1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * native hashtable management.
4  *
5  * SMP scalability work:
6  *    Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
7  */
8 
9 #undef DEBUG_LOW
10 
11 #include <linux/spinlock.h>
12 #include <linux/bitops.h>
13 #include <linux/of.h>
14 #include <linux/processor.h>
15 #include <linux/threads.h>
16 #include <linux/smp.h>
17 #include <linux/pgtable.h>
18 
19 #include <asm/machdep.h>
20 #include <asm/mmu.h>
21 #include <asm/mmu_context.h>
22 #include <asm/trace.h>
23 #include <asm/tlb.h>
24 #include <asm/cputable.h>
25 #include <asm/udbg.h>
26 #include <asm/kexec.h>
27 #include <asm/ppc-opcode.h>
28 #include <asm/feature-fixups.h>
29 
30 #include <misc/cxl-base.h>
31 
32 #ifdef DEBUG_LOW
33 #define DBG_LOW(fmt...) udbg_printf(fmt)
34 #else
35 #define DBG_LOW(fmt...)
36 #endif
37 
38 #ifdef __BIG_ENDIAN__
39 #define HPTE_LOCK_BIT 3
40 #else
41 #define HPTE_LOCK_BIT (56+3)
42 #endif
43 
44 static DEFINE_RAW_SPINLOCK(native_tlbie_lock);
45 
46 #ifdef CONFIG_LOCKDEP
47 static struct lockdep_map hpte_lock_map =
48 	STATIC_LOCKDEP_MAP_INIT("hpte_lock", &hpte_lock_map);
49 
50 static void acquire_hpte_lock(void)
51 {
52 	lock_map_acquire(&hpte_lock_map);
53 }
54 
55 static void release_hpte_lock(void)
56 {
57 	lock_map_release(&hpte_lock_map);
58 }
59 #else
60 static void acquire_hpte_lock(void)
61 {
62 }
63 
64 static void release_hpte_lock(void)
65 {
66 }
67 #endif
68 
69 static inline unsigned long  ___tlbie(unsigned long vpn, int psize,
70 						int apsize, int ssize)
71 {
72 	unsigned long va;
73 	unsigned int penc;
74 	unsigned long sllp;
75 
76 	/*
77 	 * We need 14 to 65 bits of va for a tlibe of 4K page
78 	 * With vpn we ignore the lower VPN_SHIFT bits already.
79 	 * And top two bits are already ignored because we can
80 	 * only accomodate 76 bits in a 64 bit vpn with a VPN_SHIFT
81 	 * of 12.
82 	 */
83 	va = vpn << VPN_SHIFT;
84 	/*
85 	 * clear top 16 bits of 64bit va, non SLS segment
86 	 * Older versions of the architecture (2.02 and earler) require the
87 	 * masking of the top 16 bits.
88 	 */
89 	if (mmu_has_feature(MMU_FTR_TLBIE_CROP_VA))
90 		va &= ~(0xffffULL << 48);
91 
92 	switch (psize) {
93 	case MMU_PAGE_4K:
94 		/* clear out bits after (52) [0....52.....63] */
95 		va &= ~((1ul << (64 - 52)) - 1);
96 		va |= ssize << 8;
97 		sllp = get_sllp_encoding(apsize);
98 		va |= sllp << 5;
99 		asm volatile(ASM_FTR_IFCLR("tlbie %0,0", PPC_TLBIE(%1,%0), %2)
100 			     : : "r" (va), "r"(0), "i" (CPU_FTR_ARCH_206)
101 			     : "memory");
102 		break;
103 	default:
104 		/* We need 14 to 14 + i bits of va */
105 		penc = mmu_psize_defs[psize].penc[apsize];
106 		va &= ~((1ul << mmu_psize_defs[apsize].shift) - 1);
107 		va |= penc << 12;
108 		va |= ssize << 8;
109 		/*
110 		 * AVAL bits:
111 		 * We don't need all the bits, but rest of the bits
112 		 * must be ignored by the processor.
113 		 * vpn cover upto 65 bits of va. (0...65) and we need
114 		 * 58..64 bits of va.
