xref: /linux/arch/riscv/mm/fault.c (revision f77d1a49902bc70625e3d101a16d8a687f7e97db)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  * Copyright (C) 2009 Sunplus Core Technology Co., Ltd.
4  *  Lennox Wu <lennox.wu@sunplusct.com>
5  *  Chen Liqin <liqin.chen@sunplusct.com>
6  * Copyright (C) 2012 Regents of the University of California
7  */
8 
9 
10 #include <linux/mm.h>
11 #include <linux/kernel.h>
12 #include <linux/interrupt.h>
13 #include <linux/perf_event.h>
14 #include <linux/signal.h>
15 #include <linux/uaccess.h>
16 #include <linux/kprobes.h>
17 #include <linux/kfence.h>
18 #include <linux/entry-common.h>
19 
20 #include <asm/ptrace.h>
21 #include <asm/tlbflush.h>
22 
23 #include "../kernel/head.h"
24 
25 static void die_kernel_fault(const char *msg, unsigned long addr,
26 		struct pt_regs *regs)
27 {
28 	bust_spinlocks(1);
29 
30 	pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n", msg,
31 		addr);
32 
33 	bust_spinlocks(0);
34 	die(regs, "Oops");
35 	make_task_dead(SIGKILL);
36 }
37 
38 static inline void no_context(struct pt_regs *regs, unsigned long addr)
39 {
40 	const char *msg;
41 
42 	/* Are we prepared to handle this kernel fault? */
43 	if (fixup_exception(regs))
44 		return;
45 
46 	/*
47 	 * Oops. The kernel tried to access some bad page. We'll have to
48 	 * terminate things with extreme prejudice.
49 	 */
50 	if (addr < PAGE_SIZE)
51 		msg = "NULL pointer dereference";
52 	else {
53 		if (kfence_handle_page_fault(addr, regs->cause == EXC_STORE_PAGE_FAULT, regs))
54 			return;
55 
56 		msg = "paging request";
57 	}
58 
59 	die_kernel_fault(msg, addr, regs);
60 }
61 
62 static inline void mm_fault_error(struct pt_regs *regs, unsigned long addr, vm_fault_t fault)
63 {
64 	if (fault & VM_FAULT_OOM) {
65 		/*
66 		 * We ran out of memory, call the OOM killer, and return the userspace
67 		 * (which will retry the fault, or kill us if we got oom-killed).
68 		 */
69 		if (!user_mode(regs)) {
70 			no_context(regs, addr);
71 			return;
72 		}
73 		pagefault_out_of_memory();
74 		return;
75 	} else if (fault & VM_FAULT_SIGBUS) {
76 		/* Kernel mode? Handle exceptions or die */
77 		if (!user_mode(regs)) {
78 			no_context(regs, addr);
79 			return;
80 		}
81 		do_trap(regs, SIGBUS, BUS_ADRERR, addr);
82 		return;
83 	}
84 	BUG();
85 }
86 
87 static inline void
88 bad_area_nosemaphore(struct pt_regs *regs, int code, unsigned long addr)
89 {
90 	/*
91 	 * Something tried to access memory that isn't in our memory map.
92 	 * Fix it, but check if it's kernel or user first.
93 	 */
94 	/* User mode accesses just cause a SIGSEGV */
95 	if (user_mode(regs)) {
96 		do_trap(regs, SIGSEGV, code, addr);
97 		return;
98 	}
99 
100 	no_context(regs, addr);
101 }
102 
103 static inline void
104 bad_area(struct pt_regs *regs, struct mm_struct *mm, int code,
105 	 unsigned long addr)
106 {
107 	mmap_read_unlock(mm);
108 
109 	bad_area_nosemaphore(regs, code, addr);
110 }
111 
112 static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr)
113 {
114 	pgd_t *pgd, *pgd_k;
115 	pud_t *pud_k;
116 	p4d_t *p4d_k;
117 	pmd_t *pmd_k;
118 	pte_t *pte_k;
119 	int index;
120 	unsigned long pfn;
121 
122 	/* User mode accesses just cause a SIGSEGV */
123 	if (user_mode(regs))
124 		return do_trap(regs, SIGSEGV, code, addr);
125 
126 	/*
127 	 * Synchronize this task's top level page-table
128 	 * with the 'reference' page table.
129 	 *
130 	 * Do _not_ use "tsk->active_mm->pgd" here.
