um_arch.c

  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/string_choices.h>
 16#include <linux/utsname.h>
 17#include <linux/sched.h>
 18#include <linux/sched/task.h>
 19#include <linux/kmsg_dump.h>
 20#include <linux/suspend.h>
 21#include <linux/random.h>
 22#include <linux/smp-internal.h>
 23
 24#include <asm/processor.h>
 25#include <asm/cpufeature.h>
 26#include <asm/sections.h>
 27#include <asm/setup.h>
 28#include <asm/text-patching.h>
 29#include <as-layout.h>
 30#include <arch.h>
 31#include <init.h>
 32#include <kern.h>
 33#include <kern_util.h>
 34#include <mem_user.h>
 35#include <os.h>
 36
 37#include "um_arch.h"
 38
 39#define DEFAULT_COMMAND_LINE_ROOT "root=98:0"
 40#define DEFAULT_COMMAND_LINE_CONSOLE "console=tty0"
 41
 42/* Changed in add_arg and setup_arch, which run before SMP is started */
 43static char __initdata command_line[COMMAND_LINE_SIZE] = { 0 };
 44
 45static void __init add_arg(char *arg)
 46{
 47	if (strlen(command_line) + strlen(arg) + 1 > COMMAND_LINE_SIZE) {
 48		os_warn("add_arg: Too many command line arguments!\n");
 49		exit(1);
 50	}
 51	if (strlen(command_line) > 0)
 52		strcat(command_line, " ");
 53	strcat(command_line, arg);
 54}
 55
 56/*
 57 * These fields are initialized at boot time and not changed.
 58 */
 59struct cpuinfo_um boot_cpu_data = {
 60	.loops_per_jiffy	= 0,
 61	.cache_alignment	= L1_CACHE_BYTES,
 62	.x86_capability		= { 0 }
 63};
 64
 65EXPORT_SYMBOL(boot_cpu_data);
 66
 67
 68/* Changed in setup_arch, which is called in early boot */
 69static char host_info[(__NEW_UTS_LEN + 1) * 5];
 70
 71static int show_cpuinfo(struct seq_file *m, void *v)
 72{
 73	int i = 0;
 74
 75#if IS_ENABLED(CONFIG_SMP)
 76	i = (uintptr_t) v - 1;
 77	if (!cpu_online(i))
 78		return 0;
 79#endif
 80
 81	seq_printf(m, "processor\t: %d\n", i);
 82	seq_printf(m, "vendor_id\t: User Mode Linux\n");
 83	seq_printf(m, "model name\t: UML\n");
 84	seq_printf(m, "mode\t\t: skas\n");
 85	seq_printf(m, "host\t\t: %s\n", host_info);
 86	seq_printf(m, "fpu\t\t: %s\n", str_yes_no(cpu_has(&boot_cpu_data, X86_FEATURE_FPU)));
 87	seq_printf(m, "flags\t\t:");
 88	for (i = 0; i < 32*NCAPINTS; i++)
 89		if (cpu_has(&boot_cpu_data, i) && (x86_cap_flags[i] != NULL))
 90			seq_printf(m, " %s", x86_cap_flags[i]);
 91	seq_printf(m, "\n");
 92	seq_printf(m, "cache_alignment\t: %d\n", boot_cpu_data.cache_alignment);
 93	seq_printf(m, "bogomips\t: %lu.%02lu\n",
 94		   loops_per_jiffy/(500000/HZ),
 95		   (loops_per_jiffy/(5000/HZ)) % 100);
 96
 97	return 0;
 98}
 99
100static void *c_start(struct seq_file *m, loff_t *pos)
101{
102	if (*pos < nr_cpu_ids)
103		return (void *)(uintptr_t)(*pos + 1);
104	return NULL;
105}
106
107static void *c_next(struct seq_file *m, void *v, loff_t *pos)
108{
109	++*pos;
110	return c_start(m, pos);
111}
112
113static void c_stop(struct seq_file *m, void *v)
114{
115}
116
117const struct seq_operations cpuinfo_op = {
118	.start	= c_start,
119	.next	= c_next,
120	.stop	= c_stop,
121	.show	= show_cpuinfo,
122};
123
124/* Set in linux_main */
125unsigned long uml_physmem;
126EXPORT_SYMBOL(uml_physmem);
127
128unsigned long uml_reserved; /* Also modified in mem_init */
129unsigned long start_vm;
130unsigned long end_vm;
131
132/* Set in early boot */
133static int have_root __initdata;
134static int have_console __initdata;
135
136/* Set in uml_mem_setup and modified in linux_main */
137unsigned long long physmem_size = 64 * 1024 * 1024;
138EXPORT_SYMBOL(physmem_size);
139
140static const char *usage_string =
141"User Mode Linux v%s\n"
142"	available at http://user-mode-linux.sourceforge.net/\n\n";
143
144static int __init uml_version_setup(char *line, int *add)
145{
146	/* Explicitly use printf() to show version in stdout */
147	printf("%s\n", init_utsname()->release);
148	exit(0);
149
150	return 0;
151}
152
153__uml_setup("--version", uml_version_setup,
154"--version\n"
155"    Prints the version number of the kernel.\n\n"
156);
157
158static int __init uml_root_setup(char *line, int *add)
159{
160	have_root = 1;
161	return 0;
162}
163
164__uml_setup("root=", uml_root_setup,
165"root=<file containing the root fs>\n"
166"    This is actually used by the generic kernel in exactly the same\n"