115 		 */
116 		va |= (vpn & 0xfe); /* AVAL */
117 		va |= 1; /* L */
118 		asm volatile(ASM_FTR_IFCLR("tlbie %0,1", PPC_TLBIE(%1,%0), %2)
119 			     : : "r" (va), "r"(0), "i" (CPU_FTR_ARCH_206)
120 			     : "memory");
121 		break;
122 	}
123 	return va;
124 }
125 
126 static inline void fixup_tlbie_vpn(unsigned long vpn, int psize,
127 				   int apsize, int ssize)
128 {
129 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_ERAT_BUG)) {
130 		/* Radix flush for a hash guest */
131 
132 		unsigned long rb,rs,prs,r,ric;
133 
134 		rb = PPC_BIT(52); /* IS = 2 */
135 		rs = 0;  /* lpid = 0 */
136 		prs = 0; /* partition scoped */
137 		r = 1;   /* radix format */
138 		ric = 0; /* RIC_FLSUH_TLB */
139 
140 		/*
141 		 * Need the extra ptesync to make sure we don't
142 		 * re-order the tlbie
143 		 */
144 		asm volatile("ptesync": : :"memory");
145 		asm volatile(PPC_TLBIE_5(%0, %4, %3, %2, %1)
146 			     : : "r"(rb), "i"(r), "i"(prs),
147 			       "i"(ric), "r"(rs) : "memory");
148 	}
149 
150 
151 	if (cpu_has_feature(CPU_FTR_P9_TLBIE_STQ_BUG)) {
152 		/* Need the extra ptesync to ensure we don't reorder tlbie*/
153 		asm volatile("ptesync": : :"memory");
154 		___tlbie(vpn, psize, apsize, ssize);
155 	}
156 }
157 
158 static inline void __tlbie(unsigned long vpn, int psize, int apsize, int ssize)
159 {
160 	unsigned long rb;
161 
162 	rb = ___tlbie(vpn, psize, apsize, ssize);
163 	trace_tlbie(0, 0, rb, 0, 0, 0, 0);
164 }
165 
166 static inline void __tlbiel(unsigned long vpn, int psize, int apsize, int ssize)
167 {
168 	unsigned long va;
169 	unsigned int penc;
170 	unsigned long sllp;
171 
172 	/* VPN_SHIFT can be atmost 12 */
173 	va = vpn << VPN_SHIFT;
174 	/*
175 	 * clear top 16 bits of 64 bit va, non SLS segment
176 	 * Older versions of the architecture (2.02 and earler) require the
177 	 * masking of the top 16 bits.
178 	 */
179 	if (mmu_has_feature(MMU_FTR_TLBIE_CROP_VA))
180 		va &= ~(0xffffULL << 48);
181 
182 	switch (psize) {
183 	case MMU_PAGE_4K:
184 		/* clear out bits after(52) [0....52.....63] */
185 		va &= ~((1ul << (64 - 52)) - 1);
186 		va |= ssize << 8;
187 		sllp = get_sllp_encoding(apsize);
188 		va |= sllp << 5;
189 		asm volatile(ASM_FTR_IFSET("tlbiel %0", PPC_TLBIEL_v205(%0, 0), %1)
190 			     : : "r" (va), "i" (CPU_FTR_ARCH_206)
191 			     : "memory");
192 		break;
193 	default:
194 		/* We need 14 to 14 + i bits of va */
195 		penc = mmu_psize_defs[psize].penc[apsize];
196 		va &= ~((1ul << mmu_psize_defs[apsize].shift) - 1);
197 		va |= penc << 12;
198 		va |= ssize << 8;
199 		/*
200 		 * AVAL bits:
201 		 * We don't need all the bits, but rest of the bits
202 		 * must be ignored by the processor.
203 		 * vpn cover upto 65 bits of va. (0...65) and we need
204 		 * 58..64 bits of va.
