1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * TLB flush routines for radix kernels.
4  *
5  * Copyright 2015-2016, Aneesh Kumar K.V, IBM Corporation.
6  */
7 
8 #include <linux/mm.h>
9 #include <linux/hugetlb.h>
10 #include <linux/memblock.h>
11 #include <linux/mmu_context.h>
12 #include <linux/sched/mm.h>
13 #include <linux/debugfs.h>
14 
15 #include <asm/ppc-opcode.h>
16 #include <asm/tlb.h>
17 #include <asm/tlbflush.h>
18 #include <asm/trace.h>
19 #include <asm/cputhreads.h>
20 #include <asm/plpar_wrappers.h>
21 
22 #include "internal.h"
23 
24 /*
25  * tlbiel instruction for radix, set invalidation
26  * i.e., r=1 and is=01 or is=10 or is=11
27  */
28 static __always_inline void tlbiel_radix_set_isa300(unsigned int set, unsigned int is,
29 					unsigned int pid,
30 					unsigned int ric, unsigned int prs)
31 {
32 	unsigned long rb;
33 	unsigned long rs;
34 
35 	rb = (set << PPC_BITLSHIFT(51)) | (is << PPC_BITLSHIFT(53));
36 	rs = ((unsigned long)pid << PPC_BITLSHIFT(31));
37 
38 	asm volatile(PPC_TLBIEL(%0, %1, %2, %3, 1)
39 		     : : "r"(rb), "r"(rs), "i"(ric), "i"(prs)
40 		     : "memory");
41 }
42 
43 static void tlbiel_all_isa300(unsigned int num_sets, unsigned int is)
44 {
45 	unsigned int set;
46 
47 	asm volatile("ptesync": : :"memory");
48 
49 	/*
50 	 * Flush the first set of the TLB, and the entire Page Walk Cache
51 	 * and partition table entries. Then flush the remaining sets of the
52 	 * TLB.
53 	 */
54 
55 	if (early_cpu_has_feature(CPU_FTR_HVMODE)) {
56 		/* MSR[HV] should flush partition scope translations first. */
57 		tlbiel_radix_set_isa300(0, is, 0, RIC_FLUSH_ALL, 0);
58 
59 		if (!early_cpu_has_feature(CPU_FTR_ARCH_31)) {
60 			for (set = 1; set < num_sets; set++)
61 				tlbiel_radix_set_isa300(set, is, 0,
62 							RIC_FLUSH_TLB, 0);
63 		}
64 	}
65 
66 	/* Flush process scoped entries. */
67 	tlbiel_radix_set_isa300(0, is, 0, RIC_FLUSH_ALL, 1);
68 
69 	if (!early_cpu_has_feature(CPU_FTR_ARCH_31)) {
70 		for (set = 1; set < num_sets; set++)
71 			tlbiel_radix_set_isa300(set, is, 0, RIC_FLUSH_TLB, 1);
72 	}
73 
74 	ppc_after_tlbiel_barrier();
75 }
76 
77 void radix__tlbiel_all(unsigned int action)
78 {
79 	unsigned int is;
80 
81 	switch (action) {
82 	case TLB_INVAL_SCOPE_GLOBAL:
83 		is = 3;
84 		break;
85 	case TLB_INVAL_SCOPE_LPID:
86 		is = 2;
87 		break;
88 	default:
89 		BUG();
90 	}
91 
92 	if (early_cpu_has_feature(CPU_FTR_ARCH_300))
93 		tlbiel_all_isa300(POWER9_TLB_SETS_RADIX, is);
94 	else
95 		WARN(1, "%s called on pre-POWER9 CPU\n", __func__);
96 
97 	asm volatile(PPC_ISA_3_0_INVALIDATE_ERAT "; isync" : : :"memory");
98 }
99 
100 static __always_inline void __tlbiel_pid(unsigned long pid, int set,
101 				unsigned long ric)
102 {
103 	unsigned long rb,rs,prs,r;
104 
105 	rb = PPC_BIT(53); /* IS = 1 */
106 	rb |= set << PPC_BITLSHIFT(51);
107 	rs = ((unsigned long)pid) << PPC_BITLSHIFT(31);
108 	prs = 1; /* process scoped */
109 	r = 1;   /* radix format */
110 
111 	asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
112 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
113 	trace_tlbie(0, 1, rb, rs, ric, prs, r);
114 }
115 
116 static __always_inline void __tlbie_pid(unsigned long pid, unsigned long ric)
117 {
118 	unsigned long rb,rs,prs,r;
119 
120 	rb = PPC_BIT(53); /* IS = 1 */
121 	rs = pid << PPC_BITLSHIFT(31);
122 	prs = 1; /* process scoped */
123 	r = 1;   /* radix format */
124 
125 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
126 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
127 	trace_tlbie(0, 0, rb, rs, ric, prs, r);
128 }
129 
130 static __always_inline void __tlbie_pid_lpid(unsigned long pid,
131 					     unsigned long lpid,
132 					     unsigned long ric)
133 {
134 	unsigned long rb, rs, prs, r;
135 
136 	rb = PPC_BIT(53); /* IS = 1 */
137 	rs = (pid << PPC_BITLSHIFT(31)) | (lpid & ~(PPC_BITMASK(0, 31)));
138 	prs = 1; /* process scoped */
139 	r = 1;   /* radix format */
140 
141 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
142 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
143 	trace_tlbie(0, 0, rb, rs, ric, prs, r);
144 }
145 static __always_inline void __tlbie_lpid(unsigned long lpid, unsigned long ric)
146 {
147 	unsigned long rb,rs,prs,r;
148 
149 	rb = PPC_BIT(52); /* IS = 2 */
150 	rs = lpid;
151 	prs = 0; /* partition scoped */
152 	r = 1;   /* radix format */
153 
154 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
155 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
156 	trace_tlbie(lpid, 0, rb, rs, ric, prs, r);
157 }
158 
159 static __always_inline void __tlbie_lpid_guest(unsigned long lpid, unsigned long ric)
160 {
161 	unsigned long rb,rs,prs,r;
162 
163 	rb = PPC_BIT(52); /* IS = 2 */
164 	rs = lpid;
165 	prs = 1; /* process scoped */
166 	r = 1;   /* radix format */
167 
168 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
169 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
170 	trace_tlbie(lpid, 0, rb, rs, ric, prs, r);
171 }
172 
173 static __always_inline void __tlbiel_va(unsigned long va, unsigned long pid,
174 					unsigned long ap, unsigned long ric)
175 {
176 	unsigned long rb,rs,prs,r;
177 
178 	rb = va & ~(PPC_BITMASK(52, 63));
179 	rb |= ap << PPC_BITLSHIFT(58);
180 	rs = pid << PPC_BITLSHIFT(31);
181 	prs = 1; /* process scoped */
182 	r = 1;   /* radix format */
183 
184 	asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
185 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
186 	trace_tlbie(0, 1, rb, rs, ric, prs, r);
187 }
188 
189 static __always_inline void __tlbie_va(unsigned long va, unsigned long pid,
190 				       unsigned long ap, unsigned long ric)
191 {
192 	unsigned long rb,rs,prs,r;
193 
194 	rb = va & ~(PPC_BITMASK(52, 63));
195 	rb |= ap << PPC_BITLSHIFT(58);
196 	rs = pid << PPC_BITLSHIFT(31);
197 	prs = 1; /* process scoped */
198 	r = 1;   /* radix format */
199 
200 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
201 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
202 	trace_tlbie(0, 0, rb, rs, ric, prs, r);
203 }
204 
205 static __always_inline void __tlbie_va_lpid(unsigned long va, unsigned long pid,
206 					    unsigned long lpid,
207 					    unsigned long ap, unsigned long ric)
208 {
209 	unsigned long rb, rs, prs, r;
210 
211 	rb = va & ~(PPC_BITMASK(52, 63));
212 	rb |= ap << PPC_BITLSHIFT(58);
213 	rs = (pid << PPC_BITLSHIFT(31)) | (lpid & ~(PPC_BITMASK(0, 31)));
214 	prs = 1; /* process scoped */
215 	r = 1;   /* radix format */