167"    way as in any other kernel. If you configure a number of block\n"
168"    devices and want to boot off something other than ubd0, you \n"
169"    would use something like:\n"
170"        root=/dev/ubd5\n\n"
171);
172
173static int __init uml_console_setup(char *line, int *add)
174{
175	have_console = 1;
176	return 0;
177}
178
179__uml_setup("console=", uml_console_setup,
180"console=<preferred console>\n"
181"    Specify the preferred console output driver\n\n"
182);
183
184static int __init Usage(char *line, int *add)
185{
186	const char **p;
187
188	printf(usage_string, init_utsname()->release);
189	p = &__uml_help_start;
190	/* Explicitly use printf() to show help in stdout */
191	while (p < &__uml_help_end) {
192		printf("%s", *p);
193		p++;
194	}
195	exit(0);
196	return 0;
197}
198
199__uml_setup("--help", Usage,
200"--help\n"
201"    Prints this message.\n\n"
202);
203
204static void __init uml_checksetup(char *line, int *add)
205{
206	struct uml_param *p;
207
208	p = &__uml_setup_start;
209	while (p < &__uml_setup_end) {
210		size_t n;
211
212		n = strlen(p->str);
213		if (!strncmp(line, p->str, n) && p->setup_func(line + n, add))
214			return;
215		p++;
216	}
217}
218
219static void __init uml_postsetup(void)
220{
221	initcall_t *p;
222
223	p = &__uml_postsetup_start;
224	while (p < &__uml_postsetup_end) {
225		(*p)();
226		p++;
227	}
228	return;
229}
230
231static int panic_exit(struct notifier_block *self, unsigned long unused1,
232		      void *unused2)
233{
234	kmsg_dump(KMSG_DUMP_PANIC);
235	bust_spinlocks(1);
236	bust_spinlocks(0);
237	uml_exitcode = 1;
238	os_dump_core();
239
240	return NOTIFY_DONE;
241}
242
243static struct notifier_block panic_exit_notifier = {
244	.notifier_call	= panic_exit,
245	.priority	= INT_MAX - 1, /* run as 2nd notifier, won't return */
246};
247
248void uml_finishsetup(void)
249{
250	atomic_notifier_chain_register(&panic_notifier_list,
251				       &panic_exit_notifier);
252
253	uml_postsetup();
254
255	new_thread_handler();
256}
257
258/* Set during early boot */
259unsigned long stub_start;
260unsigned long task_size;
261EXPORT_SYMBOL(task_size);
262
263unsigned long brk_start;
264
265#define MIN_VMALLOC (32 * 1024 * 1024)
266
267static void __init parse_host_cpu_flags(char *line)
268{
269	int i;
270	for (i = 0; i < 32*NCAPINTS; i++) {
271		if ((x86_cap_flags[i] != NULL) && strstr(line, x86_cap_flags[i]))
272			set_cpu_cap(&boot_cpu_data, i);
273	}
274}
275
276static void __init 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
292static unsigned long __init 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	return PAGE_ALIGN(top_addr + 1);
304}
305
306int __init linux_main(int argc, char **argv, char **envp)
307{
308	unsigned long avail, diff;
309	unsigned long virtmem_size, max_physmem;
310	unsigned long host_task_size;
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 = get_top_address(envp);
330	/* reserve a few pages for the stubs */
331	stub_start = host_task_size - STUB_SIZE;
332	host_task_size = stub_start;
333
334	/* Limit TASK_SIZE to what is addressable by the page table */
335	task_size = host_task_size;
336	if (task_size > (unsigned long long) PTRS_PER_PGD * PGDIR_SIZE)
337		task_size = PTRS_PER_PGD * PGDIR_SIZE;
338
339	/*
340	 * TASK_SIZE needs to be PGDIR_SIZE aligned or else exit_mmap craps
341	 * out
342	 */
343	task_size = task_size & PGDIR_MASK;
344
345	/* OS sanity checks that need to happen before the kernel runs */
346	os_early_checks();
347
348	get_host_cpu_features(parse_host_cpu_flags, parse_cache_line);
349
350	brk_start = (unsigned long) sbrk(0);
351
352	/*
353	 * Increase physical memory size for exec-shield users
354	 * so they actually get what they asked for. This should
355	 * add zero for non-exec shield users
356	 */
357	diff = PAGE_ALIGN(brk_start) - PAGE_ALIGN((unsigned long) &_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 += diff;
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	physmem_size = PAGE_ALIGN(physmem_size);
372	max_physmem = TASK_SIZE - uml_physmem - MIN_VMALLOC;
373	if (physmem_size > max_physmem) {
374		physmem_size = max_physmem;
375		os_info("Physical memory size shrunk to %llu bytes\n",
376			physmem_size);