205 		 */
206 		va |= (vpn & 0xfe);
207 		va |= 1; /* L */
208 		asm volatile(ASM_FTR_IFSET("tlbiel %0", PPC_TLBIEL_v205(%0, 1), %1)
209 			     : : "r" (va), "i" (CPU_FTR_ARCH_206)
210 			     : "memory");
211 		break;
212 	}
213 	trace_tlbie(0, 1, va, 0, 0, 0, 0);
214 
215 }
216 
217 static inline void tlbie(unsigned long vpn, int psize, int apsize,
218 			 int ssize, int local)
219 {
220 	unsigned int use_local;
221 	int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
222 
223 	use_local = local && mmu_has_feature(MMU_FTR_TLBIEL) && !cxl_ctx_in_use();
224 
225 	if (use_local)
226 		use_local = mmu_psize_defs[psize].tlbiel;
227 	if (lock_tlbie && !use_local)
228 		raw_spin_lock(&native_tlbie_lock);
229 	asm volatile("ptesync": : :"memory");
230 	if (use_local) {
231 		__tlbiel(vpn, psize, apsize, ssize);
232 		ppc_after_tlbiel_barrier();
233 	} else {
234 		__tlbie(vpn, psize, apsize, ssize);
235 		fixup_tlbie_vpn(vpn, psize, apsize, ssize);
236 		asm volatile("eieio; tlbsync; ptesync": : :"memory");
237 	}
238 	if (lock_tlbie && !use_local)
239 		raw_spin_unlock(&native_tlbie_lock);
240 }
241 
242 static inline void native_lock_hpte(struct hash_pte *hptep)
243 {
244 	unsigned long *word = (unsigned long *)&hptep->v;
245 
246 	acquire_hpte_lock();
247 	while (1) {
248 		if (!test_and_set_bit_lock(HPTE_LOCK_BIT, word))
249 			break;
250 		spin_begin();
251 		while(test_bit(HPTE_LOCK_BIT, word))
252 			spin_cpu_relax();
253 		spin_end();
254 	}
255 }
256 
257 static inline void native_unlock_hpte(struct hash_pte *hptep)
258 {
259 	unsigned long *word = (unsigned long *)&hptep->v;
260 
261 	release_hpte_lock();
262 	clear_bit_unlock(HPTE_LOCK_BIT, word);
263 }
264 
265 static long native_hpte_insert(unsigned long hpte_group, unsigned long vpn,
266 			unsigned long pa, unsigned long rflags,
267 			unsigned long vflags, int psize, int apsize, int ssize)
268 {
269 	struct hash_pte *hptep = htab_address + hpte_group;
270 	unsigned long hpte_v, hpte_r;
271 	unsigned long flags;
272 	int i;
273 
274 	local_irq_save(flags);
275 
276 	if (!(vflags & HPTE_V_BOLTED)) {
277 		DBG_LOW("    insert(group=%lx, vpn=%016lx, pa=%016lx,"
278 			" rflags=%lx, vflags=%lx, psize=%d)\n",
279 			hpte_group, vpn, pa, rflags, vflags, psize);
280 	}
281 
282 	for (i = 0; i < HPTES_PER_GROUP; i++) {
283 		if (! (be64_to_cpu(hptep->v) & HPTE_V_VALID)) {
284 			/* retry with lock held */
285 			native_lock_hpte(hptep);
286 			if (! (be64_to_cpu(hptep->v) & HPTE_V_VALID))
287 				break;
288 			native_unlock_hpte(hptep);
289 		}
290 
291 		hptep++;
292 	}
293 
294 	if (i == HPTES_PER_GROUP) {
295 		local_irq_restore(flags);
296 		return -1;
297 	}
298 
299 	hpte_v = hpte_encode_v(vpn, psize, apsize, ssize) | vflags | HPTE_V_VALID;
300 	hpte_r = hpte_encode_r(pa, psize, apsize) | rflags;
301 
302 	if (!(vflags & HPTE_V_BOLTED)) {
303 		DBG_LOW(" i=%x hpte_v=%016lx, hpte_r=%016lx\n",
304 			i, hpte_v, hpte_r);
305 	}
306 
307 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
308 		hpte_r = hpte_old_to_new_r(hpte_v, hpte_r);
309 		hpte_v = hpte_old_to_new_v(hpte_v);
310 	}
311 
312 	hptep->r = cpu_to_be64(hpte_r);
313 	/* Guarantee the second dword is visible before the valid bit */
314 	eieio();
315 	/*
316 	 * Now set the first dword including the valid bit
317 	 * NOTE: this also unlocks the hpte
318 	 */
319 	release_hpte_lock();
320 	hptep->v = cpu_to_be64(hpte_v);
321 
322 	__asm__ __volatile__ ("ptesync" : : : "memory");
323 
324 	local_irq_restore(flags);
325 
326 	return i | (!!(vflags & HPTE_V_SECONDARY) << 3);
327 }
328 
329 static long native_hpte_remove(unsigned long hpte_group)
330 {
331 	struct hash_pte *hptep;
