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 #endif /* CONFIG_SMP */
591 
592 void radix__local_flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
593 				       int psize)
594 {
595 	unsigned long pid;
596 
597 	preempt_disable();
598 	pid = mm->context.id;
599 	if (pid != MMU_NO_CONTEXT)
600 		_tlbiel_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
601 	preempt_enable();
602 }
603 
604 void radix__local_flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
605 {
606 #ifdef CONFIG_HUGETLB_PAGE
607 	/* need the return fix for nohash.c */
608 	if (is_vm_hugetlb_page(vma))
609 		return radix__local_flush_hugetlb_page(vma, vmaddr);
610 #endif
611 	radix__local_flush_tlb_page_psize(vma->vm_mm, vmaddr, mmu_virtual_psize);
612 }
613 EXPORT_SYMBOL(radix__local_flush_tlb_page);
614 
615 static bool mm_is_singlethreaded(struct mm_struct *mm)
616 {
617 	if (atomic_read(&mm->context.copros) > 0)
618 		return false;
619 	if (atomic_read(&mm->mm_users) <= 1 && current->mm == mm)
620 		return true;
621 	return false;
622 }
623 
624 static bool mm_needs_flush_escalation(struct mm_struct *mm)
625 {
626 	/*
627 	 * P9 nest MMU has issues with the page walk cache
628 	 * caching PTEs and not flushing them properly when
629 	 * RIC = 0 for a PID/LPID invalidate
630 	 */
631 	if (atomic_read(&mm->context.copros) > 0)
632 		return true;
633 	return false;
634 }
635 
636 #ifdef CONFIG_SMP
637 static void do_exit_flush_lazy_tlb(void *arg)
638 {
639 	struct mm_struct *mm = arg;
640 	unsigned long pid = mm->context.id;
641 
642 	if (current->mm == mm)
643 		return; /* Local CPU */
644 
645 	if (current->active_mm == mm) {
646 		/*
647 		 * Must be a kernel thread because sender is single-threaded.
648 		 */
649 		BUG_ON(current->mm);
650 		mmgrab(&init_mm);
651 		switch_mm(mm, &init_mm, current);
652 		current->active_mm = &init_mm;
653 		mmdrop(mm);
654 	}
655 	_tlbiel_pid(pid, RIC_FLUSH_ALL);
656 }
657 
658 static void exit_flush_lazy_tlbs(struct mm_struct *mm)
659 {
660 	/*
661 	 * Would be nice if this was async so it could be run in
662 	 * parallel with our local flush, but generic code does not
663 	 * give a good API for it. Could extend the generic code or
664 	 * make a special powerpc IPI for flushing TLBs.
665 	 * For now it's not too performance critical.
666 	 */
667 	smp_call_function_many(mm_cpumask(mm), do_exit_flush_lazy_tlb,
668 				(void *)mm, 1);
669 	mm_reset_thread_local(mm);
670 }
671 
672 void radix__flush_tlb_mm(struct mm_struct *mm)
673 {
674 	unsigned long pid;
675 
676 	pid = mm->context.id;
677 	if (unlikely(pid == MMU_NO_CONTEXT))
678 		return;
679 
680 	preempt_disable();
681 	/*
682 	 * Order loads of mm_cpumask vs previous stores to clear ptes before
683 	 * the invalidate. See barrier in switch_mm_irqs_off
684 	 */
685 	smp_mb();
686 	if (!mm_is_thread_local(mm)) {
687 		if (unlikely(mm_is_singlethreaded(mm))) {
688 			exit_flush_lazy_tlbs(mm);
689 			goto local;
690 		}
691 
692 		if (cputlb_use_tlbie()) {
693 			if (mm_needs_flush_escalation(mm))
