xref: /linux/arch/riscv/mm/fault.c (revision 60684c2bd35064043360e6f716d1b7c20e967b7d)
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 
19 #include <asm/ptrace.h>
20 #include <asm/tlbflush.h>
21 
22 #include "../kernel/head.h"
23 
24 static void die_kernel_fault(const char *msg, unsigned long addr,
25 		struct pt_regs *regs)
26 {
27 	bust_spinlocks(1);
28 
29 	pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n", msg,
30 		addr);
31 
32 	bust_spinlocks(0);
33 	die(regs, "Oops");
34 	make_task_dead(SIGKILL);
35 }
36 
37 static inline void no_context(struct pt_regs *regs, unsigned long addr)
38 {
39 	const char *msg;
40 
41 	/* Are we prepared to handle this kernel fault? */
42 	if (fixup_exception(regs))
43 		return;
44 
45 	/*
46 	 * Oops. The kernel tried to access some bad page. We'll have to
47 	 * terminate things with extreme prejudice.
48 	 */
49 	if (addr < PAGE_SIZE)
50 		msg = "NULL pointer dereference";
51 	else {
52 		if (kfence_handle_page_fault(addr, regs->cause == EXC_STORE_PAGE_FAULT, regs))
53 			return;
54 
55 		msg = "paging request";
56 	}
57 
58 	die_kernel_fault(msg, addr, regs);
59 }
60 
61 static inline void mm_fault_error(struct pt_regs *regs, unsigned long addr, vm_fault_t fault)
62 {
63 	if (fault & VM_FAULT_OOM) {
64 		/*
65 		 * We ran out of memory, call the OOM killer, and return the userspace
66 		 * (which will retry the fault, or kill us if we got oom-killed).
67 		 */
68 		if (!user_mode(regs)) {
69 			no_context(regs, addr);
70 			return;
71 		}
72 		pagefault_out_of_memory();
73 		return;
74 	} else if (fault & VM_FAULT_SIGBUS) {
75 		/* Kernel mode? Handle exceptions or die */
76 		if (!user_mode(regs)) {
77 			no_context(regs, addr);
78 			return;
79 		}
80 		do_trap(regs, SIGBUS, BUS_ADRERR, addr);
81 		return;
82 	}
83 	BUG();
84 }
85 
86 static inline void bad_area(struct pt_regs *regs, struct mm_struct *mm, int code, unsigned long addr)
87 {
88 	/*
89 	 * Something tried to access memory that isn't in our memory map.
90 	 * Fix it, but check if it's kernel or user first.
91 	 */
92 	mmap_read_unlock(mm);
93 	/* User mode accesses just cause a SIGSEGV */
94 	if (user_mode(regs)) {
95 		do_trap(regs, SIGSEGV, code, addr);
96 		return;
97 	}
98 
99 	no_context(regs, addr);
100 }
101 
102 static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr)
103 {
104 	pgd_t *pgd, *pgd_k;
105 	pud_t *pud_k;
106 	p4d_t *p4d_k;
107 	pmd_t *pmd_k;
108 	pte_t *pte_k;
109 	int index;
110 	unsigned long pfn;
111 
112 	/* User mode accesses just cause a SIGSEGV */
113 	if (user_mode(regs))
114 		return do_trap(regs, SIGSEGV, code, addr);
115 
116 	/*
117 	 * Synchronize this task's top level page-table
118 	 * with the 'reference' page table.
119 	 *
120 	 * Do _not_ use "tsk->active_mm->pgd" here.
121 	 * We might be inside an interrupt in the middle
122 	 * of a task switch.
