xref: /openbmc/linux/arch/riscv/mm/fault.c (revision 877013bc)
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 
18 #include <asm/ptrace.h>
19 #include <asm/tlbflush.h>
20 
21 #include "../kernel/head.h"
22 
23 static void die_kernel_fault(const char *msg, unsigned long addr,
24 		struct pt_regs *regs)
25 {
26 	bust_spinlocks(1);
27 
28 	pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n", msg,
29 		addr);
30 
31 	bust_spinlocks(0);
32 	die(regs, "Oops");
33 	do_exit(SIGKILL);
34 }
35 
36 static inline void no_context(struct pt_regs *regs, unsigned long addr)
37 {
38 	const char *msg;
39 
40 	/* Are we prepared to handle this kernel fault? */
41 	if (fixup_exception(regs))
42 		return;
43 
44 	/*
45 	 * Oops. The kernel tried to access some bad page. We'll have to
46 	 * terminate things with extreme prejudice.
47 	 */
48 	msg = (addr < PAGE_SIZE) ? "NULL pointer dereference" : "paging request";
49 	die_kernel_fault(msg, addr, regs);
50 }
51 
52 static inline void mm_fault_error(struct pt_regs *regs, unsigned long addr, vm_fault_t fault)
53 {
54 	if (fault & VM_FAULT_OOM) {
55 		/*
56 		 * We ran out of memory, call the OOM killer, and return the userspace
57 		 * (which will retry the fault, or kill us if we got oom-killed).
58 		 */
59 		if (!user_mode(regs)) {
60 			no_context(regs, addr);
61 			return;
62 		}
63 		pagefault_out_of_memory();
64 		return;
65 	} else if (fault & VM_FAULT_SIGBUS) {
66 		/* Kernel mode? Handle exceptions or die */
67 		if (!user_mode(regs)) {
68 			no_context(regs, addr);
69 			return;
70 		}
71 		do_trap(regs, SIGBUS, BUS_ADRERR, addr);
72 		return;
73 	}
74 	BUG();
75 }
76 
77 static inline void bad_area(struct pt_regs *regs, struct mm_struct *mm, int code, unsigned long addr)
78 {
79 	/*
80 	 * Something tried to access memory that isn't in our memory map.
81 	 * Fix it, but check if it's kernel or user first.
82 	 */
83 	mmap_read_unlock(mm);
84 	/* User mode accesses just cause a SIGSEGV */
85 	if (user_mode(regs)) {
86 		do_trap(regs, SIGSEGV, code, addr);
87 		return;
88 	}
89 
90 	no_context(regs, addr);
91 }
92 
93 static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr)
94 {
95 	pgd_t *pgd, *pgd_k;
96 	pud_t *pud, *pud_k;
97 	p4d_t *p4d, *p4d_k;
98 	pmd_t *pmd, *pmd_k;
99 	pte_t *pte_k;
100 	int index;
101 	unsigned long pfn;
102 
103 	/* User mode accesses just cause a SIGSEGV */
104 	if (user_mode(regs))
105 		return do_trap(regs, SIGSEGV, code, addr);
106 
107 	/*
108 	 * Synchronize this task's top level page-table
109 	 * with the 'reference' page table.
110 	 *
111 	 * Do _not_ use "tsk->active_mm->pgd" here.
112 	 * We might be inside an interrupt in the middle
113 	 * of a task switch.
114 	 */
115 	index = pgd_index(addr);
116 	pfn = csr_read(CSR_SATP) & SATP_PPN;
117 	pgd = (pgd_t *)pfn_to_virt(pfn) + index;
118 	pgd_k = init_mm.pgd + index;
119 
120 	if (!pgd_present(*pgd_k)) {
121 		no_context(regs, addr);
122 		return;
123 	}
124 	set_pgd(pgd, *pgd_k);
125 
126 	p4d = p4d_offset(pgd, addr);
127 	p4d_k = p4d_offset(pgd_k, addr);
128 	if (!p4d_present(*p4d_k)) {
129 		no_context(regs, addr);
130 		return;
131 	}
132 
133 	pud = pud_offset(p4d, addr);
134 	pud_k = pud_offset(p4d_k, addr);
135 	if (!pud_present(*pud_k)) {
136 		no_context(regs, addr);
137 		return;
138 	}
139 
140 	/*
141 	 * Since the vmalloc area is global, it is unnecessary
142 	 * to copy individual PTEs
143 	 */
144 	pmd = pmd_offset(pud, addr);
145 	pmd_k = pmd_offset(pud_k, addr);
146 	if (!pmd_present(*pmd_k)) {
147 		no_context(regs, addr);
148 		return;
149 	}
150 	set_pmd(pmd, *pmd_k);
151 
152 	/*
153 	 * Make sure the actual PTE exists as well to
154 	 * catch kernel vmalloc-area accesses to non-mapped
155 	 * addresses. If we don't do this, this will just
156 	 * silently loop forever.
157 	 */
158 	pte_k = pte_offset_kernel(pmd_k, addr);
159 	if (!pte_present(*pte_k)) {
160 		no_context(regs, addr);
161 		return;
162 	}
163 
164 	/*
165 	 * The kernel assumes that TLBs don't cache invalid
166 	 * entries, but in RISC-V, SFENCE.VMA specifies an
167 	 * ordering constraint, not a cache flush; it is
168 	 * necessary even after writing invalid entries.