694 				_tlbie_pid(pid, RIC_FLUSH_ALL);
695 			else
696 				_tlbie_pid(pid, RIC_FLUSH_TLB);
697 		} else {
698 			_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_TLB);
699 		}
700 	} else {
701 local:
702 		_tlbiel_pid(pid, RIC_FLUSH_TLB);
703 	}
704 	preempt_enable();
705 }
706 EXPORT_SYMBOL(radix__flush_tlb_mm);
707 
708 static void __flush_all_mm(struct mm_struct *mm, bool fullmm)
709 {
710 	unsigned long pid;
711 
712 	pid = mm->context.id;
713 	if (unlikely(pid == MMU_NO_CONTEXT))
714 		return;
715 
716 	preempt_disable();
717 	smp_mb(); /* see radix__flush_tlb_mm */
718 	if (!mm_is_thread_local(mm)) {
719 		if (unlikely(mm_is_singlethreaded(mm))) {
720 			if (!fullmm) {
721 				exit_flush_lazy_tlbs(mm);
722 				goto local;
723 			}
724 		}
725 		if (cputlb_use_tlbie())
726 			_tlbie_pid(pid, RIC_FLUSH_ALL);
727 		else
728 			_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_ALL);
729 	} else {
730 local:
731 		_tlbiel_pid(pid, RIC_FLUSH_ALL);
732 	}
733 	preempt_enable();
734 }
735 
736 void radix__flush_all_mm(struct mm_struct *mm)
737 {
738 	__flush_all_mm(mm, false);
739 }
740 EXPORT_SYMBOL(radix__flush_all_mm);
741 
742 void radix__flush_tlb_page_psize(struct mm_struct *mm, unsigned long vmaddr,
743 				 int psize)
744 {
745 	unsigned long pid;
746 
747 	pid = mm->context.id;
748 	if (unlikely(pid == MMU_NO_CONTEXT))
749 		return;
750 
751 	preempt_disable();
752 	smp_mb(); /* see radix__flush_tlb_mm */
753 	if (!mm_is_thread_local(mm)) {
754 		if (unlikely(mm_is_singlethreaded(mm))) {
755 			exit_flush_lazy_tlbs(mm);
756 			goto local;
757 		}
758 		if (cputlb_use_tlbie())
759 			_tlbie_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
760 		else
761 			_tlbiel_va_multicast(mm, vmaddr, pid, psize, RIC_FLUSH_TLB);
762 	} else {
763 local:
764 		_tlbiel_va(vmaddr, pid, psize, RIC_FLUSH_TLB);
765 	}
766 	preempt_enable();
767 }
768 
769 void radix__flush_tlb_page(struct vm_area_struct *vma, unsigned long vmaddr)
770 {
771 #ifdef CONFIG_HUGETLB_PAGE
772 	if (is_vm_hugetlb_page(vma))
773 		return radix__flush_hugetlb_page(vma, vmaddr);
774 #endif
775 	radix__flush_tlb_page_psize(vma->vm_mm, vmaddr, mmu_virtual_psize);
776 }
777 EXPORT_SYMBOL(radix__flush_tlb_page);
778 
779 #else /* CONFIG_SMP */
780 #define radix__flush_all_mm radix__local_flush_all_mm
781 #endif /* CONFIG_SMP */
782 
783 static void do_tlbiel_kernel(void *info)
784 {
785 	_tlbiel_pid(0, RIC_FLUSH_ALL);
786 }
787 
788 static inline void _tlbiel_kernel_broadcast(void)
789 {
790 	on_each_cpu(do_tlbiel_kernel, NULL, 1);
791 	if (tlbie_capable) {
792 		/*
793 		 * Coherent accelerators don't refcount kernel memory mappings,
794 		 * so have to always issue a tlbie for them. This is quite a
795 		 * slow path anyway.
796 		 */
797 		_tlbie_pid(0, RIC_FLUSH_ALL);
798 	}
799 }
800 
801 /*
802  * If kernel TLBIs ever become local rather than global, then
803  * drivers/misc/ocxl/link.c:ocxl_link_add_pe will need some work, as it
804  * assumes kernel TLBIs are global.
