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 constraint 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 = mm->context.id;
704 
705 	if (WARN_ON_ONCE(pid == MMU_NO_CONTEXT))
706 		return;
707 
708 	preempt_disable();
709 	_tlbiel_pid(pid, RIC_FLUSH_TLB);
710 	preempt_enable();
711 }
712 EXPORT_SYMBOL(radix__local_flush_tlb_mm);
713 
714 #ifndef CONFIG_SMP
715 void radix__local_flush_all_mm(struct mm_struct *mm)
716 {
717 	unsigned long pid = mm->context.id;
718 
719 	if (WARN_ON_ONCE(pid == MMU_NO_CONTEXT))
720 		return;
721 
722 	preempt_disable();
723 	_tlbiel_pid(pid, RIC_FLUSH_ALL);
724 	preempt_enable();
725 }
726 EXPORT_SYMBOL(radix__local_flush_all_mm);
727 
728 static void __flush_all_mm(struct mm_struct *mm, bool fullmm)
729 {
730 	radix__local_flush_all_mm(mm);
731 }
732 #endif /* CONFIG_SMP */
733 
734 void radix__local_flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
735 				       int psize)
736 {
737 	unsigned long pid = mm->context.id;
738 
739 	if (WARN_ON_ONCE(pid == MMU_NO_CONTEXT))
740 		return;
741 
742 	preempt_disable();
743 	_tlbiel_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
744 	preempt_enable();
745 }
746 
747 void radix__local_flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
748 {
749 #ifdef CONFIG_HUGETLB_PAGE
750 	/* need the return fix for nohash.c */
751 	if (is_vm_hugetlb_page(vma))
752 		return radix__local_flush_hugetlb_page(vma, vmaddr);
753 #endif
754 	radix__local_flush_tlb_page_psize(vma->vm_mm, vmaddr, mmu_virtual_psize);
755 }
756 EXPORT_SYMBOL(radix__local_flush_tlb_page);
757 
758 static bool mm_needs_flush_escalation(struct mm_struct *mm)
759 {
760 	/*
761 	 * The P9 nest MMU has issues with the page walk cache caching PTEs
762 	 * and not flushing them when RIC = 0 for a PID/LPID invalidate.
763 	 *
764 	 * This may have been fixed in shipping firmware (by disabling PWC
765 	 * or preventing it from caching PTEs), but until that is confirmed,
766 	 * this workaround is required - escalate all RIC=0 IS=1/2/3 flushes
767 	 * to RIC=2.
768 	 *
769 	 * POWER10 (and P9P) does not have this problem.
770 	 */
771 	if (cpu_has_feature(CPU_FTR_ARCH_31))
772 		return false;
773 	if (atomic_read(&mm->context.copros) > 0)
774 		return true;
775 	return false;
776 }
777 
778 /*
779  * If always_flush is true, then flush even if this CPU can't be removed
780  * from mm_cpumask.
781  */
782 void exit_lazy_flush_tlb(struct mm_struct *mm, bool always_flush)
783 {
784 	unsigned long pid = mm->context.id;
785 	int cpu = smp_processor_id();
786 
787 	/*
788 	 * A kthread could have done a mmget_not_zero() after the flushing CPU
789 	 * checked mm_cpumask, and be in the process of kthread_use_mm when
790 	 * interrupted here. In that case, current->mm will be set to mm,
791 	 * because kthread_use_mm() setting ->mm and switching to the mm is
792 	 * done with interrupts off.
793 	 */
794 	if (current->mm == mm)
795 		goto out;
796 
797 	if (current->active_mm == mm) {
798 		unsigned long flags;
799 
800 		WARN_ON_ONCE(current->mm != NULL);
801 		/*
802 		 * It is a kernel thread and is using mm as the lazy tlb, so
803 		 * switch it to init_mm. This is not always called from IPI
804 		 * (e.g., flush_type_needed), so must disable irqs.
805 		 */
806 		local_irq_save(flags);
807 		mmgrab_lazy_tlb(&init_mm);
808 		current->active_mm = &init_mm;
809 		switch_mm_irqs_off(mm, &init_mm, current);
810 		mmdrop_lazy_tlb(mm);
811 		local_irq_restore(flags);
812 	}
813 
814 	/*
815 	 * This IPI may be initiated from any source including those not
816 	 * running the mm, so there may be a racing IPI that comes after
817 	 * this one which finds the cpumask already clear. Check and avoid
818 	 * underflowing the active_cpus count in that case. The race should
819 	 * not otherwise be a problem, but the TLB must be flushed because
820 	 * that's what the caller expects.
