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 #include <linux/pgtable.h>
39 
40 #include <asm/debugfs.h>
41 #include <asm/processor.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 #include "internal.h"
70 
71 
72 #ifdef DEBUG
73 #define DBG(fmt...) udbg_printf(fmt)
74 #else
75 #define DBG(fmt...)
76 #endif
77 
78 #ifdef DEBUG_LOW
79 #define DBG_LOW(fmt...) udbg_printf(fmt)
80 #else
81 #define DBG_LOW(fmt...)
82 #endif
83 
84 #define KB (1024)
85 #define MB (1024*KB)
86 #define GB (1024L*MB)
87 
88 /*
89  * Note:  pte   --> Linux PTE
90  *        HPTE  --> PowerPC Hashed Page Table Entry
91  *
92  * Execution context:
93  *   htab_initialize is called with the MMU off (of course), but
94  *   the kernel has been copied down to zero so it can directly
95  *   reference global data.  At this point it is very difficult
96  *   to print debug info.
97  *
98  */
99 
100 static unsigned long _SDR1;
101 struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
102 EXPORT_SYMBOL_GPL(mmu_psize_defs);
103 
104 u8 hpte_page_sizes[1 << LP_BITS];
105 EXPORT_SYMBOL_GPL(hpte_page_sizes);
106 
107 struct hash_pte *htab_address;
108 unsigned long htab_size_bytes;
109 unsigned long htab_hash_mask;
110 EXPORT_SYMBOL_GPL(htab_hash_mask);
111 int mmu_linear_psize = MMU_PAGE_4K;
112 EXPORT_SYMBOL_GPL(mmu_linear_psize);
113 int mmu_virtual_psize = MMU_PAGE_4K;
114 int mmu_vmalloc_psize = MMU_PAGE_4K;
115 #ifdef CONFIG_SPARSEMEM_VMEMMAP
116 int mmu_vmemmap_psize = MMU_PAGE_4K;
117 #endif
118 int mmu_io_psize = MMU_PAGE_4K;
119 int mmu_kernel_ssize = MMU_SEGSIZE_256M;
120 EXPORT_SYMBOL_GPL(mmu_kernel_ssize);
121 int mmu_highuser_ssize = MMU_SEGSIZE_256M;
122 u16 mmu_slb_size = 64;
123 EXPORT_SYMBOL_GPL(mmu_slb_size);
124 #ifdef CONFIG_PPC_64K_PAGES
125 int mmu_ci_restrictions;
126 #endif
127 #ifdef CONFIG_DEBUG_PAGEALLOC
128 static u8 *linear_map_hash_slots;
129 static unsigned long linear_map_hash_count;
130 static DEFINE_SPINLOCK(linear_map_hash_lock);
131 #endif /* CONFIG_DEBUG_PAGEALLOC */
132 struct mmu_hash_ops mmu_hash_ops;
133 EXPORT_SYMBOL(mmu_hash_ops);
134 
135 /*
136  * These are definitions of page sizes arrays to be used when none
137  * is provided by the firmware.
138  */
139 
140 /*
141  * Fallback (4k pages only)
142  */
143 static struct mmu_psize_def mmu_psize_defaults[] = {
144 	[MMU_PAGE_4K] = {
145 		.shift	= 12,
146 		.sllp	= 0,
147 		.penc   = {[MMU_PAGE_4K] = 0, [1 ... MMU_PAGE_COUNT - 1] = -1},
148 		.avpnm	= 0,
149 		.tlbiel = 0,
150 	},
151 };
152 
153 /*
154  * POWER4, GPUL, POWER5
155  *
156  * Support for 16Mb large pages
157  */
158 static struct mmu_psize_def mmu_psize_defaults_gp[] = {
159 	[MMU_PAGE_4K] = {
160 		.shift	= 12,
161 		.sllp	= 0,
162 		.penc   = {[MMU_PAGE_4K] = 0, [1 ... MMU_PAGE_COUNT - 1] = -1},
163 		.avpnm	= 0,
164 		.tlbiel = 1,
165 	},
166 	[MMU_PAGE_16M] = {
167 		.shift	= 24,
168 		.sllp	= SLB_VSID_L,
169 		.penc   = {[0 ... MMU_PAGE_16M - 1] = -1, [MMU_PAGE_16M] = 0,
170 			    [MMU_PAGE_16M + 1 ... MMU_PAGE_COUNT - 1] = -1 },
171 		.avpnm	= 0x1UL,
172 		.tlbiel = 0,
173 	},
174 };
175 
176 /*
177  * 'R' and 'C' update notes:
178  *  - Under pHyp or KVM, the updatepp path will not set C, thus it *will*
179  *     create writeable HPTEs without C set, because the hcall H_PROTECT
180  *     that we use in that case will not update C
181  *  - The above is however not a problem, because we also don't do that
182  *     fancy "no flush" variant of eviction and we use H_REMOVE which will
183  *     do the right thing and thus we don't have the race I described earlier
184  *
185  *    - Under bare metal,  we do have the race, so we need R and C set
186  *    - We make sure R is always set and never lost
187  *    - C is _PAGE_DIRTY, and *should* always be set for a writeable mapping
188  */
189 unsigned long htab_convert_pte_flags(unsigned long pteflags)
190 {
191 	unsigned long rflags = 0;
192 
193 	/* _PAGE_EXEC -> NOEXEC */
194 	if ((pteflags & _PAGE_EXEC) == 0)
195 		rflags |= HPTE_R_N;
196 	/*
197 	 * PPP bits:
198 	 * Linux uses slb key 0 for kernel and 1 for user.
199 	 * kernel RW areas are mapped with PPP=0b000
200 	 * User area is mapped with PPP=0b010 for read/write
201 	 * or PPP=0b011 for read-only (including writeable but clean pages).
202 	 */
203 	if (pteflags & _PAGE_PRIVILEGED) {
204 		/*
205 		 * Kernel read only mapped with ppp bits 0b110
206 		 */
207 		if (!(pteflags & _PAGE_WRITE)) {
208 			if (mmu_has_feature(MMU_FTR_KERNEL_RO))
209 				rflags |= (HPTE_R_PP0 | 0x2);
210 			else
211 				rflags |= 0x3;
212 		}
213 	} else {
214 		if (pteflags & _PAGE_RWX)
215 			rflags |= 0x2;
216 		if (!((pteflags & _PAGE_WRITE) && (pteflags & _PAGE_DIRTY)))
217 			rflags |= 0x1;
218 	}
219 	/*
220 	 * We can't allow hardware to update hpte bits. Hence always
221 	 * set 'R' bit and set 'C' if it is a write fault
222 	 */
223 	rflags |=  HPTE_R_R;
224 
225 	if (pteflags & _PAGE_DIRTY)
226 		rflags |= HPTE_R_C;
227 	/*
228 	 * Add in WIG bits
229 	 */
230 
231 	if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_TOLERANT)
232 		rflags |= HPTE_R_I;
233 	else if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_NON_IDEMPOTENT)
234 		rflags |= (HPTE_R_I | HPTE_R_G);
235 	else if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_SAO)
236 		rflags |= (HPTE_R_W | HPTE_R_I | HPTE_R_M);
237 	else
238 		/*
239 		 * Add memory coherence if cache inhibited is not set
240 		 */
241 		rflags |= HPTE_R_M;
242 
243 	rflags |= pte_to_hpte_pkey_bits(pteflags);
244 	return rflags;
245 }
246 
247 int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
248 		      unsigned long pstart, unsigned long prot,
249 		      int psize, int ssize)
250 {
251 	unsigned long vaddr, paddr;
252 	unsigned int step, shift;
253 	int ret = 0;
254 
255 	shift = mmu_psize_defs[psize].shift;
256 	step = 1 << shift;
257 
258 	prot = htab_convert_pte_flags(prot);
259 
260 	DBG("htab_bolt_mapping(%lx..%lx -> %lx (%lx,%d,%d)\n",
261 	    vstart, vend, pstart, prot, psize, ssize);
262 
263 	for (vaddr = vstart, paddr = pstart; vaddr < vend;
264 	     vaddr += step, paddr += step) {
265 		unsigned long hash, hpteg;
266 		unsigned long vsid = get_kernel_vsid(vaddr, ssize);
267 		unsigned long vpn  = hpt_vpn(vaddr, vsid, ssize);
268 		unsigned long tprot = prot;
269 		bool secondary_hash = false;
270 
271 		/*
272 		 * If we hit a bad address return error.
273 		 */
274 		if (!vsid)
275 			return -1;
276 		/* Make kernel text executable */
277 		if (overlaps_kernel_text(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 repeat:
299 		ret = mmu_hash_ops.hpte_insert(hpteg, vpn, paddr, tprot,
300 					       HPTE_V_BOLTED, psize, psize,
301 					       ssize);
302 		if (ret == -1) {
303 			/*
304 			 * Try to to keep bolted entries in primary.