805  */
806 void radix__flush_tlb_kernel_range(unsigned long start, unsigned long end)
807 {
808 	if (cputlb_use_tlbie())
809 		_tlbie_pid(0, RIC_FLUSH_ALL);
810 	else
811 		_tlbiel_kernel_broadcast();
812 }
813 EXPORT_SYMBOL(radix__flush_tlb_kernel_range);
814 
815 #define TLB_FLUSH_ALL -1UL
816 
817 /*
818  * Number of pages above which we invalidate the entire PID rather than
819  * flush individual pages, for local and global flushes respectively.
820  *
821  * tlbie goes out to the interconnect and individual ops are more costly.
822  * It also does not iterate over sets like the local tlbiel variant when
823  * invalidating a full PID, so it has a far lower threshold to change from
824  * individual page flushes to full-pid flushes.
825  */
826 static unsigned long tlb_single_page_flush_ceiling __read_mostly = 33;
827 static unsigned long tlb_local_single_page_flush_ceiling __read_mostly = POWER9_TLB_SETS_RADIX * 2;
828 
829 static inline void __radix__flush_tlb_range(struct mm_struct *mm,
830 					    unsigned long start, unsigned long end)
831 
832 {
833 	unsigned long pid;
834 	unsigned int page_shift = mmu_psize_defs[mmu_virtual_psize].shift;
835 	unsigned long page_size = 1UL << page_shift;
836 	unsigned long nr_pages = (end - start) >> page_shift;
837 	bool local, full;
838 
839 	pid = mm->context.id;
840 	if (unlikely(pid == MMU_NO_CONTEXT))
841 		return;
842 
843 	preempt_disable();
844 	smp_mb(); /* see radix__flush_tlb_mm */
845 	if (!mm_is_thread_local(mm)) {
846 		if (unlikely(mm_is_singlethreaded(mm))) {
847 			if (end != TLB_FLUSH_ALL) {
848 				exit_flush_lazy_tlbs(mm);
849 				goto is_local;
850 			}
851 		}
852 		local = false;
853 		full = (end == TLB_FLUSH_ALL ||
854 				nr_pages > tlb_single_page_flush_ceiling);
855 	} else {
856 is_local:
857 		local = true;
858 		full = (end == TLB_FLUSH_ALL ||
859 				nr_pages > tlb_local_single_page_flush_ceiling);
860 	}
861 
862 	if (full) {
863 		if (local) {
864 			_tlbiel_pid(pid, RIC_FLUSH_TLB);
865 		} else {
866 			if (cputlb_use_tlbie()) {
867 				if (mm_needs_flush_escalation(mm))
868 					_tlbie_pid(pid, RIC_FLUSH_ALL);
869 				else
870 					_tlbie_pid(pid, RIC_FLUSH_TLB);
871 			} else {
872 				_tlbiel_pid_multicast(mm, pid, RIC_FLUSH_TLB);
873 			}
874 		}
875 	} else {
876 		bool hflush = false;
877 		unsigned long hstart, hend;
878 
879 		if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
880 			hstart = (start + PMD_SIZE - 1) & PMD_MASK;
881 			hend = end & PMD_MASK;
882 			if (hstart == hend)
883 				hflush = false;
884 			else
885 				hflush = true;
886 		}
887 
888 		if (local) {
889 			asm volatile("ptesync": : :"memory");
890 			__tlbiel_va_range(start, end, pid, page_size, mmu_virtual_psize);
891 			if (hflush)
892 				__tlbiel_va_range(hstart, hend, pid,
893 						PMD_SIZE, MMU_PAGE_2M);
894 			asm volatile("ptesync": : :"memory");
895 		} else if (cputlb_use_tlbie()) {
896 			asm volatile("ptesync": : :"memory");