216 
217 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
218 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
219 	trace_tlbie(0, 0, rb, rs, ric, prs, r);
220 }
221 
222 static __always_inline void __tlbie_lpid_va(unsigned long va, unsigned long lpid,
223 					    unsigned long ap, unsigned long ric)
224 {
225 	unsigned long rb,rs,prs,r;
226 
227 	rb = va & ~(PPC_BITMASK(52, 63));
228 	rb |= ap << PPC_BITLSHIFT(58);
229 	rs = lpid;
230 	prs = 0; /* partition scoped */
231 	r = 1;   /* radix format */
232 
233 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
234 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
235 	trace_tlbie(lpid, 0, rb, rs, ric, prs, r);
236 }
237 
238 
239 static inline void fixup_tlbie_va(unsigned long va, unsigned long pid,
240 				  unsigned long ap)
241 {
242 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
243 		asm volatile("ptesync": : :"memory");
244 		__tlbie_va(va, 0, ap, RIC_FLUSH_TLB);
245 	}
246 
247 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
248 		asm volatile("ptesync": : :"memory");
249 		__tlbie_va(va, pid, ap, RIC_FLUSH_TLB);
250 	}
251 }
252 
253 static inline void fixup_tlbie_va_range(unsigned long va, unsigned long pid,
254 					unsigned long ap)
255 {
256 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
257 		asm volatile("ptesync": : :"memory");
258 		__tlbie_pid(0, RIC_FLUSH_TLB);
259 	}
260 
261 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
262 		asm volatile("ptesync": : :"memory");
263 		__tlbie_va(va, pid, ap, RIC_FLUSH_TLB);
264 	}
265 }
266 
267 static inline void fixup_tlbie_va_range_lpid(unsigned long va,
268 					     unsigned long pid,
269 					     unsigned long lpid,
270 					     unsigned long ap)
271 {
272 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
273 		asm volatile("ptesync" : : : "memory");
274 		__tlbie_pid_lpid(0, lpid, RIC_FLUSH_TLB);
275 	}
276 
277 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
278 		asm volatile("ptesync" : : : "memory");
279 		__tlbie_va_lpid(va, pid, lpid, ap, RIC_FLUSH_TLB);
280 	}
281 }
282 
283 static inline void fixup_tlbie_pid(unsigned long pid)
284 {
285 	/*
286 	 * We can use any address for the invalidation, pick one which is
287 	 * probably unused as an optimisation.
288 	 */
289 	unsigned long va = ((1UL << 52) - 1);
290 
291 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
292 		asm volatile("ptesync": : :"memory");
293 		__tlbie_pid(0, RIC_FLUSH_TLB);
294 	}
295 
296 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
297 		asm volatile("ptesync": : :"memory");
298 		__tlbie_va(va, pid, mmu_get_ap(MMU_PAGE_64K), RIC_FLUSH_TLB);
299 	}
300 }
301 
302 static inline void fixup_tlbie_pid_lpid(unsigned long pid, unsigned long lpid)
303 {
304 	/*
305 	 * We can use any address for the invalidation, pick one which is
306 	 * probably unused as an optimisation.
307 	 */
308 	unsigned long va = ((1UL << 52) - 1);
309 
310 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
311 		asm volatile("ptesync" : : : "memory");
312 		__tlbie_pid_lpid(0, lpid, RIC_FLUSH_TLB);
313 	}
314 
315 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
316 		asm volatile("ptesync" : : : "memory");
317 		__tlbie_va_lpid(va, pid, lpid, mmu_get_ap(MMU_PAGE_64K),
318 				RIC_FLUSH_TLB);
319 	}
320 }
321 
322 static inline void fixup_tlbie_lpid_va(unsigned long va, unsigned long lpid,
323 				       unsigned long ap)
324 {
325 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
326 		asm volatile("ptesync": : :"memory");
327 		__tlbie_lpid_va(va, 0, ap, RIC_FLUSH_TLB);
328 	}
329 
330 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
331 		asm volatile("ptesync": : :"memory");
332 		__tlbie_lpid_va(va, lpid, ap, RIC_FLUSH_TLB);
333 	}
334 }
335 
336 static inline void fixup_tlbie_lpid(unsigned long lpid)
337 {
338 	/*
339 	 * We can use any address for the invalidation, pick one which is
340 	 * probably unused as an optimisation.
341 	 */
342 	unsigned long va = ((1UL << 52) - 1);
343 
344 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
345 		asm volatile("ptesync": : :"memory");
346 		__tlbie_lpid(0, RIC_FLUSH_TLB);
347 	}
348 
349 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
350 		asm volatile("ptesync": : :"memory");
351 		__tlbie_lpid_va(va, lpid, mmu_get_ap(MMU_PAGE_64K), RIC_FLUSH_TLB);
352 	}
353 }
354 
355 /*
356  * We use 128 set in radix mode and 256 set in hpt mode.
357  */
358 static inline void _tlbiel_pid(unsigned long pid, unsigned long ric)
359 {
360 	int set;
361 
362 	asm volatile("ptesync": : :"memory");
363 
364 	switch (ric) {
365 	case RIC_FLUSH_PWC:
366 
367 		/* For PWC, only one flush is needed */
368 		__tlbiel_pid(pid, 0, RIC_FLUSH_PWC);
369 		ppc_after_tlbiel_barrier();
370 		return;
371 	case RIC_FLUSH_TLB:
372 		__tlbiel_pid(pid, 0, RIC_FLUSH_TLB);
373 		break;
374 	case RIC_FLUSH_ALL:
375 	default:
376 		/*
377 		 * Flush the first set of the TLB, and if
378 		 * we're doing a RIC_FLUSH_ALL, also flush
379 		 * the entire Page Walk Cache.
380 		 */
381 		__tlbiel_pid(pid, 0, RIC_FLUSH_ALL);
382 	}
383 
384 	if (!cpu_has_feature(CPU_FTR_ARCH_31)) {
385 		/* For the remaining sets, just flush the TLB */
386 		for (set = 1; set < POWER9_TLB_SETS_RADIX ; set++)
387 			__tlbiel_pid(pid, set, RIC_FLUSH_TLB);
388 	}
389 
390 	ppc_after_tlbiel_barrier();
391 	asm volatile(PPC_RADIX_INVALIDATE_ERAT_USER "; isync" : : :"memory");
392 }
393 
394 static inline void _tlbie_pid(unsigned long pid, unsigned long ric)
395 {
396 	asm volatile("ptesync": : :"memory");
397 
398 	/*
399 	 * Workaround the fact that the "ric" argument to __tlbie_pid
400 	 * must be a compile-time contraint to match the "i" constraint
401 	 * in the asm statement.
402 	 */
403 	switch (ric) {
404 	case RIC_FLUSH_TLB:
405 		__tlbie_pid(pid, RIC_FLUSH_TLB);
406 		fixup_tlbie_pid(pid);
407 		break;
408 	case RIC_FLUSH_PWC:
409 		__tlbie_pid(pid, RIC_FLUSH_PWC);
410 		break;
411 	case RIC_FLUSH_ALL:
412 	default:
413 		__tlbie_pid(pid, RIC_FLUSH_ALL);
414 		fixup_tlbie_pid(pid);
415 	}
416 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
417 }
418 
419 static inline void _tlbie_pid_lpid(unsigned long pid, unsigned long lpid,
420 				   unsigned long ric)
421 {
422 	asm volatile("ptesync" : : : "memory");
423 
424 	/*
425 	 * Workaround the fact that the "ric" argument to __tlbie_pid
426 	 * must be a compile-time contraint to match the "i" constraint
427 	 * in the asm statement.
