 			disp->numa_remote_disp++;
343 			break;
344 		case 2:
345 			disp->numa_far_disp++;
346 			break;
347 		default:
348 			pr_debug_ratelimited("vcpudispatch_stats: cpu %d on %d: unexpected numa dispatch distance %d\n",
349 						 smp_processor_id(),
350 						 disp_cpu,
351 						 distance);
352 		}
353 	}
354 
355 	disp->last_disp_cpu = disp_cpu;
356 }
357 
358 static void process_dtl_buffer(struct work_struct *work)
359 {
360 	struct dtl_entry dtle;
361 	u64 i = __this_cpu_read(dtl_entry_ridx);
362 	struct dtl_entry *dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
363 	struct dtl_entry *dtl_end = local_paca->dispatch_log_end;
364 	struct lppaca *vpa = local_paca->lppaca_ptr;
365 	struct dtl_worker *d = container_of(work, struct dtl_worker, work.work);
366 
367 	if (!local_paca->dispatch_log)
368 		return;
369 
370 	/* if we have been migrated away, we cancel ourself */
371 	if (d->cpu != smp_processor_id()) {
372 		pr_debug("vcpudispatch_stats: cpu %d worker migrated -- canceling worker\n",
373 						smp_processor_id());
374 		return;
375 	}
376 
377 	if (i == be64_to_cpu(vpa->dtl_idx))
378 		goto out;
379 
380 	while (i < be64_to_cpu(vpa->dtl_idx)) {
381 		dtle = *dtl;
382 		barrier();
383 		if (i + N_DISPATCH_LOG < be64_to_cpu(vpa->dtl_idx)) {
384 			/* buffer has overflowed */
385 			pr_debug_ratelimited("vcpudispatch_stats: cpu %d lost %lld DTL samples\n",
386 				d->cpu,
387 				be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG - i);
388 			i = be64_to_cpu(vpa->dtl_idx) - N_DISPATCH_LOG;
389 			dtl = local_paca->dispatch_log + (i % N_DISPATCH_LOG);
390 			continue;
391 		}
392 		update_vcpu_disp_stat(be16_to_cpu(dtle.processor_id));
393 		++i;
394 		++dtl;
395 		if (dtl == dtl_end)
396 			dtl = local_paca->dispatch_log;
397 	}
398 
399 	__this_cpu_write(dtl_entry_ridx, i);
400 
401 out:
402 	schedule_delayed_work_on(d->cpu, to_delayed_work(work),
403 					HZ / vcpudispatch_stats_freq);
404 }
405 
406 static int dtl_worker_online(unsigned int cpu)
407 {
408 	struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
409 
410 	memset(d, 0, sizeof(*d));
411 	INIT_DELAYED_WORK(&d->work, process_dtl_buffer);
412 	d->cpu = cpu;
413 
414 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
415 	per_cpu(dtl_entry_ridx, cpu) = 0;
416 	register_dtl_buffer(cpu);
417 #else
418 	per_cpu(dtl_entry_ridx, cpu) = be64_to_cpu(lppaca_of(cpu).dtl_idx);
419 #endif
420 
421 	schedule_delayed_work_on(cpu, &d->work, HZ / vcpudispatch_stats_freq);
422 	return 0;
423 }
424 
425 static int dtl_worker_offline(unsigned int cpu)
426 {
427 	struct dtl_worker *d = &per_cpu(dtl_workers, cpu);
428 
429 	cancel_delayed_work_sync(&d->work);
430 
431 #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
432 	unregister_dtl(get_hard_smp_processor_id(cpu));
433 #endif
434 
435 	return 0;
436 }
437 
438 static void set_global_dtl_mask(u8 mask)
439 {
440 	int cpu;
441 
442 	dtl_mask = mask;
443 	for_each_present_cpu(cpu)
444 		lppaca_of(cpu).dtl_enable_mask = dtl_mask;
445 }
446 
447 static void reset_global_dtl_mask(void)
448 {
449 	int cpu;
450 
451 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
452 	dtl_mask = DTL_LOG_PREEMPT;
453 #else
454 	dtl_mask = 0;
455 #endif
456 	for_each_present_cpu(cpu)
457 		lppaca_of(cpu).dtl_enable_mask = dtl_mask;
458 }
459 
460 static int dtl_worker_enable(unsigned long *time_limit)
461 {
462 	int rc = 0, state;
463 
464 	if (!write_trylock(&dtl_access_lock)) {
465 		rc = -EBUSY;
466 		goto out;
467 	}
468 
469 	set_global_dtl_mask(DTL_LOG_ALL);
470 
471 	/* Setup dtl buffers and register those */
472 	alloc_dtl_buffers(time_limit);
473 
474 	state = cpuhp_setup_state(CPUHP_AP_ONLINE_DYN, "powerpc/dtl:online",
475 					dtl_worker_online, dtl_worker_offline);
476 	if (state < 0) {
477 		pr_err("vcpudispatch_stats: unable to setup workqueue for DTL processing\n");
478 		free_dtl_buffers(time_limit);
479 		reset_global_dtl_mask();
480 		write_unlock(&dtl_access_lock);
481 		rc = -EINVAL;
482 		goto out;
483 	}
484 	dtl_worker_state = state;
485 
486 out:
487 	return rc;
488 }
489 
490 static void dtl_worker_disable(unsigned long *time_limit)
491 {
492 	cpuhp_remove_state(dtl_worker_state);
493 	free_dtl_buffers(time_limit);
494 	reset_global_dtl_mask();
495 	write_unlock(&dtl_access_lock);
496 }
497 
498 static ssize_t vcpudispatch_stats_write(struct file *file, const char __user *p,
499 		size_t count, loff_t *ppos)
500 {
501 	unsigned long time_limit = jiffies + HZ;
502 	struct vcpu_dispatch_data *disp;
503 	int rc, cmd, cpu;
504 	char buf[16];
505 
506 	if (count > 15)
507 		return -EINVAL;
508 
509 	if (copy_from_user(buf, p, count))
510 		return -EFAULT;
511 
512 	buf[count] = 0;
513 	rc = kstrtoint(buf, 0, &cmd);
514 	if (rc || cmd < 0 || cmd > 1) {
515 		pr_err("vcpudispatch_stats: please use 0 to disable or 1 to enable dispatch statistics\n");
516 		return rc ? rc : -EINVAL;
517 	}
518 
519 	mutex_lock(&dtl_enable_mutex);
520 
521 	if ((cmd == 0 && !vcpudispatch_stats_on) ||
522 			(cmd == 1 && vcpudispatch_stats_on))
523 		goto out;
524 
525 	if (cmd) {
526 		rc = init_cpu_associativity();
527 		if (rc)
528 			goto out;
529 
530 		for_each_possible_cpu(cpu) {
531 			disp = per_cpu_ptr(&vcpu_disp_data, cpu);
532 			memset(disp, 0, sizeof(*disp));
533 			disp->last_disp_cpu = -1;
534 		}
535 
536 		rc = dtl_worker_enable(&time_limit);
537 		if (rc) {
538 			destroy_cpu_associativity();
539 			goto out;
540 		}
541 	} else {
542 		dtl_worker_disable(&time_limit);
543 		destroy_cpu_associativity();
544 	}
545 
546 	vcpudispatch_stats_on = cmd;
547 
548 out:
549 	mutex_unlock(&dtl_enable_mutex);
550 	if (rc)
551 		return rc;
552 	return count;
553 }
554 
555 static int vcpudispatch_stats_display(struct seq_file *p, void *v)
556 {
557 	int cpu;
558 	struct vcpu_dispatch_data *disp;
559 
560 	if (!vcpudispatch_stats_on) {
561 		seq_puts(p, "off\n");
562 		return 0;
563 	}
564 
565 	for_each_online_cpu(cpu) {
566 		disp = per_cpu_ptr(&vcpu_disp_data, cpu);
567 		seq_printf(p, "cpu%d", cpu);
568 		seq_put_decimal_ull(p, " ", disp->total_disp);
569 		seq_put_decimal_ull(p, " ", disp->same_cpu_disp);
570 		seq_put_decimal_ull(p, " ", disp->same_chip_disp);
571 		seq_put_decimal_ull(p, " ", disp->diff_chip_disp);
572 		seq_put_decimal_ull(p, " ", disp->far_chip_disp);
573 		seq_put_decimal_ull(p, " ", disp->numa_home_disp);
574 		seq_put_decimal_ull(p, " ", disp->numa_remote_disp);
575 		seq_put_decimal_ull(p, " ", disp->numa_far_disp);
576 		seq_puts(p, "\n");
577 	}
578 
579 	return 0;
580 }
581 
582 static int vcpudispatch_stats_open(struct inode *inode, struct file *file)