897 			__tlbie_va_range(start, end, pid, page_size, mmu_virtual_psize);
898 			if (hflush)
899 				__tlbie_va_range(hstart, hend, pid,
900 						PMD_SIZE, MMU_PAGE_2M);
901 			asm volatile("eieio; tlbsync; ptesync": : :"memory");
902 		} else {
903 			_tlbiel_va_range_multicast(mm,
904 					start, end, pid, page_size, mmu_virtual_psize, false);
905 			if (hflush)
906 				_tlbiel_va_range_multicast(mm,
907 					hstart, hend, pid, PMD_SIZE, MMU_PAGE_2M, false);
908 		}
909 	}
910 	preempt_enable();
911 }
912 
913 void radix__flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
914 		     unsigned long end)
915 
916 {
917 #ifdef CONFIG_HUGETLB_PAGE
918 	if (is_vm_hugetlb_page(vma))
919 		return radix__flush_hugetlb_tlb_range(vma, start, end);
920 #endif
921 
922 	__radix__flush_tlb_range(vma->vm_mm, start, end);
923 }
924 EXPORT_SYMBOL(radix__flush_tlb_range);
925 
926 static int radix_get_mmu_psize(int page_size)
927 {
928 	int psize;
929 
930 	if (page_size == (1UL << mmu_psize_defs[mmu_virtual_psize].shift))
931 		psize = mmu_virtual_psize;
932 	else if (page_size == (1UL << mmu_psize_defs[MMU_PAGE_2M].shift))
933 		psize = MMU_PAGE_2M;
934 	else if (page_size == (1UL << mmu_psize_defs[MMU_PAGE_1G].shift))
935 		psize = MMU_PAGE_1G;
936 	else
937 		return -1;
938 	return psize;
939 }
940 
941 /*
942  * Flush partition scoped LPID address translation for all CPUs.
943  */
944 void radix__flush_tlb_lpid_page(unsigned int lpid,
945 					unsigned long addr,
946 					unsigned long page_size)
947 {
948 	int psize = radix_get_mmu_psize(page_size);
949 
950 	_tlbie_lpid_va(addr, lpid, psize, RIC_FLUSH_TLB);
951 }
952 EXPORT_SYMBOL_GPL(radix__flush_tlb_lpid_page);
953 
954 /*
955  * Flush partition scoped PWC from LPID for all CPUs.
956  */
957 void radix__flush_pwc_lpid(unsigned int lpid)
958 {
959 	_tlbie_lpid(lpid, RIC_FLUSH_PWC);
960 }
961 EXPORT_SYMBOL_GPL(radix__flush_pwc_lpid);
962 
963 /*
964  * Flush partition scoped translations from LPID (=LPIDR)
965  */
966 void radix__flush_all_lpid(unsigned int lpid)
967 {
968 	_tlbie_lpid(lpid, RIC_FLUSH_ALL);
969 }
970 EXPORT_SYMBOL_GPL(radix__flush_all_lpid);
971 
972 /*
973  * Flush process scoped translations from LPID (=LPIDR)
974  */
975 void radix__flush_all_lpid_guest(unsigned int lpid)
976 {
977 	_tlbie_lpid_guest(lpid, RIC_FLUSH_ALL);
978 }
979 
980 static void radix__flush_tlb_pwc_range_psize(struct mm_struct *mm, unsigned long start,
981 				  unsigned long end, int psize);
982 
983 void radix__tlb_flush(struct mmu_gather *tlb)
984 {
985 	int psize = 0;
986 	struct mm_struct *mm = tlb->mm;
987 	int page_size = tlb->page_size;
988 	unsigned long start = tlb->start;
989 	unsigned long end = tlb->end;
990 
991 	/*
992 	 * if page size is not something we understand, do a full mm flush
993 	 *
994 	 * A "fullmm" flush must always do a flush_all_mm (RIC=2) flush
995 	 * that flushes the process table entry cache upon process teardown.
996 	 * See the comment for radix in arch_exit_mmap().