305 			 * Remove non bolted entries and try insert again
306 			 */
307 			ret = mmu_hash_ops.hpte_remove(hpteg);
308 			if (ret != -1)
309 				ret = mmu_hash_ops.hpte_insert(hpteg, vpn, paddr, tprot,
310 							       HPTE_V_BOLTED, psize, psize,
311 							       ssize);
312 			if (ret == -1 && !secondary_hash) {
313 				secondary_hash = true;
314 				hpteg = ((~hash & htab_hash_mask) * HPTES_PER_GROUP);
315 				goto repeat;
316 			}
317 		}
318 
319 		if (ret < 0)
320 			break;
321 
322 		cond_resched();
323 #ifdef CONFIG_DEBUG_PAGEALLOC
324 		if (debug_pagealloc_enabled() &&
325 			(paddr >> PAGE_SHIFT) < linear_map_hash_count)
326 			linear_map_hash_slots[paddr >> PAGE_SHIFT] = ret | 0x80;
327 #endif /* CONFIG_DEBUG_PAGEALLOC */
328 	}
329 	return ret < 0 ? ret : 0;
330 }
331 
332 int htab_remove_mapping(unsigned long vstart, unsigned long vend,
333 		      int psize, int ssize)
334 {
335 	unsigned long vaddr;
336 	unsigned int step, shift;
337 	int rc;
338 	int ret = 0;
339 
340 	shift = mmu_psize_defs[psize].shift;
341 	step = 1 << shift;
342 
343 	if (!mmu_hash_ops.hpte_removebolted)
344 		return -ENODEV;
345 
346 	for (vaddr = vstart; vaddr < vend; vaddr += step) {
347 		rc = mmu_hash_ops.hpte_removebolted(vaddr, psize, ssize);
348 		if (rc == -ENOENT) {
349 			ret = -ENOENT;
350 			continue;
351 		}
352 		if (rc < 0)
353 			return rc;
354 	}
355 
356 	return ret;
357 }
358 
359 static bool disable_1tb_segments = false;
360 
361 static int __init parse_disable_1tb_segments(char *p)
362 {
363 	disable_1tb_segments = true;
364 	return 0;
365 }
366 early_param("disable_1tb_segments", parse_disable_1tb_segments);
367 
368 static int __init htab_dt_scan_seg_sizes(unsigned long node,
369 					 const char *uname, int depth,
370 					 void *data)
371 {
372 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
373 	const __be32 *prop;
374 	int size = 0;
375 
376 	/* We are scanning "cpu" nodes only */
377 	if (type == NULL || strcmp(type, "cpu") != 0)
378 		return 0;
379 
380 	prop = of_get_flat_dt_prop(node, "ibm,processor-segment-sizes", &size);
381 	if (prop == NULL)
382 		return 0;
383 	for (; size >= 4; size -= 4, ++prop) {
384 		if (be32_to_cpu(prop[0]) == 40) {
385 			DBG("1T segment support detected\n");
386 
387 			if (disable_1tb_segments) {
388 				DBG("1T segments disabled by command line\n");
389 				break;
390 			}
391 
392 			cur_cpu_spec->mmu_features |= MMU_FTR_1T_SEGMENT;
393 			return 1;
394 		}
395 	}
396 	cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B;
397 	return 0;
398 }
399 
400 static int __init get_idx_from_shift(unsigned int shift)
401 {
402 	int idx = -1;
403 
404 	switch (shift) {
405 	case 0xc:
406 		idx = MMU_PAGE_4K;
407 		break;
408 	case 0x10:
409 		idx = MMU_PAGE_64K;
410 		break;
411 	case 0x14:
412 		idx = MMU_PAGE_1M;
413 		break;
414 	case 0x18:
415 		idx = MMU_PAGE_16M;
416 		break;
417 	case 0x22:
418 		idx = MMU_PAGE_16G;
419 		break;
420 	}
421 	return idx;
422 }
423 
424 static int __init htab_dt_scan_page_sizes(unsigned long node,
425 					  const char *uname, int depth,
426 					  void *data)
427 {
428 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
429 	const __be32 *prop;
430 	int size = 0;
431 
432 	/* We are scanning "cpu" nodes only */
433 	if (type == NULL || strcmp(type, "cpu") != 0)
434 		return 0;
435 
436 	prop = of_get_flat_dt_prop(node, "ibm,segment-page-sizes", &size);
437 	if (!prop)
438 		return 0;
439 
440 	pr_info("Page sizes from device-tree:\n");
441 	size /= 4;
442 	cur_cpu_spec->mmu_features &= ~(MMU_FTR_16M_PAGE);
443 	while(size > 0) {
444 		unsigned int base_shift = be32_to_cpu(prop[0]);
445 		unsigned int slbenc = be32_to_cpu(prop[1]);
446 		unsigned int lpnum = be32_to_cpu(prop[2]);
447 		struct mmu_psize_def *def;
448 		int idx, base_idx;
449 
450 		size -= 3; prop += 3;
451 		base_idx = get_idx_from_shift(base_shift);
452 		if (base_idx < 0) {
453 			/* skip the pte encoding also */
454 			prop += lpnum * 2; size -= lpnum * 2;
455 			continue;
456 		}
457 		def = &mmu_psize_defs[base_idx];
458 		if (base_idx == MMU_PAGE_16M)
459 			cur_cpu_spec->mmu_features |= MMU_FTR_16M_PAGE;
460 
461 		def->shift = base_shift;
462 		if (base_shift <= 23)
463 			def->avpnm = 0;
464 		else
465 			def->avpnm = (1 << (base_shift - 23)) - 1;
466 		def->sllp = slbenc;
467 		/*
468 		 * We don't know for sure what's up with tlbiel, so
469 		 * for now we only set it for 4K and 64K pages
470 		 */
471 		if (base_idx == MMU_PAGE_4K || base_idx == MMU_PAGE_64K)
472 			def->tlbiel = 1;
473 		else
474 			def->tlbiel = 0;
475 
476 		while (size > 0 && lpnum) {
477 			unsigned int shift = be32_to_cpu(prop[0]);
478 			int penc  = be32_to_cpu(prop[1]);
479 
480 			prop += 2; size -= 2;
481 			lpnum--;
482 
483 			idx = get_idx_from_shift(shift);
484 			if (idx < 0)
485 				continue;
486 
487 			if (penc == -1)
488 				pr_err("Invalid penc for base_shift=%d "
489 				       "shift=%d\n", base_shift, shift);
490 
491 			def->penc[idx] = penc;
492 			pr_info("base_shift=%d: shift=%d, sllp=0x%04lx,"
493 				" avpnm=0x%08lx, tlbiel=%d, penc=%d\n",
494 				base_shift, shift, def->sllp,
495 				def->avpnm, def->tlbiel, def->penc[idx]);
496 		}
497 	}
498 
499 	return 1;
500 }
501 
502 #ifdef CONFIG_HUGETLB_PAGE
503 /*
504  * Scan for 16G memory blocks that have been set aside for huge pages
505  * and reserve those blocks for 16G huge pages.
506  */
507 static int __init htab_dt_scan_hugepage_blocks(unsigned long node,
508 					const char *uname, int depth,
509 					void *data) {
510 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
511 	const __be64 *addr_prop;
512 	const __be32 *page_count_prop;
513 	unsigned int expected_pages;
514 	long unsigned int phys_addr;
515 	long unsigned int block_size;
516 
517 	/* We are scanning "memory" nodes only */
518 	if (type == NULL || strcmp(type, "memory") != 0)
519 		return 0;
520 
521 	/*
522 	 * This property is the log base 2 of the number of virtual pages that
523 	 * will represent this memory block.
524 	 */
525 	page_count_prop = of_get_flat_dt_prop(node, "ibm,expected#pages", NULL);
526 	if (page_count_prop == NULL)
527 		return 0;
528 	expected_pages = (1 << be32_to_cpu(page_count_prop[0]));
529 	addr_prop = of_get_flat_dt_prop(node, "reg", NULL);
530 	if (addr_prop == NULL)
531 		return 0;
532 	phys_addr = be64_to_cpu(addr_prop[0]);
533 	block_size = be64_to_cpu(addr_prop[1]);
534 	if (block_size != (16 * GB))
535 		return 0;
536 	printk(KERN_INFO "Huge page(16GB) memory: "
537 			"addr = 0x%lX size = 0x%lX pages = %d\n",
538 			phys_addr, block_size, expected_pages);
539 	if (phys_addr + block_size * expected_pages <= memblock_end_of_DRAM()) {
540 		memblock_reserve(phys_addr, block_size * expected_pages);
541 		pseries_add_gpage(phys_addr, block_size, expected_pages);
542 	}
543 	return 0;
544 }
545 #endif /* CONFIG_HUGETLB_PAGE */
546 
547 static void mmu_psize_set_default_penc(void)
548 {
549 	int bpsize, apsize;
550 	for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
551 		for (apsize = 0; apsize < MMU_PAGE_COUNT; apsize++)
552 			mmu_psize_defs[bpsize].penc[apsize] = -1;
553 }
554 
555 #ifdef CONFIG_PPC_64K_PAGES
556 
557 static bool might_have_hea(void)
558 {
559 	/*
560 	 * The HEA ethernet adapter requires awareness of the
561 	 * GX bus. Without that awareness we can easily assume
562 	 * we will never see an HEA ethernet device.
563 	 */
564 #ifdef CONFIG_IBMEBUS
565 	return !cpu_has_feature(CPU_FTR_ARCH_207S) &&
566 		firmware_has_feature(FW_FEATURE_SPLPAR);
567 #else
568 	return false;
569 #endif
570 }
571 
572 #endif /* #ifdef CONFIG_PPC_64K_PAGES */
573 
574 static void __init htab_scan_page_sizes(void)
575 {
576 	int rc;
577 
578 	/* se the invalid penc to -1 */
579 	mmu_psize_set_default_penc();
580 
581 	/* Default to 4K pages only */
582 	memcpy(mmu_psize_defs, mmu_psize_defaults,
583 	       sizeof(mmu_psize_defaults));
584 
585 	/*
586 	 * Try to find the available page sizes in the device-tree
587 	 */
588 	rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL);
589 	if (rc == 0 && early_mmu_has_feature(MMU_FTR_16M_PAGE)) {
590 		/*
591 		 * Nothing in the device-tree, but the CPU supports 16M pages,
592 		 * so let's fallback on a known size list for 16M capable CPUs.
