xref: /linux/arch/loongarch/kernel/process.c (revision bb3c90fe347a5321e7d176ed5b21367aa28be9ee)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Author: Huacai Chen <chenhuacai@loongson.cn>
4  * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
5  *
6  * Derived from MIPS:
7  * Copyright (C) 1994 - 1999, 2000 by Ralf Baechle and others.
8  * Copyright (C) 2005, 2006 by Ralf Baechle (ralf@linux-mips.org)
9  * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
10  * Copyright (C) 2004 Thiemo Seufer
11  * Copyright (C) 2013  Imagination Technologies Ltd.
12  */
13 #include <linux/cpu.h>
14 #include <linux/init.h>
15 #include <linux/kernel.h>
16 #include <linux/entry-common.h>
17 #include <linux/errno.h>
18 #include <linux/sched.h>
19 #include <linux/sched/debug.h>
20 #include <linux/sched/task.h>
21 #include <linux/sched/task_stack.h>
22 #include <linux/hw_breakpoint.h>
23 #include <linux/mm.h>
24 #include <linux/stddef.h>
25 #include <linux/unistd.h>
26 #include <linux/export.h>
27 #include <linux/ptrace.h>
28 #include <linux/mman.h>
29 #include <linux/personality.h>
30 #include <linux/sys.h>
31 #include <linux/completion.h>
32 #include <linux/kallsyms.h>
33 #include <linux/random.h>
34 #include <linux/prctl.h>
35 #include <linux/nmi.h>
36 
37 #include <asm/asm.h>
38 #include <asm/asm-prototypes.h>
39 #include <asm/bootinfo.h>
40 #include <asm/cpu.h>
41 #include <asm/elf.h>
42 #include <asm/exec.h>
43 #include <asm/fpu.h>
44 #include <asm/lbt.h>
45 #include <asm/io.h>
46 #include <asm/irq.h>
47 #include <asm/irq_regs.h>
48 #include <asm/loongarch.h>
49 #include <asm/pgtable.h>
50 #include <asm/processor.h>
51 #include <asm/reg.h>
52 #include <asm/switch_to.h>
53 #include <asm/unwind.h>
54 #include <asm/vdso.h>
55 #include <asm/vdso/vdso.h>
56 
57 #ifdef CONFIG_STACKPROTECTOR
58 #include <linux/stackprotector.h>
59 unsigned long __stack_chk_guard __read_mostly;
60 EXPORT_SYMBOL(__stack_chk_guard);
61 #endif
62 
63 DEFINE_PER_CPU(struct task_struct *, cpu_tasks);
64 
65 /*
66  * Idle related variables and functions
67  */
68 
69 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
70 EXPORT_SYMBOL(boot_option_idle_override);
71 
72 asmlinkage void restore_and_ret(void);
73 asmlinkage void ret_from_fork_asm(void);
74 asmlinkage void ret_from_kernel_thread_asm(void);
75 
76 void start_thread(struct pt_regs *regs, unsigned long pc, unsigned long sp)
77 {
78 	unsigned long crmd;
79 	unsigned long prmd;
80 	unsigned long euen;
81 
82 	/* New thread loses kernel privileges. */
83 	crmd = regs->csr_crmd & ~(PLV_MASK);
84 	crmd |= PLV_USER;
85 	regs->csr_crmd = crmd;
86 
87 	prmd = regs->csr_prmd & ~(PLV_MASK);
88 	prmd |= PLV_USER;
89 	regs->csr_prmd = prmd;
90 
91 	euen = regs->csr_euen & ~(CSR_EUEN_FPEN);
92 	regs->csr_euen = euen;
93 	lose_fpu(0);
94 	lose_lbt(0);
95 	current->thread.fpu.fcsr = boot_cpu_data.fpu_csr0;
96 
97 	clear_thread_flag(TIF_LSX_CTX_LIVE);
98 	clear_thread_flag(TIF_LASX_CTX_LIVE);
99 	clear_thread_flag(TIF_LBT_CTX_LIVE);
100 	clear_used_math();
101 	regs->csr_era = pc;
102 	regs->regs[3] = sp;
103 }
104 
105 void flush_thread(void)
106 {
107 	flush_ptrace_hw_breakpoint(current);
108 }
109 
110 void exit_thread(struct task_struct *tsk)
111 {
112 }
113 
114 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
115 {
116 	/*
117 	 * Save any process state which is live in hardware registers to the
118 	 * parent context prior to duplication. This prevents the new child
119 	 * state becoming stale if the parent is preempted before copy_thread()
120 	 * gets a chance to save the parent's live hardware registers to the
121 	 * child context.
