xref: /openbmc/linux/arch/um/kernel/um_arch.c (revision be0fffa5)
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 <as-layout.h>
27 #include <arch.h>
28 #include <init.h>
29 #include <kern.h>
30 #include <kern_util.h>
31 #include <mem_user.h>
32 #include <os.h>
33 
34 #include "um_arch.h"
35 
36 #define DEFAULT_COMMAND_LINE_ROOT "root=98:0"
37 #define DEFAULT_COMMAND_LINE_CONSOLE "console=tty0"
38 
39 /* Changed in add_arg and setup_arch, which run before SMP is started */
40 static char __initdata command_line[COMMAND_LINE_SIZE] = { 0 };
41 
add_arg(char * arg)42 static void __init add_arg(char *arg)
43 {
44 	if (strlen(command_line) + strlen(arg) + 1 > COMMAND_LINE_SIZE) {
45 		os_warn("add_arg: Too many command line arguments!\n");
46 		exit(1);
47 	}
48 	if (strlen(command_line) > 0)
49 		strcat(command_line, " ");
50 	strcat(command_line, arg);
51 }
52 
53 /*
54  * These fields are initialized at boot time and not changed.
55  * XXX This structure is used only in the non-SMP case.  Maybe this
56  * should be moved to smp.c.
57  */
58 struct cpuinfo_um boot_cpu_data = {
59 	.loops_per_jiffy	= 0,
60 	.ipi_pipe		= { -1, -1 },
61 	.cache_alignment	= L1_CACHE_BYTES,
62 	.x86_capability		= { 0 }
63 };
64 
65 EXPORT_SYMBOL(boot_cpu_data);
66 
67 union thread_union cpu0_irqstack
68 	__section(".data..init_irqstack") =
69 		{ .thread_info = INIT_THREAD_INFO(init_task) };
70 
71 /* Changed in setup_arch, which is called in early boot */
72 static char host_info[(__NEW_UTS_LEN + 1) * 5];
73 
show_cpuinfo(struct seq_file * m,void * v)74 static int show_cpuinfo(struct seq_file *m, void *v)
75 {
76 	int i = 0;
77 
78 	seq_printf(m, "processor\t: %d\n", i);
79 	seq_printf(m, "vendor_id\t: User Mode Linux\n");
80 	seq_printf(m, "model name\t: UML\n");
81 	seq_printf(m, "mode\t\t: skas\n");
82 	seq_printf(m, "host\t\t: %s\n", host_info);
83 	seq_printf(m, "fpu\t\t: %s\n", cpu_has(&boot_cpu_data, X86_FEATURE_FPU) ? "yes" : "no");
84 	seq_printf(m, "flags\t\t:");
85 	for (i = 0; i < 32*NCAPINTS; i++)
86 		if (cpu_has(&boot_cpu_data, i) && (x86_cap_flags[i] != NULL))
87 			seq_printf(m, " %s", x86_cap_flags[i]);
88 	seq_printf(m, "\n");
89 	seq_printf(m, "cache_alignment\t: %d\n", boot_cpu_data.cache_alignment);
90 	seq_printf(m, "bogomips\t: %lu.%02lu\n",
91 		   loops_per_jiffy/(500000/HZ),
92 		   (loops_per_jiffy/(5000/HZ)) % 100);
93 
94 
95 	return 0;
96 }
97 
c_start(struct seq_file * m,loff_t * pos)98 static void *c_start(struct seq_file *m, loff_t *pos)
99 {
100 	return *pos < nr_cpu_ids ? &boot_cpu_data + *pos : NULL;
101 }
102 
c_next(struct seq_file * m,void * v,loff_t * pos)103 static void *c_next(struct seq_file *m, void *v, loff_t *pos)
104 {
105 	++*pos;
106 	return c_start(m, pos);
107 }
108 
c_stop(struct seq_file * m,void * v)109 static void c_stop(struct seq_file *m, void *v)
110 {
111 }
112 
113 const struct seq_operations cpuinfo_op = {
114 	.start	= c_start,
115 	.next	= c_next,
116 	.stop	= c_stop,
117 	.show	= show_cpuinfo,
118 };
119 
120 /* Set in linux_main */
