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 
14 #include <asm/ppc-opcode.h>
15 #include <asm/tlb.h>
16 #include <asm/tlbflush.h>
17 #include <asm/trace.h>
18 #include <asm/cputhreads.h>
19 
20 #define RIC_FLUSH_TLB 0
21 #define RIC_FLUSH_PWC 1
22 #define RIC_FLUSH_ALL 2
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 		for (set = 1; set < num_sets; set++)
59 			tlbiel_radix_set_isa300(set, is, 0, RIC_FLUSH_TLB, 0);
60 	}
61 
62 	/* Flush process scoped entries. */
63 	tlbiel_radix_set_isa300(0, is, 0, RIC_FLUSH_ALL, 1);
64 	for (set = 1; set < num_sets; set++)
65 		tlbiel_radix_set_isa300(set, is, 0, RIC_FLUSH_TLB, 1);
66 
67 	asm volatile("ptesync": : :"memory");
68 }
69 
70 void radix__tlbiel_all(unsigned int action)
71 {
72 	unsigned int is;
73 
74 	switch (action) {
75 	case TLB_INVAL_SCOPE_GLOBAL:
76 		is = 3;
77 		break;
78 	case TLB_INVAL_SCOPE_LPID:
79 		is = 2;
80 		break;
81 	default:
82 		BUG();
83 	}
84 
85 	if (early_cpu_has_feature(CPU_FTR_ARCH_300))
86 		tlbiel_all_isa300(POWER9_TLB_SETS_RADIX, is);
87 	else
88 		WARN(1, "%s called on pre-POWER9 CPU\n", __func__);
89 
90 	asm volatile(PPC_ISA_3_0_INVALIDATE_ERAT "; isync" : : :"memory");
91 }
92 
93 static __always_inline void __tlbiel_pid(unsigned long pid, int set,
94 				unsigned long ric)
95 {
96 	unsigned long rb,rs,prs,r;
97 
98 	rb = PPC_BIT(53); /* IS = 1 */
99 	rb |= set << PPC_BITLSHIFT(51);
100 	rs = ((unsigned long)pid) << PPC_BITLSHIFT(31);
101 	prs = 1; /* process scoped */
102 	r = 1;   /* radix format */
103 
104 	asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
105 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
106 	trace_tlbie(0, 1, rb, rs, ric, prs, r);
107 }
108 
109 static __always_inline void __tlbie_pid(unsigned long pid, unsigned long ric)
110 {
111 	unsigned long rb,rs,prs,r;
112 
113 	rb = PPC_BIT(53); /* IS = 1 */
114 	rs = pid << PPC_BITLSHIFT(31);
115 	prs = 1; /* process scoped */
116 	r = 1;   /* radix format */
117 
118 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
119 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
120 	trace_tlbie(0, 0, rb, rs, ric, prs, r);
121 }
122 
123 static __always_inline void __tlbie_lpid(unsigned long lpid, unsigned long ric)
124 {
125 	unsigned long rb,rs,prs,r;
126 
127 	rb = PPC_BIT(52); /* IS = 2 */
128 	rs = lpid;
129 	prs = 0; /* partition scoped */
130 	r = 1;   /* radix format */
131 
132 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
133 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
134 	trace_tlbie(lpid, 0, rb, rs, ric, prs, r);
135 }
136 
137 static __always_inline void __tlbie_lpid_guest(unsigned long lpid, unsigned long ric)
138 {
139 	unsigned long rb,rs,prs,r;
140 
141 	rb = PPC_BIT(52); /* IS = 2 */
142 	rs = lpid;
143 	prs = 1; /* process scoped */
144 	r = 1;   /* radix format */
145 
146 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
147 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
148 	trace_tlbie(lpid, 0, rb, rs, ric, prs, r);
149 }
150 
151 static __always_inline void __tlbiel_va(unsigned long va, unsigned long pid,
152 					unsigned long ap, unsigned long ric)
153 {
154 	unsigned long rb,rs,prs,r;
155 
156 	rb = va & ~(PPC_BITMASK(52, 63));
157 	rb |= ap << PPC_BITLSHIFT(58);
158 	rs = pid << PPC_BITLSHIFT(31);
159 	prs = 1; /* process scoped */
160 	r = 1;   /* radix format */
161 
162 	asm volatile(PPC_TLBIEL(%0, %4, %3, %2, %1)
163 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
164 	trace_tlbie(0, 1, rb, rs, ric, prs, r);
165 }
166 
167 static __always_inline void __tlbie_va(unsigned long va, unsigned long pid,
168 				       unsigned long ap, unsigned long ric)
169 {
170 	unsigned long rb,rs,prs,r;
171 
172 	rb = va & ~(PPC_BITMASK(52, 63));
173 	rb |= ap << PPC_BITLSHIFT(58);
174 	rs = pid << PPC_BITLSHIFT(31);
175 	prs = 1; /* process scoped */
176 	r = 1;   /* radix format */
177 
178 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
179 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
180 	trace_tlbie(0, 0, rb, rs, ric, prs, r);
181 }
182 
183 static __always_inline void __tlbie_lpid_va(unsigned long va, unsigned long lpid,
184 					    unsigned long ap, unsigned long ric)
185 {
186 	unsigned long rb,rs,prs,r;
187 
188 	rb = va & ~(PPC_BITMASK(52, 63));
189 	rb |= ap << PPC_BITLSHIFT(58);
190 	rs = lpid;
191 	prs = 0; /* partition scoped */
192 	r = 1;   /* radix format */
193 
194 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
195 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(rs) : "memory");
196 	trace_tlbie(lpid, 0, rb, rs, ric, prs, r);
197 }
198 
199 
200 static inline void fixup_tlbie_va(unsigned long va, unsigned long pid,
201 				  unsigned long ap)
202 {
203 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
204 		asm volatile("ptesync": : :"memory");
205 		__tlbie_va(va, 0, ap, RIC_FLUSH_TLB);
206 	}
207 
208 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
209 		asm volatile("ptesync": : :"memory");
210 		__tlbie_va(va, pid, ap, RIC_FLUSH_TLB);
211 	}
212 }
213 
214 static inline void fixup_tlbie_va_range(unsigned long va, unsigned long pid,
215 					unsigned long ap)
216 {
217 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
218 		asm volatile("ptesync": : :"memory");
219 		__tlbie_pid(0, RIC_FLUSH_TLB);
220 	}
221 
222 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
223 		asm volatile("ptesync": : :"memory");
224 		__tlbie_va(va, pid, ap, RIC_FLUSH_TLB);
225 	}
226 }
227 
228 static inline void fixup_tlbie_pid(unsigned long pid)
229 {
230 	/*
231 	 * We can use any address for the invalidation, pick one which is
232 	 * probably unused as an optimisation.
