xref: /linux/arch/powerpc/mm/book3s64/hash_utils.c (revision df561f6688fef775baa341a0f5d960becd248b11)
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
236 		/*
237 		 * Add memory coherence if cache inhibited is not set
238 		 */
239 		rflags |= HPTE_R_M;
240 
241 	rflags |= pte_to_hpte_pkey_bits(pteflags);
242 	return rflags;
243 }
244 
245 int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
246 		      unsigned long pstart, unsigned long prot,
247 		      int psize, int ssize)
248 {
249 	unsigned long vaddr, paddr;
250 	unsigned int step, shift;
251 	int ret = 0;
252 
253 	shift = mmu_psize_defs[psize].shift;
254 	step = 1 << shift;
255 
256 	prot = htab_convert_pte_flags(prot);
257 
258 	DBG("htab_bolt_mapping(%lx..%lx -> %lx (%lx,%d,%d)\n",
259 	    vstart, vend, pstart, prot, psize, ssize);
260 
261 	for (vaddr = vstart, paddr = pstart; vaddr < vend;
262 	     vaddr += step, paddr += step) {
263 		unsigned long hash, hpteg;
264 		unsigned long vsid = get_kernel_vsid(vaddr, ssize);
265 		unsigned long vpn  = hpt_vpn(vaddr, vsid, ssize);
266 		unsigned long tprot = prot;
267 		bool secondary_hash = false;
268 
269 		/*
270 		 * If we hit a bad address return error.
271 		 */
272 		if (!vsid)
273 			return -1;
274 		/* Make kernel text executable */
275 		if (overlaps_kernel_text(vaddr, vaddr + step))
276 			tprot &= ~HPTE_R_N;
277 
278 		/*
279 		 * If relocatable, check if it overlaps interrupt vectors that
280 		 * are copied down to real 0. For relocatable kernel
281 		 * (e.g. kdump case) we copy interrupt vectors down to real
282 		 * address 0. Mark that region as executable. This is
283 		 * because on p8 system with relocation on exception feature
284 		 * enabled, exceptions are raised with MMU (IR=DR=1) ON. Hence
285 		 * in order to execute the interrupt handlers in virtual
286 		 * mode the vector region need to be marked as executable.
287 		 */
288 		if ((PHYSICAL_START > MEMORY_START) &&
289 			overlaps_interrupt_vector_text(vaddr, vaddr + step))
290 				tprot &= ~HPTE_R_N;
291 
292 		hash = hpt_hash(vpn, shift, ssize);
293 		hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
294 
295 		BUG_ON(!mmu_hash_ops.hpte_insert);
296 repeat:
297 		ret = mmu_hash_ops.hpte_insert(hpteg, vpn, paddr, tprot,
298 					       HPTE_V_BOLTED, psize, psize,
299 					       ssize);
300 		if (ret == -1) {
301 			/*
302 			 * Try to to keep bolted entries in primary.
303 			 * Remove non bolted entries and try insert again
304 			 */
305 			ret = mmu_hash_ops.hpte_remove(hpteg);
306 			if (ret != -1)
307 				ret = mmu_hash_ops.hpte_insert(hpteg, vpn, paddr, tprot,
308 							       HPTE_V_BOLTED, psize, psize,
309 							       ssize);
310 			if (ret == -1 && !secondary_hash) {
311 				secondary_hash = true;
312 				hpteg = ((~hash & htab_hash_mask) * HPTES_PER_GROUP);
313 				goto repeat;
314 			}
315 		}
316 
317 		if (ret < 0)
318 			break;
319 
320 		cond_resched();
321 #ifdef CONFIG_DEBUG_PAGEALLOC
322 		if (debug_pagealloc_enabled() &&
323 			(paddr >> PAGE_SHIFT) < linear_map_hash_count)
324 			linear_map_hash_slots[paddr >> PAGE_SHIFT] = ret | 0x80;
325 #endif /* CONFIG_DEBUG_PAGEALLOC */
326 	}
327 	return ret < 0 ? ret : 0;
328 }
329 
330 int htab_remove_mapping(unsigned long vstart, unsigned long vend,
331 		      int psize, int ssize)
332 {
333 	unsigned long vaddr;
334 	unsigned int step, shift;
335 	int rc;
336 	int ret = 0;
337 
338 	shift = mmu_psize_defs[psize].shift;
339 	step = 1 << shift;
340 
341 	if (!mmu_hash_ops.hpte_removebolted)
342 		return -ENODEV;
343 
344 	for (vaddr = vstart; vaddr < vend; vaddr += step) {
345 		rc = mmu_hash_ops.hpte_removebolted(vaddr, psize, ssize);
346 		if (rc == -ENOENT) {
347 			ret = -ENOENT;
348 			continue;
349 		}
350 		if (rc < 0)
351 			return rc;
352 	}
353 
354 	return ret;
355 }
356 
357 static bool disable_1tb_segments = false;
358 
359 static int __init parse_disable_1tb_segments(char *p)
360 {
361 	disable_1tb_segments = true;
362 	return 0;
363 }
364 early_param("disable_1tb_segments", parse_disable_1tb_segments);
365 
366 static int __init htab_dt_scan_seg_sizes(unsigned long node,
367 					 const char *uname, int depth,
368 					 void *data)
369 {
370 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
371 	const __be32 *prop;
372 	int size = 0;
373 
374 	/* We are scanning "cpu" nodes only */
375 	if (type == NULL || strcmp(type, "cpu") != 0)
376 		return 0;
377 
378 	prop = of_get_flat_dt_prop(node, "ibm,processor-segment-sizes", &size);
379 	if (prop == NULL)
380 		return 0;
381 	for (; size >= 4; size -= 4, ++prop) {
382 		if (be32_to_cpu(prop[0]) == 40) {
383 			DBG("1T segment support detected\n");
384 
385 			if (disable_1tb_segments) {
386 				DBG("1T segments disabled by command line\n");
387 				break;
388 			}
389 
390 			cur_cpu_spec->mmu_features |= MMU_FTR_1T_SEGMENT;
391 			return 1;
392 		}
393 	}
394 	cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B;
395 	return 0;
396 }
397 
398 static int __init get_idx_from_shift(unsigned int shift)
399 {
400 	int idx = -1;
401 
402 	switch (shift) {
403 	case 0xc:
404 		idx = MMU_PAGE_4K;
405 		break;
406 	case 0x10:
407 		idx = MMU_PAGE_64K;
408 		break;
409 	case 0x14:
410 		idx = MMU_PAGE_1M;
411 		break;
412 	case 0x18:
413 		idx = MMU_PAGE_16M;
414 		break;
415 	case 0x22:
416 		idx = MMU_PAGE_16G;
417 		break;
418 	}
419 	return idx;
420 }
421 
422 static int __init htab_dt_scan_page_sizes(unsigned long node,
423 					  const char *uname, int depth,
424 					  void *data)
425 {
426 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
427 	const __be32 *prop;
428 	int size = 0;
429 
430 	/* We are scanning "cpu" nodes only */
431 	if (type == NULL || strcmp(type, "cpu") != 0)
432 		return 0;
433 
434 	prop = of_get_flat_dt_prop(node, "ibm,segment-page-sizes", &size);
435 	if (!prop)
436 		return 0;
437 
438 	pr_info("Page sizes from device-tree:\n");
439 	size /= 4;
440 	cur_cpu_spec->mmu_features &= ~(MMU_FTR_16M_PAGE);
441 	while(size > 0) {
442 		unsigned int base_shift = be32_to_cpu(prop[0]);
443 		unsigned int slbenc = be32_to_cpu(prop[1]);
444 		unsigned int lpnum = be32_to_cpu(prop[2]);
445 		struct mmu_psize_def *def;
446 		int idx, base_idx;
447 
448 		size -= 3; prop += 3;
449 		base_idx = get_idx_from_shift(base_shift);
450 		if (base_idx < 0) {
451 			/* skip the pte encoding also */
452 			prop += lpnum * 2; size -= lpnum * 2;
453 			continue;
454 		}
455 		def = &mmu_psize_defs[base_idx];
456 		if (base_idx == MMU_PAGE_16M)
457 			cur_cpu_spec->mmu_features |= MMU_FTR_16M_PAGE;
458 
459 		def->shift = base_shift;
460 		if (base_shift <= 23)
461 			def->avpnm = 0;
462 		else
463 			def->avpnm = (1 << (base_shift - 23)) - 1;
464 		def->sllp = slbenc;
465 		/*
466 		 * We don't know for sure what's up with tlbiel, so
467 		 * for now we only set it for 4K and 64K pages
468 		 */
469 		if (base_idx == MMU_PAGE_4K || base_idx == MMU_PAGE_64K)
470 			def->tlbiel = 1;
471 		else
472 			def->tlbiel = 0;
473 
474 		while (size > 0 && lpnum) {
475 			unsigned int shift = be32_to_cpu(prop[0]);
476 			int penc  = be32_to_cpu(prop[1]);
477 
478 			prop += 2; size -= 2;
479 			lpnum--;
480 
481 			idx = get_idx_from_shift(shift);
482 			if (idx < 0)
483 				continue;
484 
485 			if (penc == -1)
486 				pr_err("Invalid penc for base_shift=%d "
487 				       "shift=%d\n", base_shift, shift);
488 
489 			def->penc[idx] = penc;
490 			pr_info("base_shift=%d: shift=%d, sllp=0x%04lx,"
491 				" avpnm=0x%08lx, tlbiel=%d, penc=%d\n",
492 				base_shift, shift, def->sllp,
493 				def->avpnm, def->tlbiel, def->penc[idx]);
494 		}
495 	}
496 
497 	return 1;
498 }
499 
500 #ifdef CONFIG_HUGETLB_PAGE
501 /*
502  * Scan for 16G memory blocks that have been set aside for huge pages
503  * and reserve those blocks for 16G huge pages.