428 	 */
429 	switch (ric) {
430 	case RIC_FLUSH_TLB:
431 		__tlbie_pid_lpid(pid, lpid, RIC_FLUSH_TLB);
432 		fixup_tlbie_pid_lpid(pid, lpid);
433 		break;
434 	case RIC_FLUSH_PWC:
435 		__tlbie_pid_lpid(pid, lpid, RIC_FLUSH_PWC);
436 		break;
437 	case RIC_FLUSH_ALL:
438 	default:
439 		__tlbie_pid_lpid(pid, lpid, RIC_FLUSH_ALL);
440 		fixup_tlbie_pid_lpid(pid, lpid);
441 	}
442 	asm volatile("eieio; tlbsync; ptesync" : : : "memory");
443 }
444 struct tlbiel_pid {
445 	unsigned long pid;
446 	unsigned long ric;
447 };
448 
449 static void do_tlbiel_pid(void *info)
450 {
451 	struct tlbiel_pid *t = info;
452 
453 	if (t->ric == RIC_FLUSH_TLB)
454 		_tlbiel_pid(t->pid, RIC_FLUSH_TLB);
455 	else if (t->ric == RIC_FLUSH_PWC)
456 		_tlbiel_pid(t->pid, RIC_FLUSH_PWC);
457 	else
458 		_tlbiel_pid(t->pid, RIC_FLUSH_ALL);
459 }
460 
461 static inline void _tlbiel_pid_multicast(struct mm_struct *mm,
462 				unsigned long pid, unsigned long ric)
463 {
464 	struct cpumask *cpus = mm_cpumask(mm);
465 	struct tlbiel_pid t = { .pid = pid, .ric = ric };
466 
467 	on_each_cpu_mask(cpus, do_tlbiel_pid, &t, 1);
468 	/*
469 	 * Always want the CPU translations to be invalidated with tlbiel in
470 	 * these paths, so while coprocessors must use tlbie, we can not
471 	 * optimise away the tlbiel component.
472 	 */
473 	if (atomic_read(&mm->context.copros) > 0)
474 		_tlbie_pid(pid, RIC_FLUSH_ALL);
475 }
476 
477 static inline void _tlbie_lpid(unsigned long lpid, unsigned long ric)
478 {
479 	asm volatile("ptesync": : :"memory");
480 
481 	/*
482 	 * Workaround the fact that the "ric" argument to __tlbie_pid
483 	 * must be a compile-time contraint to match the "i" constraint
484 	 * in the asm statement.
485 	 */
486 	switch (ric) {
487 	case RIC_FLUSH_TLB:
488 		__tlbie_lpid(lpid, RIC_FLUSH_TLB);
489 		fixup_tlbie_lpid(lpid);
490 		break;
491 	case RIC_FLUSH_PWC:
492 		__tlbie_lpid(lpid, RIC_FLUSH_PWC);
493 		break;
494 	case RIC_FLUSH_ALL:
495 	default:
496 		__tlbie_lpid(lpid, RIC_FLUSH_ALL);
497 		fixup_tlbie_lpid(lpid);
498 	}
499 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
500 }
501 
502 static __always_inline void _tlbie_lpid_guest(unsigned long lpid, unsigned long ric)
503 {
504 	/*
505 	 * Workaround the fact that the "ric" argument to __tlbie_pid
506 	 * must be a compile-time contraint to match the "i" constraint
507 	 * in the asm statement.
508 	 */
509 	switch (ric) {
510 	case RIC_FLUSH_TLB:
511 		__tlbie_lpid_guest(lpid, RIC_FLUSH_TLB);
512 		break;
513 	case RIC_FLUSH_PWC:
514 		__tlbie_lpid_guest(lpid, RIC_FLUSH_PWC);
515 		break;
516 	case RIC_FLUSH_ALL:
517 	default:
518 		__tlbie_lpid_guest(lpid, RIC_FLUSH_ALL);
519 	}
520 	fixup_tlbie_lpid(lpid);
521 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
522 }
523 
524 static inline void __tlbiel_va_range(unsigned long start, unsigned long end,
525 				    unsigned long pid, unsigned long page_size,
526 				    unsigned long psize)
527 {
528 	unsigned long addr;
529 	unsigned long ap = mmu_get_ap(psize);
530 
531 	for (addr = start; addr < end; addr += page_size)
532 		__tlbiel_va(addr, pid, ap, RIC_FLUSH_TLB);
533 }
534 
535 static __always_inline void _tlbiel_va(unsigned long va, unsigned long pid,
536 				       unsigned long psize, unsigned long ric)
537 {
538 	unsigned long ap = mmu_get_ap(psize);
539 
540 	asm volatile("ptesync": : :"memory");
541 	__tlbiel_va(va, pid, ap, ric);
542 	ppc_after_tlbiel_barrier();
543 }
544 
545 static inline void _tlbiel_va_range(unsigned long start, unsigned long end,
546 				    unsigned long pid, unsigned long page_size,
547 				    unsigned long psize, bool also_pwc)
548 {
549 	asm volatile("ptesync": : :"memory");
550 	if (also_pwc)
551 		__tlbiel_pid(pid, 0, RIC_FLUSH_PWC);
552 	__tlbiel_va_range(start, end, pid, page_size, psize);
553 	ppc_after_tlbiel_barrier();
554 }
555 
556 static inline void __tlbie_va_range(unsigned long start, unsigned long end,
557 				    unsigned long pid, unsigned long page_size,
558 				    unsigned long psize)
559 {
560 	unsigned long addr;
561 	unsigned long ap = mmu_get_ap(psize);
562 
563 	for (addr = start; addr < end; addr += page_size)
564 		__tlbie_va(addr, pid, ap, RIC_FLUSH_TLB);
565 
566 	fixup_tlbie_va_range(addr - page_size, pid, ap);
567 }
568 
569 static inline void __tlbie_va_range_lpid(unsigned long start, unsigned long end,
570 					 unsigned long pid, unsigned long lpid,
571 					 unsigned long page_size,
572 					 unsigned long psize)
573 {
574 	unsigned long addr;
575 	unsigned long ap = mmu_get_ap(psize);
576 
577 	for (addr = start; addr < end; addr += page_size)
578 		__tlbie_va_lpid(addr, pid, lpid, ap, RIC_FLUSH_TLB);
579 
580 	fixup_tlbie_va_range_lpid(addr - page_size, pid, lpid, ap);
581 }
582 
583 static __always_inline void _tlbie_va(unsigned long va, unsigned long pid,
584 				      unsigned long psize, unsigned long ric)
585 {
586 	unsigned long ap = mmu_get_ap(psize);
587 
588 	asm volatile("ptesync": : :"memory");
589 	__tlbie_va(va, pid, ap, ric);
590 	fixup_tlbie_va(va, pid, ap);
591 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
592 }
593 
594 struct tlbiel_va {
595 	unsigned long pid;
596 	unsigned long va;
597 	unsigned long psize;
598 	unsigned long ric;
599 };
600 
601 static void do_tlbiel_va(void *info)
602 {
603 	struct tlbiel_va *t = info;
604 
605 	if (t->ric == RIC_FLUSH_TLB)
606 		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_TLB);
607 	else if (t->ric == RIC_FLUSH_PWC)
608 		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_PWC);
609 	else
610 		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_ALL);
611 }
612 
613 static inline void _tlbiel_va_multicast(struct mm_struct *mm,
614 				unsigned long va, unsigned long pid,
615 				unsigned long psize, unsigned long ric)
616 {
617 	struct cpumask *cpus = mm_cpumask(mm);
618 	struct tlbiel_va t = { .va = va, .pid = pid, .psize = psize, .ric = ric };
619 	on_each_cpu_mask(cpus, do_tlbiel_va, &t, 1);
620 	if (atomic_read(&mm->context.copros) > 0)
621 		_tlbie_va(va, pid, psize, RIC_FLUSH_TLB);
622 }
623 
624 struct tlbiel_va_range {
625 	unsigned long pid;
626 	unsigned long start;
627 	unsigned long end;
628 	unsigned long page_size;
629 	unsigned long psize;
630 	bool also_pwc;
631 };
632 
633 static void do_tlbiel_va_range(void *info)
634 {
635 	struct tlbiel_va_range *t = info;