821 	 */
822 	if (cpumask_test_cpu(cpu, mm_cpumask(mm))) {
823 		atomic_dec(&mm->context.active_cpus);
824 		cpumask_clear_cpu(cpu, mm_cpumask(mm));
825 		always_flush = true;
826 	}
827 
828 out:
829 	if (always_flush)
830 		_tlbiel_pid(pid, RIC_FLUSH_ALL);
831 }
832 
833 #ifdef CONFIG_SMP
834 static void do_exit_flush_lazy_tlb(void *arg)
835 {
836 	struct mm_struct *mm = arg;
837 	exit_lazy_flush_tlb(mm, true);
838 }
839 
840 static void exit_flush_lazy_tlbs(struct mm_struct *mm)
841 {
842 	/*
843 	 * Would be nice if this was async so it could be run in
844 	 * parallel with our local flush, but generic code does not
845 	 * give a good API for it. Could extend the generic code or
846 	 * make a special powerpc IPI for flushing TLBs.
847 	 * For now it's not too performance critical.
848 	 */
849 	smp_call_function_many(mm_cpumask(mm), do_exit_flush_lazy_tlb,
850 				(void *)mm, 1);
851 }
852 
853 #else /* CONFIG_SMP */
854 static inline void exit_flush_lazy_tlbs(struct mm_struct *mm) { }
855 #endif /* CONFIG_SMP */
856 
857 static DEFINE_PER_CPU(unsigned int, mm_cpumask_trim_clock);
858 
859 /*
860  * Interval between flushes at which we send out IPIs to check whether the
861  * mm_cpumask can be trimmed for the case where it's not a single-threaded
862  * process flushing its own mm. The intent is to reduce the cost of later
863  * flushes. Don't want this to be so low that it adds noticable cost to TLB
864  * flushing, or so high that it doesn't help reduce global TLBIEs.
865  */
866 static unsigned long tlb_mm_cpumask_trim_timer = 1073;
867 
868 static bool tick_and_test_trim_clock(void)
869 {
870 	if (__this_cpu_inc_return(mm_cpumask_trim_clock) ==
871 			tlb_mm_cpumask_trim_timer) {
872 		__this_cpu_write(mm_cpumask_trim_clock, 0);
873 		return true;
874 	}
875 	return false;
876 }
877 
878 enum tlb_flush_type {
879 	FLUSH_TYPE_NONE,
880 	FLUSH_TYPE_LOCAL,
881 	FLUSH_TYPE_GLOBAL,
882 };
883 
884 static enum tlb_flush_type flush_type_needed(struct mm_struct *mm, bool fullmm)
885 {
886 	int active_cpus = atomic_read(&mm->context.active_cpus);
887 	int cpu = smp_processor_id();
888 
889 	if (active_cpus == 0)
890 		return FLUSH_TYPE_NONE;
891 	if (active_cpus == 1 && cpumask_test_cpu(cpu, mm_cpumask(mm))) {
892 		if (current->mm != mm) {
893 			/*
894 			 * Asynchronous flush sources may trim down to nothing
895 			 * if the process is not running, so occasionally try
896 			 * to trim.
897 			 */
898 			if (tick_and_test_trim_clock()) {
899 				exit_lazy_flush_tlb(mm, true);
900 				return FLUSH_TYPE_NONE;
901 			}
902 		}
903 		return FLUSH_TYPE_LOCAL;
904 	}
905 
906 	/* Coprocessors require TLBIE to invalidate nMMU. */
907 	if (atomic_read(&mm->context.copros) > 0)
908 		return FLUSH_TYPE_GLOBAL;
909 
910 	/*
911 	 * In the fullmm case there's no point doing the exit_flush_lazy_tlbs
912 	 * because the mm is being taken down anyway, and a TLBIE tends to
913 	 * be faster than an IPI+TLBIEL.
914 	 */
915 	if (fullmm)
916 		return FLUSH_TYPE_GLOBAL;
917 
918 	/*
919 	 * If we are running the only thread of a single-threaded process,
920 	 * then we should almost always be able to trim off the rest of the
921 	 * CPU mask (except in the case of use_mm() races), so always try
922 	 * trimming the mask.
923 	 */
924 	if (atomic_read(&mm->mm_users) <= 1 && current->mm == mm) {
925 		exit_flush_lazy_tlbs(mm);
926 		/*
927 		 * use_mm() race could prevent IPIs from being able to clear
928 		 * the cpumask here, however those users are established
929 		 * after our first check (and so after the PTEs are removed),
930 		 * and the TLB still gets flushed by the IPI, so this CPU
931 		 * will only require a local flush.
