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
100static unsigned long _SDR1;
101struct mmu_psize_def mmu_psize_defs[MMU_PAGE_COUNT];
102EXPORT_SYMBOL_GPL(mmu_psize_defs);
103
104u8 hpte_page_sizes[1 << LP_BITS];
105EXPORT_SYMBOL_GPL(hpte_page_sizes);
106
107struct hash_pte *htab_address;
108unsigned long htab_size_bytes;
109unsigned long htab_hash_mask;
110EXPORT_SYMBOL_GPL(htab_hash_mask);
111int mmu_linear_psize = MMU_PAGE_4K;
112EXPORT_SYMBOL_GPL(mmu_linear_psize);
113int mmu_virtual_psize = MMU_PAGE_4K;
114int mmu_vmalloc_psize = MMU_PAGE_4K;
115#ifdef CONFIG_SPARSEMEM_VMEMMAP
116int mmu_vmemmap_psize = MMU_PAGE_4K;
117#endif
118int mmu_io_psize = MMU_PAGE_4K;
119int mmu_kernel_ssize = MMU_SEGSIZE_256M;
120EXPORT_SYMBOL_GPL(mmu_kernel_ssize);
121int mmu_highuser_ssize = MMU_SEGSIZE_256M;
122u16 mmu_slb_size = 64;
123EXPORT_SYMBOL_GPL(mmu_slb_size);
124#ifdef CONFIG_PPC_64K_PAGES
125int mmu_ci_restrictions;
126#endif
127#ifdef CONFIG_DEBUG_PAGEALLOC
128static u8 *linear_map_hash_slots;
129static unsigned long linear_map_hash_count;
130static DEFINE_SPINLOCK(linear_map_hash_lock);
131#endif /* CONFIG_DEBUG_PAGEALLOC */
132struct mmu_hash_ops mmu_hash_ops;
133EXPORT_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 */
143static 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 */
158static 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 */
189unsigned long htab_convert_pte_flags(unsigned long pteflags)
190{
191	unsigned long rflags = 0;
192
193	/* _PAGE_EXEC -> NOEXEC */
194	if ((pteflags & _PAGE_EXEC) == 0)
195		rflags |= HPTE_R_N;
196	/*
197	 * PPP bits:
198	 * Linux uses slb key 0 for kernel and 1 for user.
199	 * kernel RW areas are mapped with PPP=0b000
200	 * User area is mapped with PPP=0b010 for read/write
201	 * or PPP=0b011 for read-only (including writeable but clean pages).
202	 */
203	if (pteflags & _PAGE_PRIVILEGED) {
204		/*
205		 * Kernel read only mapped with ppp bits 0b110
206		 */
207		if (!(pteflags & _PAGE_WRITE)) {
208			if (mmu_has_feature(MMU_FTR_KERNEL_RO))
209				rflags |= (HPTE_R_PP0 | 0x2);
210			else
211				rflags |= 0x3;
212		}
213	} else {
214		if (pteflags & _PAGE_RWX)
215			rflags |= 0x2;
216		if (!((pteflags & _PAGE_WRITE) && (pteflags & _PAGE_DIRTY)))
217			rflags |= 0x1;
218	}
219	/*
220	 * We can't allow hardware to update hpte bits. Hence always
221	 * set 'R' bit and set 'C' if it is a write fault
222	 */
223	rflags |=  HPTE_R_R;
224
225	if (pteflags & _PAGE_DIRTY)
226		rflags |= HPTE_R_C;
227	/*
228	 * Add in WIG bits
229	 */
230
231	if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_TOLERANT)
232		rflags |= HPTE_R_I;
233	else if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_NON_IDEMPOTENT)
234		rflags |= (HPTE_R_I | HPTE_R_G);
235	else if ((pteflags & _PAGE_CACHE_CTL) == _PAGE_SAO)
236		rflags |= (HPTE_R_W | HPTE_R_I | HPTE_R_M);
237	else
238		/*
239		 * Add memory coherence if cache inhibited is not set
240		 */
241		rflags |= HPTE_R_M;
242
243	rflags |= pte_to_hpte_pkey_bits(pteflags);
244	return rflags;
245}
246
247int htab_bolt_mapping(unsigned long vstart, unsigned long vend,
248		      unsigned long pstart, unsigned long prot,
249		      int psize, int ssize)
250{
251	unsigned long vaddr, paddr;
252	unsigned int step, shift;
253	int ret = 0;
254
255	shift = mmu_psize_defs[psize].shift;
256	step = 1 << shift;
257
258	prot = htab_convert_pte_flags(prot);
259
260	DBG("htab_bolt_mapping(%lx..%lx -> %lx (%lx,%d,%d)\n",
261	    vstart, vend, pstart, prot, psize, ssize);
262
263	/* Carefully map only the possible range */
264	vaddr = ALIGN(vstart, step);
265	paddr = ALIGN(pstart, step);
266	vend  = ALIGN_DOWN(vend, step);
267
268	for (; vaddr < vend; vaddr += step, paddr += step) {
269		unsigned long hash, hpteg;
270		unsigned long vsid = get_kernel_vsid(vaddr, ssize);
271		unsigned long vpn  = hpt_vpn(vaddr, vsid, ssize);
272		unsigned long tprot = prot;
273		bool secondary_hash = false;
274
275		/*
276		 * If we hit a bad address return error.
277		 */
278		if (!vsid)
279			return -1;
280		/* Make kernel text executable */
281		if (overlaps_kernel_text(vaddr, vaddr + step))
282			tprot &= ~HPTE_R_N;
283
284		/*
285		 * If relocatable, check if it overlaps interrupt vectors that
286		 * are copied down to real 0. For relocatable kernel
287		 * (e.g. kdump case) we copy interrupt vectors down to real
288		 * address 0. Mark that region as executable. This is
289		 * because on p8 system with relocation on exception feature
290		 * enabled, exceptions are raised with MMU (IR=DR=1) ON. Hence
291		 * in order to execute the interrupt handlers in virtual
292		 * mode the vector region need to be marked as executable.
293		 */
294		if ((PHYSICAL_START > MEMORY_START) &&
295			overlaps_interrupt_vector_text(vaddr, vaddr + step))
296				tprot &= ~HPTE_R_N;
297
298		hash = hpt_hash(vpn, shift, ssize);
299		hpteg = ((hash & htab_hash_mask) * HPTES_PER_GROUP);
300
301		BUG_ON(!mmu_hash_ops.hpte_insert);
302repeat:
303		ret = mmu_hash_ops.hpte_insert(hpteg, vpn, paddr, tprot,
304					       HPTE_V_BOLTED, psize, psize,
305					       ssize);
306		if (ret == -1) {
307			/*
308			 * Try to to keep bolted entries in primary.
309			 * Remove non bolted entries and try insert again
310			 */
311			ret = mmu_hash_ops.hpte_remove(hpteg);
312			if (ret != -1)
313				ret = mmu_hash_ops.hpte_insert(hpteg, vpn, paddr, tprot,
314							       HPTE_V_BOLTED, psize, psize,
315							       ssize);
316			if (ret == -1 && !secondary_hash) {
317				secondary_hash = true;
318				hpteg = ((~hash & htab_hash_mask) * HPTES_PER_GROUP);
319				goto repeat;
320			}
321		}
322
323		if (ret < 0)
324			break;
325
326		cond_resched();
327#ifdef CONFIG_DEBUG_PAGEALLOC
328		if (debug_pagealloc_enabled() &&
329			(paddr >> PAGE_SHIFT) < linear_map_hash_count)
330			linear_map_hash_slots[paddr >> PAGE_SHIFT] = ret | 0x80;
331#endif /* CONFIG_DEBUG_PAGEALLOC */
332	}
333	return ret < 0 ? ret : 0;
334}
335
336int htab_remove_mapping(unsigned long vstart, unsigned long vend,
337		      int psize, int ssize)
338{
339	unsigned long vaddr, time_limit;
340	unsigned int step, shift;
341	int rc;
342	int ret = 0;
343
344	shift = mmu_psize_defs[psize].shift;
345	step = 1 << shift;
346
347	if (!mmu_hash_ops.hpte_removebolted)
348		return -ENODEV;
349
350	/* Unmap the full range specificied */
351	vaddr = ALIGN_DOWN(vstart, step);
352	time_limit = jiffies + HZ;
353
354	for (;vaddr < vend; vaddr += step) {
355		rc = mmu_hash_ops.hpte_removebolted(vaddr, psize, ssize);
356
357		/*
358		 * For large number of mappings introduce a cond_resched()
359		 * to prevent softlockup warnings.