593 		 */
594 		memcpy(mmu_psize_defs, mmu_psize_defaults_gp,
595 		       sizeof(mmu_psize_defaults_gp));
596 	}
597 
598 #ifdef CONFIG_HUGETLB_PAGE
599 	if (!hugetlb_disabled) {
600 		/* Reserve 16G huge page memory sections for huge pages */
601 		of_scan_flat_dt(htab_dt_scan_hugepage_blocks, NULL);
602 	}
603 #endif /* CONFIG_HUGETLB_PAGE */
604 }
605 
606 /*
607  * Fill in the hpte_page_sizes[] array.
608  * We go through the mmu_psize_defs[] array looking for all the
609  * supported base/actual page size combinations.  Each combination
610  * has a unique pagesize encoding (penc) value in the low bits of
611  * the LP field of the HPTE.  For actual page sizes less than 1MB,
612  * some of the upper LP bits are used for RPN bits, meaning that
613  * we need to fill in several entries in hpte_page_sizes[].
614  *
615  * In diagrammatic form, with r = RPN bits and z = page size bits:
616  *        PTE LP     actual page size
617  *    rrrr rrrz		>=8KB
618  *    rrrr rrzz		>=16KB
619  *    rrrr rzzz		>=32KB
620  *    rrrr zzzz		>=64KB
621  *    ...
622  *
623  * The zzzz bits are implementation-specific but are chosen so that
624  * no encoding for a larger page size uses the same value in its
625  * low-order N bits as the encoding for the 2^(12+N) byte page size
626  * (if it exists).
627  */
628 static void init_hpte_page_sizes(void)
629 {
630 	long int ap, bp;
631 	long int shift, penc;
632 
633 	for (bp = 0; bp < MMU_PAGE_COUNT; ++bp) {
634 		if (!mmu_psize_defs[bp].shift)
635 			continue;	/* not a supported page size */
636 		for (ap = bp; ap < MMU_PAGE_COUNT; ++ap) {
637 			penc = mmu_psize_defs[bp].penc[ap];
638 			if (penc == -1 || !mmu_psize_defs[ap].shift)
639 				continue;
640 			shift = mmu_psize_defs[ap].shift - LP_SHIFT;
641 			if (shift <= 0)
642 				continue;	/* should never happen */
643 			/*
644 			 * For page sizes less than 1MB, this loop
645 			 * replicates the entry for all possible values
646 			 * of the rrrr bits.
647 			 */
648 			while (penc < (1 << LP_BITS)) {
649 				hpte_page_sizes[penc] = (ap << 4) | bp;
650 				penc += 1 << shift;
651 			}
652 		}
653 	}
654 }
655 
656 static void __init htab_init_page_sizes(void)
657 {
658 	bool aligned = true;
659 	init_hpte_page_sizes();
660 
661 	if (!debug_pagealloc_enabled()) {
662 		/*
663 		 * Pick a size for the linear mapping. Currently, we only
664 		 * support 16M, 1M and 4K which is the default
665 		 */
666 		if (IS_ENABLED(STRICT_KERNEL_RWX) &&
667 		    (unsigned long)_stext % 0x1000000) {
668 			if (mmu_psize_defs[MMU_PAGE_16M].shift)
669 				pr_warn("Kernel not 16M aligned, "
670 					"disabling 16M linear map alignment");
671 			aligned = false;
672 		}
673 
674 		if (mmu_psize_defs[MMU_PAGE_16M].shift && aligned)
675 			mmu_linear_psize = MMU_PAGE_16M;
676 		else if (mmu_psize_defs[MMU_PAGE_1M].shift)
677 			mmu_linear_psize = MMU_PAGE_1M;
678 	}
679 
680 #ifdef CONFIG_PPC_64K_PAGES
681 	/*
682 	 * Pick a size for the ordinary pages. Default is 4K, we support
683 	 * 64K for user mappings and vmalloc if supported by the processor.
684 	 * We only use 64k for ioremap if the processor
685 	 * (and firmware) support cache-inhibited large pages.
686 	 * If not, we use 4k and set mmu_ci_restrictions so that
687 	 * hash_page knows to switch processes that use cache-inhibited
688 	 * mappings to 4k pages.
689 	 */
690 	if (mmu_psize_defs[MMU_PAGE_64K].shift) {
691 		mmu_virtual_psize = MMU_PAGE_64K;
692 		mmu_vmalloc_psize = MMU_PAGE_64K;
693 		if (mmu_linear_psize == MMU_PAGE_4K)
694 			mmu_linear_psize = MMU_PAGE_64K;
695 		if (mmu_has_feature(MMU_FTR_CI_LARGE_PAGE)) {
696 			/*
697 			 * When running on pSeries using 64k pages for ioremap
698 			 * would stop us accessing the HEA ethernet. So if we
699 			 * have the chance of ever seeing one, stay at 4k.
700 			 */
701 			if (!might_have_hea())
702 				mmu_io_psize = MMU_PAGE_64K;
703 		} else
704 			mmu_ci_restrictions = 1;
705 	}
706 #endif /* CONFIG_PPC_64K_PAGES */
707 
708 #ifdef CONFIG_SPARSEMEM_VMEMMAP
709 	/*
710 	 * We try to use 16M pages for vmemmap if that is supported
711 	 * and we have at least 1G of RAM at boot
712 	 */
713 	if (mmu_psize_defs[MMU_PAGE_16M].shift &&
714 	    memblock_phys_mem_size() >= 0x40000000)
715 		mmu_vmemmap_psize = MMU_PAGE_16M;
716 	else
717 		mmu_vmemmap_psize = mmu_virtual_psize;
718 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
719 
720 	printk(KERN_DEBUG "Page orders: linear mapping = %d, "
721 	       "virtual = %d, io = %d"
722 #ifdef CONFIG_SPARSEMEM_VMEMMAP
723 	       ", vmemmap = %d"
724 #endif
725 	       "\n",
726 	       mmu_psize_defs[mmu_linear_psize].shift,
727 	       mmu_psize_defs[mmu_virtual_psize].shift,
728 	       mmu_psize_defs[mmu_io_psize].shift
729 #ifdef CONFIG_SPARSEMEM_VMEMMAP
730 	       ,mmu_psize_defs[mmu_vmemmap_psize].shift
731 #endif
732 	       );
733 }
734 
735 static int __init htab_dt_scan_pftsize(unsigned long node,
736 				       const char *uname, int depth,
737 				       void *data)
738 {
739 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
740 	const __be32 *prop;
741 
742 	/* We are scanning "cpu" nodes only */
743 	if (type == NULL || strcmp(type, "cpu") != 0)
744 		return 0;
745 
746 	prop = of_get_flat_dt_prop(node, "ibm,pft-size", NULL);
747 	if (prop != NULL) {
748 		/* pft_size[0] is the NUMA CEC cookie */
749 		ppc64_pft_size = be32_to_cpu(prop[1]);
750 		return 1;
751 	}
752 	return 0;
753 }
754 
755 unsigned htab_shift_for_mem_size(unsigned long mem_size)
756 {
757 	unsigned memshift = __ilog2(mem_size);
758 	unsigned pshift = mmu_psize_defs[mmu_virtual_psize].shift;
759 	unsigned pteg_shift;
760 
761 	/* round mem_size up to next power of 2 */
762 	if ((1UL << memshift) < mem_size)
763 		memshift += 1;
764 
765 	/* aim for 2 pages / pteg */
766 	pteg_shift = memshift - (pshift + 1);
767 
768 	/*
769 	 * 2^11 PTEGS of 128 bytes each, ie. 2^18 bytes is the minimum htab
770 	 * size permitted by the architecture.
771 	 */
772 	return max(pteg_shift + 7, 18U);
773 }
774 
775 static unsigned long __init htab_get_table_size(void)
776 {
777 	/*
778 	 * If hash size isn't already provided by the platform, we try to
779 	 * retrieve it from the device-tree. If it's not there neither, we
780 	 * calculate it now based on the total RAM size
781 	 */
782 	if (ppc64_pft_size == 0)
783 		of_scan_flat_dt(htab_dt_scan_pftsize, NULL);
784 	if (ppc64_pft_size)
785 		return 1UL << ppc64_pft_size;
786 
787 	return 1UL << htab_shift_for_mem_size(memblock_phys_mem_size());
788 }
789 
790 #ifdef CONFIG_MEMORY_HOTPLUG
791 int resize_hpt_for_hotplug(unsigned long new_mem_size)
792 {
793 	unsigned target_hpt_shift;
794 
795 	if (!mmu_hash_ops.resize_hpt)
796 		return 0;
797 
798 	target_hpt_shift = htab_shift_for_mem_size(new_mem_size);
799 
800 	/*
801 	 * To avoid lots of HPT resizes if memory size is fluctuating
802 	 * across a boundary, we deliberately have some hysterisis
803 	 * here: we immediately increase the HPT size if the target
804 	 * shift exceeds the current shift, but we won't attempt to
805 	 * reduce unless the target shift is at least 2 below the
806 	 * current shift
807 	 */
808 	if (target_hpt_shift > ppc64_pft_size ||
809 	    target_hpt_shift < ppc64_pft_size - 1)
810 		return mmu_hash_ops.resize_hpt(target_hpt_shift);
811 
812 	return 0;
813 }
814 
815 int hash__create_section_mapping(unsigned long start, unsigned long end,
816 				 int nid, pgprot_t prot)
817 {
818 	int rc;
819 
820 	if (end >= H_VMALLOC_START) {
821 		pr_warn("Outside the supported range\n");
822 		return -1;
823 	}
824 
825 	rc = htab_bolt_mapping(start, end, __pa(start),
826 			       pgprot_val(prot), mmu_linear_psize,
827 			       mmu_kernel_ssize);
828 
829 	if (rc < 0) {
830 		int rc2 = htab_remove_mapping(start, end, mmu_linear_psize,
831 					      mmu_kernel_ssize);
832 		BUG_ON(rc2 && (rc2 != -ENOENT));
833 	}
834 	return rc;
835 }
836 
837 int hash__remove_section_mapping(unsigned long start, unsigned long end)
838 {
839 	int rc = htab_remove_mapping(start, end, mmu_linear_psize,
840 				     mmu_kernel_ssize);
841 	WARN_ON(rc < 0);
842 	return rc;
843 }
844 #endif /* CONFIG_MEMORY_HOTPLUG */
845 
846 static void __init hash_init_partition_table(phys_addr_t hash_table,
847 					     unsigned long htab_size)
848 {
849 	mmu_partition_table_init();
850 
851 	/*
852 	 * PS field (VRMA page size) is not used for LPID 0, hence set to 0.
