1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Page table handling routines for radix page table.
4  *
5  * Copyright 2015-2016, Aneesh Kumar K.V, IBM Corporation.
6  */
7 
8 #define pr_fmt(fmt) "radix-mmu: " fmt
9 
10 #include <linux/io.h>
11 #include <linux/kernel.h>
12 #include <linux/sched/mm.h>
13 #include <linux/memblock.h>
14 #include <linux/of_fdt.h>
15 #include <linux/mm.h>
16 #include <linux/hugetlb.h>
17 #include <linux/string_helpers.h>
18 #include <linux/stop_machine.h>
19 
20 #include <asm/pgtable.h>
21 #include <asm/pgalloc.h>
22 #include <asm/mmu_context.h>
23 #include <asm/dma.h>
24 #include <asm/machdep.h>
25 #include <asm/mmu.h>
26 #include <asm/firmware.h>
27 #include <asm/powernv.h>
28 #include <asm/sections.h>
29 #include <asm/trace.h>
30 #include <asm/uaccess.h>
31 #include <asm/ultravisor.h>
32 
33 #include <trace/events/thp.h>
34 
35 unsigned int mmu_pid_bits;
36 unsigned int mmu_base_pid;
37 
38 static __ref void *early_alloc_pgtable(unsigned long size, int nid,
39 			unsigned long region_start, unsigned long region_end)
40 {
41 	phys_addr_t min_addr = MEMBLOCK_LOW_LIMIT;
42 	phys_addr_t max_addr = MEMBLOCK_ALLOC_ANYWHERE;
43 	void *ptr;
44 
45 	if (region_start)
46 		min_addr = region_start;
47 	if (region_end)
48 		max_addr = region_end;
49 
50 	ptr = memblock_alloc_try_nid(size, size, min_addr, max_addr, nid);
51 
52 	if (!ptr)
53 		panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d from=%pa max_addr=%pa\n",
54 		      __func__, size, size, nid, &min_addr, &max_addr);
55 
56 	return ptr;
57 }
58 
59 static int early_map_kernel_page(unsigned long ea, unsigned long pa,
60 			  pgprot_t flags,
61 			  unsigned int map_page_size,
62 			  int nid,
63 			  unsigned long region_start, unsigned long region_end)
64 {
65 	unsigned long pfn = pa >> PAGE_SHIFT;
66 	pgd_t *pgdp;
67 	pud_t *pudp;
68 	pmd_t *pmdp;
69 	pte_t *ptep;
70 
71 	pgdp = pgd_offset_k(ea);
72 	if (pgd_none(*pgdp)) {
73 		pudp = early_alloc_pgtable(PUD_TABLE_SIZE, nid,
74 						region_start, region_end);
75 		pgd_populate(&init_mm, pgdp, pudp);
76 	}
77 	pudp = pud_offset(pgdp, ea);
78 	if (map_page_size == PUD_SIZE) {
79 		ptep = (pte_t *)pudp;
80 		goto set_the_pte;
81 	}
82 	if (pud_none(*pudp)) {
83 		pmdp = early_alloc_pgtable(PMD_TABLE_SIZE, nid,
84 						region_start, region_end);
85 		pud_populate(&init_mm, pudp, pmdp);
86 	}
87 	pmdp = pmd_offset(pudp, ea);
88 	if (map_page_size == PMD_SIZE) {
89 		ptep = pmdp_ptep(pmdp);
90 		goto set_the_pte;
91 	}
92 	if (!pmd_present(*pmdp)) {
93 		ptep = early_alloc_pgtable(PAGE_SIZE, nid,
94 						region_start, region_end);
95 		pmd_populate_kernel(&init_mm, pmdp, ptep);
96 	}
97 	ptep = pte_offset_kernel(pmdp, ea);
98 
99 set_the_pte:
100 	set_pte_at(&init_mm, ea, ptep, pfn_pte(pfn, flags));
101 	smp_wmb();
102 	return 0;
103 }
104 
105 /*
106  * nid, region_start, and region_end are hints to try to place the page
107  * table memory in the same node or region.
108  */
109 static int __map_kernel_page(unsigned long ea, unsigned long pa,
110 			  pgprot_t flags,
111 			  unsigned int map_page_size,
112 			  int nid,
113 			  unsigned long region_start, unsigned long region_end)
114 {
115 	unsigned long pfn = pa >> PAGE_SHIFT;
116 	pgd_t *pgdp;
117 	pud_t *pudp;
118 	pmd_t *pmdp;
119 	pte_t *ptep;
120 	/*
121 	 * Make sure task size is correct as per the max adddr
122 	 */
123 	BUILD_BUG_ON(TASK_SIZE_USER64 > RADIX_PGTABLE_RANGE);
124 
125 #ifdef CONFIG_PPC_64K_PAGES
126 	BUILD_BUG_ON(RADIX_KERN_MAP_SIZE != (1UL << MAX_EA_BITS_PER_CONTEXT));
127 #endif
128 
129 	if (unlikely(!slab_is_available()))
130 		return early_map_kernel_page(ea, pa, flags, map_page_size,
131 						nid, region_start, region_end);
132 
133 	/*
134 	 * Should make page table allocation functions be able to take a
135 	 * node, so we can place kernel page tables on the right nodes after
136 	 * boot.