583 {
584 	return single_open(file, vcpudispatch_stats_display, NULL);
585 }
586 
587 static const struct proc_ops vcpudispatch_stats_proc_ops = {
588 	.proc_open	= vcpudispatch_stats_open,
589 	.proc_read	= seq_read,
590 	.proc_write	= vcpudispatch_stats_write,
591 	.proc_lseek	= seq_lseek,
592 	.proc_release	= single_release,
593 };
594 
595 static ssize_t vcpudispatch_stats_freq_write(struct file *file,
596 		const char __user *p, size_t count, loff_t *ppos)
597 {
598 	int rc, freq;
599 	char buf[16];
600 
601 	if (count > 15)
602 		return -EINVAL;
603 
604 	if (copy_from_user(buf, p, count))
605 		return -EFAULT;
606 
607 	buf[count] = 0;
608 	rc = kstrtoint(buf, 0, &freq);
609 	if (rc || freq < 1 || freq > HZ) {
610 		pr_err("vcpudispatch_stats_freq: please specify a frequency between 1 and %d\n",
611 				HZ);
612 		return rc ? rc : -EINVAL;
613 	}
614 
615 	vcpudispatch_stats_freq = freq;
616 
617 	return count;
618 }
619 
620 static int vcpudispatch_stats_freq_display(struct seq_file *p, void *v)
621 {
622 	seq_printf(p, "%d\n", vcpudispatch_stats_freq);
623 	return 0;
624 }
625 
626 static int vcpudispatch_stats_freq_open(struct inode *inode, struct file *file)
627 {
628 	return single_open(file, vcpudispatch_stats_freq_display, NULL);
629 }
630 
631 static const struct proc_ops vcpudispatch_stats_freq_proc_ops = {
632 	.proc_open	= vcpudispatch_stats_freq_open,
633 	.proc_read	= seq_read,
634 	.proc_write	= vcpudispatch_stats_freq_write,
635 	.proc_lseek	= seq_lseek,
636 	.proc_release	= single_release,
637 };
638 
639 static int __init vcpudispatch_stats_procfs_init(void)
640 {
641 	/*
642 	 * Avoid smp_processor_id while preemptible. All CPUs should have
643 	 * the same value for lppaca_shared_proc.
644 	 */
645 	preempt_disable();
646 	if (!lppaca_shared_proc(get_lppaca())) {
647 		preempt_enable();
648 		return 0;
649 	}
650 	preempt_enable();
651 
652 	if (!proc_create("powerpc/vcpudispatch_stats", 0600, NULL,
653 					&vcpudispatch_stats_proc_ops))
654 		pr_err("vcpudispatch_stats: error creating procfs file\n");
655 	else if (!proc_create("powerpc/vcpudispatch_stats_freq", 0600, NULL,
656 					&vcpudispatch_stats_freq_proc_ops))
657 		pr_err("vcpudispatch_stats_freq: error creating procfs file\n");
658 
659 	return 0;
660 }
661 
662 machine_device_initcall(pseries, vcpudispatch_stats_procfs_init);
663 #endif /* CONFIG_PPC_SPLPAR */
664 
665 void vpa_init(int cpu)
666 {
667 	int hwcpu = get_hard_smp_processor_id(cpu);
668 	unsigned long addr;
669 	long ret;
670 
671 	/*
672 	 * The spec says it "may be problematic" if CPU x registers the VPA of
673 	 * CPU y. We should never do that, but wail if we ever do.
674 	 */
675 	WARN_ON(cpu != smp_processor_id());
676 
677 	if (cpu_has_feature(CPU_FTR_ALTIVEC))
678 		lppaca_of(cpu).vmxregs_in_use = 1;
679 
680 	if (cpu_has_feature(CPU_FTR_ARCH_207S))
681 		lppaca_of(cpu).ebb_regs_in_use = 1;
682 
683 	addr = __pa(&lppaca_of(cpu));
684 	ret = register_vpa(hwcpu, addr);
685 
686 	if (ret) {
687 		pr_err("WARNING: VPA registration for cpu %d (hw %d) of area "
688 		       "%lx failed with %ld\n", cpu, hwcpu, addr, ret);
689 		return;
690 	}
691 
692 #ifdef CONFIG_PPC_BOOK3S_64
693 	/*
694 	 * PAPR says this feature is SLB-Buffer but firmware never
695 	 * reports that.  All SPLPAR support SLB shadow buffer.
696 	 */
697 	if (!radix_enabled() && firmware_has_feature(FW_FEATURE_SPLPAR)) {
698 		addr = __pa(paca_ptrs[cpu]->slb_shadow_ptr);
699 		ret = register_slb_shadow(hwcpu, addr);
700 		if (ret)
701 			pr_err("WARNING: SLB shadow buffer registration for "
702 			       "cpu %d (hw %d) of area %lx failed with %ld\n",
703 			       cpu, hwcpu, addr, ret);
704 	}
705 #endif /* CONFIG_PPC_BOOK3S_64 */
706 
707 	/*
708 	 * Register dispatch trace log, if one has been allocated.
709 	 */
710 	register_dtl_buffer(cpu);
711 }
712 
713 #ifdef CONFIG_PPC_BOOK3S_64
714 
715 static long pSeries_lpar_hpte_insert(unsigned long hpte_group,
716 				     unsigned long vpn, unsigned long pa,
717 				     unsigned long rflags, unsigned long vflags,
718 				     int psize, int apsize, int ssize)
719 {
720 	unsigned long lpar_rc;
721 	unsigned long flags;
722 	unsigned long slot;
723 	unsigned long hpte_v, hpte_r;
724 
725 	if (!(vflags & HPTE_V_BOLTED))
726 		pr_devel("hpte_insert(group=%lx, vpn=%016lx, "
727 			 "pa=%016lx, rflags=%lx, vflags=%lx, psize=%d)\n",
728 			 hpte_group, vpn,  pa, rflags, vflags, psize);
729 
730 	hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
731 	hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
732 
733 	if (!(vflags & HPTE_V_BOLTED))
734 		pr_devel(" hpte_v=%016lx, hpte_r=%016lx\n", hpte_v, hpte_r);
735 
736 	/* Now fill in the actual HPTE */
737 	/* Set CEC cookie to 0         */
738 	/* Zero page = 0               */
739 	/* I-cache Invalidate = 0      */
740 	/* I-cache synchronize = 0     */
741 	/* Exact = 0                   */
742 	flags = 0;
743 
744 	if (firmware_has_feature(FW_FEATURE_XCMO) && !(hpte_r & HPTE_R_N))
745 		flags |= H_COALESCE_CAND;
746 
747 	lpar_rc = plpar_pte_enter(flags, hpte_group, hpte_v, hpte_r, &slot);
748 	if (unlikely(lpar_rc == H_PTEG_FULL)) {
749 		pr_devel("Hash table group is full\n");
750 		return -1;
751 	}
752 
753 	/*
754 	 * Since we try and ioremap PHBs we don't own, the pte insert
755 	 * will fail. However we must catch the failure in hash_page
756 	 * or we will loop forever, so return -2 in this case.
757 	 */
758 	if (unlikely(lpar_rc != H_SUCCESS)) {
759 		pr_err("Failed hash pte insert with error %ld\n", lpar_rc);
760 		return -2;
761 	}
762 	if (!(vflags & HPTE_V_BOLTED))
763 		pr_devel(" -> slot: %lu\n", slot & 7);
764 
765 	/* Because of iSeries, we have to pass down the secondary
766 	 * bucket bit here as well
767 	 */
768 	return (slot & 7) | (!!(vflags & HPTE_V_SECONDARY) << 3);
769 }
770 
771 static DEFINE_SPINLOCK(pSeries_lpar_tlbie_lock);
772 
773 static long pSeries_lpar_hpte_remove(unsigned long hpte_group)
774 {
775 	unsigned long slot_offset;
776 	unsigned long lpar_rc;
777 	int i;
778 	unsigned long dummy1, dummy2;
779 
780 	/* pick a random slot to start at */
781 	slot_offset = mftb() & 0x7;
782 
783 	for (i = 0; i < HPTES_PER_GROUP; i++) {
784 
785 		/* don't remove a bolted entry */
786 		lpar_rc = plpar_pte_remove(H_ANDCOND, hpte_group + slot_offset,
787 					   HPTE_V_BOLTED, &dummy1, &dummy2);
788 		if (lpar_rc == H_SUCCESS)
789 			return i;
790 
791 		/*
792 		 * The test for adjunct partition is performed before the
793 		 * ANDCOND test.  H_RESOURCE may be returned, so we need to
794 		 * check for that as well.
