xref: /linux/arch/um/kernel/um_arch.c (revision 90daca7c8f6f33e9482591446d2e76b18a21be49)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2000 - 2007 Jeff Dike (jdike@{addtoit,linux.intel}.com)
4  */
5 
6 #include <linux/cpu.h>
7 #include <linux/delay.h>
8 #include <linux/init.h>
9 #include <linux/mm.h>
10 #include <linux/ctype.h>
11 #include <linux/module.h>
12 #include <linux/panic_notifier.h>
13 #include <linux/seq_file.h>
14 #include <linux/string.h>
15 #include <linux/utsname.h>
16 #include <linux/sched.h>
17 #include <linux/sched/task.h>
18 #include <linux/kmsg_dump.h>
19 #include <linux/suspend.h>
20 #include <linux/random.h>
21 
22 #include <asm/processor.h>
23 #include <asm/cpufeature.h>
24 #include <asm/sections.h>
25 #include <asm/setup.h>
26 #include <asm/text-patching.h>
27 #include <as-layout.h>
28 #include <arch.h>
29 #include <init.h>
30 #include <kern.h>
31 #include <kern_util.h>
32 #include <mem_user.h>
33 #include <os.h>
34 
35 #include "um_arch.h"
36 
37 #define DEFAULT_COMMAND_LINE_ROOT "root=98:0"
38 #define DEFAULT_COMMAND_LINE_CONSOLE "console=tty0"
39 
40 /* Changed in add_arg and setup_arch, which run before SMP is started */
41 static char __initdata command_line[COMMAND_LINE_SIZE] = { 0 };
42 
43 static void __init add_arg(char *arg)
44 {
45 	if (strlen(command_line) + strlen(arg) + 1 > COMMAND_LINE_SIZE) {
46 		os_warn("add_arg: Too many command line arguments!\n");
47 		exit(1);
48 	}
49 	if (strlen(command_line) > 0)
50 		strcat(command_line, " ");
51 	strcat(command_line, arg);
52 }
53 
54 /*
55  * These fields are initialized at boot time and not changed.
56  * XXX This structure is used only in the non-SMP case.  Maybe this
57  * should be moved to smp.c.
58  */
59 struct cpuinfo_um boot_cpu_data = {
60 	.loops_per_jiffy	= 0,
61 	.ipi_pipe		= { -1, -1 },
62 	.cache_alignment	= L1_CACHE_BYTES,
63 	.x86_capability		= { 0 }
64 };
65 
66 EXPORT_SYMBOL(boot_cpu_data);
67 
68 union thread_union cpu0_irqstack
69 	__section(".data..init_irqstack") =
70 		{ .thread_info = INIT_THREAD_INFO(init_task) };
71 
72 /* Changed in setup_arch, which is called in early boot */
73 static char host_info[(__NEW_UTS_LEN + 1) * 5];
74 
75 static int show_cpuinfo(struct seq_file *m, void *v)
76 {
77 	int i = 0;
78 
79 	seq_printf(m, "processor\t: %d\n", i);
80 	seq_printf(m, "vendor_id\t: User Mode Linux\n");
81 	seq_printf(m, "model name\t: UML\n");
82 	seq_printf(m, "mode\t\t: skas\n");
83 	seq_printf(m, "host\t\t: %s\n", host_info);
84 	seq_printf(m, "fpu\t\t: %s\n", cpu_has(&boot_cpu_data, X86_FEATURE_FPU) ? "yes" : "no");
85 	seq_printf(m, "flags\t\t:");
86 	for (i = 0; i < 32*NCAPINTS; i++)
87 		if (cpu_has(&boot_cpu_data, i) && (x86_cap_flags[i] != NULL))
88 			seq_printf(m, " %s", x86_cap_flags[i]);
89 	seq_printf(m, "\n");
90 	seq_printf(m, "cache_alignment\t: %d\n", boot_cpu_data.cache_alignment);
91 	seq_printf(m, "bogomips\t: %lu.%02lu\n",
92 		   loops_per_jiffy/(500000/HZ),
93 		   (loops_per_jiffy/(5000/HZ)) % 100);
94 
95 
96 	return 0;
97 }
98 
99 static void *c_start(struct seq_file *m, loff_t *pos)
100 {
101 	return *pos < nr_cpu_ids ? &boot_cpu_data + *pos : NULL;
102 }
103 
104 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
105 {
106 	++*pos;
107 	return c_start(m, pos);
108 }
109 
110 static void c_stop(struct seq_file *m, void *v)
111 {
112 }
113 
114 const struct seq_operations cpuinfo_op = {
115 	.start	= c_start,
116 	.next	= c_next,
117 	.stop	= c_stop,
118 	.show	= show_cpuinfo,
119 };
120 
121 /* Set in linux_main */
122 unsigned long uml_physmem;
123 EXPORT_SYMBOL(uml_physmem);
124 
125 unsigned long uml_reserved; /* Also modified in mem_init */
126 unsigned long start_vm;
