xref: /linux/arch/powerpc/mm/book3s64/hash_utils.c (revision 6f62a8223e65c0571e48225d5d7e56de95225bae)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * PowerPC64 port by Mike Corrigan and Dave Engebretsen
4  *   {mikejc|engebret}@us.ibm.com
5  *
6  *    Copyright (c) 2000 Mike Corrigan <mikejc@us.ibm.com>
7  *
8  * SMP scalability work:
9  *    Copyright (C) 2001 Anton Blanchard <anton@au.ibm.com>, IBM
10  *
11  *    Module name: htab.c
12  *
13  *    Description:
14  *      PowerPC Hashed Page Table functions
15  */
16 
17 #undef DEBUG
18 #undef DEBUG_LOW
19 
20 #define pr_fmt(fmt) "hash-mmu: " fmt
21 #include <linux/spinlock.h>
22 #include <linux/errno.h>
23 #include <linux/sched/mm.h>
24 #include <linux/proc_fs.h>
25 #include <linux/stat.h>
26 #include <linux/sysctl.h>
27 #include <linux/export.h>
28 #include <linux/ctype.h>
29 #include <linux/cache.h>
30 #include <linux/init.h>
31 #include <linux/signal.h>
32 #include <linux/memblock.h>
33 #include <linux/context_tracking.h>
34 #include <linux/libfdt.h>
35 #include <linux/pkeys.h>
36 #include <linux/hugetlb.h>
37 #include <linux/cpu.h>
38 
39 #include <asm/debugfs.h>
40 #include <asm/processor.h>
41 #include <asm/pgtable.h>
42 #include <asm/mmu.h>
43 #include <asm/mmu_context.h>
44 #include <asm/page.h>
45 #include <asm/types.h>
46 #include <linux/uaccess.h>
47 #include <asm/machdep.h>
48 #include <asm/prom.h>
49 #include <asm/io.h>
50 #include <asm/eeh.h>
51 #include <asm/tlb.h>
52 #include <asm/cacheflush.h>
53 #include <asm/cputable.h>
54 #include <asm/sections.h>
55 #include <asm/copro.h>
56 #include <asm/udbg.h>
57 #include <asm/code-patching.h>
58 #include <asm/fadump.h>
59 #include <asm/firmware.h>
60 #include <asm/tm.h>
61 #include <asm/trace.h>
62 #include <asm/ps3.h>
63 #include <asm/pte-walk.h>
64 #include <asm/asm-prototypes.h>
65 #include <asm/ultravisor.h>
66 
67 #include <mm/mmu_decl.h>
68 
69 #ifdef DEBUG
70 #define DBG(fmt...) udbg_printf(fmt)
71 #else
72 #define DBG(fmt...)
73 #endif
74 
75 #ifdef DEBUG_LOW
76 #define DBG_LOW(fmt...) udbg_printf(fmt)
77 #else
78 #define DBG_LOW(fmt...)
79 #endif
80 
81 #define KB (1024)
82 #define MB (1024*KB)
83 #define GB (1024L*MB)
84 
85 /*
86  * Note:  pte   --> Linux PTE
87  *        HPTE  --> PowerPC Hashed Page Table Entry
88  *
89  * Execution context:
90  *   htab_initialize is called with the MMU off (of course), but
91  *   the kernel has been copied down to zero so it can directly
92  *   reference global data.  At this point it is very difficult
93  *   to print debug info.
94  *
95  */
96 
97 static unsigned long _SDR1;
98 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
99 EXPORT_SYMBOL_GPL(mmu_psize_defs);
100 
101 u8 hpte_page_sizes[1 << LP_BITS];
102 EXPORT_SYMBOL_GPL(hpte_page_sizes);
103 
104 struct hash_pte *htab_address;
105 unsigned long htab_size_bytes;
106 unsigned long htab_hash_mask;
107 EXPORT_SYMBOL_GPL(htab_hash_mask);
108 int mmu_linear_psize = MMU_PAGE_4K;
109 EXPORT_SYMBOL_GPL(mmu_linear_psize);
110 int mmu_virtual_psize = MMU_PAGE_4K;
111 int mmu_vmalloc_psize = MMU_PAGE_4K;
112 #ifdef CONFIG_SPARSEMEM_VMEMMAP
113 int mmu_vmemmap_psize = MMU_PAGE_4K;
114 #endif
115 int mmu_io_psize = MMU_PAGE_4K;
116 int mmu_kernel_ssize = MMU_SEGSIZE_256M;
117 EXPORT_SYMBOL_GPL(mmu_kernel_ssize);
118 int mmu_highuser_ssize = MMU_SEGSIZE_256M;
119 u16 mmu_slb_size = 64;
120 EXPORT_SYMBOL_GPL(mmu_slb_size);
121 #ifdef CONFIG_PPC_64K_PAGES
122 int mmu_ci_restrictions;
123 #endif
124 #ifdef CONFIG_DEBUG_PAGEALLOC
125 static u8 *linear_map_hash_slots;
126 static unsigned long linear_map_hash_count;
127 static DEFINE_SPINLOCK(linear_map_hash_lock);
128 #endif /* CONFIG_DEBUG_PAGEALLOC */
129 struct mmu_hash_ops mmu_hash_ops;
130 EXPORT_SYMBOL(mmu_hash_ops);
131 
132 /*
133  * These are definitions of page sizes arrays to be used when none
134  * is provided by the firmware.
135  */
136 
137 /*
138  * Fallback (4k pages only)
139  */
140 static struct mmu_psize_def mmu_psize_defaults[] = {
141 	[MMU_PAGE_4K] = {
142 		.shift	= 12,
143 		.sllp	= 0,
144 		.penc   = {[MMU_PAGE_4K] = 0, [1 ... MMU_PAGE_COUNT - 1] = -1},
145 		.avpnm	= 0,
146 		.tlbiel = 0,
147 	},
148 };
149 
150 /*
151  * POWER4, GPUL, POWER5
152  *
153  * Support for 16Mb large pages
154  */
155 static struct mmu_psize_def mmu_psize_defaults_gp[] = {
156 	[MMU_PAGE_4K] = {
157 		.shift	= 12,
158 		.sllp	= 0,
159 		.penc   = {[MMU_PAGE_4K] = 0, [1 ... MMU_PAGE_COUNT - 1] = -1},
160 		.avpnm	= 0,
161 		.tlbiel = 1,
162 	},
163 	[MMU_PAGE_16M] = {
164 		.shift	= 24,
165 		.sllp	= SLB_VSID_L,
166 		.penc   = {[0 ... MMU_PAGE_16M - 1] = -1, [MMU_PAGE_16M] = 0,
167 			    [MMU_PAGE_16M + 1 ... MMU_PAGE_COUNT - 1] = -1 },
168 		.avpnm	= 0x1UL,
169 		.tlbiel = 0,
170 	},
171 };
172 
173 /*
174  * 'R' and 'C' update notes:
175  *  - Under pHyp or KVM, the updatepp path will not set C, thus it *will*
176  *     create writeable HPTEs without C set, because the hcall H_PROTECT
177  *     that we use in that case will not update C
178  *  - The above is however not a problem, because we also don't do that
179  *     fancy "no flush" variant of eviction and we use H_REMOVE which will
180  *     do the right thing and thus we don't have the race I described earlier
181  *
182  *    - Under bare metal,  we do have the race, so we need R and C set
183  *    - We make sure R is always set and never lost
184  *    - C is _PAGE_DIRTY, and *should* always be set for a writeable mapping
185  */
186 unsigned long htab_convert_pte_flags(unsigned long pteflags)
187 {
188 	unsigned long rflags = 0;
189 
190 	/* _PAGE_EXEC -> NOEXEC */
191 	if ((pteflags & _PAGE_EXEC) == 0)
192 		rflags |= HPTE_R_N;
193 	/*
194 	 * PPP bits:
195 	 * Linux uses slb key 0 for kernel and 1 for user.
196 	 * kernel RW areas are mapped with PPP=0b000
197 	 * User area is mapped with PPP=0b010 for read/write
198 	 * or PPP=0b011 for read-only (including writeable but clean pages).
199 	 */
200 	if (pteflags & _PAGE_PRIVILEGED) {
201 		/*
202 		 * Kernel read only mapped with ppp bits 0b110
203 		 */
204 		if (!(pteflags & _PAGE_WRITE)) {
205 			if (mmu_has_feature(MMU_FTR_KERNEL_RO))
206 				rflags |= (HPTE_R_PP0 | 0x2);
207 			else
208 				rflags |= 0x3;
209 		}
210 	} else {
211 		if (pteflags & _PAGE_RWX)
212 			rflags |= 0x2;
213 		if (!((pteflags & _PAGE_WRITE) && (pteflags & _PAGE_DIRTY)))
214 			rflags |= 0x1;
215 	}
216 	/*
217 	 * We can't allow hardware to update hpte bits. Hence always
218 	 * set 'R' bit and set 'C' if it is a write fault
219 	 */
220 	rflags |=  HPTE_R_R;
221 
222 	if (pteflags & _PAGE_DIRTY)
223 		rflags |= HPTE_R_C;
224 	/*
225 	 * Add in WIG bits
226 	 */
227 
228 	if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_TOLERANT)
229 		rflags |= HPTE_R_I;
230 	else if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_NON_IDEMPOTENT)
231 		rflags |= (HPTE_R_I | HPTE_R_G);
232 	else if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_SAO)
233 		rflags |= (HPTE_R_W | HPTE_R_I | HPTE_R_M);
234 	else
235 		/*
236 		 * Add memory coherence if cache inhibited is not set
237 		 */
238 		rflags |= HPTE_R_M;
239 
240 	rflags |= pte_to_hpte_pkey_bits(pteflags);
241 	return rflags;
242 }
243 
244 int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
245 		      unsigned long pstart, unsigned long prot,
246 		      int psize, int ssize)
247 {
248 	unsigned long vaddr, paddr;
249 	unsigned int step, shift;
250 	int ret = 0;
251 
252 	shift = mmu_psize_defs[psize].shift;
253 	step = 1 << shift;
254 
255 	prot = htab_convert_pte_flags(prot);
256 
257 	DBG("htab_bolt_mapping(%lx..%lx -> %lx (%lx,%d,%d)\n",
258 	    vstart, vend, pstart, prot, psize, ssize);
259 
260 	for (vaddr = vstart, paddr = pstart; vaddr < vend;
261 	     vaddr += step, paddr += step) {
262 		unsigned long hash, hpteg;
263 		unsigned long vsid = get_kernel_vsid(vaddr, ssize);
264 		unsigned long vpn  = hpt_vpn(vaddr, vsid, ssize);
265 		unsigned long tprot = prot;
266 
267 		/*
268 		 * If we hit a bad address return error.
