xref: /linux/arch/riscv/mm/fault.c (revision b7019ac550eb3916f34d79db583e9b7ea2524afa)
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 
17 #include <asm/pgalloc.h>
18 #include <asm/ptrace.h>
19 
20 /*
21  * This routine handles page faults.  It determines the address and the
22  * problem, and then passes it off to one of the appropriate routines.
23  */
24 asmlinkage void do_page_fault(struct pt_regs *regs)
25 {
26 	struct task_struct *tsk;
27 	struct vm_area_struct *vma;
28 	struct mm_struct *mm;
29 	unsigned long addr, cause;
30 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
31 	int code = SEGV_MAPERR;
32 	vm_fault_t fault;
33 
34 	cause = regs->scause;
35 	addr = regs->sbadaddr;
36 
37 	tsk = current;
38 	mm = tsk->mm;
39 
40 	/*
41 	 * Fault-in kernel-space virtual memory on-demand.
42 	 * The 'reference' page table is init_mm.pgd.
43 	 *
44 	 * NOTE! We MUST NOT take any locks for this case. We may
45 	 * be in an interrupt or a critical region, and should
46 	 * only copy the information from the master page table,
47 	 * nothing more.
48 	 */
49 	if (unlikely((addr >= VMALLOC_START) && (addr <= VMALLOC_END)))
50 		goto vmalloc_fault;
51 
52 	/* Enable interrupts if they were enabled in the parent context. */
53 	if (likely(regs->sstatus & SR_SPIE))
54 		local_irq_enable();
55 
56 	/*
57 	 * If we're in an interrupt, have no user context, or are running
58 	 * in an atomic region, then we must not take the fault.
59 	 */
60 	if (unlikely(faulthandler_disabled() || !mm))
61 		goto no_context;
62 
63 	if (user_mode(regs))
64 		flags |= FAULT_FLAG_USER;
65 
66 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
67 
68 retry:
69 	down_read(&mm->mmap_sem);
70 	vma = find_vma(mm, addr);
71 	if (unlikely(!vma))
72 		goto bad_area;
73 	if (likely(vma->vm_start <= addr))
74 		goto good_area;
75 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
76 		goto bad_area;
77 	if (unlikely(expand_stack(vma, addr)))
78 		goto bad_area;
79 
80 	/*
81 	 * Ok, we have a good vm_area for this memory access, so
82 	 * we can handle it.
83 	 */
84 good_area:
85 	code = SEGV_ACCERR;
86 
87 	switch (cause) {
88 	case EXC_INST_PAGE_FAULT:
89 		if (!(vma->vm_flags & VM_EXEC))
90 			goto bad_area;
91 		break;
92 	case EXC_LOAD_PAGE_FAULT:
93 		if (!(vma->vm_flags & VM_READ))
94 			goto bad_area;
95 		break;
96 	case EXC_STORE_PAGE_FAULT:
97 		if (!(vma->vm_flags & VM_WRITE))
98 			goto bad_area;
99 		flags |= FAULT_FLAG_WRITE;
100 		break;
101 	default:
102 		panic("%s: unhandled cause %lu", __func__, cause);
103 	}
104 
105 	/*
106 	 * If for any reason at all we could not handle the fault,
107 	 * make sure we exit gracefully rather than endlessly redo
108 	 * the fault.
109 	 */
110 	fault = handle_mm_fault(vma, addr, flags);
111 
112 	/*
113 	 * If we need to retry but a fatal signal is pending, handle the
114 	 * signal first. We do not need to release the mmap_sem because it
115 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
116 	 */
117 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(tsk))
118 		return;
119 
120 	if (unlikely(fault & VM_FAULT_ERROR)) {
121 		if (fault & VM_FAULT_OOM)
122 			goto out_of_memory;
123 		else if (fault & VM_FAULT_SIGBUS)
124 			goto do_sigbus;
125 		BUG();
126 	}
127 
128 	/*
129 	 * Major/minor page fault accounting is only done on the
130 	 * initial attempt. If we go through a retry, it is extremely
131 	 * likely that the page will be found in page cache at that point.
132 	 */
133 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
134 		if (fault & VM_FAULT_MAJOR) {
135 			tsk->maj_flt++;
136 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
137 				      1, regs, addr);
138 		} else {
139 			tsk->min_flt++;
140 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
141 				      1, regs, addr);
142 		}
143 		if (fault & VM_FAULT_RETRY) {
144 			/*
145 			 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
146 			 * of starvation.