131 	 * We might be inside an interrupt in the middle
132 	 * of a task switch.
133 	 */
134 	index = pgd_index(addr);
135 	pfn = csr_read(CSR_SATP) & SATP_PPN;
136 	pgd = (pgd_t *)pfn_to_virt(pfn) + index;
137 	pgd_k = init_mm.pgd + index;
138 
139 	if (!pgd_present(*pgd_k)) {
140 		no_context(regs, addr);
141 		return;
142 	}
143 	set_pgd(pgd, *pgd_k);
144 
145 	p4d_k = p4d_offset(pgd_k, addr);
146 	if (!p4d_present(*p4d_k)) {
147 		no_context(regs, addr);
148 		return;
149 	}
150 
151 	pud_k = pud_offset(p4d_k, addr);
152 	if (!pud_present(*pud_k)) {
153 		no_context(regs, addr);
154 		return;
155 	}
156 	if (pud_leaf(*pud_k))
157 		goto flush_tlb;
158 
159 	/*
160 	 * Since the vmalloc area is global, it is unnecessary
161 	 * to copy individual PTEs
162 	 */
163 	pmd_k = pmd_offset(pud_k, addr);
164 	if (!pmd_present(*pmd_k)) {
165 		no_context(regs, addr);
166 		return;
167 	}
168 	if (pmd_leaf(*pmd_k))
169 		goto flush_tlb;
170 
171 	/*
172 	 * Make sure the actual PTE exists as well to
173 	 * catch kernel vmalloc-area accesses to non-mapped
174 	 * addresses. If we don't do this, this will just
175 	 * silently loop forever.
176 	 */
177 	pte_k = pte_offset_kernel(pmd_k, addr);
178 	if (!pte_present(*pte_k)) {
179 		no_context(regs, addr);
180 		return;
181 	}
182 
183 	/*
184 	 * The kernel assumes that TLBs don't cache invalid
185 	 * entries, but in RISC-V, SFENCE.VMA specifies an
186 	 * ordering constraint, not a cache flush; it is
187 	 * necessary even after writing invalid entries.
188 	 */
189 flush_tlb:
190 	local_flush_tlb_page(addr);
191 }
192 
193 static inline bool access_error(unsigned long cause, struct vm_area_struct *vma)
194 {
195 	switch (cause) {
196 	case EXC_INST_PAGE_FAULT:
197 		if (!(vma->vm_flags & VM_EXEC)) {
198 			return true;
199 		}
200 		break;
201 	case EXC_LOAD_PAGE_FAULT:
202 		/* Write implies read */
203 		if (!(vma->vm_flags & (VM_READ | VM_WRITE))) {
204 			return true;
205 		}
206 		break;
207 	case EXC_STORE_PAGE_FAULT:
208 		if (!(vma->vm_flags & VM_WRITE)) {
209 			return true;
210 		}
211 		break;
212 	default:
213 		panic("%s: unhandled cause %lu", __func__, cause);
214 	}
215 	return false;
216 }
217 
218 /*
219  * This routine handles page faults.  It determines the address and the
220  * problem, and then passes it off to one of the appropriate routines.
221  */
222 void handle_page_fault(struct pt_regs *regs)
223 {
224 	struct task_struct *tsk;
225 	struct vm_area_struct *vma;
226 	struct mm_struct *mm;
227 	unsigned long addr, cause;
228 	unsigned int flags = FAULT_FLAG_DEFAULT;
229 	int code = SEGV_MAPERR;
230 	vm_fault_t fault;
231 
232 	cause = regs->cause;
233 	addr = regs->badaddr;
234 
235 	tsk = current;
236 	mm = tsk->mm;
237 
238 	if (kprobe_page_fault(regs, cause))
239 		return;
240 
241 	/*
242 	 * Fault-in kernel-space virtual memory on-demand.
243 	 * The 'reference' page table is init_mm.pgd.
244 	 *
245 	 * NOTE! We MUST NOT take any locks for this case. We may
246 	 * be in an interrupt or a critical region, and should
247 	 * only copy the information from the master page table,
248 	 * nothing more.
249 	 */
250 	if ((!IS_ENABLED(CONFIG_MMU) || !IS_ENABLED(CONFIG_64BIT)) &&
251 	    unlikely(addr >= VMALLOC_START && addr < VMALLOC_END)) {
252 		vmalloc_fault(regs, code, addr);
253 		return;
254 	}
255 
256 	/* Enable interrupts if they were enabled in the parent context. */
257 	if (!regs_irqs_disabled(regs))
258 		local_irq_enable();
259 
260 	/*
261 	 * If we're in an interrupt, have no user context, or are running
262 	 * in an atomic region, then we must not take the fault.