269 		 */
270 		if (!vsid)
271 			return -1;
272 		/* Make kernel text executable */
273 		if (overlaps_kernel_text(vaddr, vaddr + step))
274 			tprot &= ~HPTE_R_N;
275 
276 		/* Make kvm guest trampolines executable */
277 		if (overlaps_kvm_tmp(vaddr, vaddr + step))
278 			tprot &= ~HPTE_R_N;
279 
280 		/*
281 		 * If relocatable, check if it overlaps interrupt vectors that
282 		 * are copied down to real 0. For relocatable kernel
283 		 * (e.g. kdump case) we copy interrupt vectors down to real
284 		 * address 0. Mark that region as executable. This is
285 		 * because on p8 system with relocation on exception feature
286 		 * enabled, exceptions are raised with MMU (IR=DR=1) ON. Hence
287 		 * in order to execute the interrupt handlers in virtual
288 		 * mode the vector region need to be marked as executable.
289 		 */
290 		if ((PHYSICAL_START > MEMORY_START) &&
291 			overlaps_interrupt_vector_text(vaddr, vaddr + step))
292 				tprot &= ~HPTE_R_N;
293 
294 		hash = hpt_hash(vpn, shift, ssize);
295 		hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
296 
297 		BUG_ON(!mmu_hash_ops.hpte_insert);
298 		ret = mmu_hash_ops.hpte_insert(hpteg, vpn, paddr, tprot,
299 					       HPTE_V_BOLTED, psize, psize,
300 					       ssize);
301 
302 		if (ret < 0)
303 			break;
304 
305 #ifdef CONFIG_DEBUG_PAGEALLOC
306 		if (debug_pagealloc_enabled() &&
307 			(paddr >> PAGE_SHIFT) < linear_map_hash_count)
308 			linear_map_hash_slots[paddr >> PAGE_SHIFT] = ret | 0x80;
309 #endif /* CONFIG_DEBUG_PAGEALLOC */
310 	}
311 	return ret < 0 ? ret : 0;
312 }
313 
314 int htab_remove_mapping(unsigned long vstart, unsigned long vend,
315 		      int psize, int ssize)
316 {
317 	unsigned long vaddr;
318 	unsigned int step, shift;
319 	int rc;
320 	int ret = 0;
321 
322 	shift = mmu_psize_defs[psize].shift;
323 	step = 1 << shift;
324 
325 	if (!mmu_hash_ops.hpte_removebolted)
326 		return -ENODEV;
327 
328 	for (vaddr = vstart; vaddr < vend; vaddr += step) {
329 		rc = mmu_hash_ops.hpte_removebolted(vaddr, psize, ssize);
330 		if (rc == -ENOENT) {
331 			ret = -ENOENT;
332 			continue;
333 		}
334 		if (rc < 0)
335 			return rc;
336 	}
337 
338 	return ret;
339 }
340 
341 static bool disable_1tb_segments = false;
342 
343 static int __init parse_disable_1tb_segments(char *p)
344 {
345 	disable_1tb_segments = true;
346 	return 0;
347 }
348 early_param("disable_1tb_segments", parse_disable_1tb_segments);
349 
350 static int __init htab_dt_scan_seg_sizes(unsigned long node,
351 					 const char *uname, int depth,
352 					 void *data)
353 {
354 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
355 	const __be32 *prop;
356 	int size = 0;
357 
358 	/* We are scanning "cpu" nodes only */
359 	if (type == NULL || strcmp(type, "cpu") != 0)
360 		return 0;
361 
362 	prop = of_get_flat_dt_prop(node, "ibm,processor-segment-sizes", &size);
363 	if (prop == NULL)
364 		return 0;
365 	for (; size >= 4; size -= 4, ++prop) {
366 		if (be32_to_cpu(prop[0]) == 40) {
367 			DBG("1T segment support detected\n");
368 
369 			if (disable_1tb_segments) {
370 				DBG("1T segments disabled by command line\n");
371 				break;
372 			}
373 
374 			cur_cpu_spec->mmu_features |= MMU_FTR_1T_SEGMENT;
375 			return 1;
376 		}
377 	}
378 	cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B;
379 	return 0;
380 }
381 
382 static int __init get_idx_from_shift(unsigned int shift)
383 {
384 	int idx = -1;
385 
386 	switch (shift) {
387 	case 0xc:
388 		idx = MMU_PAGE_4K;
389 		break;
390 	case 0x10:
391 		idx = MMU_PAGE_64K;
392 		break;
393 	case 0x14:
394 		idx = MMU_PAGE_1M;
395 		break;
396 	case 0x18:
397 		idx = MMU_PAGE_16M;
398 		break;
399 	case 0x22:
400 		idx = MMU_PAGE_16G;
401 		break;
402 	}
403 	return idx;
404 }
405 
406 static int __init htab_dt_scan_page_sizes(unsigned long node,
407 					  const char *uname, int depth,
408 					  void *data)
409 {
410 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
411 	const __be32 *prop;
412 	int size = 0;
413 
414 	/* We are scanning "cpu" nodes only */
415 	if (type == NULL || strcmp(type, "cpu") != 0)
416 		return 0;
417 
418 	prop = of_get_flat_dt_prop(node, "ibm,segment-page-sizes", &size);
419 	if (!prop)
420 		return 0;
421 
422 	pr_info("Page sizes from device-tree:\n");
423 	size /= 4;
424 	cur_cpu_spec->mmu_features &= ~(MMU_FTR_16M_PAGE);
425 	while(size > 0) {
426 		unsigned int base_shift = be32_to_cpu(prop[0]);
427 		unsigned int slbenc = be32_to_cpu(prop[1]);
428 		unsigned int lpnum = be32_to_cpu(prop[2]);
429 		struct mmu_psize_def *def;
430 		int idx, base_idx;
431 
432 		size -= 3; prop += 3;
433 		base_idx = get_idx_from_shift(base_shift);
434 		if (base_idx < 0) {
435 			/* skip the pte encoding also */
436 			prop += lpnum * 2; size -= lpnum * 2;
437 			continue;
438 		}
439 		def = &mmu_psize_defs[base_idx];
440 		if (base_idx == MMU_PAGE_16M)
441 			cur_cpu_spec->mmu_features |= MMU_FTR_16M_PAGE;
442 
443 		def->shift = base_shift;
444 		if (base_shift <= 23)
445 			def->avpnm = 0;
446 		else
447 			def->avpnm = (1 << (base_shift - 23)) - 1;
448 		def->sllp = slbenc;
449 		/*
450 		 * We don't know for sure what's up with tlbiel, so
451 		 * for now we only set it for 4K and 64K pages
452 		 */
453 		if (base_idx == MMU_PAGE_4K || base_idx == MMU_PAGE_64K)
454 			def->tlbiel = 1;
455 		else
456 			def->tlbiel = 0;
457 
458 		while (size > 0 && lpnum) {
459 			unsigned int shift = be32_to_cpu(prop[0]);
460 			int penc  = be32_to_cpu(prop[1]);
461 
462 			prop += 2; size -= 2;
463 			lpnum--;
464 
465 			idx = get_idx_from_shift(shift);
466 			if (idx < 0)
467 				continue;
468 
469 			if (penc == -1)
470 				pr_err("Invalid penc for base_shift=%d "
471 				       "shift=%d\n", base_shift, shift);
472 
473 			def->penc[idx] = penc;
474 			pr_info("base_shift=%d: shift=%d, sllp=0x%04lx,"
475 				" avpnm=0x%08lx, tlbiel=%d, penc=%d\n",
476 				base_shift, shift, def->sllp,
477 				def->avpnm, def->tlbiel, def->penc[idx]);
478 		}
479 	}
480 
481 	return 1;
482 }
483 
484 #ifdef CONFIG_HUGETLB_PAGE
485 /*
486  * Scan for 16G memory blocks that have been set aside for huge pages
487  * and reserve those blocks for 16G huge pages.
488  */
489 static int __init htab_dt_scan_hugepage_blocks(unsigned long node,
490 					const char *uname, int depth,
491 					void *data) {
492 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
493 	const __be64 *addr_prop;
494 	const __be32 *page_count_prop;
495 	unsigned int expected_pages;
496 	long unsigned int phys_addr;
497 	long unsigned int block_size;
498 
499 	/* We are scanning "memory" nodes only */
500 	if (type == NULL || strcmp(type, "memory") != 0)
501 		return 0;
502 
503 	/*
504 	 * This property is the log base 2 of the number of virtual pages that
505 	 * will represent this memory block.
506 	 */
507 	page_count_prop = of_get_flat_dt_prop(node, "ibm,expected#pages", NULL);
508 	if (page_count_prop == NULL)
509 		return 0;
510 	expected_pages = (1 << be32_to_cpu(page_count_prop[0]));
511 	addr_prop = of_get_flat_dt_prop(node, "reg", NULL);
512 	if (addr_prop == NULL)
513 		return 0;
514 	phys_addr = be64_to_cpu(addr_prop[0]);
515 	block_size = be64_to_cpu(addr_prop[1]);
516 	if (block_size != (16 * GB))
517 		return 0;
518 	printk(KERN_INFO "Huge page(16GB) memory: "
519 			"addr = 0x%lX size = 0x%lX pages = %d\n",
520 			phys_addr, block_size, expected_pages);
521 	if (phys_addr + block_size * expected_pages <= memblock_end_of_DRAM()) {
522 		memblock_reserve(phys_addr, block_size * expected_pages);
523 		pseries_add_gpage(phys_addr, block_size, expected_pages);
524 	}
525 	return 0;
526 }
527 #endif /* CONFIG_HUGETLB_PAGE */
528 
529 static void mmu_psize_set_default_penc(void)
530 {
531 	int bpsize, apsize;
532 	for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
533 		for (apsize = 0; apsize < MMU_PAGE_COUNT; apsize++)
534 			mmu_psize_defs[bpsize].penc[apsize] = -1;
535 }
536 
537 #ifdef CONFIG_PPC_64K_PAGES
538 
539 static bool might_have_hea(void)
540 {
541 	/*
542 	 * The HEA ethernet adapter requires awareness of the
543 	 * GX bus. Without that awareness we can easily assume
544 	 * we will never see an HEA ethernet device.
