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_P9_RADIX_PREFETCH_BUG)) {
341 		if (!mmu_pid_bits)
342 			mmu_pid_bits = 20;
343 		mmu_base_pid = 1;
344 	} else if (cpu_has_feature(CPU_FTR_HVMODE)) {
345 		if (!mmu_pid_bits)
346 			mmu_pid_bits = 20;
347 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
348 		/*
349 		 * When KVM is possible, we only use the top half of the
350 		 * PID space to avoid collisions between host and guest PIDs
351 		 * which can cause problems due to prefetch when exiting the
352 		 * guest with AIL=3
353 		 */
354 		mmu_base_pid = 1 << (mmu_pid_bits - 1);
355 #else
356 		mmu_base_pid = 1;
357 #endif
358 	} else {
359 		/* The guest uses the bottom half of the PID space */
360 		if (!mmu_pid_bits)
361 			mmu_pid_bits = 19;
362 		mmu_base_pid = 1;
363 	}
364 
365 	/*
366 	 * Allocate Partition table and process table for the
367 	 * host.
368 	 */
369 	BUG_ON(PRTB_SIZE_SHIFT > 36);
370 	process_tb = early_alloc_pgtable(1UL << PRTB_SIZE_SHIFT, -1, 0, 0);
371 	/*
372 	 * Fill in the process table.
373 	 */
374 	rts_field = radix__get_tree_size();
375 	process_tb->prtb0 = cpu_to_be64(rts_field | __pa(init_mm.pgd) | RADIX_PGD_INDEX_SIZE);
376 
377 	/*
378 	 * The init_mm context is given the first available (non-zero) PID,
379 	 * which is the "guard PID" and contains no page table. PIDR should
380 	 * never be set to zero because that duplicates the kernel address
381 	 * space at the 0x0... offset (quadrant 0)!
382 	 *
383 	 * An arbitrary PID that may later be allocated by the PID allocator
384 	 * for userspace processes must not be used either, because that
385 	 * would cause stale user mappings for that PID on CPUs outside of
386 	 * the TLB invalidation scheme (because it won't be in mm_cpumask).
387 	 *
388 	 * So permanently carve out one PID for the purpose of a guard PID.
389 	 */
390 	init_mm.context.id = mmu_base_pid;
391 	mmu_base_pid++;
392 }
393 
394 static void __init radix_init_partition_table(void)
395 {
396 	unsigned long rts_field, dw0, dw1;
397 
398 	mmu_partition_table_init();
399 	rts_field = radix__get_tree_size();
400 	dw0 = rts_field | __pa(init_mm.pgd) | RADIX_PGD_INDEX_SIZE | PATB_HR;
401 	dw1 = __pa(process_tb) | (PRTB_SIZE_SHIFT - 12) | PATB_GR;
402 	mmu_partition_table_set_entry(0, dw0, dw1, false);
403 
404 	pr_info("Initializing Radix MMU\n");
405 }
406 
407 static int __init get_idx_from_shift(unsigned int shift)
408 {
409 	int idx = -1;
410 
411 	switch (shift) {
412 	case 0xc:
413 		idx = MMU_PAGE_4K;
414 		break;
415 	case 0x10:
416 		idx = MMU_PAGE_64K;
417 		break;
418 	case 0x15:
419 		idx = MMU_PAGE_2M;
420 		break;
421 	case 0x1e:
422 		idx = MMU_PAGE_1G;
423 		break;
424 	}