636 
637 	_tlbiel_va_range(t->start, t->end, t->pid, t->page_size,
638 				    t->psize, t->also_pwc);
639 }
640 
641 static __always_inline void _tlbie_lpid_va(unsigned long va, unsigned long lpid,
642 			      unsigned long psize, unsigned long ric)
643 {
644 	unsigned long ap = mmu_get_ap(psize);
645 
646 	asm volatile("ptesync": : :"memory");
647 	__tlbie_lpid_va(va, lpid, ap, ric);
648 	fixup_tlbie_lpid_va(va, lpid, ap);
649 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
650 }
651 
652 static inline void _tlbie_va_range(unsigned long start, unsigned long end,
653 				    unsigned long pid, unsigned long page_size,
654 				    unsigned long psize, bool also_pwc)
655 {
656 	asm volatile("ptesync": : :"memory");
657 	if (also_pwc)
658 		__tlbie_pid(pid, RIC_FLUSH_PWC);
659 	__tlbie_va_range(start, end, pid, page_size, psize);
660 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
661 }
662 
663 static inline void _tlbie_va_range_lpid(unsigned long start, unsigned long end,
664 					unsigned long pid, unsigned long lpid,
665 					unsigned long page_size,
666 					unsigned long psize, bool also_pwc)
667 {
668 	asm volatile("ptesync" : : : "memory");
669 	if (also_pwc)
670 		__tlbie_pid_lpid(pid, lpid, RIC_FLUSH_PWC);
671 	__tlbie_va_range_lpid(start, end, pid, lpid, page_size, psize);
672 	asm volatile("eieio; tlbsync; ptesync" : : : "memory");
673 }
674 
675 static inline void _tlbiel_va_range_multicast(struct mm_struct *mm,
676 				unsigned long start, unsigned long end,
677 				unsigned long pid, unsigned long page_size,
678 				unsigned long psize, bool also_pwc)
679 {
680 	struct cpumask *cpus = mm_cpumask(mm);
681 	struct tlbiel_va_range t = { .start = start, .end = end,
682 				.pid = pid, .page_size = page_size,
683 				.psize = psize, .also_pwc = also_pwc };
684 
685 	on_each_cpu_mask(cpus, do_tlbiel_va_range, &t, 1);
686 	if (atomic_read(&mm->context.copros) > 0)
687 		_tlbie_va_range(start, end, pid, page_size, psize, also_pwc);
688 }
689 
690 /*
691  * Base TLB flushing operations:
692  *
693  *  - flush_tlb_mm(mm) flushes the specified mm context TLB's
694  *  - flush_tlb_page(vma, vmaddr) flushes one page
695  *  - flush_tlb_range(vma, start, end) flushes a range of pages
696  *  - flush_tlb_kernel_range(start, end) flushes kernel pages
697  *
698  *  - local_* variants of page and mm only apply to the current
699  *    processor
700  */
701 void radix__local_flush_tlb_mm(struct mm_struct *mm)
702 {
703 	unsigned long pid;
704 
705 	preempt_disable();
706 	pid = mm->context.id;
707 	if (pid != MMU_NO_CONTEXT)
708 		_tlbiel_pid(pid, RIC_FLUSH_TLB);
709 	preempt_enable();
710 }
711 EXPORT_SYMBOL(radix__local_flush_tlb_mm);
712 
713 #ifndef CONFIG_SMP
714 void radix__local_flush_all_mm(struct mm_struct *mm)
715 {
716 	unsigned long pid;
717 
718 	preempt_disable();
719 	pid = mm->context.id;
720 	if (pid != MMU_NO_CONTEXT)
721 		_tlbiel_pid(pid, RIC_FLUSH_ALL);
722 	preempt_enable();
723 }
724 EXPORT_SYMBOL(radix__local_flush_all_mm);
725 
726 static void __flush_all_mm(struct mm_struct *mm, bool fullmm)
727 {
728 	radix__local_flush_all_mm(mm);
729 }
730 #endif /* CONFIG_SMP */
731 
732 void radix__local_flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
733 				       int psize)
734 {
735 	unsigned long pid;
736 
737 	preempt_disable();
738 	pid = mm->context.id;
739 	if (pid != MMU_NO_CONTEXT)
740 		_tlbiel_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
741 	preempt_enable();
742 }
743 
744 void radix__local_flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
745 {
746 #ifdef CONFIG_HUGETLB_PAGE
747 	/* need the return fix for nohash.c */
748 	if (is_vm_hugetlb_page(vma))
749 		return radix__local_flush_hugetlb_page(vma, vmaddr);
750 #endif
751 	radix__local_flush_tlb_page_psize(vma->vm_mm, vmaddr, mmu_virtual_psize);
752 }
753 EXPORT_SYMBOL(radix__local_flush_tlb_page);
754 
755 static bool mm_needs_flush_escalation(struct mm_struct *mm)
756 {
757 	/*
758 	 * P9 nest MMU has issues with the page walk cache
759 	 * caching PTEs and not flushing them properly when
760 	 * RIC = 0 for a PID/LPID invalidate
761 	 */
762 	if (atomic_read(&mm->context.copros) > 0)
763 		return true;
764 	return false;
765 }
766 
767 /*
768  * If always_flush is true, then flush even if this CPU can't be removed
769  * from mm_cpumask.
770  */
771 void exit_lazy_flush_tlb(struct mm_struct *mm, bool always_flush)
772 {
773 	unsigned long pid = mm->context.id;
774 	int cpu = smp_processor_id();
775 
776 	/*
777 	 * A kthread could have done a mmget_not_zero() after the flushing CPU
778 	 * checked mm_cpumask, and be in the process of kthread_use_mm when
779 	 * interrupted here. In that case, current->mm will be set to mm,
780 	 * because kthread_use_mm() setting ->mm and switching to the mm is
781 	 * done with interrupts off.
782 	 */
783 	if (current->mm == mm)
784 		goto out;
785 
786 	if (current->active_mm == mm) {
787 		WARN_ON_ONCE(current->mm != NULL);
788 		/* Is a kernel thread and is using mm as the lazy tlb */
789 		mmgrab(&init_mm);
790 		current->active_mm = &init_mm;
791 		switch_mm_irqs_off(mm, &init_mm, current);
792 		mmdrop(mm);
793 	}
794 
795 	/*
796 	 * This IPI may be initiated from any source including those not
797 	 * running the mm, so there may be a racing IPI that comes after
798 	 * this one which finds the cpumask already clear. Check and avoid
799 	 * underflowing the active_cpus count in that case. The race should
800 	 * not otherwise be a problem, but the TLB must be flushed because
801 	 * that's what the caller expects.
802 	 */
803 	if (cpumask_test_cpu(cpu, mm_cpumask(mm))) {
804 		atomic_dec(&mm->context.active_cpus);
805 		cpumask_clear_cpu(cpu, mm_cpumask(mm));
806 		always_flush = true;
807 	}
808 
809 out:
810 	if (always_flush)
811 		_tlbiel_pid(pid, RIC_FLUSH_ALL);
812 }
813 
814 #ifdef CONFIG_SMP
815 static void do_exit_flush_lazy_tlb(void *arg)
816 {
817 	struct mm_struct *mm = arg;
818 	exit_lazy_flush_tlb(mm, true);
819 }
820 
821 static void exit_flush_lazy_tlbs(struct mm_struct *mm)
822 {
823 	/*
824 	 * Would be nice if this was async so it could be run in
825 	 * parallel with our local flush, but generic code does not
826 	 * give a good API for it. Could extend the generic code or
827 	 * make a special powerpc IPI for flushing TLBs.