121 unsigned long uml_physmem;
122 EXPORT_SYMBOL(uml_physmem);
123 
124 unsigned long uml_reserved; /* Also modified in mem_init */
125 unsigned long start_vm;
126 unsigned long end_vm;
127 
128 /* Set in uml_ncpus_setup */
129 int ncpus = 1;
130 
131 /* Set in early boot */
132 static int have_root __initdata;
133 static int have_console __initdata;
134 
135 /* Set in uml_mem_setup and modified in linux_main */
136 long long physmem_size = 64 * 1024 * 1024;
137 EXPORT_SYMBOL(physmem_size);
138 
139 static const char *usage_string =
140 "User Mode Linux v%s\n"
141 "	available at http://user-mode-linux.sourceforge.net/\n\n";
142 
uml_version_setup(char * line,int * add)143 static int __init uml_version_setup(char *line, int *add)
144 {
145 	/* Explicitly use printf() to show version in stdout */
146 	printf("%s\n", init_utsname()->release);
147 	exit(0);
148 
149 	return 0;
150 }
151 
152 __uml_setup("--version", uml_version_setup,
153 "--version\n"
154 "    Prints the version number of the kernel.\n\n"
155 );
156 
uml_root_setup(char * line,int * add)157 static int __init uml_root_setup(char *line, int *add)
158 {
159 	have_root = 1;
160 	return 0;
161 }
162 
163 __uml_setup("root=", uml_root_setup,
164 "root=<file containing the root fs>\n"
165 "    This is actually used by the generic kernel in exactly the same\n"
166 "    way as in any other kernel. If you configure a number of block\n"
167 "    devices and want to boot off something other than ubd0, you \n"
168 "    would use something like:\n"
169 "        root=/dev/ubd5\n\n"
170 );
171 
no_skas_debug_setup(char * line,int * add)172 static int __init no_skas_debug_setup(char *line, int *add)
173 {
174 	os_warn("'debug' is not necessary to gdb UML in skas mode - run\n");
175 	os_warn("'gdb linux'\n");
176 
177 	return 0;
178 }
179 
180 __uml_setup("debug", no_skas_debug_setup,
181 "debug\n"
182 "    this flag is not needed to run gdb on UML in skas mode\n\n"
183 );
184 
uml_console_setup(char * line,int * add)185 static int __init uml_console_setup(char *line, int *add)
186 {
187 	have_console = 1;
188 	return 0;
189 }
190 
191 __uml_setup("console=", uml_console_setup,
192 "console=<preferred console>\n"
193 "    Specify the preferred console output driver\n\n"
194 );
195 
Usage(char * line,int * add)196 static int __init Usage(char *line, int *add)
197 {
198 	const char **p;
199 
200 	printf(usage_string, init_utsname()->release);
201 	p = &__uml_help_start;
202 	/* Explicitly use printf() to show help in stdout */
203 	while (p < &__uml_help_end) {
204 		printf("%s", *p);
205 		p++;
206 	}
207 	exit(0);
208 	return 0;
209 }
210 
211 __uml_setup("--help", Usage,
212 "--help\n"
213 "    Prints this message.\n\n"
214 );
215 
uml_checksetup(char * line,int * add)216 static void __init uml_checksetup(char *line, int *add)
217 {
218 	struct uml_param *p;
219 
220 	p = &__uml_setup_start;
221 	while (p < &__uml_setup_end) {
222 		size_t n;
223 
224 		n = strlen(p->str);
225 		if (!strncmp(line, p->str, n) && p->setup_func(line + n, add))
226 			return;
