xref: /openbmc/linux/arch/riscv/mm/fault.c (revision 0545810f)
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 bad_area(struct pt_regs *regs, struct mm_struct *mm, int code, unsigned long addr)
88 {
89 	/*
90 	 * Something tried to access memory that isn't in our memory map.
91 	 * Fix it, but check if it's kernel or user first.
92 	 */
93 	mmap_read_unlock(mm);
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 vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr)
104 {
105 	pgd_t *pgd, *pgd_k;
106 	pud_t *pud_k;
107 	p4d_t *p4d_k;
108 	pmd_t *pmd_k;
109 	pte_t *pte_k;
110 	int index;
111 	unsigned long pfn;
112 
113 	/* User mode accesses just cause a SIGSEGV */
114 	if (user_mode(regs))
115 		return do_trap(regs, SIGSEGV, code, addr);
116 
117 	/*
118 	 * Synchronize this task's top level page-table
119 	 * with the 'reference' page table.
120 	 *
121 	 * Do _not_ use "tsk->active_mm->pgd" here.
122 	 * We might be inside an interrupt in the middle
123 	 * of a task switch.
124 	 */
125 	index = pgd_index(addr);
126 	pfn = csr_read(CSR_SATP) & SATP_PPN;
127 	pgd = (pgd_t *)pfn_to_virt(pfn) + index;
128 	pgd_k = init_mm.pgd + index;
129 
130 	if (!pgd_present(*pgd_k)) {
131 		no_context(regs, addr);
132 		return;
133 	}
134 	set_pgd(pgd, *pgd_k);
135 
136 	p4d_k = p4d_offset(pgd_k, addr);
137 	if (!p4d_present(*p4d_k)) {
138 		no_context(regs, addr);
139 		return;
140 	}
141 
142 	pud_k = pud_offset(p4d_k, addr);
143 	if (!pud_present(*pud_k)) {
144 		no_context(regs, addr);
145 		return;
146 	}
147 	if (pud_leaf(*pud_k))
148 		goto flush_tlb;
149 
150 	/*
151 	 * Since the vmalloc area is global, it is unnecessary
152 	 * to copy individual PTEs
153 	 */
154 	pmd_k = pmd_offset(pud_k, addr);
155 	if (!pmd_present(*pmd_k)) {
156 		no_context(regs, addr);
157 		return;
158 	}
159 	if (pmd_leaf(*pmd_k))
160 		goto flush_tlb;
161 
162 	/*
163 	 * Make sure the actual PTE exists as well to
164 	 * catch kernel vmalloc-area accesses to non-mapped
165 	 * addresses. If we don't do this, this will just
166 	 * silently loop forever.
167 	 */
168 	pte_k = pte_offset_kernel(pmd_k, addr);
169 	if (!pte_present(*pte_k)) {
170 		no_context(regs, addr);
171 		return;
172 	}
173 
174 	/*
175 	 * The kernel assumes that TLBs don't cache invalid
176 	 * entries, but in RISC-V, SFENCE.VMA specifies an
177 	 * ordering constraint, not a cache flush; it is
178 	 * necessary even after writing invalid entries.
179 	 */
180 flush_tlb:
181 	local_flush_tlb_page(addr);
182 }
183 
184 static inline bool access_error(unsigned long cause, struct vm_area_struct *vma)
185 {
186 	switch (cause) {
187 	case EXC_INST_PAGE_FAULT:
188 		if (!(vma->vm_flags & VM_EXEC)) {
189 			return true;
190 		}
191 		break;
192 	case EXC_LOAD_PAGE_FAULT:
193 		/* Write implies read */
194 		if (!(vma->vm_flags & (VM_READ | VM_WRITE))) {
195 			return true;
196 		}
197 		break;
198 	case EXC_STORE_PAGE_FAULT:
199 		if (!(vma->vm_flags & VM_WRITE)) {
200 			return true;
201 		}
202 		break;
203 	default:
204 		panic("%s: unhandled cause %lu", __func__, cause);
205 	}
206 	return false;
207 }
208 
209 /*
210  * This routine handles page faults.  It determines the address and the
211  * problem, and then passes it off to one of the appropriate routines.
212  */
213 void handle_page_fault(struct pt_regs *regs)
214 {
215 	struct task_struct *tsk;
216 	struct vm_area_struct *vma;
217 	struct mm_struct *mm;
218 	unsigned long addr, cause;
219 	unsigned int flags = FAULT_FLAG_DEFAULT;
220 	int code = SEGV_MAPERR;
221 	vm_fault_t fault;
222 
223 	cause = regs->cause;
224 	addr = regs->badaddr;
225 
226 	tsk = current;
227 	mm = tsk->mm;
228 
229 	if (kprobe_page_fault(regs, cause))
230 		return;
231 
232 	/*
233 	 * Fault-in kernel-space virtual memory on-demand.