123 	 */
124 	index = pgd_index(addr);
125 	pfn = csr_read(CSR_SATP) & SATP_PPN;
126 	pgd = (pgd_t *)pfn_to_virt(pfn) + index;
127 	pgd_k = init_mm.pgd + index;
128 
129 	if (!pgd_present(*pgd_k)) {
130 		no_context(regs, addr);
131 		return;
132 	}
133 	set_pgd(pgd, *pgd_k);
134 
135 	p4d_k = p4d_offset(pgd_k, addr);
136 	if (!p4d_present(*p4d_k)) {
137 		no_context(regs, addr);
138 		return;
139 	}
140 
141 	pud_k = pud_offset(p4d_k, addr);
142 	if (!pud_present(*pud_k)) {
143 		no_context(regs, addr);
144 		return;
145 	}
146 
147 	/*
148 	 * Since the vmalloc area is global, it is unnecessary
149 	 * to copy individual PTEs
150 	 */
151 	pmd_k = pmd_offset(pud_k, addr);
152 	if (!pmd_present(*pmd_k)) {
153 		no_context(regs, addr);
154 		return;
155 	}
156 
157 	/*
158 	 * Make sure the actual PTE exists as well to
159 	 * catch kernel vmalloc-area accesses to non-mapped
160 	 * addresses. If we don't do this, this will just
161 	 * silently loop forever.
162 	 */
163 	pte_k = pte_offset_kernel(pmd_k, addr);
164 	if (!pte_present(*pte_k)) {
165 		no_context(regs, addr);
166 		return;
167 	}
168 
169 	/*
170 	 * The kernel assumes that TLBs don't cache invalid
171 	 * entries, but in RISC-V, SFENCE.VMA specifies an
172 	 * ordering constraint, not a cache flush; it is
173 	 * necessary even after writing invalid entries.
174 	 */
175 	local_flush_tlb_page(addr);
176 }
177 
178 static inline bool access_error(unsigned long cause, struct vm_area_struct *vma)
179 {
180 	switch (cause) {
181 	case EXC_INST_PAGE_FAULT:
182 		if (!(vma->vm_flags & VM_EXEC)) {
183 			return true;
184 		}
185 		break;
186 	case EXC_LOAD_PAGE_FAULT:
187 		/* Write implies read */
188 		if (!(vma->vm_flags & (VM_READ | VM_WRITE))) {
189 			return true;
190 		}
191 		break;
192 	case EXC_STORE_PAGE_FAULT:
193 		if (!(vma->vm_flags & VM_WRITE)) {
194 			return true;
195 		}
196 		break;
197 	default:
198 		panic("%s: unhandled cause %lu", __func__, cause);
199 	}
200 	return false;
201 }
202 
203 /*
204  * This routine handles page faults.  It determines the address and the
205  * problem, and then passes it off to one of the appropriate routines.
206  */
207 asmlinkage void do_page_fault(struct pt_regs *regs)
208 {
209 	struct task_struct *tsk;
210 	struct vm_area_struct *vma;
211 	struct mm_struct *mm;
212 	unsigned long addr, cause;
213 	unsigned int flags = FAULT_FLAG_DEFAULT;
214 	int code = SEGV_MAPERR;
215 	vm_fault_t fault;
216 
217 	cause = regs->cause;
218 	addr = regs->badaddr;
219 
220 	tsk = current;
221 	mm = tsk->mm;
222 
223 	if (kprobe_page_fault(regs, cause))
224 		return;
225 
226 	/*
227 	 * Fault-in kernel-space virtual memory on-demand.
228 	 * The 'reference' page table is init_mm.pgd.
229 	 *
230 	 * NOTE! We MUST NOT take any locks for this case. We may
231 	 * be in an interrupt or a critical region, and should
232 	 * only copy the information from the master page table,
233 	 * nothing more.
234 	 */
235 	if (unlikely((addr >= VMALLOC_START) && (addr < VMALLOC_END))) {
236 		vmalloc_fault(regs, code, addr);
237 		return;
238 	}
239 
240 #ifdef CONFIG_64BIT
241 	/*
242 	 * Modules in 64bit kernels lie in their own virtual region which is not
243 	 * in the vmalloc region, but dealing with page faults in this region
244 	 * or the vmalloc region amounts to doing the same thing: checking that
245 	 * the mapping exists in init_mm.pgd and updating user page table, so
246 	 * just use vmalloc_fault.