828 	 * For now it's not too performance critical.
829 	 */
830 	smp_call_function_many(mm_cpumask(mm), do_exit_flush_lazy_tlb,
831 				(void *)mm, 1);
832 }
833 
834 #else /* CONFIG_SMP */
835 static inline void exit_flush_lazy_tlbs(struct mm_struct *mm) { }
836 #endif /* CONFIG_SMP */
837 
838 static DEFINE_PER_CPU(unsigned int, mm_cpumask_trim_clock);
839 
840 /*
841  * Interval between flushes at which we send out IPIs to check whether the
842  * mm_cpumask can be trimmed for the case where it's not a single-threaded
843  * process flushing its own mm. The intent is to reduce the cost of later
844  * flushes. Don't want this to be so low that it adds noticable cost to TLB
845  * flushing, or so high that it doesn't help reduce global TLBIEs.
846  */
847 static unsigned long tlb_mm_cpumask_trim_timer = 1073;
848 
849 static bool tick_and_test_trim_clock(void)
850 {
851 	if (__this_cpu_inc_return(mm_cpumask_trim_clock) ==
852 			tlb_mm_cpumask_trim_timer) {
853 		__this_cpu_write(mm_cpumask_trim_clock, 0);
854 		return true;
855 	}
856 	return false;
857 }
858 
859 enum tlb_flush_type {
860 	FLUSH_TYPE_NONE,
861 	FLUSH_TYPE_LOCAL,
862 	FLUSH_TYPE_GLOBAL,
863 };
864 
865 static enum tlb_flush_type flush_type_needed(struct mm_struct *mm, bool fullmm)
866 {
867 	int active_cpus = atomic_read(&mm->context.active_cpus);
868 	int cpu = smp_processor_id();
869 
870 	if (active_cpus == 0)
871 		return FLUSH_TYPE_NONE;
872 	if (active_cpus == 1 && cpumask_test_cpu(cpu, mm_cpumask(mm))) {
873 		if (current->mm != mm) {
874 			/*
875 			 * Asynchronous flush sources may trim down to nothing
876 			 * if the process is not running, so occasionally try
877 			 * to trim.
878 			 */
879 			if (tick_and_test_trim_clock()) {
880 				exit_lazy_flush_tlb(mm, true);
881 				return FLUSH_TYPE_NONE;
882 			}
883 		}
884 		return FLUSH_TYPE_LOCAL;
885 	}
886 
887 	/* Coprocessors require TLBIE to invalidate nMMU. */
888 	if (atomic_read(&mm->context.copros) > 0)
889 		return FLUSH_TYPE_GLOBAL;
890 
891 	/*
892 	 * In the fullmm case there's no point doing the exit_flush_lazy_tlbs
893 	 * because the mm is being taken down anyway, and a TLBIE tends to
894 	 * be faster than an IPI+TLBIEL.
895 	 */
896 	if (fullmm)
897 		return FLUSH_TYPE_GLOBAL;
898 
899 	/*
900 	 * If we are running the only thread of a single-threaded process,
901 	 * then we should almost always be able to trim off the rest of the
902 	 * CPU mask (except in the case of use_mm() races), so always try
903 	 * trimming the mask.
904 	 */
905 	if (atomic_read(&mm->mm_users) <= 1 && current->mm == mm) {
906 		exit_flush_lazy_tlbs(mm);
907 		/*
908 		 * use_mm() race could prevent IPIs from being able to clear
909 		 * the cpumask here, however those users are established
910 		 * after our first check (and so after the PTEs are removed),
911 		 * and the TLB still gets flushed by the IPI, so this CPU
912 		 * will only require a local flush.
913 		 */
914 		return FLUSH_TYPE_LOCAL;
915 	}
916 
917 	/*
918 	 * Occasionally try to trim down the cpumask. It's possible this can
919 	 * bring the mask to zero, which results in no flush.
920 	 */
921 	if (tick_and_test_trim_clock()) {
922 		exit_flush_lazy_tlbs(mm);
923 		if (current->mm == mm)
924 			return FLUSH_TYPE_LOCAL;
925 		if (cpumask_test_cpu(cpu, mm_cpumask(mm)))
926 			exit_lazy_flush_tlb(mm, true);
927 		return FLUSH_TYPE_NONE;
928 	}
929 
930 	return FLUSH_TYPE_GLOBAL;
931 }
932 
933 #ifdef CONFIG_SMP
934 void radix__flush_tlb_mm(struct mm_struct *mm)
935 {
936 	unsigned long pid;
937 	enum tlb_flush_type type;
938 
939 	pid = mm->context.id;
940 	if (unlikely(pid == MMU_NO_CONTEXT))
941 		return;
942 
943 	preempt_disable();
944 	/*
945 	 * Order loads of mm_cpumask (in flush_type_needed) vs previous
946 	 * stores to clear ptes before the invalidate. See barrier in
947 	 * switch_mm_irqs_off
948 	 */
949 	smp_mb();
950 	type = flush_type_needed(mm, false);
951 	if (type == FLUSH_TYPE_LOCAL) {
952 		_tlbiel_pid(pid, RIC_FLUSH_TLB);
953 	} else if (type == FLUSH_TYPE_GLOBAL) {
954 		if (!mmu_has_feature(MMU_FTR_GTSE)) {
955 			unsigned long tgt = H_RPTI_TARGET_CMMU;
956 
957 			if (atomic_read(&mm->context.copros) > 0)
958 				tgt |= H_RPTI_TARGET_NMMU;
959 			pseries_rpt_invalidate(pid, tgt, H_RPTI_TYPE_TLB,
960 					       H_RPTI_PAGE_ALL, 0, -1UL);
961 		} else if (cputlb_use_tlbie()) {
962 			if (mm_needs_flush_escalation(mm))
963 				_tlbie_pid(pid, RIC_FLUSH_ALL);
964 			else
965 				_tlbie_pid(pid, RIC_FLUSH_TLB);
966 		} else {
967 			_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_TLB);
968 		}
969 	}
970 	preempt_enable();
971 }
972 EXPORT_SYMBOL(radix__flush_tlb_mm);
973 
974 static void __flush_all_mm(struct mm_struct *mm, bool fullmm)
975 {
976 	unsigned long pid;
977 	enum tlb_flush_type type;
978 
979 	pid = mm->context.id;
980 	if (unlikely(pid == MMU_NO_CONTEXT))
981 		return;
982 
983 	preempt_disable();
984 	smp_mb(); /* see radix__flush_tlb_mm */
985 	type = flush_type_needed(mm, fullmm);
986 	if (type == FLUSH_TYPE_LOCAL) {
987 		_tlbiel_pid(pid, RIC_FLUSH_ALL);
988 	} else if (type == FLUSH_TYPE_GLOBAL) {
989 		if (!mmu_has_feature(MMU_FTR_GTSE)) {
990 			unsigned long tgt = H_RPTI_TARGET_CMMU;
