xref: /linux/arch/x86/boot/compressed/misc.c (revision 3a39d672e7f48b8d6b91a09afa4b55352773b4b5)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * misc.c
4  *
5  * This is a collection of several routines used to extract the kernel
6  * which includes KASLR relocation, decompression, ELF parsing, and
7  * relocation processing. Additionally included are the screen and serial
8  * output functions and related debugging support functions.
9  *
10  * malloc by Hannu Savolainen 1993 and Matthias Urlichs 1994
11  * puts by Nick Holloway 1993, better puts by Martin Mares 1995
12  * High loaded stuff by Hans Lermen & Werner Almesberger, Feb. 1996
13  */
14 
15 #include "misc.h"
16 #include "error.h"
17 #include "pgtable.h"
18 #include "../string.h"
19 #include "../voffset.h"
20 #include <asm/bootparam_utils.h>
21 
22 /*
23  * WARNING!!
24  * This code is compiled with -fPIC and it is relocated dynamically at
25  * run time, but no relocation processing is performed. This means that
26  * it is not safe to place pointers in static structures.
27  */
28 
29 /* Macros used by the included decompressor code below. */
30 #define STATIC		static
31 /* Define an externally visible malloc()/free(). */
32 #define MALLOC_VISIBLE
33 #include <linux/decompress/mm.h>
34 
35 /*
36  * Provide definitions of memzero and memmove as some of the decompressors will
37  * try to define their own functions if these are not defined as macros.
38  */
39 #define memzero(s, n)	memset((s), 0, (n))
40 #ifndef memmove
41 #define memmove		memmove
42 /* Functions used by the included decompressor code below. */
43 void *memmove(void *dest, const void *src, size_t n);
44 #endif
45 
46 /*
47  * This is set up by the setup-routine at boot-time
48  */
49 struct boot_params *boot_params_ptr;
50 
51 struct port_io_ops pio_ops;
52 
53 memptr free_mem_ptr;
54 memptr free_mem_end_ptr;
55 int spurious_nmi_count;
56 
57 static char *vidmem;
58 static int vidport;
59 
60 /* These might be accessed before .bss is cleared, so use .data instead. */
61 static int lines __section(".data");
62 static int cols __section(".data");
63 
64 #ifdef CONFIG_KERNEL_GZIP
65 #include "../../../../lib/decompress_inflate.c"
66 #endif
67 
68 #ifdef CONFIG_KERNEL_BZIP2
69 #include "../../../../lib/decompress_bunzip2.c"
70 #endif
71 
72 #ifdef CONFIG_KERNEL_LZMA
73 #include "../../../../lib/decompress_unlzma.c"
74 #endif
75 
76 #ifdef CONFIG_KERNEL_XZ
77 #include "../../../../lib/decompress_unxz.c"
78 #endif
79 
80 #ifdef CONFIG_KERNEL_LZO
81 #include "../../../../lib/decompress_unlzo.c"
82 #endif
83 
84 #ifdef CONFIG_KERNEL_LZ4
85 #include "../../../../lib/decompress_unlz4.c"
86 #endif
87 
88 #ifdef CONFIG_KERNEL_ZSTD
89 #include "../../../../lib/decompress_unzstd.c"
90 #endif
91 /*
92  * NOTE: When adding a new decompressor, please update the analysis in
93  * ../header.S.