137 	 */
138 	pgdp = pgd_offset_k(ea);
139 	pudp = pud_alloc(&init_mm, pgdp, ea);
140 	if (!pudp)
141 		return -ENOMEM;
142 	if (map_page_size == PUD_SIZE) {
143 		ptep = (pte_t *)pudp;
144 		goto set_the_pte;
145 	}
146 	pmdp = pmd_alloc(&init_mm, pudp, ea);
147 	if (!pmdp)
148 		return -ENOMEM;
149 	if (map_page_size == PMD_SIZE) {
150 		ptep = pmdp_ptep(pmdp);
151 		goto set_the_pte;
152 	}
153 	ptep = pte_alloc_kernel(pmdp, ea);
154 	if (!ptep)
155 		return -ENOMEM;
156 
157 set_the_pte:
158 	set_pte_at(&init_mm, ea, ptep, pfn_pte(pfn, flags));
159 	smp_wmb();
160 	return 0;
161 }
162 
163 int radix__map_kernel_page(unsigned long ea, unsigned long pa,
164 			  pgprot_t flags,
165 			  unsigned int map_page_size)
166 {
167 	return __map_kernel_page(ea, pa, flags, map_page_size, -1, 0, 0);
168 }
169 
170 #ifdef CONFIG_STRICT_KERNEL_RWX
171 void radix__change_memory_range(unsigned long start, unsigned long end,
172 				unsigned long clear)
173 {
174 	unsigned long idx;
175 	pgd_t *pgdp;
176 	pud_t *pudp;
177 	pmd_t *pmdp;
178 	pte_t *ptep;
179 
180 	start = ALIGN_DOWN(start, PAGE_SIZE);
181 	end = PAGE_ALIGN(end); // aligns up
182 
183 	pr_debug("Changing flags on range %lx-%lx removing 0x%lx\n",
184 		 start, end, clear);
185 
186 	for (idx = start; idx < end; idx += PAGE_SIZE) {
187 		pgdp = pgd_offset_k(idx);
188 		pudp = pud_alloc(&init_mm, pgdp, idx);
189 		if (!pudp)
190 			continue;
191 		if (pud_is_leaf(*pudp)) {
192 			ptep = (pte_t *)pudp;
193 			goto update_the_pte;
194 		}
195 		pmdp = pmd_alloc(&init_mm, pudp, idx);
196 		if (!pmdp)
197 			continue;
198 		if (pmd_is_leaf(*pmdp)) {
199 			ptep = pmdp_ptep(pmdp);
200 			goto update_the_pte;
201 		}
202 		ptep = pte_alloc_kernel(pmdp, idx);
203 		if (!ptep)
204 			continue;
205 update_the_pte:
206 		radix__pte_update(&init_mm, idx, ptep, clear, 0, 0);
207 	}
208 
209 	radix__flush_tlb_kernel_range(start, end);
210 }
211 
212 void radix__mark_rodata_ro(void)
213 {
214 	unsigned long start, end;
215 
216 	start = (unsigned long)_stext;
217 	end = (unsigned long)__init_begin;
218 
219 	radix__change_memory_range(start, end, _PAGE_WRITE);
220 }
221 
222 void radix__mark_initmem_nx(void)
223 {
224 	unsigned long start = (unsigned long)__init_begin;
225 	unsigned long end = (unsigned long)__init_end;
226 
227 	radix__change_memory_range(start, end, _PAGE_EXEC);
228 }
229 #endif /* CONFIG_STRICT_KERNEL_RWX */
230 
231 static inline void __meminit
232 print_mapping(unsigned long start, unsigned long end, unsigned long size, bool exec)
233 {
234 	char buf[10];
235 
236 	if (end <= start)
237 		return;
238 
239 	string_get_size(size, 1, STRING_UNITS_2, buf, sizeof(buf));
240 
241 	pr_info("Mapped 0x%016lx-0x%016lx with %s pages%s\n", start, end, buf,
242 		exec ? " (exec)" : "");
243 }
244 
245 static unsigned long next_boundary(unsigned long addr, unsigned long end)
246 {
247 #ifdef CONFIG_STRICT_KERNEL_RWX
248 	if (addr < __pa_symbol(__init_begin))
249 		return __pa_symbol(__init_begin);
250 #endif
251 	return end;
252 }
253 
254 static int __meminit create_physical_mapping(unsigned long start,
255 					     unsigned long end,
256 					     int nid)
257 {
258 	unsigned long vaddr, addr, mapping_size = 0;
259 	bool prev_exec, exec = false;
260 	pgprot_t prot;
261 	int psize;
262 
263 	start = _ALIGN_UP(start, PAGE_SIZE);
264 	for (addr = start; addr < end; addr += mapping_size) {
265 		unsigned long gap, previous_size;
266 		int rc;
267 
268 		gap = next_boundary(addr, end) - addr;
269 		previous_size = mapping_size;
270 		prev_exec = exec;
271 
272 		if (IS_ALIGNED(addr, PUD_SIZE) && gap >= PUD_SIZE &&
273 		    mmu_psize_defs[MMU_PAGE_1G].shift) {
274 			mapping_size = PUD_SIZE;
275 			psize = MMU_PAGE_1G;
276 		} else if (IS_ALIGNED(addr, PMD_SIZE) && gap >= PMD_SIZE &&
277 			   mmu_psize_defs[MMU_PAGE_2M].shift) {
278 			mapping_size = PMD_SIZE;
279 			psize = MMU_PAGE_2M;
280 		} else {
281 			mapping_size = PAGE_SIZE;
282 			psize = mmu_virtual_psize;
283 		}
284 
285 		vaddr = (unsigned long)__va(addr);
286 
287 		if (overlaps_kernel_text(vaddr, vaddr + mapping_size) ||
288 		    overlaps_interrupt_vector_text(vaddr, vaddr + mapping_size)) {
289 			prot = PAGE_KERNEL_X;
290 			exec = true;
291 		} else {
292 			prot = PAGE_KERNEL;
293 			exec = false;
294 		}
295 
296 		if (mapping_size != previous_size || exec != prev_exec) {
297 			print_mapping(start, addr, previous_size, prev_exec);
298 			start = addr;
299 		}
300 
301 		rc = __map_kernel_page(vaddr, addr, prot, mapping_size, nid, start, end);
302 		if (rc)
303 			return rc;
304 
305 		update_page_count(psize, 1);
306 	}
307 
308 	print_mapping(start, addr, mapping_size, exec);
309 	return 0;
310 }
311 
312 static void __init radix_init_pgtable(void)
313 {
314 	unsigned long rts_field;
315 	struct memblock_region *reg;
316 
317 	/* We don't support slb for radix */
318 	mmu_slb_size = 0;
319 	/*
320 	 * Create the linear mapping, using standard page size for now
321 	 */
322 	for_each_memblock(memory, reg) {
323 		/*
324 		 * The memblock allocator  is up at this point, so the
325 		 * page tables will be allocated within the range. No
326 		 * need or a node (which we don't have yet).
