1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright 2005, Paul Mackerras, IBM Corporation.
4  * Copyright 2009, Benjamin Herrenschmidt, IBM Corporation.
5  * Copyright 2015-2016, Aneesh Kumar K.V, IBM Corporation.
6  */
7 
8 #include <linux/sched.h>
9 #include <linux/memblock.h>
10 #include <asm/pgalloc.h>
11 #include <asm/tlb.h>
12 #include <asm/dma.h>
13 #include <asm/code-patching.h>
14 
15 #include <mm/mmu_decl.h>
16 
17 #ifdef CONFIG_SPARSEMEM_VMEMMAP
18 /*
19  * On Book3E CPUs, the vmemmap is currently mapped in the top half of
20  * the vmalloc space using normal page tables, though the size of
21  * pages encoded in the PTEs can be different
22  */
vmemmap_create_mapping(unsigned long start,unsigned long page_size,unsigned long phys)23 int __meminit vmemmap_create_mapping(unsigned long start,
24 				     unsigned long page_size,
25 				     unsigned long phys)
26 {
27 	/* Create a PTE encoding without page size */
28 	unsigned long i, flags = _PAGE_PRESENT | _PAGE_ACCESSED |
29 		_PAGE_KERNEL_RW;
30 
31 	/* PTEs only contain page size encodings up to 32M */
32 	BUG_ON(mmu_psize_defs[mmu_vmemmap_psize].shift - 10 > 0xf);
33 
34 	/* Encode the size in the PTE */
35 	flags |= (mmu_psize_defs[mmu_vmemmap_psize].shift - 10) << 8;
36 
37 	/* For each PTE for that area, map things. Note that we don't
38 	 * increment phys because all PTEs are of the large size and
39 	 * thus must have the low bits clear
40 	 */
41 	for (i = 0; i < page_size; i += PAGE_SIZE)
42 		BUG_ON(map_kernel_page(start + i, phys, __pgprot(flags)));
43 
44 	return 0;
45 }
46 
47 #ifdef CONFIG_MEMORY_HOTPLUG
vmemmap_remove_mapping(unsigned long start,unsigned long page_size)48 void vmemmap_remove_mapping(unsigned long start,
49 			    unsigned long page_size)
50 {
51 }
52 #endif
53 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
54 
early_alloc_pgtable(unsigned long size)55 static void __init *early_alloc_pgtable(unsigned long size)
56 {
57 	void *ptr;
58 
59 	ptr = memblock_alloc_try_nid(size, size, MEMBLOCK_LOW_LIMIT,
60 				     __pa(MAX_DMA_ADDRESS), NUMA_NO_NODE);
61 
62 	if (!ptr)
63 		panic("%s: Failed to allocate %lu bytes align=0x%lx max_addr=%lx\n",
64 		      __func__, size, size, __pa(MAX_DMA_ADDRESS));
65 
66 	return ptr;
67 }
68 
69 /*
70  * map_kernel_page currently only called by __ioremap
71  * map_kernel_page adds an entry to the ioremap page table
72  * and adds an entry to the HPT, possibly bolting it
73  */
map_kernel_page(unsigned long ea,phys_addr_t pa,pgprot_t prot)74 int __ref map_kernel_page(unsigned long ea, phys_addr_t pa, pgprot_t prot)
75 {
76 	pgd_t *pgdp;
77 	p4d_t *p4dp;
78 	pud_t *pudp;
79 	pmd_t *pmdp;
80 	pte_t *ptep;
81 
82 	BUILD_BUG_ON(TASK_SIZE_USER64 > PGTABLE_RANGE);
83 	if (slab_is_available()) {
84 		pgdp = pgd_offset_k(ea);
85 		p4dp = p4d_offset(pgdp, ea);
86 		pudp = pud_alloc(&init_mm, p4dp, ea);
87 		if (!pudp)
88 			return -ENOMEM;
89 		pmdp = pmd_alloc(&init_mm, pudp, ea);
90 		if (!pmdp)
91 			return -ENOMEM;
92 		ptep = pte_alloc_kernel(pmdp, ea);
93 		if (!ptep)
94 			return -ENOMEM;
95 	} else {
96 		pgdp = pgd_offset_k(ea);
97 		p4dp = p4d_offset(pgdp, ea);
98 		if (p4d_none(*p4dp)) {
99 			pudp = early_alloc_pgtable(PUD_TABLE_SIZE);
100 			p4d_populate(&init_mm, p4dp, pudp);
101 		}
102 		pudp = pud_offset(p4dp, ea);
103 		if (pud_none(*pudp)) {
104 			pmdp = early_alloc_pgtable(PMD_TABLE_SIZE);
105 			pud_populate(&init_mm, pudp, pmdp);
106 		}
107 		pmdp = pmd_offset(pudp, ea);
108 		if (!pmd_present(*pmdp)) {
109 			ptep = early_alloc_pgtable(PTE_TABLE_SIZE);
110 			pmd_populate_kernel(&init_mm, pmdp, ptep);
111 		}
112 		ptep = pte_offset_kernel(pmdp, ea);
113 	}
114 	set_pte_at(&init_mm, ea, ptep, pfn_pte(pa >> PAGE_SHIFT, prot));
115 
116 	smp_wmb();
117 	return 0;
118 }
119 
__patch_exception(int exc,unsigned long addr)120 void __patch_exception(int exc, unsigned long addr)
121 {
122 	unsigned int *ibase = &interrupt_base_book3e;
123 
124 	/*
125 	 * Our exceptions vectors start with a NOP and -then- a branch
126 	 * to deal with single stepping from userspace which stops on
127 	 * the second instruction. Thus we need to patch the second
128 	 * instruction of the exception, not the first one.
129 	 */
130 
131 	patch_branch(ibase + (exc / 4) + 1, addr, 0);
132 }
133