263 	 */
264 	if (unlikely(faulthandler_disabled() || !mm)) {
265 		tsk->thread.bad_cause = cause;
266 		no_context(regs, addr);
267 		return;
268 	}
269 
270 	if (user_mode(regs))
271 		flags |= FAULT_FLAG_USER;
272 
273 	if (!user_mode(regs) && addr < TASK_SIZE && unlikely(!(regs->status & SR_SUM))) {
274 		if (fixup_exception(regs))
275 			return;
276 
277 		die_kernel_fault("access to user memory without uaccess routines", addr, regs);
278 	}
279 
280 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
281 
282 	if (cause == EXC_STORE_PAGE_FAULT)
283 		flags |= FAULT_FLAG_WRITE;
284 	else if (cause == EXC_INST_PAGE_FAULT)
285 		flags |= FAULT_FLAG_INSTRUCTION;
286 #ifdef CONFIG_PER_VMA_LOCK
287 	if (!(flags & FAULT_FLAG_USER))
288 		goto lock_mmap;
289 
290 	vma = lock_vma_under_rcu(mm, addr);
291 	if (!vma)
292 		goto lock_mmap;
293 
294 	if (unlikely(access_error(cause, vma))) {
295 		vma_end_read(vma);
296 		goto lock_mmap;
297 	}
298 
299 	fault = handle_mm_fault(vma, addr, flags | FAULT_FLAG_VMA_LOCK, regs);
300 	vma_end_read(vma);
301 
302 	if (!(fault & VM_FAULT_RETRY)) {
303 		count_vm_vma_lock_event(VMA_LOCK_SUCCESS);
304 		goto done;
305 	}
306 	count_vm_vma_lock_event(VMA_LOCK_RETRY);
307 
308 	if (fault_signal_pending(fault, regs)) {
309 		if (!user_mode(regs))
310 			no_context(regs, addr);
311 		return;
312 	}
313 lock_mmap:
314 #endif /* CONFIG_PER_VMA_LOCK */
315 
316 retry:
317 	vma = lock_mm_and_find_vma(mm, addr, regs);
318 	if (unlikely(!vma)) {
319 		tsk->thread.bad_cause = cause;
320 		bad_area_nosemaphore(regs, code, addr);
321 		return;
322 	}
323 
324 	/*
325 	 * Ok, we have a good vm_area for this memory access, so
326 	 * we can handle it.
327 	 */
328 	code = SEGV_ACCERR;
329 
330 	if (unlikely(access_error(cause, vma))) {
331 		tsk->thread.bad_cause = cause;
332 		bad_area(regs, mm, code, addr);
333 		return;
334 	}
335 
336 	/*
337 	 * If for any reason at all we could not handle the fault,
338 	 * make sure we exit gracefully rather than endlessly redo
339 	 * the fault.
340 	 */
341 	fault = handle_mm_fault(vma, addr, flags, regs);
342 
343 	/*
344 	 * If we need to retry but a fatal signal is pending, handle the
345 	 * signal first. We do not need to release the mmap_lock because it
346 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
347 	 */
348 	if (fault_signal_pending(fault, regs)) {
349 		if (!user_mode(regs))
350 			no_context(regs, addr);
351 		return;
352 	}
353 
354 	/* The fault is fully completed (including releasing mmap lock) */
355 	if (fault & VM_FAULT_COMPLETED)
356 		return;
357 
358 	if (unlikely(fault & VM_FAULT_RETRY)) {
359 		flags |= FAULT_FLAG_TRIED;
360 
361 		/*
362 		 * No need to mmap_read_unlock(mm) as we would
363 		 * have already released it in __lock_page_or_retry
364 		 * in mm/filemap.c.
365 		 */
366 		goto retry;
367 	}
368 
369 	mmap_read_unlock(mm);
370 
371 #ifdef CONFIG_PER_VMA_LOCK
372 done:
373 #endif
374 	if (unlikely(fault & VM_FAULT_ERROR)) {
375 		tsk->thread.bad_cause = cause;
376 		mm_fault_error(regs, addr, fault);
377 		return;
378 	}
379 	return;
380 }
381