425 	return idx;
426 }
427 
428 static int __init radix_dt_scan_page_sizes(unsigned long node,
429 					   const char *uname, int depth,
430 					   void *data)
431 {
432 	int size = 0;
433 	int shift, idx;
434 	unsigned int ap;
435 	const __be32 *prop;
436 	const char *type = of_get_flat_dt_prop(node, "device_type", NULL);
437 
438 	/* We are scanning "cpu" nodes only */
439 	if (type == NULL || strcmp(type, "cpu") != 0)
440 		return 0;
441 
442 	/* Find MMU PID size */
443 	prop = of_get_flat_dt_prop(node, "ibm,mmu-pid-bits", &size);
444 	if (prop && size == 4)
445 		mmu_pid_bits = be32_to_cpup(prop);
446 
447 	/* Grab page size encodings */
448 	prop = of_get_flat_dt_prop(node, "ibm,processor-radix-AP-encodings", &size);
449 	if (!prop)
450 		return 0;
451 
452 	pr_info("Page sizes from device-tree:\n");
453 	for (; size >= 4; size -= 4, ++prop) {
454 
455 		struct mmu_psize_def *def;
456 
457 		/* top 3 bit is AP encoding */
458 		shift = be32_to_cpu(prop[0]) & ~(0xe << 28);
459 		ap = be32_to_cpu(prop[0]) >> 29;
460 		pr_info("Page size shift = %d AP=0x%x\n", shift, ap);
461 
462 		idx = get_idx_from_shift(shift);
463 		if (idx < 0)
464 			continue;
465 
466 		def = &mmu_psize_defs[idx];
467 		def->shift = shift;
468 		def->ap  = ap;
469 	}
470 
471 	/* needed ? */
472 	cur_cpu_spec->mmu_features &= ~MMU_FTR_NO_SLBIE_B;
473 	return 1;
474 }
475 
476 void __init radix__early_init_devtree(void)
477 {
478 	int rc;
479 
480 	/*
481 	 * Try to find the available page sizes in the device-tree
482 	 */
483 	rc = of_scan_flat_dt(radix_dt_scan_page_sizes, NULL);
484 	if (rc != 0)  /* Found */
485 		goto found;
486 	/*
487 	 * let's assume we have page 4k and 64k support
488 	 */
489 	mmu_psize_defs[MMU_PAGE_4K].shift = 12;
490 	mmu_psize_defs[MMU_PAGE_4K].ap = 0x0;
491 
492 	mmu_psize_defs[MMU_PAGE_64K].shift = 16;
493 	mmu_psize_defs[MMU_PAGE_64K].ap = 0x5;
494 found:
495 	return;
496 }
497 
498 static void radix_init_amor(void)
499 {
500 	/*
501 	* In HV mode, we init AMOR (Authority Mask Override Register) so that
502 	* the hypervisor and guest can setup IAMR (Instruction Authority Mask
503 	* Register), enable key 0 and set it to 1.
504 	*
505 	* AMOR = 0b1100 .... 0000 (Mask for key 0 is 11)
506 	*/
507 	mtspr(SPRN_AMOR, (3ul << 62));
508 }
509 
510 #ifdef CONFIG_PPC_KUEP
511 void setup_kuep(bool disabled)
512 {
513 	if (disabled || !early_radix_enabled())
514 		return;
515 
516 	if (smp_processor_id() == boot_cpuid)
517 		pr_info("Activating Kernel Userspace Execution Prevention\n");
518 
519 	/*
520 	 * Radix always uses key0 of the IAMR to determine if an access is
521 	 * allowed. We set bit 0 (IBM bit 1) of key0, to prevent instruction
522 	 * fetch.