360		 */
361		if (time_after(jiffies, time_limit)) {
362			cond_resched();
363			time_limit = jiffies + HZ;
364		}
365		if (rc == -ENOENT) {
366			ret = -ENOENT;
367			continue;
368		}
369		if (rc < 0)
370			return rc;
371	}
372
373	return ret;
374}
375
376static bool disable_1tb_segments = false;
377
378static int __init parse_disable_1tb_segments(char *p)
379{
380	disable_1tb_segments = true;
381	return 0;
382}
383early_param("disable_1tb_segments", parse_disable_1tb_segments);
384
385static int __init htab_dt_scan_seg_sizes(unsigned long node,
386					 const char *uname, int depth,
387					 void *data)
388{
389	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
390	const __be32 *prop;
391	int size = 0;
392
393	/* We are scanning "cpu" nodes only */
394	if (type == NULL || strcmp(type, "cpu") != 0)
395		return 0;
396
397	prop = of_get_flat_dt_prop(node, "ibm,processor-segment-sizes", &size);
398	if (prop == NULL)
399		return 0;
400	for (; size >= 4; size -= 4, ++prop) {
401		if (be32_to_cpu(prop[0]) == 40) {
402			DBG("1T segment support detected\n");
403
404			if (disable_1tb_segments) {
405				DBG("1T segments disabled by command line\n");
406				break;
407			}
408
409			cur_cpu_spec->mmu_features |= MMU_FTR_1T_SEGMENT;
410			return 1;
411		}
412	}
413	cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B;
414	return 0;
415}
416
417static int __init get_idx_from_shift(unsigned int shift)
418{
419	int idx = -1;
420
421	switch (shift) {
422	case 0xc:
423		idx = MMU_PAGE_4K;
424		break;
425	case 0x10:
426		idx = MMU_PAGE_64K;
427		break;
428	case 0x14:
429		idx = MMU_PAGE_1M;
430		break;
431	case 0x18:
432		idx = MMU_PAGE_16M;
433		break;
434	case 0x22:
435		idx = MMU_PAGE_16G;
436		break;
437	}
438	return idx;
439}
440
441static int __init htab_dt_scan_page_sizes(unsigned long node,
442					  const char *uname, int depth,
443					  void *data)
444{
445	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
446	const __be32 *prop;
447	int size = 0;
448
449	/* We are scanning "cpu" nodes only */
450	if (type == NULL || strcmp(type, "cpu") != 0)
451		return 0;
452
453	prop = of_get_flat_dt_prop(node, "ibm,segment-page-sizes", &size);
454	if (!prop)
455		return 0;
456
457	pr_info("Page sizes from device-tree:\n");
458	size /= 4;
459	cur_cpu_spec->mmu_features &= ~(MMU_FTR_16M_PAGE);
460	while(size > 0) {
461		unsigned int base_shift = be32_to_cpu(prop[0]);
462		unsigned int slbenc = be32_to_cpu(prop[1]);
463		unsigned int lpnum = be32_to_cpu(prop[2]);
464		struct mmu_psize_def *def;
465		int idx, base_idx;
466
467		size -= 3; prop += 3;
468		base_idx = get_idx_from_shift(base_shift);
469		if (base_idx < 0) {
470			/* skip the pte encoding also */
471			prop += lpnum * 2; size -= lpnum * 2;
472			continue;
473		}
474		def = &mmu_psize_defs[base_idx];
475		if (base_idx == MMU_PAGE_16M)
476			cur_cpu_spec->mmu_features |= MMU_FTR_16M_PAGE;
477
478		def->shift = base_shift;
479		if (base_shift <= 23)
480			def->avpnm = 0;
481		else
482			def->avpnm = (1 << (base_shift - 23)) - 1;
483		def->sllp = slbenc;
484		/*
485		 * We don't know for sure what's up with tlbiel, so
486		 * for now we only set it for 4K and 64K pages
487		 */
488		if (base_idx == MMU_PAGE_4K || base_idx == MMU_PAGE_64K)
489			def->tlbiel = 1;
490		else
491			def->tlbiel = 0;
492
493		while (size > 0 && lpnum) {
494			unsigned int shift = be32_to_cpu(prop[0]);
495			int penc  = be32_to_cpu(prop[1]);
496
497			prop += 2; size -= 2;
498			lpnum--;
499
500			idx = get_idx_from_shift(shift);
501			if (idx < 0)
502				continue;
503
504			if (penc == -1)
505				pr_err("Invalid penc for base_shift=%d "
506				       "shift=%d\n", base_shift, shift);
507
508			def->penc[idx] = penc;
509			pr_info("base_shift=%d: shift=%d, sllp=0x%04lx,"
510				" avpnm=0x%08lx, tlbiel=%d, penc=%d\n",
511				base_shift, shift, def->sllp,
512				def->avpnm, def->tlbiel, def->penc[idx]);
513		}
514	}
515
516	return 1;
517}
518
519#ifdef CONFIG_HUGETLB_PAGE
520/*
521 * Scan for 16G memory blocks that have been set aside for huge pages
522 * and reserve those blocks for 16G huge pages.
523 */
524static int __init htab_dt_scan_hugepage_blocks(unsigned long node,
525					const char *uname, int depth,
526					void *data) {
527	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
528	const __be64 *addr_prop;
529	const __be32 *page_count_prop;
530	unsigned int expected_pages;
531	long unsigned int phys_addr;
532	long unsigned int block_size;
533
534	/* We are scanning "memory" nodes only */
535	if (type == NULL || strcmp(type, "memory") != 0)
536		return 0;
537
538	/*
539	 * This property is the log base 2 of the number of virtual pages that
540	 * will represent this memory block.
541	 */
542	page_count_prop = of_get_flat_dt_prop(node, "ibm,expected#pages", NULL);
543	if (page_count_prop == NULL)
544		return 0;
545	expected_pages = (1 << be32_to_cpu(page_count_prop[0]));
546	addr_prop = of_get_flat_dt_prop(node, "reg", NULL);
547	if (addr_prop == NULL)
548		return 0;
549	phys_addr = be64_to_cpu(addr_prop[0]);
550	block_size = be64_to_cpu(addr_prop[1]);
551	if (block_size != (16 * GB))
552		return 0;
553	printk(KERN_INFO "Huge page(16GB) memory: "
554			"addr = 0x%lX size = 0x%lX pages = %d\n",
555			phys_addr, block_size, expected_pages);
556	if (phys_addr + block_size * expected_pages <= memblock_end_of_DRAM()) {
557		memblock_reserve(phys_addr, block_size * expected_pages);
558		pseries_add_gpage(phys_addr, block_size, expected_pages);
559	}
560	return 0;
561}
562#endif /* CONFIG_HUGETLB_PAGE */
563
564static void mmu_psize_set_default_penc(void)
565{
566	int bpsize, apsize;
567	for (bpsize = 0; bpsize < MMU_PAGE_COUNT; bpsize++)
568		for (apsize = 0; apsize < MMU_PAGE_COUNT; apsize++)
569			mmu_psize_defs[bpsize].penc[apsize] = -1;
570}
571
572#ifdef CONFIG_PPC_64K_PAGES
573
574static bool might_have_hea(void)
575{
576	/*
577	 * The HEA ethernet adapter requires awareness of the
578	 * GX bus. Without that awareness we can easily assume
579	 * we will never see an HEA ethernet device.
580	 */
581#ifdef CONFIG_IBMEBUS
582	return !cpu_has_feature(CPU_FTR_ARCH_207S) &&
583		firmware_has_feature(FW_FEATURE_SPLPAR);
584#else
585	return false;
586#endif
587}
588
589#endif /* #ifdef CONFIG_PPC_64K_PAGES */
590
591static void __init htab_scan_page_sizes(void)
592{
593	int rc;
594
595	/* se the invalid penc to -1 */
596	mmu_psize_set_default_penc();
597
598	/* Default to 4K pages only */
599	memcpy(mmu_psize_defs, mmu_psize_defaults,
600	       sizeof(mmu_psize_defaults));
601
602	/*
603	 * Try to find the available page sizes in the device-tree
604	 */
605	rc = of_scan_flat_dt(htab_dt_scan_page_sizes, NULL);
606	if (rc == 0 && early_mmu_has_feature(MMU_FTR_16M_PAGE)) {
607		/*
608		 * Nothing in the device-tree, but the CPU supports 16M pages,
609		 * so let's fallback on a known size list for 16M capable CPUs.
610		 */
611		memcpy(mmu_psize_defs, mmu_psize_defaults_gp,
612		       sizeof(mmu_psize_defaults_gp));
613	}
614
615#ifdef CONFIG_HUGETLB_PAGE
616	if (!hugetlb_disabled && !early_radix_enabled() ) {
617		/* Reserve 16G huge page memory sections for huge pages */
618		of_scan_flat_dt(htab_dt_scan_hugepage_blocks, NULL);
619	}
620#endif /* CONFIG_HUGETLB_PAGE */
621}
622
623/*
624 * Fill in the hpte_page_sizes[] array.
625 * We go through the mmu_psize_defs[] array looking for all the
626 * supported base/actual page size combinations.  Each combination
627 * has a unique pagesize encoding (penc) value in the low bits of
628 * the LP field of the HPTE.  For actual page sizes less than 1MB,
629 * some of the upper LP bits are used for RPN bits, meaning that
630 * we need to fill in several entries in hpte_page_sizes[].
631 *
632 * In diagrammatic form, with r = RPN bits and z = page size bits:
633 *        PTE LP     actual page size
634 *    rrrr rrrz		>=8KB
635 *    rrrr rrzz		>=16KB
636 *    rrrr rzzz		>=32KB
637 *    rrrr zzzz		>=64KB
638 *    ...
639 *
640 * The zzzz bits are implementation-specific but are chosen so that
641 * no encoding for a larger page size uses the same value in its
642 * low-order N bits as the encoding for the 2^(12+N) byte page size
643 * (if it exists).
