xref: /linux/arch/x86/boot/compressed/pgtable_64.c (revision 103bf75f)
1 // SPDX-License-Identifier: GPL-2.0
2 #include "misc.h"
3 #include <asm/bootparam.h>
4 #include <asm/e820/types.h>
5 #include <asm/processor.h>
6 #include "pgtable.h"
7 #include "../string.h"
8 #include "efi.h"
9 
10 #define BIOS_START_MIN		0x20000U	/* 128K, less than this is insane */
11 #define BIOS_START_MAX		0x9f000U	/* 640K, absolute maximum */
12 
13 #ifdef CONFIG_X86_5LEVEL
14 /* __pgtable_l5_enabled needs to be in .data to avoid being cleared along with .bss */
15 unsigned int __section(".data") __pgtable_l5_enabled;
16 unsigned int __section(".data") pgdir_shift = 39;
17 unsigned int __section(".data") ptrs_per_p4d = 1;
18 #endif
19 
20 /* Buffer to preserve trampoline memory */
21 static char trampoline_save[TRAMPOLINE_32BIT_SIZE];
22 
23 /*
24  * Trampoline address will be printed by extract_kernel() for debugging
25  * purposes.
26  *
27  * Avoid putting the pointer into .bss as it will be cleared between
28  * configure_5level_paging() and extract_kernel().
29  */
30 unsigned long *trampoline_32bit __section(".data");
31 
32 int cmdline_find_option_bool(const char *option);
33 
find_trampoline_placement(void)34 static unsigned long find_trampoline_placement(void)
35 {
36 	unsigned long bios_start = 0, ebda_start = 0;
37 	struct boot_e820_entry *entry;
38 	char *signature;
39 	int i;
40 
41 	/*
42 	 * Find a suitable spot for the trampoline.
43 	 * This code is based on reserve_bios_regions().
44 	 */
45 
46 	/*
47 	 * EFI systems may not provide legacy ROM. The memory may not be mapped
48 	 * at all.
49 	 *
50 	 * Only look for values in the legacy ROM for non-EFI system.
51 	 */
52 	signature = (char *)&boot_params_ptr->efi_info.efi_loader_signature;
53 	if (strncmp(signature, EFI32_LOADER_SIGNATURE, 4) &&
54 	    strncmp(signature, EFI64_LOADER_SIGNATURE, 4)) {
55 		ebda_start = *(unsigned short *)0x40e << 4;
56 		bios_start = *(unsigned short *)0x413 << 10;
57 	}
58 
59 	if (bios_start < BIOS_START_MIN || bios_start > BIOS_START_MAX)
60 		bios_start = BIOS_START_MAX;
61 
62 	if (ebda_start > BIOS_START_MIN && ebda_start < bios_start)
63 		bios_start = ebda_start;
64 
65 	bios_start = round_down(bios_start, PAGE_SIZE);
66 
67 	/* Find the first usable memory region under bios_start. */
68 	for (i = boot_params_ptr->e820_entries - 1; i >= 0; i--) {
69 		unsigned long new = bios_start;
70 
71 		entry = &boot_params_ptr->e820_table[i];
72 
73 		/* Skip all entries above bios_start. */
74 		if (bios_start <= entry->addr)
75 			continue;
76 
77 		/* Skip non-RAM entries. */
78 		if (entry->type != E820_TYPE_RAM)
79 			continue;
80 
81 		/* Adjust bios_start to the end of the entry if needed. */
82 		if (bios_start > entry->addr + entry->size)
83 			new = entry->addr + entry->size;
84 
85 		/* Keep bios_start page-aligned. */
86 		new = round_down(new, PAGE_SIZE);
87 
88 		/* Skip the entry if it's too small. */
89 		if (new - TRAMPOLINE_32BIT_SIZE < entry->addr)
90 			continue;
91 
92 		/* Protect against underflow. */
93 		if (new - TRAMPOLINE_32BIT_SIZE > bios_start)
94 			break;
95 
96 		bios_start = new;
97 		break;
98 	}
99 
100 	/* Place the trampoline just below the end of low memory */
101 	return bios_start - TRAMPOLINE_32BIT_SIZE;
102 }
103 
configure_5level_paging(struct boot_params * bp,void * pgtable)104 asmlinkage void configure_5level_paging(struct boot_params *bp, void *pgtable)
105 {
106 	void (*toggle_la57)(void *cr3);
107 	bool l5_required = false;
108 
109 	/* Initialize boot_params. Required for cmdline_find_option_bool(). */
110 	boot_params_ptr = bp;
111 
112 	/*
113 	 * Check if LA57 is desired and supported.