991 			unsigned long type = H_RPTI_TYPE_TLB | H_RPTI_TYPE_PWC |
992 					     H_RPTI_TYPE_PRT;
993 
994 			if (atomic_read(&mm->context.copros) > 0)
995 				tgt |= H_RPTI_TARGET_NMMU;
996 			pseries_rpt_invalidate(pid, tgt, type,
997 					       H_RPTI_PAGE_ALL, 0, -1UL);
998 		} else if (cputlb_use_tlbie())
999 			_tlbie_pid(pid, RIC_FLUSH_ALL);
1000 		else
1001 			_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_ALL);
1002 	}
1003 	preempt_enable();
1004 }
1005 
1006 void radix__flush_all_mm(struct mm_struct *mm)
1007 {
1008 	__flush_all_mm(mm, false);
1009 }
1010 EXPORT_SYMBOL(radix__flush_all_mm);
1011 
1012 void radix__flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
1013 				 int psize)
1014 {
1015 	unsigned long pid;
1016 	enum tlb_flush_type type;
1017 
1018 	pid = mm->context.id;
1019 	if (unlikely(pid == MMU_NO_CONTEXT))
1020 		return;
1021 
1022 	preempt_disable();
1023 	smp_mb(); /* see radix__flush_tlb_mm */
1024 	type = flush_type_needed(mm, false);
1025 	if (type == FLUSH_TYPE_LOCAL) {
1026 		_tlbiel_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
1027 	} else if (type == FLUSH_TYPE_GLOBAL) {
1028 		if (!mmu_has_feature(MMU_FTR_GTSE)) {
1029 			unsigned long tgt, pg_sizes, size;
1030 
1031 			tgt = H_RPTI_TARGET_CMMU;
1032 			pg_sizes = psize_to_rpti_pgsize(psize);
1033 			size = 1UL << mmu_psize_to_shift(psize);
1034 
1035 			if (atomic_read(&mm->context.copros) > 0)
1036 				tgt |= H_RPTI_TARGET_NMMU;
1037 			pseries_rpt_invalidate(pid, tgt, H_RPTI_TYPE_TLB,
1038 					       pg_sizes, vmaddr,
1039 					       vmaddr + size);
1040 		} else if (cputlb_use_tlbie())
1041 			_tlbie_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
1042 		else
1043 			_tlbiel_va_multicast(mm, vmaddr, pid, psize, RIC_FLUSH_TLB);
1044 	}
1045 	preempt_enable();
1046 }
1047 
1048 void radix__flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
1049 {
1050 #ifdef CONFIG_HUGETLB_PAGE
1051 	if (is_vm_hugetlb_page(vma))
1052 		return radix__flush_hugetlb_page(vma, vmaddr);
1053 #endif
1054 	radix__flush_tlb_page_psize(vma->vm_mm, vmaddr, mmu_virtual_psize);
1055 }
1056 EXPORT_SYMBOL(radix__flush_tlb_page);
1057 
1058 #endif /* CONFIG_SMP */
1059 
1060 static void do_tlbiel_kernel(void *info)
1061 {
1062 	_tlbiel_pid(0, RIC_FLUSH_ALL);
1063 }
1064 
1065 static inline void _tlbiel_kernel_broadcast(void)
1066 {
1067 	on_each_cpu(do_tlbiel_kernel, NULL, 1);
1068 	if (tlbie_capable) {
1069 		/*
1070 		 * Coherent accelerators don't refcount kernel memory mappings,
1071 		 * so have to always issue a tlbie for them. This is quite a
1072 		 * slow path anyway.
1073 		 */
1074 		_tlbie_pid(0, RIC_FLUSH_ALL);
1075 	}
1076 }
1077 
1078 /*
1079  * If kernel TLBIs ever become local rather than global, then
1080  * drivers/misc/ocxl/link.c:ocxl_link_add_pe will need some work, as it
1081  * assumes kernel TLBIs are global.
1082  */
1083 void radix__flush_tlb_kernel_range(unsigned long start, unsigned long end)
1084 {
1085 	if (!mmu_has_feature(MMU_FTR_GTSE)) {
1086 		unsigned long tgt = H_RPTI_TARGET_CMMU | H_RPTI_TARGET_NMMU;
1087 		unsigned long type = H_RPTI_TYPE_TLB | H_RPTI_TYPE_PWC |
1088 				     H_RPTI_TYPE_PRT;
1089 
1090 		pseries_rpt_invalidate(0, tgt, type, H_RPTI_PAGE_ALL,
1091 				       start, end);
1092 	} else if (cputlb_use_tlbie())
1093 		_tlbie_pid(0, RIC_FLUSH_ALL);
1094 	else
1095 		_tlbiel_kernel_broadcast();
1096 }
1097 EXPORT_SYMBOL(radix__flush_tlb_kernel_range);
1098 
1099 #define TLB_FLUSH_ALL -1UL
1100 
1101 /*
1102  * Number of pages above which we invalidate the entire PID rather than
1103  * flush individual pages, for local and global flushes respectively.
1104  *
1105  * tlbie goes out to the interconnect and individual ops are more costly.
1106  * It also does not iterate over sets like the local tlbiel variant when
1107  * invalidating a full PID, so it has a far lower threshold to change from
1108  * individual page flushes to full-pid flushes.
1109  */
1110 static u32 tlb_single_page_flush_ceiling __read_mostly = 33;
1111 static u32 tlb_local_single_page_flush_ceiling __read_mostly = POWER9_TLB_SETS_RADIX * 2;
1112 
1113 static inline void __radix__flush_tlb_range(struct mm_struct *mm,
1114 					    unsigned long start, unsigned long end)
1115 {
1116 	unsigned long pid;
1117 	unsigned int page_shift = mmu_psize_defs[mmu_virtual_psize].shift;
1118 	unsigned long page_size = 1UL << page_shift;
1119 	unsigned long nr_pages = (end - start) >> page_shift;
1120 	bool fullmm = (end == TLB_FLUSH_ALL);
1121 	bool flush_pid, flush_pwc = false;
1122 	enum tlb_flush_type type;
1123 
1124 	pid = mm->context.id;
1125 	if (unlikely(pid == MMU_NO_CONTEXT))
1126 		return;
1127 
1128 	preempt_disable();
1129 	smp_mb(); /* see radix__flush_tlb_mm */
1130 	type = flush_type_needed(mm, fullmm);
1131 	if (type == FLUSH_TYPE_NONE)
1132 		goto out;
1133 
1134 	if (fullmm)
1135 		flush_pid = true;
1136 	else if (type == FLUSH_TYPE_GLOBAL)
1137 		flush_pid = nr_pages > tlb_single_page_flush_ceiling;
1138 	else
1139 		flush_pid = nr_pages > tlb_local_single_page_flush_ceiling;
1140 	/*
1141 	 * full pid flush already does the PWC flush. if it is not full pid
1142 	 * flush check the range is more than PMD and force a pwc flush
1143 	 * mremap() depends on this behaviour.
