xref: /linux/arch/um/kernel/um_arch.c (revision bf36793fa260cb68cc817f311f1f683788261796)
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 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 no_skas_debug_setup(char *line, int *add)
171 {
172 	os_warn("'debug' is not necessary to gdb UML in skas mode - run\n");
173 	os_warn("'gdb linux'\n");
174 
175 	return 0;
176 }
177 
178 __uml_setup("debug", no_skas_debug_setup,
179 "debug\n"
180 "    this flag is not needed to run gdb on UML in skas mode\n\n"
181 );
182 
183 static int __init uml_console_setup(char *line, int *add)
184 {
185 	have_console = 1;
186 	return 0;
187 }
188 
189 __uml_setup("console=", uml_console_setup,
190 "console=<preferred console>\n"
191 "    Specify the preferred console output driver\n\n"
192 );
193 
194 static int __init Usage(char *line, int *add)
195 {
196 	const char **p;
197 
198 	printf(usage_string, init_utsname()->release);
199 	p = &__uml_help_start;
200 	/* Explicitly use printf() to show help in stdout */
201 	while (p < &__uml_help_end) {
202 		printf("%s", *p);
203 		p++;
204 	}
205 	exit(0);
206 	return 0;
207 }
208 
209 __uml_setup("--help", Usage,
210 "--help\n"
211 "    Prints this message.\n\n"
212 );
213 
214 static void __init uml_checksetup(char *line, int *add)
215 {
216 	struct uml_param *p;
217 
218 	p = &__uml_setup_start;
219 	while (p < &__uml_setup_end) {
220 		size_t n;
221 
222 		n = strlen(p->str);
223 		if (!strncmp(line, p->str, n) && p->setup_func(line + n, add))
224 			return;
225 		p++;
226 	}
227 }
228 
229 static void __init uml_postsetup(void)
230 {
231 	initcall_t *p;
232 
233 	p = &__uml_postsetup_start;
234 	while (p < &__uml_postsetup_end) {
235 		(*p)();
236 		p++;
237 	}
238 	return;
239 }
240 
241 static int panic_exit(struct notifier_block *self, unsigned long unused1,
242 		      void *unused2)
243 {
244 	kmsg_dump(KMSG_DUMP_PANIC);
245 	bust_spinlocks(1);
246 	bust_spinlocks(0);
247 	uml_exitcode = 1;
248 	os_dump_core();
249 
250 	return NOTIFY_DONE;
251 }
252 
253 static struct notifier_block panic_exit_notifier = {
254 	.notifier_call	= panic_exit,
255 	.priority	= INT_MAX - 1, /* run as 2nd notifier, won't return */
256 };
257 
258 void uml_finishsetup(void)
259 {
260 	atomic_notifier_chain_register(&panic_notifier_list,
261 				       &panic_exit_notifier);
262 
263 	uml_postsetup();
264 
265 	new_thread_handler();
266 }
267 
268 /* Set during early boot */
269 unsigned long stub_start;
270 unsigned long task_size;
271 EXPORT_SYMBOL(task_size);
272 
273 unsigned long host_task_size;
274 
275 unsigned long brk_start;
276 unsigned long end_iomem;
277 EXPORT_SYMBOL(end_iomem);
278 
279 #define MIN_VMALLOC (32 * 1024 * 1024)
280 
281 static void parse_host_cpu_flags(char *line)
282 {
283 	int i;
284 	for (i = 0; i < 32*NCAPINTS; i++) {
285 		if ((x86_cap_flags[i] != NULL) && strstr(line, x86_cap_flags[i]))
286 			set_cpu_cap(&boot_cpu_data, i);
287 	}
288 }
289 static void parse_cache_line(char *line)
290 {
291 	long res;
292 	char *to_parse = strstr(line, ":");
293 	if (to_parse) {
294 		to_parse++;
295 		while (*to_parse != 0 && isspace(*to_parse)) {
296 			to_parse++;
297 		}
298 		if (kstrtoul(to_parse, 10, &res) == 0 && is_power_of_2(res))
299 			boot_cpu_data.cache_alignment = res;
300 		else