332 	int i;
333 	int slot_offset;
334 	unsigned long hpte_v;
335 
336 	DBG_LOW("    remove(group=%lx)\n", hpte_group);
337 
338 	/* pick a random entry to start at */
339 	slot_offset = mftb() & 0x7;
340 
341 	for (i = 0; i < HPTES_PER_GROUP; i++) {
342 		hptep = htab_address + hpte_group + slot_offset;
343 		hpte_v = be64_to_cpu(hptep->v);
344 
345 		if ((hpte_v & HPTE_V_VALID) && !(hpte_v & HPTE_V_BOLTED)) {
346 			/* retry with lock held */
347 			native_lock_hpte(hptep);
348 			hpte_v = be64_to_cpu(hptep->v);
349 			if ((hpte_v & HPTE_V_VALID)
350 			    && !(hpte_v & HPTE_V_BOLTED))
351 				break;
352 			native_unlock_hpte(hptep);
353 		}
354 
355 		slot_offset++;
356 		slot_offset &= 0x7;
357 	}
358 
359 	if (i == HPTES_PER_GROUP)
360 		return -1;
361 
362 	/* Invalidate the hpte. NOTE: this also unlocks it */
363 	release_hpte_lock();
364 	hptep->v = 0;
365 
366 	return i;
367 }
368 
369 static long native_hpte_updatepp(unsigned long slot, unsigned long newpp,
370 				 unsigned long vpn, int bpsize,
371 				 int apsize, int ssize, unsigned long flags)
372 {
373 	struct hash_pte *hptep = htab_address + slot;
374 	unsigned long hpte_v, want_v;
375 	int ret = 0, local = 0;
376 	unsigned long irqflags;
377 
378 	local_irq_save(irqflags);
379 
380 	want_v = hpte_encode_avpn(vpn, bpsize, ssize);
381 
382 	DBG_LOW("    update(vpn=%016lx, avpnv=%016lx, group=%lx, newpp=%lx)",
383 		vpn, want_v & HPTE_V_AVPN, slot, newpp);
384 
385 	hpte_v = hpte_get_old_v(hptep);
386 	/*
387 	 * We need to invalidate the TLB always because hpte_remove doesn't do
388 	 * a tlb invalidate. If a hash bucket gets full, we "evict" a more/less
389 	 * random entry from it. When we do that we don't invalidate the TLB
390 	 * (hpte_remove) because we assume the old translation is still
391 	 * technically "valid".
392 	 */
393 	if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID)) {
394 		DBG_LOW(" -> miss\n");
395 		ret = -1;
396 	} else {
397 		native_lock_hpte(hptep);
398 		/* recheck with locks held */
399 		hpte_v = hpte_get_old_v(hptep);
400 		if (unlikely(!HPTE_V_COMPARE(hpte_v, want_v) ||
401 			     !(hpte_v & HPTE_V_VALID))) {
402 			ret = -1;
403 		} else {
404 			DBG_LOW(" -> hit\n");
405 			/* Update the HPTE */
406 			hptep->r = cpu_to_be64((be64_to_cpu(hptep->r) &
407 						~(HPTE_R_PPP | HPTE_R_N)) |
408 					       (newpp & (HPTE_R_PPP | HPTE_R_N |
409 							 HPTE_R_C)));
410 		}
411 		native_unlock_hpte(hptep);
412 	}
413 
414 	if (flags & HPTE_LOCAL_UPDATE)
415 		local = 1;
416 	/*
417 	 * Ensure it is out of the tlb too if it is not a nohpte fault
418 	 */
419 	if (!(flags & HPTE_NOHPTE_UPDATE))
420 		tlbie(vpn, bpsize, apsize, ssize, local);
421 
422 	local_irq_restore(irqflags);
423 
424 	return ret;
425 }
426 
427 static long __native_hpte_find(unsigned long want_v, unsigned long slot)
428 {
429 	struct hash_pte *hptep;
430 	unsigned long hpte_v;
431 	unsigned long i;
432 
433 	for (i = 0; i < HPTES_PER_GROUP; i++) {
434 
435 		hptep = htab_address + slot;
436 		hpte_v = hpte_get_old_v(hptep);
437 		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID))
438 			/* HPTE matches */
439 			return slot;
440 		++slot;
441 	}
442 
443 	return -1;
444 }
445 
446 static long native_hpte_find(unsigned long vpn, int psize, int ssize)
447 {
448 	unsigned long hpte_group;
449 	unsigned long want_v;
450 	unsigned long hash;
451 	long slot;
452 
453 	hash = hpt_hash(vpn, mmu_psize_defs[psize].shift, ssize);
454 	want_v = hpte_encode_avpn(vpn, psize, ssize);
455 
456 	/*
457 	 * We try to keep bolted entries always in primary hash
458 	 * But in some case we can find them in secondary too.