504  */
505 static int __init htab_dt_scan_hugepage_blocks(unsigned long node,
506 					const char *uname, int depth,
507 					void *data) {
508 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
509 	const __be64 *addr_prop;
510 	const __be32 *page_count_prop;
511 	unsigned int expected_pages;
512 	long unsigned int phys_addr;
513 	long unsigned int block_size;
514 
515 	/* We are scanning "memory" nodes only */
516 	if (type == NULL || strcmp(type, "memory") != 0)
517 		return 0;
518 
519 	/*
520 	 * This property is the log base 2 of the number of virtual pages that
521 	 * will represent this memory block.
522 	 */
523 	page_count_prop = of_get_flat_dt_prop(node, "ibm,expected#pages", NULL);
524 	if (page_count_prop == NULL)
525 		return 0;
526 	expected_pages = (1 << be32_to_cpu(page_count_prop[0]));
527 	addr_prop = of_get_flat_dt_prop(node, "reg", NULL);
528 	if (addr_prop == NULL)
529 		return 0;
530 	phys_addr = be64_to_cpu(addr_prop[0]);
531 	block_size = be64_to_cpu(addr_prop[1]);
532 	if (block_size != (16 * GB))
533 		return 0;
534 	printk(KERN_INFO "Huge page(16GB) memory: "
535 			"addr = 0x%lX size = 0x%lX pages = %d\n",
536 			phys_addr, block_size, expected_pages);
537 	if (phys_addr + block_size * expected_pages <= memblock_end_of_DRAM()) {
538 		memblock_reserve(phys_addr, block_size * expected_pages);
539 		pseries_add_gpage(phys_addr, block_size, expected_pages);
540 	}
541 	return 0;
542 }
543 #endif /* CONFIG_HUGETLB_PAGE */
544 
545 static void mmu_psize_set_default_penc(void)
546 {
547 	int bpsize, apsize;
548 	for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
549 		for (apsize = 0; apsize < MMU_PAGE_COUNT; apsize++)
550 			mmu_psize_defs[bpsize].penc[apsize] = -1;
551 }
552 
553 #ifdef CONFIG_PPC_64K_PAGES
554 
555 static bool might_have_hea(void)
556 {
557 	/*
558 	 * The HEA ethernet adapter requires awareness of the
559 	 * GX bus. Without that awareness we can easily assume
560 	 * we will never see an HEA ethernet device.
561 	 */
562 #ifdef CONFIG_IBMEBUS
563 	return !cpu_has_feature(CPU_FTR_ARCH_207S) &&
564 		firmware_has_feature(FW_FEATURE_SPLPAR);
565 #else
566 	return false;
567 #endif
568 }
569 
570 #endif /* #ifdef CONFIG_PPC_64K_PAGES */
571 
572 static void __init htab_scan_page_sizes(void)
573 {
574 	int rc;
575 
576 	/* se the invalid penc to -1 */
577 	mmu_psize_set_default_penc();
578 
579 	/* Default to 4K pages only */
580 	memcpy(mmu_psize_defs, mmu_psize_defaults,
581 	       sizeof(mmu_psize_defaults));
582 
583 	/*
584 	 * Try to find the available page sizes in the device-tree
585 	 */
586 	rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL);
587 	if (rc == 0 && early_mmu_has_feature(MMU_FTR_16M_PAGE)) {
588 		/*
589 		 * Nothing in the device-tree, but the CPU supports 16M pages,
590 		 * so let's fallback on a known size list for 16M capable CPUs.
591 		 */
592 		memcpy(mmu_psize_defs, mmu_psize_defaults_gp,
593 		       sizeof(mmu_psize_defaults_gp));
594 	}
595 
596 #ifdef CONFIG_HUGETLB_PAGE
597 	if (!hugetlb_disabled && !early_radix_enabled() ) {
598 		/* Reserve 16G huge page memory sections for huge pages */
599 		of_scan_flat_dt(htab_dt_scan_hugepage_blocks, NULL);
600 	}
601 #endif /* CONFIG_HUGETLB_PAGE */
602 }
603 
604 /*
605  * Fill in the hpte_page_sizes[] array.
606  * We go through the mmu_psize_defs[] array looking for all the
607  * supported base/actual page size combinations.  Each combination
608  * has a unique pagesize encoding (penc) value in the low bits of
609  * the LP field of the HPTE.  For actual page sizes less than 1MB,
610  * some of the upper LP bits are used for RPN bits, meaning that
611  * we need to fill in several entries in hpte_page_sizes[].
612  *
613  * In diagrammatic form, with r = RPN bits and z = page size bits:
614  *        PTE LP     actual page size
615  *    rrrr rrrz		>=8KB
616  *    rrrr rrzz		>=16KB
617  *    rrrr rzzz		>=32KB
618  *    rrrr zzzz		>=64KB
619  *    ...
620  *
621  * The zzzz bits are implementation-specific but are chosen so that
622  * no encoding for a larger page size uses the same value in its
623  * low-order N bits as the encoding for the 2^(12+N) byte page size
624  * (if it exists).
625  */
626 static void init_hpte_page_sizes(void)
627 {
628 	long int ap, bp;
629 	long int shift, penc;
630 
631 	for (bp = 0; bp < MMU_PAGE_COUNT; ++bp) {
632 		if (!mmu_psize_defs[bp].shift)
633 			continue;	/* not a supported page size */
634 		for (ap = bp; ap < MMU_PAGE_COUNT; ++ap) {
635 			penc = mmu_psize_defs[bp].penc[ap];
636 			if (penc == -1 || !mmu_psize_defs[ap].shift)
637 				continue;
638 			shift = mmu_psize_defs[ap].shift - LP_SHIFT;
639 			if (shift <= 0)
640 				continue;	/* should never happen */
641 			/*
642 			 * For page sizes less than 1MB, this loop
643 			 * replicates the entry for all possible values
644 			 * of the rrrr bits.
645 			 */
646 			while (penc < (1 << LP_BITS)) {
647 				hpte_page_sizes[penc] = (ap << 4) | bp;
648 				penc += 1 << shift;
649 			}
650 		}
651 	}
652 }
653 
654 static void __init htab_init_page_sizes(void)
655 {
656 	bool aligned = true;
657 	init_hpte_page_sizes();
658 
659 	if (!debug_pagealloc_enabled()) {
660 		/*
661 		 * Pick a size for the linear mapping. Currently, we only
662 		 * support 16M, 1M and 4K which is the default
663 		 */
664 		if (IS_ENABLED(CONFIG_STRICT_KERNEL_RWX) &&
665 		    (unsigned long)_stext % 0x1000000) {
666 			if (mmu_psize_defs[MMU_PAGE_16M].shift)
667 				pr_warn("Kernel not 16M aligned, disabling 16M linear map alignment\n");
668 			aligned = false;
669 		}
670 
671 		if (mmu_psize_defs[MMU_PAGE_16M].shift && aligned)
672 			mmu_linear_psize = MMU_PAGE_16M;
673 		else if (mmu_psize_defs[MMU_PAGE_1M].shift)
674 			mmu_linear_psize = MMU_PAGE_1M;
675 	}
676 
677 #ifdef CONFIG_PPC_64K_PAGES
678 	/*
679 	 * Pick a size for the ordinary pages. Default is 4K, we support
680 	 * 64K for user mappings and vmalloc if supported by the processor.
681 	 * We only use 64k for ioremap if the processor
682 	 * (and firmware) support cache-inhibited large pages.
683 	 * If not, we use 4k and set mmu_ci_restrictions so that
684 	 * hash_page knows to switch processes that use cache-inhibited
685 	 * mappings to 4k pages.
686 	 */
687 	if (mmu_psize_defs[MMU_PAGE_64K].shift) {
688 		mmu_virtual_psize = MMU_PAGE_64K;
689 		mmu_vmalloc_psize = MMU_PAGE_64K;
690 		if (mmu_linear_psize == MMU_PAGE_4K)
691 			mmu_linear_psize = MMU_PAGE_64K;
692 		if (mmu_has_feature(MMU_FTR_CI_LARGE_PAGE)) {
693 			/*
694 			 * When running on pSeries using 64k pages for ioremap
695 			 * would stop us accessing the HEA ethernet. So if we
696 			 * have the chance of ever seeing one, stay at 4k.