233 	 */
234 	unsigned long va = ((1UL << 52) - 1);
235 
236 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
237 		asm volatile("ptesync": : :"memory");
238 		__tlbie_pid(0, RIC_FLUSH_TLB);
239 	}
240 
241 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
242 		asm volatile("ptesync": : :"memory");
243 		__tlbie_va(va, pid, mmu_get_ap(MMU_PAGE_64K), RIC_FLUSH_TLB);
244 	}
245 }
246 
247 
248 static inline void fixup_tlbie_lpid_va(unsigned long va, unsigned long lpid,
249 				       unsigned long ap)
250 {
251 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
252 		asm volatile("ptesync": : :"memory");
253 		__tlbie_lpid_va(va, 0, ap, RIC_FLUSH_TLB);
254 	}
255 
256 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
257 		asm volatile("ptesync": : :"memory");
258 		__tlbie_lpid_va(va, lpid, ap, RIC_FLUSH_TLB);
259 	}
260 }
261 
262 static inline void fixup_tlbie_lpid(unsigned long lpid)
263 {
264 	/*
265 	 * We can use any address for the invalidation, pick one which is
266 	 * probably unused as an optimisation.
267 	 */
268 	unsigned long va = ((1UL << 52) - 1);
269 
270 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
271 		asm volatile("ptesync": : :"memory");
272 		__tlbie_lpid(0, RIC_FLUSH_TLB);
273 	}
274 
275 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
276 		asm volatile("ptesync": : :"memory");
277 		__tlbie_lpid_va(va, lpid, mmu_get_ap(MMU_PAGE_64K), RIC_FLUSH_TLB);
278 	}
279 }
280 
281 /*
282  * We use 128 set in radix mode and 256 set in hpt mode.
283  */
284 static __always_inline void _tlbiel_pid(unsigned long pid, unsigned long ric)
285 {
286 	int set;
287 
288 	asm volatile("ptesync": : :"memory");
289 
290 	/*
291 	 * Flush the first set of the TLB, and if we're doing a RIC_FLUSH_ALL,
292 	 * also flush the entire Page Walk Cache.
293 	 */
294 	__tlbiel_pid(pid, 0, ric);
295 
296 	/* For PWC, only one flush is needed */
297 	if (ric == RIC_FLUSH_PWC) {
298 		asm volatile("ptesync": : :"memory");
299 		return;
300 	}
301 
302 	/* For the remaining sets, just flush the TLB */
303 	for (set = 1; set < POWER9_TLB_SETS_RADIX ; set++)
304 		__tlbiel_pid(pid, set, RIC_FLUSH_TLB);
305 
306 	asm volatile("ptesync": : :"memory");
307 	asm volatile(PPC_RADIX_INVALIDATE_ERAT_USER "; isync" : : :"memory");
308 }
309 
310 static inline void _tlbie_pid(unsigned long pid, unsigned long ric)
311 {
312 	asm volatile("ptesync": : :"memory");
313 
314 	/*
315 	 * Workaround the fact that the "ric" argument to __tlbie_pid
316 	 * must be a compile-time contraint to match the "i" constraint
317 	 * in the asm statement.
318 	 */
319 	switch (ric) {
320 	case RIC_FLUSH_TLB:
321 		__tlbie_pid(pid, RIC_FLUSH_TLB);
322 		fixup_tlbie_pid(pid);
323 		break;
324 	case RIC_FLUSH_PWC:
325 		__tlbie_pid(pid, RIC_FLUSH_PWC);
326 		break;
327 	case RIC_FLUSH_ALL:
328 	default:
329 		__tlbie_pid(pid, RIC_FLUSH_ALL);
330 		fixup_tlbie_pid(pid);
331 	}
332 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
333 }
334 
335 struct tlbiel_pid {
336 	unsigned long pid;
337 	unsigned long ric;
338 };
339 
340 static void do_tlbiel_pid(void *info)
341 {
342 	struct tlbiel_pid *t = info;
343 
344 	if (t->ric == RIC_FLUSH_TLB)
345 		_tlbiel_pid(t->pid, RIC_FLUSH_TLB);
346 	else if (t->ric == RIC_FLUSH_PWC)
347 		_tlbiel_pid(t->pid, RIC_FLUSH_PWC);
348 	else
349 		_tlbiel_pid(t->pid, RIC_FLUSH_ALL);
350 }
351 
352 static inline void _tlbiel_pid_multicast(struct mm_struct *mm,
353 				unsigned long pid, unsigned long ric)
354 {
355 	struct cpumask *cpus = mm_cpumask(mm);
356 	struct tlbiel_pid t = { .pid = pid, .ric = ric };
357 
358 	on_each_cpu_mask(cpus, do_tlbiel_pid, &t, 1);
359 	/*
360 	 * Always want the CPU translations to be invalidated with tlbiel in
361 	 * these paths, so while coprocessors must use tlbie, we can not
362 	 * optimise away the tlbiel component.
