xref: /openbmc/linux/arch/riscv/mm/fault.c (revision 1edd0337)
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 	if (pud_leaf(*pud_k))
147 		goto flush_tlb;
148 
149 	/*
150 	 * Since the vmalloc area is global, it is unnecessary
151 	 * to copy individual PTEs
152 	 */
153 	pmd_k = pmd_offset(pud_k, addr);
154 	if (!pmd_present(*pmd_k)) {
155 		no_context(regs, addr);
156 		return;
157 	}
158 	if (pmd_leaf(*pmd_k))
159 		goto flush_tlb;
160 
161 	/*
162 	 * Make sure the actual PTE exists as well to
163 	 * catch kernel vmalloc-area accesses to non-mapped
164 	 * addresses. If we don't do this, this will just
165 	 * silently loop forever.
166 	 */
167 	pte_k = pte_offset_kernel(pmd_k, addr);
168 	if (!pte_present(*pte_k)) {
169 		no_context(regs, addr);
170 		return;
171 	}
172 
173 	/*
174 	 * The kernel assumes that TLBs don't cache invalid
175 	 * entries, but in RISC-V, SFENCE.VMA specifies an
176 	 * ordering constraint, not a cache flush; it is
177 	 * necessary even after writing invalid entries.
178 	 */
179 flush_tlb:
180 	local_flush_tlb_page(addr);
181 }
182 
183 static inline bool access_error(unsigned long cause, struct vm_area_struct *vma)
184 {
185 	switch (cause) {
186 	case EXC_INST_PAGE_FAULT:
187 		if (!(vma->vm_flags & VM_EXEC)) {
188 			return true;
189 		}
190 		break;
191 	case EXC_LOAD_PAGE_FAULT:
192 		/* Write implies read */
193 		if (!(vma->vm_flags & (VM_READ | VM_WRITE))) {
194 			return true;
195 		}
196 		break;
197 	case EXC_STORE_PAGE_FAULT:
198 		if (!(vma->vm_flags & VM_WRITE)) {
199 			return true;
200 		}
201 		break;
202 	default:
203 		panic("%s: unhandled cause %lu", __func__, cause);
204 	}
205 	return false;
206 }
207 
208 /*
209  * This routine handles page faults.  It determines the address and the
210  * problem, and then passes it off to one of the appropriate routines.
211  */
212 asmlinkage void do_page_fault(struct pt_regs *regs)
213 {
214 	struct task_struct *tsk;
215 	struct vm_area_struct *vma;
216 	struct mm_struct *mm;
217 	unsigned long addr, cause;
218 	unsigned int flags = FAULT_FLAG_DEFAULT;
219 	int code = SEGV_MAPERR;
220 	vm_fault_t fault;
221 
222 	cause = regs->cause;
223 	addr = regs->badaddr;
224 
225 	tsk = current;
226 	mm = tsk->mm;
227 
228 	if (kprobe_page_fault(regs, cause))
229 		return;
230 
231 	/*
232 	 * Fault-in kernel-space virtual memory on-demand.
233 	 * The 'reference' page table is init_mm.pgd.
234 	 *
235 	 * NOTE! We MUST NOT take any locks for this case. We may
236 	 * be in an interrupt or a critical region, and should
237 	 * only copy the information from the master page table,
238 	 * nothing more.
239 	 */
240 	if (unlikely((addr >= VMALLOC_START) && (addr < VMALLOC_END))) {
241 		vmalloc_fault(regs, code, addr);
242 		return;
243 	}
244 
245 #ifdef CONFIG_64BIT
246 	/*
247 	 * Modules in 64bit kernels lie in their own virtual region which is not
248 	 * in the vmalloc region, but dealing with page faults in this region
249 	 * or the vmalloc region amounts to doing the same thing: checking that
250 	 * the mapping exists in init_mm.pgd and updating user page table, so
251 	 * just use vmalloc_fault.
