xref: /linux/arch/riscv/kernel/process.c (revision 23b0f90ba871f096474e1c27c3d14f455189d2d9)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2009 Sunplus Core Technology Co., Ltd.
4  *  Chen Liqin <liqin.chen@sunplusct.com>
5  *  Lennox Wu <lennox.wu@sunplusct.com>
6  * Copyright (C) 2012 Regents of the University of California
7  * Copyright (C) 2017 SiFive
8  */
9 
10 #include <linux/bitfield.h>
11 #include <linux/cpu.h>
12 #include <linux/kernel.h>
13 #include <linux/sched.h>
14 #include <linux/sched/debug.h>
15 #include <linux/sched/task_stack.h>
16 #include <linux/tick.h>
17 #include <linux/ptrace.h>
18 #include <linux/uaccess.h>
19 #include <linux/personality.h>
20 #include <linux/entry-common.h>
21 
22 #include <asm/asm-prototypes.h>
23 #include <asm/unistd.h>
24 #include <asm/processor.h>
25 #include <asm/csr.h>
26 #include <asm/stacktrace.h>
27 #include <asm/string.h>
28 #include <asm/switch_to.h>
29 #include <asm/thread_info.h>
30 #include <asm/cpuidle.h>
31 #include <asm/vector.h>
32 #include <asm/cpufeature.h>
33 #include <asm/exec.h>
34 #include <asm/usercfi.h>
35 
36 #if defined(CONFIG_STACKPROTECTOR) && !defined(CONFIG_STACKPROTECTOR_PER_TASK)
37 #include <linux/stackprotector.h>
38 unsigned long __stack_chk_guard __read_mostly;
39 EXPORT_SYMBOL(__stack_chk_guard);
40 #endif
41 
42 extern asmlinkage void ret_from_fork_kernel_asm(void);
43 extern asmlinkage void ret_from_fork_user_asm(void);
44 
45 void noinstr arch_cpu_idle(void)
46 {
47 	cpu_do_idle();
48 }
49 
50 int set_unalign_ctl(struct task_struct *tsk, unsigned int val)
51 {
52 	if (!unaligned_ctl_available())
53 		return -EINVAL;
54 
55 	tsk->thread.align_ctl = val;
56 	return 0;
57 }
58 
59 int get_unalign_ctl(struct task_struct *tsk, unsigned long adr)
60 {
61 	if (!unaligned_ctl_available())
62 		return -EINVAL;
63 
64 	return put_user(tsk->thread.align_ctl, (unsigned int __user *)adr);
65 }
66 
67 void __show_regs(struct pt_regs *regs)
68 {
69 	show_regs_print_info(KERN_DEFAULT);
70 
71 	if (!user_mode(regs)) {
72 		pr_cont("epc : %pS\n", (void *)regs->epc);
73 		pr_cont(" ra : %pS\n", (void *)regs->ra);
74 	}
75 
76 	pr_cont("epc : " REG_FMT " ra : " REG_FMT " sp : " REG_FMT "\n",
77 		regs->epc, regs->ra, regs->sp);
78 	pr_cont(" gp : " REG_FMT " tp : " REG_FMT " t0 : " REG_FMT "\n",
79 		regs->gp, regs->tp, regs->t0);
80 	pr_cont(" t1 : " REG_FMT " t2 : " REG_FMT " s0 : " REG_FMT "\n",
81 		regs->t1, regs->t2, regs->s0);
82 	pr_cont(" s1 : " REG_FMT " a0 : " REG_FMT " a1 : " REG_FMT "\n",
83 		regs->s1, regs->a0, regs->a1);
84 	pr_cont(" a2 : " REG_FMT " a3 : " REG_FMT " a4 : " REG_FMT "\n",
85 		regs->a2, regs->a3, regs->a4);
86 	pr_cont(" a5 : " REG_FMT " a6 : " REG_FMT " a7 : " REG_FMT "\n",
87 		regs->a5, regs->a6, regs->a7);
88 	pr_cont(" s2 : " REG_FMT " s3 : " REG_FMT " s4 : " REG_FMT "\n",
89 		regs->s2, regs->s3, regs->s4);
90 	pr_cont(" s5 : " REG_FMT " s6 : " REG_FMT " s7 : " REG_FMT "\n",
91 		regs->s5, regs->s6, regs->s7);
92 	pr_cont(" s8 : " REG_FMT " s9 : " REG_FMT " s10: " REG_FMT "\n",
93 		regs->s8, regs->s9, regs->s10);
94 	pr_cont(" s11: " REG_FMT " t3 : " REG_FMT " t4 : " REG_FMT "\n",
95 		regs->s11, regs->t3, regs->t4);
96 	pr_cont(" t5 : " REG_FMT " t6 : " REG_FMT " ssp : " REG_FMT "\n",
97 		regs->t5, regs->t6, get_active_shstk(current));
98 
99 	pr_cont("status: " REG_FMT " badaddr: " REG_FMT " cause: " REG_FMT "\n",