545 	 */
546 #ifdef CONFIG_IBMEBUS
547 	return !cpu_has_feature(CPU_FTR_ARCH_207S) &&
548 		firmware_has_feature(FW_FEATURE_SPLPAR);
549 #else
550 	return false;
551 #endif
552 }
553 
554 #endif /* #ifdef CONFIG_PPC_64K_PAGES */
555 
556 static void __init htab_scan_page_sizes(void)
557 {
558 	int rc;
559 
560 	/* se the invalid penc to -1 */
561 	mmu_psize_set_default_penc();
562 
563 	/* Default to 4K pages only */
564 	memcpy(mmu_psize_defs, mmu_psize_defaults,
565 	       sizeof(mmu_psize_defaults));
566 
567 	/*
568 	 * Try to find the available page sizes in the device-tree
569 	 */
570 	rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL);
571 	if (rc == 0 && early_mmu_has_feature(MMU_FTR_16M_PAGE)) {
572 		/*
573 		 * Nothing in the device-tree, but the CPU supports 16M pages,
574 		 * so let's fallback on a known size list for 16M capable CPUs.
575 		 */
576 		memcpy(mmu_psize_defs, mmu_psize_defaults_gp,
577 		       sizeof(mmu_psize_defaults_gp));
578 	}
579 
580 #ifdef CONFIG_HUGETLB_PAGE
581 	if (!hugetlb_disabled) {
582 		/* Reserve 16G huge page memory sections for huge pages */
583 		of_scan_flat_dt(htab_dt_scan_hugepage_blocks, NULL);
584 	}
585 #endif /* CONFIG_HUGETLB_PAGE */
586 }
587 
588 /*
589  * Fill in the hpte_page_sizes[] array.
590  * We go through the mmu_psize_defs[] array looking for all the
591  * supported base/actual page size combinations.  Each combination
592  * has a unique pagesize encoding (penc) value in the low bits of
593  * the LP field of the HPTE.  For actual page sizes less than 1MB,
594  * some of the upper LP bits are used for RPN bits, meaning that
595  * we need to fill in several entries in hpte_page_sizes[].
596  *
597  * In diagrammatic form, with r = RPN bits and z = page size bits:
598  *        PTE LP     actual page size
599  *    rrrr rrrz		>=8KB
600  *    rrrr rrzz		>=16KB
601  *    rrrr rzzz		>=32KB
602  *    rrrr zzzz		>=64KB
603  *    ...
604  *
605  * The zzzz bits are implementation-specific but are chosen so that
606  * no encoding for a larger page size uses the same value in its
607  * low-order N bits as the encoding for the 2^(12+N) byte page size
608  * (if it exists).
609  */
610 static void init_hpte_page_sizes(void)
611 {
612 	long int ap, bp;
613 	long int shift, penc;
614 
615 	for (bp = 0; bp < MMU_PAGE_COUNT; ++bp) {
616 		if (!mmu_psize_defs[bp].shift)
617 			continue;	/* not a supported page size */
618 		for (ap = bp; ap < MMU_PAGE_COUNT; ++ap) {
619 			penc = mmu_psize_defs[bp].penc[ap];
620 			if (penc == -1 || !mmu_psize_defs[ap].shift)
621 				continue;
622 			shift = mmu_psize_defs[ap].shift - LP_SHIFT;
623 			if (shift <= 0)
624 				continue;	/* should never happen */
625 			/*
626 			 * For page sizes less than 1MB, this loop
627 			 * replicates the entry for all possible values
628 			 * of the rrrr bits.
629 			 */
630 			while (penc < (1 << LP_BITS)) {
631 				hpte_page_sizes[penc] = (ap << 4) | bp;
632 				penc += 1 << shift;
633 			}
634 		}
635 	}
636 }
637 
638 static void __init htab_init_page_sizes(void)
639 {
640 	init_hpte_page_sizes();
641 
642 	if (!debug_pagealloc_enabled()) {
643 		/*
644 		 * Pick a size for the linear mapping. Currently, we only
645 		 * support 16M, 1M and 4K which is the default
646 		 */
647 		if (mmu_psize_defs[MMU_PAGE_16M].shift)
648 			mmu_linear_psize = MMU_PAGE_16M;
649 		else if (mmu_psize_defs[MMU_PAGE_1M].shift)
650 			mmu_linear_psize = MMU_PAGE_1M;
651 	}
652 
653 #ifdef CONFIG_PPC_64K_PAGES
654 	/*
655 	 * Pick a size for the ordinary pages. Default is 4K, we support
656 	 * 64K for user mappings and vmalloc if supported by the processor.
657 	 * We only use 64k for ioremap if the processor
658 	 * (and firmware) support cache-inhibited large pages.
659 	 * If not, we use 4k and set mmu_ci_restrictions so that
660 	 * hash_page knows to switch processes that use cache-inhibited
661 	 * mappings to 4k pages.
662 	 */
663 	if (mmu_psize_defs[MMU_PAGE_64K].shift) {
664 		mmu_virtual_psize = MMU_PAGE_64K;
665 		mmu_vmalloc_psize = MMU_PAGE_64K;
666 		if (mmu_linear_psize == MMU_PAGE_4K)
667 			mmu_linear_psize = MMU_PAGE_64K;
668 		if (mmu_has_feature(MMU_FTR_CI_LARGE_PAGE)) {
669 			/*
670 			 * When running on pSeries using 64k pages for ioremap
671 			 * would stop us accessing the HEA ethernet. So if we
672 			 * have the chance of ever seeing one, stay at 4k.
673 			 */
674 			if (!might_have_hea())
675 				mmu_io_psize = MMU_PAGE_64K;
676 		} else
677 			mmu_ci_restrictions = 1;
678 	}
679 #endif /* CONFIG_PPC_64K_PAGES */
680 
681 #ifdef CONFIG_SPARSEMEM_VMEMMAP
682 	/*
683 	 * We try to use 16M pages for vmemmap if that is supported
684 	 * and we have at least 1G of RAM at boot
685 	 */
686 	if (mmu_psize_defs[MMU_PAGE_16M].shift &&
687 	    memblock_phys_mem_size() >= 0x40000000)
688 		mmu_vmemmap_psize = MMU_PAGE_16M;
689 	else
690 		mmu_vmemmap_psize = mmu_virtual_psize;
691 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
692 
693 	printk(KERN_DEBUG "Page orders: linear mapping = %d, "
694 	       "virtual = %d, io = %d"
695 #ifdef CONFIG_SPARSEMEM_VMEMMAP
696 	       ", vmemmap = %d"
697 #endif
698 	       "\n",
699 	       mmu_psize_defs[mmu_linear_psize].shift,
700 	       mmu_psize_defs[mmu_virtual_psize].shift,
701 	       mmu_psize_defs[mmu_io_psize].shift
702 #ifdef CONFIG_SPARSEMEM_VMEMMAP
703 	       ,mmu_psize_defs[mmu_vmemmap_psize].shift
704 #endif
705 	       );
706 }
707 
708 static int __init htab_dt_scan_pftsize(unsigned long node,
709 				       const char *uname, int depth,
710 				       void *data)
711 {
712 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
713 	const __be32 *prop;
714 
715 	/* We are scanning "cpu" nodes only */
716 	if (type == NULL || strcmp(type, "cpu") != 0)
717 		return 0;
718 
719 	prop = of_get_flat_dt_prop(node, "ibm,pft-size", NULL);
720 	if (prop != NULL) {
721 		/* pft_size[0] is the NUMA CEC cookie */
722 		ppc64_pft_size = be32_to_cpu(prop[1]);
723 		return 1;
724 	}
725 	return 0;
726 }
727 
728 unsigned htab_shift_for_mem_size(unsigned long mem_size)
729 {
730 	unsigned memshift = __ilog2(mem_size);
731 	unsigned pshift = mmu_psize_defs[mmu_virtual_psize].shift;
732 	unsigned pteg_shift;
733 
734 	/* round mem_size up to next power of 2 */
735 	if ((1UL << memshift) < mem_size)
736 		memshift += 1;
737 
738 	/* aim for 2 pages / pteg */
739 	pteg_shift = memshift - (pshift + 1);
740 
741 	/*
742 	 * 2^11 PTEGS of 128 bytes each, ie. 2^18 bytes is the minimum htab
743 	 * size permitted by the architecture.