301 			boot_cpu_data.cache_alignment = L1_CACHE_BYTES;
302 	}
303 }
304 
305 int __init linux_main(int argc, char **argv)
306 {
307 	unsigned long avail, diff;
308 	unsigned long virtmem_size, max_physmem;
309 	unsigned long stack;
310 	unsigned int i;
311 	int add;
312 
313 	for (i = 1; i < argc; i++) {
314 		if ((i == 1) && (argv[i][0] == ' '))
315 			continue;
316 		add = 1;
317 		uml_checksetup(argv[i], &add);
318 		if (add)
319 			add_arg(argv[i]);
320 	}
321 	if (have_root == 0)
322 		add_arg(DEFAULT_COMMAND_LINE_ROOT);
323 
324 	if (have_console == 0)
325 		add_arg(DEFAULT_COMMAND_LINE_CONSOLE);
326 
327 	host_task_size = os_get_top_address();
328 	/* reserve a few pages for the stubs (taking care of data alignment) */
329 	/* align the data portion */
330 	BUILD_BUG_ON(!is_power_of_2(STUB_DATA_PAGES));
331 	stub_start = (host_task_size - 1) & ~(STUB_DATA_PAGES * PAGE_SIZE - 1);
332 	/* another page for the code portion */
333 	stub_start -= PAGE_SIZE;
334 	host_task_size = stub_start;
335 
336 	/*
337 	 * TASK_SIZE needs to be PGDIR_SIZE aligned or else exit_mmap craps
338 	 * out
339 	 */
340 	task_size = host_task_size & PGDIR_MASK;
341 
342 	/* OS sanity checks that need to happen before the kernel runs */
343 	os_early_checks();
344 
345 	get_host_cpu_features(parse_host_cpu_flags, parse_cache_line);
346 
347 	brk_start = (unsigned long) sbrk(0);
348 
349 	/*
350 	 * Increase physical memory size for exec-shield users
351 	 * so they actually get what they asked for. This should
352 	 * add zero for non-exec shield users
353 	 */
354 
355 	diff = UML_ROUND_UP(brk_start) - UML_ROUND_UP(&_end);
356 	if (diff > 1024 * 1024) {
357 		os_info("Adding %ld bytes to physical memory to account for "
358 			"exec-shield gap\n", diff);
359 		physmem_size += UML_ROUND_UP(brk_start) - UML_ROUND_UP(&_end);
360 	}
361 
362 	uml_physmem = (unsigned long) __binary_start & PAGE_MASK;
363 
364 	/* Reserve up to 4M after the current brk */
365 	uml_reserved = ROUND_4M(brk_start) + (1 << 22);
366 
367 	setup_machinename(init_utsname()->machine);
368 
369 	highmem = 0;
370 	iomem_size = (iomem_size + PAGE_SIZE - 1) & PAGE_MASK;
371 	max_physmem = TASK_SIZE - uml_physmem - iomem_size - MIN_VMALLOC;
372 
373 	/*
374 	 * Zones have to begin on a 1 << MAX_PAGE_ORDER page boundary,
375 	 * so this makes sure that's true for highmem
376 	 */
377 	max_physmem &= ~((1 << (PAGE_SHIFT + MAX_PAGE_ORDER)) - 1);
378 	if (physmem_size + iomem_size > max_physmem) {
379 		highmem = physmem_size + iomem_size - max_physmem;
380 		physmem_size -= highmem;
381 	}
382 
383 	high_physmem = uml_physmem + physmem_size;
384 	end_iomem = high_physmem + iomem_size;
385 	high_memory = (void *) end_iomem;
386 
387 	start_vm = VMALLOC_START;
388 
389 	virtmem_size = physmem_size;
390 	stack = (unsigned long) argv;
391 	stack &= ~(1024 * 1024 - 1);
392 	avail = stack - start_vm;
393 	if (physmem_size > avail)
394 		virtmem_size = avail;
395 	end_vm = start_vm + virtmem_size;
396 
397 	if (virtmem_size < physmem_size)
398 		os_info("Kernel virtual memory size shrunk to %lu bytes\n",
399 			virtmem_size);
400 
401 	os_flush_stdout();
402 
403 	return start_uml();
404 }
405 
406 int __init __weak read_initrd(void)
407 {
408 	return 0;
409 }
410 
411 void __init setup_arch(char **cmdline_p)