363 	 */
364 	if (atomic_read(&mm->context.copros) > 0)
365 		_tlbie_pid(pid, RIC_FLUSH_ALL);
366 }
367 
368 static inline void _tlbie_lpid(unsigned long lpid, unsigned long ric)
369 {
370 	asm volatile("ptesync": : :"memory");
371 
372 	/*
373 	 * Workaround the fact that the "ric" argument to __tlbie_pid
374 	 * must be a compile-time contraint to match the "i" constraint
375 	 * in the asm statement.
376 	 */
377 	switch (ric) {
378 	case RIC_FLUSH_TLB:
379 		__tlbie_lpid(lpid, RIC_FLUSH_TLB);
380 		fixup_tlbie_lpid(lpid);
381 		break;
382 	case RIC_FLUSH_PWC:
383 		__tlbie_lpid(lpid, RIC_FLUSH_PWC);
384 		break;
385 	case RIC_FLUSH_ALL:
386 	default:
387 		__tlbie_lpid(lpid, RIC_FLUSH_ALL);
388 		fixup_tlbie_lpid(lpid);
389 	}
390 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
391 }
392 
393 static __always_inline void _tlbie_lpid_guest(unsigned long lpid, unsigned long ric)
394 {
395 	/*
396 	 * Workaround the fact that the "ric" argument to __tlbie_pid
397 	 * must be a compile-time contraint to match the "i" constraint
398 	 * in the asm statement.
399 	 */
400 	switch (ric) {
401 	case RIC_FLUSH_TLB:
402 		__tlbie_lpid_guest(lpid, RIC_FLUSH_TLB);
403 		break;
404 	case RIC_FLUSH_PWC:
405 		__tlbie_lpid_guest(lpid, RIC_FLUSH_PWC);
406 		break;
407 	case RIC_FLUSH_ALL:
408 	default:
409 		__tlbie_lpid_guest(lpid, RIC_FLUSH_ALL);
410 	}
411 	fixup_tlbie_lpid(lpid);
412 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
413 }
414 
415 static inline void __tlbiel_va_range(unsigned long start, unsigned long end,
416 				    unsigned long pid, unsigned long page_size,
417 				    unsigned long psize)
418 {
419 	unsigned long addr;
420 	unsigned long ap = mmu_get_ap(psize);
421 
422 	for (addr = start; addr < end; addr += page_size)
423 		__tlbiel_va(addr, pid, ap, RIC_FLUSH_TLB);
424 }
425 
426 static __always_inline void _tlbiel_va(unsigned long va, unsigned long pid,
427 				       unsigned long psize, unsigned long ric)
428 {
429 	unsigned long ap = mmu_get_ap(psize);
430 
431 	asm volatile("ptesync": : :"memory");
432 	__tlbiel_va(va, pid, ap, ric);
433 	asm volatile("ptesync": : :"memory");
434 }
435 
436 static inline void _tlbiel_va_range(unsigned long start, unsigned long end,
437 				    unsigned long pid, unsigned long page_size,
438 				    unsigned long psize, bool also_pwc)
439 {
440 	asm volatile("ptesync": : :"memory");
441 	if (also_pwc)
442 		__tlbiel_pid(pid, 0, RIC_FLUSH_PWC);
443 	__tlbiel_va_range(start, end, pid, page_size, psize);
444 	asm volatile("ptesync": : :"memory");
445 }
446 
447 static inline void __tlbie_va_range(unsigned long start, unsigned long end,
448 				    unsigned long pid, unsigned long page_size,
449 				    unsigned long psize)
450 {
451 	unsigned long addr;
452 	unsigned long ap = mmu_get_ap(psize);
453 
454 	for (addr = start; addr < end; addr += page_size)
455 		__tlbie_va(addr, pid, ap, RIC_FLUSH_TLB);
456 
457 	fixup_tlbie_va_range(addr - page_size, pid, ap);
458 }
459 
460 static __always_inline void _tlbie_va(unsigned long va, unsigned long pid,
461 				      unsigned long psize, unsigned long ric)
462 {
463 	unsigned long ap = mmu_get_ap(psize);
464 
465 	asm volatile("ptesync": : :"memory");
466 	__tlbie_va(va, pid, ap, ric);
467 	fixup_tlbie_va(va, pid, ap);
468 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
469 }
470 
471 struct tlbiel_va {
472 	unsigned long pid;
473 	unsigned long va;
474 	unsigned long psize;
475 	unsigned long ric;
476 };
477 
478 static void do_tlbiel_va(void *info)
479 {
480 	struct tlbiel_va *t = info;
481 
482 	if (t->ric == RIC_FLUSH_TLB)
483 		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_TLB);
484 	else if (t->ric == RIC_FLUSH_PWC)
485 		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_PWC);
486 	else
487 		_tlbiel_va(t->va, t->pid, t->psize, RIC_FLUSH_ALL);
488 }
489 
490 static inline void _tlbiel_va_multicast(struct mm_struct *mm,
491 				unsigned long va, unsigned long pid,
492 				unsigned long psize, unsigned long ric)
493 {
494 	struct cpumask *cpus = mm_cpumask(mm);
495 	struct tlbiel_va t = { .va = va, .pid = pid, .psize = psize, .ric = ric };
496 	on_each_cpu_mask(cpus, do_tlbiel_va, &t, 1);
497 	if (atomic_read(&mm->context.copros) > 0)
498 		_tlbie_va(va, pid, psize, RIC_FLUSH_TLB);
499 }
500 
501 struct tlbiel_va_range {
502 	unsigned long pid;
503 	unsigned long start;
504 	unsigned long end;
505 	unsigned long page_size;
506 	unsigned long psize;