932 		 */
933 		return FLUSH_TYPE_LOCAL;
934 	}
935 
936 	/*
937 	 * Occasionally try to trim down the cpumask. It's possible this can
938 	 * bring the mask to zero, which results in no flush.
939 	 */
940 	if (tick_and_test_trim_clock()) {
941 		exit_flush_lazy_tlbs(mm);
942 		if (current->mm == mm)
943 			return FLUSH_TYPE_LOCAL;
944 		if (cpumask_test_cpu(cpu, mm_cpumask(mm)))
945 			exit_lazy_flush_tlb(mm, true);
946 		return FLUSH_TYPE_NONE;
947 	}
948 
949 	return FLUSH_TYPE_GLOBAL;
950 }
951 
952 #ifdef CONFIG_SMP
953 void radix__flush_tlb_mm(struct mm_struct *mm)
954 {
955 	unsigned long pid;
956 	enum tlb_flush_type type;
957 
958 	pid = mm->context.id;
959 	if (WARN_ON_ONCE(pid == MMU_NO_CONTEXT))
960 		return;
961 
962 	preempt_disable();
963 	/*
964 	 * Order loads of mm_cpumask (in flush_type_needed) vs previous
965 	 * stores to clear ptes before the invalidate. See barrier in
966 	 * switch_mm_irqs_off
967 	 */
968 	smp_mb();
969 	type = flush_type_needed(mm, false);
970 	if (type == FLUSH_TYPE_LOCAL) {
971 		_tlbiel_pid(pid, RIC_FLUSH_TLB);
972 	} else if (type == FLUSH_TYPE_GLOBAL) {
973 		if (!mmu_has_feature(MMU_FTR_GTSE)) {
974 			unsigned long tgt = H_RPTI_TARGET_CMMU;
975 
976 			if (atomic_read(&mm->context.copros) > 0)
977 				tgt |= H_RPTI_TARGET_NMMU;
978 			pseries_rpt_invalidate(pid, tgt, H_RPTI_TYPE_TLB,
979 					       H_RPTI_PAGE_ALL, 0, -1UL);
980 		} else if (cputlb_use_tlbie()) {
981 			if (mm_needs_flush_escalation(mm))
982 				_tlbie_pid(pid, RIC_FLUSH_ALL);
983 			else
984 				_tlbie_pid(pid, RIC_FLUSH_TLB);
985 		} else {
986 			_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_TLB);
987 		}
988 	}
989 	preempt_enable();
990 }
991 EXPORT_SYMBOL(radix__flush_tlb_mm);
992 
993 static void __flush_all_mm(struct mm_struct *mm, bool fullmm)
994 {
995 	unsigned long pid;
996 	enum tlb_flush_type type;
997 
998 	pid = mm->context.id;
999 	if (WARN_ON_ONCE(pid == MMU_NO_CONTEXT))
1000 		return;
1001 
1002 	preempt_disable();
1003 	smp_mb(); /* see radix__flush_tlb_mm */
1004 	type = flush_type_needed(mm, fullmm);
1005 	if (type == FLUSH_TYPE_LOCAL) {
1006 		_tlbiel_pid(pid, RIC_FLUSH_ALL);
1007 	} else if (type == FLUSH_TYPE_GLOBAL) {
1008 		if (!mmu_has_feature(MMU_FTR_GTSE)) {
1009 			unsigned long tgt = H_RPTI_TARGET_CMMU;
1010 			unsigned long type = H_RPTI_TYPE_TLB | H_RPTI_TYPE_PWC |
1011 					     H_RPTI_TYPE_PRT;
1012 
1013 			if (atomic_read(&mm->context.copros) > 0)
1014 				tgt |= H_RPTI_TARGET_NMMU;
1015 			pseries_rpt_invalidate(pid, tgt, type,
1016 					       H_RPTI_PAGE_ALL, 0, -1UL);
1017 		} else if (cputlb_use_tlbie())
1018 			_tlbie_pid(pid, RIC_FLUSH_ALL);
1019 		else
1020 			_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_ALL);
1021 	}
1022 	preempt_enable();
1023 }
1024 
1025 void radix__flush_all_mm(struct mm_struct *mm)
1026 {
1027 	__flush_all_mm(mm, false);
1028 }
1029 EXPORT_SYMBOL(radix__flush_all_mm);
1030 
1031 void radix__flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
1032 				 int psize)
1033 {
1034 	unsigned long pid;
1035 	enum tlb_flush_type type;
1036 
1037 	pid = mm->context.id;
1038 	if (WARN_ON_ONCE(pid == MMU_NO_CONTEXT))