744 	 */
745 	return max(pteg_shift + 7, 18U);
746 }
747 
748 static unsigned long __init htab_get_table_size(void)
749 {
750 	/*
751 	 * If hash size isn't already provided by the platform, we try to
752 	 * retrieve it from the device-tree. If it's not there neither, we
753 	 * calculate it now based on the total RAM size
754 	 */
755 	if (ppc64_pft_size == 0)
756 		of_scan_flat_dt(htab_dt_scan_pftsize, NULL);
757 	if (ppc64_pft_size)
758 		return 1UL << ppc64_pft_size;
759 
760 	return 1UL << htab_shift_for_mem_size(memblock_phys_mem_size());
761 }
762 
763 #ifdef CONFIG_MEMORY_HOTPLUG
764 int resize_hpt_for_hotplug(unsigned long new_mem_size)
765 {
766 	unsigned target_hpt_shift;
767 
768 	if (!mmu_hash_ops.resize_hpt)
769 		return 0;
770 
771 	target_hpt_shift = htab_shift_for_mem_size(new_mem_size);
772 
773 	/*
774 	 * To avoid lots of HPT resizes if memory size is fluctuating
775 	 * across a boundary, we deliberately have some hysterisis
776 	 * here: we immediately increase the HPT size if the target
777 	 * shift exceeds the current shift, but we won't attempt to
778 	 * reduce unless the target shift is at least 2 below the
779 	 * current shift
780 	 */
781 	if (target_hpt_shift > ppc64_pft_size ||
782 	    target_hpt_shift < ppc64_pft_size - 1)
783 		return mmu_hash_ops.resize_hpt(target_hpt_shift);
784 
785 	return 0;
786 }
787 
788 int hash__create_section_mapping(unsigned long start, unsigned long end, int nid)
789 {
790 	int rc;
791 
792 	if (end >= H_VMALLOC_START) {
793 		pr_warn("Outside the supported range\n");
794 		return -1;
795 	}
796 
797 	rc = htab_bolt_mapping(start, end, __pa(start),
798 			       pgprot_val(PAGE_KERNEL), mmu_linear_psize,
799 			       mmu_kernel_ssize);
800 
801 	if (rc < 0) {
802 		int rc2 = htab_remove_mapping(start, end, mmu_linear_psize,
803 					      mmu_kernel_ssize);
804 		BUG_ON(rc2 && (rc2 != -ENOENT));
805 	}
806 	return rc;
807 }
808 
809 int hash__remove_section_mapping(unsigned long start, unsigned long end)
810 {
811 	int rc = htab_remove_mapping(start, end, mmu_linear_psize,
812 				     mmu_kernel_ssize);
813 	WARN_ON(rc < 0);
814 	return rc;
815 }
816 #endif /* CONFIG_MEMORY_HOTPLUG */
817 
818 static void __init hash_init_partition_table(phys_addr_t hash_table,
819 					     unsigned long htab_size)
820 {
821 	mmu_partition_table_init();
822 
823 	/*
824 	 * PS field (VRMA page size) is not used for LPID 0, hence set to 0.
825 	 * For now, UPRT is 0 and we have no segment table.
826 	 */
827 	htab_size =  __ilog2(htab_size) - 18;
828 	mmu_partition_table_set_entry(0, hash_table | htab_size, 0, false);
829 	pr_info("Partition table %p\n", partition_tb);
830 }
831 
832 static void __init htab_initialize(void)
833 {
834 	unsigned long table;
835 	unsigned long pteg_count;
836 	unsigned long prot;
837 	unsigned long base = 0, size = 0;
838 	struct memblock_region *reg;
839 
840 	DBG(" -> htab_initialize()\n");
841 
842 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT)) {
843 		mmu_kernel_ssize = MMU_SEGSIZE_1T;
844 		mmu_highuser_ssize = MMU_SEGSIZE_1T;
845 		printk(KERN_INFO "Using 1TB segments\n");
846 	}
847 
848 	/*
849 	 * Calculate the required size of the htab.  We want the number of
850 	 * PTEGs to equal one half the number of real pages.
851 	 */
852 	htab_size_bytes = htab_get_table_size();
853 	pteg_count = htab_size_bytes >> 7;
854 
855 	htab_hash_mask = pteg_count - 1;
856 
857 	if (firmware_has_feature(FW_FEATURE_LPAR) ||
858 	    firmware_has_feature(FW_FEATURE_PS3_LV1)) {
859 		/* Using a hypervisor which owns the htab */
860 		htab_address = NULL;
861 		_SDR1 = 0;
862 #ifdef CONFIG_FA_DUMP
863 		/*
864 		 * If firmware assisted dump is active firmware preserves
865 		 * the contents of htab along with entire partition memory.
866 		 * Clear the htab if firmware assisted dump is active so
867 		 * that we dont end up using old mappings.
868 		 */
869 		if (is_fadump_active() && mmu_hash_ops.hpte_clear_all)
870 			mmu_hash_ops.hpte_clear_all();
871 #endif
872 	} else {
873 		unsigned long limit = MEMBLOCK_ALLOC_ANYWHERE;
874 
875 #ifdef CONFIG_PPC_CELL
876 		/*
877 		 * Cell may require the hash table down low when using the
878 		 * Axon IOMMU in order to fit the dynamic region over it, see
879 		 * comments in cell/iommu.c
880 		 */
881 		if (fdt_subnode_offset(initial_boot_params, 0, "axon") > 0) {
882 			limit = 0x80000000;
883 			pr_info("Hash table forced below 2G for Axon IOMMU\n");
884 		}
885 #endif /* CONFIG_PPC_CELL */
886 
887 		table = memblock_phys_alloc_range(htab_size_bytes,
888 						  htab_size_bytes,
889 						  0, limit);
890 		if (!table)
891 			panic("ERROR: Failed to allocate %pa bytes below %pa\n",
892 			      &htab_size_bytes, &limit);
893 
894 		DBG("Hash table allocated at %lx, size: %lx\n", table,
895 		    htab_size_bytes);
896 
897 		htab_address = __va(table);
898 
899 		/* htab absolute addr + encoded htabsize */
900 		_SDR1 = table + __ilog2(htab_size_bytes) - 18;
901 
902 		/* Initialize the HPT with no entries */
903 		memset((void *)table, 0, htab_size_bytes);
904 
905 		if (!cpu_has_feature(CPU_FTR_ARCH_300))
906 			/* Set SDR1 */
907 			mtspr(SPRN_SDR1, _SDR1);
908 		else
909 			hash_init_partition_table(table, htab_size_bytes);
910 	}
911 
912 	prot = pgprot_val(PAGE_KERNEL);
913 
914 #ifdef CONFIG_DEBUG_PAGEALLOC
915 	if (debug_pagealloc_enabled()) {
916 		linear_map_hash_count = memblock_end_of_DRAM() >> PAGE_SHIFT;
917 		linear_map_hash_slots = memblock_alloc_try_nid(
918 				linear_map_hash_count, 1, MEMBLOCK_LOW_LIMIT,
919 				ppc64_rma_size,	NUMA_NO_NODE);
920 		if (!linear_map_hash_slots)
921 			panic("%s: Failed to allocate %lu bytes max_addr=%pa\n",
922 			      __func__, linear_map_hash_count, &ppc64_rma_size);
923 	}
924 #endif /* CONFIG_DEBUG_PAGEALLOC */
925 
926 	/* create bolted the linear mapping in the hash table */
927 	for_each_memblock(memory, reg) {
928 		base = (unsigned long)__va(reg->base);
929 		size = reg->size;
930 
931 		DBG("creating mapping for region: %lx..%lx (prot: %lx)\n",
932 		    base, size, prot);
933 
934 		if ((base + size) >= H_VMALLOC_START) {
935 			pr_warn("Outside the supported range\n");
936 			continue;
937 		}
938 
939 		BUG_ON(htab_bolt_mapping(base, base + size, __pa(base),
940 				prot, mmu_linear_psize, mmu_kernel_ssize));
941 	}
942 	memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
943 
944 	/*
945 	 * If we have a memory_limit and we've allocated TCEs then we need to
946 	 * explicitly map the TCE area at the top of RAM. We also cope with the
947 	 * case that the TCEs start below memory_limit.
948 	 * tce_alloc_start/end are 16MB aligned so the mapping should work
949 	 * for either 4K or 16MB pages.
950 	 */
951 	if (tce_alloc_start) {
952 		tce_alloc_start = (unsigned long)__va(tce_alloc_start);
953 		tce_alloc_end = (unsigned long)__va(tce_alloc_end);
954 
955 		if (base + size >= tce_alloc_start)
956 			tce_alloc_start = base + size + 1;
957 
958 		BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end,
959 					 __pa(tce_alloc_start), prot,
960 					 mmu_linear_psize, mmu_kernel_ssize));
961 	}
962 
963 
964 	DBG(" <- htab_initialize()\n");
965 }
966 #undef KB
967 #undef MB
968 
969 void __init hash__early_init_devtree(void)
970 {
971 	/* Initialize segment sizes */
972 	of_scan_flat_dt(htab_dt_scan_seg_sizes, NULL);
973 
974 	/* Initialize page sizes */
975 	htab_scan_page_sizes();
976 }
977 
978 static struct hash_mm_context init_hash_mm_context;
979 void __init hash__early_init_mmu(void)
980 {
981 #ifndef CONFIG_PPC_64K_PAGES
982 	/*
983 	 * We have code in __hash_page_4K() and elsewhere, which assumes it can
984 	 * do the following:
985 	 *   new_pte |= (slot << H_PAGE_F_GIX_SHIFT) & (H_PAGE_F_SECOND | H_PAGE_F_GIX);
986 	 *
987 	 * Where the slot number is between 0-15, and values of 8-15 indicate
988 	 * the secondary bucket. For that code to work H_PAGE_F_SECOND and
989 	 * H_PAGE_F_GIX must occupy four contiguous bits in the PTE, and
990 	 * H_PAGE_F_SECOND must be placed above H_PAGE_F_GIX. Assert that here
991 	 * with a BUILD_BUG_ON().