523 	 */
524 	mtspr(SPRN_IAMR, (1ul << 62));
525 }
526 #endif
527 
528 #ifdef CONFIG_PPC_KUAP
529 void setup_kuap(bool disabled)
530 {
531 	if (disabled || !early_radix_enabled())
532 		return;
533 
534 	if (smp_processor_id() == boot_cpuid) {
535 		pr_info("Activating Kernel Userspace Access Prevention\n");
536 		cur_cpu_spec->mmu_features |= MMU_FTR_RADIX_KUAP;
537 	}
538 
539 	/* Make sure userspace can't change the AMR */
540 	mtspr(SPRN_UAMOR, 0);
541 	mtspr(SPRN_AMR, AMR_KUAP_BLOCKED);
542 	isync();
543 }
544 #endif
545 
546 void __init radix__early_init_mmu(void)
547 {
548 	unsigned long lpcr;
549 
550 #ifdef CONFIG_PPC_64K_PAGES
551 	/* PAGE_SIZE mappings */
552 	mmu_virtual_psize = MMU_PAGE_64K;
553 #else
554 	mmu_virtual_psize = MMU_PAGE_4K;
555 #endif
556 
557 #ifdef CONFIG_SPARSEMEM_VMEMMAP
558 	/* vmemmap mapping */
559 	if (mmu_psize_defs[MMU_PAGE_2M].shift) {
560 		/*
561 		 * map vmemmap using 2M if available
562 		 */
563 		mmu_vmemmap_psize = MMU_PAGE_2M;
564 	} else
565 		mmu_vmemmap_psize = mmu_virtual_psize;
566 #endif
567 	/*
568 	 * initialize page table size
569 	 */
570 	__pte_index_size = RADIX_PTE_INDEX_SIZE;
571 	__pmd_index_size = RADIX_PMD_INDEX_SIZE;
572 	__pud_index_size = RADIX_PUD_INDEX_SIZE;
573 	__pgd_index_size = RADIX_PGD_INDEX_SIZE;
574 	__pud_cache_index = RADIX_PUD_INDEX_SIZE;
575 	__pte_table_size = RADIX_PTE_TABLE_SIZE;
576 	__pmd_table_size = RADIX_PMD_TABLE_SIZE;
577 	__pud_table_size = RADIX_PUD_TABLE_SIZE;
578 	__pgd_table_size = RADIX_PGD_TABLE_SIZE;
579 
580 	__pmd_val_bits = RADIX_PMD_VAL_BITS;
581 	__pud_val_bits = RADIX_PUD_VAL_BITS;
582 	__pgd_val_bits = RADIX_PGD_VAL_BITS;
583 
584 	__kernel_virt_start = RADIX_KERN_VIRT_START;
585 	__vmalloc_start = RADIX_VMALLOC_START;
586 	__vmalloc_end = RADIX_VMALLOC_END;
587 	__kernel_io_start = RADIX_KERN_IO_START;
588 	__kernel_io_end = RADIX_KERN_IO_END;
589 	vmemmap = (struct page *)RADIX_VMEMMAP_START;
590 	ioremap_bot = IOREMAP_BASE;
591 
592 #ifdef CONFIG_PCI
593 	pci_io_base = ISA_IO_BASE;
594 #endif
595 	__pte_frag_nr = RADIX_PTE_FRAG_NR;
596 	__pte_frag_size_shift = RADIX_PTE_FRAG_SIZE_SHIFT;
597 	__pmd_frag_nr = RADIX_PMD_FRAG_NR;
598 	__pmd_frag_size_shift = RADIX_PMD_FRAG_SIZE_SHIFT;
599 
600 	radix_init_pgtable();
601 
602 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
603 		lpcr = mfspr(SPRN_LPCR);
604 		mtspr(SPRN_LPCR, lpcr | LPCR_UPRT | LPCR_HR);
605 		radix_init_partition_table();
606 		radix_init_amor();
607 	} else {
608 		radix_init_pseries();
609 	}
610 
611 	memblock_set_current_limit(MEMBLOCK_ALLOC_ANYWHERE);
612 
613 	/* Switch to the guard PID before turning on MMU */
614 	radix__switch_mmu_context(NULL, &init_mm);
615 	tlbiel_all();
616 }
617 
618 void radix__early_init_mmu_secondary(void)
619 {
620 	unsigned long lpcr;
621 	/*
622 	 * update partition table control register and UPRT
623 	 */
624 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
625 		lpcr = mfspr(SPRN_LPCR);
626 		mtspr(SPRN_LPCR, lpcr | LPCR_UPRT | LPCR_HR);
627 
628 		set_ptcr_when_no_uv(__pa(partition_tb) |
629 				    (PATB_SIZE_SHIFT - 12));
630 
631 		radix_init_amor();
632 	}
633 
634 	radix__switch_mmu_context(NULL, &init_mm);
635 	tlbiel_all();
636 }
637 
638 void radix__mmu_cleanup_all(void)
639 {
640 	unsigned long lpcr;
641 
642 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
643 		lpcr = mfspr(SPRN_LPCR);
644 		mtspr(SPRN_LPCR, lpcr & ~LPCR_UPRT);
645 		set_ptcr_when_no_uv(0);
646 		powernv_set_nmmu_ptcr(0);
647 		radix__flush_tlb_all();
648 	}
649 }
650 
651 void radix__setup_initial_memory_limit(phys_addr_t first_memblock_base,
652 				phys_addr_t first_memblock_size)
653 {
654 	/*
655 	 * We don't currently support the first MEMBLOCK not mapping 0
656 	 * physical on those processors
657 	 */
658 	BUG_ON(first_memblock_base != 0);
659 
660 	/*
661 	 * Radix mode is not limited by RMA / VRMA addressing.