507 	bool also_pwc;
508 };
509 
510 static void do_tlbiel_va_range(void *info)
511 {
512 	struct tlbiel_va_range *t = info;
513 
514 	_tlbiel_va_range(t->start, t->end, t->pid, t->page_size,
515 				    t->psize, t->also_pwc);
516 }
517 
518 static __always_inline void _tlbie_lpid_va(unsigned long va, unsigned long lpid,
519 			      unsigned long psize, unsigned long ric)
520 {
521 	unsigned long ap = mmu_get_ap(psize);
522 
523 	asm volatile("ptesync": : :"memory");
524 	__tlbie_lpid_va(va, lpid, ap, ric);
525 	fixup_tlbie_lpid_va(va, lpid, ap);
526 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
527 }
528 
529 static inline void _tlbie_va_range(unsigned long start, unsigned long end,
530 				    unsigned long pid, unsigned long page_size,
531 				    unsigned long psize, bool also_pwc)
532 {
533 	asm volatile("ptesync": : :"memory");
534 	if (also_pwc)
535 		__tlbie_pid(pid, RIC_FLUSH_PWC);
536 	__tlbie_va_range(start, end, pid, page_size, psize);
537 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
538 }
539 
540 static inline void _tlbiel_va_range_multicast(struct mm_struct *mm,
541 				unsigned long start, unsigned long end,
542 				unsigned long pid, unsigned long page_size,
543 				unsigned long psize, bool also_pwc)
544 {
545 	struct cpumask *cpus = mm_cpumask(mm);
546 	struct tlbiel_va_range t = { .start = start, .end = end,
547 				.pid = pid, .page_size = page_size,
548 				.psize = psize, .also_pwc = also_pwc };
549 
550 	on_each_cpu_mask(cpus, do_tlbiel_va_range, &t, 1);
551 	if (atomic_read(&mm->context.copros) > 0)
552 		_tlbie_va_range(start, end, pid, page_size, psize, also_pwc);
553 }
554 
555 /*
556  * Base TLB flushing operations:
557  *
558  *  - flush_tlb_mm(mm) flushes the specified mm context TLB's
559  *  - flush_tlb_page(vma, vmaddr) flushes one page
560  *  - flush_tlb_range(vma, start, end) flushes a range of pages
561  *  - flush_tlb_kernel_range(start, end) flushes kernel pages
562  *
563  *  - local_* variants of page and mm only apply to the current
564  *    processor
565  */
566 void radix__local_flush_tlb_mm(struct mm_struct *mm)
567 {
568 	unsigned long pid;
569 
570 	preempt_disable();
571 	pid = mm->context.id;
572 	if (pid != MMU_NO_CONTEXT)
573 		_tlbiel_pid(pid, RIC_FLUSH_TLB);
574 	preempt_enable();
575 }
576 EXPORT_SYMBOL(radix__local_flush_tlb_mm);
577 
578 #ifndef CONFIG_SMP
579 void radix__local_flush_all_mm(struct mm_struct *mm)
580 {
581 	unsigned long pid;
582 
583 	preempt_disable();
584 	pid = mm->context.id;
585 	if (pid != MMU_NO_CONTEXT)
586 		_tlbiel_pid(pid, RIC_FLUSH_ALL);
587 	preempt_enable();
588 }
589 EXPORT_SYMBOL(radix__local_flush_all_mm);
590 
591 static void __flush_all_mm(struct mm_struct *mm, bool fullmm)
592 {
593 	radix__local_flush_all_mm(mm);
594 }
595 #endif /* CONFIG_SMP */
596 
597 void radix__local_flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
598 				       int psize)
599 {
600 	unsigned long pid;
601 
602 	preempt_disable();
603 	pid = mm->context.id;
604 	if (pid != MMU_NO_CONTEXT)
605 		_tlbiel_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
606 	preempt_enable();
607 }
608 
609 void radix__local_flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
610 {
611 #ifdef CONFIG_HUGETLB_PAGE
612 	/* need the return fix for nohash.c */
613 	if (is_vm_hugetlb_page(vma))
614 		return radix__local_flush_hugetlb_page(vma, vmaddr);
615 #endif
616 	radix__local_flush_tlb_page_psize(vma->vm_mm, vmaddr, mmu_virtual_psize);
617 }
618 EXPORT_SYMBOL(radix__local_flush_tlb_page);
619 
620 static bool mm_is_singlethreaded(struct mm_struct *mm)
621 {
622 	if (atomic_read(&mm->context.copros) > 0)
623 		return false;
624 	if (atomic_read(&mm->mm_users) <= 1 && current->mm == mm)
625 		return true;
626 	return false;
627 }
628 
629 static bool mm_needs_flush_escalation(struct mm_struct *mm)
630 {
631 	/*
632 	 * P9 nest MMU has issues with the page walk cache
633 	 * caching PTEs and not flushing them properly when
634 	 * RIC = 0 for a PID/LPID invalidate
635 	 */
636 	if (atomic_read(&mm->context.copros) > 0)
637 		return true;
638 	return false;
639 }
640 
641 #ifdef CONFIG_SMP
642 static void do_exit_flush_lazy_tlb(void *arg)
643 {
644 	struct mm_struct *mm = arg;
645 	unsigned long pid = mm->context.id;
646 
647 	if (current->mm == mm)
648 		return; /* Local CPU */
649 
650 	if (current->active_mm == mm) {
651 		/*
652 		 * Must be a kernel thread because sender is single-threaded.