1144 	 */
1145 	if (!flush_pid && (end - start) >= PMD_SIZE)
1146 		flush_pwc = true;
1147 
1148 	if (!mmu_has_feature(MMU_FTR_GTSE) && type == FLUSH_TYPE_GLOBAL) {
1149 		unsigned long type = H_RPTI_TYPE_TLB;
1150 		unsigned long tgt = H_RPTI_TARGET_CMMU;
1151 		unsigned long pg_sizes = psize_to_rpti_pgsize(mmu_virtual_psize);
1152 
1153 		if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1154 			pg_sizes |= psize_to_rpti_pgsize(MMU_PAGE_2M);
1155 		if (atomic_read(&mm->context.copros) > 0)
1156 			tgt |= H_RPTI_TARGET_NMMU;
1157 		if (flush_pwc)
1158 			type |= H_RPTI_TYPE_PWC;
1159 		pseries_rpt_invalidate(pid, tgt, type, pg_sizes, start, end);
1160 	} else if (flush_pid) {
1161 		/*
1162 		 * We are now flushing a range larger than PMD size force a RIC_FLUSH_ALL
1163 		 */
1164 		if (type == FLUSH_TYPE_LOCAL) {
1165 			_tlbiel_pid(pid, RIC_FLUSH_ALL);
1166 		} else {
1167 			if (cputlb_use_tlbie()) {
1168 				_tlbie_pid(pid, RIC_FLUSH_ALL);
1169 			} else {
1170 				_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_ALL);
1171 			}
1172 		}
1173 	} else {
1174 		bool hflush = false;
1175 		unsigned long hstart, hend;
1176 
1177 		if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1178 			hstart = (start + PMD_SIZE - 1) & PMD_MASK;
1179 			hend = end & PMD_MASK;
1180 			if (hstart < hend)
1181 				hflush = true;
1182 		}
1183 
1184 		if (type == FLUSH_TYPE_LOCAL) {
1185 			asm volatile("ptesync": : :"memory");
1186 			if (flush_pwc)
1187 				/* For PWC, only one flush is needed */
1188 				__tlbiel_pid(pid, 0, RIC_FLUSH_PWC);
1189 			__tlbiel_va_range(start, end, pid, page_size, mmu_virtual_psize);
1190 			if (hflush)
1191 				__tlbiel_va_range(hstart, hend, pid,
1192 						PMD_SIZE, MMU_PAGE_2M);
1193 			ppc_after_tlbiel_barrier();
1194 		} else if (cputlb_use_tlbie()) {
1195 			asm volatile("ptesync": : :"memory");
1196 			if (flush_pwc)
1197 				__tlbie_pid(pid, RIC_FLUSH_PWC);
1198 			__tlbie_va_range(start, end, pid, page_size, mmu_virtual_psize);
1199 			if (hflush)
1200 				__tlbie_va_range(hstart, hend, pid,
1201 						PMD_SIZE, MMU_PAGE_2M);
1202 			asm volatile("eieio; tlbsync; ptesync": : :"memory");
1203 		} else {
1204 			_tlbiel_va_range_multicast(mm,
1205 					start, end, pid, page_size, mmu_virtual_psize, flush_pwc);
1206 			if (hflush)
1207 				_tlbiel_va_range_multicast(mm,
1208 					hstart, hend, pid, PMD_SIZE, MMU_PAGE_2M, flush_pwc);
1209 		}
1210 	}
1211 out:
1212 	preempt_enable();
1213 }
1214 
1215 void radix__flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
1216 		     unsigned long end)
1217 
1218 {
1219 #ifdef CONFIG_HUGETLB_PAGE
1220 	if (is_vm_hugetlb_page(vma))
1221 		return radix__flush_hugetlb_tlb_range(vma, start, end);
1222 #endif
1223 
1224 	__radix__flush_tlb_range(vma->vm_mm, start, end);
1225 }
1226 EXPORT_SYMBOL(radix__flush_tlb_range);
1227 
1228 static int radix_get_mmu_psize(int page_size)
1229 {
1230 	int psize;
1231 
1232 	if (page_size == (1UL << mmu_psize_defs[mmu_virtual_psize].shift))
1233 		psize = mmu_virtual_psize;
1234 	else if (page_size == (1UL << mmu_psize_defs[MMU_PAGE_2M].shift))
1235 		psize = MMU_PAGE_2M;
1236 	else if (page_size == (1UL << mmu_psize_defs[MMU_PAGE_1G].shift))
1237 		psize = MMU_PAGE_1G;
1238 	else
1239 		return -1;
1240 	return psize;
1241 }
1242 
1243 /*
1244  * Flush partition scoped LPID address translation for all CPUs.
1245  */
1246 void radix__flush_tlb_lpid_page(unsigned int lpid,
1247 					unsigned long addr,
1248 					unsigned long page_size)
1249 {
1250 	int psize = radix_get_mmu_psize(page_size);
1251 
1252 	_tlbie_lpid_va(addr, lpid, psize, RIC_FLUSH_TLB);
1253 }
1254 EXPORT_SYMBOL_GPL(radix__flush_tlb_lpid_page);
1255 
1256 /*
1257  * Flush partition scoped PWC from LPID for all CPUs.
1258  */
1259 void radix__flush_pwc_lpid(unsigned int lpid)
1260 {
1261 	_tlbie_lpid(lpid, RIC_FLUSH_PWC);
1262 }
1263 EXPORT_SYMBOL_GPL(radix__flush_pwc_lpid);
1264 
1265 /*
1266  * Flush partition scoped translations from LPID (=LPIDR)
1267  */
1268 void radix__flush_all_lpid(unsigned int lpid)
1269 {
1270 	_tlbie_lpid(lpid, RIC_FLUSH_ALL);
1271 }
1272 EXPORT_SYMBOL_GPL(radix__flush_all_lpid);
1273 
1274 /*
1275  * Flush process scoped translations from LPID (=LPIDR)
1276  */
1277 void radix__flush_all_lpid_guest(unsigned int lpid)
1278 {
1279 	_tlbie_lpid_guest(lpid, RIC_FLUSH_ALL);
1280 }
1281 
1282 void radix__tlb_flush(struct mmu_gather *tlb)
1283 {
1284 	int psize = 0;
1285 	struct mm_struct *mm = tlb->mm;
1286 	int page_size = tlb->page_size;
1287 	unsigned long start = tlb->start;
1288 	unsigned long end = tlb->end;
1289 
1290 	/*
1291 	 * if page size is not something we understand, do a full mm flush
1292 	 *
1293 	 * A "fullmm" flush must always do a flush_all_mm (RIC=2) flush
1294 	 * that flushes the process table entry cache upon process teardown.
1295 	 * See the comment for radix in arch_exit_mmap().
1296 	 */
1297 	if (tlb->fullmm || tlb->need_flush_all) {
1298 		__flush_all_mm(mm, true);
1299 	} else if ( (psize = radix_get_mmu_psize(page_size)) == -1) {
1300 		if (!tlb->freed_tables)
1301 			radix__flush_tlb_mm(mm);
1302 		else
1303 			radix__flush_all_mm(mm);
1304 	} else {
1305 		if (!tlb->freed_tables)
1306 			radix__flush_tlb_range_psize(mm, start, end, psize);
1307 		else
1308 			radix__flush_tlb_pwc_range_psize(mm, start, end, psize);
1309 	}
1310 }
1311 
1312 static void __radix__flush_tlb_range_psize(struct mm_struct *mm,
1313 				unsigned long start, unsigned long end,
1314 				int psize, bool also_pwc)
1315 {
1316 	unsigned long pid;
1317 	unsigned int page_shift = mmu_psize_defs[psize].shift;
1318 	unsigned long page_size = 1UL << page_shift;
1319 	unsigned long nr_pages = (end - start) >> page_shift;
1320 	bool fullmm = (end == TLB_FLUSH_ALL);
1321 	bool flush_pid;
1322 	enum tlb_flush_type type;
1323 
1324 	pid = mm->context.id;
1325 	if (unlikely(pid == MMU_NO_CONTEXT))
1326 		return;
1327 
1328 	fullmm = (end == TLB_FLUSH_ALL);
1329 
1330 	preempt_disable();
1331 	smp_mb(); /* see radix__flush_tlb_mm */
1332 	type = flush_type_needed(mm, fullmm);
1333 	if (type == FLUSH_TYPE_NONE)
1334 		goto out;
1335 
1336 	if (fullmm)
1337 		flush_pid = true;
1338 	else if (type == FLUSH_TYPE_GLOBAL)
1339 		flush_pid = nr_pages > tlb_single_page_flush_ceiling;
1340 	else
1341 		flush_pid = nr_pages > tlb_local_single_page_flush_ceiling;
1342 
1343 	if (!mmu_has_feature(MMU_FTR_GTSE) && type == FLUSH_TYPE_GLOBAL) {
1344 		unsigned long tgt = H_RPTI_TARGET_CMMU;
1345 		unsigned long type = H_RPTI_TYPE_TLB;
1346 		unsigned long pg_sizes = psize_to_rpti_pgsize(psize);