100 		regs->status, regs->badaddr, regs->cause);
101 }
102 void show_regs(struct pt_regs *regs)
103 {
104 	__show_regs(regs);
105 	if (!user_mode(regs))
106 		dump_backtrace(regs, NULL, KERN_DEFAULT);
107 }
108 
109 unsigned long arch_align_stack(unsigned long sp)
110 {
111 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
112 		sp -= get_random_u32_below(PAGE_SIZE);
113 	return sp & ~0xf;
114 }
115 
116 #ifdef CONFIG_COMPAT
117 static bool compat_mode_supported __read_mostly;
118 
119 bool compat_elf_check_arch(Elf32_Ehdr *hdr)
120 {
121 	return compat_mode_supported &&
122 	       hdr->e_machine == EM_RISCV &&
123 	       hdr->e_ident[EI_CLASS] == ELFCLASS32;
124 }
125 
126 static int __init compat_mode_detect(void)
127 {
128 	unsigned long tmp = csr_read(CSR_STATUS);
129 
130 	csr_write(CSR_STATUS, (tmp & ~SR_UXL) | SR_UXL_32);
131 	compat_mode_supported =
132 			(csr_read(CSR_STATUS) & SR_UXL) == SR_UXL_32;
133 
134 	csr_write(CSR_STATUS, tmp);
135 
136 	pr_info("riscv: ELF compat mode %s",
137 			compat_mode_supported ? "supported" : "unsupported");
138 
139 	return 0;
140 }
141 early_initcall(compat_mode_detect);
142 #endif
143 
144 void start_thread(struct pt_regs *regs, unsigned long pc,
145 	unsigned long sp)
146 {
147 	regs->status = SR_PIE;
148 	if (has_fpu()) {
149 		regs->status |= SR_FS_INITIAL;
150 		/*
151 		 * Restore the initial value to the FP register
152 		 * before starting the user program.
153 		 */
154 		fstate_restore(current, regs);
155 	}
156 	regs->epc = pc;
157 	regs->sp = sp;
158 
159 	/*
160 	 * clear shadow stack state on exec.
161 	 * libc will set it later via prctl.
162 	 */
163 	set_shstk_status(current, false);
164 	set_shstk_base(current, 0, 0);
165 	set_active_shstk(current, 0);
166 	/*
167 	 * disable indirect branch tracking on exec.
168 	 * libc will enable it later via prctl.
169 	 */
170 	set_indir_lp_status(current, false);
171 
172 #ifdef CONFIG_64BIT
173 	regs->status &= ~SR_UXL;
174 
175 	if (is_compat_task())
176 		regs->status |= SR_UXL_32;
177 	else
178 		regs->status |= SR_UXL_64;
179 #endif
180 }
181 
182 void flush_thread(void)
183 {
184 #ifdef CONFIG_FPU
185 	/*
186 	 * Reset FPU state and context
187 	 *	frm: round to nearest, ties to even (IEEE default)
188 	 *	fflags: accrued exceptions cleared
189 	 */
190 	fstate_off(current, task_pt_regs(current));
191 	memset(&current->thread.fstate, 0, sizeof(current->thread.fstate));
192 #endif
193 #ifdef CONFIG_RISCV_ISA_V
194 	/* Reset vector state */
195 	riscv_v_vstate_ctrl_init(current);
196 	riscv_v_vstate_off(task_pt_regs(current));
197 	kfree(current->thread.vstate.datap);
198 	memset(&current->thread.vstate, 0, sizeof(struct __riscv_v_ext_state));
199 	clear_tsk_thread_flag(current, TIF_RISCV_V_DEFER_RESTORE);
200 #endif
201 #ifdef CONFIG_RISCV_ISA_SUPM
202 	if (riscv_has_extension_unlikely(RISCV_ISA_EXT_SUPM))
203 		envcfg_update_bits(current, ENVCFG_PMM, ENVCFG_PMM_PMLEN_0);
204 #endif
205 }
206 
207 void arch_release_task_struct(struct task_struct *tsk)
208 {
209 	/* Free the vector context of datap. */
210 	if (has_vector() || has_xtheadvector())
211 		riscv_v_thread_free(tsk);
212 }
213 
214 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
215 {
216 	fstate_save(src, task_pt_regs(src));
217 	*dst = *src;
218 	/* clear entire V context, including datap for a new task */