653 		 */
654 		BUG_ON(current->mm);
655 		mmgrab(&init_mm);
656 		switch_mm(mm, &init_mm, current);
657 		current->active_mm = &init_mm;
658 		mmdrop(mm);
659 	}
660 	_tlbiel_pid(pid, RIC_FLUSH_ALL);
661 }
662 
663 static void exit_flush_lazy_tlbs(struct mm_struct *mm)
664 {
665 	/*
666 	 * Would be nice if this was async so it could be run in
667 	 * parallel with our local flush, but generic code does not
668 	 * give a good API for it. Could extend the generic code or
669 	 * make a special powerpc IPI for flushing TLBs.
670 	 * For now it's not too performance critical.
671 	 */
672 	smp_call_function_many(mm_cpumask(mm), do_exit_flush_lazy_tlb,
673 				(void *)mm, 1);
674 	mm_reset_thread_local(mm);
675 }
676 
677 void radix__flush_tlb_mm(struct mm_struct *mm)
678 {
679 	unsigned long pid;
680 
681 	pid = mm->context.id;
682 	if (unlikely(pid == MMU_NO_CONTEXT))
683 		return;
684 
685 	preempt_disable();
686 	/*
687 	 * Order loads of mm_cpumask vs previous stores to clear ptes before
688 	 * the invalidate. See barrier in switch_mm_irqs_off
689 	 */
690 	smp_mb();
691 	if (!mm_is_thread_local(mm)) {
692 		if (unlikely(mm_is_singlethreaded(mm))) {
693 			exit_flush_lazy_tlbs(mm);
694 			goto local;
695 		}
696 
697 		if (cputlb_use_tlbie()) {
698 			if (mm_needs_flush_escalation(mm))
699 				_tlbie_pid(pid, RIC_FLUSH_ALL);
700 			else
701 				_tlbie_pid(pid, RIC_FLUSH_TLB);
702 		} else {
703 			_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_TLB);
704 		}
705 	} else {
706 local:
707 		_tlbiel_pid(pid, RIC_FLUSH_TLB);
708 	}
709 	preempt_enable();
710 }
711 EXPORT_SYMBOL(radix__flush_tlb_mm);
712 
713 static void __flush_all_mm(struct mm_struct *mm, bool fullmm)
714 {
715 	unsigned long pid;
716 
717 	pid = mm->context.id;
718 	if (unlikely(pid == MMU_NO_CONTEXT))
719 		return;
720 
721 	preempt_disable();
722 	smp_mb(); /* see radix__flush_tlb_mm */
723 	if (!mm_is_thread_local(mm)) {
724 		if (unlikely(mm_is_singlethreaded(mm))) {
725 			if (!fullmm) {
726 				exit_flush_lazy_tlbs(mm);
727 				goto local;
728 			}
729 		}
730 		if (cputlb_use_tlbie())
731 			_tlbie_pid(pid, RIC_FLUSH_ALL);
732 		else
733 			_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_ALL);
734 	} else {
735 local:
736 		_tlbiel_pid(pid, RIC_FLUSH_ALL);
737 	}
738 	preempt_enable();
739 }
740 
741 void radix__flush_all_mm(struct mm_struct *mm)
742 {
743 	__flush_all_mm(mm, false);
744 }
745 EXPORT_SYMBOL(radix__flush_all_mm);
746 
747 void radix__flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
748 				 int psize)
749 {
750 	unsigned long pid;
751 
752 	pid = mm->context.id;
753 	if (unlikely(pid == MMU_NO_CONTEXT))
754 		return;
755 
756 	preempt_disable();
757 	smp_mb(); /* see radix__flush_tlb_mm */
758 	if (!mm_is_thread_local(mm)) {
759 		if (unlikely(mm_is_singlethreaded(mm))) {
760 			exit_flush_lazy_tlbs(mm);
761 			goto local;
762 		}
763 		if (cputlb_use_tlbie())
764 			_tlbie_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
765 		else
766 			_tlbiel_va_multicast(mm, vmaddr, pid, psize, RIC_FLUSH_TLB);
767 	} else {
768 local:
769 		_tlbiel_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
770 	}
771 	preempt_enable();
772 }
773 
774 void radix__flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
775 {
776 #ifdef CONFIG_HUGETLB_PAGE
777 	if (is_vm_hugetlb_page(vma))
778 		return radix__flush_hugetlb_page(vma, vmaddr);
779 #endif
780 	radix__flush_tlb_page_psize(vma->vm_mm, vmaddr, mmu_virtual_psize);
781 }
782 EXPORT_SYMBOL(radix__flush_tlb_page);
783 
784 #else /* CONFIG_SMP */
785 static inline void exit_flush_lazy_tlbs(struct mm_struct *mm) { }
786 #endif /* CONFIG_SMP */
787 
788 static void do_tlbiel_kernel(void *info)
789 {
790 	_tlbiel_pid(0, RIC_FLUSH_ALL);
791 }
792 
793 static inline void _tlbiel_kernel_broadcast(void)
794 {
795 	on_each_cpu(do_tlbiel_kernel, NULL, 1);
796 	if (tlbie_capable) {
797 		/*
798 		 * Coherent accelerators don't refcount kernel memory mappings,
799 		 * so have to always issue a tlbie for them. This is quite a
800 		 * slow path anyway.
