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