412 {
413 	u8 rng_seed[32];
414 
415 	stack_protections((unsigned long) &init_thread_info);
416 	setup_physmem(uml_physmem, uml_reserved, physmem_size, highmem);
417 	mem_total_pages(physmem_size, iomem_size, highmem);
418 	uml_dtb_init();
419 	read_initrd();
420 
421 	paging_init();
422 	strscpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
423 	*cmdline_p = command_line;
424 	setup_hostinfo(host_info, sizeof host_info);
425 
426 	if (os_getrandom(rng_seed, sizeof(rng_seed), 0) == sizeof(rng_seed)) {
427 		add_bootloader_randomness(rng_seed, sizeof(rng_seed));
428 		memzero_explicit(rng_seed, sizeof(rng_seed));
429 	}
430 }
431 
432 void __init arch_cpu_finalize_init(void)
433 {
434 	arch_check_bugs();
435 	os_check_bugs();
436 }
437 
438 void apply_seal_endbr(s32 *start, s32 *end)
439 {
440 }
441 
442 void apply_retpolines(s32 *start, s32 *end)
443 {
444 }
445 
446 void apply_returns(s32 *start, s32 *end)
447 {
448 }
449 
450 void apply_fineibt(s32 *start_retpoline, s32 *end_retpoline,
451 		   s32 *start_cfi, s32 *end_cfi)
452 {
453 }
454 
455 void apply_alternatives(struct alt_instr *start, struct alt_instr *end)
456 {
457 }
458 
459 void *text_poke(void *addr, const void *opcode, size_t len)
460 {
461 	/*
462 	 * In UML, the only reference to this function is in
463 	 * apply_relocate_add(), which shouldn't ever actually call this
464 	 * because UML doesn't have live patching.
465 	 */
466 	WARN_ON(1);
467 
468 	return memcpy(addr, opcode, len);
469 }
470 
471 void text_poke_sync(void)
472 {
473 }
474 
475 void uml_pm_wake(void)
476 {
477 	pm_system_wakeup();
478 }
479 
480 #ifdef CONFIG_PM_SLEEP
481 static int um_suspend_valid(suspend_state_t state)
482 {
483 	return state == PM_SUSPEND_MEM;
484 }
485 
486 static int um_suspend_prepare(void)
487 {
488 	um_irqs_suspend();
489 	return 0;
490 }
491 
492 static int um_suspend_enter(suspend_state_t state)
493 {
494 	if (WARN_ON(state != PM_SUSPEND_MEM))
495 		return -EINVAL;
496 
497 	/*
498 	 * This is identical to the idle sleep, but we've just
499 	 * (during suspend) turned off all interrupt sources
500 	 * except for the ones we want, so now we can only wake
501 	 * up on something we actually want to wake up on. All
502 	 * timing has also been suspended.
503 	 */
504 	um_idle_sleep();
505 	return 0;
506 }
507 
508 static void um_suspend_finish(void)
509 {
510 	um_irqs_resume();
511 }
512 
513 const struct platform_suspend_ops um_suspend_ops = {
514 	.valid = um_suspend_valid,
515 	.prepare = um_suspend_prepare,
516 	.enter = um_suspend_enter,
517 	.finish = um_suspend_finish,
518 };
519 
520 static int init_pm_wake_signal(void)
521 {
522 	/*
523 	 * In external time-travel mode we can't use signals to wake up
524 	 * since that would mess with the scheduling. We'll have to do
525 	 * some additional work to support wakeup on virtio devices or
526 	 * similar, perhaps implementing a fake RTC controller that can
527 	 * trigger wakeup (and request the appropriate scheduling from
528 	 * the external scheduler when going to suspend.)
529 	 */
530 	if (time_travel_mode != TT_MODE_EXTERNAL)
531 		register_pm_wake_signal();
532 
533 	suspend_set_ops(&um_suspend_ops);
534 
535 	return 0;
536 }
537 
538 late_initcall(init_pm_wake_signal);
539 #endif
540