1039 		return;
1040 
1041 	preempt_disable();
1042 	smp_mb(); /* see radix__flush_tlb_mm */
1043 	type = flush_type_needed(mm, false);
1044 	if (type == FLUSH_TYPE_LOCAL) {
1045 		_tlbiel_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
1046 	} else if (type == FLUSH_TYPE_GLOBAL) {
1047 		if (!mmu_has_feature(MMU_FTR_GTSE)) {
1048 			unsigned long tgt, pg_sizes, size;
1049 
1050 			tgt = H_RPTI_TARGET_CMMU;
1051 			pg_sizes = psize_to_rpti_pgsize(psize);
1052 			size = 1UL << mmu_psize_to_shift(psize);
1053 
1054 			if (atomic_read(&mm->context.copros) > 0)
1055 				tgt |= H_RPTI_TARGET_NMMU;
1056 			pseries_rpt_invalidate(pid, tgt, H_RPTI_TYPE_TLB,
1057 					       pg_sizes, vmaddr,
1058 					       vmaddr + size);
1059 		} else if (cputlb_use_tlbie())
1060 			_tlbie_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
1061 		else
1062 			_tlbiel_va_multicast(mm, vmaddr, pid, psize, RIC_FLUSH_TLB);
1063 	}
1064 	preempt_enable();
1065 }
1066 
1067 void radix__flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
1068 {
1069 #ifdef CONFIG_HUGETLB_PAGE
1070 	if (is_vm_hugetlb_page(vma))
1071 		return radix__flush_hugetlb_page(vma, vmaddr);
1072 #endif
1073 	radix__flush_tlb_page_psize(vma->vm_mm, vmaddr, mmu_virtual_psize);
1074 }
1075 EXPORT_SYMBOL(radix__flush_tlb_page);
1076 
1077 #endif /* CONFIG_SMP */
1078 
1079 static void do_tlbiel_kernel(void *info)
1080 {
1081 	_tlbiel_pid(0, RIC_FLUSH_ALL);
1082 }
1083 
1084 static inline void _tlbiel_kernel_broadcast(void)
1085 {
1086 	on_each_cpu(do_tlbiel_kernel, NULL, 1);
1087 	if (tlbie_capable) {
1088 		/*
1089 		 * Coherent accelerators don't refcount kernel memory mappings,
1090 		 * so have to always issue a tlbie for them. This is quite a
1091 		 * slow path anyway.
1092 		 */
1093 		_tlbie_pid(0, RIC_FLUSH_ALL);
1094 	}
1095 }
1096 
1097 /*
1098  * If kernel TLBIs ever become local rather than global, then
1099  * drivers/misc/ocxl/link.c:ocxl_link_add_pe will need some work, as it
1100  * assumes kernel TLBIs are global.
1101  */
1102 void radix__flush_tlb_kernel_range(unsigned long start, unsigned long end)
1103 {
1104 	if (!mmu_has_feature(MMU_FTR_GTSE)) {
1105 		unsigned long tgt = H_RPTI_TARGET_CMMU | H_RPTI_TARGET_NMMU;
1106 		unsigned long type = H_RPTI_TYPE_TLB | H_RPTI_TYPE_PWC |
1107 				     H_RPTI_TYPE_PRT;
1108 
1109 		pseries_rpt_invalidate(0, tgt, type, H_RPTI_PAGE_ALL,
1110 				       start, end);
1111 	} else if (cputlb_use_tlbie())
1112 		_tlbie_pid(0, RIC_FLUSH_ALL);
1113 	else
1114 		_tlbiel_kernel_broadcast();
1115 }
1116 EXPORT_SYMBOL(radix__flush_tlb_kernel_range);
1117 
1118 /*
1119  * Doesn't appear to be used anywhere. Remove.
1120  */
1121 #define TLB_FLUSH_ALL -1UL
1122 
1123 /*
1124  * Number of pages above which we invalidate the entire PID rather than
1125  * flush individual pages, for local and global flushes respectively.
1126  *
1127  * tlbie goes out to the interconnect and individual ops are more costly.
1128  * It also does not iterate over sets like the local tlbiel variant when
1129  * invalidating a full PID, so it has a far lower threshold to change from
1130  * individual page flushes to full-pid flushes.