795 		 */
796 		BUG_ON(lpar_rc != H_NOT_FOUND && lpar_rc != H_RESOURCE);
797 
798 		slot_offset++;
799 		slot_offset &= 0x7;
800 	}
801 
802 	return -1;
803 }
804 
805 /* Called during kexec sequence with MMU off */
806 static notrace void manual_hpte_clear_all(void)
807 {
808 	unsigned long size_bytes = 1UL << ppc64_pft_size;
809 	unsigned long hpte_count = size_bytes >> 4;
810 	struct {
811 		unsigned long pteh;
812 		unsigned long ptel;
813 	} ptes[4];
814 	long lpar_rc;
815 	unsigned long i, j;
816 
817 	/* Read in batches of 4,
818 	 * invalidate only valid entries not in the VRMA
819 	 * hpte_count will be a multiple of 4
820          */
821 	for (i = 0; i < hpte_count; i += 4) {
822 		lpar_rc = plpar_pte_read_4_raw(0, i, (void *)ptes);
823 		if (lpar_rc != H_SUCCESS) {
824 			pr_info("Failed to read hash page table at %ld err %ld\n",
825 				i, lpar_rc);
826 			continue;
827 		}
828 		for (j = 0; j < 4; j++){
829 			if ((ptes[j].pteh & HPTE_V_VRMA_MASK) ==
830 				HPTE_V_VRMA_MASK)
831 				continue;
832 			if (ptes[j].pteh & HPTE_V_VALID)
833 				plpar_pte_remove_raw(0, i + j, 0,
834 					&(ptes[j].pteh), &(ptes[j].ptel));
835 		}
836 	}
837 }
838 
839 /* Called during kexec sequence with MMU off */
840 static notrace int hcall_hpte_clear_all(void)
841 {
842 	int rc;
843 
844 	do {
845 		rc = plpar_hcall_norets(H_CLEAR_HPT);
846 	} while (rc == H_CONTINUE);
847 
848 	return rc;
849 }
850 
851 /* Called during kexec sequence with MMU off */
852 static notrace void pseries_hpte_clear_all(void)
853 {
854 	int rc;
855 
856 	rc = hcall_hpte_clear_all();
857 	if (rc != H_SUCCESS)
858 		manual_hpte_clear_all();
859 
860 #ifdef __LITTLE_ENDIAN__
861 	/*
862 	 * Reset exceptions to big endian.
863 	 *
864 	 * FIXME this is a hack for kexec, we need to reset the exception
865 	 * endian before starting the new kernel and this is a convenient place
866 	 * to do it.
867 	 *
868 	 * This is also called on boot when a fadump happens. In that case we
869 	 * must not change the exception endian mode.
870 	 */
871 	if (firmware_has_feature(FW_FEATURE_SET_MODE) && !is_fadump_active())
872 		pseries_big_endian_exceptions();
873 #endif
874 }
875 
876 /*
877  * NOTE: for updatepp ops we are fortunate that the linux "newpp" bits and
878  * the low 3 bits of flags happen to line up.  So no transform is needed.
879  * We can probably optimize here and assume the high bits of newpp are
880  * already zero.  For now I am paranoid.
881  */
882 static long pSeries_lpar_hpte_updatepp(unsigned long slot,
883 				       unsigned long newpp,
884 				       unsigned long vpn,
885 				       int psize, int apsize,
886 				       int ssize, unsigned long inv_flags)
887 {
888 	unsigned long lpar_rc;
889 	unsigned long flags;
890 	unsigned long want_v;
891 
892 	want_v = hpte_encode_avpn(vpn, psize, ssize);
893 
894 	flags = (newpp & (HPTE_R_PP | HPTE_R_N | HPTE_R_KEY_LO)) | H_AVPN;
895 	flags |= (newpp & HPTE_R_KEY_HI) >> 48;
896 	if (mmu_has_feature(MMU_FTR_KERNEL_RO))
897 		/* Move pp0 into bit 8 (IBM 55) */
898 		flags |= (newpp & HPTE_R_PP0) >> 55;
899 
900 	pr_devel("    update: avpnv=%016lx, hash=%016lx, f=%lx, psize: %d ...",
901 		 want_v, slot, flags, psize);
902 
903 	lpar_rc = plpar_pte_protect(flags, slot, want_v);
904 
905 	if (lpar_rc == H_NOT_FOUND) {
906 		pr_devel("not found !\n");
907 		return -1;
908 	}
909 
910 	pr_devel("ok\n");
911 
912 	BUG_ON(lpar_rc != H_SUCCESS);
913 
914 	return 0;
915 }
916 
917 static long __pSeries_lpar_hpte_find(unsigned long want_v, unsigned long hpte_group)
918 {
919 	long lpar_rc;
920 	unsigned long i, j;
921 	struct {
922 		unsigned long pteh;
923 		unsigned long ptel;
924 	} ptes[4];
925 
926 	for (i = 0; i < HPTES_PER_GROUP; i += 4, hpte_group += 4) {
927 
928 		lpar_rc = plpar_pte_read_4(0, hpte_group, (void *)ptes);
929 		if (lpar_rc != H_SUCCESS) {
930 			pr_info("Failed to read hash page table at %ld err %ld\n",
931 				hpte_group, lpar_rc);
932 			continue;
933 		}
934 
935 		for (j = 0; j < 4; j++) {
936 			if (HPTE_V_COMPARE(ptes[j].pteh, want_v) &&
937 			    (ptes[j].pteh & HPTE_V_VALID))
938 				return i + j;
939 		}
940 	}
941 
942 	return -1;
943 }
944 
945 static long pSeries_lpar_hpte_find(unsigned long vpn, int psize, int ssize)
946 {
947 	long slot;
948 	unsigned long hash;
949 	unsigned long want_v;
950 	unsigned long hpte_group;
951 
952 	hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
953 	want_v = hpte_encode_avpn(vpn, psize, ssize);
954 
955 	/*
956 	 * We try to keep bolted entries always in primary hash
957 	 * But in some case we can find them in secondary too.
958 	 */
959 	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
960 	slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
961 	if (slot < 0) {
962 		/* Try in secondary */
963 		hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
964 		slot = __pSeries_lpar_hpte_find(want_v, hpte_group);
965 		if (slot < 0)
966 			return -1;
967 	}
968 	return hpte_group + slot;
969 }
970 
971 static void pSeries_lpar_hpte_updateboltedpp(unsigned long newpp,
972 					     unsigned long ea,
973 					     int psize, int ssize)
974 {
975 	unsigned long vpn;
976 	unsigned long lpar_rc, slot, vsid, flags;
977 
978 	vsid = get_kernel_vsid(ea, ssize);
979 	vpn = hpt_vpn(ea, vsid, ssize);
980 
981 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
982 	BUG_ON(slot == -1);
983 
984 	flags = newpp & (HPTE_R_PP | HPTE_R_N);
985 	if (mmu_has_feature(MMU_FTR_KERNEL_RO))
986 		/* Move pp0 into bit 8 (IBM 55) */
987 		flags |= (newpp & HPTE_R_PP0) >> 55;
988 
989 	flags |= ((newpp & HPTE_R_KEY_HI) >> 48) | (newpp & HPTE_R_KEY_LO);
990 
991 	lpar_rc = plpar_pte_protect(flags, slot, 0);
992 
993 	BUG_ON(lpar_rc != H_SUCCESS);
994 }
995 
996 static void pSeries_lpar_hpte_invalidate(unsigned long slot, unsigned long vpn,
997 					 int psize, int apsize,
998 					 int ssize, int local)
999 {
1000 	unsigned long want_v;
1001 	unsigned long lpar_rc;
1002 	unsigned long dummy1, dummy2;
1003 
1004 	pr_devel("    inval : slot=%lx, vpn=%016lx, psize: %d, local: %d\n",
1005 		 slot, vpn, psize, local);
1006 
1007 	want_v = hpte_encode_avpn(vpn, psize, ssize);
1008 	lpar_rc = plpar_pte_remove(H_AVPN, slot, want_v, &dummy1, &dummy2);
1009 	if (lpar_rc == H_NOT_FOUND)
1010 		return;
1011 
1012 	BUG_ON(lpar_rc != H_SUCCESS);
1013 }
1014 
1015 
1016 /*
1017  * As defined in the PAPR's section 14.5.4.1.8
1018  * The control mask doesn't include the returned reference and change bit from
1019  * the processed PTE.