459 	 */
460 	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
461 	slot = __native_hpte_find(want_v, hpte_group);
462 	if (slot < 0) {
463 		/* Try in secondary */
464 		hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
465 		slot = __native_hpte_find(want_v, hpte_group);
466 		if (slot < 0)
467 			return -1;
468 	}
469 
470 	return slot;
471 }
472 
473 /*
474  * Update the page protection bits. Intended to be used to create
475  * guard pages for kernel data structures on pages which are bolted
476  * in the HPT. Assumes pages being operated on will not be stolen.
477  *
478  * No need to lock here because we should be the only user.
479  */
480 static void native_hpte_updateboltedpp(unsigned long newpp, unsigned long ea,
481 				       int psize, int ssize)
482 {
483 	unsigned long vpn;
484 	unsigned long vsid;
485 	long slot;
486 	struct hash_pte *hptep;
487 	unsigned long flags;
488 
489 	local_irq_save(flags);
490 
491 	vsid = get_kernel_vsid(ea, ssize);
492 	vpn = hpt_vpn(ea, vsid, ssize);
493 
494 	slot = native_hpte_find(vpn, psize, ssize);
495 	if (slot == -1)
496 		panic("could not find page to bolt\n");
497 	hptep = htab_address + slot;
498 
499 	/* Update the HPTE */
500 	hptep->r = cpu_to_be64((be64_to_cpu(hptep->r) &
501 				~(HPTE_R_PPP | HPTE_R_N)) |
502 			       (newpp & (HPTE_R_PPP | HPTE_R_N)));
503 	/*
504 	 * Ensure it is out of the tlb too. Bolted entries base and
505 	 * actual page size will be same.
506 	 */
507 	tlbie(vpn, psize, psize, ssize, 0);
508 
509 	local_irq_restore(flags);
510 }
511 
512 /*
513  * Remove a bolted kernel entry. Memory hotplug uses this.
514  *
515  * No need to lock here because we should be the only user.
516  */
517 static int native_hpte_removebolted(unsigned long ea, int psize, int ssize)
518 {
519 	unsigned long vpn;
520 	unsigned long vsid;
521 	long slot;
522 	struct hash_pte *hptep;
523 	unsigned long flags;
524 
525 	local_irq_save(flags);
526 
527 	vsid = get_kernel_vsid(ea, ssize);
528 	vpn = hpt_vpn(ea, vsid, ssize);
529 
530 	slot = native_hpte_find(vpn, psize, ssize);
531 	if (slot == -1)
532 		return -ENOENT;
533 
534 	hptep = htab_address + slot;
535 
536 	VM_WARN_ON(!(be64_to_cpu(hptep->v) & HPTE_V_BOLTED));
537 
538 	/* Invalidate the hpte */
539 	hptep->v = 0;
540 
541 	/* Invalidate the TLB */
542 	tlbie(vpn, psize, psize, ssize, 0);
543 
544 	local_irq_restore(flags);
545 
546 	return 0;
547 }
548 
549 
550 static void native_hpte_invalidate(unsigned long slot, unsigned long vpn,
551 				   int bpsize, int apsize, int ssize, int local)
552 {
553 	struct hash_pte *hptep = htab_address + slot;
554 	unsigned long hpte_v;
555 	unsigned long want_v;
556 	unsigned long flags;
557 
558 	local_irq_save(flags);
559 
560 	DBG_LOW("    invalidate(vpn=%016lx, hash: %lx)\n", vpn, slot);
561 
562 	want_v = hpte_encode_avpn(vpn, bpsize, ssize);
563 	hpte_v = hpte_get_old_v(hptep);
564 
565 	if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
566 		native_lock_hpte(hptep);
567 		/* recheck with locks held */
568 		hpte_v = hpte_get_old_v(hptep);
569 
570 		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
571 			/* Invalidate the hpte. NOTE: this also unlocks it */
572 			release_hpte_lock();
573 			hptep->v = 0;
574 		} else
575 			native_unlock_hpte(hptep);
576 	}
577 	/*
578 	 * We need to invalidate the TLB always because hpte_remove doesn't do
579 	 * a tlb invalidate. If a hash bucket gets full, we "evict" a more/less
580 	 * random entry from it. When we do that we don't invalidate the TLB
581 	 * (hpte_remove) because we assume the old translation is still
582 	 * technically "valid".
