xref: /linux/arch/x86/boot/compressed/misc.c (revision ab93e0dd72c37d378dd936f031ffb83ff2bd87ce)
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 "../string.h"
18 #include "../voffset.h"
19 #include <asm/bootparam_utils.h>
20 
21 /*
22  * WARNING!!
23  * This code is compiled with -fPIC and it is relocated dynamically at
24  * run time, but no relocation processing is performed. This means that
25  * it is not safe to place pointers in static structures.
26  */
27 
28 /* Macros used by the included decompressor code below. */
29 #define STATIC		static
30 /* Define an externally visible malloc()/free(). */
31 #define MALLOC_VISIBLE
32 #include <linux/decompress/mm.h>
33 
34 /*
35  * Provide definitions of memzero and memmove as some of the decompressors will
36  * try to define their own functions if these are not defined as macros.
37  */
38 #define memzero(s, n)	memset((s), 0, (n))
39 #ifndef memmove
40 #define memmove		memmove
41 /* Functions used by the included decompressor code below. */
42 void *memmove(void *dest, const void *src, size_t n);
43 #endif
44 
45 /*
46  * This is set up by the setup-routine at boot-time
47  */
48 struct boot_params *boot_params_ptr;
49 
50 struct port_io_ops pio_ops;
51 
52 memptr free_mem_ptr;
53 memptr free_mem_end_ptr;
54 int spurious_nmi_count;
55 
56 static char *vidmem;
57 static int vidport;
58 
59 /* These might be accessed before .bss is cleared, so use .data instead. */
60 static int lines __section(".data");
61 static int cols __section(".data");
62 
63 #ifdef CONFIG_KERNEL_GZIP
64 #include "../../../../lib/decompress_inflate.c"
65 #endif
66 
67 #ifdef CONFIG_KERNEL_BZIP2
68 #include "../../../../lib/decompress_bunzip2.c"
69 #endif
70 
71 #ifdef CONFIG_KERNEL_LZMA
72 #include "../../../../lib/decompress_unlzma.c"
73 #endif
74 
75 #ifdef CONFIG_KERNEL_XZ
76 #include "../../../../lib/decompress_unxz.c"
77 #endif
78 
79 #ifdef CONFIG_KERNEL_LZO
80 #include "../../../../lib/decompress_unlzo.c"
81 #endif
82 
83 #ifdef CONFIG_KERNEL_LZ4
84 #include "../../../../lib/decompress_unlz4.c"
85 #endif
86 
87 #ifdef CONFIG_KERNEL_ZSTD
88 #include "../../../../lib/decompress_unzstd.c"
89 #endif
90 /*
91  * NOTE: When adding a new decompressor, please update the analysis in
92  * ../header.S.
93  */
94 
scroll(void)95 static void scroll(void)
96 {
97 	int i;
98 
99 	memmove(vidmem, vidmem + cols * 2, (lines - 1) * cols * 2);
100 	for (i = (lines - 1) * cols * 2; i < lines * cols * 2; i += 2)
101 		vidmem[i] = ' ';
102 }
103 
104 #define XMTRDY          0x20
105 
106 #define TXR             0       /*  Transmit register (WRITE) */
107 #define LSR             5       /*  Line Status               */
serial_putchar(int ch)108 static void serial_putchar(int ch)
109 {
110 	unsigned timeout = 0xffff;
111 
112 	while ((inb(early_serial_base + LSR) & XMTRDY) == 0 && --timeout)
113 		cpu_relax();
114 
115 	outb(ch, early_serial_base + TXR);
116 }
117 
__putstr(const char * s)118 void __putstr(const char *s)
119 {
120 	int x, y, pos;
121 	char c;
122 
123 	if (early_serial_base) {
124 		const char *str = s;
125 		while (*str) {
126 			if (*str == '\n')
127 				serial_putchar('\r');
128 			serial_putchar(*str++);
129 		}
130 	}
131 
132 	if (lines == 0 || cols == 0)
133 		return;
134 
135 	x = boot_params_ptr->screen_info.orig_x;
136 	y = boot_params_ptr->screen_info.orig_y;
137 
138 	while ((c = *s++) != '\0') {