1131  */
1132 static u32 tlb_single_page_flush_ceiling __read_mostly = 33;
1133 static u32 tlb_local_single_page_flush_ceiling __read_mostly = POWER9_TLB_SETS_RADIX * 2;
1134 
1135 static inline void __radix__flush_tlb_range(struct mm_struct *mm,
1136 					    unsigned long start, unsigned long end)
1137 {
1138 	unsigned long pid;
1139 	unsigned int page_shift = mmu_psize_defs[mmu_virtual_psize].shift;
1140 	unsigned long page_size = 1UL << page_shift;
1141 	unsigned long nr_pages = (end - start) >> page_shift;
1142 	bool flush_pid, flush_pwc = false;
1143 	enum tlb_flush_type type;
1144 
1145 	pid = mm->context.id;
1146 	if (WARN_ON_ONCE(pid == MMU_NO_CONTEXT))
1147 		return;
1148 
1149 	WARN_ON_ONCE(end == TLB_FLUSH_ALL);
1150 
1151 	preempt_disable();
1152 	smp_mb(); /* see radix__flush_tlb_mm */
1153 	type = flush_type_needed(mm, false);
1154 	if (type == FLUSH_TYPE_NONE)
1155 		goto out;
1156 
1157 	if (type == FLUSH_TYPE_GLOBAL)
1158 		flush_pid = nr_pages > tlb_single_page_flush_ceiling;
1159 	else
1160 		flush_pid = nr_pages > tlb_local_single_page_flush_ceiling;
1161 	/*
1162 	 * full pid flush already does the PWC flush. if it is not full pid
1163 	 * flush check the range is more than PMD and force a pwc flush
1164 	 * mremap() depends on this behaviour.
1165 	 */
1166 	if (!flush_pid && (end - start) >= PMD_SIZE)
1167 		flush_pwc = true;
1168 
1169 	if (!mmu_has_feature(MMU_FTR_GTSE) && type == FLUSH_TYPE_GLOBAL) {
1170 		unsigned long type = H_RPTI_TYPE_TLB;
1171 		unsigned long tgt = H_RPTI_TARGET_CMMU;
1172 		unsigned long pg_sizes = psize_to_rpti_pgsize(mmu_virtual_psize);
1173 
1174 		if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1175 			pg_sizes |= psize_to_rpti_pgsize(MMU_PAGE_2M);
1176 		if (atomic_read(&mm->context.copros) > 0)
1177 			tgt |= H_RPTI_TARGET_NMMU;
1178 		if (flush_pwc)
1179 			type |= H_RPTI_TYPE_PWC;
1180 		pseries_rpt_invalidate(pid, tgt, type, pg_sizes, start, end);
1181 	} else if (flush_pid) {
1182 		/*
1183 		 * We are now flushing a range larger than PMD size force a RIC_FLUSH_ALL
1184 		 */
1185 		if (type == FLUSH_TYPE_LOCAL) {
1186 			_tlbiel_pid(pid, RIC_FLUSH_ALL);
1187 		} else {
1188 			if (cputlb_use_tlbie()) {
1189 				_tlbie_pid(pid, RIC_FLUSH_ALL);
1190 			} else {
1191 				_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_ALL);
1192 			}
1193 		}
1194 	} else {
1195 		bool hflush;
1196 		unsigned long hstart, hend;
1197 
1198 		hstart = (start + PMD_SIZE - 1) & PMD_MASK;
1199 		hend = end & PMD_MASK;
1200 		hflush = IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) && hstart < hend;
1201 
1202 		if (type == FLUSH_TYPE_LOCAL) {
1203 			asm volatile("ptesync": : :"memory");
1204 			if (flush_pwc)
1205 				/* For PWC, only one flush is needed */
1206 				__tlbiel_pid(pid, 0, RIC_FLUSH_PWC);
1207 			__tlbiel_va_range(start, end, pid, page_size, mmu_virtual_psize);
1208 			if (hflush)
1209 				__tlbiel_va_range(hstart, hend, pid,
1210 						PMD_SIZE, MMU_PAGE_2M);
1211 			ppc_after_tlbiel_barrier();
1212 		} else if (cputlb_use_tlbie()) {
1213 			asm volatile("ptesync": : :"memory");
1214 			if (flush_pwc)
1215 				__tlbie_pid(pid, RIC_FLUSH_PWC);
1216 			__tlbie_va_range(start, end, pid, page_size, mmu_virtual_psize);
1217 			if (hflush)
1218 				__tlbie_va_range(hstart, hend, pid,
1219 						PMD_SIZE, MMU_PAGE_2M);
1220 			asm volatile("eieio; tlbsync; ptesync": : :"memory");
1221 		} else {
1222 			_tlbiel_va_range_multicast(mm,
1223 					start, end, pid, page_size, mmu_virtual_psize, flush_pwc);
1224 			if (hflush)
1225 				_tlbiel_va_range_multicast(mm,
1226 					hstart, hend, pid, PMD_SIZE, MMU_PAGE_2M, flush_pwc);
1227 		}
1228 	}
1229 out:
1230 	preempt_enable();
1231 }
1232 
1233 void radix__flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
1234 		     unsigned long end)
1235 
1236 {
1237 #ifdef CONFIG_HUGETLB_PAGE
1238 	if (is_vm_hugetlb_page(vma))
1239 		return radix__flush_hugetlb_tlb_range(vma, start, end);
1240 #endif
1241 
1242 	__radix__flush_tlb_range(vma->vm_mm, start, end);
1243 }
1244 EXPORT_SYMBOL(radix__flush_tlb_range);
1245 
1246 static int radix_get_mmu_psize(int page_size)
1247 {
1248 	int psize;
1249 
1250 	if (page_size == (1UL << mmu_psize_defs[mmu_virtual_psize].shift))
1251 		psize = mmu_virtual_psize;
1252 	else if (page_size == (1UL << mmu_psize_defs[MMU_PAGE_2M].shift))
1253 		psize = MMU_PAGE_2M;
1254 	else if (page_size == (1UL << mmu_psize_defs[MMU_PAGE_1G].shift))
1255 		psize = MMU_PAGE_1G;
1256 	else
1257 		return -1;
1258 	return psize;
1259 }
1260 
1261 /*
1262  * Flush partition scoped LPID address translation for all CPUs.