114 	 *
115 	 * There are several parts to the check:
116 	 *   - if the kernel supports 5-level paging: CONFIG_X86_5LEVEL=y
117 	 *   - if user asked to disable 5-level paging: no5lvl in cmdline
118 	 *   - if the machine supports 5-level paging:
119 	 *     + CPUID leaf 7 is supported
120 	 *     + the leaf has the feature bit set
121 	 *
122 	 * That's substitute for boot_cpu_has() in early boot code.
123 	 */
124 	if (IS_ENABLED(CONFIG_X86_5LEVEL) &&
125 			!cmdline_find_option_bool("no5lvl") &&
126 			native_cpuid_eax(0) >= 7 &&
127 			(native_cpuid_ecx(7) & (1 << (X86_FEATURE_LA57 & 31)))) {
128 		l5_required = true;
129 
130 		/* Initialize variables for 5-level paging */
131 		__pgtable_l5_enabled = 1;
132 		pgdir_shift = 48;
133 		ptrs_per_p4d = 512;
134 	}
135 
136 	/*
137 	 * The trampoline will not be used if the paging mode is already set to
138 	 * the desired one.
139 	 */
140 	if (l5_required == !!(native_read_cr4() & X86_CR4_LA57))
141 		return;
142 
143 	trampoline_32bit = (unsigned long *)find_trampoline_placement();
144 
145 	/* Preserve trampoline memory */
146 	memcpy(trampoline_save, trampoline_32bit, TRAMPOLINE_32BIT_SIZE);
147 
148 	/* Clear trampoline memory first */
149 	memset(trampoline_32bit, 0, TRAMPOLINE_32BIT_SIZE);
150 
151 	/* Copy trampoline code in place */
152 	toggle_la57 = memcpy(trampoline_32bit +
153 			TRAMPOLINE_32BIT_CODE_OFFSET / sizeof(unsigned long),
154 			&trampoline_32bit_src, TRAMPOLINE_32BIT_CODE_SIZE);
155 
156 	/*
157 	 * Avoid the need for a stack in the 32-bit trampoline code, by using
158 	 * LJMP rather than LRET to return back to long mode. LJMP takes an
159 	 * immediate absolute address, which needs to be adjusted based on the
160 	 * placement of the trampoline.
161 	 */
162 	*(u32 *)((u8 *)toggle_la57 + trampoline_ljmp_imm_offset) +=
163 						(unsigned long)toggle_la57;
164 
165 	/*
166 	 * The code below prepares page table in trampoline memory.
167 	 *
168 	 * The new page table will be used by trampoline code for switching
169 	 * from 4- to 5-level paging or vice versa.
170 	 */
171 
172 	if (l5_required) {
173 		/*
174 		 * For 4- to 5-level paging transition, set up current CR3 as
175 		 * the first and the only entry in a new top-level page table.
176 		 */
177 		*trampoline_32bit = __native_read_cr3() | _PAGE_TABLE_NOENC;
178 	} else {
179 		unsigned long src;
180 
181 		/*
182 		 * For 5- to 4-level paging transition, copy page table pointed
183 		 * by first entry in the current top-level page table as our
184 		 * new top-level page table.
185 		 *
186 		 * We cannot just point to the page table from trampoline as it
187 		 * may be above 4G.
188 		 */
189 		src = *(unsigned long *)__native_read_cr3() & PAGE_MASK;
190 		memcpy(trampoline_32bit, (void *)src, PAGE_SIZE);
191 	}
192 
193 	toggle_la57(trampoline_32bit);
194 
195 	/*
196 	 * Move the top level page table out of trampoline memory.
197 	 */
198 	memcpy(pgtable, trampoline_32bit, PAGE_SIZE);
199 	native_write_cr3((unsigned long)pgtable);
200 
201 	/* Restore trampoline memory */
202 	memcpy(trampoline_32bit, trampoline_save, TRAMPOLINE_32BIT_SIZE);
203 }
204