127 unsigned long end_vm;
128 
129 /* Set in early boot */
130 static int have_root __initdata;
131 static int have_console __initdata;
132 
133 /* Set in uml_mem_setup and modified in linux_main */
134 unsigned long long physmem_size = 64 * 1024 * 1024;
135 EXPORT_SYMBOL(physmem_size);
136 
137 static const char *usage_string =
138 "User Mode Linux v%s\n"
139 "	available at http://user-mode-linux.sourceforge.net/\n\n";
140 
141 static int __init uml_version_setup(char *line, int *add)
142 {
143 	/* Explicitly use printf() to show version in stdout */
144 	printf("%s\n", init_utsname()->release);
145 	exit(0);
146 
147 	return 0;
148 }
149 
150 __uml_setup("--version", uml_version_setup,
151 "--version\n"
152 "    Prints the version number of the kernel.\n\n"
153 );
154 
155 static int __init uml_root_setup(char *line, int *add)
156 {
157 	have_root = 1;
158 	return 0;
159 }
160 
161 __uml_setup("root=", uml_root_setup,
162 "root=<file containing the root fs>\n"
163 "    This is actually used by the generic kernel in exactly the same\n"
164 "    way as in any other kernel. If you configure a number of block\n"
165 "    devices and want to boot off something other than ubd0, you \n"
166 "    would use something like:\n"
167 "        root=/dev/ubd5\n\n"
168 );
169 
170 static int __init uml_console_setup(char *line, int *add)
171 {
172 	have_console = 1;
173 	return 0;
174 }
175 
176 __uml_setup("console=", uml_console_setup,
177 "console=<preferred console>\n"
178 "    Specify the preferred console output driver\n\n"
179 );
180 
181 static int __init Usage(char *line, int *add)
182 {
183 	const char **p;
184 
185 	printf(usage_string, init_utsname()->release);
186 	p = &__uml_help_start;
187 	/* Explicitly use printf() to show help in stdout */
188 	while (p < &__uml_help_end) {
189 		printf("%s", *p);
190 		p++;
191 	}
192 	exit(0);
193 	return 0;
194 }
195 
196 __uml_setup("--help", Usage,
197 "--help\n"
198 "    Prints this message.\n\n"
199 );
200 
201 static void __init uml_checksetup(char *line, int *add)
202 {
203 	struct uml_param *p;
204 
205 	p = &__uml_setup_start;
206 	while (p < &__uml_setup_end) {
207 		size_t n;
208 
209 		n = strlen(p->str);
210 		if (!strncmp(line, p->str, n) && p->setup_func(line + n, add))
211 			return;
212 		p++;
213 	}
214 }
215 
216 static void __init uml_postsetup(void)
217 {
218 	initcall_t *p;
219 
220 	p = &__uml_postsetup_start;
221 	while (p < &__uml_postsetup_end) {
222 		(*p)();
223 		p++;
224 	}
225 	return;
226 }
227 
228 static int panic_exit(struct notifier_block *self, unsigned long unused1,
229 		      void *unused2)
230 {
231 	kmsg_dump(KMSG_DUMP_PANIC);
232 	bust_spinlocks(1);
233 	bust_spinlocks(0);
234 	uml_exitcode = 1;
235 	os_dump_core();
236 
237 	return NOTIFY_DONE;
238 }
239 
240 static struct notifier_block panic_exit_notifier = {
241 	.notifier_call	= panic_exit,
242 	.priority	= INT_MAX - 1, /* run as 2nd notifier, won't return */
243 };
244 
245 void uml_finishsetup(void)
246 {
247 	atomic_notifier_chain_register(&panic_notifier_list,
248 				       &panic_exit_notifier);
249 
250 	uml_postsetup();
251 
252 	new_thread_handler();
253 }
254 
255 /* Set during early boot */
256 unsigned long stub_start;
257 unsigned long task_size;
258 EXPORT_SYMBOL(task_size);
259 
260 unsigned long host_task_size;
261 
262 unsigned long brk_start;
263 unsigned long end_iomem;
264 EXPORT_SYMBOL(end_iomem);
265 
266 #define MIN_VMALLOC (32 * 1024 * 1024)
267 
268 static void parse_host_cpu_flags(char *line)
269 {
270 	int i;
271 	for (i = 0; i < 32*NCAPINTS; i++) {
272 		if ((x86_cap_flags[i] != NULL) && strstr(line, x86_cap_flags[i]))
273 			set_cpu_cap(&boot_cpu_data, i);