227 		p++;
228 	}
229 }
230 
uml_postsetup(void)231 static void __init uml_postsetup(void)
232 {
233 	initcall_t *p;
234 
235 	p = &__uml_postsetup_start;
236 	while (p < &__uml_postsetup_end) {
237 		(*p)();
238 		p++;
239 	}
240 	return;
241 }
242 
panic_exit(struct notifier_block * self,unsigned long unused1,void * unused2)243 static int panic_exit(struct notifier_block *self, unsigned long unused1,
244 		      void *unused2)
245 {
246 	kmsg_dump(KMSG_DUMP_PANIC);
247 	bust_spinlocks(1);
248 	bust_spinlocks(0);
249 	uml_exitcode = 1;
250 	os_dump_core();
251 
252 	return NOTIFY_DONE;
253 }
254 
255 static struct notifier_block panic_exit_notifier = {
256 	.notifier_call	= panic_exit,
257 	.priority	= INT_MAX - 1, /* run as 2nd notifier, won't return */
258 };
259 
uml_finishsetup(void)260 void uml_finishsetup(void)
261 {
262 	atomic_notifier_chain_register(&panic_notifier_list,
263 				       &panic_exit_notifier);
264 
265 	uml_postsetup();
266 
267 	new_thread_handler();
268 }
269 
270 /* Set during early boot */
271 unsigned long stub_start;
272 unsigned long task_size;
273 EXPORT_SYMBOL(task_size);
274 
275 unsigned long host_task_size;
276 
277 unsigned long brk_start;
278 unsigned long end_iomem;
279 EXPORT_SYMBOL(end_iomem);
280 
281 #define MIN_VMALLOC (32 * 1024 * 1024)
282 
parse_host_cpu_flags(char * line)283 static void parse_host_cpu_flags(char *line)
284 {
285 	int i;
286 	for (i = 0; i < 32*NCAPINTS; i++) {
287 		if ((x86_cap_flags[i] != NULL) && strstr(line, x86_cap_flags[i]))
288 			set_cpu_cap(&boot_cpu_data, i);
289 	}
290 }
parse_cache_line(char * line)291 static void parse_cache_line(char *line)
292 {
293 	long res;
294 	char *to_parse = strstr(line, ":");
295 	if (to_parse) {
296 		to_parse++;
297 		while (*to_parse != 0 && isspace(*to_parse)) {
298 			to_parse++;
299 		}
300 		if (kstrtoul(to_parse, 10, &res) == 0 && is_power_of_2(res))
301 			boot_cpu_data.cache_alignment = res;
302 		else
303 			boot_cpu_data.cache_alignment = L1_CACHE_BYTES;
304 	}
305 }
306 
linux_main(int argc,char ** argv)307 int __init linux_main(int argc, char **argv)
308 {
309 	unsigned long avail, diff;
310 	unsigned long virtmem_size, max_physmem;
311 	unsigned long stack;
312 	unsigned int i;
313 	int add;
314 
315 	for (i = 1; i < argc; i++) {
316 		if ((i == 1) && (argv[i][0] == ' '))
317 			continue;
318 		add = 1;
319 		uml_checksetup(argv[i], &add);
320 		if (add)
321 			add_arg(argv[i]);
322 	}
323 	if (have_root == 0)
324 		add_arg(DEFAULT_COMMAND_LINE_ROOT);
325 
326 	if (have_console == 0)
327 		add_arg(DEFAULT_COMMAND_LINE_CONSOLE);
328 
329 	host_task_size = os_get_top_address();
330 	/* reserve a few pages for the stubs (taking care of data alignment) */
331 	/* align the data portion */
332 	BUILD_BUG_ON(!is_power_of_2(STUB_DATA_PAGES));
333 	stub_start = (host_task_size - 1) & ~(STUB_DATA_PAGES * PAGE_SIZE - 1);
334 	/* another page for the code portion */
335 	stub_start -= PAGE_SIZE;
336 	host_task_size = stub_start;
337 
338 	/*
339 	 * TASK_SIZE needs to be PGDIR_SIZE aligned or else exit_mmap craps