801 		 */
802 		_tlbie_pid(0, RIC_FLUSH_ALL);
803 	}
804 }
805 
806 /*
807  * If kernel TLBIs ever become local rather than global, then
808  * drivers/misc/ocxl/link.c:ocxl_link_add_pe will need some work, as it
809  * assumes kernel TLBIs are global.
810  */
811 void radix__flush_tlb_kernel_range(unsigned long start, unsigned long end)
812 {
813 	if (cputlb_use_tlbie())
814 		_tlbie_pid(0, RIC_FLUSH_ALL);
815 	else
816 		_tlbiel_kernel_broadcast();
817 }
818 EXPORT_SYMBOL(radix__flush_tlb_kernel_range);
819 
820 #define TLB_FLUSH_ALL -1UL
821 
822 /*
823  * Number of pages above which we invalidate the entire PID rather than
824  * flush individual pages, for local and global flushes respectively.
825  *
826  * tlbie goes out to the interconnect and individual ops are more costly.
827  * It also does not iterate over sets like the local tlbiel variant when
828  * invalidating a full PID, so it has a far lower threshold to change from
829  * individual page flushes to full-pid flushes.
830  */
831 static unsigned long tlb_single_page_flush_ceiling __read_mostly = 33;
832 static unsigned long tlb_local_single_page_flush_ceiling __read_mostly = POWER9_TLB_SETS_RADIX * 2;
833 
834 static inline void __radix__flush_tlb_range(struct mm_struct *mm,
835 					    unsigned long start, unsigned long end)
836 
837 {
838 	unsigned long pid;
839 	unsigned int page_shift = mmu_psize_defs[mmu_virtual_psize].shift;
840 	unsigned long page_size = 1UL << page_shift;
841 	unsigned long nr_pages = (end - start) >> page_shift;
842 	bool local, full;
843 
844 	pid = mm->context.id;
845 	if (unlikely(pid == MMU_NO_CONTEXT))
846 		return;
847 
848 	preempt_disable();
849 	smp_mb(); /* see radix__flush_tlb_mm */
850 	if (!mm_is_thread_local(mm)) {
851 		if (unlikely(mm_is_singlethreaded(mm))) {
852 			if (end != TLB_FLUSH_ALL) {
853 				exit_flush_lazy_tlbs(mm);
854 				goto is_local;
855 			}
856 		}
857 		local = false;
858 		full = (end == TLB_FLUSH_ALL ||
859 				nr_pages > tlb_single_page_flush_ceiling);
860 	} else {
861 is_local:
862 		local = true;
863 		full = (end == TLB_FLUSH_ALL ||
864 				nr_pages > tlb_local_single_page_flush_ceiling);
865 	}
866 
867 	if (full) {
868 		if (local) {
869 			_tlbiel_pid(pid, RIC_FLUSH_TLB);
870 		} else {
871 			if (cputlb_use_tlbie()) {
872 				if (mm_needs_flush_escalation(mm))
873 					_tlbie_pid(pid, RIC_FLUSH_ALL);
874 				else
875 					_tlbie_pid(pid, RIC_FLUSH_TLB);
876 			} else {
877 				_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_TLB);
878 			}
879 		}
880 	} else {
881 		bool hflush = false;
882 		unsigned long hstart, hend;
883 
884 		if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
885 			hstart = (start + PMD_SIZE - 1) & PMD_MASK;
886 			hend = end & PMD_MASK;
887 			if (hstart < hend)
888 				hflush = true;
889 		}
890 
891 		if (local) {
892 			asm volatile("ptesync": : :"memory");
893 			__tlbiel_va_range(start, end, pid, page_size, mmu_virtual_psize);
894 			if (hflush)
895 				__tlbiel_va_range(hstart, hend, pid,
896 						PMD_SIZE, MMU_PAGE_2M);
897 			asm volatile("ptesync": : :"memory");
898 		} else if (cputlb_use_tlbie()) {
899 			asm volatile("ptesync": : :"memory");
900 			__tlbie_va_range(start, end, pid, page_size, mmu_virtual_psize);
901 			if (hflush)
902 				__tlbie_va_range(hstart, hend, pid,
903 						PMD_SIZE, MMU_PAGE_2M);
904 			asm volatile("eieio; tlbsync; ptesync": : :"memory");
905 		} else {
906 			_tlbiel_va_range_multicast(mm,
907 					start, end, pid, page_size, mmu_virtual_psize, false);
908 			if (hflush)
909 				_tlbiel_va_range_multicast(mm,
910 					hstart, hend, pid, PMD_SIZE, MMU_PAGE_2M, false);
911 		}
912 	}
913 	preempt_enable();
914 }
915 
916 void radix__flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
917 		     unsigned long end)
918 
919 {
920 #ifdef CONFIG_HUGETLB_PAGE
921 	if (is_vm_hugetlb_page(vma))
922 		return radix__flush_hugetlb_tlb_range(vma, start, end);
923 #endif
924 
925 	__radix__flush_tlb_range(vma->vm_mm, start, end);
926 }
927 EXPORT_SYMBOL(radix__flush_tlb_range);
928 
929 static int radix_get_mmu_psize(int page_size)
930 {
931 	int psize;
932 
933 	if (page_size == (1UL << mmu_psize_defs[mmu_virtual_psize].shift))
934 		psize = mmu_virtual_psize;
935 	else if (page_size == (1UL << mmu_psize_defs[MMU_PAGE_2M].shift))
936 		psize = MMU_PAGE_2M;
937 	else if (page_size == (1UL << mmu_psize_defs[MMU_PAGE_1G].shift))
938 		psize = MMU_PAGE_1G;
939 	else
940 		return -1;
941 	return psize;
942 }
943 
944 /*
945  * Flush partition scoped LPID address translation for all CPUs.