122 	 */
123 	preempt_disable();
124 
125 	if (is_fpu_owner()) {
126 		if (is_lasx_enabled())
127 			save_lasx(current);
128 		else if (is_lsx_enabled())
129 			save_lsx(current);
130 		else
131 			save_fp(current);
132 	}
133 
134 	preempt_enable();
135 
136 	if (IS_ENABLED(CONFIG_RANDSTRUCT)) {
137 		memcpy(dst, src, sizeof(struct task_struct));
138 		return 0;
139 	}
140 
141 	dst->thread.fpu.fcsr =  src->thread.fpu.fcsr;
142 
143 	if (!used_math())
144 		memcpy(dst, src, offsetof(struct task_struct, thread.fpu.fpr));
145 	else
146 		memcpy(dst, src, offsetof(struct task_struct, thread.lbt.scr0));
147 
148 #ifdef CONFIG_CPU_HAS_LBT
149 	memcpy(&dst->thread.lbt, &src->thread.lbt, sizeof(struct loongarch_lbt));
150 #endif
151 
152 	return 0;
153 }
154 
155 asmlinkage void noinstr __no_stack_protector ret_from_fork(struct task_struct *prev,
156 							   struct pt_regs *regs)
157 {
158 	schedule_tail(prev);
159 	syscall_exit_to_user_mode(regs);
160 }
161 
162 asmlinkage void noinstr __no_stack_protector ret_from_kernel_thread(struct task_struct *prev,
163 								    struct pt_regs *regs,
164 								    int (*fn)(void *),
165 								    void *fn_arg)
166 {
167 	schedule_tail(prev);
168 	fn(fn_arg);
169 	syscall_exit_to_user_mode(regs);
170 }
171 
172 /*
173  * Copy architecture-specific thread state
174  */
175 int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
176 {
177 	unsigned long childksp;
178 	unsigned long tls = args->tls;
179 	unsigned long usp = args->stack;
180 	u64 clone_flags = args->flags;
181 	struct pt_regs *childregs, *regs = current_pt_regs();
182 
183 	childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
184 
185 	/* set up new TSS. */
186 	childregs = (struct pt_regs *) childksp - 1;
187 	/*  Put the stack after the struct pt_regs.  */
188 	childksp = (unsigned long) childregs;
189 	p->thread.sched_cfa = 0;
190 	p->thread.csr_euen = 0;
191 	p->thread.csr_crmd = csr_read32(LOONGARCH_CSR_CRMD);
192 	p->thread.csr_prmd = csr_read32(LOONGARCH_CSR_PRMD);
193 	p->thread.csr_ecfg = csr_read32(LOONGARCH_CSR_ECFG);
194 	if (unlikely(args->fn)) {
195 		/* kernel thread */
196 		p->thread.reg03 = childksp;
197 		p->thread.reg23 = (unsigned long)args->fn;
198 		p->thread.reg24 = (unsigned long)args->fn_arg;
199 		p->thread.reg01 = (unsigned long)ret_from_kernel_thread_asm;
200 		p->thread.sched_ra = (unsigned long)ret_from_kernel_thread_asm;
201 		memset(childregs, 0, sizeof(struct pt_regs));
202 		childregs->csr_euen = p->thread.csr_euen;
203 		childregs->csr_crmd = p->thread.csr_crmd;
204 		childregs->csr_prmd = p->thread.csr_prmd;
205 		childregs->csr_ecfg = p->thread.csr_ecfg;
206 		goto out;
207 	}
208 
209 	/* user thread */
210 	*childregs = *regs;
211 	childregs->regs[4] = 0; /* Child gets zero as return value */
212 	if (usp)
213 		childregs->regs[3] = usp;
214 
215 	p->thread.reg03 = (unsigned long) childregs;
216 	p->thread.reg01 = (unsigned long) ret_from_fork_asm;
217 	p->thread.sched_ra = (unsigned long) ret_from_fork_asm;
218 
219 	/*
220 	 * New tasks lose permission to use the fpu. This accelerates context
221 	 * switching for most programs since they don't use the fpu.