997 	 */
998 	if (tlb->fullmm || tlb->need_flush_all) {
999 		__flush_all_mm(mm, true);
1000 	} else if ( (psize = radix_get_mmu_psize(page_size)) == -1) {
1001 		if (!tlb->freed_tables)
1002 			radix__flush_tlb_mm(mm);
1003 		else
1004 			radix__flush_all_mm(mm);
1005 	} else {
1006 		if (!tlb->freed_tables)
1007 			radix__flush_tlb_range_psize(mm, start, end, psize);
1008 		else
1009 			radix__flush_tlb_pwc_range_psize(mm, start, end, psize);
1010 	}
1011 }
1012 
1013 static __always_inline void __radix__flush_tlb_range_psize(struct mm_struct *mm,
1014 				unsigned long start, unsigned long end,
1015 				int psize, bool also_pwc)
1016 {
1017 	unsigned long pid;
1018 	unsigned int page_shift = mmu_psize_defs[psize].shift;
1019 	unsigned long page_size = 1UL << page_shift;
1020 	unsigned long nr_pages = (end - start) >> page_shift;
1021 	bool local, full;
1022 
1023 	pid = mm->context.id;
1024 	if (unlikely(pid == MMU_NO_CONTEXT))
1025 		return;
1026 
1027 	preempt_disable();
1028 	smp_mb(); /* see radix__flush_tlb_mm */
1029 	if (!mm_is_thread_local(mm)) {
1030 		if (unlikely(mm_is_singlethreaded(mm))) {
1031 			if (end != TLB_FLUSH_ALL) {
1032 				exit_flush_lazy_tlbs(mm);
1033 				goto is_local;
1034 			}
1035 		}
1036 		local = false;
1037 		full = (end == TLB_FLUSH_ALL ||
1038 				nr_pages > tlb_single_page_flush_ceiling);
1039 	} else {
1040 is_local:
1041 		local = true;
1042 		full = (end == TLB_FLUSH_ALL ||
1043 				nr_pages > tlb_local_single_page_flush_ceiling);
1044 	}
1045 
1046 	if (full) {
1047 		if (local) {
1048 			_tlbiel_pid(pid, also_pwc ? RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1049 		} else {
1050 			if (cputlb_use_tlbie()) {
1051 				if (mm_needs_flush_escalation(mm))
1052 					also_pwc = true;
1053 
1054 				_tlbie_pid(pid,
1055 					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1056 			} else {
1057 				_tlbiel_pid_multicast(mm, pid,
1058 					also_pwc ?  RIC_FLUSH_ALL : RIC_FLUSH_TLB);
1059 			}
1060 
1061 		}
1062 	} else {
1063 		if (local)
1064 			_tlbiel_va_range(start, end, pid, page_size, psize, also_pwc);
1065 		else if (cputlb_use_tlbie())
1066 			_tlbie_va_range(start, end, pid, page_size, psize, also_pwc);
1067 		else
1068 			_tlbiel_va_range_multicast(mm,
1069 					start, end, pid, page_size, psize, also_pwc);
1070 	}
1071 	preempt_enable();
1072 }
1073 
1074 void radix__flush_tlb_range_psize(struct mm_struct *mm, unsigned long start,
1075 				  unsigned long end, int psize)
1076 {
1077 	return __radix__flush_tlb_range_psize(mm, start, end, psize, false);
1078 }
1079 
1080 static void radix__flush_tlb_pwc_range_psize(struct mm_struct *mm, unsigned long start,
1081 				  unsigned long end, int psize)
1082 {
1083 	__radix__flush_tlb_range_psize(mm, start, end, psize, true);
1084 }
1085 
1086 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1087 void radix__flush_tlb_collapsed_pmd(struct mm_struct *mm, unsigned long addr)
1088 {
1089 	unsigned long pid, end;
1090 
1091 	pid = mm->context.id;
1092 	if (unlikely(pid == MMU_NO_CONTEXT))
1093 		return;
1094 
1095 	/* 4k page size, just blow the world */
1096 	if (PAGE_SIZE == 0x1000) {
1097 		radix__flush_all_mm(mm);
1098 		return;
1099 	}
1100 
1101 	end = addr + HPAGE_PMD_SIZE;
1102 
1103 	/* Otherwise first do the PWC, then iterate the pages. */
1104 	preempt_disable();
1105 	smp_mb(); /* see radix__flush_tlb_mm */