644 */
645static void init_hpte_page_sizes(void)
646{
647	long int ap, bp;
648	long int shift, penc;
649
650	for (bp = 0; bp < MMU_PAGE_COUNT; ++bp) {
651		if (!mmu_psize_defs[bp].shift)
652			continue;	/* not a supported page size */
653		for (ap = bp; ap < MMU_PAGE_COUNT; ++ap) {
654			penc = mmu_psize_defs[bp].penc[ap];
655			if (penc == -1 || !mmu_psize_defs[ap].shift)
656				continue;
657			shift = mmu_psize_defs[ap].shift - LP_SHIFT;
658			if (shift <= 0)
659				continue;	/* should never happen */
660			/*
661			 * For page sizes less than 1MB, this loop
662			 * replicates the entry for all possible values
663			 * of the rrrr bits.
664			 */
665			while (penc < (1 << LP_BITS)) {
666				hpte_page_sizes[penc] = (ap << 4) | bp;
667				penc += 1 << shift;
668			}
669		}
670	}
671}
672
673static void __init htab_init_page_sizes(void)
674{
675	bool aligned = true;
676	init_hpte_page_sizes();
677
678	if (!debug_pagealloc_enabled()) {
679		/*
680		 * Pick a size for the linear mapping. Currently, we only
681		 * support 16M, 1M and 4K which is the default
682		 */
683		if (IS_ENABLED(CONFIG_STRICT_KERNEL_RWX) &&
684		    (unsigned long)_stext % 0x1000000) {
685			if (mmu_psize_defs[MMU_PAGE_16M].shift)
686				pr_warn("Kernel not 16M aligned, disabling 16M linear map alignment\n");
687			aligned = false;
688		}
689
690		if (mmu_psize_defs[MMU_PAGE_16M].shift && aligned)
691			mmu_linear_psize = MMU_PAGE_16M;
692		else if (mmu_psize_defs[MMU_PAGE_1M].shift)
693			mmu_linear_psize = MMU_PAGE_1M;
694	}
695
696#ifdef CONFIG_PPC_64K_PAGES
697	/*
698	 * Pick a size for the ordinary pages. Default is 4K, we support
699	 * 64K for user mappings and vmalloc if supported by the processor.
700	 * We only use 64k for ioremap if the processor
701	 * (and firmware) support cache-inhibited large pages.
702	 * If not, we use 4k and set mmu_ci_restrictions so that
703	 * hash_page knows to switch processes that use cache-inhibited
704	 * mappings to 4k pages.
705	 */
706	if (mmu_psize_defs[MMU_PAGE_64K].shift) {
707		mmu_virtual_psize = MMU_PAGE_64K;
708		mmu_vmalloc_psize = MMU_PAGE_64K;
709		if (mmu_linear_psize == MMU_PAGE_4K)
710			mmu_linear_psize = MMU_PAGE_64K;
711		if (mmu_has_feature(MMU_FTR_CI_LARGE_PAGE)) {
712			/*
713			 * When running on pSeries using 64k pages for ioremap
714			 * would stop us accessing the HEA ethernet. So if we
715			 * have the chance of ever seeing one, stay at 4k.
716			 */
717			if (!might_have_hea())
718				mmu_io_psize = MMU_PAGE_64K;
719		} else
720			mmu_ci_restrictions = 1;
721	}
722#endif /* CONFIG_PPC_64K_PAGES */
723
724#ifdef CONFIG_SPARSEMEM_VMEMMAP
725	/*
726	 * We try to use 16M pages for vmemmap if that is supported
727	 * and we have at least 1G of RAM at boot
728	 */
729	if (mmu_psize_defs[MMU_PAGE_16M].shift &&
730	    memblock_phys_mem_size() >= 0x40000000)
731		mmu_vmemmap_psize = MMU_PAGE_16M;
732	else
733		mmu_vmemmap_psize = mmu_virtual_psize;
734#endif /* CONFIG_SPARSEMEM_VMEMMAP */
735
736	printk(KERN_DEBUG "Page orders: linear mapping = %d, "
737	       "virtual = %d, io = %d"
738#ifdef CONFIG_SPARSEMEM_VMEMMAP
739	       ", vmemmap = %d"
740#endif
741	       "\n",
742	       mmu_psize_defs[mmu_linear_psize].shift,
743	       mmu_psize_defs[mmu_virtual_psize].shift,
744	       mmu_psize_defs[mmu_io_psize].shift
745#ifdef CONFIG_SPARSEMEM_VMEMMAP
746	       ,mmu_psize_defs[mmu_vmemmap_psize].shift
747#endif
748	       );
749}
750
751static int __init htab_dt_scan_pftsize(unsigned long node,
752				       const char *uname, int depth,
753				       void *data)
754{
755	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
756	const __be32 *prop;
757
758	/* We are scanning "cpu" nodes only */
759	if (type == NULL || strcmp(type, "cpu") != 0)
760		return 0;
761
762	prop = of_get_flat_dt_prop(node, "ibm,pft-size", NULL);
763	if (prop != NULL) {
764		/* pft_size[0] is the NUMA CEC cookie */
765		ppc64_pft_size = be32_to_cpu(prop[1]);
766		return 1;
767	}
768	return 0;
769}
770
771unsigned htab_shift_for_mem_size(unsigned long mem_size)
772{
773	unsigned memshift = __ilog2(mem_size);
774	unsigned pshift = mmu_psize_defs[mmu_virtual_psize].shift;
775	unsigned pteg_shift;
776
777	/* round mem_size up to next power of 2 */
778	if ((1UL << memshift) < mem_size)
779		memshift += 1;
780
781	/* aim for 2 pages / pteg */
782	pteg_shift = memshift - (pshift + 1);
783
784	/*
785	 * 2^11 PTEGS of 128 bytes each, ie. 2^18 bytes is the minimum htab
786	 * size permitted by the architecture.
787	 */
788	return max(pteg_shift + 7, 18U);
789}
790
791static unsigned long __init htab_get_table_size(void)
792{
793	/*
794	 * If hash size isn't already provided by the platform, we try to
795	 * retrieve it from the device-tree. If it's not there neither, we
796	 * calculate it now based on the total RAM size
797	 */
798	if (ppc64_pft_size == 0)
799		of_scan_flat_dt(htab_dt_scan_pftsize, NULL);
800	if (ppc64_pft_size)
801		return 1UL << ppc64_pft_size;
802
803	return 1UL << htab_shift_for_mem_size(memblock_phys_mem_size());
804}
805
806#ifdef CONFIG_MEMORY_HOTPLUG
807static int resize_hpt_for_hotplug(unsigned long new_mem_size)
808{
809	unsigned target_hpt_shift;
810
811	if (!mmu_hash_ops.resize_hpt)
812		return 0;
813
814	target_hpt_shift = htab_shift_for_mem_size(new_mem_size);
815
816	/*
817	 * To avoid lots of HPT resizes if memory size is fluctuating
818	 * across a boundary, we deliberately have some hysterisis
819	 * here: we immediately increase the HPT size if the target
820	 * shift exceeds the current shift, but we won't attempt to
821	 * reduce unless the target shift is at least 2 below the
822	 * current shift
823	 */
824	if (target_hpt_shift > ppc64_pft_size ||
825	    target_hpt_shift < ppc64_pft_size - 1)
826		return mmu_hash_ops.resize_hpt(target_hpt_shift);
827
828	return 0;
829}
830
831int hash__create_section_mapping(unsigned long start, unsigned long end,
832				 int nid, pgprot_t prot)
833{
834	int rc;
835
836	if (end >= H_VMALLOC_START) {
837		pr_warn("Outside the supported range\n");
838		return -1;
839	}
840
841	resize_hpt_for_hotplug(memblock_phys_mem_size());
842
843	rc = htab_bolt_mapping(start, end, __pa(start),
844			       pgprot_val(prot), mmu_linear_psize,
845			       mmu_kernel_ssize);
846
847	if (rc < 0) {
848		int rc2 = htab_remove_mapping(start, end, mmu_linear_psize,
849					      mmu_kernel_ssize);
850		BUG_ON(rc2 && (rc2 != -ENOENT));
851	}
852	return rc;
853}
854
855int hash__remove_section_mapping(unsigned long start, unsigned long end)
856{
857	int rc = htab_remove_mapping(start, end, mmu_linear_psize,
858				     mmu_kernel_ssize);
859	WARN_ON(rc < 0);
860
861	if (resize_hpt_for_hotplug(memblock_phys_mem_size()) == -ENOSPC)
862		pr_warn("Hash collision while resizing HPT\n");
863
864	return rc;
865}
866#endif /* CONFIG_MEMORY_HOTPLUG */
867
868static void __init hash_init_partition_table(phys_addr_t hash_table,
869					     unsigned long htab_size)
870{
871	mmu_partition_table_init();
872
873	/*
874	 * PS field (VRMA page size) is not used for LPID 0, hence set to 0.
875	 * For now, UPRT is 0 and we have no segment table.
876	 */
877	htab_size =  __ilog2(htab_size) - 18;
878	mmu_partition_table_set_entry(0, hash_table | htab_size, 0, false);
879	pr_info("Partition table %p\n", partition_tb);
880}
881
882static void __init htab_initialize(void)
883{
884	unsigned long table;
885	unsigned long pteg_count;
886	unsigned long prot;
887	phys_addr_t base = 0, size = 0, end;
888	u64 i;
889
890	DBG(" -> htab_initialize()\n");
891
892	if (mmu_has_feature(MMU_FTR_1T_SEGMENT)) {
893		mmu_kernel_ssize = MMU_SEGSIZE_1T;
894		mmu_highuser_ssize = MMU_SEGSIZE_1T;
895		printk(KERN_INFO "Using 1TB segments\n");
896	}
897
898	if (stress_slb_enabled)
899		static_branch_enable(&stress_slb_key);
900
901	/*
902	 * Calculate the required size of the htab.  We want the number of
903	 * PTEGs to equal one half the number of real pages.