1020  */
1021 #define HBLKR_AVPN		0x0100000000000000UL
1022 #define HBLKR_CTRL_MASK		0xf800000000000000UL
1023 #define HBLKR_CTRL_SUCCESS	0x8000000000000000UL
1024 #define HBLKR_CTRL_ERRNOTFOUND	0x8800000000000000UL
1025 #define HBLKR_CTRL_ERRBUSY	0xa000000000000000UL
1026 
1027 /*
1028  * Returned true if we are supporting this block size for the specified segment
1029  * base page size and actual page size.
1030  *
1031  * Currently, we only support 8 size block.
1032  */
1033 static inline bool is_supported_hlbkrm(int bpsize, int psize)
1034 {
1035 	return (hblkrm_size[bpsize][psize] == HBLKRM_SUPPORTED_BLOCK_SIZE);
1036 }
1037 
1038 /**
1039  * H_BLOCK_REMOVE caller.
1040  * @idx should point to the latest @param entry set with a PTEX.
1041  * If PTE cannot be processed because another CPUs has already locked that
1042  * group, those entries are put back in @param starting at index 1.
1043  * If entries has to be retried and @retry_busy is set to true, these entries
1044  * are retried until success. If @retry_busy is set to false, the returned
1045  * is the number of entries yet to process.
1046  */
1047 static unsigned long call_block_remove(unsigned long idx, unsigned long *param,
1048 				       bool retry_busy)
1049 {
1050 	unsigned long i, rc, new_idx;
1051 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1052 
1053 	if (idx < 2) {
1054 		pr_warn("Unexpected empty call to H_BLOCK_REMOVE");
1055 		return 0;
1056 	}
1057 again:
1058 	new_idx = 0;
1059 	if (idx > PLPAR_HCALL9_BUFSIZE) {
1060 		pr_err("Too many PTEs (%lu) for H_BLOCK_REMOVE", idx);
1061 		idx = PLPAR_HCALL9_BUFSIZE;
1062 	} else if (idx < PLPAR_HCALL9_BUFSIZE)
1063 		param[idx] = HBR_END;
1064 
1065 	rc = plpar_hcall9(H_BLOCK_REMOVE, retbuf,
1066 			  param[0], /* AVA */
1067 			  param[1],  param[2],  param[3],  param[4], /* TS0-7 */
1068 			  param[5],  param[6],  param[7],  param[8]);
1069 	if (rc == H_SUCCESS)
1070 		return 0;
1071 
1072 	BUG_ON(rc != H_PARTIAL);
1073 
1074 	/* Check that the unprocessed entries were 'not found' or 'busy' */
1075 	for (i = 0; i < idx-1; i++) {
1076 		unsigned long ctrl = retbuf[i] & HBLKR_CTRL_MASK;
1077 
1078 		if (ctrl == HBLKR_CTRL_ERRBUSY) {
1079 			param[++new_idx] = param[i+1];
1080 			continue;
1081 		}
1082 
1083 		BUG_ON(ctrl != HBLKR_CTRL_SUCCESS
1084 		       && ctrl != HBLKR_CTRL_ERRNOTFOUND);
1085 	}
1086 
1087 	/*
1088 	 * If there were entries found busy, retry these entries if requested,
1089 	 * of if all the entries have to be retried.
1090 	 */
1091 	if (new_idx && (retry_busy || new_idx == (PLPAR_HCALL9_BUFSIZE-1))) {
1092 		idx = new_idx + 1;
1093 		goto again;
1094 	}
1095 
1096 	return new_idx;
1097 }
1098 
1099 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1100 /*
1101  * Limit iterations holding pSeries_lpar_tlbie_lock to 3. We also need
1102  * to make sure that we avoid bouncing the hypervisor tlbie lock.
1103  */
1104 #define PPC64_HUGE_HPTE_BATCH 12
1105 
1106 static void hugepage_block_invalidate(unsigned long *slot, unsigned long *vpn,
1107 				      int count, int psize, int ssize)
1108 {
1109 	unsigned long param[PLPAR_HCALL9_BUFSIZE];
1110 	unsigned long shift, current_vpgb, vpgb;
1111 	int i, pix = 0;
1112 
1113 	shift = mmu_psize_defs[psize].shift;
1114 
1115 	for (i = 0; i < count; i++) {
1116 		/*
1117 		 * Shifting 3 bits more on the right to get a
1118 		 * 8 pages aligned virtual addresse.
1119 		 */
1120 		vpgb = (vpn[i] >> (shift - VPN_SHIFT + 3));
1121 		if (!pix || vpgb != current_vpgb) {
1122 			/*
1123 			 * Need to start a new 8 pages block, flush
1124 			 * the current one if needed.
1125 			 */
1126 			if (pix)
1127 				(void)call_block_remove(pix, param, true);
1128 			current_vpgb = vpgb;
1129 			param[0] = hpte_encode_avpn(vpn[i], psize, ssize);
1130 			pix = 1;
1131 		}
1132 
1133 		param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot[i];
1134 		if (pix == PLPAR_HCALL9_BUFSIZE) {
1135 			pix = call_block_remove(pix, param, false);
1136 			/*
1137 			 * pix = 0 means that all the entries were
1138 			 * removed, we can start a new block.
1139 			 * Otherwise, this means that there are entries
1140 			 * to retry, and pix points to latest one, so
1141 			 * we should increment it and try to continue
1142 			 * the same block.