327 		 */
328 
329 		if ((reg->base + reg->size) >= RADIX_VMALLOC_START) {
330 			pr_warn("Outside the supported range\n");
331 			continue;
332 		}
333 
334 		WARN_ON(create_physical_mapping(reg->base,
335 						reg->base + reg->size,
336 						-1));
337 	}
338 
339 	/* Find out how many PID bits are supported */
340 	if (cpu_has_feature(CPU_FTR_HVMODE)) {
341 		if (!mmu_pid_bits)
342 			mmu_pid_bits = 20;
343 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
344 		/*
345 		 * When KVM is possible, we only use the top half of the
346 		 * PID space to avoid collisions between host and guest PIDs
347 		 * which can cause problems due to prefetch when exiting the
348 		 * guest with AIL=3
349 		 */
350 		mmu_base_pid = 1 << (mmu_pid_bits - 1);
351 #else
352 		mmu_base_pid = 1;
353 #endif
354 	} else {
355 		/* The guest uses the bottom half of the PID space */
356 		if (!mmu_pid_bits)
357 			mmu_pid_bits = 19;
358 		mmu_base_pid = 1;
359 	}
360 
361 	/*
362 	 * Allocate Partition table and process table for the
363 	 * host.
364 	 */
365 	BUG_ON(PRTB_SIZE_SHIFT > 36);
366 	process_tb = early_alloc_pgtable(1UL << PRTB_SIZE_SHIFT, -1, 0, 0);
367 	/*
368 	 * Fill in the process table.
369 	 */
370 	rts_field = radix__get_tree_size();
371 	process_tb->prtb0 = cpu_to_be64(rts_field | __pa(init_mm.pgd) | RADIX_PGD_INDEX_SIZE);
372 
373 	/*
374 	 * The init_mm context is given the first available (non-zero) PID,
375 	 * which is the "guard PID" and contains no page table. PIDR should
376 	 * never be set to zero because that duplicates the kernel address
377 	 * space at the 0x0... offset (quadrant 0)!
378 	 *
379 	 * An arbitrary PID that may later be allocated by the PID allocator
380 	 * for userspace processes must not be used either, because that
381 	 * would cause stale user mappings for that PID on CPUs outside of
382 	 * the TLB invalidation scheme (because it won't be in mm_cpumask).
383 	 *
384 	 * So permanently carve out one PID for the purpose of a guard PID.
385 	 */
386 	init_mm.context.id = mmu_base_pid;
387 	mmu_base_pid++;
388 }
389 
390 static void __init radix_init_partition_table(void)
391 {
392 	unsigned long rts_field, dw0, dw1;
393 
394 	mmu_partition_table_init();
395 	rts_field = radix__get_tree_size();
396 	dw0 = rts_field | __pa(init_mm.pgd) | RADIX_PGD_INDEX_SIZE | PATB_HR;
397 	dw1 = __pa(process_tb) | (PRTB_SIZE_SHIFT - 12) | PATB_GR;
398 	mmu_partition_table_set_entry(0, dw0, dw1, false);
399 
400 	pr_info("Initializing Radix MMU\n");
401 }
402 
403 static int __init get_idx_from_shift(unsigned int shift)
404 {
405 	int idx = -1;
406 
407 	switch (shift) {
408 	case 0xc:
409 		idx = MMU_PAGE_4K;
410 		break;
411 	case 0x10:
412 		idx = MMU_PAGE_64K;
413 		break;
414 	case 0x15:
415 		idx = MMU_PAGE_2M;
416 		break;
417 	case 0x1e:
418 		idx = MMU_PAGE_1G;
419 		break;
420 	}
421 	return idx;
422 }
423 
424 static int __init radix_dt_scan_page_sizes(unsigned long node,
425 					   const char *uname, int depth,
426 					   void *data)
427 {
428 	int size = 0;
429 	int shift, idx;
430 	unsigned int ap;
431 	const __be32 *prop;
432 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
433 
434 	/* We are scanning "cpu" nodes only */
435 	if (type == NULL || strcmp(type, "cpu") != 0)
436 		return 0;
437 
438 	/* Find MMU PID size */
439 	prop = of_get_flat_dt_prop(node, "ibm,mmu-pid-bits", &size);
440 	if (prop && size == 4)
441 		mmu_pid_bits = be32_to_cpup(prop);
442 
443 	/* Grab page size encodings */
444 	prop = of_get_flat_dt_prop(node, "ibm,processor-radix-AP-encodings", &size);
445 	if (!prop)
446 		return 0;
447 
448 	pr_info("Page sizes from device-tree:\n");
449 	for (; size >= 4; size -= 4, ++prop) {
450 
451 		struct mmu_psize_def *def;
452 
453 		/* top 3 bit is AP encoding */