252 	 */
253 	if (unlikely(addr >= MODULES_VADDR && addr < MODULES_END)) {
254 		vmalloc_fault(regs, code, addr);
255 		return;
256 	}
257 #endif
258 	/* Enable interrupts if they were enabled in the parent context. */
259 	if (likely(regs->status & SR_PIE))
260 		local_irq_enable();
261 
262 	/*
263 	 * If we're in an interrupt, have no user context, or are running
264 	 * in an atomic region, then we must not take the fault.
265 	 */
266 	if (unlikely(faulthandler_disabled() || !mm)) {
267 		tsk->thread.bad_cause = cause;
268 		no_context(regs, addr);
269 		return;
270 	}
271 
272 	if (user_mode(regs))
273 		flags |= FAULT_FLAG_USER;
274 
275 	if (!user_mode(regs) && addr < TASK_SIZE && unlikely(!(regs->status & SR_SUM))) {
276 		if (fixup_exception(regs))
277 			return;
278 
279 		die_kernel_fault("access to user memory without uaccess routines", addr, regs);
280 	}
281 
282 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
283 
284 	if (cause == EXC_STORE_PAGE_FAULT)
285 		flags |= FAULT_FLAG_WRITE;
286 	else if (cause == EXC_INST_PAGE_FAULT)
287 		flags |= FAULT_FLAG_INSTRUCTION;
288 retry:
289 	mmap_read_lock(mm);
290 	vma = find_vma(mm, addr);
291 	if (unlikely(!vma)) {
292 		tsk->thread.bad_cause = cause;
293 		bad_area(regs, mm, code, addr);
294 		return;
295 	}
296 	if (likely(vma->vm_start <= addr))
297 		goto good_area;
298 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
299 		tsk->thread.bad_cause = cause;
300 		bad_area(regs, mm, code, addr);
301 		return;
302 	}
303 	if (unlikely(expand_stack(vma, addr))) {
304 		tsk->thread.bad_cause = cause;
305 		bad_area(regs, mm, code, addr);
306 		return;
307 	}
308 
309 	/*
310 	 * Ok, we have a good vm_area for this memory access, so
311 	 * we can handle it.
312 	 */
313 good_area:
314 	code = SEGV_ACCERR;
315 
316 	if (unlikely(access_error(cause, vma))) {
317 		tsk->thread.bad_cause = cause;
318 		bad_area(regs, mm, code, addr);
319 		return;
320 	}
321 
322 	/*
323 	 * If for any reason at all we could not handle the fault,
324 	 * make sure we exit gracefully rather than endlessly redo
325 	 * the fault.
326 	 */
327 	fault = handle_mm_fault(vma, addr, flags, regs);
328 
329 	/*
330 	 * If we need to retry but a fatal signal is pending, handle the
331 	 * signal first. We do not need to release the mmap_lock because it
332 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
333 	 */
334 	if (fault_signal_pending(fault, regs)) {
335 		if (!user_mode(regs))
336 			no_context(regs, addr);
337 		return;
338 	}
339 
340 	/* The fault is fully completed (including releasing mmap lock) */
341 	if (fault & VM_FAULT_COMPLETED)
342 		return;
343 
344 	if (unlikely(fault & VM_FAULT_RETRY)) {
345 		flags |= FAULT_FLAG_TRIED;
346 
347 		/*
348 		 * No need to mmap_read_unlock(mm) as we would
349 		 * have already released it in __lock_page_or_retry
350 		 * in mm/filemap.c.
351 		 */
352 		goto retry;
353 	}
354 
355 	mmap_read_unlock(mm);
356 
357 	if (unlikely(fault & VM_FAULT_ERROR)) {
358 		tsk->thread.bad_cause = cause;
359 		mm_fault_error(regs, addr, fault);
360 		return;
361 	}
362 	return;
363 }
364 NOKPROBE_SYMBOL(do_page_fault);
365