992 	 */
993 	BUILD_BUG_ON(H_PAGE_F_SECOND != (1ul  << (H_PAGE_F_GIX_SHIFT + 3)));
994 #endif /* CONFIG_PPC_64K_PAGES */
995 
996 	htab_init_page_sizes();
997 
998 	/*
999 	 * initialize page table size
1000 	 */
1001 	__pte_frag_nr = H_PTE_FRAG_NR;
1002 	__pte_frag_size_shift = H_PTE_FRAG_SIZE_SHIFT;
1003 	__pmd_frag_nr = H_PMD_FRAG_NR;
1004 	__pmd_frag_size_shift = H_PMD_FRAG_SIZE_SHIFT;
1005 
1006 	__pte_index_size = H_PTE_INDEX_SIZE;
1007 	__pmd_index_size = H_PMD_INDEX_SIZE;
1008 	__pud_index_size = H_PUD_INDEX_SIZE;
1009 	__pgd_index_size = H_PGD_INDEX_SIZE;
1010 	__pud_cache_index = H_PUD_CACHE_INDEX;
1011 	__pte_table_size = H_PTE_TABLE_SIZE;
1012 	__pmd_table_size = H_PMD_TABLE_SIZE;
1013 	__pud_table_size = H_PUD_TABLE_SIZE;
1014 	__pgd_table_size = H_PGD_TABLE_SIZE;
1015 	/*
1016 	 * 4k use hugepd format, so for hash set then to
1017 	 * zero
1018 	 */
1019 	__pmd_val_bits = HASH_PMD_VAL_BITS;
1020 	__pud_val_bits = HASH_PUD_VAL_BITS;
1021 	__pgd_val_bits = HASH_PGD_VAL_BITS;
1022 
1023 	__kernel_virt_start = H_KERN_VIRT_START;
1024 	__vmalloc_start = H_VMALLOC_START;
1025 	__vmalloc_end = H_VMALLOC_END;
1026 	__kernel_io_start = H_KERN_IO_START;
1027 	__kernel_io_end = H_KERN_IO_END;
1028 	vmemmap = (struct page *)H_VMEMMAP_START;
1029 	ioremap_bot = IOREMAP_BASE;
1030 
1031 #ifdef CONFIG_PCI
1032 	pci_io_base = ISA_IO_BASE;
1033 #endif
1034 
1035 	/* Select appropriate backend */
1036 	if (firmware_has_feature(FW_FEATURE_PS3_LV1))
1037 		ps3_early_mm_init();
1038 	else if (firmware_has_feature(FW_FEATURE_LPAR))
1039 		hpte_init_pseries();
1040 	else if (IS_ENABLED(CONFIG_PPC_NATIVE))
1041 		hpte_init_native();
1042 
1043 	if (!mmu_hash_ops.hpte_insert)
1044 		panic("hash__early_init_mmu: No MMU hash ops defined!\n");
1045 
1046 	/*
1047 	 * Initialize the MMU Hash table and create the linear mapping
1048 	 * of memory. Has to be done before SLB initialization as this is
1049 	 * currently where the page size encoding is obtained.
1050 	 */
1051 	htab_initialize();
1052 
1053 	init_mm.context.hash_context = &init_hash_mm_context;
1054 	mm_ctx_set_slb_addr_limit(&init_mm.context, SLB_ADDR_LIMIT_DEFAULT);
1055 
1056 	pr_info("Initializing hash mmu with SLB\n");
1057 	/* Initialize SLB management */
1058 	slb_initialize();
1059 
1060 	if (cpu_has_feature(CPU_FTR_ARCH_206)
1061 			&& cpu_has_feature(CPU_FTR_HVMODE))
1062 		tlbiel_all();
1063 }
1064 
1065 #ifdef CONFIG_SMP
1066 void hash__early_init_mmu_secondary(void)
1067 {
1068 	/* Initialize hash table for that CPU */
1069 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
1070 
1071 		if (!cpu_has_feature(CPU_FTR_ARCH_300))
1072 			mtspr(SPRN_SDR1, _SDR1);
1073 		else
1074 			set_ptcr_when_no_uv(__pa(partition_tb) |
1075 					    (PATB_SIZE_SHIFT - 12));
1076 	}
1077 	/* Initialize SLB */
1078 	slb_initialize();
1079 
1080 	if (cpu_has_feature(CPU_FTR_ARCH_206)
1081 			&& cpu_has_feature(CPU_FTR_HVMODE))
1082 		tlbiel_all();
1083 }
1084 #endif /* CONFIG_SMP */
1085 
1086 /*
1087  * Called by asm hashtable.S for doing lazy icache flush
1088  */
1089 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap)
1090 {
1091 	struct page *page;
1092 
1093 	if (!pfn_valid(pte_pfn(pte)))
1094 		return pp;
1095 
1096 	page = pte_page(pte);
1097 
1098 	/* page is dirty */
1099 	if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) {
1100 		if (trap == 0x400) {
1101 			flush_dcache_icache_page(page);
1102 			set_bit(PG_arch_1, &page->flags);
1103 		} else
1104 			pp |= HPTE_R_N;
1105 	}
1106 	return pp;
1107 }
1108 
1109 #ifdef CONFIG_PPC_MM_SLICES
1110 static unsigned int get_paca_psize(unsigned long addr)
1111 {
1112 	unsigned char *psizes;
1113 	unsigned long index, mask_index;
1114 
1115 	if (addr < SLICE_LOW_TOP) {
1116 		psizes = get_paca()->mm_ctx_low_slices_psize;
1117 		index = GET_LOW_SLICE_INDEX(addr);
1118 	} else {
1119 		psizes = get_paca()->mm_ctx_high_slices_psize;
1120 		index = GET_HIGH_SLICE_INDEX(addr);
1121 	}
1122 	mask_index = index & 0x1;
1123 	return (psizes[index >> 1] >> (mask_index * 4)) & 0xF;
1124 }
1125 
1126 #else
1127 unsigned int get_paca_psize(unsigned long addr)
1128 {
1129 	return get_paca()->mm_ctx_user_psize;
1130 }
1131 #endif
1132 
1133 /*
1134  * Demote a segment to using 4k pages.
1135  * For now this makes the whole process use 4k pages.
1136  */
1137 #ifdef CONFIG_PPC_64K_PAGES
1138 void demote_segment_4k(struct mm_struct *mm, unsigned long addr)
1139 {
1140 	if (get_slice_psize(mm, addr) == MMU_PAGE_4K)
1141 		return;
1142 	slice_set_range_psize(mm, addr, 1, MMU_PAGE_4K);
1143 	copro_flush_all_slbs(mm);
1144 	if ((get_paca_psize(addr) != MMU_PAGE_4K) && (current->mm == mm)) {
1145 
1146 		copy_mm_to_paca(mm);
1147 		slb_flush_and_restore_bolted();
1148 	}
1149 }
1150 #endif /* CONFIG_PPC_64K_PAGES */
1151 
1152 #ifdef CONFIG_PPC_SUBPAGE_PROT
1153 /*
1154  * This looks up a 2-bit protection code for a 4k subpage of a 64k page.
1155  * Userspace sets the subpage permissions using the subpage_prot system call.
1156  *
1157  * Result is 0: full permissions, _PAGE_RW: read-only,
1158  * _PAGE_RWX: no access.
1159  */
1160 static int subpage_protection(struct mm_struct *mm, unsigned long ea)
1161 {
1162 	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
1163 	u32 spp = 0;
1164 	u32 **sbpm, *sbpp;
1165 
1166 	if (!spt)
1167 		return 0;
1168 
1169 	if (ea >= spt->maxaddr)
1170 		return 0;
1171 	if (ea < 0x100000000UL) {
1172 		/* addresses below 4GB use spt->low_prot */
1173 		sbpm = spt->low_prot;
1174 	} else {
1175 		sbpm = spt->protptrs[ea >> SBP_L3_SHIFT];
1176 		if (!sbpm)
1177 			return 0;
1178 	}
1179 	sbpp = sbpm[(ea >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
1180 	if (!sbpp)
1181 		return 0;
1182 	spp = sbpp[(ea >> PAGE_SHIFT) & (SBP_L1_COUNT - 1)];
1183 
1184 	/* extract 2-bit bitfield for this 4k subpage */
1185 	spp >>= 30 - 2 * ((ea >> 12) & 0xf);
1186 
1187 	/*
1188 	 * 0 -> full premission
1189 	 * 1 -> Read only
1190 	 * 2 -> no access.
1191 	 * We return the flag that need to be cleared.
1192 	 */
1193 	spp = ((spp & 2) ? _PAGE_RWX : 0) | ((spp & 1) ? _PAGE_WRITE : 0);
1194 	return spp;
1195 }
1196 
1197 #else /* CONFIG_PPC_SUBPAGE_PROT */
1198 static inline int subpage_protection(struct mm_struct *mm, unsigned long ea)
1199 {
1200 	return 0;
1201 }
1202 #endif
1203 
1204 void hash_failure_debug(unsigned long ea, unsigned long access,
1205 			unsigned long vsid, unsigned long trap,
1206 			int ssize, int psize, int lpsize, unsigned long pte)
1207 {
1208 	if (!printk_ratelimit())
1209 		return;
1210 	pr_info("mm: Hashing failure ! EA=0x%lx access=0x%lx current=%s\n",
1211 		ea, access, current->comm);
1212 	pr_info("    trap=0x%lx vsid=0x%lx ssize=%d base psize=%d psize %d pte=0x%lx\n",
1213 		trap, vsid, ssize, psize, lpsize, pte);
1214 }
1215 
1216 static void check_paca_psize(unsigned long ea, struct mm_struct *mm,
1217 			     int psize, bool user_region)
1218 {
1219 	if (user_region) {
1220 		if (psize != get_paca_psize(ea)) {
1221 			copy_mm_to_paca(mm);
1222 			slb_flush_and_restore_bolted();
1223 		}
1224 	} else if (get_paca()->vmalloc_sllp !=
1225 		   mmu_psize_defs[mmu_vmalloc_psize].sllp) {
1226 		get_paca()->vmalloc_sllp =
1227 			mmu_psize_defs[mmu_vmalloc_psize].sllp;
1228 		slb_vmalloc_update();
1229 	}
1230 }
1231 
1232 /*
1233  * Result code is:
1234  *  0 - handled
1235  *  1 - normal page fault
1236  * -1 - critical hash insertion error
1237  * -2 - access not permitted by subpage protection mechanism
1238  */
1239 int hash_page_mm(struct mm_struct *mm, unsigned long ea,
1240 		 unsigned long access, unsigned long trap,
1241 		 unsigned long flags)
1242 {
1243 	bool is_thp;
1244 	enum ctx_state prev_state = exception_enter();
1245 	pgd_t *pgdir;
1246 	unsigned long vsid;
1247 	pte_t *ptep;
1248 	unsigned hugeshift;
1249 	int rc, user_region = 0;
1250 	int psize, ssize;
1251 
1252 	DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n",
1253 		ea, access, trap);
1254 	trace_hash_fault(ea, access, trap);
1255 
1256 	/* Get region & vsid */
1257 	switch (get_region_id(ea)) {
1258 	case USER_REGION_ID:
1259 		user_region = 1;
1260 		if (! mm) {
1261 			DBG_LOW(" user region with no mm !\n");
1262 			rc = 1;
1263 			goto bail;
1264 		}
1265 		psize = get_slice_psize(mm, ea);
1266 		ssize = user_segment_size(ea);
1267 		vsid = get_user_vsid(&mm->context, ea, ssize);
1268 		break;
1269 	case VMALLOC_REGION_ID:
1270 		vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
1271 		psize = mmu_vmalloc_psize;
1272 		ssize = mmu_kernel_ssize;
1273 		break;
1274 
1275 	case IO_REGION_ID:
1276 		vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
1277 		psize = mmu_io_psize;
1278 		ssize = mmu_kernel_ssize;
1279 		break;
1280 	default:
1281 		/*
1282 		 * Not a valid range
1283 		 * Send the problem up to do_page_fault()
1284 		 */
1285 		rc = 1;
1286 		goto bail;
1287 	}
1288 	DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid);
1289 
1290 	/* Bad address. */
1291 	if (!vsid) {
1292 		DBG_LOW("Bad address!\n");
1293 		rc = 1;
1294 		goto bail;
1295 	}
1296 	/* Get pgdir */
1297 	pgdir = mm->pgd;
1298 	if (pgdir == NULL) {
1299 		rc = 1;
1300 		goto bail;
1301 	}
1302 
1303 	/* Check CPU locality */
1304 	if (user_region && mm_is_thread_local(mm))
1305 		flags |= HPTE_LOCAL_UPDATE;
1306 
1307 #ifndef CONFIG_PPC_64K_PAGES
1308 	/*
1309 	 * If we use 4K pages and our psize is not 4K, then we might
1310 	 * be hitting a special driver mapping, and need to align the
1311 	 * address before we fetch the PTE.