454 		shift = be32_to_cpu(prop[0]) & ~(0xe << 28);
455 		ap = be32_to_cpu(prop[0]) >> 29;
456 		pr_info("Page size shift = %d AP=0x%x\n", shift, ap);
457 
458 		idx = get_idx_from_shift(shift);
459 		if (idx < 0)
460 			continue;
461 
462 		def = &mmu_psize_defs[idx];
463 		def->shift = shift;
464 		def->ap  = ap;
465 	}
466 
467 	/* needed ? */
468 	cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B;
469 	return 1;
470 }
471 
472 void __init radix__early_init_devtree(void)
473 {
474 	int rc;
475 
476 	/*
477 	 * Try to find the available page sizes in the device-tree
478 	 */
479 	rc = of_scan_flat_dt(radix_dt_scan_page_sizes, NULL);
480 	if (rc != 0)  /* Found */
481 		goto found;
482 	/*
483 	 * let's assume we have page 4k and 64k support
484 	 */
485 	mmu_psize_defs[MMU_PAGE_4K].shift = 12;
486 	mmu_psize_defs[MMU_PAGE_4K].ap = 0x0;
487 
488 	mmu_psize_defs[MMU_PAGE_64K].shift = 16;
489 	mmu_psize_defs[MMU_PAGE_64K].ap = 0x5;
490 found:
491 	return;
492 }
493 
494 static void radix_init_amor(void)
495 {
496 	/*
497 	* In HV mode, we init AMOR (Authority Mask Override Register) so that
498 	* the hypervisor and guest can setup IAMR (Instruction Authority Mask
499 	* Register), enable key 0 and set it to 1.
500 	*
501 	* AMOR = 0b1100 .... 0000 (Mask for key 0 is 11)
502 	*/
503 	mtspr(SPRN_AMOR, (3ul << 62));
504 }
505 
506 #ifdef CONFIG_PPC_KUEP
507 void setup_kuep(bool disabled)
508 {
509 	if (disabled || !early_radix_enabled())
510 		return;
511 
512 	if (smp_processor_id() == boot_cpuid)
513 		pr_info("Activating Kernel Userspace Execution Prevention\n");
514 
515 	/*
516 	 * Radix always uses key0 of the IAMR to determine if an access is
517 	 * allowed. We set bit 0 (IBM bit 1) of key0, to prevent instruction
518 	 * fetch.
519 	 */
520 	mtspr(SPRN_IAMR, (1ul << 62));
521 }
522 #endif
523 
524 #ifdef CONFIG_PPC_KUAP
525 void setup_kuap(bool disabled)
526 {
527 	if (disabled || !early_radix_enabled())
528 		return;
529 
530 	if (smp_processor_id() == boot_cpuid) {
531 		pr_info("Activating Kernel Userspace Access Prevention\n");
532 		cur_cpu_spec->mmu_features |= MMU_FTR_RADIX_KUAP;
533 	}
534 
535 	/* Make sure userspace can't change the AMR */
536 	mtspr(SPRN_UAMOR, 0);
537 	mtspr(SPRN_AMR, AMR_KUAP_BLOCKED);
538 	isync();
539 }
540 #endif
541 
542 void __init radix__early_init_mmu(void)
543 {
544 	unsigned long lpcr;
545 
546 #ifdef CONFIG_PPC_64K_PAGES
547 	/* PAGE_SIZE mappings */
548 	mmu_virtual_psize = MMU_PAGE_64K;
549 #else
550 	mmu_virtual_psize = MMU_PAGE_4K;
551 #endif
552 
553 #ifdef CONFIG_SPARSEMEM_VMEMMAP
554 	/* vmemmap mapping */
555 	if (mmu_psize_defs[MMU_PAGE_2M].shift) {
556 		/*
557 		 * map vmemmap using 2M if available
558 		 */
559 		mmu_vmemmap_psize = MMU_PAGE_2M;
560 	} else
561 		mmu_vmemmap_psize = mmu_virtual_psize;
562 #endif
563 	/*
564 	 * initialize page table size
565 	 */
566 	__pte_index_size = RADIX_PTE_INDEX_SIZE;
567 	__pmd_index_size = RADIX_PMD_INDEX_SIZE;
568 	__pud_index_size = RADIX_PUD_INDEX_SIZE;
569 	__pgd_index_size = RADIX_PGD_INDEX_SIZE;
570 	__pud_cache_index = RADIX_PUD_INDEX_SIZE;
571 	__pte_table_size = RADIX_PTE_TABLE_SIZE;
572 	__pmd_table_size = RADIX_PMD_TABLE_SIZE;
573 	__pud_table_size = RADIX_PUD_TABLE_SIZE;
574 	__pgd_table_size = RADIX_PGD_TABLE_SIZE;
575 
576 	__pmd_val_bits = RADIX_PMD_VAL_BITS;
577 	__pud_val_bits = RADIX_PUD_VAL_BITS;
578 	__pgd_val_bits = RADIX_PGD_VAL_BITS;
579 
580 	__kernel_virt_start = RADIX_KERN_VIRT_START;
581 	__vmalloc_start = RADIX_VMALLOC_START;
582 	__vmalloc_end = RADIX_VMALLOC_END;
583 	__kernel_io_start = RADIX_KERN_IO_START;
584 	__kernel_io_end = RADIX_KERN_IO_END;
585 	vmemmap = (struct page *)RADIX_VMEMMAP_START;
586 	ioremap_bot = IOREMAP_BASE;
587 
588 #ifdef CONFIG_PCI
589 	pci_io_base = ISA_IO_BASE;
590 #endif