904	 */
905	htab_size_bytes = htab_get_table_size();
906	pteg_count = htab_size_bytes >> 7;
907
908	htab_hash_mask = pteg_count - 1;
909
910	if (firmware_has_feature(FW_FEATURE_LPAR) ||
911	    firmware_has_feature(FW_FEATURE_PS3_LV1)) {
912		/* Using a hypervisor which owns the htab */
913		htab_address = NULL;
914		_SDR1 = 0;
915#ifdef CONFIG_FA_DUMP
916		/*
917		 * If firmware assisted dump is active firmware preserves
918		 * the contents of htab along with entire partition memory.
919		 * Clear the htab if firmware assisted dump is active so
920		 * that we dont end up using old mappings.
921		 */
922		if (is_fadump_active() && mmu_hash_ops.hpte_clear_all)
923			mmu_hash_ops.hpte_clear_all();
924#endif
925	} else {
926		unsigned long limit = MEMBLOCK_ALLOC_ANYWHERE;
927
928#ifdef CONFIG_PPC_CELL
929		/*
930		 * Cell may require the hash table down low when using the
931		 * Axon IOMMU in order to fit the dynamic region over it, see
932		 * comments in cell/iommu.c
933		 */
934		if (fdt_subnode_offset(initial_boot_params, 0, "axon") > 0) {
935			limit = 0x80000000;
936			pr_info("Hash table forced below 2G for Axon IOMMU\n");
937		}
938#endif /* CONFIG_PPC_CELL */
939
940		table = memblock_phys_alloc_range(htab_size_bytes,
941						  htab_size_bytes,
942						  0, limit);
943		if (!table)
944			panic("ERROR: Failed to allocate %pa bytes below %pa\n",
945			      &htab_size_bytes, &limit);
946
947		DBG("Hash table allocated at %lx, size: %lx\n", table,
948		    htab_size_bytes);
949
950		htab_address = __va(table);
951
952		/* htab absolute addr + encoded htabsize */
953		_SDR1 = table + __ilog2(htab_size_bytes) - 18;
954
955		/* Initialize the HPT with no entries */
956		memset((void *)table, 0, htab_size_bytes);
957
958		if (!cpu_has_feature(CPU_FTR_ARCH_300))
959			/* Set SDR1 */
960			mtspr(SPRN_SDR1, _SDR1);
961		else
962			hash_init_partition_table(table, htab_size_bytes);
963	}
964
965	prot = pgprot_val(PAGE_KERNEL);
966
967#ifdef CONFIG_DEBUG_PAGEALLOC
968	if (debug_pagealloc_enabled()) {
969		linear_map_hash_count = memblock_end_of_DRAM() >> PAGE_SHIFT;
970		linear_map_hash_slots = memblock_alloc_try_nid(
971				linear_map_hash_count, 1, MEMBLOCK_LOW_LIMIT,
972				ppc64_rma_size,	NUMA_NO_NODE);
973		if (!linear_map_hash_slots)
974			panic("%s: Failed to allocate %lu bytes max_addr=%pa\n",
975			      __func__, linear_map_hash_count, &ppc64_rma_size);
976	}
977#endif /* CONFIG_DEBUG_PAGEALLOC */
978
979	/* create bolted the linear mapping in the hash table */
980	for_each_mem_range(i, &base, &end) {
981		size = end - base;
982		base = (unsigned long)__va(base);
983
984		DBG("creating mapping for region: %lx..%lx (prot: %lx)\n",
985		    base, size, prot);
986
987		if ((base + size) >= H_VMALLOC_START) {
988			pr_warn("Outside the supported range\n");
989			continue;
990		}
991
992		BUG_ON(htab_bolt_mapping(base, base + size, __pa(base),
993				prot, mmu_linear_psize, mmu_kernel_ssize));
994	}
995	memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
996
997	/*
998	 * If we have a memory_limit and we've allocated TCEs then we need to
999	 * explicitly map the TCE area at the top of RAM. We also cope with the
1000	 * case that the TCEs start below memory_limit.
1001	 * tce_alloc_start/end are 16MB aligned so the mapping should work
1002	 * for either 4K or 16MB pages.
1003	 */
1004	if (tce_alloc_start) {
1005		tce_alloc_start = (unsigned long)__va(tce_alloc_start);
1006		tce_alloc_end = (unsigned long)__va(tce_alloc_end);
1007
1008		if (base + size >= tce_alloc_start)
1009			tce_alloc_start = base + size + 1;
1010
1011		BUG_ON(htab_bolt_mapping(tce_alloc_start, tce_alloc_end,
1012					 __pa(tce_alloc_start), prot,
1013					 mmu_linear_psize, mmu_kernel_ssize));
1014	}
1015
1016
1017	DBG(" <- htab_initialize()\n");
1018}
1019#undef KB
1020#undef MB
1021
1022void __init hash__early_init_devtree(void)
1023{
1024	/* Initialize segment sizes */
1025	of_scan_flat_dt(htab_dt_scan_seg_sizes, NULL);
1026
1027	/* Initialize page sizes */
1028	htab_scan_page_sizes();
1029}
1030
1031static struct hash_mm_context init_hash_mm_context;
1032void __init hash__early_init_mmu(void)
1033{
1034#ifndef CONFIG_PPC_64K_PAGES
1035	/*
1036	 * We have code in __hash_page_4K() and elsewhere, which assumes it can
1037	 * do the following:
1038	 *   new_pte |= (slot << H_PAGE_F_GIX_SHIFT) & (H_PAGE_F_SECOND | H_PAGE_F_GIX);
1039	 *
1040	 * Where the slot number is between 0-15, and values of 8-15 indicate
1041	 * the secondary bucket. For that code to work H_PAGE_F_SECOND and
1042	 * H_PAGE_F_GIX must occupy four contiguous bits in the PTE, and
1043	 * H_PAGE_F_SECOND must be placed above H_PAGE_F_GIX. Assert that here
1044	 * with a BUILD_BUG_ON().
1045	 */
1046	BUILD_BUG_ON(H_PAGE_F_SECOND != (1ul  << (H_PAGE_F_GIX_SHIFT + 3)));
1047#endif /* CONFIG_PPC_64K_PAGES */
1048
1049	htab_init_page_sizes();
1050
1051	/*
1052	 * initialize page table size
1053	 */
1054	__pte_frag_nr = H_PTE_FRAG_NR;
1055	__pte_frag_size_shift = H_PTE_FRAG_SIZE_SHIFT;
1056	__pmd_frag_nr = H_PMD_FRAG_NR;
1057	__pmd_frag_size_shift = H_PMD_FRAG_SIZE_SHIFT;
1058
1059	__pte_index_size = H_PTE_INDEX_SIZE;
1060	__pmd_index_size = H_PMD_INDEX_SIZE;
1061	__pud_index_size = H_PUD_INDEX_SIZE;
1062	__pgd_index_size = H_PGD_INDEX_SIZE;
1063	__pud_cache_index = H_PUD_CACHE_INDEX;
1064	__pte_table_size = H_PTE_TABLE_SIZE;
1065	__pmd_table_size = H_PMD_TABLE_SIZE;
1066	__pud_table_size = H_PUD_TABLE_SIZE;
1067	__pgd_table_size = H_PGD_TABLE_SIZE;
1068	/*
1069	 * 4k use hugepd format, so for hash set then to
1070	 * zero
1071	 */
1072	__pmd_val_bits = HASH_PMD_VAL_BITS;
1073	__pud_val_bits = HASH_PUD_VAL_BITS;
1074	__pgd_val_bits = HASH_PGD_VAL_BITS;
1075
1076	__kernel_virt_start = H_KERN_VIRT_START;
1077	__vmalloc_start = H_VMALLOC_START;
1078	__vmalloc_end = H_VMALLOC_END;
1079	__kernel_io_start = H_KERN_IO_START;
1080	__kernel_io_end = H_KERN_IO_END;
1081	vmemmap = (struct page *)H_VMEMMAP_START;
1082	ioremap_bot = IOREMAP_BASE;
1083
1084#ifdef CONFIG_PCI
1085	pci_io_base = ISA_IO_BASE;
1086#endif
1087
1088	/* Select appropriate backend */
1089	if (firmware_has_feature(FW_FEATURE_PS3_LV1))
1090		ps3_early_mm_init();
1091	else if (firmware_has_feature(FW_FEATURE_LPAR))
1092		hpte_init_pseries();
1093	else if (IS_ENABLED(CONFIG_PPC_NATIVE))
1094		hpte_init_native();
1095
1096	if (!mmu_hash_ops.hpte_insert)
1097		panic("hash__early_init_mmu: No MMU hash ops defined!\n");
1098
1099	/*
1100	 * Initialize the MMU Hash table and create the linear mapping
1101	 * of memory. Has to be done before SLB initialization as this is
1102	 * currently where the page size encoding is obtained.