946  */
947 void radix__flush_tlb_lpid_page(unsigned int lpid,
948 					unsigned long addr,
949 					unsigned long page_size)
950 {
951 	int psize = radix_get_mmu_psize(page_size);
952 
953 	_tlbie_lpid_va(addr, lpid, psize, RIC_FLUSH_TLB);
954 }
955 EXPORT_SYMBOL_GPL(radix__flush_tlb_lpid_page);
956 
957 /*
958  * Flush partition scoped PWC from LPID for all CPUs.
959  */
960 void radix__flush_pwc_lpid(unsigned int lpid)
961 {
962 	_tlbie_lpid(lpid, RIC_FLUSH_PWC);
963 }
964 EXPORT_SYMBOL_GPL(radix__flush_pwc_lpid);
965 
966 /*
967  * Flush partition scoped translations from LPID (=LPIDR)
968  */
969 void radix__flush_all_lpid(unsigned int lpid)
970 {
971 	_tlbie_lpid(lpid, RIC_FLUSH_ALL);
972 }
973 EXPORT_SYMBOL_GPL(radix__flush_all_lpid);
974 
975 /*
976  * Flush process scoped translations from LPID (=LPIDR)
977  */
978 void radix__flush_all_lpid_guest(unsigned int lpid)
979 {
980 	_tlbie_lpid_guest(lpid, RIC_FLUSH_ALL);
981 }
982 
983 static void radix__flush_tlb_pwc_range_psize(struct mm_struct *mm, unsigned long start,
984 				  unsigned long end, int psize);
985 
986 void radix__tlb_flush(struct mmu_gather *tlb)
987 {
988 	int psize = 0;
989 	struct mm_struct *mm = tlb->mm;
990 	int page_size = tlb->page_size;
991 	unsigned long start = tlb->start;
992 	unsigned long end = tlb->end;
993 
994 	/*
995 	 * if page size is not something we understand, do a full mm flush
996 	 *
997 	 * A "fullmm" flush must always do a flush_all_mm (RIC=2) flush
998 	 * that flushes the process table entry cache upon process teardown.
999 	 * See the comment for radix in arch_exit_mmap().
1000 	 */
1001 	if (tlb->fullmm || tlb->need_flush_all) {
1002 		__flush_all_mm(mm, true);
1003 	} else if ( (psize = radix_get_mmu_psize(page_size)) == -1) {
1004 		if (!tlb->freed_tables)
1005 			radix__flush_tlb_mm(mm);
1006 		else
1007 			radix__flush_all_mm(mm);
1008 	} else {
1009 		if (!tlb->freed_tables)
1010 			radix__flush_tlb_range_psize(mm, start, end, psize);
1011 		else
1012 			radix__flush_tlb_pwc_range_psize(mm, start, end, psize);
1013 	}
1014 }
1015 
1016 static __always_inline void __radix__flush_tlb_range_psize(struct mm_struct *mm,
1017 				unsigned long start, unsigned long end,
1018 				int psize, bool also_pwc)
1019 {
1020 	unsigned long pid;
1021 	unsigned int page_shift = mmu_psize_defs[psize].shift;
1022 	unsigned long page_size = 1UL << page_shift;
1023 	unsigned long nr_pages = (end - start) >> page_shift;
1024 	bool local, full;
1025 
1026 	pid = mm->context.id;
1027 	if (unlikely(pid == MMU_NO_CONTEXT))
1028 		return;
1029 
1030 	preempt_disable();
1031 	smp_mb(); /* see radix__flush_tlb_mm */
1032 	if (!mm_is_thread_local(mm)) {
1033 		if (unlikely(mm_is_singlethreaded(mm))) {
1034 			if (end != TLB_FLUSH_ALL) {
1035 				exit_flush_lazy_tlbs(mm);
1036 				goto is_local;
1037 			}
1038 		}
1039 		local = false;
1040 		full = (end == TLB_FLUSH_ALL ||
1041 				nr_pages > tlb_single_page_flush_ceiling);
1042 	} else {
1043 is_local:
1044 		local = true;
1045 		full = (end == TLB_FLUSH_ALL ||
1046 				nr_pages > tlb_local_single_page_flush_ceiling);
1047 	}
1048 
1049 	if (full) {
1050 		if (local) {
1051 			_tlbiel_pid(pid, also_pwc ? RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1052 		} else {
1053 			if (cputlb_use_tlbie()) {
1054 				if (mm_needs_flush_escalation(mm))
1055 					also_pwc = true;
1056 
1057 				_tlbie_pid(pid,
1058 					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1059 			} else {
1060 				_tlbiel_pid_multicast(mm, pid,
1061 					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1062 			}
1063 
1064 		}
1065 	} else {
1066 		if (local)
1067 			_tlbiel_va_range(start, end, pid, page_size, psize, also_pwc);
1068 		else if (cputlb_use_tlbie())
1069 			_tlbie_va_range(start, end, pid, page_size, psize, also_pwc);
1070 		else
1071 			_tlbiel_va_range_multicast(mm,
1072 					start, end, pid, page_size, psize, also_pwc);
1073 	}
1074 	preempt_enable();
1075 }
1076 
1077 void radix__flush_tlb_range_psize(struct mm_struct *mm, unsigned long start,
1078 				  unsigned long end, int psize)
1079 {
1080 	return __radix__flush_tlb_range_psize(mm, start, end, psize, false);
1081 }
1082 