247 	 */
248 	if (unlikely(addr >= MODULES_VADDR && addr < MODULES_END)) {
249 		vmalloc_fault(regs, code, addr);
250 		return;
251 	}
252 #endif
253 	/* Enable interrupts if they were enabled in the parent context. */
254 	if (likely(regs->status & SR_PIE))
255 		local_irq_enable();
256 
257 	/*
258 	 * If we're in an interrupt, have no user context, or are running
259 	 * in an atomic region, then we must not take the fault.
260 	 */
261 	if (unlikely(faulthandler_disabled() || !mm)) {
262 		tsk->thread.bad_cause = cause;
263 		no_context(regs, addr);
264 		return;
265 	}
266 
267 	if (user_mode(regs))
268 		flags |= FAULT_FLAG_USER;
269 
270 	if (!user_mode(regs) && addr < TASK_SIZE && unlikely(!(regs->status & SR_SUM))) {
271 		if (fixup_exception(regs))
272 			return;
273 
274 		die_kernel_fault("access to user memory without uaccess routines", addr, regs);
275 	}
276 
277 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
278 
279 	if (cause == EXC_STORE_PAGE_FAULT)
280 		flags |= FAULT_FLAG_WRITE;
281 	else if (cause == EXC_INST_PAGE_FAULT)
282 		flags |= FAULT_FLAG_INSTRUCTION;
283 retry:
284 	mmap_read_lock(mm);
285 	vma = find_vma(mm, addr);
286 	if (unlikely(!vma)) {
287 		tsk->thread.bad_cause = cause;
288 		bad_area(regs, mm, code, addr);
289 		return;
290 	}
291 	if (likely(vma->vm_start <= addr))
292 		goto good_area;
293 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
294 		tsk->thread.bad_cause = cause;
295 		bad_area(regs, mm, code, addr);
296 		return;
297 	}
298 	if (unlikely(expand_stack(vma, addr))) {
299 		tsk->thread.bad_cause = cause;
300 		bad_area(regs, mm, code, addr);
301 		return;
302 	}
303 
304 	/*
305 	 * Ok, we have a good vm_area for this memory access, so
306 	 * we can handle it.
307 	 */
308 good_area:
309 	code = SEGV_ACCERR;
310 
311 	if (unlikely(access_error(cause, vma))) {
312 		tsk->thread.bad_cause = cause;
313 		bad_area(regs, mm, code, addr);
314 		return;
315 	}
316 
317 	/*
318 	 * If for any reason at all we could not handle the fault,
319 	 * make sure we exit gracefully rather than endlessly redo
320 	 * the fault.
321 	 */
322 	fault = handle_mm_fault(vma, addr, flags, regs);
323 
324 	/*
325 	 * If we need to retry but a fatal signal is pending, handle the
326 	 * signal first. We do not need to release the mmap_lock because it
327 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
328 	 */
329 	if (fault_signal_pending(fault, regs)) {
330 		if (!user_mode(regs))
331 			no_context(regs, addr);
332 		return;
333 	}
334 
335 	/* The fault is fully completed (including releasing mmap lock) */
336 	if (fault & VM_FAULT_COMPLETED)
337 		return;
338 
339 	if (unlikely(fault & VM_FAULT_RETRY)) {
340 		flags |= FAULT_FLAG_TRIED;
341 
342 		/*
343 		 * No need to mmap_read_unlock(mm) as we would
344 		 * have already released it in __lock_page_or_retry
345 		 * in mm/filemap.c.
346 		 */
347 		goto retry;
348 	}
349 
350 	mmap_read_unlock(mm);
351 
352 	if (unlikely(fault & VM_FAULT_ERROR)) {
353 		tsk->thread.bad_cause = cause;
354 		mm_fault_error(regs, addr, fault);
355 		return;
356 	}
357 	return;
358 }
359 NOKPROBE_SYMBOL(do_page_fault);
360