1312 	 *
1313 	 * It could also be a hugepage mapping, in which case this is
1314 	 * not necessary, but it's not harmful, either.
1315 	 */
1316 	if (psize != MMU_PAGE_4K)
1317 		ea &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
1318 #endif /* CONFIG_PPC_64K_PAGES */
1319 
1320 	/* Get PTE and page size from page tables */
1321 	ptep = find_linux_pte(pgdir, ea, &is_thp, &hugeshift);
1322 	if (ptep == NULL || !pte_present(*ptep)) {
1323 		DBG_LOW(" no PTE !\n");
1324 		rc = 1;
1325 		goto bail;
1326 	}
1327 
1328 	/* Add _PAGE_PRESENT to the required access perm */
1329 	access |= _PAGE_PRESENT;
1330 
1331 	/*
1332 	 * Pre-check access permissions (will be re-checked atomically
1333 	 * in __hash_page_XX but this pre-check is a fast path
1334 	 */
1335 	if (!check_pte_access(access, pte_val(*ptep))) {
1336 		DBG_LOW(" no access !\n");
1337 		rc = 1;
1338 		goto bail;
1339 	}
1340 
1341 	if (hugeshift) {
1342 		if (is_thp)
1343 			rc = __hash_page_thp(ea, access, vsid, (pmd_t *)ptep,
1344 					     trap, flags, ssize, psize);
1345 #ifdef CONFIG_HUGETLB_PAGE
1346 		else
1347 			rc = __hash_page_huge(ea, access, vsid, ptep, trap,
1348 					      flags, ssize, hugeshift, psize);
1349 #else
1350 		else {
1351 			/*
1352 			 * if we have hugeshift, and is not transhuge with
1353 			 * hugetlb disabled, something is really wrong.
1354 			 */
1355 			rc = 1;
1356 			WARN_ON(1);
1357 		}
1358 #endif
1359 		if (current->mm == mm)
1360 			check_paca_psize(ea, mm, psize, user_region);
1361 
1362 		goto bail;
1363 	}
1364 
1365 #ifndef CONFIG_PPC_64K_PAGES
1366 	DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep));
1367 #else
1368 	DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep),
1369 		pte_val(*(ptep + PTRS_PER_PTE)));
1370 #endif
1371 	/* Do actual hashing */
1372 #ifdef CONFIG_PPC_64K_PAGES
1373 	/* If H_PAGE_4K_PFN is set, make sure this is a 4k segment */
1374 	if ((pte_val(*ptep) & H_PAGE_4K_PFN) && psize == MMU_PAGE_64K) {
1375 		demote_segment_4k(mm, ea);
1376 		psize = MMU_PAGE_4K;
1377 	}
1378 
1379 	/*
1380 	 * If this PTE is non-cacheable and we have restrictions on
1381 	 * using non cacheable large pages, then we switch to 4k
1382 	 */
1383 	if (mmu_ci_restrictions && psize == MMU_PAGE_64K && pte_ci(*ptep)) {
1384 		if (user_region) {
1385 			demote_segment_4k(mm, ea);
1386 			psize = MMU_PAGE_4K;
1387 		} else if (ea < VMALLOC_END) {
1388 			/*
1389 			 * some driver did a non-cacheable mapping
1390 			 * in vmalloc space, so switch vmalloc
1391 			 * to 4k pages
1392 			 */
1393 			printk(KERN_ALERT "Reducing vmalloc segment "
1394 			       "to 4kB pages because of "
1395 			       "non-cacheable mapping\n");
1396 			psize = mmu_vmalloc_psize = MMU_PAGE_4K;
1397 			copro_flush_all_slbs(mm);
1398 		}
1399 	}
1400 
1401 #endif /* CONFIG_PPC_64K_PAGES */
1402 
1403 	if (current->mm == mm)
1404 		check_paca_psize(ea, mm, psize, user_region);
1405 
1406 #ifdef CONFIG_PPC_64K_PAGES
1407 	if (psize == MMU_PAGE_64K)
1408 		rc = __hash_page_64K(ea, access, vsid, ptep, trap,
1409 				     flags, ssize);
1410 	else
1411 #endif /* CONFIG_PPC_64K_PAGES */
1412 	{
1413 		int spp = subpage_protection(mm, ea);
1414 		if (access & spp)
1415 			rc = -2;
1416 		else
1417 			rc = __hash_page_4K(ea, access, vsid, ptep, trap,
1418 					    flags, ssize, spp);
1419 	}
1420 
1421 	/*
1422 	 * Dump some info in case of hash insertion failure, they should
1423 	 * never happen so it is really useful to know if/when they do
1424 	 */
1425 	if (rc == -1)
1426 		hash_failure_debug(ea, access, vsid, trap, ssize, psize,
1427 				   psize, pte_val(*ptep));
1428 #ifndef CONFIG_PPC_64K_PAGES
1429 	DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep));
1430 #else
1431 	DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep),
1432 		pte_val(*(ptep + PTRS_PER_PTE)));
1433 #endif
1434 	DBG_LOW(" -> rc=%d\n", rc);
1435 
1436 bail:
1437 	exception_exit(prev_state);
1438 	return rc;
1439 }
1440 EXPORT_SYMBOL_GPL(hash_page_mm);
1441 
1442 int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
1443 	      unsigned long dsisr)
1444 {
1445 	unsigned long flags = 0;
1446 	struct mm_struct *mm = current->mm;
1447 
1448 	if ((get_region_id(ea) == VMALLOC_REGION_ID) ||
1449 	    (get_region_id(ea) == IO_REGION_ID))
1450 		mm = &init_mm;
1451 
1452 	if (dsisr & DSISR_NOHPTE)
1453 		flags |= HPTE_NOHPTE_UPDATE;
1454 
1455 	return hash_page_mm(mm, ea, access, trap, flags);
1456 }
1457 EXPORT_SYMBOL_GPL(hash_page);
1458 
1459 int __hash_page(unsigned long trap, unsigned long ea, unsigned long dsisr,
1460 		unsigned long msr)
1461 {
1462 	unsigned long access = _PAGE_PRESENT | _PAGE_READ;
1463 	unsigned long flags = 0;
1464 	struct mm_struct *mm = current->mm;
1465 	unsigned int region_id = get_region_id(ea);
1466 
1467 	if ((region_id == VMALLOC_REGION_ID) || (region_id == IO_REGION_ID))
1468 		mm = &init_mm;
1469 
1470 	if (dsisr & DSISR_NOHPTE)
1471 		flags |= HPTE_NOHPTE_UPDATE;
1472 
1473 	if (dsisr & DSISR_ISSTORE)
1474 		access |= _PAGE_WRITE;
1475 	/*
1476 	 * We set _PAGE_PRIVILEGED only when
1477 	 * kernel mode access kernel space.
1478 	 *
1479 	 * _PAGE_PRIVILEGED is NOT set
1480 	 * 1) when kernel mode access user space
1481 	 * 2) user space access kernel space.
1482 	 */
1483 	access |= _PAGE_PRIVILEGED;
1484 	if ((msr & MSR_PR) || (region_id == USER_REGION_ID))
1485 		access &= ~_PAGE_PRIVILEGED;
1486 
1487 	if (trap == 0x400)
1488 		access |= _PAGE_EXEC;
1489 
1490 	return hash_page_mm(mm, ea, access, trap, flags);
1491 }
1492 
1493 #ifdef CONFIG_PPC_MM_SLICES
1494 static bool should_hash_preload(struct mm_struct *mm, unsigned long ea)
1495 {
1496 	int psize = get_slice_psize(mm, ea);
1497 
1498 	/* We only prefault standard pages for now */
1499 	if (unlikely(psize != mm_ctx_user_psize(&mm->context)))
1500 		return false;
1501 
1502 	/*
1503 	 * Don't prefault if subpage protection is enabled for the EA.