1083 static void radix__flush_tlb_pwc_range_psize(struct mm_struct *mm, unsigned long start,
1084 				  unsigned long end, int psize)
1085 {
1086 	__radix__flush_tlb_range_psize(mm, start, end, psize, true);
1087 }
1088 
1089 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1090 void radix__flush_tlb_collapsed_pmd(struct mm_struct *mm, unsigned long addr)
1091 {
1092 	unsigned long pid, end;
1093 
1094 	pid = mm->context.id;
1095 	if (unlikely(pid == MMU_NO_CONTEXT))
1096 		return;
1097 
1098 	/* 4k page size, just blow the world */
1099 	if (PAGE_SIZE == 0x1000) {
1100 		radix__flush_all_mm(mm);
1101 		return;
1102 	}
1103 
1104 	end = addr + HPAGE_PMD_SIZE;
1105 
1106 	/* Otherwise first do the PWC, then iterate the pages. */
1107 	preempt_disable();
1108 	smp_mb(); /* see radix__flush_tlb_mm */
1109 	if (!mm_is_thread_local(mm)) {
1110 		if (unlikely(mm_is_singlethreaded(mm))) {
1111 			exit_flush_lazy_tlbs(mm);
1112 			goto local;
1113 		}
1114 		if (cputlb_use_tlbie())
1115 			_tlbie_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1116 		else
1117 			_tlbiel_va_range_multicast(mm,
1118 					addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1119 	} else {
1120 local:
1121 		_tlbiel_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1122 	}
1123 
1124 	preempt_enable();
1125 }
1126 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1127 
1128 void radix__flush_pmd_tlb_range(struct vm_area_struct *vma,
1129 				unsigned long start, unsigned long end)
1130 {
1131 	radix__flush_tlb_range_psize(vma->vm_mm, start, end, MMU_PAGE_2M);
1132 }
1133 EXPORT_SYMBOL(radix__flush_pmd_tlb_range);
1134 
1135 void radix__flush_tlb_all(void)
1136 {
1137 	unsigned long rb,prs,r,rs;
1138 	unsigned long ric = RIC_FLUSH_ALL;
1139 
1140 	rb = 0x3 << PPC_BITLSHIFT(53); /* IS = 3 */
1141 	prs = 0; /* partition scoped */
1142 	r = 1;   /* radix format */
1143 	rs = 1 & ((1UL << 32) - 1); /* any LPID value to flush guest mappings */
1144 
1145 	asm volatile("ptesync": : :"memory");
1146 	/*
1147 	 * now flush guest entries by passing PRS = 1 and LPID != 0
1148 	 */
1149 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
1150 		     : : "r"(rb), "i"(r), "i"(1), "i"(ric), "r"(rs) : "memory");
1151 	/*
1152 	 * now flush host entires by passing PRS = 0 and LPID == 0
1153 	 */
1154 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
1155 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(0) : "memory");
1156 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
1157 }
1158 
1159 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1160 extern void radix_kvm_prefetch_workaround(struct mm_struct *mm)
1161 {
1162 	unsigned long pid = mm->context.id;
1163 
1164 	if (unlikely(pid == MMU_NO_CONTEXT))
1165 		return;
1166 
1167 	if (!cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
1168 		return;
1169 
1170 	/*
1171 	 * If this context hasn't run on that CPU before and KVM is
1172 	 * around, there's a slim chance that the guest on another
1173 	 * CPU just brought in obsolete translation into the TLB of
1174 	 * this CPU due to a bad prefetch using the guest PID on
1175 	 * the way into the hypervisor.
1176 	 *
1177 	 * We work around this here. If KVM is possible, we check if
1178 	 * any sibling thread is in KVM. If it is, the window may exist
1179 	 * and thus we flush that PID from the core.
1180 	 *
1181 	 * A potential future improvement would be to mark which PIDs
1182 	 * have never been used on the system and avoid it if the PID
1183 	 * is new and the process has no other cpumask bit set.
1184 	 */
1185 	if (cpu_has_feature(CPU_FTR_HVMODE) && radix_enabled()) {
1186 		int cpu = smp_processor_id();
1187 		int sib = cpu_first_thread_sibling(cpu);
1188 		bool flush = false;
1189 
1190 		for (; sib <= cpu_last_thread_sibling(cpu) && !flush; sib++) {
1191 			if (sib == cpu)
1192 				continue;
1193 			if (!cpu_possible(sib))
1194 				continue;
1195 			if (paca_ptrs[sib]->kvm_hstate.kvm_vcpu)
1196 				flush = true;
1197 		}
1198 		if (flush)
1199 			_tlbiel_pid(pid, RIC_FLUSH_ALL);
1200 	}
1201 }
1202 EXPORT_SYMBOL_GPL(radix_kvm_prefetch_workaround);
1203 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1204