1103	 */
1104	htab_initialize();
1105
1106	init_mm.context.hash_context = &init_hash_mm_context;
1107	mm_ctx_set_slb_addr_limit(&init_mm.context, SLB_ADDR_LIMIT_DEFAULT);
1108
1109	pr_info("Initializing hash mmu with SLB\n");
1110	/* Initialize SLB management */
1111	slb_initialize();
1112
1113	if (cpu_has_feature(CPU_FTR_ARCH_206)
1114			&& cpu_has_feature(CPU_FTR_HVMODE))
1115		tlbiel_all();
1116}
1117
1118#ifdef CONFIG_SMP
1119void hash__early_init_mmu_secondary(void)
1120{
1121	/* Initialize hash table for that CPU */
1122	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
1123
1124		if (!cpu_has_feature(CPU_FTR_ARCH_300))
1125			mtspr(SPRN_SDR1, _SDR1);
1126		else
1127			set_ptcr_when_no_uv(__pa(partition_tb) |
1128					    (PATB_SIZE_SHIFT - 12));
1129	}
1130	/* Initialize SLB */
1131	slb_initialize();
1132
1133	if (cpu_has_feature(CPU_FTR_ARCH_206)
1134			&& cpu_has_feature(CPU_FTR_HVMODE))
1135		tlbiel_all();
1136
1137#ifdef CONFIG_PPC_MEM_KEYS
1138	if (mmu_has_feature(MMU_FTR_PKEY))
1139		mtspr(SPRN_UAMOR, default_uamor);
1140#endif
1141}
1142#endif /* CONFIG_SMP */
1143
1144/*
1145 * Called by asm hashtable.S for doing lazy icache flush
1146 */
1147unsigned int hash_page_do_lazy_icache(unsigned int pp, pte_t pte, int trap)
1148{
1149	struct page *page;
1150
1151	if (!pfn_valid(pte_pfn(pte)))
1152		return pp;
1153
1154	page = pte_page(pte);
1155
1156	/* page is dirty */
1157	if (!test_bit(PG_arch_1, &page->flags) && !PageReserved(page)) {
1158		if (trap == 0x400) {
1159			flush_dcache_icache_page(page);
1160			set_bit(PG_arch_1, &page->flags);
1161		} else
1162			pp |= HPTE_R_N;
1163	}
1164	return pp;
1165}
1166
1167#ifdef CONFIG_PPC_MM_SLICES
1168static unsigned int get_paca_psize(unsigned long addr)
1169{
1170	unsigned char *psizes;
1171	unsigned long index, mask_index;
1172
1173	if (addr < SLICE_LOW_TOP) {
1174		psizes = get_paca()->mm_ctx_low_slices_psize;
1175		index = GET_LOW_SLICE_INDEX(addr);
1176	} else {
1177		psizes = get_paca()->mm_ctx_high_slices_psize;
1178		index = GET_HIGH_SLICE_INDEX(addr);
1179	}
1180	mask_index = index & 0x1;
1181	return (psizes[index >> 1] >> (mask_index * 4)) & 0xF;
1182}
1183
1184#else
1185unsigned int get_paca_psize(unsigned long addr)
1186{
1187	return get_paca()->mm_ctx_user_psize;
1188}
1189#endif
1190
1191/*
1192 * Demote a segment to using 4k pages.
1193 * For now this makes the whole process use 4k pages.
1194 */
1195#ifdef CONFIG_PPC_64K_PAGES
1196void demote_segment_4k(struct mm_struct *mm, unsigned long addr)
1197{
1198	if (get_slice_psize(mm, addr) == MMU_PAGE_4K)
1199		return;
1200	slice_set_range_psize(mm, addr, 1, MMU_PAGE_4K);
1201	copro_flush_all_slbs(mm);
1202	if ((get_paca_psize(addr) != MMU_PAGE_4K) && (current->mm == mm)) {
1203
1204		copy_mm_to_paca(mm);
1205		slb_flush_and_restore_bolted();
1206	}
1207}
1208#endif /* CONFIG_PPC_64K_PAGES */
1209
1210#ifdef CONFIG_PPC_SUBPAGE_PROT
1211/*
1212 * This looks up a 2-bit protection code for a 4k subpage of a 64k page.
1213 * Userspace sets the subpage permissions using the subpage_prot system call.
1214 *
1215 * Result is 0: full permissions, _PAGE_RW: read-only,
1216 * _PAGE_RWX: no access.
1217 */
1218static int subpage_protection(struct mm_struct *mm, unsigned long ea)
1219{
1220	struct subpage_prot_table *spt = mm_ctx_subpage_prot(&mm->context);
1221	u32 spp = 0;
1222	u32 **sbpm, *sbpp;
1223
1224	if (!spt)
1225		return 0;
1226
1227	if (ea >= spt->maxaddr)
1228		return 0;
1229	if (ea < 0x100000000UL) {
1230		/* addresses below 4GB use spt->low_prot */
1231		sbpm = spt->low_prot;
1232	} else {
1233		sbpm = spt->protptrs[ea >> SBP_L3_SHIFT];
1234		if (!sbpm)
1235			return 0;
1236	}
1237	sbpp = sbpm[(ea >> SBP_L2_SHIFT) & (SBP_L2_COUNT - 1)];
1238	if (!sbpp)
1239		return 0;
1240	spp = sbpp[(ea >> PAGE_SHIFT) & (SBP_L1_COUNT - 1)];
1241
1242	/* extract 2-bit bitfield for this 4k subpage */
1243	spp >>= 30 - 2 * ((ea >> 12) & 0xf);
1244
1245	/*
1246	 * 0 -> full premission
1247	 * 1 -> Read only
1248	 * 2 -> no access.
1249	 * We return the flag that need to be cleared.
1250	 */
1251	spp = ((spp & 2) ? _PAGE_RWX : 0) | ((spp & 1) ? _PAGE_WRITE : 0);
1252	return spp;
1253}
1254
1255#else /* CONFIG_PPC_SUBPAGE_PROT */
1256static inline int subpage_protection(struct mm_struct *mm, unsigned long ea)
1257{
1258	return 0;
1259}
1260#endif
1261
1262void hash_failure_debug(unsigned long ea, unsigned long access,
1263			unsigned long vsid, unsigned long trap,
1264			int ssize, int psize, int lpsize, unsigned long pte)
1265{
1266	if (!printk_ratelimit())
1267		return;
1268	pr_info("mm: Hashing failure ! EA=0x%lx access=0x%lx current=%s\n",
1269		ea, access, current->comm);
1270	pr_info("    trap=0x%lx vsid=0x%lx ssize=%d base psize=%d psize %d pte=0x%lx\n",
1271		trap, vsid, ssize, psize, lpsize, pte);
1272}
1273
1274static void check_paca_psize(unsigned long ea, struct mm_struct *mm,
1275			     int psize, bool user_region)
1276{
1277	if (user_region) {
1278		if (psize != get_paca_psize(ea)) {
1279			copy_mm_to_paca(mm);
1280			slb_flush_and_restore_bolted();
1281		}
1282	} else if (get_paca()->vmalloc_sllp !=
1283		   mmu_psize_defs[mmu_vmalloc_psize].sllp) {
1284		get_paca()->vmalloc_sllp =
1285			mmu_psize_defs[mmu_vmalloc_psize].sllp;
1286		slb_vmalloc_update();
1287	}
1288}
1289
1290/*
1291 * Result code is:
1292 *  0 - handled
1293 *  1 - normal page fault
1294 * -1 - critical hash insertion error
1295 * -2 - access not permitted by subpage protection mechanism
1296 */
1297int hash_page_mm(struct mm_struct *mm, unsigned long ea,
1298		 unsigned long access, unsigned long trap,
1299		 unsigned long flags)
1300{
1301	bool is_thp;
1302	enum ctx_state prev_state = exception_enter();
1303	pgd_t *pgdir;
1304	unsigned long vsid;
1305	pte_t *ptep;
1306	unsigned hugeshift;
1307	int rc, user_region = 0;
1308	int psize, ssize;
1309
1310	DBG_LOW("hash_page(ea=%016lx, access=%lx, trap=%lx\n",
1311		ea, access, trap);
1312	trace_hash_fault(ea, access, trap);
1313
1314	/* Get region & vsid */
1315	switch (get_region_id(ea)) {
1316	case USER_REGION_ID:
1317		user_region = 1;
1318		if (! mm) {
1319			DBG_LOW(" user region with no mm !\n");
1320			rc = 1;
1321			goto bail;
1322		}
1323		psize = get_slice_psize(mm, ea);
1324		ssize = user_segment_size(ea);
1325		vsid = get_user_vsid(&mm->context, ea, ssize);
1326		break;
1327	case VMALLOC_REGION_ID:
1328		vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
1329		psize = mmu_vmalloc_psize;
1330		ssize = mmu_kernel_ssize;
1331		break;
1332
1333	case IO_REGION_ID:
1334		vsid = get_kernel_vsid(ea, mmu_kernel_ssize);
1335		psize = mmu_io_psize;
1336		ssize = mmu_kernel_ssize;
1337		break;
1338	default:
1339		/*
1340		 * Not a valid range
1341		 * Send the problem up to do_page_fault()
1342		 */
1343		rc = 1;
1344		goto bail;
1345	}
1346	DBG_LOW(" mm=%p, mm->pgdir=%p, vsid=%016lx\n", mm, mm->pgd, vsid);
1347
1348	/* Bad address. */
1349	if (!vsid) {
1350		DBG_LOW("Bad address!\n");
1351		rc = 1;
1352		goto bail;
1353	}
1354	/* Get pgdir */
1355	pgdir = mm->pgd;
1356	if (pgdir == NULL) {
1357		rc = 1;
1358		goto bail;
1359	}
1360
1361	/* Check CPU locality */
1362	if (user_region && mm_is_thread_local(mm))
1363		flags |= HPTE_LOCAL_UPDATE;
1364
1365#ifndef CONFIG_PPC_64K_PAGES
1366	/*
1367	 * If we use 4K pages and our psize is not 4K, then we might
1368	 * be hitting a special driver mapping, and need to align the
1369	 * address before we fetch the PTE.