94  */
95 
scroll(void)96 static void scroll(void)
97 {
98 	int i;
99 
100 	memmove(vidmem, vidmem + cols * 2, (lines - 1) * cols * 2);
101 	for (i = (lines - 1) * cols * 2; i < lines * cols * 2; i += 2)
102 		vidmem[i] = ' ';
103 }
104 
105 #define XMTRDY          0x20
106 
107 #define TXR             0       /*  Transmit register (WRITE) */
108 #define LSR             5       /*  Line Status               */
serial_putchar(int ch)109 static void serial_putchar(int ch)
110 {
111 	unsigned timeout = 0xffff;
112 
113 	while ((inb(early_serial_base + LSR) & XMTRDY) == 0 && --timeout)
114 		cpu_relax();
115 
116 	outb(ch, early_serial_base + TXR);
117 }
118 
__putstr(const char * s)119 void __putstr(const char *s)
120 {
121 	int x, y, pos;
122 	char c;
123 
124 	if (early_serial_base) {
125 		const char *str = s;
126 		while (*str) {
127 			if (*str == '\n')
128 				serial_putchar('\r');
129 			serial_putchar(*str++);
130 		}
131 	}
132 
133 	if (lines == 0 || cols == 0)
134 		return;
135 
136 	x = boot_params_ptr->screen_info.orig_x;
137 	y = boot_params_ptr->screen_info.orig_y;
138 
139 	while ((c = *s++) != '\0') {
140 		if (c == '\n') {
141 			x = 0;
142 			if (++y >= lines) {
143 				scroll();
144 				y--;
145 			}
146 		} else {
147 			vidmem[(x + cols * y) * 2] = c;
148 			if (++x >= cols) {
149 				x = 0;
150 				if (++y >= lines) {
151 					scroll();
152 					y--;
153 				}
154 			}
155 		}
156 	}
157 
158 	boot_params_ptr->screen_info.orig_x = x;
159 	boot_params_ptr->screen_info.orig_y = y;
160 
161 	pos = (x + cols * y) * 2;	/* Update cursor position */
162 	outb(14, vidport);
163 	outb(0xff & (pos >> 9), vidport+1);
164 	outb(15, vidport);
165 	outb(0xff & (pos >> 1), vidport+1);
166 }
167 
__putnum(unsigned long value,unsigned int base,int mindig)168 static noinline void __putnum(unsigned long value, unsigned int base,
169 			      int mindig)
170 {
171 	char buf[8*sizeof(value)+1];
172 	char *p;
173 
174 	p = buf + sizeof(buf);
175 	*--p = '\0';
176 
177 	while (mindig-- > 0 || value) {
178 		unsigned char digit = value % base;
179 		digit += (digit >= 10) ? ('a'-10) : '0';
180 		*--p = digit;
181 
182 		value /= base;
183 	}
184 
185 	__putstr(p);
186 }
187 
__puthex(unsigned long value)188 void __puthex(unsigned long value)
189 {
190 	__putnum(value, 16, sizeof(value)*2);
191 }
192 
__putdec(unsigned long value)193 void __putdec(unsigned long value)
194 {
195 	__putnum(value, 10, 1);
196 }
197 
198 #ifdef CONFIG_X86_NEED_RELOCS
handle_relocations(void * output,unsigned long output_len,unsigned long virt_addr)199 static void handle_relocations(void *output, unsigned long output_len,
200 			       unsigned long virt_addr)
201 {
202 	int *reloc;
203 	unsigned long delta, map, ptr;
204 	unsigned long min_addr = (unsigned long)output;
205 	unsigned long max_addr = min_addr + (VO___bss_start - VO__text);
206 
207 	/*
208 	 * Calculate the delta between where vmlinux was linked to load
209 	 * and where it was actually loaded.
210 	 */
211 	delta = min_addr - LOAD_PHYSICAL_ADDR;
212 
213 	/*
214 	 * The kernel contains a table of relocation addresses. Those
215 	 * addresses have the final load address of the kernel in virtual
216 	 * memory. We are currently working in the self map. So we need to
217 	 * create an adjustment for kernel memory addresses to the self map.
218 	 * This will involve subtracting out the base address of the kernel.
219 	 */
220 	map = delta - __START_KERNEL_map;
221 
222 	/*
223 	 * 32-bit always performs relocations. 64-bit relocations are only
224 	 * needed if KASLR has chosen a different starting address offset
225 	 * from __START_KERNEL_map.
226 	 */
227 	if (IS_ENABLED(CONFIG_X86_64))
228 		delta = virt_addr - LOAD_PHYSICAL_ADDR;
229 
230 	if (!delta) {
231 		debug_putstr("No relocation needed... ");
232 		return;
233 	}
234 	debug_putstr("Performing relocations... ");
235 
236 	/*
237 	 * Process relocations: 32 bit relocations first then 64 bit after.