234 	 * The 'reference' page table is init_mm.pgd.
235 	 *
236 	 * NOTE! We MUST NOT take any locks for this case. We may
237 	 * be in an interrupt or a critical region, and should
238 	 * only copy the information from the master page table,
239 	 * nothing more.
240 	 */
241 	if (unlikely((addr >= VMALLOC_START) && (addr < VMALLOC_END))) {
242 		vmalloc_fault(regs, code, addr);
243 		return;
244 	}
245 
246 #ifdef CONFIG_64BIT
247 	/*
248 	 * Modules in 64bit kernels lie in their own virtual region which is not
249 	 * in the vmalloc region, but dealing with page faults in this region
250 	 * or the vmalloc region amounts to doing the same thing: checking that
251 	 * the mapping exists in init_mm.pgd and updating user page table, so
252 	 * just use vmalloc_fault.
253 	 */
254 	if (unlikely(addr >= MODULES_VADDR && addr < MODULES_END)) {
255 		vmalloc_fault(regs, code, addr);
256 		return;
257 	}
258 #endif
259 	/* Enable interrupts if they were enabled in the parent context. */
260 	if (!regs_irqs_disabled(regs))
261 		local_irq_enable();
262 
263 	/*
264 	 * If we're in an interrupt, have no user context, or are running
265 	 * in an atomic region, then we must not take the fault.
266 	 */
267 	if (unlikely(faulthandler_disabled() || !mm)) {
268 		tsk->thread.bad_cause = cause;
269 		no_context(regs, addr);
270 		return;
271 	}
272 
273 	if (user_mode(regs))
274 		flags |= FAULT_FLAG_USER;
275 
276 	if (!user_mode(regs) && addr < TASK_SIZE && unlikely(!(regs->status & SR_SUM))) {
277 		if (fixup_exception(regs))
278 			return;
279 
280 		die_kernel_fault("access to user memory without uaccess routines", addr, regs);
281 	}
282 
283 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
284 
285 	if (cause == EXC_STORE_PAGE_FAULT)
286 		flags |= FAULT_FLAG_WRITE;
287 	else if (cause == EXC_INST_PAGE_FAULT)
288 		flags |= FAULT_FLAG_INSTRUCTION;
289 retry:
290 	mmap_read_lock(mm);
291 	vma = find_vma(mm, addr);
292 	if (unlikely(!vma)) {
293 		tsk->thread.bad_cause = cause;
294 		bad_area(regs, mm, code, addr);
295 		return;
296 	}
297 	if (likely(vma->vm_start <= addr))
298 		goto good_area;
299 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
300 		tsk->thread.bad_cause = cause;
301 		bad_area(regs, mm, code, addr);
302 		return;
303 	}
304 	if (unlikely(expand_stack(vma, addr))) {
305 		tsk->thread.bad_cause = cause;
306 		bad_area(regs, mm, code, addr);
307 		return;
308 	}
309 
310 	/*
311 	 * Ok, we have a good vm_area for this memory access, so
312 	 * we can handle it.
313 	 */
314 good_area:
315 	code = SEGV_ACCERR;
316 
317 	if (unlikely(access_error(cause, vma))) {
318 		tsk->thread.bad_cause = cause;
319 		bad_area(regs, mm, code, addr);
320 		return;
321 	}
322 
323 	/*
324 	 * If for any reason at all we could not handle the fault,
325 	 * make sure we exit gracefully rather than endlessly redo
326 	 * the fault.
327 	 */
328 	fault = handle_mm_fault(vma, addr, flags, regs);
329 
330 	/*
331 	 * If we need to retry but a fatal signal is pending, handle the
332 	 * signal first. We do not need to release the mmap_lock because it
333 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
334 	 */
335 	if (fault_signal_pending(fault, regs)) {
336 		if (!user_mode(regs))
337 			no_context(regs, addr);
338 		return;
339 	}
340 
341 	/* The fault is fully completed (including releasing mmap lock) */
342 	if (fault & VM_FAULT_COMPLETED)
343 		return;
344 
345 	if (unlikely(fault & VM_FAULT_RETRY)) {
346 		flags |= FAULT_FLAG_TRIED;
347 
348 		/*
349 		 * No need to mmap_read_unlock(mm) as we would
350 		 * have already released it in __lock_page_or_retry
351 		 * in mm/filemap.c.
352 		 */
353 		goto retry;
354 	}
355 
356 	mmap_read_unlock(mm);
357 
358 	if (unlikely(fault & VM_FAULT_ERROR)) {
359 		tsk->thread.bad_cause = cause;
360 		mm_fault_error(regs, addr, fault);
361 		return;
362 	}
363 	return;
364 }
365