583 	 */
584 	tlbie(vpn, bpsize, apsize, ssize, local);
585 
586 	local_irq_restore(flags);
587 }
588 
589 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
590 static void native_hugepage_invalidate(unsigned long vsid,
591 				       unsigned long addr,
592 				       unsigned char *hpte_slot_array,
593 				       int psize, int ssize, int local)
594 {
595 	int i;
596 	struct hash_pte *hptep;
597 	int actual_psize = MMU_PAGE_16M;
598 	unsigned int max_hpte_count, valid;
599 	unsigned long flags, s_addr = addr;
600 	unsigned long hpte_v, want_v, shift;
601 	unsigned long hidx, vpn = 0, hash, slot;
602 
603 	shift = mmu_psize_defs[psize].shift;
604 	max_hpte_count = 1U << (PMD_SHIFT - shift);
605 
606 	local_irq_save(flags);
607 	for (i = 0; i < max_hpte_count; i++) {
608 		valid = hpte_valid(hpte_slot_array, i);
609 		if (!valid)
610 			continue;
611 		hidx =  hpte_hash_index(hpte_slot_array, i);
612 
613 		/* get the vpn */
614 		addr = s_addr + (i * (1ul << shift));
615 		vpn = hpt_vpn(addr, vsid, ssize);
616 		hash = hpt_hash(vpn, shift, ssize);
617 		if (hidx & _PTEIDX_SECONDARY)
618 			hash = ~hash;
619 
620 		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
621 		slot += hidx & _PTEIDX_GROUP_IX;
622 
623 		hptep = htab_address + slot;
624 		want_v = hpte_encode_avpn(vpn, psize, ssize);
625 		hpte_v = hpte_get_old_v(hptep);
626 
627 		/* Even if we miss, we need to invalidate the TLB */
628 		if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
629 			/* recheck with locks held */
630 			native_lock_hpte(hptep);
631 			hpte_v = hpte_get_old_v(hptep);
632 
633 			if (HPTE_V_COMPARE(hpte_v, want_v) && (hpte_v & HPTE_V_VALID)) {
634 				/* Invalidate the hpte. NOTE: this also unlocks it */
635 				release_hpte_lock();
636 				hptep->v = 0;
637 			} else
638 				native_unlock_hpte(hptep);
639 		}
640 		/*
641 		 * We need to do tlb invalidate for all the address, tlbie
642 		 * instruction compares entry_VA in tlb with the VA specified
643 		 * here
644 		 */
645 		tlbie(vpn, psize, actual_psize, ssize, local);
646 	}
647 	local_irq_restore(flags);
648 }
649 #else
650 static void native_hugepage_invalidate(unsigned long vsid,
651 				       unsigned long addr,
652 				       unsigned char *hpte_slot_array,
653 				       int psize, int ssize, int local)
654 {
655 	WARN(1, "%s called without THP support\n", __func__);
656 }
657 #endif
658 
659 static void hpte_decode(struct hash_pte *hpte, unsigned long slot,
660 			int *psize, int *apsize, int *ssize, unsigned long *vpn)
661 {
662 	unsigned long avpn, pteg, vpi;
663 	unsigned long hpte_v = be64_to_cpu(hpte->v);
664 	unsigned long hpte_r = be64_to_cpu(hpte->r);
665 	unsigned long vsid, seg_off;
666 	int size, a_size, shift;
667 	/* Look at the 8 bit LP value */
668 	unsigned int lp = (hpte_r >> LP_SHIFT) & ((1 << LP_BITS) - 1);
669 
670 	if (cpu_has_feature(CPU_FTR_ARCH_300)) {
671 		hpte_v = hpte_new_to_old_v(hpte_v, hpte_r);
672 		hpte_r = hpte_new_to_old_r(hpte_r);