853 	 * For now, UPRT is 0 and we have no segment table.
854 	 */
855 	htab_size =  __ilog2(htab_size) - 18;
856 	mmu_partition_table_set_entry(0, hash_table | htab_size, 0, false);
857 	pr_info("Partition table %p\n", partition_tb);
858 }
859 
860 static void __init htab_initialize(void)
861 {
862 	unsigned long table;
863 	unsigned long pteg_count;
864 	unsigned long prot;
865 	unsigned long base = 0, size = 0;
866 	struct memblock_region *reg;
867 
868 	DBG(" -> htab_initialize()\n");
869 
870 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT)) {
871 		mmu_kernel_ssize = MMU_SEGSIZE_1T;
872 		mmu_highuser_ssize = MMU_SEGSIZE_1T;
873 		printk(KERN_INFO "Using 1TB segments\n");
874 	}
875 
876 	if (stress_slb_enabled)
877 		static_branch_enable(&stress_slb_key);
878 
879 	/*
880 	 * Calculate the required size of the htab.  We want the number of
881 	 * PTEGs to equal one half the number of real pages.
882 	 */
883 	htab_size_bytes = htab_get_table_size();
884 	pteg_count = htab_size_bytes >> 7;
885 
886 	htab_hash_mask = pteg_count - 1;
887 
888 	if (firmware_has_feature(FW_FEATURE_LPAR) ||
889 	    firmware_has_feature(FW_FEATURE_PS3_LV1)) {
890 		/* Using a hypervisor which owns the htab */
891 		htab_address = NULL;
892 		_SDR1 = 0;
893 #ifdef CONFIG_FA_DUMP
894 		/*
895 		 * If firmware assisted dump is active firmware preserves
896 		 * the contents of htab along with entire partition memory.
897 		 * Clear the htab if firmware assisted dump is active so
898 		 * that we dont end up using old mappings.
899 		 */
900 		if (is_fadump_active() && mmu_hash_ops.hpte_clear_all)
901 			mmu_hash_ops.hpte_clear_all();
902 #endif
903 	} else {
904 		unsigned long limit = MEMBLOCK_ALLOC_ANYWHERE;
905 
906 #ifdef CONFIG_PPC_CELL
907 		/*
908 		 * Cell may require the hash table down low when using the
909 		 * Axon IOMMU in order to fit the dynamic region over it, see
910 		 * comments in cell/iommu.c
911 		 */
912 		if (fdt_subnode_offset(initial_boot_params, 0, "axon") > 0) {
913 			limit = 0x80000000;
914 			pr_info("Hash table forced below 2G for Axon IOMMU\n");
915 		}
916 #endif /* CONFIG_PPC_CELL */
917 
918 		table = memblock_phys_alloc_range(htab_size_bytes,
919 						  htab_size_bytes,
920 						  0, limit);
921 		if (!table)
922 			panic("ERROR: Failed to allocate %pa bytes below %pa\n",
923 			      &htab_size_bytes, &limit);
924 
925 		DBG("Hash table allocated at %lx, size: %lx\n", table,
926 		    htab_size_bytes);
927 
928 		htab_address = __va(table);
929 
930 		/* htab absolute addr + encoded htabsize */
931 		_SDR1 = table + __ilog2(htab_size_bytes) - 18;
932 
933 		/* Initialize the HPT with no entries */
934 		memset((void *)table, 0, htab_size_bytes);
935 
936 		if (!cpu_has_feature(CPU_FTR_ARCH_300))
937 			/* Set SDR1 */
938 			mtspr(SPRN_SDR1, _SDR1);
939 		else
940 			hash_init_partition_table(table, htab_size_bytes);
941 	}
942 
943 	prot = pgprot_val(PAGE_KERNEL);
944 
945 #ifdef CONFIG_DEBUG_PAGEALLOC
946 	if (debug_pagealloc_enabled()) {
947 		linear_map_hash_count = memblock_end_of_DRAM() >> PAGE_SHIFT;
948 		linear_map_hash_slots = memblock_alloc_try_nid(
949 				linear_map_hash_count, 1, MEMBLOCK_LOW_LIMIT,
950 				ppc64_rma_size,	NUMA_NO_NODE);
951 		if (!linear_map_hash_slots)
952 			panic("%s: Failed to allocate %lu bytes max_addr=%pa\n",
953 			      __func__, linear_map_hash_count, &ppc64_rma_size);
954 	}
955 #endif /* CONFIG_DEBUG_PAGEALLOC */
956 
957 	/* create bolted the linear mapping in the hash table */
958 	for_each_memblock(memory, reg) {
959 		base = (unsigned long)__va(reg->base);
960 		size = reg->size;
961 
962 		DBG("creating mapping for region: %lx..%lx (prot: %lx)\n",
963 		    base, size, prot);
964 
965 		if ((base + size) >= H_VMALLOC_START) {
966 			pr_warn("Outside the supported range\n");
967 			continue;
968 		}
969 
970 		BUG_ON(htab_bolt_mapping(base, base + size, __pa(base),
971 				prot, mmu_linear_psize, mmu_kernel_ssize));
972 	}
973 	memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
974 
975 	/*
976 	 * If we have a memory_limit and we've allocated TCEs then we need to
977 	 * explicitly map the TCE area at the top of RAM. We also cope with the
978 	 * case that the TCEs start below memory_limit.
979 	 * tce_alloc_start/end are 16MB aligned so the mapping should work
980 	 * for either 4K or 16MB pages.
981 	 */
982 	if (tce_alloc_start) {
983 		tce_alloc_start = (unsigned long)__va(tce_alloc_start);
984 		tce_alloc_end = (unsigned long)__va(tce_alloc_end);
985 
986 		if (base + size >= tce_alloc_start)
987 			tce_alloc_start = base + size + 1;
988 
989 		BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end,
990 					 __pa(tce_alloc_start), prot,
991 					 mmu_linear_psize, mmu_kernel_ssize));
992 	}
993 
994 
995 	DBG(" <- htab_initialize()\n");
996 }
997 #undef KB
998 #undef MB
999 
1000 void __init hash__early_init_devtree(void)
1001 {
1002 	/* Initialize segment sizes */
1003 	of_scan_flat_dt(htab_dt_scan_seg_sizes, NULL);
1004 
1005 	/* Initialize page sizes */
1006 	htab_scan_page_sizes();
1007 }
1008 
1009 static struct hash_mm_context init_hash_mm_context;
1010 void __init hash__early_init_mmu(void)
1011 {
1012 #ifndef CONFIG_PPC_64K_PAGES
1013 	/*
1014 	 * We have code in __hash_page_4K() and elsewhere, which assumes it can
1015 	 * do the following:
1016 	 *   new_pte |= (slot << H_PAGE_F_GIX_SHIFT) & (H_PAGE_F_SECOND | H_PAGE_F_GIX);
1017 	 *
1018 	 * Where the slot number is between 0-15, and values of 8-15 indicate
1019 	 * the secondary bucket. For that code to work H_PAGE_F_SECOND and
1020 	 * H_PAGE_F_GIX must occupy four contiguous bits in the PTE, and
1021 	 * H_PAGE_F_SECOND must be placed above H_PAGE_F_GIX. Assert that here
1022 	 * with a BUILD_BUG_ON().