222 	 */
223 	childregs->csr_euen = 0;
224 
225 	if (clone_flags & CLONE_SETTLS)
226 		childregs->regs[2] = tls;
227 
228 out:
229 	ptrace_hw_copy_thread(p);
230 	clear_tsk_thread_flag(p, TIF_USEDFPU);
231 	clear_tsk_thread_flag(p, TIF_USEDSIMD);
232 	clear_tsk_thread_flag(p, TIF_USEDLBT);
233 	clear_tsk_thread_flag(p, TIF_LSX_CTX_LIVE);
234 	clear_tsk_thread_flag(p, TIF_LASX_CTX_LIVE);
235 	clear_tsk_thread_flag(p, TIF_LBT_CTX_LIVE);
236 
237 	return 0;
238 }
239 
240 unsigned long __get_wchan(struct task_struct *task)
241 {
242 	unsigned long pc = 0;
243 	struct unwind_state state;
244 
245 	if (!try_get_task_stack(task))
246 		return 0;
247 
248 	for (unwind_start(&state, task, NULL);
249 	     !unwind_done(&state); unwind_next_frame(&state)) {
250 		pc = unwind_get_return_address(&state);
251 		if (!pc)
252 			break;
253 		if (in_sched_functions(pc))
254 			continue;
255 		break;
256 	}
257 
258 	put_task_stack(task);
259 
260 	return pc;
261 }
262 
263 bool in_irq_stack(unsigned long stack, struct stack_info *info)
264 {
265 	unsigned long nextsp;
266 	unsigned long begin = (unsigned long)this_cpu_read(irq_stack);
267 	unsigned long end = begin + IRQ_STACK_START;
268 
269 	if (stack < begin || stack >= end)
270 		return false;
271 
272 	nextsp = *(unsigned long *)end;
273 	if (nextsp & (SZREG - 1))
274 		return false;
275 
276 	info->begin = begin;
277 	info->end = end;
278 	info->next_sp = nextsp;
279 	info->type = STACK_TYPE_IRQ;
280 
281 	return true;
282 }
283 
284 bool in_task_stack(unsigned long stack, struct task_struct *task,
285 			struct stack_info *info)
286 {
287 	unsigned long begin = (unsigned long)task_stack_page(task);
288 	unsigned long end = begin + THREAD_SIZE;
289 
290 	if (stack < begin || stack >= end)
291 		return false;
292 
293 	info->begin = begin;
294 	info->end = end;
295 	info->next_sp = 0;
296 	info->type = STACK_TYPE_TASK;
297 
298 	return true;
299 }
300 
301 int get_stack_info(unsigned long stack, struct task_struct *task,
302 		   struct stack_info *info)
303 {
304 	task = task ? : current;
305 
306 	if (!stack || stack & (SZREG - 1))
307 		goto unknown;
308 
309 	if (in_task_stack(stack, task, info))
310 		return 0;
311 
312 	if (task != current)
313 		goto unknown;
314 
315 	if (in_irq_stack(stack, info))
316 		return 0;
317 
318 unknown:
319 	info->type = STACK_TYPE_UNKNOWN;
320 	return -EINVAL;
321 }
322 
323 unsigned long stack_top(void)
324 {
325 	unsigned long top = TASK_SIZE & PAGE_MASK;
326 
327 	if (current->thread.vdso) {
328 		/* Space for the VDSO & data page */
329 		top -= PAGE_ALIGN(current->thread.vdso->size);
330 		top -= VVAR_SIZE;
331 
332 		/* Space to randomize the VDSO base */
333 		if (current->flags & PF_RANDOMIZE)
334 			top -= VDSO_RANDOMIZE_SIZE;
335 	}
336 
337 	return top;
338 }
339 
340 /*
341  * Don't forget that the stack pointer must be aligned on a 8 bytes
342  * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
343  */
344 unsigned long arch_align_stack(unsigned long sp)
345 {
346 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
347 		sp -= get_random_u32_below(PAGE_SIZE);
348 
349 	return sp & STACK_ALIGN;
350 }
351 
352 static DEFINE_PER_CPU(call_single_data_t, backtrace_csd);
353 static struct cpumask backtrace_csd_busy;
354 
355 static void handle_backtrace(void *info)
356 {
357 	nmi_cpu_backtrace(get_irq_regs());
358 	cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy);
359 }
360 
361 static void raise_backtrace(cpumask_t *mask)
362 {
363 	call_single_data_t *csd;
364 	int cpu;
365 
366 	for_each_cpu(cpu, mask) {
367 		/*
368 		 * If we previously sent an IPI to the target CPU & it hasn't
369 		 * cleared its bit in the busy cpumask then it didn't handle
370 		 * our previous IPI & it's not safe for us to reuse the
371 		 * call_single_data_t.
372 		 */
373 		if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) {
374 			pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n",
375 				cpu);
376 			continue;
377 		}
378 
379 		csd = &per_cpu(backtrace_csd, cpu);
380 		csd->func = handle_backtrace;
381 		smp_call_function_single_async(cpu, csd);
382 	}
383 }
384 
385 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int exclude_cpu)
386 {
387 	nmi_trigger_cpumask_backtrace(mask, exclude_cpu, raise_backtrace);
388 }
389 
390 #ifdef CONFIG_32BIT
391 void loongarch_dump_regs32(u32 *uregs, const struct pt_regs *regs)
392 #else
393 void loongarch_dump_regs64(u64 *uregs, const struct pt_regs *regs)
394 #endif
395 {
396 	unsigned int i;
397 
398 	for (i = LOONGARCH_EF_R1; i <= LOONGARCH_EF_R31; i++) {
399 		uregs[i] = regs->regs[i - LOONGARCH_EF_R0];
400 	}
401 
402 	uregs[LOONGARCH_EF_ORIG_A0] = regs->orig_a0;
403 	uregs[LOONGARCH_EF_CSR_ERA] = regs->csr_era;
404 	uregs[LOONGARCH_EF_CSR_BADV] = regs->csr_badvaddr;
405 	uregs[LOONGARCH_EF_CSR_CRMD] = regs->csr_crmd;
406 	uregs[LOONGARCH_EF_CSR_PRMD] = regs->csr_prmd;
407 	uregs[LOONGARCH_EF_CSR_EUEN] = regs->csr_euen;
408 	uregs[LOONGARCH_EF_CSR_ECFG] = regs->csr_ecfg;
409 	uregs[LOONGARCH_EF_CSR_ESTAT] = regs->csr_estat;
410 }
411