1504 	 */
1505 	if (unlikely((psize == MMU_PAGE_4K) && subpage_protection(mm, ea)))
1506 		return false;
1507 
1508 	return true;
1509 }
1510 #else
1511 static bool should_hash_preload(struct mm_struct *mm, unsigned long ea)
1512 {
1513 	return true;
1514 }
1515 #endif
1516 
1517 static void hash_preload(struct mm_struct *mm, unsigned long ea,
1518 			 bool is_exec, unsigned long trap)
1519 {
1520 	int hugepage_shift;
1521 	unsigned long vsid;
1522 	pgd_t *pgdir;
1523 	pte_t *ptep;
1524 	unsigned long flags;
1525 	int rc, ssize, update_flags = 0;
1526 	unsigned long access = _PAGE_PRESENT | _PAGE_READ | (is_exec ? _PAGE_EXEC : 0);
1527 
1528 	BUG_ON(get_region_id(ea) != USER_REGION_ID);
1529 
1530 	if (!should_hash_preload(mm, ea))
1531 		return;
1532 
1533 	DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx,"
1534 		" trap=%lx\n", mm, mm->pgd, ea, access, trap);
1535 
1536 	/* Get Linux PTE if available */
1537 	pgdir = mm->pgd;
1538 	if (pgdir == NULL)
1539 		return;
1540 
1541 	/* Get VSID */
1542 	ssize = user_segment_size(ea);
1543 	vsid = get_user_vsid(&mm->context, ea, ssize);
1544 	if (!vsid)
1545 		return;
1546 	/*
1547 	 * Hash doesn't like irqs. Walking linux page table with irq disabled
1548 	 * saves us from holding multiple locks.
1549 	 */
1550 	local_irq_save(flags);
1551 
1552 	/*
1553 	 * THP pages use update_mmu_cache_pmd. We don't do
1554 	 * hash preload there. Hence can ignore THP here
1555 	 */
1556 	ptep = find_current_mm_pte(pgdir, ea, NULL, &hugepage_shift);
1557 	if (!ptep)
1558 		goto out_exit;
1559 
1560 	WARN_ON(hugepage_shift);
1561 #ifdef CONFIG_PPC_64K_PAGES
1562 	/* If either H_PAGE_4K_PFN or cache inhibited is set (and we are on
1563 	 * a 64K kernel), then we don't preload, hash_page() will take
1564 	 * care of it once we actually try to access the page.
1565 	 * That way we don't have to duplicate all of the logic for segment
1566 	 * page size demotion here
1567 	 */
1568 	if ((pte_val(*ptep) & H_PAGE_4K_PFN) || pte_ci(*ptep))
1569 		goto out_exit;
1570 #endif /* CONFIG_PPC_64K_PAGES */
1571 
1572 	/* Is that local to this CPU ? */
1573 	if (mm_is_thread_local(mm))
1574 		update_flags |= HPTE_LOCAL_UPDATE;
1575 
1576 	/* Hash it in */
1577 #ifdef CONFIG_PPC_64K_PAGES
1578 	if (mm_ctx_user_psize(&mm->context) == MMU_PAGE_64K)
1579 		rc = __hash_page_64K(ea, access, vsid, ptep, trap,
1580 				     update_flags, ssize);
1581 	else
1582 #endif /* CONFIG_PPC_64K_PAGES */
1583 		rc = __hash_page_4K(ea, access, vsid, ptep, trap, update_flags,
1584 				    ssize, subpage_protection(mm, ea));
1585 
1586 	/* Dump some info in case of hash insertion failure, they should
1587 	 * never happen so it is really useful to know if/when they do
1588 	 */
1589 	if (rc == -1)
1590 		hash_failure_debug(ea, access, vsid, trap, ssize,
1591 				   mm_ctx_user_psize(&mm->context),
1592 				   mm_ctx_user_psize(&mm->context),
1593 				   pte_val(*ptep));
1594 out_exit:
1595 	local_irq_restore(flags);
1596 }
1597 
1598 /*
1599  * This is called at the end of handling a user page fault, when the
1600  * fault has been handled by updating a PTE in the linux page tables.
1601  * We use it to preload an HPTE into the hash table corresponding to
1602  * the updated linux PTE.
1603  *
1604  * This must always be called with the pte lock held.
1605  */
1606 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
1607 		      pte_t *ptep)
1608 {
1609 	/*
1610 	 * We don't need to worry about _PAGE_PRESENT here because we are
1611 	 * called with either mm->page_table_lock held or ptl lock held
1612 	 */
1613 	unsigned long trap;
1614 	bool is_exec;
1615 
1616 	if (radix_enabled()) {
1617 		prefetch((void *)address);
1618 		return;
1619 	}
1620 
1621 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
1622 	if (!pte_young(*ptep) || address >= TASK_SIZE)
1623 		return;
1624 
1625 	/*
1626 	 * We try to figure out if we are coming from an instruction
1627 	 * access fault and pass that down to __hash_page so we avoid
1628 	 * double-faulting on execution of fresh text. We have to test
1629 	 * for regs NULL since init will get here first thing at boot.
1630 	 *
1631 	 * We also avoid filling the hash if not coming from a fault.
1632 	 */
1633 
1634 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
1635 	switch (trap) {
1636 	case 0x300:
1637 		is_exec = false;
1638 		break;
1639 	case 0x400:
1640 		is_exec = true;
1641 		break;
1642 	default:
1643 		return;
1644 	}
1645 
1646 	hash_preload(vma->vm_mm, address, is_exec, trap);
1647 }
1648 
1649 #ifdef CONFIG_PPC_MEM_KEYS
1650 /*
1651  * Return the protection key associated with the given address and the
1652  * mm_struct.
1653  */
1654 u16 get_mm_addr_key(struct mm_struct *mm, unsigned long address)
1655 {
1656 	pte_t *ptep;
1657 	u16 pkey = 0;
1658 	unsigned long flags;
1659 
1660 	if (!mm || !mm->pgd)
1661 		return 0;
1662 
1663 	local_irq_save(flags);
1664 	ptep = find_linux_pte(mm->pgd, address, NULL, NULL);
1665 	if (ptep)
1666 		pkey = pte_to_pkey_bits(pte_val(READ_ONCE(*ptep)));
1667 	local_irq_restore(flags);
1668 
1669 	return pkey;
1670 }
1671 #endif /* CONFIG_PPC_MEM_KEYS */
1672 
1673 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1674 static inline void tm_flush_hash_page(int local)
1675 {
1676 	/*
1677 	 * Transactions are not aborted by tlbiel, only tlbie. Without, syncing a
1678 	 * page back to a block device w/PIO could pick up transactional data
1679 	 * (bad!) so we force an abort here. Before the sync the page will be
1680 	 * made read-only, which will flush_hash_page. BIG ISSUE here: if the
1681 	 * kernel uses a page from userspace without unmapping it first, it may
1682 	 * see the speculated version.
1683 	 */
1684 	if (local && cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
1685 	    MSR_TM_ACTIVE(current->thread.regs->msr)) {
1686 		tm_enable();
1687 		tm_abort(TM_CAUSE_TLBI);
1688 	}
1689 }
1690 #else
1691 static inline void tm_flush_hash_page(int local)
1692 {
1693 }
1694 #endif
1695 
1696 /*
1697  * Return the global hash slot, corresponding to the given PTE, which contains
1698  * the HPTE.
1699  */
1700 unsigned long pte_get_hash_gslot(unsigned long vpn, unsigned long shift,
1701 		int ssize, real_pte_t rpte, unsigned int subpg_index)
1702 {
1703 	unsigned long hash, gslot, hidx;
1704 
1705 	hash = hpt_hash(vpn, shift, ssize);
1706 	hidx = __rpte_to_hidx(rpte, subpg_index);
1707 	if (hidx & _PTEIDX_SECONDARY)
1708 		hash = ~hash;
1709 	gslot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1710 	gslot += hidx & _PTEIDX_GROUP_IX;
1711 	return gslot;
1712 }
1713 
1714 /*
1715  * WARNING: This is called from hash_low_64.S, if you change this prototype,
1716  *          do not forget to update the assembly call site !
1717  */
1718 void flush_hash_page(unsigned long vpn, real_pte_t pte, int psize, int ssize,
1719 		     unsigned long flags)
1720 {
1721 	unsigned long index, shift, gslot;
1722 	int local = flags & HPTE_LOCAL_UPDATE;
1723 
1724 	DBG_LOW("flush_hash_page(vpn=%016lx)\n", vpn);
1725 	pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1726 		gslot = pte_get_hash_gslot(vpn, shift, ssize, pte, index);
1727 		DBG_LOW(" sub %ld: gslot=%lx\n", index, gslot);
1728 		/*
1729 		 * We use same base page size and actual psize, because we don't
1730 		 * use these functions for hugepage
1731 		 */
1732 		mmu_hash_ops.hpte_invalidate(gslot, vpn, psize, psize,
1733 					     ssize, local);
1734 	} pte_iterate_hashed_end();
1735 
1736 	tm_flush_hash_page(local);
1737 }
1738 
1739 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1740 void flush_hash_hugepage(unsigned long vsid, unsigned long addr,
1741 			 pmd_t *pmdp, unsigned int psize, int ssize,
1742 			 unsigned long flags)
1743 {
1744 	int i, max_hpte_count, valid;
1745 	unsigned long s_addr;
1746 	unsigned char *hpte_slot_array;
1747 	unsigned long hidx, shift, vpn, hash, slot;
1748 	int local = flags & HPTE_LOCAL_UPDATE;
1749 
1750 	s_addr = addr & HPAGE_PMD_MASK;
1751 	hpte_slot_array = get_hpte_slot_array(pmdp);
1752 	/*
1753 	 * IF we try to do a HUGE PTE update after a withdraw is done.