1023 	 */
1024 	BUILD_BUG_ON(H_PAGE_F_SECOND != (1ul  << (H_PAGE_F_GIX_SHIFT + 3)));
1025 #endif /* CONFIG_PPC_64K_PAGES */
1026 
1027 	htab_init_page_sizes();
1028 
1029 	/*
1030 	 * initialize page table size
1031 	 */
1032 	__pte_frag_nr = H_PTE_FRAG_NR;
1033 	__pte_frag_size_shift = H_PTE_FRAG_SIZE_SHIFT;
1034 	__pmd_frag_nr = H_PMD_FRAG_NR;
1035 	__pmd_frag_size_shift = H_PMD_FRAG_SIZE_SHIFT;
1036 
1037 	__pte_index_size = H_PTE_INDEX_SIZE;
1038 	__pmd_index_size = H_PMD_INDEX_SIZE;
1039 	__pud_index_size = H_PUD_INDEX_SIZE;
1040 	__pgd_index_size = H_PGD_INDEX_SIZE;
1041 	__pud_cache_index = H_PUD_CACHE_INDEX;
1042 	__pte_table_size = H_PTE_TABLE_SIZE;
1043 	__pmd_table_size = H_PMD_TABLE_SIZE;
1044 	__pud_table_size = H_PUD_TABLE_SIZE;
1045 	__pgd_table_size = H_PGD_TABLE_SIZE;
1046 	/*
1047 	 * 4k use hugepd format, so for hash set then to
1048 	 * zero
1049 	 */
1050 	__pmd_val_bits = HASH_PMD_VAL_BITS;
1051 	__pud_val_bits = HASH_PUD_VAL_BITS;
1052 	__pgd_val_bits = HASH_PGD_VAL_BITS;
1053 
1054 	__kernel_virt_start = H_KERN_VIRT_START;
1055 	__vmalloc_start = H_VMALLOC_START;
1056 	__vmalloc_end = H_VMALLOC_END;
1057 	__kernel_io_start = H_KERN_IO_START;
1058 	__kernel_io_end = H_KERN_IO_END;
1059 	vmemmap = (struct page *)H_VMEMMAP_START;
1060 	ioremap_bot = IOREMAP_BASE;
1061 
1062 #ifdef CONFIG_PCI
1063 	pci_io_base = ISA_IO_BASE;
1064 #endif
1065 
1066 	/* Select appropriate backend */
1067 	if (firmware_has_feature(FW_FEATURE_PS3_LV1))
1068 		ps3_early_mm_init();
1069 	else if (firmware_has_feature(FW_FEATURE_LPAR))
1070 		hpte_init_pseries();
1071 	else if (IS_ENABLED(CONFIG_PPC_NATIVE))
1072 		hpte_init_native();
1073 
1074 	if (!mmu_hash_ops.hpte_insert)
1075 		panic("hash__early_init_mmu: No MMU hash ops defined!\n");
1076 
1077 	/*
1078 	 * Initialize the MMU Hash table and create the linear mapping
1079 	 * of memory. Has to be done before SLB initialization as this is
1080 	 * currently where the page size encoding is obtained.
1081 	 */
1082 	htab_initialize();
1083 
1084 	init_mm.context.hash_context = &init_hash_mm_context;
1085 	mm_ctx_set_slb_addr_limit(&init_mm.context, SLB_ADDR_LIMIT_DEFAULT);
1086 
1087 	pr_info("Initializing hash mmu with SLB\n");
1088 	/* Initialize SLB management */
1089 	slb_initialize();
1090 
1091 	if (cpu_has_feature(CPU_FTR_ARCH_206)
1092 			&& cpu_has_feature(CPU_FTR_HVMODE))
1093 		tlbiel_all();
1094 }
1095 
1096 #ifdef CONFIG_SMP
1097 void hash__early_init_mmu_secondary(void)
1098 {
1099 	/* Initialize hash table for that CPU */
1100 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
1101 
1102 		if (!cpu_has_feature(CPU_FTR_ARCH_300))
1103 			mtspr(SPRN_SDR1, _SDR1);
1104 		else
1105 			set_ptcr_when_no_uv(__pa(partition_tb) |
1106 					    (PATB_SIZE_SHIFT - 12));
1107 	}
1108 	/* Initialize SLB */
1109 	slb_initialize();
1110 
1111 	if (cpu_has_feature(CPU_FTR_ARCH_206)
1112 			&& cpu_has_feature(CPU_FTR_HVMODE))
1113 		tlbiel_all();
1114 }
1115 #endif /* CONFIG_SMP */
1116 
1117 /*
1118  * Called by asm hashtable.S for doing lazy icache flush
1119  */
1120 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap)
1121 {
1122 	struct page *page;
1123 
1124 	if (!pfn_valid(pte_pfn(pte)))
1125 		return pp;
1126 
1127 	page = pte_page(pte);
1128 
1129 	/* page is dirty */
1130 	if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) {
1131 		if (trap == 0x400) {
1132 			flush_dcache_icache_page(page);
1133 			set_bit(PG_arch_1, &page->flags);
1134 		} else
1135 			pp |= HPTE_R_N;
1136 	}
1137 	return pp;
1138 }
1139 
1140 #ifdef CONFIG_PPC_MM_SLICES
1141 static unsigned int get_paca_psize(unsigned long addr)
1142 {
1143 	unsigned char *psizes;
1144 	unsigned long index, mask_index;
1145 
1146 	if (addr < SLICE_LOW_TOP) {
1147 		psizes = get_paca()->mm_ctx_low_slices_psize;
1148 		index = GET_LOW_SLICE_INDEX(addr);
1149 	} else {
1150 		psizes = get_paca()->mm_ctx_high_slices_psize;
1151 		index = GET_HIGH_SLICE_INDEX(addr);
1152 	}
1153 	mask_index = index & 0x1;
1154 	return (psizes[index >> 1] >> (mask_index * 4)) & 0xF;
1155 }
1156 
1157 #else
1158 unsigned int get_paca_psize(unsigned long addr)
1159 {
1160 	return get_paca()->mm_ctx_user_psize;
1161 }
1162 #endif
1163 
1164 /*
1165  * Demote a segment to using 4k pages.
1166  * For now this makes the whole process use 4k pages.
1167  */
1168 #ifdef CONFIG_PPC_64K_PAGES
1169 void demote_segment_4k(struct mm_struct *mm, unsigned long addr)
1170 {
1171 	if (get_slice_psize(mm, addr) == MMU_PAGE_4K)
1172 		return;
1173 	slice_set_range_psize(mm, addr, 1, MMU_PAGE_4K);
1174 	copro_flush_all_slbs(mm);
1175 	if ((get_paca_psize(addr) != MMU_PAGE_4K) && (current->mm == mm)) {
1176 
1177 		copy_mm_to_paca(mm);
1178 		slb_flush_and_restore_bolted();
1179 	}
1180 }
1181 #endif /* CONFIG_PPC_64K_PAGES */
1182 
1183 #ifdef CONFIG_PPC_SUBPAGE_PROT
1184 /*
1185  * This looks up a 2-bit protection code for a 4k subpage of a 64k page.
1186  * Userspace sets the subpage permissions using the subpage_prot system call.
1187  *
1188  * Result is 0: full permissions, _PAGE_RW: read-only,
1189  * _PAGE_RWX: no access.
1190  */
1191 static int subpage_protection(struct mm_struct *mm, unsigned long ea)
1192 {
1193 	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
1194 	u32 spp = 0;
1195 	u32 **sbpm, *sbpp;
1196 
1197 	if (!spt)
1198 		return 0;
1199 
1200 	if (ea >= spt->maxaddr)
1201 		return 0;
1202 	if (ea < 0x100000000UL) {
1203 		/* addresses below 4GB use spt->low_prot */
1204 		sbpm = spt->low_prot;
1205 	} else {
1206 		sbpm = spt->protptrs[ea >> SBP_L3_SHIFT];
1207 		if (!sbpm)
1208 			return 0;
1209 	}
1210 	sbpp = sbpm[(ea >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
1211 	if (!sbpp)
1212 		return 0;
1213 	spp = sbpp[(ea >> PAGE_SHIFT) & (SBP_L1_COUNT - 1)];
1214 
1215 	/* extract 2-bit bitfield for this 4k subpage */
1216 	spp >>= 30 - 2 * ((ea >> 12) & 0xf);
1217 
1218 	/*
1219 	 * 0 -> full premission
1220 	 * 1 -> Read only
1221 	 * 2 -> no access.
1222 	 * We return the flag that need to be cleared.
1223 	 */
1224 	spp = ((spp & 2) ? _PAGE_RWX : 0) | ((spp & 1) ? _PAGE_WRITE : 0);
1225 	return spp;
1226 }
1227 
1228 #else /* CONFIG_PPC_SUBPAGE_PROT */
1229 static inline int subpage_protection(struct mm_struct *mm, unsigned long ea)
1230 {
1231 	return 0;
1232 }
1233 #endif
1234 
1235 void hash_failure_debug(unsigned long ea, unsigned long access,
1236 			unsigned long vsid, unsigned long trap,
1237 			int ssize, int psize, int lpsize, unsigned long pte)
1238 {
1239 	if (!printk_ratelimit())
1240 		return;
1241 	pr_info("mm: Hashing failure ! EA=0x%lx access=0x%lx current=%s\n",
1242 		ea, access, current->comm);
1243 	pr_info("    trap=0x%lx vsid=0x%lx ssize=%d base psize=%d psize %d pte=0x%lx\n",
1244 		trap, vsid, ssize, psize, lpsize, pte);
1245 }
1246 
1247 static void check_paca_psize(unsigned long ea, struct mm_struct *mm,
1248 			     int psize, bool user_region)
1249 {
1250 	if (user_region) {
1251 		if (psize != get_paca_psize(ea)) {
1252 			copy_mm_to_paca(mm);
1253 			slb_flush_and_restore_bolted();
1254 		}
1255 	} else if (get_paca()->vmalloc_sllp !=
1256 		   mmu_psize_defs[mmu_vmalloc_psize].sllp) {
1257 		get_paca()->vmalloc_sllp =
1258 			mmu_psize_defs[mmu_vmalloc_psize].sllp;
1259 		slb_vmalloc_update();
1260 	}
1261 }
1262 
1263 /*
1264  * Result code is:
1265  *  0 - handled
1266  *  1 - normal page fault
1267  * -1 - critical hash insertion error
1268  * -2 - access not permitted by subpage protection mechanism
1269  */
1270 int hash_page_mm(struct mm_struct *mm, unsigned long ea,
1271 		 unsigned long access, unsigned long trap,
1272 		 unsigned long flags)
1273 {
1274 	bool is_thp;
1275 	enum ctx_state prev_state = exception_enter();
1276 	pgd_t *pgdir;
1277 	unsigned long vsid;
1278 	pte_t *ptep;
1279 	unsigned hugeshift;
1280 	int rc, user_region = 0;
1281 	int psize, ssize;
1282 
1283 	DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n",
1284 		ea, access, trap);
1285 	trace_hash_fault(ea, access, trap);
1286 
1287 	/* Get region & vsid */
1288 	switch (get_region_id(ea)) {
1289 	case USER_REGION_ID:
1290 		user_region = 1;
1291 		if (! mm) {
1292 			DBG_LOW(" user region with no mm !\n");
1293 			rc = 1;
1294 			goto bail;
1295 		}
1296 		psize = get_slice_psize(mm, ea);
1297 		ssize = user_segment_size(ea);
1298 		vsid = get_user_vsid(&mm->context, ea, ssize);
1299 		break;
1300 	case VMALLOC_REGION_ID:
1301 		vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
1302 		psize = mmu_vmalloc_psize;
1303 		ssize = mmu_kernel_ssize;
1304 		break;
1305 
1306 	case IO_REGION_ID:
1307 		vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
1308 		psize = mmu_io_psize;
1309 		ssize = mmu_kernel_ssize;
1310 		break;
1311 	default:
1312 		/*
1313 		 * Not a valid range
1314 		 * Send the problem up to do_page_fault()
1315 		 */
1316 		rc = 1;
1317 		goto bail;
1318 	}
1319 	DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid);
1320 
1321 	/* Bad address. */
1322 	if (!vsid) {
1323 		DBG_LOW("Bad address!\n");
1324 		rc = 1;
1325 		goto bail;
1326 	}
1327 	/* Get pgdir */
1328 	pgdir = mm->pgd;
1329 	if (pgdir == NULL) {
1330 		rc = 1;
1331 		goto bail;
1332 	}
1333 
1334 	/* Check CPU locality */
1335 	if (user_region && mm_is_thread_local(mm))
1336 		flags |= HPTE_LOCAL_UPDATE;
1337 
1338 #ifndef CONFIG_PPC_64K_PAGES
1339 	/*
1340 	 * If we use 4K pages and our psize is not 4K, then we might
1341 	 * be hitting a special driver mapping, and need to align the
1342 	 * address before we fetch the PTE.