1263  */
1264 void radix__flush_tlb_lpid_page(unsigned int lpid,
1265 					unsigned long addr,
1266 					unsigned long page_size)
1267 {
1268 	int psize = radix_get_mmu_psize(page_size);
1269 
1270 	_tlbie_lpid_va(addr, lpid, psize, RIC_FLUSH_TLB);
1271 }
1272 EXPORT_SYMBOL_GPL(radix__flush_tlb_lpid_page);
1273 
1274 /*
1275  * Flush partition scoped PWC from LPID for all CPUs.
1276  */
1277 void radix__flush_pwc_lpid(unsigned int lpid)
1278 {
1279 	_tlbie_lpid(lpid, RIC_FLUSH_PWC);
1280 }
1281 EXPORT_SYMBOL_GPL(radix__flush_pwc_lpid);
1282 
1283 /*
1284  * Flush partition scoped translations from LPID (=LPIDR)
1285  */
1286 void radix__flush_all_lpid(unsigned int lpid)
1287 {
1288 	_tlbie_lpid(lpid, RIC_FLUSH_ALL);
1289 }
1290 EXPORT_SYMBOL_GPL(radix__flush_all_lpid);
1291 
1292 /*
1293  * Flush process scoped translations from LPID (=LPIDR)
1294  */
1295 void radix__flush_all_lpid_guest(unsigned int lpid)
1296 {
1297 	_tlbie_lpid_guest(lpid, RIC_FLUSH_ALL);
1298 }
1299 
1300 void radix__tlb_flush(struct mmu_gather *tlb)
1301 {
1302 	int psize = 0;
1303 	struct mm_struct *mm = tlb->mm;
1304 	int page_size = tlb->page_size;
1305 	unsigned long start = tlb->start;
1306 	unsigned long end = tlb->end;
1307 
1308 	/*
1309 	 * if page size is not something we understand, do a full mm flush
1310 	 *
1311 	 * A "fullmm" flush must always do a flush_all_mm (RIC=2) flush
1312 	 * that flushes the process table entry cache upon process teardown.
1313 	 * See the comment for radix in arch_exit_mmap().