591 	__pte_frag_nr = RADIX_PTE_FRAG_NR;
592 	__pte_frag_size_shift = RADIX_PTE_FRAG_SIZE_SHIFT;
593 	__pmd_frag_nr = RADIX_PMD_FRAG_NR;
594 	__pmd_frag_size_shift = RADIX_PMD_FRAG_SIZE_SHIFT;
595 
596 	radix_init_pgtable();
597 
598 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
599 		lpcr = mfspr(SPRN_LPCR);
600 		mtspr(SPRN_LPCR, lpcr | LPCR_UPRT | LPCR_HR);
601 		radix_init_partition_table();
602 		radix_init_amor();
603 	} else {
604 		radix_init_pseries();
605 	}
606 
607 	memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
608 
609 	/* Switch to the guard PID before turning on MMU */
610 	radix__switch_mmu_context(NULL, &init_mm);
611 	tlbiel_all();
612 }
613 
614 void radix__early_init_mmu_secondary(void)
615 {
616 	unsigned long lpcr;
617 	/*
618 	 * update partition table control register and UPRT
619 	 */
620 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
621 		lpcr = mfspr(SPRN_LPCR);
622 		mtspr(SPRN_LPCR, lpcr | LPCR_UPRT | LPCR_HR);
623 
624 		set_ptcr_when_no_uv(__pa(partition_tb) |
625 				    (PATB_SIZE_SHIFT - 12));
626 
627 		radix_init_amor();
628 	}
629 
630 	radix__switch_mmu_context(NULL, &init_mm);
631 	tlbiel_all();
632 }
633 
634 void radix__mmu_cleanup_all(void)
635 {
636 	unsigned long lpcr;
637 
638 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
639 		lpcr = mfspr(SPRN_LPCR);
640 		mtspr(SPRN_LPCR, lpcr & ~LPCR_UPRT);
641 		set_ptcr_when_no_uv(0);
642 		powernv_set_nmmu_ptcr(0);
643 		radix__flush_tlb_all();
644 	}
645 }
646 
647 void radix__setup_initial_memory_limit(phys_addr_t first_memblock_base,
648 				phys_addr_t first_memblock_size)
649 {
650 	/*
651 	 * We don't currently support the first MEMBLOCK not mapping 0
652 	 * physical on those processors
653 	 */
654 	BUG_ON(first_memblock_base != 0);
655 
656 	/*
657 	 * Radix mode is not limited by RMA / VRMA addressing.
658 	 */
659 	ppc64_rma_size = ULONG_MAX;
660 }
661 
662 #ifdef CONFIG_MEMORY_HOTPLUG
663 static void free_pte_table(pte_t *pte_start, pmd_t *pmd)
664 {
665 	pte_t *pte;
666 	int i;
667 
668 	for (i = 0; i < PTRS_PER_PTE; i++) {
669 		pte = pte_start + i;
670 		if (!pte_none(*pte))
671 			return;
672 	}
673 
674 	pte_free_kernel(&init_mm, pte_start);
675 	pmd_clear(pmd);
676 }
677 
678 static void free_pmd_table(pmd_t *pmd_start, pud_t *pud)
679 {
680 	pmd_t *pmd;
681 	int i;
682 
683 	for (i = 0; i < PTRS_PER_PMD; i++) {
684 		pmd = pmd_start + i;
685 		if (!pmd_none(*pmd))
686 			return;
687 	}
688 
689 	pmd_free(&init_mm, pmd_start);
690 	pud_clear(pud);
691 }
692 
693 struct change_mapping_params {
694 	pte_t *pte;
695 	unsigned long start;
696 	unsigned long end;
697 	unsigned long aligned_start;
698 	unsigned long aligned_end;
699 };
700 
701 static int __meminit stop_machine_change_mapping(void *data)
702 {
703 	struct change_mapping_params *params =
704 			(struct change_mapping_params *)data;
705 
706 	if (!data)
707 		return -1;
708 
709 	spin_unlock(&init_mm.page_table_lock);
710 	pte_clear(&init_mm, params->aligned_start, params->pte);
711 	create_physical_mapping(__pa(params->aligned_start), __pa(params->start), -1);
712 	create_physical_mapping(__pa(params->end), __pa(params->aligned_end), -1);
713 	spin_lock(&init_mm.page_table_lock);
714 	return 0;
715 }
716 
717 static void remove_pte_table(pte_t *pte_start, unsigned long addr,
718 			     unsigned long end)
719 {
720 	unsigned long next;
721 	pte_t *pte;
722 
723 	pte = pte_start + pte_index(addr);
724 	for (; addr < end; addr = next, pte++) {
725 		next = (addr + PAGE_SIZE) & PAGE_MASK;
726 		if (next > end)
727 			next = end;
728 
729 		if (!pte_present(*pte))
730 			continue;
731 
732 		if (!PAGE_ALIGNED(addr) || !PAGE_ALIGNED(next)) {
733 			/*
734 			 * The vmemmap_free() and remove_section_mapping()
735 			 * codepaths call us with aligned addresses.