1343 	 *
1344 	 * It could also be a hugepage mapping, in which case this is
1345 	 * not necessary, but it's not harmful, either.
1346 	 */
1347 	if (psize != MMU_PAGE_4K)
1348 		ea &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
1349 #endif /* CONFIG_PPC_64K_PAGES */
1350 
1351 	/* Get PTE and page size from page tables */
1352 	ptep = find_linux_pte(pgdir, ea, &is_thp, &hugeshift);
1353 	if (ptep == NULL || !pte_present(*ptep)) {
1354 		DBG_LOW(" no PTE !\n");
1355 		rc = 1;
1356 		goto bail;
1357 	}
1358 
1359 	/*
1360 	 * Add _PAGE_PRESENT to the required access perm. If there are parallel
1361 	 * updates to the pte that can possibly clear _PAGE_PTE, catch that too.
1362 	 *
1363 	 * We can safely use the return pte address in rest of the function
1364 	 * because we do set H_PAGE_BUSY which prevents further updates to pte
1365 	 * from generic code.
1366 	 */
1367 	access |= _PAGE_PRESENT | _PAGE_PTE;
1368 
1369 	/*
1370 	 * Pre-check access permissions (will be re-checked atomically
1371 	 * in __hash_page_XX but this pre-check is a fast path
1372 	 */
1373 	if (!check_pte_access(access, pte_val(*ptep))) {
1374 		DBG_LOW(" no access !\n");
1375 		rc = 1;
1376 		goto bail;
1377 	}
1378 
1379 	if (hugeshift) {
1380 		if (is_thp)
1381 			rc = __hash_page_thp(ea, access, vsid, (pmd_t *)ptep,
1382 					     trap, flags, ssize, psize);
1383 #ifdef CONFIG_HUGETLB_PAGE
1384 		else
1385 			rc = __hash_page_huge(ea, access, vsid, ptep, trap,
1386 					      flags, ssize, hugeshift, psize);
1387 #else
1388 		else {
1389 			/*
1390 			 * if we have hugeshift, and is not transhuge with
1391 			 * hugetlb disabled, something is really wrong.
1392 			 */
1393 			rc = 1;
1394 			WARN_ON(1);
1395 		}
1396 #endif
1397 		if (current->mm == mm)
1398 			check_paca_psize(ea, mm, psize, user_region);
1399 
1400 		goto bail;
1401 	}
1402 
1403 #ifndef CONFIG_PPC_64K_PAGES
1404 	DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep));
1405 #else
1406 	DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep),
1407 		pte_val(*(ptep + PTRS_PER_PTE)));
1408 #endif
1409 	/* Do actual hashing */
1410 #ifdef CONFIG_PPC_64K_PAGES
1411 	/* If H_PAGE_4K_PFN is set, make sure this is a 4k segment */
1412 	if ((pte_val(*ptep) & H_PAGE_4K_PFN) && psize == MMU_PAGE_64K) {
1413 		demote_segment_4k(mm, ea);
1414 		psize = MMU_PAGE_4K;
1415 	}
1416 
1417 	/*
1418 	 * If this PTE is non-cacheable and we have restrictions on
1419 	 * using non cacheable large pages, then we switch to 4k
1420 	 */
1421 	if (mmu_ci_restrictions && psize == MMU_PAGE_64K && pte_ci(*ptep)) {
1422 		if (user_region) {
1423 			demote_segment_4k(mm, ea);
1424 			psize = MMU_PAGE_4K;
1425 		} else if (ea < VMALLOC_END) {
1426 			/*
1427 			 * some driver did a non-cacheable mapping
1428 			 * in vmalloc space, so switch vmalloc
1429 			 * to 4k pages
1430 			 */
1431 			printk(KERN_ALERT "Reducing vmalloc segment "
1432 			       "to 4kB pages because of "
1433 			       "non-cacheable mapping\n");
1434 			psize = mmu_vmalloc_psize = MMU_PAGE_4K;
1435 			copro_flush_all_slbs(mm);
1436 		}
1437 	}
1438 
1439 #endif /* CONFIG_PPC_64K_PAGES */
1440 
1441 	if (current->mm == mm)
1442 		check_paca_psize(ea, mm, psize, user_region);
1443 
1444 #ifdef CONFIG_PPC_64K_PAGES
1445 	if (psize == MMU_PAGE_64K)
1446 		rc = __hash_page_64K(ea, access, vsid, ptep, trap,
1447 				     flags, ssize);
1448 	else
1449 #endif /* CONFIG_PPC_64K_PAGES */
1450 	{
1451 		int spp = subpage_protection(mm, ea);
1452 		if (access & spp)
1453 			rc = -2;
1454 		else
1455 			rc = __hash_page_4K(ea, access, vsid, ptep, trap,
1456 					    flags, ssize, spp);
1457 	}
1458 
1459 	/*
1460 	 * Dump some info in case of hash insertion failure, they should
1461 	 * never happen so it is really useful to know if/when they do
1462 	 */
1463 	if (rc == -1)
1464 		hash_failure_debug(ea, access, vsid, trap, ssize, psize,
1465 				   psize, pte_val(*ptep));
1466 #ifndef CONFIG_PPC_64K_PAGES
1467 	DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep));
1468 #else
1469 	DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep),
1470 		pte_val(*(ptep + PTRS_PER_PTE)));
1471 #endif
1472 	DBG_LOW(" -> rc=%d\n", rc);
1473 
1474 bail:
1475 	exception_exit(prev_state);
1476 	return rc;
1477 }
1478 EXPORT_SYMBOL_GPL(hash_page_mm);
1479 
1480 int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
1481 	      unsigned long dsisr)
1482 {
1483 	unsigned long flags = 0;
1484 	struct mm_struct *mm = current->mm;
1485 
1486 	if ((get_region_id(ea) == VMALLOC_REGION_ID) ||
1487 	    (get_region_id(ea) == IO_REGION_ID))
1488 		mm = &init_mm;
1489 
1490 	if (dsisr & DSISR_NOHPTE)
1491 		flags |= HPTE_NOHPTE_UPDATE;
1492 
1493 	return hash_page_mm(mm, ea, access, trap, flags);
1494 }
1495 EXPORT_SYMBOL_GPL(hash_page);
1496 
1497 int __hash_page(unsigned long trap, unsigned long ea, unsigned long dsisr,
1498 		unsigned long msr)
1499 {
1500 	unsigned long access = _PAGE_PRESENT | _PAGE_READ;
1501 	unsigned long flags = 0;
1502 	struct mm_struct *mm = current->mm;
1503 	unsigned int region_id = get_region_id(ea);
1504 
1505 	if ((region_id == VMALLOC_REGION_ID) || (region_id == IO_REGION_ID))
1506 		mm = &init_mm;
1507 
1508 	if (dsisr & DSISR_NOHPTE)
1509 		flags |= HPTE_NOHPTE_UPDATE;
1510 
1511 	if (dsisr & DSISR_ISSTORE)
1512 		access |= _PAGE_WRITE;
1513 	/*
1514 	 * We set _PAGE_PRIVILEGED only when
1515 	 * kernel mode access kernel space.
1516 	 *
1517 	 * _PAGE_PRIVILEGED is NOT set
1518 	 * 1) when kernel mode access user space
1519 	 * 2) user space access kernel space.