238 	 * Three sets of binary relocations are added to the end of the kernel
239 	 * before compression. Each relocation table entry is the kernel
240 	 * address of the location which needs to be updated stored as a
241 	 * 32-bit value which is sign extended to 64 bits.
242 	 *
243 	 * Format is:
244 	 *
245 	 * kernel bits...
246 	 * 0 - zero terminator for 64 bit relocations
247 	 * 64 bit relocation repeated
248 	 * 0 - zero terminator for inverse 32 bit relocations
249 	 * 32 bit inverse relocation repeated
250 	 * 0 - zero terminator for 32 bit relocations
251 	 * 32 bit relocation repeated
252 	 *
253 	 * So we work backwards from the end of the decompressed image.
254 	 */
255 	for (reloc = output + output_len - sizeof(*reloc); *reloc; reloc--) {
256 		long extended = *reloc;
257 		extended += map;
258 
259 		ptr = (unsigned long)extended;
260 		if (ptr < min_addr || ptr > max_addr)
261 			error("32-bit relocation outside of kernel!\n");
262 
263 		*(uint32_t *)ptr += delta;
264 	}
265 #ifdef CONFIG_X86_64
266 	while (*--reloc) {
267 		long extended = *reloc;
268 		extended += map;
269 
270 		ptr = (unsigned long)extended;
271 		if (ptr < min_addr || ptr > max_addr)
272 			error("inverse 32-bit relocation outside of kernel!\n");
273 
274 		*(int32_t *)ptr -= delta;
275 	}
276 	for (reloc--; *reloc; reloc--) {
277 		long extended = *reloc;
278 		extended += map;
279 
280 		ptr = (unsigned long)extended;
281 		if (ptr < min_addr || ptr > max_addr)
282 			error("64-bit relocation outside of kernel!\n");
283 
284 		*(uint64_t *)ptr += delta;
285 	}
286 #endif
287 }
288 #else
handle_relocations(void * output,unsigned long output_len,unsigned long virt_addr)289 static inline void handle_relocations(void *output, unsigned long output_len,
290 				      unsigned long virt_addr)
291 { }
292 #endif
293 
parse_elf(void * output)294 static size_t parse_elf(void *output)
295 {
296 #ifdef CONFIG_X86_64
297 	Elf64_Ehdr ehdr;
298 	Elf64_Phdr *phdrs, *phdr;
299 #else
300 	Elf32_Ehdr ehdr;
301 	Elf32_Phdr *phdrs, *phdr;
302 #endif
303 	void *dest;
304 	int i;
305 
306 	memcpy(&ehdr, output, sizeof(ehdr));
307 	if (ehdr.e_ident[EI_MAG0] != ELFMAG0 ||
308 	   ehdr.e_ident[EI_MAG1] != ELFMAG1 ||
309 	   ehdr.e_ident[EI_MAG2] != ELFMAG2 ||
310 	   ehdr.e_ident[EI_MAG3] != ELFMAG3)
311 		error("Kernel is not a valid ELF file");
312 
313 	debug_putstr("Parsing ELF... ");
314 
315 	phdrs = malloc(sizeof(*phdrs) * ehdr.e_phnum);
316 	if (!phdrs)
317 		error("Failed to allocate space for phdrs");
318 
319 	memcpy(phdrs, output + ehdr.e_phoff, sizeof(*phdrs) * ehdr.e_phnum);
320 
321 	for (i = 0; i < ehdr.e_phnum; i++) {
322 		phdr = &phdrs[i];
323 
324 		switch (phdr->p_type) {
325 		case PT_LOAD:
326 #ifdef CONFIG_X86_64
327 			if ((phdr->p_align % 0x200000) != 0)
328 				error("Alignment of LOAD segment isn't multiple of 2MB");
329 #endif
330 #ifdef CONFIG_RELOCATABLE
331 			dest = output;
332 			dest += (phdr->p_paddr - LOAD_PHYSICAL_ADDR);
333 #else
334 			dest = (void *)(phdr->p_paddr);
335 #endif
336 			memmove(dest, output + phdr->p_offset, phdr->p_filesz);
337 			break;
338 		default: /* Ignore other PT_* */ break;
339 		}
340 	}
341 
342 	free(phdrs);
343 
344 	return ehdr.e_entry - LOAD_PHYSICAL_ADDR;
345 }
346 