736 			 */
737 			WARN_ONCE(1, "%s: unaligned range\n", __func__);
738 			continue;
739 		}
740 
741 		pte_clear(&init_mm, addr, pte);
742 	}
743 }
744 
745 /*
746  * clear the pte and potentially split the mapping helper
747  */
748 static void __meminit split_kernel_mapping(unsigned long addr, unsigned long end,
749 				unsigned long size, pte_t *pte)
750 {
751 	unsigned long mask = ~(size - 1);
752 	unsigned long aligned_start = addr & mask;
753 	unsigned long aligned_end = addr + size;
754 	struct change_mapping_params params;
755 	bool split_region = false;
756 
757 	if ((end - addr) < size) {
758 		/*
759 		 * We're going to clear the PTE, but not flushed
760 		 * the mapping, time to remap and flush. The
761 		 * effects if visible outside the processor or
762 		 * if we are running in code close to the
763 		 * mapping we cleared, we are in trouble.
764 		 */
765 		if (overlaps_kernel_text(aligned_start, addr) ||
766 			overlaps_kernel_text(end, aligned_end)) {
767 			/*
768 			 * Hack, just return, don't pte_clear
769 			 */
770 			WARN_ONCE(1, "Linear mapping %lx->%lx overlaps kernel "
771 				  "text, not splitting\n", addr, end);
772 			return;
773 		}
774 		split_region = true;
775 	}
776 
777 	if (split_region) {
778 		params.pte = pte;
779 		params.start = addr;
780 		params.end = end;
781 		params.aligned_start = addr & ~(size - 1);
782 		params.aligned_end = min_t(unsigned long, aligned_end,
783 				(unsigned long)__va(memblock_end_of_DRAM()));
784 		stop_machine(stop_machine_change_mapping, &params, NULL);
785 		return;
786 	}
787 
788 	pte_clear(&init_mm, addr, pte);
789 }
790 
791 static void remove_pmd_table(pmd_t *pmd_start, unsigned long addr,
792 			     unsigned long end)
793 {
794 	unsigned long next;
795 	pte_t *pte_base;
796 	pmd_t *pmd;
797 
798 	pmd = pmd_start + pmd_index(addr);
799 	for (; addr < end; addr = next, pmd++) {
800 		next = pmd_addr_end(addr, end);
801 
802 		if (!pmd_present(*pmd))
803 			continue;
804 
805 		if (pmd_is_leaf(*pmd)) {
806 			split_kernel_mapping(addr, end, PMD_SIZE, (pte_t *)pmd);
807 			continue;
808 		}
809 
810 		pte_base = (pte_t *)pmd_page_vaddr(*pmd);
811 		remove_pte_table(pte_base, addr, next);
812 		free_pte_table(pte_base, pmd);
813 	}
814 }
815 
816 static void remove_pud_table(pud_t *pud_start, unsigned long addr,
817 			     unsigned long end)
818 {
819 	unsigned long next;
820 	pmd_t *pmd_base;
821 	pud_t *pud;
822 
823 	pud = pud_start + pud_index(addr);
824 	for (; addr < end; addr = next, pud++) {
825 		next = pud_addr_end(addr, end);
826 
827 		if (!pud_present(*pud))
828 			continue;
829 
830 		if (pud_is_leaf(*pud)) {
831 			split_kernel_mapping(addr, end, PUD_SIZE, (pte_t *)pud);
832 			continue;
833 		}
834 
835 		pmd_base = (pmd_t *)pud_page_vaddr(*pud);
836 		remove_pmd_table(pmd_base, addr, next);
837 		free_pmd_table(pmd_base, pud);
838 	}
839 }
840 
841 static void __meminit remove_pagetable(unsigned long start, unsigned long end)
842 {
843 	unsigned long addr, next;
844 	pud_t *pud_base;
845 	pgd_t *pgd;
846 
847 	spin_lock(&init_mm.page_table_lock);
848 
849 	for (addr = start; addr < end; addr = next) {
850 		next = pgd_addr_end(addr, end);
851 
852 		pgd = pgd_offset_k(addr);
853 		if (!pgd_present(*pgd))
854 			continue;
855 
856 		if (pgd_is_leaf(*pgd)) {
857 			split_kernel_mapping(addr, end, PGDIR_SIZE, (pte_t *)pgd);
858 			continue;
859 		}
860 
861 		pud_base = (pud_t *)pgd_page_vaddr(*pgd);
862 		remove_pud_table(pud_base, addr, next);
863 	}
864 
865 	spin_unlock(&init_mm.page_table_lock);
866 	radix__flush_tlb_kernel_range(start, end);
867 }
868 
869 int __meminit radix__create_section_mapping(unsigned long start, unsigned long end, int nid)
870 {
871 	if (end >= RADIX_VMALLOC_START) {
872 		pr_warn("Outside the supported range\n");
873 		return -1;
874 	}
875 
876 	return create_physical_mapping(__pa(start), __pa(end), nid);
877 }
878 
879 int __meminit radix__remove_section_mapping(unsigned long start, unsigned long end)
880 {
881 	remove_pagetable(start, end);
882 	return 0;
883 }
884 #endif /* CONFIG_MEMORY_HOTPLUG */
885 
886 #ifdef CONFIG_SPARSEMEM_VMEMMAP
887 static int __map_kernel_page_nid(unsigned long ea, unsigned long pa,