697 			 */
698 			if (!might_have_hea())
699 				mmu_io_psize = MMU_PAGE_64K;
700 		} else
701 			mmu_ci_restrictions = 1;
702 	}
703 #endif /* CONFIG_PPC_64K_PAGES */
704 
705 #ifdef CONFIG_SPARSEMEM_VMEMMAP
706 	/*
707 	 * We try to use 16M pages for vmemmap if that is supported
708 	 * and we have at least 1G of RAM at boot
709 	 */
710 	if (mmu_psize_defs[MMU_PAGE_16M].shift &&
711 	    memblock_phys_mem_size() >= 0x40000000)
712 		mmu_vmemmap_psize = MMU_PAGE_16M;
713 	else
714 		mmu_vmemmap_psize = mmu_virtual_psize;
715 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
716 
717 	printk(KERN_DEBUG "Page orders: linear mapping = %d, "
718 	       "virtual = %d, io = %d"
719 #ifdef CONFIG_SPARSEMEM_VMEMMAP
720 	       ", vmemmap = %d"
721 #endif
722 	       "\n",
723 	       mmu_psize_defs[mmu_linear_psize].shift,
724 	       mmu_psize_defs[mmu_virtual_psize].shift,
725 	       mmu_psize_defs[mmu_io_psize].shift
726 #ifdef CONFIG_SPARSEMEM_VMEMMAP
727 	       ,mmu_psize_defs[mmu_vmemmap_psize].shift
728 #endif
729 	       );
730 }
731 
732 static int __init htab_dt_scan_pftsize(unsigned long node,
733 				       const char *uname, int depth,
734 				       void *data)
735 {
736 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
737 	const __be32 *prop;
738 
739 	/* We are scanning "cpu" nodes only */
740 	if (type == NULL || strcmp(type, "cpu") != 0)
741 		return 0;
742 
743 	prop = of_get_flat_dt_prop(node, "ibm,pft-size", NULL);
744 	if (prop != NULL) {
745 		/* pft_size[0] is the NUMA CEC cookie */
746 		ppc64_pft_size = be32_to_cpu(prop[1]);
747 		return 1;
748 	}
749 	return 0;
750 }
751 
752 unsigned htab_shift_for_mem_size(unsigned long mem_size)
753 {
754 	unsigned memshift = __ilog2(mem_size);
755 	unsigned pshift = mmu_psize_defs[mmu_virtual_psize].shift;
756 	unsigned pteg_shift;
757 
758 	/* round mem_size up to next power of 2 */
759 	if ((1UL << memshift) < mem_size)
760 		memshift += 1;
761 
762 	/* aim for 2 pages / pteg */
763 	pteg_shift = memshift - (pshift + 1);
764 
765 	/*
766 	 * 2^11 PTEGS of 128 bytes each, ie. 2^18 bytes is the minimum htab
767 	 * size permitted by the architecture.
768 	 */
769 	return max(pteg_shift + 7, 18U);
770 }
771 
772 static unsigned long __init htab_get_table_size(void)
773 {
774 	/*
775 	 * If hash size isn't already provided by the platform, we try to
776 	 * retrieve it from the device-tree. If it's not there neither, we
777 	 * calculate it now based on the total RAM size
778 	 */
779 	if (ppc64_pft_size == 0)
780 		of_scan_flat_dt(htab_dt_scan_pftsize, NULL);
781 	if (ppc64_pft_size)
782 		return 1UL << ppc64_pft_size;
783 
784 	return 1UL << htab_shift_for_mem_size(memblock_phys_mem_size());
785 }
786 
787 #ifdef CONFIG_MEMORY_HOTPLUG
788 static int resize_hpt_for_hotplug(unsigned long new_mem_size)
789 {
790 	unsigned target_hpt_shift;
791 
792 	if (!mmu_hash_ops.resize_hpt)
793 		return 0;
794 
795 	target_hpt_shift = htab_shift_for_mem_size(new_mem_size);
796 
797 	/*
798 	 * To avoid lots of HPT resizes if memory size is fluctuating
799 	 * across a boundary, we deliberately have some hysterisis
800 	 * here: we immediately increase the HPT size if the target
801 	 * shift exceeds the current shift, but we won't attempt to
802 	 * reduce unless the target shift is at least 2 below the
803 	 * current shift
804 	 */
805 	if (target_hpt_shift > ppc64_pft_size ||
806 	    target_hpt_shift < ppc64_pft_size - 1)
807 		return mmu_hash_ops.resize_hpt(target_hpt_shift);
808 
809 	return 0;
810 }
811 
812 int hash__create_section_mapping(unsigned long start, unsigned long end,
813 				 int nid, pgprot_t prot)
814 {
815 	int rc;
816 
817 	if (end >= H_VMALLOC_START) {
818 		pr_warn("Outside the supported range\n");
819 		return -1;
820 	}
821 
822 	resize_hpt_for_hotplug(memblock_phys_mem_size());
823 
824 	rc = htab_bolt_mapping(start, end, __pa(start),
825 			       pgprot_val(prot), mmu_linear_psize,
826 			       mmu_kernel_ssize);
827 
828 	if (rc < 0) {
829 		int rc2 = htab_remove_mapping(start, end, mmu_linear_psize,
830 					      mmu_kernel_ssize);
831 		BUG_ON(rc2 && (rc2 != -ENOENT));
832 	}
833 	return rc;
834 }
835 
836 int hash__remove_section_mapping(unsigned long start, unsigned long end)
837 {
838 	int rc = htab_remove_mapping(start, end, mmu_linear_psize,
839 				     mmu_kernel_ssize);
840 	WARN_ON(rc < 0);
841 
842 	if (resize_hpt_for_hotplug(memblock_phys_mem_size()) == -ENOSPC)
843 		pr_warn("Hash collision while resizing HPT\n");
844 
845 	return rc;
846 }
847 #endif /* CONFIG_MEMORY_HOTPLUG */
848 
849 static void __init hash_init_partition_table(phys_addr_t hash_table,
850 					     unsigned long htab_size)
851 {
852 	mmu_partition_table_init();
853 
854 	/*
855 	 * PS field (VRMA page size) is not used for LPID 0, hence set to 0.
856 	 * For now, UPRT is 0 and we have no segment table.
857 	 */
858 	htab_size =  __ilog2(htab_size) - 18;
859 	mmu_partition_table_set_entry(0, hash_table | htab_size, 0, false);
860 	pr_info("Partition table %p\n", partition_tb);
861 }
862 
863 static void __init htab_initialize(void)
864 {
865 	unsigned long table;
866 	unsigned long pteg_count;
867 	unsigned long prot;
868 	unsigned long base = 0, size = 0;
869 	struct memblock_region *reg;
870 
871 	DBG(" -> htab_initialize()\n");
872 
873 	if (mmu_has_feature(MMU_FTR_1T_SEGMENT)) {
874 		mmu_kernel_ssize = MMU_SEGSIZE_1T;
875 		mmu_highuser_ssize = MMU_SEGSIZE_1T;
876 		printk(KERN_INFO "Using 1TB segments\n");
877 	}
878 
879 	if (stress_slb_enabled)
880 		static_branch_enable(&stress_slb_key);
881 
882 	/*
883 	 * Calculate the required size of the htab.  We want the number of
884 	 * PTEGs to equal one half the number of real pages.
885 	 */
886 	htab_size_bytes = htab_get_table_size();
887 	pteg_count = htab_size_bytes >> 7;
888 
889 	htab_hash_mask = pteg_count - 1;
890 
891 	if (firmware_has_feature(FW_FEATURE_LPAR) ||
892 	    firmware_has_feature(FW_FEATURE_PS3_LV1)) {
893 		/* Using a hypervisor which owns the htab */
894 		htab_address = NULL;
895 		_SDR1 = 0;
896 #ifdef CONFIG_FA_DUMP
897 		/*
898 		 * If firmware assisted dump is active firmware preserves
899 		 * the contents of htab along with entire partition memory.
900 		 * Clear the htab if firmware assisted dump is active so
901 		 * that we dont end up using old mappings.