274 	}
275 }
276 static void parse_cache_line(char *line)
277 {
278 	long res;
279 	char *to_parse = strstr(line, ":");
280 	if (to_parse) {
281 		to_parse++;
282 		while (*to_parse != 0 && isspace(*to_parse)) {
283 			to_parse++;
284 		}
285 		if (kstrtoul(to_parse, 10, &res) == 0 && is_power_of_2(res))
286 			boot_cpu_data.cache_alignment = res;
287 		else
288 			boot_cpu_data.cache_alignment = L1_CACHE_BYTES;
289 	}
290 }
291 
292 static unsigned long get_top_address(char **envp)
293 {
294 	unsigned long top_addr = (unsigned long) &top_addr;
295 	int i;
296 
297 	/* The earliest variable should be after the program name in ELF */
298 	for (i = 0; envp[i]; i++) {
299 		if ((unsigned long) envp[i] > top_addr)
300 			top_addr = (unsigned long) envp[i];
301 	}
302 
303 	top_addr &= ~(UM_KERN_PAGE_SIZE - 1);
304 	top_addr += UM_KERN_PAGE_SIZE;
305 
306 	return top_addr;
307 }
308 
309 int __init linux_main(int argc, char **argv, char **envp)
310 {
311 	unsigned long avail, diff;
312 	unsigned long virtmem_size, max_physmem;
313 	unsigned long stack;
314 	unsigned int i;
315 	int add;
316 
317 	for (i = 1; i < argc; i++) {
318 		if ((i == 1) && (argv[i][0] == ' '))
319 			continue;
320 		add = 1;
321 		uml_checksetup(argv[i], &add);
322 		if (add)
323 			add_arg(argv[i]);
324 	}
325 	if (have_root == 0)
326 		add_arg(DEFAULT_COMMAND_LINE_ROOT);
327 
328 	if (have_console == 0)
329 		add_arg(DEFAULT_COMMAND_LINE_CONSOLE);
330 
331 	host_task_size = get_top_address(envp);
332 	/* reserve a few pages for the stubs */
333 	stub_start = host_task_size - STUB_DATA_PAGES * PAGE_SIZE;
334 	/* another page for the code portion */
335 	stub_start -= PAGE_SIZE;
336 	host_task_size = stub_start;
337 
338 	/* Limit TASK_SIZE to what is addressable by the page table */
339 	task_size = host_task_size;
340 	if (task_size > (unsigned long long) PTRS_PER_PGD * PGDIR_SIZE)
341 		task_size = PTRS_PER_PGD * PGDIR_SIZE;
342 
343 	/*
344 	 * TASK_SIZE needs to be PGDIR_SIZE aligned or else exit_mmap craps
345 	 * out
346 	 */
347 	task_size = task_size & PGDIR_MASK;
348 
349 	/* OS sanity checks that need to happen before the kernel runs */
350 	os_early_checks();
351 
352 	get_host_cpu_features(parse_host_cpu_flags, parse_cache_line);
353 
354 	brk_start = (unsigned long) sbrk(0);
355 
356 	/*
357 	 * Increase physical memory size for exec-shield users
358 	 * so they actually get what they asked for. This should
359 	 * add zero for non-exec shield users
360 	 */
361 
362 	diff = UML_ROUND_UP(brk_start) - UML_ROUND_UP(&_end);
363 	if (diff > 1024 * 1024) {
364 		os_info("Adding %ld bytes to physical memory to account for "
365 			"exec-shield gap\n", diff);
366 		physmem_size += UML_ROUND_UP(brk_start) - UML_ROUND_UP(&_end);
367 	}
368 
369 	uml_physmem = (unsigned long) __binary_start & PAGE_MASK;
370 
371 	/* Reserve up to 4M after the current brk */
372 	uml_reserved = ROUND_4M(brk_start) + (1 << 22);
373 
374 	setup_machinename(init_utsname()->machine);
375 
376 	physmem_size = (physmem_size + PAGE_SIZE - 1) & PAGE_MASK;
377 	iomem_size = (iomem_size + PAGE_SIZE - 1) & PAGE_MASK;
378 
379 	max_physmem = TASK_SIZE - uml_physmem - iomem_size - MIN_VMALLOC;
380 
381 	if (physmem_size + iomem_size > max_physmem) {
382 		physmem_size = max_physmem - iomem_size;
383 		os_info("Physical memory size shrunk to %llu bytes\n",
384 			physmem_size);
385 	}
386 
387 	high_physmem = uml_physmem + physmem_size;
388 	end_iomem = high_physmem + iomem_size;
389 	high_memory = (void *) end_iomem;
390 
391 	start_vm = VMALLOC_START;
392 
393 	virtmem_size = physmem_size;
394 	stack = (unsigned long) argv;
395 	stack &= ~(1024 * 1024 - 1);