147 			 */
148 			flags &= ~(FAULT_FLAG_ALLOW_RETRY);
149 			flags |= FAULT_FLAG_TRIED;
150 
151 			/*
152 			 * No need to up_read(&mm->mmap_sem) as we would
153 			 * have already released it in __lock_page_or_retry
154 			 * in mm/filemap.c.
155 			 */
156 			goto retry;
157 		}
158 	}
159 
160 	up_read(&mm->mmap_sem);
161 	return;
162 
163 	/*
164 	 * Something tried to access memory that isn't in our memory map.
165 	 * Fix it, but check if it's kernel or user first.
166 	 */
167 bad_area:
168 	up_read(&mm->mmap_sem);
169 	/* User mode accesses just cause a SIGSEGV */
170 	if (user_mode(regs)) {
171 		do_trap(regs, SIGSEGV, code, addr, tsk);
172 		return;
173 	}
174 
175 no_context:
176 	/* Are we prepared to handle this kernel fault? */
177 	if (fixup_exception(regs))
178 		return;
179 
180 	/*
181 	 * Oops. The kernel tried to access some bad page. We'll have to
182 	 * terminate things with extreme prejudice.
183 	 */
184 	bust_spinlocks(1);
185 	pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n",
186 		(addr < PAGE_SIZE) ? "NULL pointer dereference" :
187 		"paging request", addr);
188 	die(regs, "Oops");
189 	do_exit(SIGKILL);
190 
191 	/*
192 	 * We ran out of memory, call the OOM killer, and return the userspace
193 	 * (which will retry the fault, or kill us if we got oom-killed).
194 	 */
195 out_of_memory:
196 	up_read(&mm->mmap_sem);
197 	if (!user_mode(regs))
198 		goto no_context;
199 	pagefault_out_of_memory();
200 	return;
201 
202 do_sigbus:
203 	up_read(&mm->mmap_sem);
204 	/* Kernel mode? Handle exceptions or die */
205 	if (!user_mode(regs))
206 		goto no_context;
207 	do_trap(regs, SIGBUS, BUS_ADRERR, addr, tsk);
208 	return;
209 
210 vmalloc_fault:
211 	{
212 		pgd_t *pgd, *pgd_k;
213 		pud_t *pud, *pud_k;
214 		p4d_t *p4d, *p4d_k;
215 		pmd_t *pmd, *pmd_k;
216 		pte_t *pte_k;
217 		int index;
218 
219 		/* User mode accesses just cause a SIGSEGV */
220 		if (user_mode(regs))
221 			return do_trap(regs, SIGSEGV, code, addr, tsk);
222 
223 		/*
224 		 * Synchronize this task's top level page-table
225 		 * with the 'reference' page table.
226 		 *
227 		 * Do _not_ use "tsk->active_mm->pgd" here.
228 		 * We might be inside an interrupt in the middle
229 		 * of a task switch.
230 		 */
231 		index = pgd_index(addr);
232 		pgd = (pgd_t *)pfn_to_virt(csr_read(CSR_SATP)) + index;
233 		pgd_k = init_mm.pgd + index;
234 
235 		if (!pgd_present(*pgd_k))
236 			goto no_context;
237 		set_pgd(pgd, *pgd_k);
238 
239 		p4d = p4d_offset(pgd, addr);
240 		p4d_k = p4d_offset(pgd_k, addr);
241 		if (!p4d_present(*p4d_k))
242 			goto no_context;
243 
244 		pud = pud_offset(p4d, addr);
245 		pud_k = pud_offset(p4d_k, addr);
246 		if (!pud_present(*pud_k))
247 			goto no_context;
248 
249 		/*
250 		 * Since the vmalloc area is global, it is unnecessary
251 		 * to copy individual PTEs
252 		 */
253 		pmd = pmd_offset(pud, addr);
254 		pmd_k = pmd_offset(pud_k, addr);
255 		if (!pmd_present(*pmd_k))
256 			goto no_context;
257 		set_pmd(pmd, *pmd_k);
258 
259 		/*
260 		 * Make sure the actual PTE exists as well to
261 		 * catch kernel vmalloc-area accesses to non-mapped
262 		 * addresses. If we don't do this, this will just
263 		 * silently loop forever.
264 		 */
265 		pte_k = pte_offset_kernel(pmd_k, addr);
266 		if (!pte_present(*pte_k))
267 			goto no_context;
268 		return;
269 	}
270 }
271