902 		 */
903 		if (is_fadump_active() && mmu_hash_ops.hpte_clear_all)
904 			mmu_hash_ops.hpte_clear_all();
905 #endif
906 	} else {
907 		unsigned long limit = MEMBLOCK_ALLOC_ANYWHERE;
908 
909 #ifdef CONFIG_PPC_CELL
910 		/*
911 		 * Cell may require the hash table down low when using the
912 		 * Axon IOMMU in order to fit the dynamic region over it, see
913 		 * comments in cell/iommu.c
914 		 */
915 		if (fdt_subnode_offset(initial_boot_params, 0, "axon") > 0) {
916 			limit = 0x80000000;
917 			pr_info("Hash table forced below 2G for Axon IOMMU\n");
918 		}
919 #endif /* CONFIG_PPC_CELL */
920 
921 		table = memblock_phys_alloc_range(htab_size_bytes,
922 						  htab_size_bytes,
923 						  0, limit);
924 		if (!table)
925 			panic("ERROR: Failed to allocate %pa bytes below %pa\n",
926 			      &htab_size_bytes, &limit);
927 
928 		DBG("Hash table allocated at %lx, size: %lx\n", table,
929 		    htab_size_bytes);
930 
931 		htab_address = __va(table);
932 
933 		/* htab absolute addr + encoded htabsize */
934 		_SDR1 = table + __ilog2(htab_size_bytes) - 18;
935 
936 		/* Initialize the HPT with no entries */
937 		memset((void *)table, 0, htab_size_bytes);
938 
939 		if (!cpu_has_feature(CPU_FTR_ARCH_300))
940 			/* Set SDR1 */
941 			mtspr(SPRN_SDR1, _SDR1);
942 		else
943 			hash_init_partition_table(table, htab_size_bytes);
944 	}
945 
946 	prot = pgprot_val(PAGE_KERNEL);
947 
948 #ifdef CONFIG_DEBUG_PAGEALLOC
949 	if (debug_pagealloc_enabled()) {
950 		linear_map_hash_count = memblock_end_of_DRAM() >> PAGE_SHIFT;
951 		linear_map_hash_slots = memblock_alloc_try_nid(
952 				linear_map_hash_count, 1, MEMBLOCK_LOW_LIMIT,
953 				ppc64_rma_size,	NUMA_NO_NODE);
954 		if (!linear_map_hash_slots)
955 			panic("%s: Failed to allocate %lu bytes max_addr=%pa\n",
956 			      __func__, linear_map_hash_count, &ppc64_rma_size);
957 	}
958 #endif /* CONFIG_DEBUG_PAGEALLOC */
959 
960 	/* create bolted the linear mapping in the hash table */
961 	for_each_memblock(memory, reg) {
962 		base = (unsigned long)__va(reg->base);
963 		size = reg->size;
964 
965 		DBG("creating mapping for region: %lx..%lx (prot: %lx)\n",
966 		    base, size, prot);
967 
968 		if ((base + size) >= H_VMALLOC_START) {
969 			pr_warn("Outside the supported range\n");
970 			continue;
971 		}
972 
973 		BUG_ON(htab_bolt_mapping(base, base + size, __pa(base),
974 				prot, mmu_linear_psize, mmu_kernel_ssize));
975 	}
976 	memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
977 
978 	/*
979 	 * If we have a memory_limit and we've allocated TCEs then we need to
980 	 * explicitly map the TCE area at the top of RAM. We also cope with the
981 	 * case that the TCEs start below memory_limit.
982 	 * tce_alloc_start/end are 16MB aligned so the mapping should work
983 	 * for either 4K or 16MB pages.
984 	 */
985 	if (tce_alloc_start) {
986 		tce_alloc_start = (unsigned long)__va(tce_alloc_start);
987 		tce_alloc_end = (unsigned long)__va(tce_alloc_end);
988 
989 		if (base + size >= tce_alloc_start)
990 			tce_alloc_start = base + size + 1;
991 
992 		BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end,
993 					 __pa(tce_alloc_start), prot,
994 					 mmu_linear_psize, mmu_kernel_ssize));
995 	}
996 
997 
998 	DBG(" <- htab_initialize()\n");
999 }
1000 #undef KB
1001 #undef MB
1002 
1003 void __init hash__early_init_devtree(void)
1004 {
1005 	/* Initialize segment sizes */
1006 	of_scan_flat_dt(htab_dt_scan_seg_sizes, NULL);
1007 
1008 	/* Initialize page sizes */
1009 	htab_scan_page_sizes();
1010 }
1011 
1012 static struct hash_mm_context init_hash_mm_context;
1013 void __init hash__early_init_mmu(void)
1014 {
1015 #ifndef CONFIG_PPC_64K_PAGES
1016 	/*
1017 	 * We have code in __hash_page_4K() and elsewhere, which assumes it can
1018 	 * do the following:
1019 	 *   new_pte |= (slot << H_PAGE_F_GIX_SHIFT) & (H_PAGE_F_SECOND | H_PAGE_F_GIX);
1020 	 *
1021 	 * Where the slot number is between 0-15, and values of 8-15 indicate
1022 	 * the secondary bucket. For that code to work H_PAGE_F_SECOND and
1023 	 * H_PAGE_F_GIX must occupy four contiguous bits in the PTE, and
1024 	 * H_PAGE_F_SECOND must be placed above H_PAGE_F_GIX. Assert that here
1025 	 * with a BUILD_BUG_ON().
1026 	 */
1027 	BUILD_BUG_ON(H_PAGE_F_SECOND != (1ul  << (H_PAGE_F_GIX_SHIFT + 3)));
1028 #endif /* CONFIG_PPC_64K_PAGES */
1029 
1030 	htab_init_page_sizes();
1031 
1032 	/*
1033 	 * initialize page table size
1034 	 */
1035 	__pte_frag_nr = H_PTE_FRAG_NR;
1036 	__pte_frag_size_shift = H_PTE_FRAG_SIZE_SHIFT;
1037 	__pmd_frag_nr = H_PMD_FRAG_NR;
1038 	__pmd_frag_size_shift = H_PMD_FRAG_SIZE_SHIFT;
1039 
1040 	__pte_index_size = H_PTE_INDEX_SIZE;
1041 	__pmd_index_size = H_PMD_INDEX_SIZE;
1042 	__pud_index_size = H_PUD_INDEX_SIZE;
1043 	__pgd_index_size = H_PGD_INDEX_SIZE;
1044 	__pud_cache_index = H_PUD_CACHE_INDEX;
1045 	__pte_table_size = H_PTE_TABLE_SIZE;
1046 	__pmd_table_size = H_PMD_TABLE_SIZE;
1047 	__pud_table_size = H_PUD_TABLE_SIZE;
1048 	__pgd_table_size = H_PGD_TABLE_SIZE;
1049 	/*
1050 	 * 4k use hugepd format, so for hash set then to
1051 	 * zero
1052 	 */
1053 	__pmd_val_bits = HASH_PMD_VAL_BITS;
1054 	__pud_val_bits = HASH_PUD_VAL_BITS;
1055 	__pgd_val_bits = HASH_PGD_VAL_BITS;
1056 
1057 	__kernel_virt_start = H_KERN_VIRT_START;
1058 	__vmalloc_start = H_VMALLOC_START;
1059 	__vmalloc_end = H_VMALLOC_END;
1060 	__kernel_io_start = H_KERN_IO_START;
1061 	__kernel_io_end = H_KERN_IO_END;
1062 	vmemmap = (struct page *)H_VMEMMAP_START;
1063 	ioremap_bot = IOREMAP_BASE;
1064 
1065 #ifdef CONFIG_PCI
1066 	pci_io_base = ISA_IO_BASE;
1067 #endif
1068 
1069 	/* Select appropriate backend */
1070 	if (firmware_has_feature(FW_FEATURE_PS3_LV1))
1071 		ps3_early_mm_init();
1072 	else if (firmware_has_feature(FW_FEATURE_LPAR))
1073 		hpte_init_pseries();
1074 	else if (IS_ENABLED(CONFIG_PPC_NATIVE))
1075 		hpte_init_native();
1076 
1077 	if (!mmu_hash_ops.hpte_insert)
1078 		panic("hash__early_init_mmu: No MMU hash ops defined!\n");
1079 
1080 	/*
1081 	 * Initialize the MMU Hash table and create the linear mapping
1082 	 * of memory. Has to be done before SLB initialization as this is
1083 	 * currently where the page size encoding is obtained.
1084 	 */
1085 	htab_initialize();
1086 
1087 	init_mm.context.hash_context = &init_hash_mm_context;
1088 	mm_ctx_set_slb_addr_limit(&init_mm.context, SLB_ADDR_LIMIT_DEFAULT);
1089 
1090 	pr_info("Initializing hash mmu with SLB\n");
1091 	/* Initialize SLB management */
1092 	slb_initialize();
1093 
1094 	if (cpu_has_feature(CPU_FTR_ARCH_206)
1095 			&& cpu_has_feature(CPU_FTR_HVMODE))
1096 		tlbiel_all();
1097 }
1098 
1099 #ifdef CONFIG_SMP
1100 void hash__early_init_mmu_secondary(void)
1101 {
1102 	/* Initialize hash table for that CPU */
1103 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
1104 
1105 		if (!cpu_has_feature(CPU_FTR_ARCH_300))
1106 			mtspr(SPRN_SDR1, _SDR1);
1107 		else
1108 			set_ptcr_when_no_uv(__pa(partition_tb) |
1109 					    (PATB_SIZE_SHIFT - 12));
1110 	}
1111 	/* Initialize SLB */
1112 	slb_initialize();
1113 
1114 	if (cpu_has_feature(CPU_FTR_ARCH_206)
1115 			&& cpu_has_feature(CPU_FTR_HVMODE))
1116 		tlbiel_all();
1117 
1118 #ifdef CONFIG_PPC_MEM_KEYS
1119 	if (mmu_has_feature(MMU_FTR_PKEY))
1120 		mtspr(SPRN_UAMOR, default_uamor);
1121 #endif
1122 }
1123 #endif /* CONFIG_SMP */
1124 
1125 /*
1126  * Called by asm hashtable.S for doing lazy icache flush
1127  */
1128 unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap)
1129 {
1130 	struct page *page;
1131 
1132 	if (!pfn_valid(pte_pfn(pte)))
1133 		return pp;
1134 
1135 	page = pte_page(pte);
1136 
1137 	/* page is dirty */
1138 	if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) {
1139 		if (trap == 0x400) {
1140 			flush_dcache_icache_page(page);
1141 			set_bit(PG_arch_1, &page->flags);
1142 		} else
1143 			pp |= HPTE_R_N;
1144 	}
1145 	return pp;
1146 }
1147 
1148 #ifdef CONFIG_PPC_MM_SLICES
1149 static unsigned int get_paca_psize(unsigned long addr)
1150 {
1151 	unsigned char *psizes;
1152 	unsigned long index, mask_index;
1153 
1154 	if (addr < SLICE_LOW_TOP) {
1155 		psizes = get_paca()->mm_ctx_low_slices_psize;
1156 		index = GET_LOW_SLICE_INDEX(addr);
1157 	} else {
1158 		psizes = get_paca()->mm_ctx_high_slices_psize;
1159 		index = GET_HIGH_SLICE_INDEX(addr);
1160 	}
1161 	mask_index = index & 0x1;
1162 	return (psizes[index >> 1] >> (mask_index * 4)) & 0xF;
1163 }
1164 
1165 #else
1166 unsigned int get_paca_psize(unsigned long addr)
1167 {
1168 	return get_paca()->mm_ctx_user_psize;
1169 }
1170 #endif
1171 
1172 /*
1173  * Demote a segment to using 4k pages.