1143 			 */
1144 			if (pix)
1145 				pix++;
1146 		}
1147 	}
1148 	if (pix)
1149 		(void)call_block_remove(pix, param, true);
1150 }
1151 
1152 static void hugepage_bulk_invalidate(unsigned long *slot, unsigned long *vpn,
1153 				     int count, int psize, int ssize)
1154 {
1155 	unsigned long param[PLPAR_HCALL9_BUFSIZE];
1156 	int i = 0, pix = 0, rc;
1157 
1158 	for (i = 0; i < count; i++) {
1159 
1160 		if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1161 			pSeries_lpar_hpte_invalidate(slot[i], vpn[i], psize, 0,
1162 						     ssize, 0);
1163 		} else {
1164 			param[pix] = HBR_REQUEST | HBR_AVPN | slot[i];
1165 			param[pix+1] = hpte_encode_avpn(vpn[i], psize, ssize);
1166 			pix += 2;
1167 			if (pix == 8) {
1168 				rc = plpar_hcall9(H_BULK_REMOVE, param,
1169 						  param[0], param[1], param[2],
1170 						  param[3], param[4], param[5],
1171 						  param[6], param[7]);
1172 				BUG_ON(rc != H_SUCCESS);
1173 				pix = 0;
1174 			}
1175 		}
1176 	}
1177 	if (pix) {
1178 		param[pix] = HBR_END;
1179 		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1180 				  param[2], param[3], param[4], param[5],
1181 				  param[6], param[7]);
1182 		BUG_ON(rc != H_SUCCESS);
1183 	}
1184 }
1185 
1186 static inline void __pSeries_lpar_hugepage_invalidate(unsigned long *slot,
1187 						      unsigned long *vpn,
1188 						      int count, int psize,
1189 						      int ssize)
1190 {
1191 	unsigned long flags = 0;
1192 	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1193 
1194 	if (lock_tlbie)
1195 		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1196 
1197 	/* Assuming THP size is 16M */
1198 	if (is_supported_hlbkrm(psize, MMU_PAGE_16M))
1199 		hugepage_block_invalidate(slot, vpn, count, psize, ssize);
1200 	else
1201 		hugepage_bulk_invalidate(slot, vpn, count, psize, ssize);
1202 
1203 	if (lock_tlbie)
1204 		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1205 }
1206 
1207 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1208 					     unsigned long addr,
1209 					     unsigned char *hpte_slot_array,
1210 					     int psize, int ssize, int local)
1211 {
1212 	int i, index = 0;
1213 	unsigned long s_addr = addr;
1214 	unsigned int max_hpte_count, valid;
1215 	unsigned long vpn_array[PPC64_HUGE_HPTE_BATCH];
1216 	unsigned long slot_array[PPC64_HUGE_HPTE_BATCH];
1217 	unsigned long shift, hidx, vpn = 0, hash, slot;
1218 
1219 	shift = mmu_psize_defs[psize].shift;
1220 	max_hpte_count = 1U << (PMD_SHIFT - shift);
1221 
1222 	for (i = 0; i < max_hpte_count; i++) {
1223 		valid = hpte_valid(hpte_slot_array, i);
1224 		if (!valid)
1225 			continue;
1226 		hidx =  hpte_hash_index(hpte_slot_array, i);
1227 
1228 		/* get the vpn */
1229 		addr = s_addr + (i * (1ul << shift));
1230 		vpn = hpt_vpn(addr, vsid, ssize);
1231 		hash = hpt_hash(vpn, shift, ssize);
1232 		if (hidx & _PTEIDX_SECONDARY)
1233 			hash = ~hash;
1234 
1235 		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1236 		slot += hidx & _PTEIDX_GROUP_IX;
1237 
1238 		slot_array[index] = slot;
1239 		vpn_array[index] = vpn;
1240 		if (index == PPC64_HUGE_HPTE_BATCH - 1) {
1241 			/*
1242 			 * Now do a bluk invalidate
1243 			 */
1244 			__pSeries_lpar_hugepage_invalidate(slot_array,
1245 							   vpn_array,
1246 							   PPC64_HUGE_HPTE_BATCH,
1247 							   psize, ssize);
1248 			index = 0;
1249 		} else
1250 			index++;
1251 	}
1252 	if (index)
1253 		__pSeries_lpar_hugepage_invalidate(slot_array, vpn_array,
1254 						   index, psize, ssize);
1255 }
1256 #else
1257 static void pSeries_lpar_hugepage_invalidate(unsigned long vsid,
1258 					     unsigned long addr,
1259 					     unsigned char *hpte_slot_array,
1260 					     int psize, int ssize, int local)
1261 {
1262 	WARN(1, "%s called without THP support\n", __func__);
1263 }
1264 #endif
1265 
1266 static int pSeries_lpar_hpte_removebolted(unsigned long ea,
1267 					  int psize, int ssize)
1268 {
1269 	unsigned long vpn;
1270 	unsigned long slot, vsid;
1271 
1272 	vsid = get_kernel_vsid(ea, ssize);
1273 	vpn = hpt_vpn(ea, vsid, ssize);
1274 
1275 	slot = pSeries_lpar_hpte_find(vpn, psize, ssize);
1276 	if (slot == -1)
1277 		return -ENOENT;
1278 
1279 	/*
1280 	 * lpar doesn't use the passed actual page size
1281 	 */
1282 	pSeries_lpar_hpte_invalidate(slot, vpn, psize, 0, ssize, 0);
1283 	return 0;
1284 }
1285 
1286 
1287 static inline unsigned long compute_slot(real_pte_t pte,
1288 					 unsigned long vpn,
1289 					 unsigned long index,
1290 					 unsigned long shift,
1291 					 int ssize)
1292 {
1293 	unsigned long slot, hash, hidx;
1294 
1295 	hash = hpt_hash(vpn, shift, ssize);
1296 	hidx = __rpte_to_hidx(pte, index);
1297 	if (hidx & _PTEIDX_SECONDARY)
1298 		hash = ~hash;
1299 	slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1300 	slot += hidx & _PTEIDX_GROUP_IX;
1301 	return slot;
1302 }
1303 
1304 /**
1305  * The hcall H_BLOCK_REMOVE implies that the virtual pages to processed are
1306  * "all within the same naturally aligned 8 page virtual address block".
1307  */
1308 static void do_block_remove(unsigned long number, struct ppc64_tlb_batch *batch,
1309 			    unsigned long *param)
1310 {
1311 	unsigned long vpn;
1312 	unsigned long i, pix = 0;
1313 	unsigned long index, shift, slot, current_vpgb, vpgb;
1314 	real_pte_t pte;
1315 	int psize, ssize;
1316 
1317 	psize = batch->psize;
1318 	ssize = batch->ssize;
1319 
1320 	for (i = 0; i < number; i++) {
1321 		vpn = batch->vpn[i];
1322 		pte = batch->pte[i];
1323 		pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1324 			/*
1325 			 * Shifting 3 bits more on the right to get a
1326 			 * 8 pages aligned virtual addresse.
1327 			 */
1328 			vpgb = (vpn >> (shift - VPN_SHIFT + 3));
1329 			if (!pix || vpgb != current_vpgb) {
1330 				/*
1331 				 * Need to start a new 8 pages block, flush
1332 				 * the current one if needed.
1333 				 */
1334 				if (pix)
1335 					(void)call_block_remove(pix, param,
1336 								true);
1337 				current_vpgb = vpgb;
1338 				param[0] = hpte_encode_avpn(vpn, psize,
1339 							    ssize);
1340 				pix = 1;
1341 			}
1342 
1343 			slot = compute_slot(pte, vpn, index, shift, ssize);
1344 			param[pix++] = HBR_REQUEST | HBLKR_AVPN | slot;
1345 
1346 			if (pix == PLPAR_HCALL9_BUFSIZE) {
1347 				pix = call_block_remove(pix, param, false);
1348 				/*
1349 				 * pix = 0 means that all the entries were
1350 				 * removed, we can start a new block.
1351 				 * Otherwise, this means that there are entries
1352 				 * to retry, and pix points to latest one, so
1353 				 * we should increment it and try to continue
1354 				 * the same block.
1355 				 */
1356 				if (pix)
1357 					pix++;
1358 			}
1359 		} pte_iterate_hashed_end();
1360 	}
1361 
1362 	if (pix)
1363 		(void)call_block_remove(pix, param, true);
1364 }
1365 
1366 /*
1367  * TLB Block Invalidate Characteristics
1368  *
1369  * These characteristics define the size of the block the hcall H_BLOCK_REMOVE
1370  * is able to process for each couple segment base page size, actual page size.
1371  *
1372  * The ibm,get-system-parameter properties is returning a buffer with the
1373  * following layout:
1374  *
1375  * [ 2 bytes size of the RTAS buffer (excluding these 2 bytes) ]
1376  * -----------------
1377  * TLB Block Invalidate Specifiers:
1378  * [ 1 byte LOG base 2 of the TLB invalidate block size being specified ]
1379  * [ 1 byte Number of page sizes (N) that are supported for the specified
1380  *          TLB invalidate block size ]
1381  * [ 1 byte Encoded segment base page size and actual page size
1382  *          MSB=0 means 4k segment base page size and actual page size
1383  *          MSB=1 the penc value in mmu_psize_def ]
1384  * ...
1385  * -----------------
1386  * Next TLB Block Invalidate Specifiers...
1387  * -----------------
1388  * [ 0 ]
1389  */
1390 static inline void set_hblkrm_bloc_size(int bpsize, int psize,
1391 					unsigned int block_size)
1392 {
1393 	if (block_size > hblkrm_size[bpsize][psize])
1394 		hblkrm_size[bpsize][psize] = block_size;
1395 }
1396 
1397 /*
1398  * Decode the Encoded segment base page size and actual page size.
1399  * PAPR specifies:
1400  *   - bit 7 is the L bit
1401  *   - bits 0-5 are the penc value
1402  * If the L bit is 0, this means 4K segment base page size and actual page size
1403  * otherwise the penc value should be read.
1404  */
1405 #define HBLKRM_L_MASK		0x80
1406 #define HBLKRM_PENC_MASK	0x3f
1407 static inline void __init check_lp_set_hblkrm(unsigned int lp,
1408 					      unsigned int block_size)
1409 {
1410 	unsigned int bpsize, psize;
1411 
1412 	/* First, check the L bit, if not set, this means 4K */
1413 	if ((lp & HBLKRM_L_MASK) == 0) {
1414 		set_hblkrm_bloc_size(MMU_PAGE_4K, MMU_PAGE_4K, block_size);
1415 		return;
1416 	}
1417 
1418 	lp &= HBLKRM_PENC_MASK;
1419 	for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++) {
1420 		struct mmu_psize_def *def = &mmu_psize_defs[bpsize];
1421 
1422 		for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1423 			if (def->penc[psize] == lp) {
1424 				set_hblkrm_bloc_size(bpsize, psize, block_size);
1425 				return;
1426 			}
1427 		}
1428 	}
1429 }
1430 
1431 #define SPLPAR_TLB_BIC_TOKEN		50
1432 
1433 /*
1434  * The size of the TLB Block Invalidate Characteristics is variable. But at the
1435  * maximum it will be the number of possible page sizes *2 + 10 bytes.