1370	 *
1371	 * It could also be a hugepage mapping, in which case this is
1372	 * not necessary, but it's not harmful, either.
1373	 */
1374	if (psize != MMU_PAGE_4K)
1375		ea &= ~((1ul << mmu_psize_defs[psize].shift) - 1);
1376#endif /* CONFIG_PPC_64K_PAGES */
1377
1378	/* Get PTE and page size from page tables */
1379	ptep = find_linux_pte(pgdir, ea, &is_thp, &hugeshift);
1380	if (ptep == NULL || !pte_present(*ptep)) {
1381		DBG_LOW(" no PTE !\n");
1382		rc = 1;
1383		goto bail;
1384	}
1385
1386	/*
1387	 * Add _PAGE_PRESENT to the required access perm. If there are parallel
1388	 * updates to the pte that can possibly clear _PAGE_PTE, catch that too.
1389	 *
1390	 * We can safely use the return pte address in rest of the function
1391	 * because we do set H_PAGE_BUSY which prevents further updates to pte
1392	 * from generic code.
1393	 */
1394	access |= _PAGE_PRESENT | _PAGE_PTE;
1395
1396	/*
1397	 * Pre-check access permissions (will be re-checked atomically
1398	 * in __hash_page_XX but this pre-check is a fast path
1399	 */
1400	if (!check_pte_access(access, pte_val(*ptep))) {
1401		DBG_LOW(" no access !\n");
1402		rc = 1;
1403		goto bail;
1404	}
1405
1406	if (hugeshift) {
1407		if (is_thp)
1408			rc = __hash_page_thp(ea, access, vsid, (pmd_t *)ptep,
1409					     trap, flags, ssize, psize);
1410#ifdef CONFIG_HUGETLB_PAGE
1411		else
1412			rc = __hash_page_huge(ea, access, vsid, ptep, trap,
1413					      flags, ssize, hugeshift, psize);
1414#else
1415		else {
1416			/*
1417			 * if we have hugeshift, and is not transhuge with
1418			 * hugetlb disabled, something is really wrong.
1419			 */
1420			rc = 1;
1421			WARN_ON(1);
1422		}
1423#endif
1424		if (current->mm == mm)
1425			check_paca_psize(ea, mm, psize, user_region);
1426
1427		goto bail;
1428	}
1429
1430#ifndef CONFIG_PPC_64K_PAGES
1431	DBG_LOW(" i-pte: %016lx\n", pte_val(*ptep));
1432#else
1433	DBG_LOW(" i-pte: %016lx %016lx\n", pte_val(*ptep),
1434		pte_val(*(ptep + PTRS_PER_PTE)));
1435#endif
1436	/* Do actual hashing */
1437#ifdef CONFIG_PPC_64K_PAGES
1438	/* If H_PAGE_4K_PFN is set, make sure this is a 4k segment */
1439	if ((pte_val(*ptep) & H_PAGE_4K_PFN) && psize == MMU_PAGE_64K) {
1440		demote_segment_4k(mm, ea);
1441		psize = MMU_PAGE_4K;
1442	}
1443
1444	/*
1445	 * If this PTE is non-cacheable and we have restrictions on
1446	 * using non cacheable large pages, then we switch to 4k
1447	 */
1448	if (mmu_ci_restrictions && psize == MMU_PAGE_64K && pte_ci(*ptep)) {
1449		if (user_region) {
1450			demote_segment_4k(mm, ea);
1451			psize = MMU_PAGE_4K;
1452		} else if (ea < VMALLOC_END) {
1453			/*
1454			 * some driver did a non-cacheable mapping
1455			 * in vmalloc space, so switch vmalloc
1456			 * to 4k pages
1457			 */
1458			printk(KERN_ALERT "Reducing vmalloc segment "
1459			       "to 4kB pages because of "
1460			       "non-cacheable mapping\n");
1461			psize = mmu_vmalloc_psize = MMU_PAGE_4K;
1462			copro_flush_all_slbs(mm);
1463		}
1464	}
1465
1466#endif /* CONFIG_PPC_64K_PAGES */
1467
1468	if (current->mm == mm)
1469		check_paca_psize(ea, mm, psize, user_region);
1470
1471#ifdef CONFIG_PPC_64K_PAGES
1472	if (psize == MMU_PAGE_64K)
1473		rc = __hash_page_64K(ea, access, vsid, ptep, trap,
1474				     flags, ssize);
1475	else
1476#endif /* CONFIG_PPC_64K_PAGES */
1477	{
1478		int spp = subpage_protection(mm, ea);
1479		if (access & spp)
1480			rc = -2;
1481		else
1482			rc = __hash_page_4K(ea, access, vsid, ptep, trap,
1483					    flags, ssize, spp);
1484	}
1485
1486	/*
1487	 * Dump some info in case of hash insertion failure, they should
1488	 * never happen so it is really useful to know if/when they do
1489	 */
1490	if (rc == -1)
1491		hash_failure_debug(ea, access, vsid, trap, ssize, psize,
1492				   psize, pte_val(*ptep));
1493#ifndef CONFIG_PPC_64K_PAGES
1494	DBG_LOW(" o-pte: %016lx\n", pte_val(*ptep));
1495#else
1496	DBG_LOW(" o-pte: %016lx %016lx\n", pte_val(*ptep),
1497		pte_val(*(ptep + PTRS_PER_PTE)));
1498#endif
1499	DBG_LOW(" -> rc=%d\n", rc);
1500
1501bail:
1502	exception_exit(prev_state);
1503	return rc;
1504}
1505EXPORT_SYMBOL_GPL(hash_page_mm);
1506
1507int hash_page(unsigned long ea, unsigned long access, unsigned long trap,
1508	      unsigned long dsisr)
1509{
1510	unsigned long flags = 0;
1511	struct mm_struct *mm = current->mm;
1512
1513	if ((get_region_id(ea) == VMALLOC_REGION_ID) ||
1514	    (get_region_id(ea) == IO_REGION_ID))
1515		mm = &init_mm;
1516
1517	if (dsisr & DSISR_NOHPTE)
1518		flags |= HPTE_NOHPTE_UPDATE;
1519
1520	return hash_page_mm(mm, ea, access, trap, flags);
1521}
1522EXPORT_SYMBOL_GPL(hash_page);
1523
1524int __hash_page(unsigned long trap, unsigned long ea, unsigned long dsisr,
1525		unsigned long msr)
1526{
1527	unsigned long access = _PAGE_PRESENT | _PAGE_READ;
1528	unsigned long flags = 0;
1529	struct mm_struct *mm = current->mm;
1530	unsigned int region_id = get_region_id(ea);
1531
1532	if ((region_id == VMALLOC_REGION_ID) || (region_id == IO_REGION_ID))
1533		mm = &init_mm;
1534
1535	if (dsisr & DSISR_NOHPTE)
1536		flags |= HPTE_NOHPTE_UPDATE;
1537
1538	if (dsisr & DSISR_ISSTORE)
1539		access |= _PAGE_WRITE;
1540	/*
1541	 * We set _PAGE_PRIVILEGED only when
1542	 * kernel mode access kernel space.
1543	 *
1544	 * _PAGE_PRIVILEGED is NOT set
1545	 * 1) when kernel mode access user space
1546	 * 2) user space access kernel space.
1547	 */
1548	access |= _PAGE_PRIVILEGED;
1549	if ((msr & MSR_PR) || (region_id == USER_REGION_ID))
1550		access &= ~_PAGE_PRIVILEGED;
1551
1552	if (trap == 0x400)
1553		access |= _PAGE_EXEC;
1554
1555	return hash_page_mm(mm, ea, access, trap, flags);
1556}
1557
1558#ifdef CONFIG_PPC_MM_SLICES
1559static bool should_hash_preload(struct mm_struct *mm, unsigned long ea)
1560{
1561	int psize = get_slice_psize(mm, ea);
1562
1563	/* We only prefault standard pages for now */
1564	if (unlikely(psize != mm_ctx_user_psize(&mm->context)))
1565		return false;
1566
1567	/*
1568	 * Don't prefault if subpage protection is enabled for the EA.