169 	 */
170 	local_flush_tlb_page(addr);
171 }
172 
173 static inline bool access_error(unsigned long cause, struct vm_area_struct *vma)
174 {
175 	switch (cause) {
176 	case EXC_INST_PAGE_FAULT:
177 		if (!(vma->vm_flags & VM_EXEC)) {
178 			return true;
179 		}
180 		break;
181 	case EXC_LOAD_PAGE_FAULT:
182 		if (!(vma->vm_flags & VM_READ)) {
183 			return true;
184 		}
185 		break;
186 	case EXC_STORE_PAGE_FAULT:
187 		if (!(vma->vm_flags & VM_WRITE)) {
188 			return true;
189 		}
190 		break;
191 	default:
192 		panic("%s: unhandled cause %lu", __func__, cause);
193 	}
194 	return false;
195 }
196 
197 /*
198  * This routine handles page faults.  It determines the address and the
199  * problem, and then passes it off to one of the appropriate routines.
200  */
201 asmlinkage void do_page_fault(struct pt_regs *regs)
202 {
203 	struct task_struct *tsk;
204 	struct vm_area_struct *vma;
205 	struct mm_struct *mm;
206 	unsigned long addr, cause;
207 	unsigned int flags = FAULT_FLAG_DEFAULT;
208 	int code = SEGV_MAPERR;
209 	vm_fault_t fault;
210 
211 	cause = regs->cause;
212 	addr = regs->badaddr;
213 
214 	tsk = current;
215 	mm = tsk->mm;
216 
217 	if (kprobe_page_fault(regs, cause))
218 		return;
219 
220 	/*
221 	 * Fault-in kernel-space virtual memory on-demand.
222 	 * The 'reference' page table is init_mm.pgd.
223 	 *
224 	 * NOTE! We MUST NOT take any locks for this case. We may
225 	 * be in an interrupt or a critical region, and should
226 	 * only copy the information from the master page table,
227 	 * nothing more.
228 	 */
229 	if (unlikely((addr >= VMALLOC_START) && (addr <= VMALLOC_END))) {
230 		vmalloc_fault(regs, code, addr);
231 		return;
232 	}
233 
234 #ifdef CONFIG_64BIT
235 	/*
236 	 * Modules in 64bit kernels lie in their own virtual region which is not
237 	 * in the vmalloc region, but dealing with page faults in this region
238 	 * or the vmalloc region amounts to doing the same thing: checking that
239 	 * the mapping exists in init_mm.pgd and updating user page table, so
240 	 * just use vmalloc_fault.
241 	 */
242 	if (unlikely(addr >= MODULES_VADDR && addr < MODULES_END)) {
243 		vmalloc_fault(regs, code, addr);
244 		return;
245 	}
246 #endif
247 	/* Enable interrupts if they were enabled in the parent context. */
248 	if (likely(regs->status & SR_PIE))
249 		local_irq_enable();
250 
251 	/*
252 	 * If we're in an interrupt, have no user context, or are running
253 	 * in an atomic region, then we must not take the fault.
254 	 */
255 	if (unlikely(faulthandler_disabled() || !mm)) {
256 		tsk->thread.bad_cause = cause;
257 		no_context(regs, addr);
258 		return;
259 	}
260 
261 	if (user_mode(regs))
262 		flags |= FAULT_FLAG_USER;
263 
264 	if (!user_mode(regs) && addr < TASK_SIZE &&
265 			unlikely(!(regs->status & SR_SUM)))
266 		die_kernel_fault("access to user memory without uaccess routines",
267 				addr, regs);
268 
269 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
270 
271 	if (cause == EXC_STORE_PAGE_FAULT)
272 		flags |= FAULT_FLAG_WRITE;
273 	else if (cause == EXC_INST_PAGE_FAULT)
274 		flags |= FAULT_FLAG_INSTRUCTION;
275 retry:
276 	mmap_read_lock(mm);
277 	vma = find_vma(mm, addr);
278 	if (unlikely(!vma)) {
279 		tsk->thread.bad_cause = cause;
280 		bad_area(regs, mm, code, addr);
281 		return;
282 	}
283 	if (likely(vma->vm_start <= addr))
284 		goto good_area;
285 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
286 		tsk->thread.bad_cause = cause;
287 		bad_area(regs, mm, code, addr);
288 		return;
289 	}
290 	if (unlikely(expand_stack(vma, addr))) {
291 		tsk->thread.bad_cause = cause;
292 		bad_area(regs, mm, code, addr);
293 		return;
294 	}
295 
296 	/*
297 	 * Ok, we have a good vm_area for this memory access, so
298 	 * we can handle it.
299 	 */
300 good_area:
301 	code = SEGV_ACCERR;
302 
303 	if (unlikely(access_error(cause, vma))) {
304 		tsk->thread.bad_cause = cause;
305 		bad_area(regs, mm, code, addr);
306 		return;
307 	}
308 
309 	/*
310 	 * If for any reason at all we could not handle the fault,
311 	 * make sure we exit gracefully rather than endlessly redo
312 	 * the fault.
313 	 */
314 	fault = handle_mm_fault(vma, addr, flags, regs);
315 
316 	/*
317 	 * If we need to retry but a fatal signal is pending, handle the
318 	 * signal first. We do not need to release the mmap_lock because it
319 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
320 	 */
321 	if (fault_signal_pending(fault, regs))
322 		return;
323 
324 	if (unlikely((fault & VM_FAULT_RETRY) && (flags & FAULT_FLAG_ALLOW_RETRY))) {
325 		flags |= FAULT_FLAG_TRIED;
326 
327 		/*
328 		 * No need to mmap_read_unlock(mm) as we would
329 		 * have already released it in __lock_page_or_retry
330 		 * in mm/filemap.c.
331 		 */
332 		goto retry;
333 	}
334 
335 	mmap_read_unlock(mm);
336 
337 	if (unlikely(fault & VM_FAULT_ERROR)) {
338 		tsk->thread.bad_cause = cause;
339 		mm_fault_error(regs, addr, fault);
340 		return;
341 	}
342 	return;
343 }
344 NOKPROBE_SYMBOL(do_page_fault);
345