1314 	 */
1315 	if (tlb->fullmm) {
1316 		__flush_all_mm(mm, true);
1317 	} else if ( (psize = radix_get_mmu_psize(page_size)) == -1) {
1318 		if (!tlb->freed_tables)
1319 			radix__flush_tlb_mm(mm);
1320 		else
1321 			radix__flush_all_mm(mm);
1322 	} else {
1323 		if (!tlb->freed_tables)
1324 			radix__flush_tlb_range_psize(mm, start, end, psize);
1325 		else
1326 			radix__flush_tlb_pwc_range_psize(mm, start, end, psize);
1327 	}
1328 }
1329 
1330 static void __radix__flush_tlb_range_psize(struct mm_struct *mm,
1331 				unsigned long start, unsigned long end,
1332 				int psize, bool also_pwc)
1333 {
1334 	unsigned long pid;
1335 	unsigned int page_shift = mmu_psize_defs[psize].shift;
1336 	unsigned long page_size = 1UL << page_shift;
1337 	unsigned long nr_pages = (end - start) >> page_shift;
1338 	bool flush_pid;
1339 	enum tlb_flush_type type;
1340 
1341 	pid = mm->context.id;
1342 	if (WARN_ON_ONCE(pid == MMU_NO_CONTEXT))
1343 		return;
1344 
1345 	WARN_ON_ONCE(end == TLB_FLUSH_ALL);
1346 
1347 	preempt_disable();
1348 	smp_mb(); /* see radix__flush_tlb_mm */
1349 	type = flush_type_needed(mm, false);
1350 	if (type == FLUSH_TYPE_NONE)
1351 		goto out;
1352 
1353 	if (type == FLUSH_TYPE_GLOBAL)
1354 		flush_pid = nr_pages > tlb_single_page_flush_ceiling;
1355 	else
1356 		flush_pid = nr_pages > tlb_local_single_page_flush_ceiling;
1357 
1358 	if (!mmu_has_feature(MMU_FTR_GTSE) && type == FLUSH_TYPE_GLOBAL) {
1359 		unsigned long tgt = H_RPTI_TARGET_CMMU;
1360 		unsigned long type = H_RPTI_TYPE_TLB;
1361 		unsigned long pg_sizes = psize_to_rpti_pgsize(psize);
1362 
1363 		if (also_pwc)
1364 			type |= H_RPTI_TYPE_PWC;
1365 		if (atomic_read(&mm->context.copros) > 0)
1366 			tgt |= H_RPTI_TARGET_NMMU;
1367 		pseries_rpt_invalidate(pid, tgt, type, pg_sizes, start, end);
1368 	} else if (flush_pid) {
1369 		if (type == FLUSH_TYPE_LOCAL) {
1370 			_tlbiel_pid(pid, also_pwc ? RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1371 		} else {
1372 			if (cputlb_use_tlbie()) {
1373 				if (mm_needs_flush_escalation(mm))
1374 					also_pwc = true;
1375 
1376 				_tlbie_pid(pid,
1377 					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1378 			} else {
1379 				_tlbiel_pid_multicast(mm, pid,
1380 					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1381 			}
1382 
1383 		}
1384 	} else {
1385 		if (type == FLUSH_TYPE_LOCAL)
1386 			_tlbiel_va_range(start, end, pid, page_size, psize, also_pwc);
1387 		else if (cputlb_use_tlbie())
1388 			_tlbie_va_range(start, end, pid, page_size, psize, also_pwc);
1389 		else
1390 			_tlbiel_va_range_multicast(mm,
1391 					start, end, pid, page_size, psize, also_pwc);
1392 	}
1393 out:
1394 	preempt_enable();
1395 }
1396 
1397 void radix__flush_tlb_range_psize(struct mm_struct *mm, unsigned long start,
1398 				  unsigned long end, int psize)
1399 {
1400 	return __radix__flush_tlb_range_psize(mm, start, end, psize, false);
1401 }
1402 
1403 void radix__flush_tlb_pwc_range_psize(struct mm_struct *mm, unsigned long start,
1404 				      unsigned long end, int psize)
1405 {
1406 	__radix__flush_tlb_range_psize(mm, start, end, psize, true);
1407 }
1408 
1409 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1410 void radix__flush_tlb_collapsed_pmd(struct mm_struct *mm, unsigned long addr)
1411 {
1412 	unsigned long pid, end;
1413 	enum tlb_flush_type type;
1414 
1415 	pid = mm->context.id;
1416 	if (WARN_ON_ONCE(pid == MMU_NO_CONTEXT))
1417 		return;
1418 
1419 	/* 4k page size, just blow the world */
1420 	if (PAGE_SIZE == 0x1000) {
1421 		radix__flush_all_mm(mm);
1422 		return;
1423 	}
1424 
1425 	end = addr + HPAGE_PMD_SIZE;
1426 
1427 	/* Otherwise first do the PWC, then iterate the pages. */
1428 	preempt_disable();
1429 	smp_mb(); /* see radix__flush_tlb_mm */
1430 	type = flush_type_needed(mm, false);
1431 	if (type == FLUSH_TYPE_LOCAL) {
1432 		_tlbiel_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1433 	} else if (type == FLUSH_TYPE_GLOBAL) {
1434 		if (!mmu_has_feature(MMU_FTR_GTSE)) {
1435 			unsigned long tgt, type, pg_sizes;
1436 
1437 			tgt = H_RPTI_TARGET_CMMU;