1436  * Currently MMU_PAGE_COUNT is 16, which means 42 bytes. Use a cache line size
1437  * (128 bytes) for the buffer to get plenty of space.
1438  */
1439 #define SPLPAR_TLB_BIC_MAXLENGTH	128
1440 
1441 void __init pseries_lpar_read_hblkrm_characteristics(void)
1442 {
1443 	unsigned char local_buffer[SPLPAR_TLB_BIC_MAXLENGTH];
1444 	int call_status, len, idx, bpsize;
1445 
1446 	if (!firmware_has_feature(FW_FEATURE_BLOCK_REMOVE))
1447 		return;
1448 
1449 	spin_lock(&rtas_data_buf_lock);
1450 	memset(rtas_data_buf, 0, RTAS_DATA_BUF_SIZE);
1451 	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,
1452 				NULL,
1453 				SPLPAR_TLB_BIC_TOKEN,
1454 				__pa(rtas_data_buf),
1455 				RTAS_DATA_BUF_SIZE);
1456 	memcpy(local_buffer, rtas_data_buf, SPLPAR_TLB_BIC_MAXLENGTH);
1457 	local_buffer[SPLPAR_TLB_BIC_MAXLENGTH - 1] = '\0';
1458 	spin_unlock(&rtas_data_buf_lock);
1459 
1460 	if (call_status != 0) {
1461 		pr_warn("%s %s Error calling get-system-parameter (0x%x)\n",
1462 			__FILE__, __func__, call_status);
1463 		return;
1464 	}
1465 
1466 	/*
1467 	 * The first two (2) bytes of the data in the buffer are the length of
1468 	 * the returned data, not counting these first two (2) bytes.
1469 	 */
1470 	len = be16_to_cpu(*((u16 *)local_buffer)) + 2;
1471 	if (len > SPLPAR_TLB_BIC_MAXLENGTH) {
1472 		pr_warn("%s too large returned buffer %d", __func__, len);
1473 		return;
1474 	}
1475 
1476 	idx = 2;
1477 	while (idx < len) {
1478 		u8 block_shift = local_buffer[idx++];
1479 		u32 block_size;
1480 		unsigned int npsize;
1481 
1482 		if (!block_shift)
1483 			break;
1484 
1485 		block_size = 1 << block_shift;
1486 
1487 		for (npsize = local_buffer[idx++];
1488 		     npsize > 0 && idx < len; npsize--)
1489 			check_lp_set_hblkrm((unsigned int) local_buffer[idx++],
1490 					    block_size);
1491 	}
1492 
1493 	for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
1494 		for (idx = 0; idx < MMU_PAGE_COUNT; idx++)
1495 			if (hblkrm_size[bpsize][idx])
1496 				pr_info("H_BLOCK_REMOVE supports base psize:%d psize:%d block size:%d",
1497 					bpsize, idx, hblkrm_size[bpsize][idx]);
1498 }
1499 
1500 /*
1501  * Take a spinlock around flushes to avoid bouncing the hypervisor tlbie
1502  * lock.
1503  */
1504 static void pSeries_lpar_flush_hash_range(unsigned long number, int local)
1505 {
1506 	unsigned long vpn;
1507 	unsigned long i, pix, rc;
1508 	unsigned long flags = 0;
1509 	struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
1510 	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
1511 	unsigned long param[PLPAR_HCALL9_BUFSIZE];
1512 	unsigned long index, shift, slot;
1513 	real_pte_t pte;
1514 	int psize, ssize;
1515 
1516 	if (lock_tlbie)
1517 		spin_lock_irqsave(&pSeries_lpar_tlbie_lock, flags);
1518 
1519 	if (is_supported_hlbkrm(batch->psize, batch->psize)) {
1520 		do_block_remove(number, batch, param);
1521 		goto out;
1522 	}
1523 
1524 	psize = batch->psize;
1525 	ssize = batch->ssize;
1526 	pix = 0;
1527 	for (i = 0; i < number; i++) {
1528 		vpn = batch->vpn[i];
1529 		pte = batch->pte[i];
1530 		pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1531 			slot = compute_slot(pte, vpn, index, shift, ssize);
1532 			if (!firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1533 				/*
1534 				 * lpar doesn't use the passed actual page size
1535 				 */
1536 				pSeries_lpar_hpte_invalidate(slot, vpn, psize,
1537 							     0, ssize, local);
1538 			} else {
1539 				param[pix] = HBR_REQUEST | HBR_AVPN | slot;
1540 				param[pix+1] = hpte_encode_avpn(vpn, psize,
1541 								ssize);
1542 				pix += 2;
1543 				if (pix == 8) {
1544 					rc = plpar_hcall9(H_BULK_REMOVE, param,
1545 						param[0], param[1], param[2],
1546 						param[3], param[4], param[5],
1547 						param[6], param[7]);
1548 					BUG_ON(rc != H_SUCCESS);
1549 					pix = 0;
1550 				}
1551 			}
1552 		} pte_iterate_hashed_end();
1553 	}
1554 	if (pix) {
1555 		param[pix] = HBR_END;
1556 		rc = plpar_hcall9(H_BULK_REMOVE, param, param[0], param[1],
1557 				  param[2], param[3], param[4], param[5],
1558 				  param[6], param[7]);
1559 		BUG_ON(rc != H_SUCCESS);
1560 	}
1561 
1562 out:
1563 	if (lock_tlbie)
1564 		spin_unlock_irqrestore(&pSeries_lpar_tlbie_lock, flags);
1565 }
1566 
1567 static int __init disable_bulk_remove(char *str)
1568 {
1569 	if (strcmp(str, "off") == 0 &&
1570 	    firmware_has_feature(FW_FEATURE_BULK_REMOVE)) {
1571 		pr_info("Disabling BULK_REMOVE firmware feature");
1572 		powerpc_firmware_features &= ~FW_FEATURE_BULK_REMOVE;
1573 	}
1574 	return 1;
1575 }
1576 
1577 __setup("bulk_remove=", disable_bulk_remove);
1578 
1579 #define HPT_RESIZE_TIMEOUT	10000 /* ms */
1580 
1581 struct hpt_resize_state {
1582 	unsigned long shift;
1583 	int commit_rc;
1584 };
1585 
1586 static int pseries_lpar_resize_hpt_commit(void *data)
1587 {
1588 	struct hpt_resize_state *state = data;
1589 
1590 	state->commit_rc = plpar_resize_hpt_commit(0, state->shift);
1591 	if (state->commit_rc != H_SUCCESS)
1592 		return -EIO;
1593 
1594 	/* Hypervisor has transitioned the HTAB, update our globals */
1595 	ppc64_pft_size = state->shift;
1596 	htab_size_bytes = 1UL << ppc64_pft_size;
1597 	htab_hash_mask = (htab_size_bytes >> 7) - 1;
1598 
1599 	return 0;
1600 }
1601 
1602 /*
1603  * Must be called in process context. The caller must hold the
1604  * cpus_lock.