340 	 * out
341 	 */
342 	task_size = host_task_size & PGDIR_MASK;
343 
344 	/* OS sanity checks that need to happen before the kernel runs */
345 	os_early_checks();
346 
347 	get_host_cpu_features(parse_host_cpu_flags, parse_cache_line);
348 
349 	brk_start = (unsigned long) sbrk(0);
350 
351 	/*
352 	 * Increase physical memory size for exec-shield users
353 	 * so they actually get what they asked for. This should
354 	 * add zero for non-exec shield users
355 	 */
356 
357 	diff = UML_ROUND_UP(brk_start) - UML_ROUND_UP(&_end);
358 	if (diff > 1024 * 1024) {
359 		os_info("Adding %ld bytes to physical memory to account for "
360 			"exec-shield gap\n", diff);
361 		physmem_size += UML_ROUND_UP(brk_start) - UML_ROUND_UP(&_end);
362 	}
363 
364 	uml_physmem = (unsigned long) __binary_start & PAGE_MASK;
365 
366 	/* Reserve up to 4M after the current brk */
367 	uml_reserved = ROUND_4M(brk_start) + (1 << 22);
368 
369 	setup_machinename(init_utsname()->machine);
370 
371 	highmem = 0;
372 	iomem_size = (iomem_size + PAGE_SIZE - 1) & PAGE_MASK;
373 	max_physmem = TASK_SIZE - uml_physmem - iomem_size - MIN_VMALLOC;
374 
375 	/*
376 	 * Zones have to begin on a 1 << MAX_ORDER page boundary,
377 	 * so this makes sure that's true for highmem
378 	 */
379 	max_physmem &= ~((1 << (PAGE_SHIFT + MAX_ORDER)) - 1);
380 	if (physmem_size + iomem_size > max_physmem) {
381 		highmem = physmem_size + iomem_size - max_physmem;
382 		physmem_size -= highmem;
383 	}
384 
385 	high_physmem = uml_physmem + physmem_size;
386 	end_iomem = high_physmem + iomem_size;
387 	high_memory = (void *) end_iomem;
388 
389 	start_vm = VMALLOC_START;
390 
391 	virtmem_size = physmem_size;
392 	stack = (unsigned long) argv;
393 	stack &= ~(1024 * 1024 - 1);
394 	avail = stack - start_vm;
395 	if (physmem_size > avail)
396 		virtmem_size = avail;
397 	end_vm = start_vm + virtmem_size;
398 
399 	if (virtmem_size < physmem_size)
400 		os_info("Kernel virtual memory size shrunk to %lu bytes\n",
401 			virtmem_size);
402 
403 	os_flush_stdout();
404 
405 	return start_uml();
406 }
407 
read_initrd(void)408 int __init __weak read_initrd(void)
409 {
410 	return 0;
411 }
412 
setup_arch(char ** cmdline_p)413 void __init setup_arch(char **cmdline_p)
414 {
415 	u8 rng_seed[32];
416 
417 	stack_protections((unsigned long) &init_thread_info);
418 	setup_physmem(uml_physmem, uml_reserved, physmem_size, highmem);
419 	mem_total_pages(physmem_size, iomem_size, highmem);
420 	uml_dtb_init();
421 	read_initrd();
422 
423 	paging_init();
424 	strscpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
425 	*cmdline_p = command_line;
426 	setup_hostinfo(host_info, sizeof host_info);
427 
428 	if (os_getrandom(rng_seed, sizeof(rng_seed), 0) == sizeof(rng_seed)) {
429 		add_bootloader_randomness(rng_seed, sizeof(rng_seed));
430 		memzero_explicit(rng_seed, sizeof(rng_seed));
431 	}
432 }
433 
arch_cpu_finalize_init(void)434 void __init arch_cpu_finalize_init(void)
435 {
436 	arch_check_bugs();
437 	os_check_bugs();