662 	 */
663 	ppc64_rma_size = ULONG_MAX;
664 }
665 
666 #ifdef CONFIG_MEMORY_HOTPLUG
667 static void free_pte_table(pte_t *pte_start, pmd_t *pmd)
668 {
669 	pte_t *pte;
670 	int i;
671 
672 	for (i = 0; i < PTRS_PER_PTE; i++) {
673 		pte = pte_start + i;
674 		if (!pte_none(*pte))
675 			return;
676 	}
677 
678 	pte_free_kernel(&init_mm, pte_start);
679 	pmd_clear(pmd);
680 }
681 
682 static void free_pmd_table(pmd_t *pmd_start, pud_t *pud)
683 {
684 	pmd_t *pmd;
685 	int i;
686 
687 	for (i = 0; i < PTRS_PER_PMD; i++) {
688 		pmd = pmd_start + i;
689 		if (!pmd_none(*pmd))
690 			return;
691 	}
692 
693 	pmd_free(&init_mm, pmd_start);
694 	pud_clear(pud);
695 }
696 
697 struct change_mapping_params {
698 	pte_t *pte;
699 	unsigned long start;
700 	unsigned long end;
701 	unsigned long aligned_start;
702 	unsigned long aligned_end;
703 };
704 
705 static int __meminit stop_machine_change_mapping(void *data)
706 {
707 	struct change_mapping_params *params =
708 			(struct change_mapping_params *)data;
709 
710 	if (!data)
711 		return -1;
712 
713 	spin_unlock(&init_mm.page_table_lock);
714 	pte_clear(&init_mm, params->aligned_start, params->pte);
715 	create_physical_mapping(__pa(params->aligned_start), __pa(params->start), -1);
716 	create_physical_mapping(__pa(params->end), __pa(params->aligned_end), -1);
717 	spin_lock(&init_mm.page_table_lock);
718 	return 0;
719 }
720 
721 static void remove_pte_table(pte_t *pte_start, unsigned long addr,
722 			     unsigned long end)
723 {
724 	unsigned long next;
725 	pte_t *pte;
726 
727 	pte = pte_start + pte_index(addr);
728 	for (; addr < end; addr = next, pte++) {
729 		next = (addr + PAGE_SIZE) & PAGE_MASK;
730 		if (next > end)
731 			next = end;
732 
733 		if (!pte_present(*pte))
734 			continue;
735 
736 		if (!PAGE_ALIGNED(addr) || !PAGE_ALIGNED(next)) {
737 			/*
738 			 * The vmemmap_free() and remove_section_mapping()
739 			 * codepaths call us with aligned addresses.
740 			 */
741 			WARN_ONCE(1, "%s: unaligned range\n", __func__);
742 			continue;
743 		}
744 
745 		pte_clear(&init_mm, addr, pte);
746 	}
747 }
748 
749 /*
750  * clear the pte and potentially split the mapping helper
751  */
752 static void __meminit split_kernel_mapping(unsigned long addr, unsigned long end,
753 				unsigned long size, pte_t *pte)
754 {
755 	unsigned long mask = ~(size - 1);
756 	unsigned long aligned_start = addr & mask;
757 	unsigned long aligned_end = addr + size;
758 	struct change_mapping_params params;
759 	bool split_region = false;
760 
761 	if ((end - addr) < size) {
762 		/*
763 		 * We're going to clear the PTE, but not flushed
764 		 * the mapping, time to remap and flush. The
765 		 * effects if visible outside the processor or
766 		 * if we are running in code close to the
767 		 * mapping we cleared, we are in trouble.