377	}
378
379	high_physmem = uml_physmem + physmem_size;
380
381	start_vm = VMALLOC_START;
382
383	virtmem_size = physmem_size;
384	stack = (unsigned long) argv;
385	stack &= ~(1024 * 1024 - 1);
386	avail = stack - start_vm;
387	if (physmem_size > avail)
388		virtmem_size = avail;
389	end_vm = start_vm + virtmem_size;
390
391	if (virtmem_size < physmem_size)
392		os_info("Kernel virtual memory size shrunk to %lu bytes\n",
393			virtmem_size);
394
395	arch_task_struct_size = sizeof(struct task_struct) + host_fp_size;
396
397	os_flush_stdout();
398
399	return start_uml();
400}
401
402int __init __weak read_initrd(void)
403{
404	return 0;
405}
406
407void __init setup_arch(char **cmdline_p)
408{
409	u8 rng_seed[32];
410
411	stack_protections((unsigned long) init_task.stack);
412	setup_physmem(uml_physmem, uml_reserved, physmem_size);
413	uml_dtb_init();
414	read_initrd();
415
416	paging_init();
417	strscpy(boot_command_line, command_line, COMMAND_LINE_SIZE);
418	*cmdline_p = command_line;
419	setup_hostinfo(host_info, sizeof host_info);
420	prefill_possible_map();
421
422	if (os_getrandom(rng_seed, sizeof(rng_seed), 0) == sizeof(rng_seed)) {
423		add_bootloader_randomness(rng_seed, sizeof(rng_seed));
424		memzero_explicit(rng_seed, sizeof(rng_seed));
425	}
426}
427
428void __init arch_cpu_finalize_init(void)
429{
430	arch_check_bugs();
431	os_check_bugs();
432}
433
434void apply_seal_endbr(s32 *start, s32 *end)
435{
436}
437
438void apply_retpolines(s32 *start, s32 *end)
439{
440}
441
442void apply_returns(s32 *start, s32 *end)
443{
444}
445
446void apply_fineibt(s32 *start_retpoline, s32 *end_retpoline,
447		   s32 *start_cfi, s32 *end_cfi)
448{
449}
450
451void apply_alternatives(struct alt_instr *start, struct alt_instr *end)
452{
453}
454
455#if IS_ENABLED(CONFIG_SMP)
456void alternatives_smp_module_add(struct module *mod, char *name,
457				 void *locks, void *locks_end,
458				 void *text,  void *text_end)
459{
460}
461
462void alternatives_smp_module_del(struct module *mod)
463{
464}
465#endif
466
467void *text_poke(void *addr, const void *opcode, size_t len)
468{
469	/*
470	 * In UML, the only reference to this function is in
471	 * apply_relocate_add(), which shouldn't ever actually call this
472	 * because UML doesn't have live patching.
473	 */
474	WARN_ON(1);
475
476	return memcpy(addr, opcode, len);
477}
478
479void *text_poke_copy(void *addr, const void *opcode, size_t len)
480{
481	return text_poke(addr, opcode, len);
482}
483
484void smp_text_poke_sync_each_cpu(void)
485{
486}
487
488void uml_pm_wake(void)
489{
490	pm_system_wakeup();
491}
492
493#ifdef CONFIG_PM_SLEEP
494static int um_suspend_valid(suspend_state_t state)
495{
496	return state == PM_SUSPEND_MEM;
497}
498
499static int um_suspend_prepare(void)
500{
501	um_irqs_suspend();
502	return 0;
503}
504
505static int um_suspend_enter(suspend_state_t state)
506{
507	if (WARN_ON(state != PM_SUSPEND_MEM))
508		return -EINVAL;
509
510	/*
511	 * This is identical to the idle sleep, but we've just
512	 * (during suspend) turned off all interrupt sources
513	 * except for the ones we want, so now we can only wake
514	 * up on something we actually want to wake up on. All
515	 * timing has also been suspended.
516	 */
517	um_idle_sleep();
518	return 0;
519}
520
521static void um_suspend_finish(void)
522{
523	um_irqs_resume();
524}
525
526const struct platform_suspend_ops um_suspend_ops = {
527	.valid = um_suspend_valid,
528	.prepare = um_suspend_prepare,
529	.enter = um_suspend_enter,
530	.finish = um_suspend_finish,
531};
532
533static int init_pm_wake_signal(void)
534{
535	/*
536	 * In external time-travel mode we can't use signals to wake up
537	 * since that would mess with the scheduling. We'll have to do
538	 * some additional work to support wakeup on virtio devices or
539	 * similar, perhaps implementing a fake RTC controller that can
540	 * trigger wakeup (and request the appropriate scheduling from
541	 * the external scheduler when going to suspend.)
542	 */
543	if (time_travel_mode != TT_MODE_EXTERNAL)
544		register_pm_wake_signal();
545
546	suspend_set_ops(&um_suspend_ops);
547
548	return 0;
549}
550
551late_initcall(init_pm_wake_signal);
552#endif