xref: /OK3568_Linux_fs/kernel/arch/x86/boot/compressed/pgtable_64.c (revision 4882a59341e53eb6f0b4789bf948001014eff981)
1*4882a593Smuzhiyun #include <linux/efi.h>
2*4882a593Smuzhiyun #include <asm/e820/types.h>
3*4882a593Smuzhiyun #include <asm/processor.h>
4*4882a593Smuzhiyun #include <asm/efi.h>
5*4882a593Smuzhiyun #include "pgtable.h"
6*4882a593Smuzhiyun #include "../string.h"
7*4882a593Smuzhiyun 
8*4882a593Smuzhiyun #define BIOS_START_MIN		0x20000U	/* 128K, less than this is insane */
9*4882a593Smuzhiyun #define BIOS_START_MAX		0x9f000U	/* 640K, absolute maximum */
10*4882a593Smuzhiyun 
11*4882a593Smuzhiyun #ifdef CONFIG_X86_5LEVEL
12*4882a593Smuzhiyun /* __pgtable_l5_enabled needs to be in .data to avoid being cleared along with .bss */
13*4882a593Smuzhiyun unsigned int __section(".data") __pgtable_l5_enabled;
14*4882a593Smuzhiyun unsigned int __section(".data") pgdir_shift = 39;
15*4882a593Smuzhiyun unsigned int __section(".data") ptrs_per_p4d = 1;
16*4882a593Smuzhiyun #endif
17*4882a593Smuzhiyun 
18*4882a593Smuzhiyun struct paging_config {
19*4882a593Smuzhiyun 	unsigned long trampoline_start;
20*4882a593Smuzhiyun 	unsigned long l5_required;
21*4882a593Smuzhiyun };
22*4882a593Smuzhiyun 
23*4882a593Smuzhiyun /* Buffer to preserve trampoline memory */
24*4882a593Smuzhiyun static char trampoline_save[TRAMPOLINE_32BIT_SIZE];
25*4882a593Smuzhiyun 
26*4882a593Smuzhiyun /*
27*4882a593Smuzhiyun  * Trampoline address will be printed by extract_kernel() for debugging
28*4882a593Smuzhiyun  * purposes.
29*4882a593Smuzhiyun  *
30*4882a593Smuzhiyun  * Avoid putting the pointer into .bss as it will be cleared between
31*4882a593Smuzhiyun  * paging_prepare() and extract_kernel().
32*4882a593Smuzhiyun  */
33*4882a593Smuzhiyun unsigned long *trampoline_32bit __section(".data");
34*4882a593Smuzhiyun 
35*4882a593Smuzhiyun extern struct boot_params *boot_params;
36*4882a593Smuzhiyun int cmdline_find_option_bool(const char *option);
37*4882a593Smuzhiyun 
find_trampoline_placement(void)38*4882a593Smuzhiyun static unsigned long find_trampoline_placement(void)
39*4882a593Smuzhiyun {
40*4882a593Smuzhiyun 	unsigned long bios_start = 0, ebda_start = 0;
41*4882a593Smuzhiyun 	struct boot_e820_entry *entry;
42*4882a593Smuzhiyun 	char *signature;
43*4882a593Smuzhiyun 	int i;
44*4882a593Smuzhiyun 
45*4882a593Smuzhiyun 	/*
46*4882a593Smuzhiyun 	 * Find a suitable spot for the trampoline.
47*4882a593Smuzhiyun 	 * This code is based on reserve_bios_regions().
48*4882a593Smuzhiyun 	 */
49*4882a593Smuzhiyun 
50*4882a593Smuzhiyun 	/*
51*4882a593Smuzhiyun 	 * EFI systems may not provide legacy ROM. The memory may not be mapped
52*4882a593Smuzhiyun 	 * at all.
53*4882a593Smuzhiyun 	 *
54*4882a593Smuzhiyun 	 * Only look for values in the legacy ROM for non-EFI system.