347 const unsigned long kernel_text_size = VO___start_rodata - VO__text;
348 const unsigned long kernel_total_size = VO__end - VO__text;
349 
350 static u8 boot_heap[BOOT_HEAP_SIZE] __aligned(4);
351 
352 extern unsigned char input_data[];
353 extern unsigned int input_len, output_len;
354 
decompress_kernel(unsigned char * outbuf,unsigned long virt_addr,void (* error)(char * x))355 unsigned long decompress_kernel(unsigned char *outbuf, unsigned long virt_addr,
356 				void (*error)(char *x))
357 {
358 	unsigned long entry;
359 
360 	if (!free_mem_ptr) {
361 		free_mem_ptr     = (unsigned long)boot_heap;
362 		free_mem_end_ptr = (unsigned long)boot_heap + sizeof(boot_heap);
363 	}
364 
365 	if (__decompress(input_data, input_len, NULL, NULL, outbuf, output_len,
366 			 NULL, error) < 0)
367 		return ULONG_MAX;
368 
369 	entry = parse_elf(outbuf);
370 	handle_relocations(outbuf, output_len, virt_addr);
371 
372 	return entry;
373 }
374 
375 /*
376  * Set the memory encryption xloadflag based on the mem_encrypt= command line
377  * parameter, if provided.
378  */
parse_mem_encrypt(struct setup_header * hdr)379 static void parse_mem_encrypt(struct setup_header *hdr)
380 {
381 	int on = cmdline_find_option_bool("mem_encrypt=on");
382 	int off = cmdline_find_option_bool("mem_encrypt=off");
383 
384 	if (on > off)
385 		hdr->xloadflags |= XLF_MEM_ENCRYPTION;
386 }
387 
388 /*
389  * The compressed kernel image (ZO), has been moved so that its position
390  * is against the end of the buffer used to hold the uncompressed kernel
391  * image (VO) and the execution environment (.bss, .brk), which makes sure
392  * there is room to do the in-place decompression. (See header.S for the
393  * calculations.)
394  *
395  *                             |-----compressed kernel image------|
396  *                             V                                  V
397  * 0                       extract_offset                      +INIT_SIZE
398  * |-----------|---------------|-------------------------|--------|
399  *             |               |                         |        |
400  *           VO__text      startup_32 of ZO          VO__end    ZO__end
401  *             ^                                         ^
402  *             |-------uncompressed kernel image---------|
403  *
404  */
extract_kernel(void * rmode,unsigned char * output)405 asmlinkage __visible void *extract_kernel(void *rmode, unsigned char *output)
406 {
407 	unsigned long virt_addr = LOAD_PHYSICAL_ADDR;
408 	memptr heap = (memptr)boot_heap;
409 	unsigned long needed_size;
410 	size_t entry_offset;
411 
412 	/* Retain x86 boot parameters pointer passed from startup_32/64. */
413 	boot_params_ptr = rmode;
414 
415 	/* Clear flags intended for solely in-kernel use. */
416 	boot_params_ptr->hdr.loadflags &= ~KASLR_FLAG;
417 
418 	parse_mem_encrypt(&boot_params_ptr->hdr);
419 
420 	sanitize_boot_params(boot_params_ptr);
421 
422 	if (boot_params_ptr->screen_info.orig_video_mode == 7) {
423 		vidmem = (char *) 0xb0000;
424 		vidport = 0x3b4;
425 	} else {
426 		vidmem = (char *) 0xb8000;
427 		vidport = 0x3d4;
428 	}
429 
430 	lines = boot_params_ptr->screen_info.orig_video_lines;
431 	cols = boot_params_ptr->screen_info.orig_video_cols;
432 
433 	init_default_io_ops();
434 
435 	/*
436 	 * Detect TDX guest environment.
