xref: /openbmc/linux/arch/x86/mm/cpu_entry_area.c (revision d9fd5a71)
1 // SPDX-License-Identifier: GPL-2.0
2 
3 #include <linux/spinlock.h>
4 #include <linux/percpu.h>
5 #include <linux/kallsyms.h>
6 #include <linux/kcore.h>
7 #include <linux/pgtable.h>
8 
9 #include <asm/cpu_entry_area.h>
10 #include <asm/fixmap.h>
11 #include <asm/desc.h>
12 
13 static DEFINE_PER_CPU_PAGE_ALIGNED(struct entry_stack_page, entry_stack_storage);
14 
15 #ifdef CONFIG_X86_64
16 static DEFINE_PER_CPU_PAGE_ALIGNED(struct exception_stacks, exception_stacks);
17 DEFINE_PER_CPU(struct cea_exception_stacks*, cea_exception_stacks);
18 #endif
19 
20 #ifdef CONFIG_X86_32
21 DECLARE_PER_CPU_PAGE_ALIGNED(struct doublefault_stack, doublefault_stack);
22 #endif
23 
24 /* Is called from entry code, so must be noinstr */
25 noinstr struct cpu_entry_area *get_cpu_entry_area(int cpu)
26 {
27 	unsigned long va = CPU_ENTRY_AREA_PER_CPU + cpu * CPU_ENTRY_AREA_SIZE;
28 	BUILD_BUG_ON(sizeof(struct cpu_entry_area) % PAGE_SIZE != 0);
29 
30 	return (struct cpu_entry_area *) va;
31 }
32 EXPORT_SYMBOL(get_cpu_entry_area);
33 
34 void cea_set_pte(void *cea_vaddr, phys_addr_t pa, pgprot_t flags)
35 {
36 	unsigned long va = (unsigned long) cea_vaddr;
37 	pte_t pte = pfn_pte(pa >> PAGE_SHIFT, flags);
38 
39 	/*
40 	 * The cpu_entry_area is shared between the user and kernel
41 	 * page tables.  All of its ptes can safely be global.
42 	 * _PAGE_GLOBAL gets reused to help indicate PROT_NONE for
43 	 * non-present PTEs, so be careful not to set it in that
44 	 * case to avoid confusion.
45 	 */
46 	if (boot_cpu_has(X86_FEATURE_PGE) &&
47 	    (pgprot_val(flags) & _PAGE_PRESENT))
48 		pte = pte_set_flags(pte, _PAGE_GLOBAL);
49 
50 	set_pte_vaddr(va, pte);
51 }
52 
53 static void __init
54 cea_map_percpu_pages(void *cea_vaddr, void *ptr, int pages, pgprot_t prot)
55 {
56 	for ( ; pages; pages--, cea_vaddr+= PAGE_SIZE, ptr += PAGE_SIZE)
57 		cea_set_pte(cea_vaddr, per_cpu_ptr_to_phys(ptr), prot);
58 }
59 
60 static void __init percpu_setup_debug_store(unsigned int cpu)
61 {
62 #ifdef CONFIG_CPU_SUP_INTEL
63 	unsigned int npages;
64 	void *cea;
65 
66 	if (boot_cpu_data.x86_vendor != X86_VENDOR_INTEL)
67 		return;
68 
69 	cea = &get_cpu_entry_area(cpu)->cpu_debug_store;
70 	npages = sizeof(struct debug_store) / PAGE_SIZE;
71 	BUILD_BUG_ON(sizeof(struct debug_store) % PAGE_SIZE != 0);
72 	cea_map_percpu_pages(cea, &per_cpu(cpu_debug_store, cpu), npages,
73 			     PAGE_KERNEL);
74 
75 	cea = &get_cpu_entry_area(cpu)->cpu_debug_buffers;
76 	/*
77 	 * Force the population of PMDs for not yet allocated per cpu
78 	 * memory like debug store buffers.
79 	 */
80 	npages = sizeof(struct debug_store_buffers) / PAGE_SIZE;
81 	for (; npages; npages--, cea += PAGE_SIZE)
82 		cea_set_pte(cea, 0, PAGE_NONE);
83 #endif
84 }
85 
86 #ifdef CONFIG_X86_64
87 
88 #define cea_map_stack(name) do {					\
89 	npages = sizeof(estacks->name## _stack) / PAGE_SIZE;		\
90 	cea_map_percpu_pages(cea->estacks.name## _stack,		\
91 			estacks->name## _stack, npages, PAGE_KERNEL);	\
92 	} while (0)
93 
94 static void __init percpu_setup_exception_stacks(unsigned int cpu)
95 {
96 	struct exception_stacks *estacks = per_cpu_ptr(&exception_stacks, cpu);
97 	struct cpu_entry_area *cea = get_cpu_entry_area(cpu);
98 	unsigned int npages;
99 
100 	BUILD_BUG_ON(sizeof(exception_stacks) % PAGE_SIZE != 0);
101 
102 	per_cpu(cea_exception_stacks, cpu) = &cea->estacks;
103 
104 	/*
105 	 * The exceptions stack mappings in the per cpu area are protected
106 	 * by guard pages so each stack must be mapped separately. DB2 is
107 	 * not mapped; it just exists to catch triple nesting of #DB.