55*4882a593Smuzhiyun 	 */
56*4882a593Smuzhiyun 	signature = (char *)&boot_params->efi_info.efi_loader_signature;
57*4882a593Smuzhiyun 	if (strncmp(signature, EFI32_LOADER_SIGNATURE, 4) &&
58*4882a593Smuzhiyun 	    strncmp(signature, EFI64_LOADER_SIGNATURE, 4)) {
59*4882a593Smuzhiyun 		ebda_start = *(unsigned short *)0x40e << 4;
60*4882a593Smuzhiyun 		bios_start = *(unsigned short *)0x413 << 10;
61*4882a593Smuzhiyun 	}
62*4882a593Smuzhiyun 
63*4882a593Smuzhiyun 	if (bios_start < BIOS_START_MIN || bios_start > BIOS_START_MAX)
64*4882a593Smuzhiyun 		bios_start = BIOS_START_MAX;
65*4882a593Smuzhiyun 
66*4882a593Smuzhiyun 	if (ebda_start > BIOS_START_MIN && ebda_start < bios_start)
67*4882a593Smuzhiyun 		bios_start = ebda_start;
68*4882a593Smuzhiyun 
69*4882a593Smuzhiyun 	bios_start = round_down(bios_start, PAGE_SIZE);
70*4882a593Smuzhiyun 
71*4882a593Smuzhiyun 	/* Find the first usable memory region under bios_start. */
72*4882a593Smuzhiyun 	for (i = boot_params->e820_entries - 1; i >= 0; i--) {
73*4882a593Smuzhiyun 		unsigned long new = bios_start;
74*4882a593Smuzhiyun 
75*4882a593Smuzhiyun 		entry = &boot_params->e820_table[i];
76*4882a593Smuzhiyun 
77*4882a593Smuzhiyun 		/* Skip all entries above bios_start. */
78*4882a593Smuzhiyun 		if (bios_start <= entry->addr)
79*4882a593Smuzhiyun 			continue;
80*4882a593Smuzhiyun 
81*4882a593Smuzhiyun 		/* Skip non-RAM entries. */
82*4882a593Smuzhiyun 		if (entry->type != E820_TYPE_RAM)
83*4882a593Smuzhiyun 			continue;
84*4882a593Smuzhiyun 
85*4882a593Smuzhiyun 		/* Adjust bios_start to the end of the entry if needed. */
86*4882a593Smuzhiyun 		if (bios_start > entry->addr + entry->size)
87*4882a593Smuzhiyun 			new = entry->addr + entry->size;
88*4882a593Smuzhiyun 
89*4882a593Smuzhiyun 		/* Keep bios_start page-aligned. */
90*4882a593Smuzhiyun 		new = round_down(new, PAGE_SIZE);
91*4882a593Smuzhiyun 
92*4882a593Smuzhiyun 		/* Skip the entry if it's too small. */
93*4882a593Smuzhiyun 		if (new - TRAMPOLINE_32BIT_SIZE < entry->addr)
94*4882a593Smuzhiyun 			continue;
95*4882a593Smuzhiyun 
96*4882a593Smuzhiyun 		/* Protect against underflow. */
97*4882a593Smuzhiyun 		if (new - TRAMPOLINE_32BIT_SIZE > bios_start)
98*4882a593Smuzhiyun 			break;
99*4882a593Smuzhiyun 
100*4882a593Smuzhiyun 		bios_start = new;
101*4882a593Smuzhiyun 		break;
102*4882a593Smuzhiyun 	}
103*4882a593Smuzhiyun 
104*4882a593Smuzhiyun 	/* Place the trampoline just below the end of low memory */
105*4882a593Smuzhiyun 	return bios_start - TRAMPOLINE_32BIT_SIZE;
106*4882a593Smuzhiyun }
107*4882a593Smuzhiyun 
paging_prepare(void * rmode)108*4882a593Smuzhiyun struct paging_config paging_prepare(void *rmode)
109*4882a593Smuzhiyun {
110*4882a593Smuzhiyun 	struct paging_config paging_config = {};
111*4882a593Smuzhiyun 
112*4882a593Smuzhiyun 	/* Initialize boot_params. Required for cmdline_find_option_bool(). */
113*4882a593Smuzhiyun 	boot_params = rmode;
114*4882a593Smuzhiyun 
115*4882a593Smuzhiyun 	/*
116*4882a593Smuzhiyun 	 * Check if LA57 is desired and supported.