1174  * For now this makes the whole process use 4k pages.
1175  */
1176 #ifdef CONFIG_PPC_64K_PAGES
1177 void demote_segment_4k(struct mm_struct *mm, unsigned long addr)
1178 {
1179 	if (get_slice_psize(mm, addr) == MMU_PAGE_4K)
1180 		return;
1181 	slice_set_range_psize(mm, addr, 1, MMU_PAGE_4K);
1182 	copro_flush_all_slbs(mm);
1183 	if ((get_paca_psize(addr) != MMU_PAGE_4K) && (current->mm == mm)) {
1184 
1185 		copy_mm_to_paca(mm);
1186 		slb_flush_and_restore_bolted();
1187 	}
1188 }
1189 #endif /* CONFIG_PPC_64K_PAGES */
1190 
1191 #ifdef CONFIG_PPC_SUBPAGE_PROT
1192 /*
1193  * This looks up a 2-bit protection code for a 4k subpage of a 64k page.
1194  * Userspace sets the subpage permissions using the subpage_prot system call.
1195  *
1196  * Result is 0: full permissions, _PAGE_RW: read-only,
1197  * _PAGE_RWX: no access.
1198  */
1199 static int subpage_protection(struct mm_struct *mm, unsigned long ea)
1200 {
1201 	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
1202 	u32 spp = 0;
1203 	u32 **sbpm, *sbpp;
1204 
1205 	if (!spt)
1206 		return 0;
1207 
1208 	if (ea >= spt->maxaddr)
1209 		return 0;
1210 	if (ea < 0x100000000UL) {
1211 		/* addresses below 4GB use spt->low_prot */
1212 		sbpm = spt->low_prot;
1213 	} else {
1214 		sbpm = spt->protptrs[ea >> SBP_L3_SHIFT];
1215 		if (!sbpm)
1216 			return 0;
1217 	}
1218 	sbpp = sbpm[(ea >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
1219 	if (!sbpp)
1220 		return 0;
1221 	spp = sbpp[(ea >> PAGE_SHIFT) & (SBP_L1_COUNT - 1)];
1222 
1223 	/* extract 2-bit bitfield for this 4k subpage */
1224 	spp >>= 30 - 2 * ((ea >> 12) & 0xf);
1225 
1226 	/*
1227 	 * 0 -> full premission
1228 	 * 1 -> Read only
1229 	 * 2 -> no access.
1230 	 * We return the flag that need to be cleared.
1231 	 */
1232 	spp = ((spp & 2) ? _PAGE_RWX : 0) | ((spp & 1) ? _PAGE_WRITE : 0);
1233 	return spp;
1234 }
1235 
1236 #else /* CONFIG_PPC_SUBPAGE_PROT */
1237 static inline int subpage_protection(struct mm_struct *mm, unsigned long ea)
1238 {
1239 	return 0;
1240 }
1241 #endif
1242 
1243 void hash_failure_debug(unsigned long ea, unsigned long access,
1244 			unsigned long vsid, unsigned long trap,
1245 			int ssize, int psize, int lpsize, unsigned long pte)
1246 {
1247 	if (!printk_ratelimit())
1248 		return;
1249 	pr_info("mm: Hashing failure ! EA=0x%lx access=0x%lx current=%s\n",
1250 		ea, access, current->comm);
1251 	pr_info("    trap=0x%lx vsid=0x%lx ssize=%d base psize=%d psize %d pte=0x%lx\n",
1252 		trap, vsid, ssize, psize, lpsize, pte);
1253 }
1254 
1255 static void check_paca_psize(unsigned long ea, struct mm_struct *mm,
1256 			     int psize, bool user_region)
1257 {
1258 	if (user_region) {
1259 		if (psize != get_paca_psize(ea)) {
1260 			copy_mm_to_paca(mm);
1261 			slb_flush_and_restore_bolted();
1262 		}
1263 	} else if (get_paca()->vmalloc_sllp !=
1264 		   mmu_psize_defs[mmu_vmalloc_psize].sllp) {
1265 		get_paca()->vmalloc_sllp =
1266 			mmu_psize_defs[mmu_vmalloc_psize].sllp;
1267 		slb_vmalloc_update();
1268 	}
1269 }
1270 
1271 /*
1272  * Result code is:
1273  *  0 - handled
1274  *  1 - normal page fault
1275  * -1 - critical hash insertion error
1276  * -2 - access not permitted by subpage protection mechanism
1277  */
1278 int hash_page_mm(struct mm_struct *mm, unsigned long ea,
1279 		 unsigned long access, unsigned long trap,
1280 		 unsigned long flags)
1281 {
1282 	bool is_thp;
1283 	enum ctx_state prev_state = exception_enter();
1284 	pgd_t *pgdir;
1285 	unsigned long vsid;
1286 	pte_t *ptep;
1287 	unsigned hugeshift;
1288 	int rc, user_region = 0;
1289 	int psize, ssize;
1290 
1291 	DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n",
1292 		ea, access, trap);
1293 	trace_hash_fault(ea, access, trap);
1294 
1295 	/* Get region & vsid */
1296 	switch (get_region_id(ea)) {
1297 	case USER_REGION_ID:
1298 		user_region = 1;
1299 		if (! mm) {
1300 			DBG_LOW(" user region with no mm !\n");
1301 			rc = 1;
1302 			goto bail;
1303 		}
1304 		psize = get_slice_psize(mm, ea);
1305 		ssize = user_segment_size(ea);
1306 		vsid = get_user_vsid(&mm->context, ea, ssize);
1307 		break;
1308 	case VMALLOC_REGION_ID:
1309 		vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
1310 		psize = mmu_vmalloc_psize;
1311 		ssize = mmu_kernel_ssize;
1312 		break;
1313 
1314 	case IO_REGION_ID:
1315 		vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
1316 		psize = mmu_io_psize;
1317 		ssize = mmu_kernel_ssize;
1318 		break;
1319 	default:
1320 		/*
1321 		 * Not a valid range
1322 		 * Send the problem up to do_page_fault()
1323 		 */
1324 		rc = 1;
1325 		goto bail;
1326 	}
1327 	DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid);
1328 
1329 	/* Bad address. */
1330 	if (!vsid) {
1331 		DBG_LOW("Bad address!\n");
1332 		rc = 1;
1333 		goto bail;
1334 	}
1335 	/* Get pgdir */
1336 	pgdir = mm->pgd;
1337 	if (pgdir == NULL) {
1338 		rc = 1;
1339 		goto bail;
1340 	}
1341 
1342 	/* Check CPU locality */
1343 	if (user_region && mm_is_thread_local(mm))
1344 		flags |= HPTE_LOCAL_UPDATE;
1345 
1346 #ifndef CONFIG_PPC_64K_PAGES
1347 	/*
1348 	 * If we use 4K pages and our psize is not 4K, then we might
1349 	 * be hitting a special driver mapping, and need to align the
1350 	 * address before we fetch the PTE.
1351 	 *
1352 	 * It could also be a hugepage mapping, in which case this is
1353 	 * not necessary, but it's not harmful, either.
1354 	 */
1355 	if (psize != MMU_PAGE_4K)
1356 		ea &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
1357 #endif /* CONFIG_PPC_64K_PAGES */
1358 
1359 	/* Get PTE and page size from page tables */
1360 	ptep = find_linux_pte(pgdir, ea, &is_thp, &hugeshift);
1361 	if (ptep == NULL || !pte_present(*ptep)) {
1362 		DBG_LOW(" no PTE !\n");
1363 		rc = 1;
1364 		goto bail;
1365 	}
1366 
1367 	/*
1368 	 * Add _PAGE_PRESENT to the required access perm. If there are parallel
1369 	 * updates to the pte that can possibly clear _PAGE_PTE, catch that too.