1569	 */
1570	if (unlikely((psize == MMU_PAGE_4K) && subpage_protection(mm, ea)))
1571		return false;
1572
1573	return true;
1574}
1575#else
1576static bool should_hash_preload(struct mm_struct *mm, unsigned long ea)
1577{
1578	return true;
1579}
1580#endif
1581
1582static void hash_preload(struct mm_struct *mm, pte_t *ptep, unsigned long ea,
1583			 bool is_exec, unsigned long trap)
1584{
1585	unsigned long vsid;
1586	pgd_t *pgdir;
1587	int rc, ssize, update_flags = 0;
1588	unsigned long access = _PAGE_PRESENT | _PAGE_READ | (is_exec ? _PAGE_EXEC : 0);
1589	unsigned long flags;
1590
1591	BUG_ON(get_region_id(ea) != USER_REGION_ID);
1592
1593	if (!should_hash_preload(mm, ea))
1594		return;
1595
1596	DBG_LOW("hash_preload(mm=%p, mm->pgdir=%p, ea=%016lx, access=%lx,"
1597		" trap=%lx\n", mm, mm->pgd, ea, access, trap);
1598
1599	/* Get Linux PTE if available */
1600	pgdir = mm->pgd;
1601	if (pgdir == NULL)
1602		return;
1603
1604	/* Get VSID */
1605	ssize = user_segment_size(ea);
1606	vsid = get_user_vsid(&mm->context, ea, ssize);
1607	if (!vsid)
1608		return;
1609
1610#ifdef CONFIG_PPC_64K_PAGES
1611	/* If either H_PAGE_4K_PFN or cache inhibited is set (and we are on
1612	 * a 64K kernel), then we don't preload, hash_page() will take
1613	 * care of it once we actually try to access the page.
1614	 * That way we don't have to duplicate all of the logic for segment
1615	 * page size demotion here
1616	 * Called with  PTL held, hence can be sure the value won't change in
1617	 * between.
1618	 */
1619	if ((pte_val(*ptep) & H_PAGE_4K_PFN) || pte_ci(*ptep))
1620		return;
1621#endif /* CONFIG_PPC_64K_PAGES */
1622
1623	/*
1624	 * __hash_page_* must run with interrupts off, as it sets the
1625	 * H_PAGE_BUSY bit. It's possible for perf interrupts to hit at any
1626	 * time and may take a hash fault reading the user stack, see
1627	 * read_user_stack_slow() in the powerpc/perf code.
1628	 *
1629	 * If that takes a hash fault on the same page as we lock here, it
1630	 * will bail out when seeing H_PAGE_BUSY set, and retry the access
1631	 * leading to an infinite loop.
1632	 *
1633	 * Disabling interrupts here does not prevent perf interrupts, but it
1634	 * will prevent them taking hash faults (see the NMI test in
1635	 * do_hash_page), then read_user_stack's copy_from_user_nofault will
1636	 * fail and perf will fall back to read_user_stack_slow(), which
1637	 * walks the Linux page tables.
1638	 *
1639	 * Interrupts must also be off for the duration of the
1640	 * mm_is_thread_local test and update, to prevent preempt running the
1641	 * mm on another CPU (XXX: this may be racy vs kthread_use_mm).
1642	 */
1643	local_irq_save(flags);
1644
1645	/* Is that local to this CPU ? */
1646	if (mm_is_thread_local(mm))
1647		update_flags |= HPTE_LOCAL_UPDATE;
1648
1649	/* Hash it in */
1650#ifdef CONFIG_PPC_64K_PAGES
1651	if (mm_ctx_user_psize(&mm->context) == MMU_PAGE_64K)
1652		rc = __hash_page_64K(ea, access, vsid, ptep, trap,
1653				     update_flags, ssize);
1654	else
1655#endif /* CONFIG_PPC_64K_PAGES */
1656		rc = __hash_page_4K(ea, access, vsid, ptep, trap, update_flags,
1657				    ssize, subpage_protection(mm, ea));
1658
1659	/* Dump some info in case of hash insertion failure, they should
1660	 * never happen so it is really useful to know if/when they do
1661	 */
1662	if (rc == -1)
1663		hash_failure_debug(ea, access, vsid, trap, ssize,
1664				   mm_ctx_user_psize(&mm->context),
1665				   mm_ctx_user_psize(&mm->context),
1666				   pte_val(*ptep));
1667
1668	local_irq_restore(flags);
1669}
1670
1671/*
1672 * This is called at the end of handling a user page fault, when the
1673 * fault has been handled by updating a PTE in the linux page tables.
1674 * We use it to preload an HPTE into the hash table corresponding to
1675 * the updated linux PTE.
1676 *
1677 * This must always be called with the pte lock held.
1678 */
1679void update_mmu_cache(struct vm_area_struct *vma, unsigned long address,
1680		      pte_t *ptep)
1681{
1682	/*
1683	 * We don't need to worry about _PAGE_PRESENT here because we are
1684	 * called with either mm->page_table_lock held or ptl lock held
1685	 */
1686	unsigned long trap;
1687	bool is_exec;
1688
1689	if (radix_enabled())
1690		return;
1691
1692	/* We only want HPTEs for linux PTEs that have _PAGE_ACCESSED set */
1693	if (!pte_young(*ptep) || address >= TASK_SIZE)
1694		return;
1695
1696	/*
1697	 * We try to figure out if we are coming from an instruction
1698	 * access fault and pass that down to __hash_page so we avoid
1699	 * double-faulting on execution of fresh text. We have to test
1700	 * for regs NULL since init will get here first thing at boot.
1701	 *
1702	 * We also avoid filling the hash if not coming from a fault.
1703	 */
1704
1705	trap = current->thread.regs ? TRAP(current->thread.regs) : 0UL;
1706	switch (trap) {
1707	case 0x300:
1708		is_exec = false;
1709		break;
1710	case 0x400:
1711		is_exec = true;
1712		break;
1713	default:
1714		return;
1715	}
1716
1717	hash_preload(vma->vm_mm, ptep, address, is_exec, trap);
1718}
1719
1720#ifdef CONFIG_PPC_TRANSACTIONAL_MEM
1721static inline void tm_flush_hash_page(int local)
1722{
1723	/*
1724	 * Transactions are not aborted by tlbiel, only tlbie. Without, syncing a
1725	 * page back to a block device w/PIO could pick up transactional data
1726	 * (bad!) so we force an abort here. Before the sync the page will be
1727	 * made read-only, which will flush_hash_page. BIG ISSUE here: if the
1728	 * kernel uses a page from userspace without unmapping it first, it may
1729	 * see the speculated version.
1730	 */
1731	if (local && cpu_has_feature(CPU_FTR_TM) && current->thread.regs &&
1732	    MSR_TM_ACTIVE(current->thread.regs->msr)) {
1733		tm_enable();
1734		tm_abort(TM_CAUSE_TLBI);
1735	}
1736}
1737#else
1738static inline void tm_flush_hash_page(int local)
1739{
1740}
1741#endif
1742
1743/*
1744 * Return the global hash slot, corresponding to the given PTE, which contains
1745 * the HPTE.
1746 */
1747unsigned long pte_get_hash_gslot(unsigned long vpn, unsigned long shift,
1748		int ssize, real_pte_t rpte, unsigned int subpg_index)
1749{
1750	unsigned long hash, gslot, hidx;
1751
1752	hash = hpt_hash(vpn, shift, ssize);
1753	hidx = __rpte_to_hidx(rpte, subpg_index);
1754	if (hidx & _PTEIDX_SECONDARY)
1755		hash = ~hash;
1756	gslot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1757	gslot += hidx & _PTEIDX_GROUP_IX;
1758	return gslot;
1759}
1760
1761void flush_hash_page(unsigned long vpn, real_pte_t pte, int psize, int ssize,
1762		     unsigned long flags)
1763{
1764	unsigned long index, shift, gslot;
1765	int local = flags & HPTE_LOCAL_UPDATE;
1766
1767	DBG_LOW("flush_hash_page(vpn=%016lx)\n", vpn);
1768	pte_iterate_hashed_subpages(pte, psize, vpn, index, shift) {
1769		gslot = pte_get_hash_gslot(vpn, shift, ssize, pte, index);
1770		DBG_LOW(" sub %ld: gslot=%lx\n", index, gslot);
1771		/*
1772		 * We use same base page size and actual psize, because we don't
1773		 * use these functions for hugepage
1774		 */
1775		mmu_hash_ops.hpte_invalidate(gslot, vpn, psize, psize,
1776					     ssize, local);
1777	} pte_iterate_hashed_end();
1778
1779	tm_flush_hash_page(local);
1780}
1781
1782#ifdef CONFIG_TRANSPARENT_HUGEPAGE
1783void flush_hash_hugepage(unsigned long vsid, unsigned long addr,
1784			 pmd_t *pmdp, unsigned int psize, int ssize,
1785			 unsigned long flags)
1786{
1787	int i, max_hpte_count, valid;
1788	unsigned long s_addr;
1789	unsigned char *hpte_slot_array;
1790	unsigned long hidx, shift, vpn, hash, slot;
1791	int local = flags & HPTE_LOCAL_UPDATE;
1792
1793	s_addr = addr & HPAGE_PMD_MASK;
1794	hpte_slot_array = get_hpte_slot_array(pmdp);
1795	/*
1796	 * IF we try to do a HUGE PTE update after a withdraw is done.