1106 	if (!mm_is_thread_local(mm)) {
1107 		if (unlikely(mm_is_singlethreaded(mm))) {
1108 			exit_flush_lazy_tlbs(mm);
1109 			goto local;
1110 		}
1111 		if (cputlb_use_tlbie())
1112 			_tlbie_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1113 		else
1114 			_tlbiel_va_range_multicast(mm,
1115 					addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1116 	} else {
1117 local:
1118 		_tlbiel_va_range(addr, end, pid, PAGE_SIZE, mmu_virtual_psize, true);
1119 	}
1120 
1121 	preempt_enable();
1122 }
1123 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1124 
1125 void radix__flush_pmd_tlb_range(struct vm_area_struct *vma,
1126 				unsigned long start, unsigned long end)
1127 {
1128 	radix__flush_tlb_range_psize(vma->vm_mm, start, end, MMU_PAGE_2M);
1129 }
1130 EXPORT_SYMBOL(radix__flush_pmd_tlb_range);
1131 
1132 void radix__flush_tlb_all(void)
1133 {
1134 	unsigned long rb,prs,r,rs;
1135 	unsigned long ric = RIC_FLUSH_ALL;
1136 
1137 	rb = 0x3 << PPC_BITLSHIFT(53); /* IS = 3 */
1138 	prs = 0; /* partition scoped */
1139 	r = 1;   /* radix format */
1140 	rs = 1 & ((1UL << 32) - 1); /* any LPID value to flush guest mappings */
1141 
1142 	asm volatile("ptesync": : :"memory");
1143 	/*
1144 	 * now flush guest entries by passing PRS = 1 and LPID != 0
1145 	 */
1146 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
1147 		     : : "r"(rb), "i"(r), "i"(1), "i"(ric), "r"(rs) : "memory");
1148 	/*
1149 	 * now flush host entires by passing PRS = 0 and LPID == 0
1150 	 */
1151 	asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
1152 		     : : "r"(rb), "i"(r), "i"(prs), "i"(ric), "r"(0) : "memory");
1153 	asm volatile("eieio; tlbsync; ptesync": : :"memory");
1154 }
1155 
1156 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
1157 extern void radix_kvm_prefetch_workaround(struct mm_struct *mm)
1158 {
1159 	unsigned long pid = mm->context.id;
1160 
1161 	if (unlikely(pid == MMU_NO_CONTEXT))
1162 		return;
1163 
1164 	if (!cpu_has_feature(CPU_FTR_P9_RADIX_PREFETCH_BUG))
1165 		return;
1166 
1167 	/*
1168 	 * If this context hasn't run on that CPU before and KVM is
1169 	 * around, there's a slim chance that the guest on another
1170 	 * CPU just brought in obsolete translation into the TLB of
1171 	 * this CPU due to a bad prefetch using the guest PID on
1172 	 * the way into the hypervisor.
1173 	 *
1174 	 * We work around this here. If KVM is possible, we check if
1175 	 * any sibling thread is in KVM. If it is, the window may exist
1176 	 * and thus we flush that PID from the core.
1177 	 *
1178 	 * A potential future improvement would be to mark which PIDs
1179 	 * have never been used on the system and avoid it if the PID
1180 	 * is new and the process has no other cpumask bit set.
1181 	 */
1182 	if (cpu_has_feature(CPU_FTR_HVMODE) && radix_enabled()) {
1183 		int cpu = smp_processor_id();
1184 		int sib = cpu_first_thread_sibling(cpu);
1185 		bool flush = false;
1186 
1187 		for (; sib <= cpu_last_thread_sibling(cpu) && !flush; sib++) {
1188 			if (sib == cpu)
1189 				continue;
1190 			if (!cpu_possible(sib))
1191 				continue;
1192 			if (paca_ptrs[sib]->kvm_hstate.kvm_vcpu)
1193 				flush = true;
1194 		}
1195 		if (flush)
1196 			_tlbiel_pid(pid, RIC_FLUSH_ALL);
1197 	}
1198 }
1199 EXPORT_SYMBOL_GPL(radix_kvm_prefetch_workaround);
1200 #endif /* CONFIG_KVM_BOOK3S_HV_POSSIBLE */
1201