1754 	 * we will find the below NULL. This happens when we do
1755 	 * split_huge_page_pmd
1756 	 */
1757 	if (!hpte_slot_array)
1758 		return;
1759 
1760 	if (mmu_hash_ops.hugepage_invalidate) {
1761 		mmu_hash_ops.hugepage_invalidate(vsid, s_addr, hpte_slot_array,
1762 						 psize, ssize, local);
1763 		goto tm_abort;
1764 	}
1765 	/*
1766 	 * No bluk hpte removal support, invalidate each entry
1767 	 */
1768 	shift = mmu_psize_defs[psize].shift;
1769 	max_hpte_count = HPAGE_PMD_SIZE >> shift;
1770 	for (i = 0; i < max_hpte_count; i++) {
1771 		/*
1772 		 * 8 bits per each hpte entries
1773 		 * 000| [ secondary group (one bit) | hidx (3 bits) | valid bit]
1774 		 */
1775 		valid = hpte_valid(hpte_slot_array, i);
1776 		if (!valid)
1777 			continue;
1778 		hidx =  hpte_hash_index(hpte_slot_array, i);
1779 
1780 		/* get the vpn */
1781 		addr = s_addr + (i * (1ul << shift));
1782 		vpn = hpt_vpn(addr, vsid, ssize);
1783 		hash = hpt_hash(vpn, shift, ssize);
1784 		if (hidx & _PTEIDX_SECONDARY)
1785 			hash = ~hash;
1786 
1787 		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1788 		slot += hidx & _PTEIDX_GROUP_IX;
1789 		mmu_hash_ops.hpte_invalidate(slot, vpn, psize,
1790 					     MMU_PAGE_16M, ssize, local);
1791 	}
1792 tm_abort:
1793 	tm_flush_hash_page(local);
1794 }
1795 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1796 
1797 void flush_hash_range(unsigned long number, int local)
1798 {
1799 	if (mmu_hash_ops.flush_hash_range)
1800 		mmu_hash_ops.flush_hash_range(number, local);
1801 	else {
1802 		int i;
1803 		struct ppc64_tlb_batch *batch =
1804 			this_cpu_ptr(&ppc64_tlb_batch);
1805 
1806 		for (i = 0; i < number; i++)
1807 			flush_hash_page(batch->vpn[i], batch->pte[i],
1808 					batch->psize, batch->ssize, local);
1809 	}
1810 }
1811 
1812 /*
1813  * low_hash_fault is called when we the low level hash code failed
1814  * to instert a PTE due to an hypervisor error
1815  */
1816 void low_hash_fault(struct pt_regs *regs, unsigned long address, int rc)
1817 {
1818 	enum ctx_state prev_state = exception_enter();
1819 
1820 	if (user_mode(regs)) {
1821 #ifdef CONFIG_PPC_SUBPAGE_PROT
1822 		if (rc == -2)
1823 			_exception(SIGSEGV, regs, SEGV_ACCERR, address);
1824 		else
1825 #endif
1826 			_exception(SIGBUS, regs, BUS_ADRERR, address);
1827 	} else
1828 		bad_page_fault(regs, address, SIGBUS);
1829 
1830 	exception_exit(prev_state);
1831 }
1832 
1833 long hpte_insert_repeating(unsigned long hash, unsigned long vpn,
1834 			   unsigned long pa, unsigned long rflags,
1835 			   unsigned long vflags, int psize, int ssize)
1836 {
1837 	unsigned long hpte_group;
1838 	long slot;
1839 
1840 repeat:
1841 	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1842 
1843 	/* Insert into the hash table, primary slot */
1844 	slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, vflags,
1845 					psize, psize, ssize);
1846 
1847 	/* Primary is full, try the secondary */
1848 	if (unlikely(slot == -1)) {
1849 		hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
1850 		slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags,
1851 						vflags | HPTE_V_SECONDARY,
1852 						psize, psize, ssize);
1853 		if (slot == -1) {
1854 			if (mftb() & 0x1)
1855 				hpte_group = (hash & htab_hash_mask) *
1856 						HPTES_PER_GROUP;
1857 
1858 			mmu_hash_ops.hpte_remove(hpte_group);
1859 			goto repeat;
1860 		}
1861 	}
1862 
1863 	return slot;
1864 }
1865 
1866 #ifdef CONFIG_DEBUG_PAGEALLOC
1867 static void kernel_map_linear_page(unsigned long vaddr, unsigned long lmi)
1868 {
1869 	unsigned long hash;
1870 	unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1871 	unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize);
1872 	unsigned long mode = htab_convert_pte_flags(pgprot_val(PAGE_KERNEL));
1873 	long ret;
1874 
1875 	hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize);
1876 
1877 	/* Don't create HPTE entries for bad address */
1878 	if (!vsid)
1879 		return;
1880 
1881 	ret = hpte_insert_repeating(hash, vpn, __pa(vaddr), mode,
1882 				    HPTE_V_BOLTED,
1883 				    mmu_linear_psize, mmu_kernel_ssize);
1884 
1885 	BUG_ON (ret < 0);
1886 	spin_lock(&linear_map_hash_lock);
1887 	BUG_ON(linear_map_hash_slots[lmi] & 0x80);
1888 	linear_map_hash_slots[lmi] = ret | 0x80;
1889 	spin_unlock(&linear_map_hash_lock);
1890 }
1891 
1892 static void kernel_unmap_linear_page(unsigned long vaddr, unsigned long lmi)
1893 {
1894 	unsigned long hash, hidx, slot;
1895 	unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1896 	unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize);
1897 
1898 	hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize);
1899 	spin_lock(&linear_map_hash_lock);
1900 	BUG_ON(!(linear_map_hash_slots[lmi] & 0x80));
1901 	hidx = linear_map_hash_slots[lmi] & 0x7f;
1902 	linear_map_hash_slots[lmi] = 0;
1903 	spin_unlock(&linear_map_hash_lock);
1904 	if (hidx & _PTEIDX_SECONDARY)
1905 		hash = ~hash;
1906 	slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1907 	slot += hidx & _PTEIDX_GROUP_IX;
1908 	mmu_hash_ops.hpte_invalidate(slot, vpn, mmu_linear_psize,
1909 				     mmu_linear_psize,
1910 				     mmu_kernel_ssize, 0);
1911 }
1912 
1913 void __kernel_map_pages(struct page *page, int numpages, int enable)
1914 {
1915 	unsigned long flags, vaddr, lmi;
1916 	int i;
1917 
1918 	local_irq_save(flags);
1919 	for (i = 0; i < numpages; i++, page++) {
1920 		vaddr = (unsigned long)page_address(page);
1921 		lmi = __pa(vaddr) >> PAGE_SHIFT;
1922 		if (lmi >= linear_map_hash_count)
1923 			continue;
1924 		if (enable)
1925 			kernel_map_linear_page(vaddr, lmi);
1926 		else
1927 			kernel_unmap_linear_page(vaddr, lmi);
1928 	}
1929 	local_irq_restore(flags);
1930 }
1931 #endif /* CONFIG_DEBUG_PAGEALLOC */
1932 
1933 void hash__setup_initial_memory_limit(phys_addr_t first_memblock_base,
1934 				phys_addr_t first_memblock_size)
1935 {
1936 	/*
1937 	 * We don't currently support the first MEMBLOCK not mapping 0
1938 	 * physical on those processors
1939 	 */
1940 	BUG_ON(first_memblock_base != 0);
1941 
1942 	/*
1943 	 * On virtualized systems the first entry is our RMA region aka VRMA,
1944 	 * non-virtualized 64-bit hash MMU systems don't have a limitation
1945 	 * on real mode access.
1946 	 *
1947 	 * For guests on platforms before POWER9, we clamp the it limit to 1G
1948 	 * to avoid some funky things such as RTAS bugs etc...
1949 	 *
1950 	 * On POWER9 we limit to 1TB in case the host erroneously told us that
1951 	 * the RMA was >1TB. Effective address bits 0:23 are treated as zero
1952 	 * (meaning the access is aliased to zero i.e. addr = addr % 1TB)
1953 	 * for virtual real mode addressing and so it doesn't make sense to
1954 	 * have an area larger than 1TB as it can't be addressed.
1955 	 */
1956 	if (!early_cpu_has_feature(CPU_FTR_HVMODE)) {
1957 		ppc64_rma_size = first_memblock_size;
1958 		if (!early_cpu_has_feature(CPU_FTR_ARCH_300))
1959 			ppc64_rma_size = min_t(u64, ppc64_rma_size, 0x40000000);
1960 		else
1961 			ppc64_rma_size = min_t(u64, ppc64_rma_size,
1962 					       1UL << SID_SHIFT_1T);
1963 
1964 		/* Finally limit subsequent allocations */
1965 		memblock_set_current_limit(ppc64_rma_size);
1966 	} else {
1967 		ppc64_rma_size = ULONG_MAX;
1968 	}
1969 }
1970 
1971 #ifdef CONFIG_DEBUG_FS
1972 
1973 static int hpt_order_get(void *data, u64 *val)
1974 {
1975 	*val = ppc64_pft_size;
1976 	return 0;
1977 }
1978 
1979 static int hpt_order_set(void *data, u64 val)
1980 {
1981 	int ret;
1982 
1983 	if (!mmu_hash_ops.resize_hpt)
1984 		return -ENODEV;
1985 
1986 	cpus_read_lock();
1987 	ret = mmu_hash_ops.resize_hpt(val);
1988 	cpus_read_unlock();
1989 
1990 	return ret;
1991 }
1992 
1993 DEFINE_DEBUGFS_ATTRIBUTE(fops_hpt_order, hpt_order_get, hpt_order_set, "%llu\n");
1994 
1995 static int __init hash64_debugfs(void)
1996 {
1997 	if (!debugfs_create_file_unsafe("hpt_order", 0600, powerpc_debugfs_root,
1998 					NULL, &fops_hpt_order)) {
1999 		pr_err("lpar: unable to create hpt_order debugsfs file\n");
2000 	}
2001 
2002 	return 0;
2003 }
2004 machine_device_initcall(pseries, hash64_debugfs);
2005 #endif /* CONFIG_DEBUG_FS */
2006 
2007 void __init print_system_hash_info(void)
2008 {
2009 	pr_info("ppc64_pft_size    = 0x%llx\n", ppc64_pft_size);
2010 
2011 	if (htab_hash_mask)
2012 		pr_info("htab_hash_mask    = 0x%lx\n", htab_hash_mask);
2013 }
2014