437 	 *
438 	 * It has to be done before console_init() in order to use
439 	 * paravirtualized port I/O operations if needed.
440 	 */
441 	early_tdx_detect();
442 
443 	console_init();
444 
445 	/*
446 	 * Save RSDP address for later use. Have this after console_init()
447 	 * so that early debugging output from the RSDP parsing code can be
448 	 * collected.
449 	 */
450 	boot_params_ptr->acpi_rsdp_addr = get_rsdp_addr();
451 
452 	debug_putstr("early console in extract_kernel\n");
453 
454 	free_mem_ptr     = heap;	/* Heap */
455 	free_mem_end_ptr = heap + BOOT_HEAP_SIZE;
456 
457 	/*
458 	 * The memory hole needed for the kernel is the larger of either
459 	 * the entire decompressed kernel plus relocation table, or the
460 	 * entire decompressed kernel plus .bss and .brk sections.
461 	 *
462 	 * On X86_64, the memory is mapped with PMD pages. Round the
463 	 * size up so that the full extent of PMD pages mapped is
464 	 * included in the check against the valid memory table
465 	 * entries. This ensures the full mapped area is usable RAM
466 	 * and doesn't include any reserved areas.
467 	 */
468 	needed_size = max_t(unsigned long, output_len, kernel_total_size);
469 #ifdef CONFIG_X86_64
470 	needed_size = ALIGN(needed_size, MIN_KERNEL_ALIGN);
471 #endif
472 
473 	/* Report initial kernel position details. */
474 	debug_putaddr(input_data);
475 	debug_putaddr(input_len);
476 	debug_putaddr(output);
477 	debug_putaddr(output_len);
478 	debug_putaddr(kernel_total_size);
479 	debug_putaddr(needed_size);
480 
481 #ifdef CONFIG_X86_64
482 	/* Report address of 32-bit trampoline */
483 	debug_putaddr(trampoline_32bit);
484 #endif
485 
486 	choose_random_location((unsigned long)input_data, input_len,
487 				(unsigned long *)&output,
488 				needed_size,
489 				&virt_addr);
490 
491 	/* Validate memory location choices. */
492 	if ((unsigned long)output & (MIN_KERNEL_ALIGN - 1))
493 		error("Destination physical address inappropriately aligned");
494 	if (virt_addr & (MIN_KERNEL_ALIGN - 1))
495 		error("Destination virtual address inappropriately aligned");
496 #ifdef CONFIG_X86_64
497 	if (heap > 0x3fffffffffffUL)
498 		error("Destination address too large");
499 	if (virt_addr + needed_size > KERNEL_IMAGE_SIZE)
500 		error("Destination virtual address is beyond the kernel mapping area");
501 #else
502 	if (heap > ((-__PAGE_OFFSET-(128<<20)-1) & 0x7fffffff))
503 		error("Destination address too large");
504 #endif
505 #ifndef CONFIG_RELOCATABLE
506 	if (virt_addr != LOAD_PHYSICAL_ADDR)
507 		error("Destination virtual address changed when not relocatable");
508 #endif
509 
510 	debug_putstr("\nDecompressing Linux... ");
511 
512 	if (init_unaccepted_memory()) {
513 		debug_putstr("Accepting memory... ");
514 		accept_memory(__pa(output), needed_size);
515 	}
516 
517 	entry_offset = decompress_kernel(output, virt_addr, error);
518 
519 	debug_putstr("done.\nBooting the kernel (entry_offset: 0x");
520 	debug_puthex(entry_offset);
521 	debug_putstr(").\n");
522 
523 	/* Disable exception handling before booting the kernel */
524 	cleanup_exception_handling();
525 
526 	if (spurious_nmi_count) {
527 		error_putstr("Spurious early NMIs ignored: ");
528 		error_putdec(spurious_nmi_count);
529 		error_putstr("\n");
530 	}
531 
532 	return output + entry_offset;
533 }
534