1797	 * we will find the below NULL. This happens when we do
1798	 * split_huge_pmd
1799	 */
1800	if (!hpte_slot_array)
1801		return;
1802
1803	if (mmu_hash_ops.hugepage_invalidate) {
1804		mmu_hash_ops.hugepage_invalidate(vsid, s_addr, hpte_slot_array,
1805						 psize, ssize, local);
1806		goto tm_abort;
1807	}
1808	/*
1809	 * No bluk hpte removal support, invalidate each entry
1810	 */
1811	shift = mmu_psize_defs[psize].shift;
1812	max_hpte_count = HPAGE_PMD_SIZE >> shift;
1813	for (i = 0; i < max_hpte_count; i++) {
1814		/*
1815		 * 8 bits per each hpte entries
1816		 * 000| [ secondary group (one bit) | hidx (3 bits) | valid bit]
1817		 */
1818		valid = hpte_valid(hpte_slot_array, i);
1819		if (!valid)
1820			continue;
1821		hidx =  hpte_hash_index(hpte_slot_array, i);
1822
1823		/* get the vpn */
1824		addr = s_addr + (i * (1ul << shift));
1825		vpn = hpt_vpn(addr, vsid, ssize);
1826		hash = hpt_hash(vpn, shift, ssize);
1827		if (hidx & _PTEIDX_SECONDARY)
1828			hash = ~hash;
1829
1830		slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1831		slot += hidx & _PTEIDX_GROUP_IX;
1832		mmu_hash_ops.hpte_invalidate(slot, vpn, psize,
1833					     MMU_PAGE_16M, ssize, local);
1834	}
1835tm_abort:
1836	tm_flush_hash_page(local);
1837}
1838#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1839
1840void flush_hash_range(unsigned long number, int local)
1841{
1842	if (mmu_hash_ops.flush_hash_range)
1843		mmu_hash_ops.flush_hash_range(number, local);
1844	else {
1845		int i;
1846		struct ppc64_tlb_batch *batch =
1847			this_cpu_ptr(&ppc64_tlb_batch);
1848
1849		for (i = 0; i < number; i++)
1850			flush_hash_page(batch->vpn[i], batch->pte[i],
1851					batch->psize, batch->ssize, local);
1852	}
1853}
1854
1855/*
1856 * low_hash_fault is called when we the low level hash code failed
1857 * to instert a PTE due to an hypervisor error
1858 */
1859void low_hash_fault(struct pt_regs *regs, unsigned long address, int rc)
1860{
1861	enum ctx_state prev_state = exception_enter();
1862
1863	if (user_mode(regs)) {
1864#ifdef CONFIG_PPC_SUBPAGE_PROT
1865		if (rc == -2)
1866			_exception(SIGSEGV, regs, SEGV_ACCERR, address);
1867		else
1868#endif
1869			_exception(SIGBUS, regs, BUS_ADRERR, address);
1870	} else
1871		bad_page_fault(regs, address, SIGBUS);
1872
1873	exception_exit(prev_state);
1874}
1875
1876long hpte_insert_repeating(unsigned long hash, unsigned long vpn,
1877			   unsigned long pa, unsigned long rflags,
1878			   unsigned long vflags, int psize, int ssize)
1879{
1880	unsigned long hpte_group;
1881	long slot;
1882
1883repeat:
1884	hpte_group = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1885
1886	/* Insert into the hash table, primary slot */
1887	slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags, vflags,
1888					psize, psize, ssize);
1889
1890	/* Primary is full, try the secondary */
1891	if (unlikely(slot == -1)) {
1892		hpte_group = (~hash & htab_hash_mask) * HPTES_PER_GROUP;
1893		slot = mmu_hash_ops.hpte_insert(hpte_group, vpn, pa, rflags,
1894						vflags | HPTE_V_SECONDARY,
1895						psize, psize, ssize);
1896		if (slot == -1) {
1897			if (mftb() & 0x1)
1898				hpte_group = (hash & htab_hash_mask) *
1899						HPTES_PER_GROUP;
1900
1901			mmu_hash_ops.hpte_remove(hpte_group);
1902			goto repeat;
1903		}
1904	}
1905
1906	return slot;
1907}
1908
1909#ifdef CONFIG_DEBUG_PAGEALLOC
1910static void kernel_map_linear_page(unsigned long vaddr, unsigned long lmi)
1911{
1912	unsigned long hash;
1913	unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1914	unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize);
1915	unsigned long mode = htab_convert_pte_flags(pgprot_val(PAGE_KERNEL));
1916	long ret;
1917
1918	hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize);
1919
1920	/* Don't create HPTE entries for bad address */
1921	if (!vsid)
1922		return;
1923
1924	ret = hpte_insert_repeating(hash, vpn, __pa(vaddr), mode,
1925				    HPTE_V_BOLTED,
1926				    mmu_linear_psize, mmu_kernel_ssize);
1927
1928	BUG_ON (ret < 0);
1929	spin_lock(&linear_map_hash_lock);
1930	BUG_ON(linear_map_hash_slots[lmi] & 0x80);
1931	linear_map_hash_slots[lmi] = ret | 0x80;
1932	spin_unlock(&linear_map_hash_lock);
1933}
1934
1935static void kernel_unmap_linear_page(unsigned long vaddr, unsigned long lmi)
1936{
1937	unsigned long hash, hidx, slot;
1938	unsigned long vsid = get_kernel_vsid(vaddr, mmu_kernel_ssize);
1939	unsigned long vpn = hpt_vpn(vaddr, vsid, mmu_kernel_ssize);
1940
1941	hash = hpt_hash(vpn, PAGE_SHIFT, mmu_kernel_ssize);
1942	spin_lock(&linear_map_hash_lock);
1943	BUG_ON(!(linear_map_hash_slots[lmi] & 0x80));
1944	hidx = linear_map_hash_slots[lmi] & 0x7f;
1945	linear_map_hash_slots[lmi] = 0;
1946	spin_unlock(&linear_map_hash_lock);
1947	if (hidx & _PTEIDX_SECONDARY)
1948		hash = ~hash;
1949	slot = (hash & htab_hash_mask) * HPTES_PER_GROUP;
1950	slot += hidx & _PTEIDX_GROUP_IX;
1951	mmu_hash_ops.hpte_invalidate(slot, vpn, mmu_linear_psize,
1952				     mmu_linear_psize,
1953				     mmu_kernel_ssize, 0);
1954}
1955
1956void __kernel_map_pages(struct page *page, int numpages, int enable)
1957{
1958	unsigned long flags, vaddr, lmi;
1959	int i;
1960
1961	local_irq_save(flags);
1962	for (i = 0; i < numpages; i++, page++) {
1963		vaddr = (unsigned long)page_address(page);
1964		lmi = __pa(vaddr) >> PAGE_SHIFT;
1965		if (lmi >= linear_map_hash_count)
1966			continue;
1967		if (enable)
1968			kernel_map_linear_page(vaddr, lmi);
1969		else
1970			kernel_unmap_linear_page(vaddr, lmi);
1971	}
1972	local_irq_restore(flags);
1973}
1974#endif /* CONFIG_DEBUG_PAGEALLOC */
1975
1976void hash__setup_initial_memory_limit(phys_addr_t first_memblock_base,
1977				phys_addr_t first_memblock_size)
1978{
1979	/*
1980	 * We don't currently support the first MEMBLOCK not mapping 0
1981	 * physical on those processors
1982	 */
1983	BUG_ON(first_memblock_base != 0);
1984
1985	/*
1986	 * On virtualized systems the first entry is our RMA region aka VRMA,
1987	 * non-virtualized 64-bit hash MMU systems don't have a limitation
1988	 * on real mode access.
1989	 *
1990	 * For guests on platforms before POWER9, we clamp the it limit to 1G
1991	 * to avoid some funky things such as RTAS bugs etc...
1992	 *
1993	 * On POWER9 we limit to 1TB in case the host erroneously told us that
1994	 * the RMA was >1TB. Effective address bits 0:23 are treated as zero
1995	 * (meaning the access is aliased to zero i.e. addr = addr % 1TB)
1996	 * for virtual real mode addressing and so it doesn't make sense to
1997	 * have an area larger than 1TB as it can't be addressed.
1998	 */
1999	if (!early_cpu_has_feature(CPU_FTR_HVMODE)) {
2000		ppc64_rma_size = first_memblock_size;
2001		if (!early_cpu_has_feature(CPU_FTR_ARCH_300))
2002			ppc64_rma_size = min_t(u64, ppc64_rma_size, 0x40000000);
2003		else
2004			ppc64_rma_size = min_t(u64, ppc64_rma_size,
2005					       1UL << SID_SHIFT_1T);
2006
2007		/* Finally limit subsequent allocations */
2008		memblock_set_current_limit(ppc64_rma_size);
2009	} else {
2010		ppc64_rma_size = ULONG_MAX;
2011	}
2012}
2013
2014#ifdef CONFIG_DEBUG_FS
2015
2016static int hpt_order_get(void *data, u64 *val)
2017{
2018	*val = ppc64_pft_size;
2019	return 0;
2020}
2021
2022static int hpt_order_set(void *data, u64 val)
2023{
2024	int ret;
2025
2026	if (!mmu_hash_ops.resize_hpt)
2027		return -ENODEV;
2028
2029	cpus_read_lock();
2030	ret = mmu_hash_ops.resize_hpt(val);
2031	cpus_read_unlock();
2032
2033	return ret;
2034}
2035
2036DEFINE_DEBUGFS_ATTRIBUTE(fops_hpt_order, hpt_order_get, hpt_order_set, "%llu\n");
2037
2038static int __init hash64_debugfs(void)
2039{
2040	debugfs_create_file("hpt_order", 0600, powerpc_debugfs_root, NULL,
2041			    &fops_hpt_order);
2042	return 0;
2043}
2044machine_device_initcall(pseries, hash64_debugfs);
2045#endif /* CONFIG_DEBUG_FS */
2046
2047void __init print_system_hash_info(void)
2048{
2049	pr_info("ppc64_pft_size    = 0x%llx\n", ppc64_pft_size);
2050
2051	if (htab_hash_mask)
2052		pr_info("htab_hash_mask    = 0x%lx\n", htab_hash_mask);
2053}
2054