219 	memset(&dst->thread.vstate, 0, sizeof(struct __riscv_v_ext_state));
220 	memset(&dst->thread.kernel_vstate, 0, sizeof(struct __riscv_v_ext_state));
221 	clear_tsk_thread_flag(dst, TIF_RISCV_V_DEFER_RESTORE);
222 
223 	return 0;
224 }
225 
226 asmlinkage void ret_from_fork_kernel(void *fn_arg, int (*fn)(void *), struct pt_regs *regs)
227 {
228 	fn(fn_arg);
229 
230 	syscall_exit_to_user_mode(regs);
231 }
232 
233 asmlinkage void ret_from_fork_user(struct pt_regs *regs)
234 {
235 	syscall_exit_to_user_mode(regs);
236 }
237 
238 int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
239 {
240 	u64 clone_flags = args->flags;
241 	unsigned long usp = args->stack;
242 	unsigned long tls = args->tls;
243 	unsigned long ssp = 0;
244 	struct pt_regs *childregs = task_pt_regs(p);
245 
246 	/* Ensure all threads in this mm have the same pointer masking mode. */
247 	if (IS_ENABLED(CONFIG_RISCV_ISA_SUPM) && p->mm && (clone_flags & CLONE_VM))
248 		set_bit(MM_CONTEXT_LOCK_PMLEN, &p->mm->context.flags);
249 
250 	memset(&p->thread.s, 0, sizeof(p->thread.s));
251 
252 	/* p->thread holds context to be restored by __switch_to() */
253 	if (unlikely(args->fn)) {
254 		/* Kernel thread */
255 		memset(childregs, 0, sizeof(struct pt_regs));
256 		/* Supervisor/Machine, irqs on: */
257 		childregs->status = SR_PP | SR_PIE;
258 
259 		p->thread.s[0] = (unsigned long)args->fn;
260 		p->thread.s[1] = (unsigned long)args->fn_arg;
261 		p->thread.ra = (unsigned long)ret_from_fork_kernel_asm;
262 	} else {
263 		/* allocate new shadow stack if needed. In case of CLONE_VM we have to */
264 		ssp = shstk_alloc_thread_stack(p, args);
265 		if (IS_ERR_VALUE(ssp))
266 			return PTR_ERR((void *)ssp);
267 
268 		*childregs = *(current_pt_regs());
269 		/* Turn off status.VS */
270 		riscv_v_vstate_off(childregs);
271 		if (usp) /* User fork */
272 			childregs->sp = usp;
273 		/* if needed, set new ssp */
274 		if (ssp)
275 			set_active_shstk(p, ssp);
276 		if (clone_flags & CLONE_SETTLS)
277 			childregs->tp = tls;
278 		childregs->a0 = 0; /* Return value of fork() */
279 		p->thread.ra = (unsigned long)ret_from_fork_user_asm;
280 	}
281 	p->thread.riscv_v_flags = 0;
282 	if (has_vector() || has_xtheadvector())
283 		riscv_v_thread_alloc(p);
284 	p->thread.sp = (unsigned long)childregs; /* kernel sp */
285 	return 0;
286 }
287 
288 void __init arch_task_cache_init(void)
289 {
290 	riscv_v_setup_ctx_cache();
291 }
292 
293 #ifdef CONFIG_RISCV_ISA_SUPM
294 enum {
295 	PMLEN_0 = 0,
296 	PMLEN_7 = 7,
297 	PMLEN_16 = 16,
298 };
299 
300 static bool have_user_pmlen_7;
301 static bool have_user_pmlen_16;
302 
303 /*
304  * Control the relaxed ABI allowing tagged user addresses into the kernel.
305  */
306 static unsigned int tagged_addr_disabled;
307 
308 long set_tagged_addr_ctrl(struct task_struct *task, unsigned long arg)
309 {
310 	unsigned long valid_mask = PR_PMLEN_MASK | PR_TAGGED_ADDR_ENABLE;
311 	struct thread_info *ti = task_thread_info(task);
312 	struct mm_struct *mm = task->mm;
313 	unsigned long pmm;
314 	u8 pmlen;
315 
316 	if (!riscv_has_extension_unlikely(RISCV_ISA_EXT_SUPM))
317 		return -EINVAL;
318 
319 	if (is_compat_thread(ti))
320 		return -EINVAL;
321 
322 	if (arg & ~valid_mask)
323 		return -EINVAL;
324 
325 	/*
326 	 * Prefer the smallest PMLEN that satisfies the user's request,
327 	 * in case choosing a larger PMLEN has a performance impact.