1438 			type = H_RPTI_TYPE_TLB | H_RPTI_TYPE_PWC |
1439 			       H_RPTI_TYPE_PRT;
1440 			pg_sizes = psize_to_rpti_pgsize(mmu_virtual_psize);
1441 
1442 			if (atomic_read(&mm->context.copros) > 0)
1443 				tgt |= H_RPTI_TARGET_NMMU;
1444 			pseries_rpt_invalidate(pid, tgt, type, pg_sizes,
1445 					       addr, end);
1446 		} else if (cputlb_use_tlbie())
1447 			_tlbie_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1448 		else
1449 			_tlbiel_va_range_multicast(mm,
1450 					addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1451 	}
1452 
1453 	preempt_enable();
1454 }
1455 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1456 
1457 void radix__flush_pmd_tlb_range(struct vm_area_struct *vma,
1458 				unsigned long start, unsigned long end)
1459 {
1460 	radix__flush_tlb_range_psize(vma->vm_mm, start, end, MMU_PAGE_2M);
1461 }
1462 EXPORT_SYMBOL(radix__flush_pmd_tlb_range);
1463 
1464 void radix__flush_tlb_all(void)
1465 {
1466 	unsigned long rb,prs,r,rs;
1467 	unsigned long ric = RIC_FLUSH_ALL;
1468 
1469 	rb = 0x3 << PPC_BITLSHIFT(53); /* IS = 3 */
1470 	prs = 0; /* partition scoped */
1471 	r = 1;   /* radix format */
1472 	rs = 1 & ((1UL << 32) - 1); /* any LPID value to flush guest mappings */
1473 
1474 	asm volatile("ptesync": : :"memory");
1475 	/*
1476 	 * now flush guest entries by passing PRS = 1 and LPID != 0
1477 	 */
1478 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
1479 		     : : "r"(rb), "i"(r), "i"(1), "i"(ric), "r"(rs) : "memory");
1480 	/*
1481 	 * now flush host entires by passing PRS = 0 and LPID == 0
1482 	 */
1483 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
1484 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(0) : "memory");
1485 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
1486 }
1487 
1488 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1489 /*
1490  * Performs process-scoped invalidations for a given LPID
1491  * as part of H_RPT_INVALIDATE hcall.
1492  */
1493 void do_h_rpt_invalidate_prt(unsigned long pid, unsigned long lpid,
1494 			     unsigned long type, unsigned long pg_sizes,
1495 			     unsigned long start, unsigned long end)
1496 {
1497 	unsigned long psize, nr_pages;
1498 	struct mmu_psize_def *def;
1499 	bool flush_pid;
1500 
1501 	/*
1502 	 * A H_RPTI_TYPE_ALL request implies RIC=3, hence
1503 	 * do a single IS=1 based flush.
1504 	 */
1505 	if ((type & H_RPTI_TYPE_ALL) == H_RPTI_TYPE_ALL) {
1506 		_tlbie_pid_lpid(pid, lpid, RIC_FLUSH_ALL);
1507 		return;
1508 	}
1509 
1510 	if (type & H_RPTI_TYPE_PWC)
1511 		_tlbie_pid_lpid(pid, lpid, RIC_FLUSH_PWC);
1512 
1513 	/* Full PID flush */
1514 	if (start == 0 && end == -1)
1515 		return _tlbie_pid_lpid(pid, lpid, RIC_FLUSH_TLB);
1516 
1517 	/* Do range invalidation for all the valid page sizes */
1518 	for (psize = 0; psize < MMU_PAGE_COUNT; psize++) {
1519 		def = &mmu_psize_defs[psize];
1520 		if (!(pg_sizes & def->h_rpt_pgsize))
1521 			continue;
1522 
1523 		nr_pages = (end - start) >> def->shift;
1524 		flush_pid = nr_pages > tlb_single_page_flush_ceiling;
1525 
1526 		/*
1527 		 * If the number of pages spanning the range is above
1528 		 * the ceiling, convert the request into a full PID flush.
1529 		 * And since PID flush takes out all the page sizes, there
1530 		 * is no need to consider remaining page sizes.
1531 		 */
1532 		if (flush_pid) {
1533 			_tlbie_pid_lpid(pid, lpid, RIC_FLUSH_TLB);
1534 			return;
1535 		}
1536 		_tlbie_va_range_lpid(start, end, pid, lpid,
1537 				     (1UL << def->shift), psize, false);
1538 	}
1539 }
1540 EXPORT_SYMBOL_GPL(do_h_rpt_invalidate_prt);
1541 
1542 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1543 
1544 static int __init create_tlb_single_page_flush_ceiling(void)
1545 {
1546 	debugfs_create_u32("tlb_single_page_flush_ceiling", 0600,
1547 			   arch_debugfs_dir, &tlb_single_page_flush_ceiling);
1548 	debugfs_create_u32("tlb_local_single_page_flush_ceiling", 0600,
1549 			   arch_debugfs_dir, &tlb_local_single_page_flush_ceiling);
1550 	return 0;
1551 }
1552 late_initcall(create_tlb_single_page_flush_ceiling);
1553 
1554