139 		if (c == '\n') {
140 			x = 0;
141 			if (++y >= lines) {
142 				scroll();
143 				y--;
144 			}
145 		} else {
146 			vidmem[(x + cols * y) * 2] = c;
147 			if (++x >= cols) {
148 				x = 0;
149 				if (++y >= lines) {
150 					scroll();
151 					y--;
152 				}
153 			}
154 		}
155 	}
156 
157 	boot_params_ptr->screen_info.orig_x = x;
158 	boot_params_ptr->screen_info.orig_y = y;
159 
160 	pos = (x + cols * y) * 2;	/* Update cursor position */
161 	outb(14, vidport);
162 	outb(0xff & (pos >> 9), vidport+1);
163 	outb(15, vidport);
164 	outb(0xff & (pos >> 1), vidport+1);
165 }
166 
__putnum(unsigned long value,unsigned int base,int mindig)167 static noinline void __putnum(unsigned long value, unsigned int base,
168 			      int mindig)
169 {
170 	char buf[8*sizeof(value)+1];
171 	char *p;
172 
173 	p = buf + sizeof(buf);
174 	*--p = '\0';
175 
176 	while (mindig-- > 0 || value) {
177 		unsigned char digit = value % base;
178 		digit += (digit >= 10) ? ('a'-10) : '0';
179 		*--p = digit;
180 
181 		value /= base;
182 	}
183 
184 	__putstr(p);
185 }
186 
__puthex(unsigned long value)187 void __puthex(unsigned long value)
188 {
189 	__putnum(value, 16, sizeof(value)*2);
190 }
191 
__putdec(unsigned long value)192 void __putdec(unsigned long value)
193 {
194 	__putnum(value, 10, 1);
195 }
196 
197 #ifdef CONFIG_X86_NEED_RELOCS
handle_relocations(void * output,unsigned long output_len,unsigned long virt_addr)198 static void handle_relocations(void *output, unsigned long output_len,
199 			       unsigned long virt_addr)
200 {
201 	int *reloc;
202 	unsigned long delta, map, ptr;
203 	unsigned long min_addr = (unsigned long)output;
204 	unsigned long max_addr = min_addr + (VO___bss_start - VO__text);
205 
206 	/*
207 	 * Calculate the delta between where vmlinux was linked to load
208 	 * and where it was actually loaded.
209 	 */
210 	delta = min_addr - LOAD_PHYSICAL_ADDR;
211 
212 	/*
213 	 * The kernel contains a table of relocation addresses. Those
214 	 * addresses have the final load address of the kernel in virtual
215 	 * memory. We are currently working in the self map. So we need to
216 	 * create an adjustment for kernel memory addresses to the self map.
217 	 * This will involve subtracting out the base address of the kernel.
218 	 */
219 	map = delta - __START_KERNEL_map;
220 
221 	/*
222 	 * 32-bit always performs relocations. 64-bit relocations are only
223 	 * needed if KASLR has chosen a different starting address offset
224 	 * from __START_KERNEL_map.
225 	 */
226 	if (IS_ENABLED(CONFIG_X86_64))
227 		delta = virt_addr - LOAD_PHYSICAL_ADDR;
228 
229 	if (!delta) {
230 		debug_putstr("No relocation needed... ");
231 		return;
232 	}
233 	debug_putstr("Performing relocations... ");
234 
235 	/*
236 	 * Process relocations: 32 bit relocations first then 64 bit after.
237 	 * Two sets of binary relocations are added to the end of the kernel
238 	 * before compression. Each relocation table entry is the kernel
239 	 * address of the location which needs to be updated stored as a
240 	 * 32-bit value which is sign extended to 64 bits.
241 	 *
242 	 * Format is:
243 	 *
244 	 * kernel bits...
245 	 * 0 - zero terminator for 64 bit relocations
246 	 * 64 bit relocation repeated
247 	 * 0 - zero terminator for 32 bit relocations
248 	 * 32 bit relocation repeated
249 	 *
250 	 * So we work backwards from the end of the decompressed image.