117*4882a593Smuzhiyun 	 *
118*4882a593Smuzhiyun 	 * There are several parts to the check:
119*4882a593Smuzhiyun 	 *   - if the kernel supports 5-level paging: CONFIG_X86_5LEVEL=y
120*4882a593Smuzhiyun 	 *   - if user asked to disable 5-level paging: no5lvl in cmdline
121*4882a593Smuzhiyun 	 *   - if the machine supports 5-level paging:
122*4882a593Smuzhiyun 	 *     + CPUID leaf 7 is supported
123*4882a593Smuzhiyun 	 *     + the leaf has the feature bit set
124*4882a593Smuzhiyun 	 *
125*4882a593Smuzhiyun 	 * That's substitute for boot_cpu_has() in early boot code.
126*4882a593Smuzhiyun 	 */
127*4882a593Smuzhiyun 	if (IS_ENABLED(CONFIG_X86_5LEVEL) &&
128*4882a593Smuzhiyun 			!cmdline_find_option_bool("no5lvl") &&
129*4882a593Smuzhiyun 			native_cpuid_eax(0) >= 7 &&
130*4882a593Smuzhiyun 			(native_cpuid_ecx(7) & (1 << (X86_FEATURE_LA57 & 31)))) {
131*4882a593Smuzhiyun 		paging_config.l5_required = 1;
132*4882a593Smuzhiyun 	}
133*4882a593Smuzhiyun 
134*4882a593Smuzhiyun 	paging_config.trampoline_start = find_trampoline_placement();
135*4882a593Smuzhiyun 
136*4882a593Smuzhiyun 	trampoline_32bit = (unsigned long *)paging_config.trampoline_start;
137*4882a593Smuzhiyun 
138*4882a593Smuzhiyun 	/* Preserve trampoline memory */
139*4882a593Smuzhiyun 	memcpy(trampoline_save, trampoline_32bit, TRAMPOLINE_32BIT_SIZE);
140*4882a593Smuzhiyun 
141*4882a593Smuzhiyun 	/* Clear trampoline memory first */
142*4882a593Smuzhiyun 	memset(trampoline_32bit, 0, TRAMPOLINE_32BIT_SIZE);
143*4882a593Smuzhiyun 
144*4882a593Smuzhiyun 	/* Copy trampoline code in place */
145*4882a593Smuzhiyun 	memcpy(trampoline_32bit + TRAMPOLINE_32BIT_CODE_OFFSET / sizeof(unsigned long),
146*4882a593Smuzhiyun 			&trampoline_32bit_src, TRAMPOLINE_32BIT_CODE_SIZE);
147*4882a593Smuzhiyun 
148*4882a593Smuzhiyun 	/*
149*4882a593Smuzhiyun 	 * The code below prepares page table in trampoline memory.
150*4882a593Smuzhiyun 	 *
151*4882a593Smuzhiyun 	 * The new page table will be used by trampoline code for switching
152*4882a593Smuzhiyun 	 * from 4- to 5-level paging or vice versa.
153*4882a593Smuzhiyun 	 *
154*4882a593Smuzhiyun 	 * If switching is not required, the page table is unused: trampoline
155*4882a593Smuzhiyun 	 * code wouldn't touch CR3.
156*4882a593Smuzhiyun 	 */
157*4882a593Smuzhiyun 
158*4882a593Smuzhiyun 	/*
159*4882a593Smuzhiyun 	 * We are not going to use the page table in trampoline memory if we
160*4882a593Smuzhiyun 	 * are already in the desired paging mode.