768 		 */
769 		if (overlaps_kernel_text(aligned_start, addr) ||
770 			overlaps_kernel_text(end, aligned_end)) {
771 			/*
772 			 * Hack, just return, don't pte_clear
773 			 */
774 			WARN_ONCE(1, "Linear mapping %lx->%lx overlaps kernel "
775 				  "text, not splitting\n", addr, end);
776 			return;
777 		}
778 		split_region = true;
779 	}
780 
781 	if (split_region) {
782 		params.pte = pte;
783 		params.start = addr;
784 		params.end = end;
785 		params.aligned_start = addr & ~(size - 1);
786 		params.aligned_end = min_t(unsigned long, aligned_end,
787 				(unsigned long)__va(memblock_end_of_DRAM()));
788 		stop_machine(stop_machine_change_mapping, &params, NULL);
789 		return;
790 	}
791 
792 	pte_clear(&init_mm, addr, pte);
793 }
794 
795 static void remove_pmd_table(pmd_t *pmd_start, unsigned long addr,
796 			     unsigned long end)
797 {
798 	unsigned long next;
799 	pte_t *pte_base;
800 	pmd_t *pmd;
801 
802 	pmd = pmd_start + pmd_index(addr);
803 	for (; addr < end; addr = next, pmd++) {
804 		next = pmd_addr_end(addr, end);
805 
806 		if (!pmd_present(*pmd))
807 			continue;
808 
809 		if (pmd_is_leaf(*pmd)) {
810 			split_kernel_mapping(addr, end, PMD_SIZE, (pte_t *)pmd);
811 			continue;
812 		}
813 
814 		pte_base = (pte_t *)pmd_page_vaddr(*pmd);
815 		remove_pte_table(pte_base, addr, next);
816 		free_pte_table(pte_base, pmd);
817 	}
818 }
819 
820 static void remove_pud_table(pud_t *pud_start, unsigned long addr,
821 			     unsigned long end)
822 {
823 	unsigned long next;
824 	pmd_t *pmd_base;
825 	pud_t *pud;
826 
827 	pud = pud_start + pud_index(addr);
828 	for (; addr < end; addr = next, pud++) {
829 		next = pud_addr_end(addr, end);
830 
831 		if (!pud_present(*pud))
832 			continue;
833 
834 		if (pud_is_leaf(*pud)) {
835 			split_kernel_mapping(addr, end, PUD_SIZE, (pte_t *)pud);
836 			continue;
837 		}
838 
839 		pmd_base = (pmd_t *)pud_page_vaddr(*pud);
840 		remove_pmd_table(pmd_base, addr, next);
841 		free_pmd_table(pmd_base, pud);
842 	}
843 }
844 
845 static void __meminit remove_pagetable(unsigned long start, unsigned long end)
846 {
847 	unsigned long addr, next;
848 	pud_t *pud_base;
849 	pgd_t *pgd;
850 
851 	spin_lock(&init_mm.page_table_lock);
852 
853 	for (addr = start; addr < end; addr = next) {
854 		next = pgd_addr_end(addr, end);
855 
856 		pgd = pgd_offset_k(addr);
857 		if (!pgd_present(*pgd))
858 			continue;
859 
860 		if (pgd_is_leaf(*pgd)) {
861 			split_kernel_mapping(addr, end, PGDIR_SIZE, (pte_t *)pgd);
862 			continue;
863 		}
864 
865 		pud_base = (pud_t *)pgd_page_vaddr(*pgd);
866 		remove_pud_table(pud_base, addr, next);
867 	}
868 
869 	spin_unlock(&init_mm.page_table_lock);
870 	radix__flush_tlb_kernel_range(start, end);
871 }
872 
873 int __meminit radix__create_section_mapping(unsigned long start, unsigned long end, int nid)