1370 	 *
1371 	 * We can safely use the return pte address in rest of the function
1372 	 * because we do set H_PAGE_BUSY which prevents further updates to pte
1373 	 * from generic code.
1374 	 */
1375 	access |= _PAGE_PRESENT | _PAGE_PTE;
1376 
1377 	/*
1378 	 * Pre-check access permissions (will be re-checked atomically
1379 	 * in __hash_page_XX but this pre-check is a fast path
1380 	 */
1381 	if (!check_pte_access(access, pte_val(*ptep))) {
1382 		DBG_LOW(" no access !\n");
1383 		rc = 1;
1384 		goto bail;
1385 	}
1386 
1387 	if (hugeshift) {
1388 		if (is_thp)
1389 			rc = __hash_page_thp(ea, access, vsid, (pmd_t *)ptep,
1390 					     trap, flags, ssize, psize);
1391 #ifdef CONFIG_HUGETLB_PAGE
1392 		else
1393 			rc = __hash_page_huge(ea, access, vsid, ptep, trap,
1394 					      flags, ssize, hugeshift, psize);
1395 #else
1396 		else {
1397 			/*
1398 			 * if we have hugeshift, and is not transhuge with
1399 			 * hugetlb disabled, something is really wrong.
1400 			 */
1401 			rc = 1;
1402 			WARN_ON(1);
1403 		}
1404 #endif
1405 		if (current->mm == mm)
1406 			check_paca_psize(ea, mm, psize, user_region);
1407 
1408 		goto bail;
1409 	}
1410 
1411 #ifndef CONFIG_PPC_64K_PAGES
1412 	DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep));
1413 #else
1414 	DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep),
1415 		pte_val(*(ptep + PTRS_PER_PTE)));
1416 #endif
1417 	/* Do actual hashing */
1418 #ifdef CONFIG_PPC_64K_PAGES
1419 	/* If H_PAGE_4K_PFN is set, make sure this is a 4k segment */
1420 	if ((pte_val(*ptep) & H_PAGE_4K_PFN) && psize == MMU_PAGE_64K) {
1421 		demote_segment_4k(mm, ea);
1422 		psize = MMU_PAGE_4K;
1423 	}
1424 
1425 	/*
1426 	 * If this PTE is non-cacheable and we have restrictions on
1427 	 * using non cacheable large pages, then we switch to 4k
1428 	 */
1429 	if (mmu_ci_restrictions && psize == MMU_PAGE_64K && pte_ci(*ptep)) {
1430 		if (user_region) {
1431 			demote_segment_4k(mm, ea);
1432 			psize = MMU_PAGE_4K;
1433 		} else if (ea < VMALLOC_END) {
1434 			/*
1435 			 * some driver did a non-cacheable mapping
1436 			 * in vmalloc space, so switch vmalloc
1437 			 * to 4k pages
1438 			 */
1439 			printk(KERN_ALERT "Reducing vmalloc segment "
1440 			       "to 4kB pages because of "
1441 			       "non-cacheable mapping\n");
1442 			psize = mmu_vmalloc_psize = MMU_PAGE_4K;
1443 			copro_flush_all_slbs(mm);
1444 		}
1445 	}
1446 
1447 #endif /* CONFIG_PPC_64K_PAGES */
1448 
1449 	if (current->mm == mm)
1450 		check_paca_psize(ea, mm, psize, user_region);
1451 
1452 #ifdef CONFIG_PPC_64K_PAGES
1453 	if (psize == MMU_PAGE_64K)
1454 		rc = __hash_page_64K(ea, access, vsid, ptep, trap,
1455 				     flags, ssize);
1456 	else
1457 #endif /* CONFIG_PPC_64K_PAGES */
1458 	{
1459 		int spp = subpage_protection(mm, ea);
1460 		if (access & spp)
1461 			rc = -2;
1462 		else
1463 			rc = __hash_page_4K(ea, access, vsid, ptep, trap,
1464 					    flags, ssize, spp);
1465 	}
1466 
1467 	/*
1468 	 * Dump some info in case of hash insertion failure, they should
1469 	 * never happen so it is really useful to know if/when they do
1470 	 */
1471 	if (rc == -1)
1472 		hash_failure_debug(ea, access, vsid, trap, ssize, psize,
1473 				   psize, pte_val(*ptep));
1474 #ifndef CONFIG_PPC_64K_PAGES
1475 	DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep));
1476 #else
1477 	DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep),
1478 		pte_val(*(ptep + PTRS_PER_PTE)));
1479 #endif
1480 	DBG_LOW(" -> rc=%d\n", rc);
1481 
1482 bail:
1483 	exception_exit(prev_state);
1484 	return rc;
1485 }
1486 EXPORT_SYMBOL_GPL(hash_page_mm);
1487 
1488 int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
1489 	      unsigned long dsisr)
1490 {
1491 	unsigned long flags = 0;
1492 	struct mm_struct *mm = current->mm;
1493 
1494 	if ((get_region_id(ea) == VMALLOC_REGION_ID) ||
1495 	    (get_region_id(ea) == IO_REGION_ID))
1496 		mm = &init_mm;
1497 
1498 	if (dsisr & DSISR_NOHPTE)
1499 		flags |= HPTE_NOHPTE_UPDATE;
1500 
1501 	return hash_page_mm(mm, ea, access, trap, flags);
1502 }
1503 EXPORT_SYMBOL_GPL(hash_page);
1504 
1505 int __hash_page(unsigned long trap, unsigned long ea, unsigned long dsisr,
1506 		unsigned long msr)
1507 {
1508 	unsigned long access = _PAGE_PRESENT | _PAGE_READ;
1509 	unsigned long flags = 0;
1510 	struct mm_struct *mm = current->mm;
1511 	unsigned int region_id = get_region_id(ea);
1512 
1513 	if ((region_id == VMALLOC_REGION_ID) || (region_id == IO_REGION_ID))
1514 		mm = &init_mm;
1515 
1516 	if (dsisr & DSISR_NOHPTE)
1517 		flags |= HPTE_NOHPTE_UPDATE;
1518 
1519 	if (dsisr & DSISR_ISSTORE)
1520 		access |= _PAGE_WRITE;
1521 	/*
1522 	 * We set _PAGE_PRIVILEGED only when
1523 	 * kernel mode access kernel space.
1524 	 *
1525 	 * _PAGE_PRIVILEGED is NOT set
1526 	 * 1) when kernel mode access user space
1527 	 * 2) user space access kernel space.
1528 	 */
1529 	access |= _PAGE_PRIVILEGED;
1530 	if ((msr & MSR_PR) || (region_id == USER_REGION_ID))
1531 		access &= ~_PAGE_PRIVILEGED;
1532 
1533 	if (trap == 0x400)
1534 		access |= _PAGE_EXEC;
1535 
1536 	return hash_page_mm(mm, ea, access, trap, flags);
1537 }
1538 
1539 #ifdef CONFIG_PPC_MM_SLICES
1540 static bool should_hash_preload(struct mm_struct *mm, unsigned long ea)
1541 {
1542 	int psize = get_slice_psize(mm, ea);
1543 
1544 	/* We only prefault standard pages for now */
1545 	if (unlikely(psize != mm_ctx_user_psize(&mm->context)))
1546 		return false;
1547 
1548 	/*
1549 	 * Don't prefault if subpage protection is enabled for the EA.
1550 	 */
1551 	if (unlikely((psize == MMU_PAGE_4K) && subpage_protection(mm, ea)))
1552 		return false;
1553 
1554 	return true;
1555 }
1556 #else
1557 static bool should_hash_preload(struct mm_struct *mm, unsigned long ea)
1558 {
1559 	return true;
1560 }
1561 #endif
1562 
1563 static void hash_preload(struct mm_struct *mm, pte_t *ptep, unsigned long ea,
1564 			 bool is_exec, unsigned long trap)
1565 {
1566 	unsigned long vsid;
1567 	pgd_t *pgdir;
1568 	int rc, ssize, update_flags = 0;
1569 	unsigned long access = _PAGE_PRESENT | _PAGE_READ | (is_exec ? _PAGE_EXEC : 0);
1570 	unsigned long flags;
1571 
1572 	BUG_ON(get_region_id(ea) != USER_REGION_ID);
1573 
1574 	if (!should_hash_preload(mm, ea))
1575 		return;
1576 
1577 	DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx,"
1578 		" trap=%lx\n", mm, mm->pgd, ea, access, trap);
1579 
1580 	/* Get Linux PTE if available */
1581 	pgdir = mm->pgd;
1582 	if (pgdir == NULL)
1583 		return;
1584 
1585 	/* Get VSID */
1586 	ssize = user_segment_size(ea);
1587 	vsid = get_user_vsid(&mm->context, ea, ssize);
1588 	if (!vsid)
1589 		return;
1590 
1591 #ifdef CONFIG_PPC_64K_PAGES
1592 	/* If either H_PAGE_4K_PFN or cache inhibited is set (and we are on
1593 	 * a 64K kernel), then we don't preload, hash_page() will take
1594 	 * care of it once we actually try to access the page.