108 	 */
109 	cea_map_stack(DF);
110 	cea_map_stack(NMI);
111 	cea_map_stack(DB);
112 	cea_map_stack(MCE);
113 }
114 #else
115 static inline void percpu_setup_exception_stacks(unsigned int cpu)
116 {
117 	struct cpu_entry_area *cea = get_cpu_entry_area(cpu);
118 
119 	cea_map_percpu_pages(&cea->doublefault_stack,
120 			     &per_cpu(doublefault_stack, cpu), 1, PAGE_KERNEL);
121 }
122 #endif
123 
124 /* Setup the fixmap mappings only once per-processor */
125 static void __init setup_cpu_entry_area(unsigned int cpu)
126 {
127 	struct cpu_entry_area *cea = get_cpu_entry_area(cpu);
128 #ifdef CONFIG_X86_64
129 	/* On 64-bit systems, we use a read-only fixmap GDT and TSS. */
130 	pgprot_t gdt_prot = PAGE_KERNEL_RO;
131 	pgprot_t tss_prot = PAGE_KERNEL_RO;
132 #else
133 	/*
134 	 * On native 32-bit systems, the GDT cannot be read-only because
135 	 * our double fault handler uses a task gate, and entering through
136 	 * a task gate needs to change an available TSS to busy.  If the
137 	 * GDT is read-only, that will triple fault.  The TSS cannot be
138 	 * read-only because the CPU writes to it on task switches.
139 	 *
140 	 * On Xen PV, the GDT must be read-only because the hypervisor
141 	 * requires it.
142 	 */
143 	pgprot_t gdt_prot = boot_cpu_has(X86_FEATURE_XENPV) ?
144 		PAGE_KERNEL_RO : PAGE_KERNEL;
145 	pgprot_t tss_prot = PAGE_KERNEL;
146 #endif
147 
148 	cea_set_pte(&cea->gdt, get_cpu_gdt_paddr(cpu), gdt_prot);
149 
150 	cea_map_percpu_pages(&cea->entry_stack_page,
151 			     per_cpu_ptr(&entry_stack_storage, cpu), 1,
152 			     PAGE_KERNEL);
153 
154 	/*
155 	 * The Intel SDM says (Volume 3, 7.2.1):
156 	 *
157 	 *  Avoid placing a page boundary in the part of the TSS that the
158 	 *  processor reads during a task switch (the first 104 bytes). The
159 	 *  processor may not correctly perform address translations if a
160 	 *  boundary occurs in this area. During a task switch, the processor
161 	 *  reads and writes into the first 104 bytes of each TSS (using
162 	 *  contiguous physical addresses beginning with the physical address
163 	 *  of the first byte of the TSS). So, after TSS access begins, if
164 	 *  part of the 104 bytes is not physically contiguous, the processor
165 	 *  will access incorrect information without generating a page-fault
166 	 *  exception.
167 	 *
168 	 * There are also a lot of errata involving the TSS spanning a page
169 	 * boundary.  Assert that we're not doing that.
170 	 */
171 	BUILD_BUG_ON((offsetof(struct tss_struct, x86_tss) ^
172 		      offsetofend(struct tss_struct, x86_tss)) & PAGE_MASK);
173 	BUILD_BUG_ON(sizeof(struct tss_struct) % PAGE_SIZE != 0);
174 	/*
175 	 * VMX changes the host TR limit to 0x67 after a VM exit. This is
176 	 * okay, since 0x67 covers the size of struct x86_hw_tss. Make sure
177 	 * that this is correct.
178 	 */
179 	BUILD_BUG_ON(offsetof(struct tss_struct, x86_tss) != 0);
180 	BUILD_BUG_ON(sizeof(struct x86_hw_tss) != 0x68);
181 
182 	cea_map_percpu_pages(&cea->tss, &per_cpu(cpu_tss_rw, cpu),
183 			     sizeof(struct tss_struct) / PAGE_SIZE, tss_prot);
184 
185 #ifdef CONFIG_X86_32
186 	per_cpu(cpu_entry_area, cpu) = cea;
187 #endif
188 
189 	percpu_setup_exception_stacks(cpu);
190 
191 	percpu_setup_debug_store(cpu);
192 }
193 
194 static __init void setup_cpu_entry_area_ptes(void)
195 {
196 #ifdef CONFIG_X86_32
197 	unsigned long start, end;
198 
199 	/* The +1 is for the readonly IDT: */
200 	BUILD_BUG_ON((CPU_ENTRY_AREA_PAGES+1)*PAGE_SIZE != CPU_ENTRY_AREA_MAP_SIZE);
201 	BUILD_BUG_ON(CPU_ENTRY_AREA_TOTAL_SIZE != CPU_ENTRY_AREA_MAP_SIZE);
202 	BUG_ON(CPU_ENTRY_AREA_BASE & ~PMD_MASK);
203 
204 	start = CPU_ENTRY_AREA_BASE;
205 	end = start + CPU_ENTRY_AREA_MAP_SIZE;
206 
207 	/* Careful here: start + PMD_SIZE might wrap around */
208 	for (; start < end && start >= CPU_ENTRY_AREA_BASE; start += PMD_SIZE)
209 		populate_extra_pte(start);
210 #endif
211 }
212 
213 void __init setup_cpu_entry_areas(void)
214 {
215 	unsigned int cpu;
216 
217 	setup_cpu_entry_area_ptes();
218 
219 	for_each_possible_cpu(cpu)
220 		setup_cpu_entry_area(cpu);
221 
222 	/*
223 	 * This is the last essential update to swapper_pgdir which needs
224 	 * to be synchronized to initial_page_table on 32bit.
225 	 */
226 	sync_initial_page_table();
227 }
228