874 {
875 	if (end >= RADIX_VMALLOC_START) {
876 		pr_warn("Outside the supported range\n");
877 		return -1;
878 	}
879 
880 	return create_physical_mapping(__pa(start), __pa(end), nid);
881 }
882 
883 int __meminit radix__remove_section_mapping(unsigned long start, unsigned long end)
884 {
885 	remove_pagetable(start, end);
886 	return 0;
887 }
888 #endif /* CONFIG_MEMORY_HOTPLUG */
889 
890 #ifdef CONFIG_SPARSEMEM_VMEMMAP
891 static int __map_kernel_page_nid(unsigned long ea, unsigned long pa,
892 				 pgprot_t flags, unsigned int map_page_size,
893 				 int nid)
894 {
895 	return __map_kernel_page(ea, pa, flags, map_page_size, nid, 0, 0);
896 }
897 
898 int __meminit radix__vmemmap_create_mapping(unsigned long start,
899 				      unsigned long page_size,
900 				      unsigned long phys)
901 {
902 	/* Create a PTE encoding */
903 	unsigned long flags = _PAGE_PRESENT | _PAGE_ACCESSED | _PAGE_KERNEL_RW;
904 	int nid = early_pfn_to_nid(phys >> PAGE_SHIFT);
905 	int ret;
906 
907 	if ((start + page_size) >= RADIX_VMEMMAP_END) {
908 		pr_warn("Outside the supported range\n");
909 		return -1;
910 	}
911 
912 	ret = __map_kernel_page_nid(start, phys, __pgprot(flags), page_size, nid);
913 	BUG_ON(ret);
914 
915 	return 0;
916 }
917 
918 #ifdef CONFIG_MEMORY_HOTPLUG
919 void __meminit radix__vmemmap_remove_mapping(unsigned long start, unsigned long page_size)
920 {
921 	remove_pagetable(start, start + page_size);
922 }
923 #endif
924 #endif
925 
926 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
927 
928 unsigned long radix__pmd_hugepage_update(struct mm_struct *mm, unsigned long addr,
929 				  pmd_t *pmdp, unsigned long clr,
930 				  unsigned long set)
931 {
932 	unsigned long old;
933 
934 #ifdef CONFIG_DEBUG_VM
935 	WARN_ON(!radix__pmd_trans_huge(*pmdp) && !pmd_devmap(*pmdp));
936 	assert_spin_locked(pmd_lockptr(mm, pmdp));
937 #endif
938 
939 	old = radix__pte_update(mm, addr, (pte_t *)pmdp, clr, set, 1);
940 	trace_hugepage_update(addr, old, clr, set);
941 
942 	return old;
943 }
944 
945 pmd_t radix__pmdp_collapse_flush(struct vm_area_struct *vma, unsigned long address,
946 			pmd_t *pmdp)
947 
948 {
949 	pmd_t pmd;
950 
951 	VM_BUG_ON(address & ~HPAGE_PMD_MASK);
952 	VM_BUG_ON(radix__pmd_trans_huge(*pmdp));
953 	VM_BUG_ON(pmd_devmap(*pmdp));
954 	/*
955 	 * khugepaged calls this for normal pmd
956 	 */
957 	pmd = *pmdp;
958 	pmd_clear(pmdp);
959 
960 	/*FIXME!!  Verify whether we need this kick below */
961 	serialize_against_pte_lookup(vma->vm_mm);
962 
963 	radix__flush_tlb_collapsed_pmd(vma->vm_mm, address);
964 
965 	return pmd;
966 }
967 
968 /*
969  * For us pgtable_t is pte_t *. Inorder to save the deposisted
970  * page table, we consider the allocated page table as a list
971  * head. On withdraw we need to make sure we zero out the used
972  * list_head memory area.