396 	avail = stack - start_vm;
397 	if (physmem_size > avail)
398 		virtmem_size = avail;
399 	end_vm = start_vm + virtmem_size;
400 
401 	if (virtmem_size < physmem_size)
402 		os_info("Kernel virtual memory size shrunk to %lu bytes\n",
403 			virtmem_size);
404 
405 	os_flush_stdout();
406 
407 	return start_uml();
408 }
409 
410 int __init __weak read_initrd(void)
411 {
412 	return 0;
413 }
414 
415 void __init setup_arch(char **cmdline_p)
416 {
417 	u8 rng_seed[32];
418 
419 	stack_protections((unsigned long) &init_thread_info);
420 	setup_physmem(uml_physmem, uml_reserved, physmem_size);
421 	mem_total_pages(physmem_size, iomem_size);
422 	uml_dtb_init();
423 	read_initrd();
424 
425 	paging_init();
426 	strscpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
427 	*cmdline_p = command_line;
428 	setup_hostinfo(host_info, sizeof host_info);
429 
430 	if (os_getrandom(rng_seed, sizeof(rng_seed), 0) == sizeof(rng_seed)) {
431 		add_bootloader_randomness(rng_seed, sizeof(rng_seed));
432 		memzero_explicit(rng_seed, sizeof(rng_seed));
433 	}
434 }
435 
436 void __init arch_cpu_finalize_init(void)
437 {
438 	arch_check_bugs();
439 	os_check_bugs();
440 }
441 
442 void apply_seal_endbr(s32 *start, s32 *end)
443 {
444 }
445 
446 void apply_retpolines(s32 *start, s32 *end)
447 {
448 }
449 
450 void apply_returns(s32 *start, s32 *end)
451 {
452 }
453 
454 void apply_fineibt(s32 *start_retpoline, s32 *end_retpoline,
455 		   s32 *start_cfi, s32 *end_cfi)
456 {
457 }
458 
459 void apply_alternatives(struct alt_instr *start, struct alt_instr *end)
460 {
461 }
462 
463 void *text_poke(void *addr, const void *opcode, size_t len)
464 {
465 	/*
466 	 * In UML, the only reference to this function is in
467 	 * apply_relocate_add(), which shouldn't ever actually call this
468 	 * because UML doesn't have live patching.
469 	 */
470 	WARN_ON(1);
471 
472 	return memcpy(addr, opcode, len);
473 }
474 
475 void text_poke_sync(void)
476 {
477 }
478 
479 void uml_pm_wake(void)
480 {
481 	pm_system_wakeup();
482 }
483 
484 #ifdef CONFIG_PM_SLEEP
485 static int um_suspend_valid(suspend_state_t state)
486 {
487 	return state == PM_SUSPEND_MEM;
488 }
489 
490 static int um_suspend_prepare(void)
491 {
492 	um_irqs_suspend();
493 	return 0;
494 }
495 
496 static int um_suspend_enter(suspend_state_t state)
497 {
498 	if (WARN_ON(state != PM_SUSPEND_MEM))
499 		return -EINVAL;
500 
501 	/*
502 	 * This is identical to the idle sleep, but we've just
503 	 * (during suspend) turned off all interrupt sources
504 	 * except for the ones we want, so now we can only wake
505 	 * up on something we actually want to wake up on. All
506 	 * timing has also been suspended.
507 	 */
508 	um_idle_sleep();
509 	return 0;
510 }
511 
512 static void um_suspend_finish(void)
513 {
514 	um_irqs_resume();
515 }
516 
517 const struct platform_suspend_ops um_suspend_ops = {
518 	.valid = um_suspend_valid,
519 	.prepare = um_suspend_prepare,
520 	.enter = um_suspend_enter,
521 	.finish = um_suspend_finish,
522 };
523 
524 static int init_pm_wake_signal(void)
525 {
526 	/*
527 	 * In external time-travel mode we can't use signals to wake up
528 	 * since that would mess with the scheduling. We'll have to do
529 	 * some additional work to support wakeup on virtio devices or
530 	 * similar, perhaps implementing a fake RTC controller that can
531 	 * trigger wakeup (and request the appropriate scheduling from
532 	 * the external scheduler when going to suspend.)
533 	 */
534 	if (time_travel_mode != TT_MODE_EXTERNAL)
535 		register_pm_wake_signal();
536 
537 	suspend_set_ops(&um_suspend_ops);
538 
539 	return 0;
540 }
541 
542 late_initcall(init_pm_wake_signal);
543 #endif
544