1595 	 * That way we don't have to duplicate all of the logic for segment
1596 	 * page size demotion here
1597 	 * Called with  PTL held, hence can be sure the value won't change in
1598 	 * between.
1599 	 */
1600 	if ((pte_val(*ptep) & H_PAGE_4K_PFN) || pte_ci(*ptep))
1601 		return;
1602 #endif /* CONFIG_PPC_64K_PAGES */
1603 
1604 	/*
1605 	 * __hash_page_* must run with interrupts off, as it sets the
1606 	 * H_PAGE_BUSY bit. It's possible for perf interrupts to hit at any
1607 	 * time and may take a hash fault reading the user stack, see
1608 	 * read_user_stack_slow() in the powerpc/perf code.
1609 	 *
1610 	 * If that takes a hash fault on the same page as we lock here, it
1611 	 * will bail out when seeing H_PAGE_BUSY set, and retry the access
1612 	 * leading to an infinite loop.
1613 	 *
1614 	 * Disabling interrupts here does not prevent perf interrupts, but it
1615 	 * will prevent them taking hash faults (see the NMI test in
1616 	 * do_hash_page), then read_user_stack's copy_from_user_nofault will
1617 	 * fail and perf will fall back to read_user_stack_slow(), which
1618 	 * walks the Linux page tables.
1619 	 *
1620 	 * Interrupts must also be off for the duration of the
1621 	 * mm_is_thread_local test and update, to prevent preempt running the
1622 	 * mm on another CPU (XXX: this may be racy vs kthread_use_mm).
1623 	 */
1624 	local_irq_save(flags);
1625 
1626 	/* Is that local to this CPU ? */
1627 	if (mm_is_thread_local(mm))
1628 		update_flags |= HPTE_LOCAL_UPDATE;
1629 
1630 	/* Hash it in */
1631 #ifdef CONFIG_PPC_64K_PAGES
1632 	if (mm_ctx_user_psize(&mm->context) == MMU_PAGE_64K)
1633 		rc = __hash_page_64K(ea, access, vsid, ptep, trap,
1634 				     update_flags, ssize);
1635 	else
1636 #endif /* CONFIG_PPC_64K_PAGES */
1637 		rc = __hash_page_4K(ea, access, vsid, ptep, trap, update_flags,
1638 				    ssize, subpage_protection(mm, ea));
1639 
1640 	/* Dump some info in case of hash insertion failure, they should
1641 	 * never happen so it is really useful to know if/when they do
1642 	 */
1643 	if (rc == -1)
1644 		hash_failure_debug(ea, access, vsid, trap, ssize,
1645 				   mm_ctx_user_psize(&mm->context),
1646 				   mm_ctx_user_psize(&mm->context),
1647 				   pte_val(*ptep));
1648 
1649 	local_irq_restore(flags);
1650 }
1651 
1652 /*
1653  * This is called at the end of handling a user page fault, when the
1654  * fault has been handled by updating a PTE in the linux page tables.
1655  * We use it to preload an HPTE into the hash table corresponding to
1656  * the updated linux PTE.
1657  *
1658  * This must always be called with the pte lock held.
1659  */
1660 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
1661 		      pte_t *ptep)
1662 {
1663 	/*
1664 	 * We don't need to worry about _PAGE_PRESENT here because we are
1665 	 * called with either mm->page_table_lock held or ptl lock held
1666 	 */
1667 	unsigned long trap;
1668 	bool is_exec;
1669 
1670 	if (radix_enabled())
1671 		return;
1672 
1673 	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
1674 	if (!pte_young(*ptep) || address >= TASK_SIZE)
1675 		return;
1676 
1677 	/*
1678 	 * We try to figure out if we are coming from an instruction
1679 	 * access fault and pass that down to __hash_page so we avoid
1680 	 * double-faulting on execution of fresh text. We have to test
1681 	 * for regs NULL since init will get here first thing at boot.
1682 	 *
1683 	 * We also avoid filling the hash if not coming from a fault.
1684 	 */
1685 
1686 	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
1687 	switch (trap) {
1688 	case 0x300:
1689 		is_exec = false;
1690 		break;
1691 	case 0x400:
1692 		is_exec = true;
1693 		break;
1694 	default:
1695 		return;
1696 	}
1697 
1698 	hash_preload(vma->vm_mm, ptep, address, is_exec, trap);
1699 }
1700 
1701 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1702 static inline void tm_flush_hash_page(int local)
1703 {
1704 	/*
1705 	 * Transactions are not aborted by tlbiel, only tlbie. Without, syncing a
1706 	 * page back to a block device w/PIO could pick up transactional data
1707 	 * (bad!) so we force an abort here. Before the sync the page will be
1708 	 * made read-only, which will flush_hash_page. BIG ISSUE here: if the
1709 	 * kernel uses a page from userspace without unmapping it first, it may
1710 	 * see the speculated version.
1711 	 */
1712 	if (local && cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
1713 	    MSR_TM_ACTIVE(current->thread.regs->msr)) {
1714 		tm_enable();
1715 		tm_abort(TM_CAUSE_TLBI);
1716 	}
1717 }
1718 #else
1719 static inline void tm_flush_hash_page(int local)
1720 {
1721 }
1722 #endif
1723 
1724 /*
1725  * Return the global hash slot, corresponding to the given PTE, which contains
1726  * the HPTE.
1727  */
1728 unsigned long pte_get_hash_gslot(unsigned long vpn, unsigned long shift,
1729 		int ssize, real_pte_t rpte, unsigned int subpg_index)
1730 {
1731 	unsigned long hash, gslot, hidx;
1732 
1733 	hash = hpt_hash(vpn, shift, ssize);
1734 	hidx = __rpte_to_hidx(rpte, subpg_index);
1735 	if (hidx & _PTEIDX_SECONDARY)
1736 		hash = ~hash;
1737 	gslot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1738 	gslot += hidx & _PTEIDX_GROUP_IX;
1739 	return gslot;
1740 }
1741 
1742 void flush_hash_page(unsigned long vpn, real_pte_t pte, int psize, int ssize,
1743 		     unsigned long flags)
1744 {
1745 	unsigned long index, shift, gslot;
1746 	int local = flags & HPTE_LOCAL_UPDATE;
1747 
1748 	DBG_LOW("flush_hash_page(vpn=%016lx)\n", vpn);
1749 	pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1750 		gslot = pte_get_hash_gslot(vpn, shift, ssize, pte, index);
1751 		DBG_LOW(" sub %ld: gslot=%lx\n", index, gslot);
1752 		/*
1753 		 * We use same base page size and actual psize, because we don't
1754 		 * use these functions for hugepage
1755 		 */
1756 		mmu_hash_ops.hpte_invalidate(gslot, vpn, psize, psize,
1757 					     ssize, local);
1758 	} pte_iterate_hashed_end();
1759 
1760 	tm_flush_hash_page(local);
1761 }
1762 
1763 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
1764 void flush_hash_hugepage(unsigned long vsid, unsigned long addr,
1765 			 pmd_t *pmdp, unsigned int psize, int ssize,
1766 			 unsigned long flags)
1767 {
1768 	int i, max_hpte_count, valid;
1769 	unsigned long s_addr;
1770 	unsigned char *hpte_slot_array;
1771 	unsigned long hidx, shift, vpn, hash, slot;
1772 	int local = flags & HPTE_LOCAL_UPDATE;
1773 
1774 	s_addr = addr & HPAGE_PMD_MASK;
1775 	hpte_slot_array = get_hpte_slot_array(pmdp);
1776 	/*
1777 	 * IF we try to do a HUGE PTE update after a withdraw is done.