973  */
974 void radix__pgtable_trans_huge_deposit(struct mm_struct *mm, pmd_t *pmdp,
975 				 pgtable_t pgtable)
976 {
977 	struct list_head *lh = (struct list_head *) pgtable;
978 
979 	assert_spin_locked(pmd_lockptr(mm, pmdp));
980 
981 	/* FIFO */
982 	if (!pmd_huge_pte(mm, pmdp))
983 		INIT_LIST_HEAD(lh);
984 	else
985 		list_add(lh, (struct list_head *) pmd_huge_pte(mm, pmdp));
986 	pmd_huge_pte(mm, pmdp) = pgtable;
987 }
988 
989 pgtable_t radix__pgtable_trans_huge_withdraw(struct mm_struct *mm, pmd_t *pmdp)
990 {
991 	pte_t *ptep;
992 	pgtable_t pgtable;
993 	struct list_head *lh;
994 
995 	assert_spin_locked(pmd_lockptr(mm, pmdp));
996 
997 	/* FIFO */
998 	pgtable = pmd_huge_pte(mm, pmdp);
999 	lh = (struct list_head *) pgtable;
1000 	if (list_empty(lh))
1001 		pmd_huge_pte(mm, pmdp) = NULL;
1002 	else {
1003 		pmd_huge_pte(mm, pmdp) = (pgtable_t) lh->next;
1004 		list_del(lh);
1005 	}
1006 	ptep = (pte_t *) pgtable;
1007 	*ptep = __pte(0);
1008 	ptep++;
1009 	*ptep = __pte(0);
1010 	return pgtable;
1011 }
1012 
1013 pmd_t radix__pmdp_huge_get_and_clear(struct mm_struct *mm,
1014 				     unsigned long addr, pmd_t *pmdp)
1015 {
1016 	pmd_t old_pmd;
1017 	unsigned long old;
1018 
1019 	old = radix__pmd_hugepage_update(mm, addr, pmdp, ~0UL, 0);
1020 	old_pmd = __pmd(old);
1021 	/*
1022 	 * Serialize against find_current_mm_pte which does lock-less
1023 	 * lookup in page tables with local interrupts disabled. For huge pages
1024 	 * it casts pmd_t to pte_t. Since format of pte_t is different from
1025 	 * pmd_t we want to prevent transit from pmd pointing to page table
1026 	 * to pmd pointing to huge page (and back) while interrupts are disabled.
1027 	 * We clear pmd to possibly replace it with page table pointer in
1028 	 * different code paths. So make sure we wait for the parallel
1029 	 * find_current_mm_pte to finish.
1030 	 */
1031 	serialize_against_pte_lookup(mm);
1032 	return old_pmd;
1033 }
1034 
1035 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
1036 
1037 void radix__ptep_set_access_flags(struct vm_area_struct *vma, pte_t *ptep,
1038 				  pte_t entry, unsigned long address, int psize)
1039 {
1040 	struct mm_struct *mm = vma->vm_mm;
1041 	unsigned long set = pte_val(entry) & (_PAGE_DIRTY | _PAGE_ACCESSED |
1042 					      _PAGE_RW | _PAGE_EXEC);
1043 
1044 	unsigned long change = pte_val(entry) ^ pte_val(*ptep);
1045 	/*
1046 	 * To avoid NMMU hang while relaxing access, we need mark
1047 	 * the pte invalid in between.
1048 	 */
1049 	if ((change & _PAGE_RW) && atomic_read(&mm->context.copros) > 0) {
1050 		unsigned long old_pte, new_pte;
1051 
1052 		old_pte = __radix_pte_update(ptep, _PAGE_PRESENT, _PAGE_INVALID);
1053 		/*
1054 		 * new value of pte
1055 		 */
1056 		new_pte = old_pte | set;
1057 		radix__flush_tlb_page_psize(mm, address, psize);
1058 		__radix_pte_update(ptep, _PAGE_INVALID, new_pte);
1059 	} else {
1060 		__radix_pte_update(ptep, 0, set);
1061 		/*
1062 		 * Book3S does not require a TLB flush when relaxing access
1063 		 * restrictions when the address space is not attached to a
1064 		 * NMMU, because the core MMU will reload the pte after taking
1065 		 * an access fault, which is defined by the architectue.