673 	}
674 	if (!(hpte_v & HPTE_V_LARGE)) {
675 		size   = MMU_PAGE_4K;
676 		a_size = MMU_PAGE_4K;
677 	} else {
678 		size = hpte_page_sizes[lp] & 0xf;
679 		a_size = hpte_page_sizes[lp] >> 4;
680 	}
681 	/* This works for all page sizes, and for 256M and 1T segments */
682 	*ssize = hpte_v >> HPTE_V_SSIZE_SHIFT;
683 	shift = mmu_psize_defs[size].shift;
684 
685 	avpn = (HPTE_V_AVPN_VAL(hpte_v) & ~mmu_psize_defs[size].avpnm);
686 	pteg = slot / HPTES_PER_GROUP;
687 	if (hpte_v & HPTE_V_SECONDARY)
688 		pteg = ~pteg;
689 
690 	switch (*ssize) {
691 	case MMU_SEGSIZE_256M:
692 		/* We only have 28 - 23 bits of seg_off in avpn */
693 		seg_off = (avpn & 0x1f) << 23;
694 		vsid    =  avpn >> 5;
695 		/* We can find more bits from the pteg value */
696 		if (shift < 23) {
697 			vpi = (vsid ^ pteg) & htab_hash_mask;
698 			seg_off |= vpi << shift;
699 		}
700 		*vpn = vsid << (SID_SHIFT - VPN_SHIFT) | seg_off >> VPN_SHIFT;
701 		break;
702 	case MMU_SEGSIZE_1T:
703 		/* We only have 40 - 23 bits of seg_off in avpn */
704 		seg_off = (avpn & 0x1ffff) << 23;
705 		vsid    = avpn >> 17;
706 		if (shift < 23) {
707 			vpi = (vsid ^ (vsid << 25) ^ pteg) & htab_hash_mask;
708 			seg_off |= vpi << shift;
709 		}
710 		*vpn = vsid << (SID_SHIFT_1T - VPN_SHIFT) | seg_off >> VPN_SHIFT;
711 		break;
712 	default:
713 		*vpn = size = 0;
714 	}
715 	*psize  = size;
716 	*apsize = a_size;
717 }
718 
719 /*
720  * clear all mappings on kexec.  All cpus are in real mode (or they will
721  * be when they isi), and we are the only one left.  We rely on our kernel
722  * mapping being 0xC0's and the hardware ignoring those two real bits.
723  *
724  * This must be called with interrupts disabled.
725  *
726  * Taking the native_tlbie_lock is unsafe here due to the possibility of
727  * lockdep being on. On pre POWER5 hardware, not taking the lock could
728  * cause deadlock. POWER5 and newer not taking the lock is fine. This only
729  * gets called during boot before secondary CPUs have come up and during
730  * crashdump and all bets are off anyway.
731  *
732  * TODO: add batching support when enabled.  remember, no dynamic memory here,
733  * although there is the control page available...
734  */
735 static notrace void native_hpte_clear(void)
736 {
737 	unsigned long vpn = 0;
738 	unsigned long slot, slots;
739 	struct hash_pte *hptep = htab_address;
740 	unsigned long hpte_v;
741 	unsigned long pteg_count;
742 	int psize, apsize, ssize;
743 
744 	pteg_count = htab_hash_mask + 1;
745 
746 	slots = pteg_count * HPTES_PER_GROUP;
747 
748 	for (slot = 0; slot < slots; slot++, hptep++) {
749 		/*
750 		 * we could lock the pte here, but we are the only cpu
751 		 * running,  right?  and for crash dump, we probably
752 		 * don't want to wait for a maybe bad cpu.
753 		 */
754 		hpte_v = be64_to_cpu(hptep->v);
755 
756 		/*
757 		 * Call __tlbie() here rather than tlbie() since we can't take the
758 		 * native_tlbie_lock.