161*4882a593Smuzhiyun 	 */
162*4882a593Smuzhiyun 	if (paging_config.l5_required == !!(native_read_cr4() & X86_CR4_LA57))
163*4882a593Smuzhiyun 		goto out;
164*4882a593Smuzhiyun 
165*4882a593Smuzhiyun 	if (paging_config.l5_required) {
166*4882a593Smuzhiyun 		/*
167*4882a593Smuzhiyun 		 * For 4- to 5-level paging transition, set up current CR3 as
168*4882a593Smuzhiyun 		 * the first and the only entry in a new top-level page table.
169*4882a593Smuzhiyun 		 */
170*4882a593Smuzhiyun 		trampoline_32bit[TRAMPOLINE_32BIT_PGTABLE_OFFSET] = __native_read_cr3() | _PAGE_TABLE_NOENC;
171*4882a593Smuzhiyun 	} else {
172*4882a593Smuzhiyun 		unsigned long src;
173*4882a593Smuzhiyun 
174*4882a593Smuzhiyun 		/*
175*4882a593Smuzhiyun 		 * For 5- to 4-level paging transition, copy page table pointed
176*4882a593Smuzhiyun 		 * by first entry in the current top-level page table as our
177*4882a593Smuzhiyun 		 * new top-level page table.
178*4882a593Smuzhiyun 		 *
179*4882a593Smuzhiyun 		 * We cannot just point to the page table from trampoline as it
180*4882a593Smuzhiyun 		 * may be above 4G.
181*4882a593Smuzhiyun 		 */
182*4882a593Smuzhiyun 		src = *(unsigned long *)__native_read_cr3() & PAGE_MASK;
183*4882a593Smuzhiyun 		memcpy(trampoline_32bit + TRAMPOLINE_32BIT_PGTABLE_OFFSET / sizeof(unsigned long),
184*4882a593Smuzhiyun 		       (void *)src, PAGE_SIZE);
185*4882a593Smuzhiyun 	}
186*4882a593Smuzhiyun 
187*4882a593Smuzhiyun out:
188*4882a593Smuzhiyun 	return paging_config;
189*4882a593Smuzhiyun }
190*4882a593Smuzhiyun 
cleanup_trampoline(void * pgtable)191*4882a593Smuzhiyun void cleanup_trampoline(void *pgtable)
192*4882a593Smuzhiyun {
193*4882a593Smuzhiyun 	void *trampoline_pgtable;
194*4882a593Smuzhiyun 
195*4882a593Smuzhiyun 	trampoline_pgtable = trampoline_32bit + TRAMPOLINE_32BIT_PGTABLE_OFFSET / sizeof(unsigned long);
196*4882a593Smuzhiyun 
197*4882a593Smuzhiyun 	/*
198*4882a593Smuzhiyun 	 * Move the top level page table out of trampoline memory,
199*4882a593Smuzhiyun 	 * if it's there.
200*4882a593Smuzhiyun 	 */
201*4882a593Smuzhiyun 	if ((void *)__native_read_cr3() == trampoline_pgtable) {
202*4882a593Smuzhiyun 		memcpy(pgtable, trampoline_pgtable, PAGE_SIZE);
203*4882a593Smuzhiyun 		native_write_cr3((unsigned long)pgtable);
204*4882a593Smuzhiyun 	}
205*4882a593Smuzhiyun 
206*4882a593Smuzhiyun 	/* Restore trampoline memory */
207*4882a593Smuzhiyun 	memcpy(trampoline_32bit, trampoline_save, TRAMPOLINE_32BIT_SIZE);
208*4882a593Smuzhiyun 
209*4882a593Smuzhiyun 	/* Initialize variables for 5-level paging */
210*4882a593Smuzhiyun #ifdef CONFIG_X86_5LEVEL
211*4882a593Smuzhiyun 	if (__read_cr4() & X86_CR4_LA57) {
212*4882a593Smuzhiyun 		__pgtable_l5_enabled = 1;
213*4882a593Smuzhiyun 		pgdir_shift = 48;
214*4882a593Smuzhiyun 		ptrs_per_p4d = 512;
215*4882a593Smuzhiyun 	}
216*4882a593Smuzhiyun #endif
217*4882a593Smuzhiyun }
218