251 	 */
252 	for (reloc = output + output_len - sizeof(*reloc); *reloc; reloc--) {
253 		long extended = *reloc;
254 		extended += map;
255 
256 		ptr = (unsigned long)extended;
257 		if (ptr < min_addr || ptr > max_addr)
258 			error("32-bit relocation outside of kernel!\n");
259 
260 		*(uint32_t *)ptr += delta;
261 	}
262 #ifdef CONFIG_X86_64
263 	for (reloc--; *reloc; reloc--) {
264 		long extended = *reloc;
265 		extended += map;
266 
267 		ptr = (unsigned long)extended;
268 		if (ptr < min_addr || ptr > max_addr)
269 			error("64-bit relocation outside of kernel!\n");
270 
271 		*(uint64_t *)ptr += delta;
272 	}
273 #endif
274 }
275 #else
handle_relocations(void * output,unsigned long output_len,unsigned long virt_addr)276 static inline void handle_relocations(void *output, unsigned long output_len,
277 				      unsigned long virt_addr)
278 { }
279 #endif
280 
parse_elf(void * output)281 static size_t parse_elf(void *output)
282 {
283 #ifdef CONFIG_X86_64
284 	Elf64_Ehdr ehdr;
285 	Elf64_Phdr *phdrs, *phdr;
286 #else
287 	Elf32_Ehdr ehdr;
288 	Elf32_Phdr *phdrs, *phdr;
289 #endif
290 	void *dest;
291 	int i;
292 
293 	memcpy(&ehdr, output, sizeof(ehdr));
294 	if (ehdr.e_ident[EI_MAG0] != ELFMAG0 ||
295 	   ehdr.e_ident[EI_MAG1] != ELFMAG1 ||
296 	   ehdr.e_ident[EI_MAG2] != ELFMAG2 ||
297 	   ehdr.e_ident[EI_MAG3] != ELFMAG3)
298 		error("Kernel is not a valid ELF file");
299 
300 	debug_putstr("Parsing ELF... ");
301 
302 	phdrs = malloc(sizeof(*phdrs) * ehdr.e_phnum);
303 	if (!phdrs)
304 		error("Failed to allocate space for phdrs");
305 
306 	memcpy(phdrs, output + ehdr.e_phoff, sizeof(*phdrs) * ehdr.e_phnum);
307 
308 	for (i = 0; i < ehdr.e_phnum; i++) {
309 		phdr = &phdrs[i];
310 
311 		switch (phdr->p_type) {
312 		case PT_LOAD:
313 #ifdef CONFIG_X86_64
314 			if ((phdr->p_align % 0x200000) != 0)
315 				error("Alignment of LOAD segment isn't multiple of 2MB");
316 #endif
317 #ifdef CONFIG_RELOCATABLE
318 			dest = output;
319 			dest += (phdr->p_paddr - LOAD_PHYSICAL_ADDR);
320 #else
321 			dest = (void *)(phdr->p_paddr);
322 #endif
323 			memmove(dest, output + phdr->p_offset, phdr->p_filesz);
324 			break;
325 		default: /* Ignore other PT_* */ break;
326 		}
327 	}
328 
329 	free(phdrs);
330 
331 	return ehdr.e_entry - LOAD_PHYSICAL_ADDR;
332 }
333 
334 const unsigned long kernel_text_size = VO___start_rodata - VO__text;
335 const unsigned long kernel_total_size = VO__end - VO__text;
336 
337 static u8 boot_heap[BOOT_HEAP_SIZE] __aligned(4);
338 
339 extern unsigned char input_data[];
340 extern unsigned int input_len, output_len;
341 
decompress_kernel(unsigned char * outbuf,unsigned long virt_addr,void (* error)(char * x))342 unsigned long decompress_kernel(unsigned char *outbuf, unsigned long virt_addr,
343 				void (*error)(char *x))
344 {
345 	unsigned long entry;
346 
347 	if (!free_mem_ptr) {
348 		free_mem_ptr     = (unsigned long)boot_heap;
349 		free_mem_end_ptr = (unsigned long)boot_heap + sizeof(boot_heap);
350 	}
351 
352 	if (__decompress(input_data, input_len, NULL, NULL, outbuf, output_len,
353 			 NULL, error) < 0)
354 		return ULONG_MAX;
355 
356 	entry = parse_elf(outbuf);
357 	handle_relocations(outbuf, output_len, virt_addr);
358 
359 	return entry;
360 }
361 
362 /*
363  * Set the memory encryption xloadflag based on the mem_encrypt= command line
364  * parameter, if provided.