1605  */
1606 static int pseries_lpar_resize_hpt(unsigned long shift)
1607 {
1608 	struct hpt_resize_state state = {
1609 		.shift = shift,
1610 		.commit_rc = H_FUNCTION,
1611 	};
1612 	unsigned int delay, total_delay = 0;
1613 	int rc;
1614 	ktime_t t0, t1, t2;
1615 
1616 	might_sleep();
1617 
1618 	if (!firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1619 		return -ENODEV;
1620 
1621 	pr_info("Attempting to resize HPT to shift %lu\n", shift);
1622 
1623 	t0 = ktime_get();
1624 
1625 	rc = plpar_resize_hpt_prepare(0, shift);
1626 	while (H_IS_LONG_BUSY(rc)) {
1627 		delay = get_longbusy_msecs(rc);
1628 		total_delay += delay;
1629 		if (total_delay > HPT_RESIZE_TIMEOUT) {
1630 			/* prepare with shift==0 cancels an in-progress resize */
1631 			rc = plpar_resize_hpt_prepare(0, 0);
1632 			if (rc != H_SUCCESS)
1633 				pr_warn("Unexpected error %d cancelling timed out HPT resize\n",
1634 				       rc);
1635 			return -ETIMEDOUT;
1636 		}
1637 		msleep(delay);
1638 		rc = plpar_resize_hpt_prepare(0, shift);
1639 	}
1640 
1641 	switch (rc) {
1642 	case H_SUCCESS:
1643 		/* Continue on */
1644 		break;
1645 
1646 	case H_PARAMETER:
1647 		pr_warn("Invalid argument from H_RESIZE_HPT_PREPARE\n");
1648 		return -EINVAL;
1649 	case H_RESOURCE:
1650 		pr_warn("Operation not permitted from H_RESIZE_HPT_PREPARE\n");
1651 		return -EPERM;
1652 	default:
1653 		pr_warn("Unexpected error %d from H_RESIZE_HPT_PREPARE\n", rc);
1654 		return -EIO;
1655 	}
1656 
1657 	t1 = ktime_get();
1658 
1659 	rc = stop_machine_cpuslocked(pseries_lpar_resize_hpt_commit,
1660 				     &state, NULL);
1661 
1662 	t2 = ktime_get();
1663 
1664 	if (rc != 0) {
1665 		switch (state.commit_rc) {
1666 		case H_PTEG_FULL:
1667 			return -ENOSPC;
1668 
1669 		default:
1670 			pr_warn("Unexpected error %d from H_RESIZE_HPT_COMMIT\n",
1671 				state.commit_rc);
1672 			return -EIO;
1673 		};
1674 	}
1675 
1676 	pr_info("HPT resize to shift %lu complete (%lld ms / %lld ms)\n",
1677 		shift, (long long) ktime_ms_delta(t1, t0),
1678 		(long long) ktime_ms_delta(t2, t1));
1679 
1680 	return 0;
1681 }
1682 
1683 static int pseries_lpar_register_process_table(unsigned long base,
1684 			unsigned long page_size, unsigned long table_size)
1685 {
1686 	long rc;
1687 	unsigned long flags = 0;
1688 
1689 	if (table_size)
1690 		flags |= PROC_TABLE_NEW;
1691 	if (radix_enabled()) {
1692 		flags |= PROC_TABLE_RADIX;
1693 		if (mmu_has_feature(MMU_FTR_GTSE))
1694 			flags |= PROC_TABLE_GTSE;
1695 	} else
1696 		flags |= PROC_TABLE_HPT_SLB;
1697 	for (;;) {
1698 		rc = plpar_hcall_norets(H_REGISTER_PROC_TBL, flags, base,
1699 					page_size, table_size);
1700 		if (!H_IS_LONG_BUSY(rc))
1701 			break;
1702 		mdelay(get_longbusy_msecs(rc));
1703 	}
1704 	if (rc != H_SUCCESS) {
1705 		pr_err("Failed to register process table (rc=%ld)\n", rc);
1706 		BUG();
1707 	}
1708 	return rc;
1709 }
1710 
1711 void __init hpte_init_pseries(void)
1712 {
1713 	mmu_hash_ops.hpte_invalidate	 = pSeries_lpar_hpte_invalidate;
1714 	mmu_hash_ops.hpte_updatepp	 = pSeries_lpar_hpte_updatepp;
1715 	mmu_hash_ops.hpte_updateboltedpp = pSeries_lpar_hpte_updateboltedpp;
1716 	mmu_hash_ops.hpte_insert	 = pSeries_lpar_hpte_insert;
1717 	mmu_hash_ops.hpte_remove	 = pSeries_lpar_hpte_remove;
1718 	mmu_hash_ops.hpte_removebolted   = pSeries_lpar_hpte_removebolted;
1719 	mmu_hash_ops.flush_hash_range	 = pSeries_lpar_flush_hash_range;
1720 	mmu_hash_ops.hpte_clear_all      = pseries_hpte_clear_all;
1721 	mmu_hash_ops.hugepage_invalidate = pSeries_lpar_hugepage_invalidate;
1722 
1723 	if (firmware_has_feature(FW_FEATURE_HPT_RESIZE))
1724 		mmu_hash_ops.resize_hpt = pseries_lpar_resize_hpt;
1725 
1726 	/*
1727 	 * On POWER9, we need to do a H_REGISTER_PROC_TBL hcall
1728 	 * to inform the hypervisor that we wish to use the HPT.
1729 	 */
1730 	if (cpu_has_feature(CPU_FTR_ARCH_300))
1731 		pseries_lpar_register_process_table(0, 0, 0);
1732 }
1733 
1734 #ifdef CONFIG_PPC_RADIX_MMU
1735 void radix_init_pseries(void)
1736 {
1737 	pr_info("Using radix MMU under hypervisor\n");
1738 
1739 	pseries_lpar_register_process_table(__pa(process_tb),
1740 						0, PRTB_SIZE_SHIFT - 12);
1741 }
1742 #endif
1743 
1744 #ifdef CONFIG_PPC_SMLPAR
1745 #define CMO_FREE_HINT_DEFAULT 1
1746 static int cmo_free_hint_flag = CMO_FREE_HINT_DEFAULT;
1747 
1748 static int __init cmo_free_hint(char *str)
1749 {
1750 	char *parm;
1751 	parm = strstrip(str);
1752 
1753 	if (strcasecmp(parm, "no") == 0 || strcasecmp(parm, "off") == 0) {
1754 		pr_info("%s: CMO free page hinting is not active.\n", __func__);
1755 		cmo_free_hint_flag = 0;
1756 		return 1;
1757 	}
1758 
1759 	cmo_free_hint_flag = 1;
1760 	pr_info("%s: CMO free page hinting is active.\n", __func__);
1761 
1762 	if (strcasecmp(parm, "yes") == 0 || strcasecmp(parm, "on") == 0)
1763 		return 1;
1764 
1765 	return 0;
1766 }
1767 
1768 __setup("cmo_free_hint=", cmo_free_hint);
1769 
1770 static void pSeries_set_page_state(struct page *page, int order,
1771 				   unsigned long state)
1772 {
1773 	int i, j;
1774 	unsigned long cmo_page_sz, addr;
1775 
1776 	cmo_page_sz = cmo_get_page_size();
1777 	addr = __pa((unsigned long)page_address(page));
1778 
1779 	for (i = 0; i < (1 << order); i++, addr += PAGE_SIZE) {
1780 		for (j = 0; j < PAGE_SIZE; j += cmo_page_sz)
1781 			plpar_hcall_norets(H_PAGE_INIT, state, addr + j, 0);
1782 	}
1783 }
1784 
1785 void arch_free_page(struct page *page, int order)
1786 {
1787 	if (radix_enabled())
1788 		return;
1789 	if (!cmo_free_hint_flag || !firmware_has_feature(FW_FEATURE_CMO))
1790 		return;
1791 
1792 	pSeries_set_page_state(page, order, H_PAGE_SET_UNUSED);
1793 }
1794 EXPORT_SYMBOL(arch_free_page);
1795 
1796 #endif /* CONFIG_PPC_SMLPAR */
1797 #endif /* CONFIG_PPC_BOOK3S_64 */
1798 
1799 #ifdef CONFIG_TRACEPOINTS
1800 #ifdef CONFIG_JUMP_LABEL
1801 struct static_key hcall_tracepoint_key = STATIC_KEY_INIT;
1802 
1803 int hcall_tracepoint_regfunc(void)
1804 {
1805 	static_key_slow_inc(&hcall_tracepoint_key);
1806 	return 0;
1807 }
1808 
1809 void hcall_tracepoint_unregfunc(void)
1810 {
1811 	static_key_slow_dec(&hcall_tracepoint_key);
1812 }
1813 #else
1814 /*
1815  * We optimise our hcall path by placing hcall_tracepoint_refcount
1816  * directly in the TOC so we can check if the hcall tracepoints are
1817  * enabled via a single load.