1066 		 */
1067 	}
1068 	/* See ptesync comment in radix__set_pte_at */
1069 }
1070 
1071 void radix__ptep_modify_prot_commit(struct vm_area_struct *vma,
1072 				    unsigned long addr, pte_t *ptep,
1073 				    pte_t old_pte, pte_t pte)
1074 {
1075 	struct mm_struct *mm = vma->vm_mm;
1076 
1077 	/*
1078 	 * To avoid NMMU hang while relaxing access we need to flush the tlb before
1079 	 * we set the new value. We need to do this only for radix, because hash
1080 	 * translation does flush when updating the linux pte.
1081 	 */
1082 	if (is_pte_rw_upgrade(pte_val(old_pte), pte_val(pte)) &&
1083 	    (atomic_read(&mm->context.copros) > 0))
1084 		radix__flush_tlb_page(vma, addr);
1085 
1086 	set_pte_at(mm, addr, ptep, pte);
1087 }
1088 
1089 int __init arch_ioremap_pud_supported(void)
1090 {
1091 	/* HPT does not cope with large pages in the vmalloc area */
1092 	return radix_enabled();
1093 }
1094 
1095 int __init arch_ioremap_pmd_supported(void)
1096 {
1097 	return radix_enabled();
1098 }
1099 
1100 int p4d_free_pud_page(p4d_t *p4d, unsigned long addr)
1101 {
1102 	return 0;
1103 }
1104 
1105 int pud_set_huge(pud_t *pud, phys_addr_t addr, pgprot_t prot)
1106 {
1107 	pte_t *ptep = (pte_t *)pud;
1108 	pte_t new_pud = pfn_pte(__phys_to_pfn(addr), prot);
1109 
1110 	if (!radix_enabled())
1111 		return 0;
1112 
1113 	set_pte_at(&init_mm, 0 /* radix unused */, ptep, new_pud);
1114 
1115 	return 1;
1116 }
1117 
1118 int pud_clear_huge(pud_t *pud)
1119 {
1120 	if (pud_huge(*pud)) {
1121 		pud_clear(pud);
1122 		return 1;
1123 	}
1124 
1125 	return 0;
1126 }
1127 
1128 int pud_free_pmd_page(pud_t *pud, unsigned long addr)
1129 {
1130 	pmd_t *pmd;
1131 	int i;
1132 
1133 	pmd = (pmd_t *)pud_page_vaddr(*pud);
1134 	pud_clear(pud);
1135 
1136 	flush_tlb_kernel_range(addr, addr + PUD_SIZE);
1137 
1138 	for (i = 0; i < PTRS_PER_PMD; i++) {
1139 		if (!pmd_none(pmd[i])) {
1140 			pte_t *pte;
1141 			pte = (pte_t *)pmd_page_vaddr(pmd[i]);
1142 
1143 			pte_free_kernel(&init_mm, pte);
1144 		}
1145 	}
1146 
1147 	pmd_free(&init_mm, pmd);
1148 
1149 	return 1;
1150 }
1151 
1152 int pmd_set_huge(pmd_t *pmd, phys_addr_t addr, pgprot_t prot)
1153 {
1154 	pte_t *ptep = (pte_t *)pmd;
1155 	pte_t new_pmd = pfn_pte(__phys_to_pfn(addr), prot);
1156 
1157 	if (!radix_enabled())
1158 		return 0;
1159 
1160 	set_pte_at(&init_mm, 0 /* radix unused */, ptep, new_pmd);
1161 
1162 	return 1;
1163 }
1164 
1165 int pmd_clear_huge(pmd_t *pmd)
1166 {
1167 	if (pmd_huge(*pmd)) {
1168 		pmd_clear(pmd);
1169 		return 1;
1170 	}
1171 
1172 	return 0;
1173 }
1174 
1175 int pmd_free_pte_page(pmd_t *pmd, unsigned long addr)
1176 {
1177 	pte_t *pte;
1178 
1179 	pte = (pte_t *)pmd_page_vaddr(*pmd);
1180 	pmd_clear(pmd);
1181 
1182 	flush_tlb_kernel_range(addr, addr + PMD_SIZE);
1183 
1184 	pte_free_kernel(&init_mm, pte);
1185 
1186 	return 1;
1187 }
1188 
1189 int __init arch_ioremap_p4d_supported(void)
1190 {
1191 	return 0;
1192 }
1193