365  */
parse_mem_encrypt(struct setup_header * hdr)366 static void parse_mem_encrypt(struct setup_header *hdr)
367 {
368 	int on = cmdline_find_option_bool("mem_encrypt=on");
369 	int off = cmdline_find_option_bool("mem_encrypt=off");
370 
371 	if (on > off)
372 		hdr->xloadflags |= XLF_MEM_ENCRYPTION;
373 }
374 
early_sev_detect(void)375 static void early_sev_detect(void)
376 {
377 	/*
378 	 * Accessing video memory causes guest termination because
379 	 * the boot stage2 #VC handler of SEV-ES/SNP guests does not
380 	 * support MMIO handling and kexec -c adds screen_info to the
381 	 * boot parameters passed to the kexec kernel, which causes
382 	 * console output to be dumped to both video and serial.
383 	 */
384 	if (sev_status & MSR_AMD64_SEV_ES_ENABLED)
385 		lines = cols = 0;
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 	early_sev_detect();
444 
445 	console_init();
446 
447 	/*
448 	 * Save RSDP address for later use. Have this after console_init()
449 	 * so that early debugging output from the RSDP parsing code can be
450 	 * collected.
451 	 */
452 	boot_params_ptr->acpi_rsdp_addr = get_rsdp_addr();
453 
454 	debug_putstr("early console in extract_kernel\n");
455 
456 	free_mem_ptr     = heap;	/* Heap */
457 	free_mem_end_ptr = heap + BOOT_HEAP_SIZE;
458 
459 	/*
460 	 * The memory hole needed for the kernel is the larger of either
461 	 * the entire decompressed kernel plus relocation table, or the
462 	 * entire decompressed kernel plus .bss and .brk sections.
463 	 *
464 	 * On X86_64, the memory is mapped with PMD pages. Round the
465 	 * size up so that the full extent of PMD pages mapped is
466 	 * included in the check against the valid memory table
467 	 * entries. This ensures the full mapped area is usable RAM
468 	 * and doesn't include any reserved areas.
469 	 */
470 	needed_size = max_t(unsigned long, output_len, kernel_total_size);
471 #ifdef CONFIG_X86_64
472 	needed_size = ALIGN(needed_size, MIN_KERNEL_ALIGN);
473 #endif
474 
475 	/* Report initial kernel position details. */
476 	debug_putaddr(input_data);
477 	debug_putaddr(input_len);
478 	debug_putaddr(output);
479 	debug_putaddr(output_len);
480 	debug_putaddr(kernel_total_size);
481 	debug_putaddr(needed_size);
482 
483 #ifdef CONFIG_X86_64
484 	/* Report address of 32-bit trampoline */
485 	debug_putaddr(trampoline_32bit);
486 #endif
487 
488 	choose_random_location((unsigned long)input_data, input_len,
489 				(unsigned long *)&output,
490 				needed_size,
491 				&virt_addr);
492 
493 	/* Validate memory location choices. */
494 	if ((unsigned long)output & (MIN_KERNEL_ALIGN - 1))
495 		error("Destination physical address inappropriately aligned");
496 	if (virt_addr & (MIN_KERNEL_ALIGN - 1))
497 		error("Destination virtual address inappropriately aligned");
498 #ifdef CONFIG_X86_64
499 	if (heap > 0x3fffffffffffUL)
500 		error("Destination address too large");
501 	if (virt_addr + needed_size > KERNEL_IMAGE_SIZE)
502 		error("Destination virtual address is beyond the kernel mapping area");
503 #else
504 	if (heap > ((-__PAGE_OFFSET-(128<<20)-1) & 0x7fffffff))
505 		error("Destination address too large");
506 #endif
507 #ifndef CONFIG_RELOCATABLE
508 	if (virt_addr != LOAD_PHYSICAL_ADDR)
509 		error("Destination virtual address changed when not relocatable");
510 #endif
511 
512 	debug_putstr("\nDecompressing Linux... ");
513 
514 	if (init_unaccepted_memory()) {
515 		debug_putstr("Accepting memory... ");
516 		accept_memory(__pa(output), needed_size);
517 	}
518 
519 	entry_offset = decompress_kernel(output, virt_addr, error);
520 
521 	debug_putstr("done.\nBooting the kernel (entry_offset: 0x");
522 	debug_puthex(entry_offset);
523 	debug_putstr(").\n");
524 
525 	/* Disable exception handling before booting the kernel */
526 	cleanup_exception_handling();
527 
528 	if (spurious_nmi_count) {
529 		error_putstr("Spurious early NMIs ignored: ");
530 		error_putdec(spurious_nmi_count);
531 		error_putstr("\n");
532 	}
533 
534 	return output + entry_offset;
535 }
536