1778 	 * we will find the below NULL. This happens when we do
1779 	 * split_huge_pmd
1780 	 */
1781 	if (!hpte_slot_array)
1782 		return;
1783 
1784 	if (mmu_hash_ops.hugepage_invalidate) {
1785 		mmu_hash_ops.hugepage_invalidate(vsid, s_addr, hpte_slot_array,
1786 						 psize, ssize, local);
1787 		goto tm_abort;
1788 	}
1789 	/*
1790 	 * No bluk hpte removal support, invalidate each entry
1791 	 */
1792 	shift = mmu_psize_defs[psize].shift;
1793 	max_hpte_count = HPAGE_PMD_SIZE >> shift;
1794 	for (i = 0; i < max_hpte_count; i++) {
1795 		/*
1796 		 * 8 bits per each hpte entries
1797 		 * 000| [ secondary group (one bit) | hidx (3 bits) | valid bit]
1798 		 */
1799 		valid = hpte_valid(hpte_slot_array, i);
1800 		if (!valid)
1801 			continue;
1802 		hidx =  hpte_hash_index(hpte_slot_array, i);
1803 
1804 		/* get the vpn */
1805 		addr = s_addr + (i * (1ul << shift));
1806 		vpn = hpt_vpn(addr, vsid, ssize);
1807 		hash = hpt_hash(vpn, shift, ssize);
1808 		if (hidx & _PTEIDX_SECONDARY)
1809 			hash = ~hash;
1810 
1811 		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1812 		slot += hidx & _PTEIDX_GROUP_IX;
1813 		mmu_hash_ops.hpte_invalidate(slot, vpn, psize,
1814 					     MMU_PAGE_16M, ssize, local);
1815 	}
1816 tm_abort:
1817 	tm_flush_hash_page(local);
1818 }
1819 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1820 
1821 void flush_hash_range(unsigned long number, int local)
1822 {
1823 	if (mmu_hash_ops.flush_hash_range)
1824 		mmu_hash_ops.flush_hash_range(number, local);
1825 	else {
1826 		int i;
1827 		struct ppc64_tlb_batch *batch =
1828 			this_cpu_ptr(&ppc64_tlb_batch);
1829 
1830 		for (i = 0; i < number; i++)
1831 			flush_hash_page(batch->vpn[i], batch->pte[i],
1832 					batch->psize, batch->ssize, local);
1833 	}
1834 }
1835 
1836 /*
1837  * low_hash_fault is called when we the low level hash code failed
1838  * to instert a PTE due to an hypervisor error
1839  */
1840 void low_hash_fault(struct pt_regs *regs, unsigned long address, int rc)
1841 {
1842 	enum ctx_state prev_state = exception_enter();
1843 
1844 	if (user_mode(regs)) {
1845 #ifdef CONFIG_PPC_SUBPAGE_PROT
1846 		if (rc == -2)
1847 			_exception(SIGSEGV, regs, SEGV_ACCERR, address);
1848 		else
1849 #endif
1850 			_exception(SIGBUS, regs, BUS_ADRERR, address);
1851 	} else
1852 		bad_page_fault(regs, address, SIGBUS);
1853 
1854 	exception_exit(prev_state);
1855 }
1856 
1857 long hpte_insert_repeating(unsigned long hash, unsigned long vpn,
1858 			   unsigned long pa, unsigned long rflags,
1859 			   unsigned long vflags, int psize, int ssize)
1860 {
1861 	unsigned long hpte_group;
1862 	long slot;
1863 
1864 repeat:
1865 	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1866 
1867 	/* Insert into the hash table, primary slot */
1868 	slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, vflags,
1869 					psize, psize, ssize);
1870 
1871 	/* Primary is full, try the secondary */
1872 	if (unlikely(slot == -1)) {
1873 		hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
1874 		slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags,
1875 						vflags | HPTE_V_SECONDARY,
1876 						psize, psize, ssize);
1877 		if (slot == -1) {
1878 			if (mftb() & 0x1)
1879 				hpte_group = (hash & htab_hash_mask) *
1880 						HPTES_PER_GROUP;
1881 
1882 			mmu_hash_ops.hpte_remove(hpte_group);
1883 			goto repeat;
1884 		}
1885 	}
1886 
1887 	return slot;
1888 }
1889 
1890 #ifdef CONFIG_DEBUG_PAGEALLOC
1891 static void kernel_map_linear_page(unsigned long vaddr, unsigned long lmi)
1892 {
1893 	unsigned long hash;
1894 	unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1895 	unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize);
1896 	unsigned long mode = htab_convert_pte_flags(pgprot_val(PAGE_KERNEL));
1897 	long ret;
1898 
1899 	hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize);
1900 
1901 	/* Don't create HPTE entries for bad address */
1902 	if (!vsid)
1903 		return;
1904 
1905 	ret = hpte_insert_repeating(hash, vpn, __pa(vaddr), mode,
1906 				    HPTE_V_BOLTED,
1907 				    mmu_linear_psize, mmu_kernel_ssize);
1908 
1909 	BUG_ON (ret < 0);
1910 	spin_lock(&linear_map_hash_lock);
1911 	BUG_ON(linear_map_hash_slots[lmi] & 0x80);
1912 	linear_map_hash_slots[lmi] = ret | 0x80;
1913 	spin_unlock(&linear_map_hash_lock);
1914 }
1915 
1916 static void kernel_unmap_linear_page(unsigned long vaddr, unsigned long lmi)
1917 {
1918 	unsigned long hash, hidx, slot;
1919 	unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1920 	unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize);
1921 
1922 	hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize);
1923 	spin_lock(&linear_map_hash_lock);
1924 	BUG_ON(!(linear_map_hash_slots[lmi] & 0x80));
1925 	hidx = linear_map_hash_slots[lmi] & 0x7f;
1926 	linear_map_hash_slots[lmi] = 0;
1927 	spin_unlock(&linear_map_hash_lock);
1928 	if (hidx & _PTEIDX_SECONDARY)
1929 		hash = ~hash;
1930 	slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1931 	slot += hidx & _PTEIDX_GROUP_IX;
1932 	mmu_hash_ops.hpte_invalidate(slot, vpn, mmu_linear_psize,
1933 				     mmu_linear_psize,
1934 				     mmu_kernel_ssize, 0);
1935 }
1936 
1937 void __kernel_map_pages(struct page *page, int numpages, int enable)
1938 {
1939 	unsigned long flags, vaddr, lmi;
1940 	int i;
1941 
1942 	local_irq_save(flags);
1943 	for (i = 0; i < numpages; i++, page++) {
1944 		vaddr = (unsigned long)page_address(page);
1945 		lmi = __pa(vaddr) >> PAGE_SHIFT;
1946 		if (lmi >= linear_map_hash_count)
1947 			continue;
1948 		if (enable)
1949 			kernel_map_linear_page(vaddr, lmi);
1950 		else
1951 			kernel_unmap_linear_page(vaddr, lmi);
1952 	}
1953 	local_irq_restore(flags);
1954 }
1955 #endif /* CONFIG_DEBUG_PAGEALLOC */
1956 
1957 void hash__setup_initial_memory_limit(phys_addr_t first_memblock_base,
1958 				phys_addr_t first_memblock_size)
1959 {
1960 	/*
1961 	 * We don't currently support the first MEMBLOCK not mapping 0
1962 	 * physical on those processors
1963 	 */
1964 	BUG_ON(first_memblock_base != 0);
1965 
1966 	/*
1967 	 * On virtualized systems the first entry is our RMA region aka VRMA,
1968 	 * non-virtualized 64-bit hash MMU systems don't have a limitation
1969 	 * on real mode access.
1970 	 *
1971 	 * For guests on platforms before POWER9, we clamp the it limit to 1G
1972 	 * to avoid some funky things such as RTAS bugs etc...
1973 	 *
1974 	 * On POWER9 we limit to 1TB in case the host erroneously told us that
1975 	 * the RMA was >1TB. Effective address bits 0:23 are treated as zero
1976 	 * (meaning the access is aliased to zero i.e. addr = addr % 1TB)
1977 	 * for virtual real mode addressing and so it doesn't make sense to
1978 	 * have an area larger than 1TB as it can't be addressed.
1979 	 */
1980 	if (!early_cpu_has_feature(CPU_FTR_HVMODE)) {
1981 		ppc64_rma_size = first_memblock_size;
1982 		if (!early_cpu_has_feature(CPU_FTR_ARCH_300))
1983 			ppc64_rma_size = min_t(u64, ppc64_rma_size, 0x40000000);
1984 		else
1985 			ppc64_rma_size = min_t(u64, ppc64_rma_size,
1986 					       1UL << SID_SHIFT_1T);
1987 
1988 		/* Finally limit subsequent allocations */
1989 		memblock_set_current_limit(ppc64_rma_size);
1990 	} else {
1991 		ppc64_rma_size = ULONG_MAX;
1992 	}
1993 }
1994 
1995 #ifdef CONFIG_DEBUG_FS
1996 
1997 static int hpt_order_get(void *data, u64 *val)
1998 {
1999 	*val = ppc64_pft_size;
2000 	return 0;
2001 }
2002 
2003 static int hpt_order_set(void *data, u64 val)
2004 {
2005 	int ret;
2006 
2007 	if (!mmu_hash_ops.resize_hpt)
2008 		return -ENODEV;
2009 
2010 	cpus_read_lock();
2011 	ret = mmu_hash_ops.resize_hpt(val);
2012 	cpus_read_unlock();
2013 
2014 	return ret;
2015 }
2016 
2017 DEFINE_DEBUGFS_ATTRIBUTE(fops_hpt_order, hpt_order_get, hpt_order_set, "%llu\n");
2018 
2019 static int __init hash64_debugfs(void)
2020 {
2021 	debugfs_create_file("hpt_order", 0600, powerpc_debugfs_root, NULL,
2022 			    &fops_hpt_order);
2023 	return 0;
2024 }
2025 machine_device_initcall(pseries, hash64_debugfs);
2026 #endif /* CONFIG_DEBUG_FS */
2027 
2028 void __init print_system_hash_info(void)
2029 {
2030 	pr_info("ppc64_pft_size    = 0x%llx\n", ppc64_pft_size);
2031 
2032 	if (htab_hash_mask)
2033 		pr_info("htab_hash_mask    = 0x%lx\n", htab_hash_mask);
2034 }
2035