888 				 pgprot_t flags, unsigned int map_page_size,
889 				 int nid)
890 {
891 	return __map_kernel_page(ea, pa, flags, map_page_size, nid, 0, 0);
892 }
893 
894 int __meminit radix__vmemmap_create_mapping(unsigned long start,
895 				      unsigned long page_size,
896 				      unsigned long phys)
897 {
898 	/* Create a PTE encoding */
899 	unsigned long flags = _PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_KERNEL_RW;
900 	int nid = early_pfn_to_nid(phys >> PAGE_SHIFT);
901 	int ret;
902 
903 	if ((start + page_size) >= RADIX_VMEMMAP_END) {
904 		pr_warn("Outside the supported range\n");
905 		return -1;
906 	}
907 
908 	ret = __map_kernel_page_nid(start, phys, __pgprot(flags), page_size, nid);
909 	BUG_ON(ret);
910 
911 	return 0;
912 }
913 
914 #ifdef CONFIG_MEMORY_HOTPLUG
915 void __meminit radix__vmemmap_remove_mapping(unsigned long start, unsigned long page_size)
916 {
917 	remove_pagetable(start, start + page_size);
918 }
919 #endif
920 #endif
921 
922 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
923 
924 unsigned long radix__pmd_hugepage_update(struct mm_struct *mm, unsigned long addr,
925 				  pmd_t *pmdp, unsigned long clr,
926 				  unsigned long set)
927 {
928 	unsigned long old;
929 
930 #ifdef CONFIG_DEBUG_VM
931 	WARN_ON(!radix__pmd_trans_huge(*pmdp) && !pmd_devmap(*pmdp));
932 	assert_spin_locked(pmd_lockptr(mm, pmdp));
933 #endif
934 
935 	old = radix__pte_update(mm, addr, (pte_t *)pmdp, clr, set, 1);
936 	trace_hugepage_update(addr, old, clr, set);
937 
938 	return old;
939 }
940 
941 pmd_t radix__pmdp_collapse_flush(struct vm_area_struct *vma, unsigned long address,
942 			pmd_t *pmdp)
943 
944 {
945 	pmd_t pmd;
946 
947 	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
948 	VM_BUG_ON(radix__pmd_trans_huge(*pmdp));
949 	VM_BUG_ON(pmd_devmap(*pmdp));
950 	/*
951 	 * khugepaged calls this for normal pmd
952 	 */
953 	pmd = *pmdp;
954 	pmd_clear(pmdp);
955 
956 	/*FIXME!!  Verify whether we need this kick below */
957 	serialize_against_pte_lookup(vma->vm_mm);
958 
959 	radix__flush_tlb_collapsed_pmd(vma->vm_mm, address);
960 
961 	return pmd;
962 }
963 
964 /*
965  * For us pgtable_t is pte_t *. Inorder to save the deposisted
966  * page table, we consider the allocated page table as a list
967  * head. On withdraw we need to make sure we zero out the used
968  * list_head memory area.
969  */
970 void radix__pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
971 				 pgtable_t pgtable)
972 {
973 	struct list_head *lh = (struct list_head *) pgtable;
974 
975 	assert_spin_locked(pmd_lockptr(mm, pmdp));
976 
977 	/* FIFO */
978 	if (!pmd_huge_pte(mm, pmdp))
979 		INIT_LIST_HEAD(lh);
980 	else
981 		list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
982 	pmd_huge_pte(mm, pmdp) = pgtable;
983 }
984 
985 pgtable_t radix__pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
986 {
987 	pte_t *ptep;
988 	pgtable_t pgtable;
989 	struct list_head *lh;
990 
991 	assert_spin_locked(pmd_lockptr(mm, pmdp));
992 
993 	/* FIFO */
994 	pgtable = pmd_huge_pte(mm, pmdp);
995 	lh = (struct list_head *) pgtable;
996 	if (list_empty(lh))
997 		pmd_huge_pte(mm, pmdp) = NULL;
998 	else {
999 		pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
1000 		list_del(lh);
1001 	}
1002 	ptep = (pte_t *) pgtable;
1003 	*ptep = __pte(0);
1004 	ptep++;
1005 	*ptep = __pte(0);
1006 	return pgtable;
1007 }
1008 
1009 pmd_t radix__pmdp_huge_get_and_clear(struct mm_struct *mm,
1010 				     unsigned long addr, pmd_t *pmdp)
1011 {
1012 	pmd_t old_pmd;
1013 	unsigned long old;
1014 
1015 	old = radix__pmd_hugepage_update(mm, addr, pmdp, ~0UL, 0);
1016 	old_pmd = __pmd(old);
1017 	/*
1018 	 * Serialize against find_current_mm_pte which does lock-less
1019 	 * lookup in page tables with local interrupts disabled. For huge pages
1020 	 * it casts pmd_t to pte_t. Since format of pte_t is different from
1021 	 * pmd_t we want to prevent transit from pmd pointing to page table
1022 	 * to pmd pointing to huge page (and back) while interrupts are disabled.
1023 	 * We clear pmd to possibly replace it with page table pointer in
1024 	 * different code paths. So make sure we wait for the parallel
1025 	 * find_current_mm_pte to finish.