1520 	 */
1521 	access |= _PAGE_PRIVILEGED;
1522 	if ((msr & MSR_PR) || (region_id == USER_REGION_ID))
1523 		access &= ~_PAGE_PRIVILEGED;
1524 
1525 	if (trap == 0x400)
1526 		access |= _PAGE_EXEC;
1527 
1528 	return hash_page_mm(mm, ea, access, trap, flags);
1529 }
1530 
1531 #ifdef CONFIG_PPC_MM_SLICES
1532 static bool should_hash_preload(struct mm_struct *mm, unsigned long ea)
1533 {
1534 	int psize = get_slice_psize(mm, ea);
1535 
1536 	/* We only prefault standard pages for now */
1537 	if (unlikely(psize != mm_ctx_user_psize(&mm->context)))
1538 		return false;
1539 
1540 	/*
1541 	 * Don't prefault if subpage protection is enabled for the EA.
1542 	 */
1543 	if (unlikely((psize == MMU_PAGE_4K) && subpage_protection(mm, ea)))
1544 		return false;
1545 
1546 	return true;
1547 }
1548 #else
1549 static bool should_hash_preload(struct mm_struct *mm, unsigned long ea)
1550 {
1551 	return true;
1552 }
1553 #endif
1554 
1555 static void hash_preload(struct mm_struct *mm, pte_t *ptep, unsigned long ea,
1556 			 bool is_exec, unsigned long trap)
1557 {
1558 	unsigned long vsid;
1559 	pgd_t *pgdir;
1560 	int rc, ssize, update_flags = 0;
1561 	unsigned long access = _PAGE_PRESENT | _PAGE_READ | (is_exec ? _PAGE_EXEC : 0);
1562 
1563 	BUG_ON(get_region_id(ea) != USER_REGION_ID);
1564 
1565 	if (!should_hash_preload(mm, ea))
1566 		return;
1567 
1568 	DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx,"
1569 		" trap=%lx\n", mm, mm->pgd, ea, access, trap);
1570 
1571 	/* Get Linux PTE if available */
1572 	pgdir = mm->pgd;
1573 	if (pgdir == NULL)
1574 		return;
1575 
1576 	/* Get VSID */
1577 	ssize = user_segment_size(ea);
1578 	vsid = get_user_vsid(&mm->context, ea, ssize);
1579 	if (!vsid)
1580 		return;
1581 
1582 #ifdef CONFIG_PPC_64K_PAGES
1583 	/* If either H_PAGE_4K_PFN or cache inhibited is set (and we are on
1584 	 * a 64K kernel), then we don't preload, hash_page() will take
1585 	 * care of it once we actually try to access the page.
1586 	 * That way we don't have to duplicate all of the logic for segment
1587 	 * page size demotion here
1588 	 * Called with  PTL held, hence can be sure the value won't change in
1589 	 * between.
1590 	 */
1591 	if ((pte_val(*ptep) & H_PAGE_4K_PFN) || pte_ci(*ptep))
1592 		return;
1593 #endif /* CONFIG_PPC_64K_PAGES */
1594 
1595 	/* Is that local to this CPU ? */
1596 	if (mm_is_thread_local(mm))
1597 		update_flags |= HPTE_LOCAL_UPDATE;
1598 
1599 	/* Hash it in */
1600 #ifdef CONFIG_PPC_64K_PAGES
1601 	if (mm_ctx_user_psize(&mm->context) == MMU_PAGE_64K)
1602 		rc = __hash_page_64K(ea, access, vsid, ptep, trap,
1603 				     update_flags, ssize);
1604 	else
1605 #endif /* CONFIG_PPC_64K_PAGES */
1606 		rc = __hash_page_4K(ea, access, vsid, ptep, trap, update_flags,
1607 				    ssize, subpage_protection(mm, ea));
1608 
1609 	/* Dump some info in case of hash insertion failure, they should
1610 	 * never happen so it is really useful to know if/when they do
1611 	 */
1612 	if (rc == -1)
1613 		hash_failure_debug(ea, access, vsid, trap, ssize,
1614 				   mm_ctx_user_psize(&mm->context),
1615 				   mm_ctx_user_psize(&mm->context),
1616 				   pte_val(*ptep));
1617 }
1618 
1619 /*
1620  * This is called at the end of handling a user page fault, when the
1621  * fault has been handled by updating a PTE in the linux page tables.
1622  * We use it to preload an HPTE into the hash table corresponding to
1623  * the updated linux PTE.
1624  *
1625  * This must always be called with the pte lock held.
1626  */
1627 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
1628 		      pte_t *ptep)
1629 {
1630 	/*
1631 	 * We don't need to worry about _PAGE_PRESENT here because we are
1632 	 * called with either mm->page_table_lock held or ptl lock held
1633 	 */
1634 	unsigned long trap;
1635 	bool is_exec;
1636 
1637 	if (radix_enabled())
1638 		return;
1639 
1640 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
1641 	if (!pte_young(*ptep) || address >= TASK_SIZE)
1642 		return;
1643 
1644 	/*
1645 	 * We try to figure out if we are coming from an instruction
1646 	 * access fault and pass that down to __hash_page so we avoid
1647 	 * double-faulting on execution of fresh text. We have to test
1648 	 * for regs NULL since init will get here first thing at boot.
1649 	 *
1650 	 * We also avoid filling the hash if not coming from a fault.
1651 	 */
1652 
1653 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
1654 	switch (trap) {
1655 	case 0x300:
1656 		is_exec = false;
1657 		break;
1658 	case 0x400:
1659 		is_exec = true;
1660 		break;
1661 	default:
1662 		return;
1663 	}
1664 
1665 	hash_preload(vma->vm_mm, ptep, address, is_exec, trap);
1666 }
1667 
1668 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1669 static inline void tm_flush_hash_page(int local)
1670 {
1671 	/*
1672 	 * Transactions are not aborted by tlbiel, only tlbie. Without, syncing a
1673 	 * page back to a block device w/PIO could pick up transactional data
1674 	 * (bad!) so we force an abort here. Before the sync the page will be
1675 	 * made read-only, which will flush_hash_page. BIG ISSUE here: if the
1676 	 * kernel uses a page from userspace without unmapping it first, it may
1677 	 * see the speculated version.
1678 	 */
1679 	if (local && cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
1680 	    MSR_TM_ACTIVE(current->thread.regs->msr)) {
1681 		tm_enable();
1682 		tm_abort(TM_CAUSE_TLBI);
1683 	}
1684 }
1685 #else
1686 static inline void tm_flush_hash_page(int local)
1687 {
1688 }
1689 #endif
1690 
1691 /*
1692  * Return the global hash slot, corresponding to the given PTE, which contains
1693  * the HPTE.
1694  */
1695 unsigned long pte_get_hash_gslot(unsigned long vpn, unsigned long shift,
1696 		int ssize, real_pte_t rpte, unsigned int subpg_index)
1697 {
1698 	unsigned long hash, gslot, hidx;
1699 
1700 	hash = hpt_hash(vpn, shift, ssize);
1701 	hidx = __rpte_to_hidx(rpte, subpg_index);
1702 	if (hidx & _PTEIDX_SECONDARY)
1703 		hash = ~hash;
1704 	gslot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1705 	gslot += hidx & _PTEIDX_GROUP_IX;
1706 	return gslot;
1707 }
1708 
1709 /*
1710  * WARNING: This is called from hash_low_64.S, if you change this prototype,
1711  *          do not forget to update the assembly call site !
1712  */
1713 void flush_hash_page(unsigned long vpn, real_pte_t pte, int psize, int ssize,
1714 		     unsigned long flags)
1715 {
1716 	unsigned long index, shift, gslot;
1717 	int local = flags & HPTE_LOCAL_UPDATE;
1718 
1719 	DBG_LOW("flush_hash_page(vpn=%016lx)\n", vpn);
1720 	pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1721 		gslot = pte_get_hash_gslot(vpn, shift, ssize, pte, index);
1722 		DBG_LOW(" sub %ld: gslot=%lx\n", index, gslot);
1723 		/*
1724 		 * We use same base page size and actual psize, because we don't
1725 		 * use these functions for hugepage
1726 		 */
1727 		mmu_hash_ops.hpte_invalidate(gslot, vpn, psize, psize,
1728 					     ssize, local);
1729 	} pte_iterate_hashed_end();
1730 
1731 	tm_flush_hash_page(local);
1732 }
1733 
1734 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1735 void flush_hash_hugepage(unsigned long vsid, unsigned long addr,
1736 			 pmd_t *pmdp, unsigned int psize, int ssize,
1737 			 unsigned long flags)
1738 {
1739 	int i, max_hpte_count, valid;
1740 	unsigned long s_addr;
1741 	unsigned char *hpte_slot_array;
1742 	unsigned long hidx, shift, vpn, hash, slot;
1743 	int local = flags & HPTE_LOCAL_UPDATE;
1744 
1745 	s_addr = addr & HPAGE_PMD_MASK;
1746 	hpte_slot_array = get_hpte_slot_array(pmdp);
1747 	/*
1748 	 * IF we try to do a HUGE PTE update after a withdraw is done.