759 		 */
760 		if (hpte_v & HPTE_V_VALID) {
761 			hpte_decode(hptep, slot, &psize, &apsize, &ssize, &vpn);
762 			hptep->v = 0;
763 			___tlbie(vpn, psize, apsize, ssize);
764 		}
765 	}
766 
767 	asm volatile("eieio; tlbsync; ptesync":::"memory");
768 }
769 
770 /*
771  * Batched hash table flush, we batch the tlbie's to avoid taking/releasing
772  * the lock all the time
773  */
774 static void native_flush_hash_range(unsigned long number, int local)
775 {
776 	unsigned long vpn = 0;
777 	unsigned long hash, index, hidx, shift, slot;
778 	struct hash_pte *hptep;
779 	unsigned long hpte_v;
780 	unsigned long want_v;
781 	unsigned long flags;
782 	real_pte_t pte;
783 	struct ppc64_tlb_batch *batch = this_cpu_ptr(&ppc64_tlb_batch);
784 	unsigned long psize = batch->psize;
785 	int ssize = batch->ssize;
786 	int i;
787 	unsigned int use_local;
788 
789 	use_local = local && mmu_has_feature(MMU_FTR_TLBIEL) &&
790 		mmu_psize_defs[psize].tlbiel && !cxl_ctx_in_use();
791 
792 	local_irq_save(flags);
793 
794 	for (i = 0; i < number; i++) {
795 		vpn = batch->vpn[i];
796 		pte = batch->pte[i];
797 
798 		pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
799 			hash = hpt_hash(vpn, shift, ssize);
800 			hidx = __rpte_to_hidx(pte, index);
801 			if (hidx & _PTEIDX_SECONDARY)
802 				hash = ~hash;
803 			slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
804 			slot += hidx & _PTEIDX_GROUP_IX;
805 			hptep = htab_address + slot;
806 			want_v = hpte_encode_avpn(vpn, psize, ssize);
807 			hpte_v = hpte_get_old_v(hptep);
808 
809 			if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID))
810 				continue;
811 			/* lock and try again */
812 			native_lock_hpte(hptep);
813 			hpte_v = hpte_get_old_v(hptep);
814 
815 			if (!HPTE_V_COMPARE(hpte_v, want_v) || !(hpte_v & HPTE_V_VALID))
816 				native_unlock_hpte(hptep);
817 			else {
818 				release_hpte_lock();
819 				hptep->v = 0;
820 			}
821 
822 		} pte_iterate_hashed_end();
823 	}
824 
825 	if (use_local) {
826 		asm volatile("ptesync":::"memory");
827 		for (i = 0; i < number; i++) {
828 			vpn = batch->vpn[i];
829 			pte = batch->pte[i];
830 
831 			pte_iterate_hashed_subpages(pte, psize,
832 						    vpn, index, shift) {
833 				__tlbiel(vpn, psize, psize, ssize);
834 			} pte_iterate_hashed_end();
835 		}
836 		ppc_after_tlbiel_barrier();
837 	} else {
838 		int lock_tlbie = !mmu_has_feature(MMU_FTR_LOCKLESS_TLBIE);
839 
840 		if (lock_tlbie)
841 			raw_spin_lock(&native_tlbie_lock);
842 
843 		asm volatile("ptesync":::"memory");
844 		for (i = 0; i < number; i++) {
845 			vpn = batch->vpn[i];
846 			pte = batch->pte[i];
847 
848 			pte_iterate_hashed_subpages(pte, psize,
849 						    vpn, index, shift) {
850 				__tlbie(vpn, psize, psize, ssize);
851 			} pte_iterate_hashed_end();
852 		}
853 		/*
854 		 * Just do one more with the last used values.
855 		 */
856 		fixup_tlbie_vpn(vpn, psize, psize, ssize);
857 		asm volatile("eieio; tlbsync; ptesync":::"memory");
858 
859 		if (lock_tlbie)
860 			raw_spin_unlock(&native_tlbie_lock);
861 	}
862 
863 	local_irq_restore(flags);
864 }
865 
866 void __init hpte_init_native(void)
867 {
868 	mmu_hash_ops.hpte_invalidate	= native_hpte_invalidate;
869 	mmu_hash_ops.hpte_updatepp	= native_hpte_updatepp;
870 	mmu_hash_ops.hpte_updateboltedpp = native_hpte_updateboltedpp;
871 	mmu_hash_ops.hpte_removebolted = native_hpte_removebolted;
872 	mmu_hash_ops.hpte_insert	= native_hpte_insert;
873 	mmu_hash_ops.hpte_remove	= native_hpte_remove;
874 	mmu_hash_ops.hpte_clear_all	= native_hpte_clear;
875 	mmu_hash_ops.flush_hash_range = native_flush_hash_range;
876 	mmu_hash_ops.hugepage_invalidate   = native_hugepage_invalidate;
877 }
878