1026 	 */
1027 	serialize_against_pte_lookup(mm);
1028 	return old_pmd;
1029 }
1030 
1031 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1032 
1033 void radix__ptep_set_access_flags(struct vm_area_struct *vma, pte_t *ptep,
1034 				  pte_t entry, unsigned long address, int psize)
1035 {
1036 	struct mm_struct *mm = vma->vm_mm;
1037 	unsigned long set = pte_val(entry) & (_PAGE_DIRTY | _PAGE_ACCESSED |
1038 					      _PAGE_RW | _PAGE_EXEC);
1039 
1040 	unsigned long change = pte_val(entry) ^ pte_val(*ptep);
1041 	/*
1042 	 * To avoid NMMU hang while relaxing access, we need mark
1043 	 * the pte invalid in between.
1044 	 */
1045 	if ((change & _PAGE_RW) && atomic_read(&mm->context.copros) > 0) {
1046 		unsigned long old_pte, new_pte;
1047 
1048 		old_pte = __radix_pte_update(ptep, _PAGE_PRESENT, _PAGE_INVALID);
1049 		/*
1050 		 * new value of pte
1051 		 */
1052 		new_pte = old_pte | set;
1053 		radix__flush_tlb_page_psize(mm, address, psize);
1054 		__radix_pte_update(ptep, _PAGE_INVALID, new_pte);
1055 	} else {
1056 		__radix_pte_update(ptep, 0, set);
1057 		/*
1058 		 * Book3S does not require a TLB flush when relaxing access
1059 		 * restrictions when the address space is not attached to a
1060 		 * NMMU, because the core MMU will reload the pte after taking
1061 		 * an access fault, which is defined by the architectue.
1062 		 */
1063 	}
1064 	/* See ptesync comment in radix__set_pte_at */
1065 }
1066 
1067 void radix__ptep_modify_prot_commit(struct vm_area_struct *vma,
1068 				    unsigned long addr, pte_t *ptep,
1069 				    pte_t old_pte, pte_t pte)
1070 {
1071 	struct mm_struct *mm = vma->vm_mm;
1072 
1073 	/*
1074 	 * To avoid NMMU hang while relaxing access we need to flush the tlb before
1075 	 * we set the new value. We need to do this only for radix, because hash
1076 	 * translation does flush when updating the linux pte.
1077 	 */
1078 	if (is_pte_rw_upgrade(pte_val(old_pte), pte_val(pte)) &&
1079 	    (atomic_read(&mm->context.copros) > 0))
1080 		radix__flush_tlb_page(vma, addr);
1081 
1082 	set_pte_at(mm, addr, ptep, pte);
1083 }
1084 
1085 int __init arch_ioremap_pud_supported(void)
1086 {
1087 	/* HPT does not cope with large pages in the vmalloc area */
1088 	return radix_enabled();
1089 }
1090 
1091 int __init arch_ioremap_pmd_supported(void)
1092 {
1093 	return radix_enabled();
1094 }
1095 
1096 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1097 {
1098 	return 0;
1099 }
1100 
1101 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1102 {
1103 	pte_t *ptep = (pte_t *)pud;
1104 	pte_t new_pud = pfn_pte(__phys_to_pfn(addr), prot);
1105 
1106 	if (!radix_enabled())
1107 		return 0;
1108 
1109 	set_pte_at(&init_mm, 0 /* radix unused */, ptep, new_pud);
1110 
1111 	return 1;
1112 }
1113 
1114 int pud_clear_huge(pud_t *pud)
1115 {
1116 	if (pud_huge(*pud)) {
1117 		pud_clear(pud);
1118 		return 1;
1119 	}
1120 
1121 	return 0;
1122 }
1123 
1124 int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1125 {
1126 	pmd_t *pmd;
1127 	int i;
1128 
1129 	pmd = (pmd_t *)pud_page_vaddr(*pud);
1130 	pud_clear(pud);
1131 
1132 	flush_tlb_kernel_range(addr, addr + PUD_SIZE);
1133 
1134 	for (i = 0; i < PTRS_PER_PMD; i++) {
1135 		if (!pmd_none(pmd[i])) {
1136 			pte_t *pte;
1137 			pte = (pte_t *)pmd_page_vaddr(pmd[i]);
1138 
1139 			pte_free_kernel(&init_mm, pte);
1140 		}
1141 	}
1142 
1143 	pmd_free(&init_mm, pmd);
1144 
1145 	return 1;
1146 }
1147 
1148 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1149 {
1150 	pte_t *ptep = (pte_t *)pmd;
1151 	pte_t new_pmd = pfn_pte(__phys_to_pfn(addr), prot);
1152 
1153 	if (!radix_enabled())
1154 		return 0;
1155 
1156 	set_pte_at(&init_mm, 0 /* radix unused */, ptep, new_pmd);
1157 
1158 	return 1;
1159 }
1160 
1161 int pmd_clear_huge(pmd_t *pmd)
1162 {
1163 	if (pmd_huge(*pmd)) {
1164 		pmd_clear(pmd);
1165 		return 1;
1166 	}
1167 
1168 	return 0;
1169 }
1170 
1171 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1172 {
1173 	pte_t *pte;
1174 
1175 	pte = (pte_t *)pmd_page_vaddr(*pmd);
1176 	pmd_clear(pmd);
1177 
1178 	flush_tlb_kernel_range(addr, addr + PMD_SIZE);
1179 
1180 	pte_free_kernel(&init_mm, pte);
1181 
1182 	return 1;
1183 }
1184 
1185 int __init arch_ioremap_p4d_supported(void)
1186 {
1187 	return 0;
1188 }
1189