1347 
1348 		if (also_pwc)
1349 			type |= H_RPTI_TYPE_PWC;
1350 		if (atomic_read(&mm->context.copros) > 0)
1351 			tgt |= H_RPTI_TARGET_NMMU;
1352 		pseries_rpt_invalidate(pid, tgt, type, pg_sizes, start, end);
1353 	} else if (flush_pid) {
1354 		if (type == FLUSH_TYPE_LOCAL) {
1355 			_tlbiel_pid(pid, also_pwc ? RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1356 		} else {
1357 			if (cputlb_use_tlbie()) {
1358 				if (mm_needs_flush_escalation(mm))
1359 					also_pwc = true;
1360 
1361 				_tlbie_pid(pid,
1362 					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1363 			} else {
1364 				_tlbiel_pid_multicast(mm, pid,
1365 					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1366 			}
1367 
1368 		}
1369 	} else {
1370 		if (type == FLUSH_TYPE_LOCAL)
1371 			_tlbiel_va_range(start, end, pid, page_size, psize, also_pwc);
1372 		else if (cputlb_use_tlbie())
1373 			_tlbie_va_range(start, end, pid, page_size, psize, also_pwc);
1374 		else
1375 			_tlbiel_va_range_multicast(mm,
1376 					start, end, pid, page_size, psize, also_pwc);
1377 	}
1378 out:
1379 	preempt_enable();
1380 }
1381 
1382 void radix__flush_tlb_range_psize(struct mm_struct *mm, unsigned long start,
1383 				  unsigned long end, int psize)
1384 {
1385 	return __radix__flush_tlb_range_psize(mm, start, end, psize, false);
1386 }
1387 
1388 void radix__flush_tlb_pwc_range_psize(struct mm_struct *mm, unsigned long start,
1389 				      unsigned long end, int psize)
1390 {
1391 	__radix__flush_tlb_range_psize(mm, start, end, psize, true);
1392 }
1393 
1394 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1395 void radix__flush_tlb_collapsed_pmd(struct mm_struct *mm, unsigned long addr)
1396 {
1397 	unsigned long pid, end;
1398 	enum tlb_flush_type type;
1399 
1400 	pid = mm->context.id;
1401 	if (unlikely(pid == MMU_NO_CONTEXT))
1402 		return;
1403 
1404 	/* 4k page size, just blow the world */
1405 	if (PAGE_SIZE == 0x1000) {
1406 		radix__flush_all_mm(mm);
1407 		return;
1408 	}
1409 
1410 	end = addr + HPAGE_PMD_SIZE;
1411 
1412 	/* Otherwise first do the PWC, then iterate the pages. */
1413 	preempt_disable();
1414 	smp_mb(); /* see radix__flush_tlb_mm */
1415 	type = flush_type_needed(mm, false);
1416 	if (type == FLUSH_TYPE_LOCAL) {
1417 		_tlbiel_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1418 	} else if (type == FLUSH_TYPE_GLOBAL) {
1419 		if (!mmu_has_feature(MMU_FTR_GTSE)) {
1420 			unsigned long tgt, type, pg_sizes;
1421 
1422 			tgt = H_RPTI_TARGET_CMMU;
1423 			type = H_RPTI_TYPE_TLB | H_RPTI_TYPE_PWC |
1424 			       H_RPTI_TYPE_PRT;
1425 			pg_sizes = psize_to_rpti_pgsize(mmu_virtual_psize);
1426 
1427 			if (atomic_read(&mm->context.copros) > 0)
1428 				tgt |= H_RPTI_TARGET_NMMU;
1429 			pseries_rpt_invalidate(pid, tgt, type, pg_sizes,
1430 					       addr, end);
1431 		} else if (cputlb_use_tlbie())
1432 			_tlbie_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1433 		else
1434 			_tlbiel_va_range_multicast(mm,
1435 					addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1436 	}
1437 
1438 	preempt_enable();
1439 }
1440 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1441 
1442 void radix__flush_pmd_tlb_range(struct vm_area_struct *vma,
1443 				unsigned long start, unsigned long end)
1444 {
1445 	radix__flush_tlb_range_psize(vma->vm_mm, start, end, MMU_PAGE_2M);
1446 }
1447 EXPORT_SYMBOL(radix__flush_pmd_tlb_range);
1448 
1449 void radix__flush_tlb_all(void)
1450 {
1451 	unsigned long rb,prs,r,rs;
1452 	unsigned long ric = RIC_FLUSH_ALL;
1453 
1454 	rb = 0x3 << PPC_BITLSHIFT(53); /* IS = 3 */
1455 	prs = 0; /* partition scoped */
1456 	r = 1;   /* radix format */
1457 	rs = 1 & ((1UL << 32) - 1); /* any LPID value to flush guest mappings */
1458 
1459 	asm volatile("ptesync": : :"memory");
1460 	/*
1461 	 * now flush guest entries by passing PRS = 1 and LPID != 0
1462 	 */
1463 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
1464 		     : : "r"(rb), "i"(r), "i"(1), "i"(ric), "r"(rs) : "memory");
1465 	/*
1466 	 * now flush host entires by passing PRS = 0 and LPID == 0
1467 	 */
1468 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
1469 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(0) : "memory");
1470 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
1471 }
1472 
1473 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1474 /*
1475  * Performs process-scoped invalidations for a given LPID
1476  * as part of H_RPT_INVALIDATE hcall.
1477  */
1478 void do_h_rpt_invalidate_prt(unsigned long pid, unsigned long lpid,
1479 			     unsigned long type, unsigned long pg_sizes,
1480 			     unsigned long start, unsigned long end)
1481 {
1482 	unsigned long psize, nr_pages;
1483 	struct mmu_psize_def *def;
1484 	bool flush_pid;
1485 
1486 	/*
1487 	 * A H_RPTI_TYPE_ALL request implies RIC=3, hence
1488 	 * do a single IS=1 based flush.
1489 	 */
1490 	if ((type & H_RPTI_TYPE_ALL) == H_RPTI_TYPE_ALL) {
1491 		_tlbie_pid_lpid(pid, lpid, RIC_FLUSH_ALL);
1492 		return;
1493 	}
1494 
1495 	if (type & H_RPTI_TYPE_PWC)
1496 		_tlbie_pid_lpid(pid, lpid, RIC_FLUSH_PWC);
1497 
1498 	/* Full PID flush */
1499 	if (start == 0 && end == -1)
1500 		return _tlbie_pid_lpid(pid, lpid, RIC_FLUSH_TLB);
1501 
1502 	/* Do range invalidation for all the valid page sizes */
1503 	for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1504 		def = &mmu_psize_defs[psize];
1505 		if (!(pg_sizes & def->h_rpt_pgsize))
1506 			continue;
1507 
1508 		nr_pages = (end - start) >> def->shift;
1509 		flush_pid = nr_pages > tlb_single_page_flush_ceiling;
1510 
1511 		/*
1512 		 * If the number of pages spanning the range is above
1513 		 * the ceiling, convert the request into a full PID flush.
1514 		 * And since PID flush takes out all the page sizes, there
1515 		 * is no need to consider remaining page sizes.
1516 		 */
1517 		if (flush_pid) {
1518 			_tlbie_pid_lpid(pid, lpid, RIC_FLUSH_TLB);
1519 			return;
1520 		}
1521 		_tlbie_va_range_lpid(start, end, pid, lpid,
1522 				     (1UL << def->shift), psize, false);
1523 	}
1524 }
1525 EXPORT_SYMBOL_GPL(do_h_rpt_invalidate_prt);
1526 
1527 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1528 
1529 static int __init create_tlb_single_page_flush_ceiling(void)
1530 {
1531 	debugfs_create_u32("tlb_single_page_flush_ceiling", 0600,
1532 			   arch_debugfs_dir, &tlb_single_page_flush_ceiling);
1533 	debugfs_create_u32("tlb_local_single_page_flush_ceiling", 0600,
1534 			   arch_debugfs_dir, &tlb_local_single_page_flush_ceiling);
1535 	return 0;
1536 }
1537 late_initcall(create_tlb_single_page_flush_ceiling);
1538 
1539