438 }
439 
apply_seal_endbr(s32 * start,s32 * end)440 void apply_seal_endbr(s32 *start, s32 *end)
441 {
442 }
443 
apply_retpolines(s32 * start,s32 * end)444 void apply_retpolines(s32 *start, s32 *end)
445 {
446 }
447 
apply_returns(s32 * start,s32 * end)448 void apply_returns(s32 *start, s32 *end)
449 {
450 }
451 
apply_fineibt(s32 * start_retpoline,s32 * end_retpoline,s32 * start_cfi,s32 * end_cfi)452 void apply_fineibt(s32 *start_retpoline, s32 *end_retpoline,
453 		   s32 *start_cfi, s32 *end_cfi)
454 {
455 }
456 
apply_alternatives(struct alt_instr * start,struct alt_instr * end)457 void apply_alternatives(struct alt_instr *start, struct alt_instr *end)
458 {
459 }
460 
text_poke(void * addr,const void * opcode,size_t len)461 void *text_poke(void *addr, const void *opcode, size_t len)
462 {
463 	/*
464 	 * In UML, the only reference to this function is in
465 	 * apply_relocate_add(), which shouldn't ever actually call this
466 	 * because UML doesn't have live patching.
467 	 */
468 	WARN_ON(1);
469 
470 	return memcpy(addr, opcode, len);
471 }
472 
text_poke_sync(void)473 void text_poke_sync(void)
474 {
475 }
476 
uml_pm_wake(void)477 void uml_pm_wake(void)
478 {
479 	pm_system_wakeup();
480 }
481 
482 #ifdef CONFIG_PM_SLEEP
um_suspend_valid(suspend_state_t state)483 static int um_suspend_valid(suspend_state_t state)
484 {
485 	return state == PM_SUSPEND_MEM;
486 }
487 
um_suspend_prepare(void)488 static int um_suspend_prepare(void)
489 {
490 	um_irqs_suspend();
491 	return 0;
492 }
493 
um_suspend_enter(suspend_state_t state)494 static int um_suspend_enter(suspend_state_t state)
495 {
496 	if (WARN_ON(state != PM_SUSPEND_MEM))
497 		return -EINVAL;
498 
499 	/*
500 	 * This is identical to the idle sleep, but we've just
501 	 * (during suspend) turned off all interrupt sources
502 	 * except for the ones we want, so now we can only wake
503 	 * up on something we actually want to wake up on. All
504 	 * timing has also been suspended.
505 	 */
506 	um_idle_sleep();
507 	return 0;
508 }
509 
um_suspend_finish(void)510 static void um_suspend_finish(void)
511 {
512 	um_irqs_resume();
513 }
514 
515 const struct platform_suspend_ops um_suspend_ops = {
516 	.valid = um_suspend_valid,
517 	.prepare = um_suspend_prepare,
518 	.enter = um_suspend_enter,
519 	.finish = um_suspend_finish,
520 };
521 
init_pm_wake_signal(void)522 static int init_pm_wake_signal(void)
523 {
524 	/*
525 	 * In external time-travel mode we can't use signals to wake up
526 	 * since that would mess with the scheduling. We'll have to do
527 	 * some additional work to support wakeup on virtio devices or
528 	 * similar, perhaps implementing a fake RTC controller that can
529 	 * trigger wakeup (and request the appropriate scheduling from
530 	 * the external scheduler when going to suspend.)
531 	 */
532 	if (time_travel_mode != TT_MODE_EXTERNAL)
533 		register_pm_wake_signal();
534 
535 	suspend_set_ops(&um_suspend_ops);
536 
537 	return 0;
538 }
539 
540 late_initcall(init_pm_wake_signal);
541 #endif
542