1818  */
1819 
1820 /* NB: reg/unreg are called while guarded with the tracepoints_mutex */
1821 extern long hcall_tracepoint_refcount;
1822 
1823 int hcall_tracepoint_regfunc(void)
1824 {
1825 	hcall_tracepoint_refcount++;
1826 	return 0;
1827 }
1828 
1829 void hcall_tracepoint_unregfunc(void)
1830 {
1831 	hcall_tracepoint_refcount--;
1832 }
1833 #endif
1834 
1835 /*
1836  * Keep track of hcall tracing depth and prevent recursion. Warn if any is
1837  * detected because it may indicate a problem. This will not catch all
1838  * problems with tracing code making hcalls, because the tracing might have
1839  * been invoked from a non-hcall, so the first hcall could recurse into it
1840  * without warning here, but this better than nothing.
1841  *
1842  * Hcalls with specific problems being traced should use the _notrace
1843  * plpar_hcall variants.
1844  */
1845 static DEFINE_PER_CPU(unsigned int, hcall_trace_depth);
1846 
1847 
1848 notrace void __trace_hcall_entry(unsigned long opcode, unsigned long *args)
1849 {
1850 	unsigned long flags;
1851 	unsigned int *depth;
1852 
1853 	local_irq_save(flags);
1854 
1855 	depth = this_cpu_ptr(&hcall_trace_depth);
1856 
1857 	if (WARN_ON_ONCE(*depth))
1858 		goto out;
1859 
1860 	(*depth)++;
1861 	preempt_disable();
1862 	trace_hcall_entry(opcode, args);
1863 	(*depth)--;
1864 
1865 out:
1866 	local_irq_restore(flags);
1867 }
1868 
1869 notrace void __trace_hcall_exit(long opcode, long retval, unsigned long *retbuf)
1870 {
1871 	unsigned long flags;
1872 	unsigned int *depth;
1873 
1874 	local_irq_save(flags);
1875 
1876 	depth = this_cpu_ptr(&hcall_trace_depth);
1877 
1878 	if (*depth) /* Don't warn again on the way out */
1879 		goto out;
1880 
1881 	(*depth)++;
1882 	trace_hcall_exit(opcode, retval, retbuf);
1883 	preempt_enable();
1884 	(*depth)--;
1885 
1886 out:
1887 	local_irq_restore(flags);
1888 }
1889 #endif
1890 
1891 /**
1892  * h_get_mpp
1893  * H_GET_MPP hcall returns info in 7 parms
1894  */
1895 int h_get_mpp(struct hvcall_mpp_data *mpp_data)
1896 {
1897 	int rc;
1898 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE];
1899 
1900 	rc = plpar_hcall9(H_GET_MPP, retbuf);
1901 
1902 	mpp_data->entitled_mem = retbuf[0];
1903 	mpp_data->mapped_mem = retbuf[1];
1904 
1905 	mpp_data->group_num = (retbuf[2] >> 2 * 8) & 0xffff;
1906 	mpp_data->pool_num = retbuf[2] & 0xffff;
1907 
1908 	mpp_data->mem_weight = (retbuf[3] >> 7 * 8) & 0xff;
1909 	mpp_data->unallocated_mem_weight = (retbuf[3] >> 6 * 8) & 0xff;
1910 	mpp_data->unallocated_entitlement = retbuf[3] & 0xffffffffffffUL;
1911 
1912 	mpp_data->pool_size = retbuf[4];
1913 	mpp_data->loan_request = retbuf[5];
1914 	mpp_data->backing_mem = retbuf[6];
1915 
1916 	return rc;
1917 }
1918 EXPORT_SYMBOL(h_get_mpp);
1919 
1920 int h_get_mpp_x(struct hvcall_mpp_x_data *mpp_x_data)
1921 {
1922 	int rc;
1923 	unsigned long retbuf[PLPAR_HCALL9_BUFSIZE] = { 0 };
1924 
1925 	rc = plpar_hcall9(H_GET_MPP_X, retbuf);
1926 
1927 	mpp_x_data->coalesced_bytes = retbuf[0];
1928 	mpp_x_data->pool_coalesced_bytes = retbuf[1];
1929 	mpp_x_data->pool_purr_cycles = retbuf[2];
1930 	mpp_x_data->pool_spurr_cycles = retbuf[3];
1931 
1932 	return rc;
1933 }
1934 
1935 static unsigned long vsid_unscramble(unsigned long vsid, int ssize)
1936 {
1937 	unsigned long protovsid;
1938 	unsigned long va_bits = VA_BITS;
1939 	unsigned long modinv, vsid_modulus;
1940 	unsigned long max_mod_inv, tmp_modinv;
1941 
1942 	if (!mmu_has_feature(MMU_FTR_68_BIT_VA))
1943 		va_bits = 65;
1944 
1945 	if (ssize == MMU_SEGSIZE_256M) {
1946 		modinv = VSID_MULINV_256M;
1947 		vsid_modulus = ((1UL << (va_bits - SID_SHIFT)) - 1);
1948 	} else {
1949 		modinv = VSID_MULINV_1T;
1950 		vsid_modulus = ((1UL << (va_bits - SID_SHIFT_1T)) - 1);
1951 	}
1952 
1953 	/*
1954 	 * vsid outside our range.
1955 	 */
1956 	if (vsid >= vsid_modulus)
1957 		return 0;
1958 
1959 	/*
1960 	 * If modinv is the modular multiplicate inverse of (x % vsid_modulus)
1961 	 * and vsid = (protovsid * x) % vsid_modulus, then we say:
1962 	 *   protovsid = (vsid * modinv) % vsid_modulus
1963 	 */
1964 
1965 	/* Check if (vsid * modinv) overflow (63 bits) */
1966 	max_mod_inv = 0x7fffffffffffffffull / vsid;
1967 	if (modinv < max_mod_inv)
1968 		return (vsid * modinv) % vsid_modulus;
1969 
1970 	tmp_modinv = modinv/max_mod_inv;
1971 	modinv %= max_mod_inv;
1972 
1973 	protovsid = (((vsid * max_mod_inv) % vsid_modulus) * tmp_modinv) % vsid_modulus;
1974 	protovsid = (protovsid + vsid * modinv) % vsid_modulus;
1975 
1976 	return protovsid;
1977 }
1978 
1979 static int __init reserve_vrma_context_id(void)
1980 {
1981 	unsigned long protovsid;
1982 
1983 	/*
1984 	 * Reserve context ids which map to reserved virtual addresses. For now
1985 	 * we only reserve the context id which maps to the VRMA VSID. We ignore
1986 	 * the addresses in "ibm,adjunct-virtual-addresses" because we don't
1987 	 * enable adjunct support via the "ibm,client-architecture-support"
1988 	 * interface.
1989 	 */
1990 	protovsid = vsid_unscramble(VRMA_VSID, MMU_SEGSIZE_1T);
1991 	hash__reserve_context_id(protovsid >> ESID_BITS_1T);
1992 	return 0;
1993 }
1994 machine_device_initcall(pseries, reserve_vrma_context_id);
1995 
1996 #ifdef CONFIG_DEBUG_FS
1997 /* debugfs file interface for vpa data */
1998 static ssize_t vpa_file_read(struct file *filp, char __user *buf, size_t len,
1999 			      loff_t *pos)
2000 {
2001 	int cpu = (long)filp->private_data;
2002 	struct lppaca *lppaca = &lppaca_of(cpu);
2003 
2004 	return simple_read_from_buffer(buf, len, pos, lppaca,
2005 				sizeof(struct lppaca));
2006 }
2007 
2008 static const struct file_operations vpa_fops = {
2009 	.open		= simple_open,
2010 	.read		= vpa_file_read,
2011 	.llseek		= default_llseek,
2012 };
2013 
2014 static int __init vpa_debugfs_init(void)
2015 {
2016 	char name[16];
2017 	long i;
2018 	struct dentry *vpa_dir;
2019 
2020 	if (!firmware_has_feature(FW_FEATURE_SPLPAR))
2021 		return 0;
2022 
2023 	vpa_dir = debugfs_create_dir("vpa", arch_debugfs_dir);
2024 
2025 	/* set up the per-cpu vpa file*/
2026 	for_each_possible_cpu(i) {
2027 		sprintf(name, "cpu-%ld", i);
2028 		debugfs_create_file(name, 0400, vpa_dir, (void *)i, &vpa_fops);
2029 	}
2030 
2031 	return 0;
2032 }
2033 machine_arch_initcall(pseries, vpa_debugfs_init);
2034 #endif /* CONFIG_DEBUG_FS */
2035