328 	 */
329 	pmlen = FIELD_GET(PR_PMLEN_MASK, arg);
330 	if (pmlen == PMLEN_0) {
331 		pmm = ENVCFG_PMM_PMLEN_0;
332 	} else if (pmlen <= PMLEN_7 && have_user_pmlen_7) {
333 		pmlen = PMLEN_7;
334 		pmm = ENVCFG_PMM_PMLEN_7;
335 	} else if (pmlen <= PMLEN_16 && have_user_pmlen_16) {
336 		pmlen = PMLEN_16;
337 		pmm = ENVCFG_PMM_PMLEN_16;
338 	} else {
339 		return -EINVAL;
340 	}
341 
342 	/*
343 	 * Do not allow the enabling of the tagged address ABI if globally
344 	 * disabled via sysctl abi.tagged_addr_disabled, if pointer masking
345 	 * is disabled for userspace.
346 	 */
347 	if (arg & PR_TAGGED_ADDR_ENABLE && (tagged_addr_disabled || !pmlen))
348 		return -EINVAL;
349 
350 	if (!(arg & PR_TAGGED_ADDR_ENABLE))
351 		pmlen = PMLEN_0;
352 
353 	if (mmap_write_lock_killable(mm))
354 		return -EINTR;
355 
356 	if (test_bit(MM_CONTEXT_LOCK_PMLEN, &mm->context.flags) && mm->context.pmlen != pmlen) {
357 		mmap_write_unlock(mm);
358 		return -EBUSY;
359 	}
360 
361 	envcfg_update_bits(task, ENVCFG_PMM, pmm);
362 	mm->context.pmlen = pmlen;
363 
364 	mmap_write_unlock(mm);
365 
366 	return 0;
367 }
368 
369 long get_tagged_addr_ctrl(struct task_struct *task)
370 {
371 	struct thread_info *ti = task_thread_info(task);
372 	long ret = 0;
373 
374 	if (!riscv_has_extension_unlikely(RISCV_ISA_EXT_SUPM))
375 		return -EINVAL;
376 
377 	if (is_compat_thread(ti))
378 		return -EINVAL;
379 
380 	/*
381 	 * The mm context's pmlen is set only when the tagged address ABI is
382 	 * enabled, so the effective PMLEN must be extracted from envcfg.PMM.
383 	 */
384 	switch (task->thread.envcfg & ENVCFG_PMM) {
385 	case ENVCFG_PMM_PMLEN_7:
386 		ret = FIELD_PREP(PR_PMLEN_MASK, PMLEN_7);
387 		break;
388 	case ENVCFG_PMM_PMLEN_16:
389 		ret = FIELD_PREP(PR_PMLEN_MASK, PMLEN_16);
390 		break;
391 	}
392 
393 	if (task->mm->context.pmlen)
394 		ret |= PR_TAGGED_ADDR_ENABLE;
395 
396 	return ret;
397 }
398 
399 static bool try_to_set_pmm(unsigned long value)
400 {
401 	csr_set(CSR_ENVCFG, value);
402 	return (csr_read_clear(CSR_ENVCFG, ENVCFG_PMM) & ENVCFG_PMM) == value;
403 }
404 
405 /*
406  * Global sysctl to disable the tagged user addresses support. This control
407  * only prevents the tagged address ABI enabling via prctl() and does not
408  * disable it for tasks that already opted in to the relaxed ABI.
409  */
410 
411 static const struct ctl_table tagged_addr_sysctl_table[] = {
412 	{
413 		.procname	= "tagged_addr_disabled",
414 		.mode		= 0644,
415 		.data		= &tagged_addr_disabled,
416 		.maxlen		= sizeof(int),
417 		.proc_handler	= proc_dointvec_minmax,
418 		.extra1		= SYSCTL_ZERO,
419 		.extra2		= SYSCTL_ONE,
420 	},
421 };
422 
423 static int __init tagged_addr_init(void)
424 {
425 	if (!riscv_has_extension_unlikely(RISCV_ISA_EXT_SUPM))
426 		return 0;
427 
428 	/*
429 	 * envcfg.PMM is a WARL field. Detect which values are supported.
430 	 * Assume the supported PMLEN values are the same on all harts.
431 	 */
432 	csr_clear(CSR_ENVCFG, ENVCFG_PMM);
433 	have_user_pmlen_7 = try_to_set_pmm(ENVCFG_PMM_PMLEN_7);
434 	have_user_pmlen_16 = try_to_set_pmm(ENVCFG_PMM_PMLEN_16);
435 
436 	if (!register_sysctl("abi", tagged_addr_sysctl_table))
437 		return -EINVAL;
438 
439 	return 0;
440 }
441 core_initcall(tagged_addr_init);
442 #endif	/* CONFIG_RISCV_ISA_SUPM */
443