1749 	 * we will find the below NULL. This happens when we do
1750 	 * split_huge_pmd
1751 	 */
1752 	if (!hpte_slot_array)
1753 		return;
1754 
1755 	if (mmu_hash_ops.hugepage_invalidate) {
1756 		mmu_hash_ops.hugepage_invalidate(vsid, s_addr, hpte_slot_array,
1757 						 psize, ssize, local);
1758 		goto tm_abort;
1759 	}
1760 	/*
1761 	 * No bluk hpte removal support, invalidate each entry
1762 	 */
1763 	shift = mmu_psize_defs[psize].shift;
1764 	max_hpte_count = HPAGE_PMD_SIZE >> shift;
1765 	for (i = 0; i < max_hpte_count; i++) {
1766 		/*
1767 		 * 8 bits per each hpte entries
1768 		 * 000| [ secondary group (one bit) | hidx (3 bits) | valid bit]
1769 		 */
1770 		valid = hpte_valid(hpte_slot_array, i);
1771 		if (!valid)
1772 			continue;
1773 		hidx =  hpte_hash_index(hpte_slot_array, i);
1774 
1775 		/* get the vpn */
1776 		addr = s_addr + (i * (1ul << shift));
1777 		vpn = hpt_vpn(addr, vsid, ssize);
1778 		hash = hpt_hash(vpn, shift, ssize);
1779 		if (hidx & _PTEIDX_SECONDARY)
1780 			hash = ~hash;
1781 
1782 		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1783 		slot += hidx & _PTEIDX_GROUP_IX;
1784 		mmu_hash_ops.hpte_invalidate(slot, vpn, psize,
1785 					     MMU_PAGE_16M, ssize, local);
1786 	}
1787 tm_abort:
1788 	tm_flush_hash_page(local);
1789 }
1790 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1791 
1792 void flush_hash_range(unsigned long number, int local)
1793 {
1794 	if (mmu_hash_ops.flush_hash_range)
1795 		mmu_hash_ops.flush_hash_range(number, local);
1796 	else {
1797 		int i;
1798 		struct ppc64_tlb_batch *batch =
1799 			this_cpu_ptr(&ppc64_tlb_batch);
1800 
1801 		for (i = 0; i < number; i++)
1802 			flush_hash_page(batch->vpn[i], batch->pte[i],
1803 					batch->psize, batch->ssize, local);
1804 	}
1805 }
1806 
1807 /*
1808  * low_hash_fault is called when we the low level hash code failed
1809  * to instert a PTE due to an hypervisor error
1810  */
1811 void low_hash_fault(struct pt_regs *regs, unsigned long address, int rc)
1812 {
1813 	enum ctx_state prev_state = exception_enter();
1814 
1815 	if (user_mode(regs)) {
1816 #ifdef CONFIG_PPC_SUBPAGE_PROT
1817 		if (rc == -2)
1818 			_exception(SIGSEGV, regs, SEGV_ACCERR, address);
1819 		else
1820 #endif
1821 			_exception(SIGBUS, regs, BUS_ADRERR, address);
1822 	} else
1823 		bad_page_fault(regs, address, SIGBUS);
1824 
1825 	exception_exit(prev_state);
1826 }
1827 
1828 long hpte_insert_repeating(unsigned long hash, unsigned long vpn,
1829 			   unsigned long pa, unsigned long rflags,
1830 			   unsigned long vflags, int psize, int ssize)
1831 {
1832 	unsigned long hpte_group;
1833 	long slot;
1834 
1835 repeat:
1836 	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1837 
1838 	/* Insert into the hash table, primary slot */
1839 	slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, vflags,
1840 					psize, psize, ssize);
1841 
1842 	/* Primary is full, try the secondary */
1843 	if (unlikely(slot == -1)) {
1844 		hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
1845 		slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags,
1846 						vflags | HPTE_V_SECONDARY,
1847 						psize, psize, ssize);
1848 		if (slot == -1) {
1849 			if (mftb() & 0x1)
1850 				hpte_group = (hash & htab_hash_mask) *
1851 						HPTES_PER_GROUP;
1852 
1853 			mmu_hash_ops.hpte_remove(hpte_group);
1854 			goto repeat;
1855 		}
1856 	}
1857 
1858 	return slot;
1859 }
1860 
1861 #ifdef CONFIG_DEBUG_PAGEALLOC
1862 static void kernel_map_linear_page(unsigned long vaddr, unsigned long lmi)
1863 {
1864 	unsigned long hash;
1865 	unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1866 	unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize);
1867 	unsigned long mode = htab_convert_pte_flags(pgprot_val(PAGE_KERNEL));
1868 	long ret;
1869 
1870 	hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize);
1871 
1872 	/* Don't create HPTE entries for bad address */
1873 	if (!vsid)
1874 		return;
1875 
1876 	ret = hpte_insert_repeating(hash, vpn, __pa(vaddr), mode,
1877 				    HPTE_V_BOLTED,
1878 				    mmu_linear_psize, mmu_kernel_ssize);
1879 
1880 	BUG_ON (ret < 0);
1881 	spin_lock(&linear_map_hash_lock);
1882 	BUG_ON(linear_map_hash_slots[lmi] & 0x80);
1883 	linear_map_hash_slots[lmi] = ret | 0x80;
1884 	spin_unlock(&linear_map_hash_lock);
1885 }
1886 
1887 static void kernel_unmap_linear_page(unsigned long vaddr, unsigned long lmi)
1888 {
1889 	unsigned long hash, hidx, slot;
1890 	unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1891 	unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize);
1892 
1893 	hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize);
1894 	spin_lock(&linear_map_hash_lock);
1895 	BUG_ON(!(linear_map_hash_slots[lmi] & 0x80));
1896 	hidx = linear_map_hash_slots[lmi] & 0x7f;
1897 	linear_map_hash_slots[lmi] = 0;
1898 	spin_unlock(&linear_map_hash_lock);
1899 	if (hidx & _PTEIDX_SECONDARY)
1900 		hash = ~hash;
1901 	slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1902 	slot += hidx & _PTEIDX_GROUP_IX;
1903 	mmu_hash_ops.hpte_invalidate(slot, vpn, mmu_linear_psize,
1904 				     mmu_linear_psize,
1905 				     mmu_kernel_ssize, 0);
1906 }
1907 
1908 void __kernel_map_pages(struct page *page, int numpages, int enable)
1909 {
1910 	unsigned long flags, vaddr, lmi;
1911 	int i;
1912 
1913 	local_irq_save(flags);
1914 	for (i = 0; i < numpages; i++, page++) {
1915 		vaddr = (unsigned long)page_address(page);
1916 		lmi = __pa(vaddr) >> PAGE_SHIFT;
1917 		if (lmi >= linear_map_hash_count)
1918 			continue;
1919 		if (enable)
1920 			kernel_map_linear_page(vaddr, lmi);
1921 		else
1922 			kernel_unmap_linear_page(vaddr, lmi);
1923 	}
1924 	local_irq_restore(flags);
1925 }
1926 #endif /* CONFIG_DEBUG_PAGEALLOC */
1927 
1928 void hash__setup_initial_memory_limit(phys_addr_t first_memblock_base,
1929 				phys_addr_t first_memblock_size)
1930 {
1931 	/*
1932 	 * We don't currently support the first MEMBLOCK not mapping 0
1933 	 * physical on those processors
1934 	 */
1935 	BUG_ON(first_memblock_base != 0);
1936 
1937 	/*
1938 	 * On virtualized systems the first entry is our RMA region aka VRMA,
1939 	 * non-virtualized 64-bit hash MMU systems don't have a limitation
1940 	 * on real mode access.
1941 	 *
1942 	 * For guests on platforms before POWER9, we clamp the it limit to 1G
1943 	 * to avoid some funky things such as RTAS bugs etc...
1944 	 *
1945 	 * On POWER9 we limit to 1TB in case the host erroneously told us that
1946 	 * the RMA was >1TB. Effective address bits 0:23 are treated as zero
1947 	 * (meaning the access is aliased to zero i.e. addr = addr % 1TB)
1948 	 * for virtual real mode addressing and so it doesn't make sense to
1949 	 * have an area larger than 1TB as it can't be addressed.
1950 	 */
1951 	if (!early_cpu_has_feature(CPU_FTR_HVMODE)) {
1952 		ppc64_rma_size = first_memblock_size;
1953 		if (!early_cpu_has_feature(CPU_FTR_ARCH_300))
1954 			ppc64_rma_size = min_t(u64, ppc64_rma_size, 0x40000000);
1955 		else
1956 			ppc64_rma_size = min_t(u64, ppc64_rma_size,
1957 					       1UL << SID_SHIFT_1T);
1958 
1959 		/* Finally limit subsequent allocations */
1960 		memblock_set_current_limit(ppc64_rma_size);
1961 	} else {
1962 		ppc64_rma_size = ULONG_MAX;
1963 	}
1964 }
1965 
1966 #ifdef CONFIG_DEBUG_FS
1967 
1968 static int hpt_order_get(void *data, u64 *val)
1969 {
1970 	*val = ppc64_pft_size;
1971 	return 0;
1972 }
1973 
1974 static int hpt_order_set(void *data, u64 val)
1975 {
1976 	int ret;
1977 
1978 	if (!mmu_hash_ops.resize_hpt)
1979 		return -ENODEV;
1980 
1981 	cpus_read_lock();
1982 	ret = mmu_hash_ops.resize_hpt(val);
1983 	cpus_read_unlock();
1984 
1985 	return ret;
1986 }
1987 
1988 DEFINE_DEBUGFS_ATTRIBUTE(fops_hpt_order, hpt_order_get, hpt_order_set, "%llu\n");
1989 
1990 static int __init hash64_debugfs(void)
1991 {
1992 	debugfs_create_file("hpt_order", 0600, powerpc_debugfs_root, NULL,
1993 			    &fops_hpt_order);
1994 	return 0;
1995 }
1996 machine_device_initcall(pseries, hash64_debugfs);
1997 #endif /* CONFIG_DEBUG_FS */
1998 
1999 void __init print_system_hash_info(void)
2000 {
2001 	pr_info("ppc64_pft_size    = 0x%llx\n", ppc64_pft_size);
2002 
2003 	if (htab_hash_mask)
2004 		pr_info("htab_hash_mask    = 0x%lx\n", htab_hash_mask);
2005 }
2006