xref: /openbmc/linux/arch/riscv/mm/fault.c (revision f9834f18)
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 #include <asm/tlbflush.h>
20 
21 #include "../kernel/head.h"
22 
23 /*
24  * This routine handles page faults.  It determines the address and the
25  * problem, and then passes it off to one of the appropriate routines.
26  */
27 asmlinkage void do_page_fault(struct pt_regs *regs)
28 {
29 	struct task_struct *tsk;
30 	struct vm_area_struct *vma;
31 	struct mm_struct *mm;
32 	unsigned long addr, cause;
33 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
34 	int code = SEGV_MAPERR;
35 	vm_fault_t fault;
36 
37 	cause = regs->cause;
38 	addr = regs->badaddr;
39 
40 	tsk = current;
41 	mm = tsk->mm;
42 
43 	/*
44 	 * Fault-in kernel-space virtual memory on-demand.
45 	 * The 'reference' page table is init_mm.pgd.
46 	 *
47 	 * NOTE! We MUST NOT take any locks for this case. We may
48 	 * be in an interrupt or a critical region, and should
49 	 * only copy the information from the master page table,
50 	 * nothing more.
51 	 */
52 	if (unlikely((addr >= VMALLOC_START) && (addr <= VMALLOC_END)))
53 		goto vmalloc_fault;
54 
55 	/* Enable interrupts if they were enabled in the parent context. */
56 	if (likely(regs->status & SR_PIE))
57 		local_irq_enable();
58 
59 	/*
60 	 * If we're in an interrupt, have no user context, or are running
61 	 * in an atomic region, then we must not take the fault.
62 	 */
63 	if (unlikely(faulthandler_disabled() || !mm))
64 		goto no_context;
65 
66 	if (user_mode(regs))
67 		flags |= FAULT_FLAG_USER;
68 
69 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
70 
71 retry:
72 	down_read(&mm->mmap_sem);
73 	vma = find_vma(mm, addr);
74 	if (unlikely(!vma))
75 		goto bad_area;
76 	if (likely(vma->vm_start <= addr))
77 		goto good_area;
78 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
79 		goto bad_area;
80 	if (unlikely(expand_stack(vma, addr)))
81 		goto bad_area;
82 
83 	/*
84 	 * Ok, we have a good vm_area for this memory access, so
85 	 * we can handle it.
86 	 */
87 good_area:
88 	code = SEGV_ACCERR;
89 
90 	switch (cause) {
91 	case EXC_INST_PAGE_FAULT:
92 		if (!(vma->vm_flags & VM_EXEC))
93 			goto bad_area;
94 		break;
95 	case EXC_LOAD_PAGE_FAULT:
96 		if (!(vma->vm_flags & VM_READ))
97 			goto bad_area;
98 		break;
99 	case EXC_STORE_PAGE_FAULT:
100 		if (!(vma->vm_flags & VM_WRITE))
101 			goto bad_area;
102 		flags |= FAULT_FLAG_WRITE;
103 		break;
104 	default:
105 		panic("%s: unhandled cause %lu", __func__, cause);
106 	}
107 
108 	/*
109 	 * If for any reason at all we could not handle the fault,
110 	 * make sure we exit gracefully rather than endlessly redo
111 	 * the fault.
112 	 */
113 	fault = handle_mm_fault(vma, addr, flags);
114 
115 	/*
116 	 * If we need to retry but a fatal signal is pending, handle the
117 	 * signal first. We do not need to release the mmap_sem because it
118 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
119 	 */
120 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(tsk))
121 		return;
122 
123 	if (unlikely(fault & VM_FAULT_ERROR)) {
124 		if (fault & VM_FAULT_OOM)
125 			goto out_of_memory;
126 		else if (fault & VM_FAULT_SIGBUS)
127 			goto do_sigbus;
128 		BUG();
129 	}
130 
131 	/*
132 	 * Major/minor page fault accounting is only done on the
133 	 * initial attempt. If we go through a retry, it is extremely
134 	 * likely that the page will be found in page cache at that point.
135 	 */
136 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
137 		if (fault & VM_FAULT_MAJOR) {
138 			tsk->maj_flt++;
139 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
140 				      1, regs, addr);
141 		} else {
142 			tsk->min_flt++;
143 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
144 				      1, regs, addr);
145 		}
146 		if (fault & VM_FAULT_RETRY) {
147 			/*
148 			 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
149 			 * of starvation.
150 			 */
151 			flags &= ~(FAULT_FLAG_ALLOW_RETRY);
152 			flags |= FAULT_FLAG_TRIED;
153 
154 			/*
155 			 * No need to up_read(&mm->mmap_sem) as we would
156 			 * have already released it in __lock_page_or_retry
157 			 * in mm/filemap.c.
158 			 */
159 			goto retry;
160 		}
161 	}
162 
163 	up_read(&mm->mmap_sem);
164 	return;
165 
166 	/*
167 	 * Something tried to access memory that isn't in our memory map.
168 	 * Fix it, but check if it's kernel or user first.
169 	 */
170 bad_area:
171 	up_read(&mm->mmap_sem);
172 	/* User mode accesses just cause a SIGSEGV */
173 	if (user_mode(regs)) {
174 		do_trap(regs, SIGSEGV, code, addr);
175 		return;
176 	}
177 
178 no_context:
179 	/* Are we prepared to handle this kernel fault? */
180 	if (fixup_exception(regs))
181 		return;
182 
183 	/*
184 	 * Oops. The kernel tried to access some bad page. We'll have to
185 	 * terminate things with extreme prejudice.
186 	 */
187 	bust_spinlocks(1);
188 	pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n",
189 		(addr < PAGE_SIZE) ? "NULL pointer dereference" :
190 		"paging request", addr);
191 	die(regs, "Oops");
192 	do_exit(SIGKILL);
193 
194 	/*
195 	 * We ran out of memory, call the OOM killer, and return the userspace
196 	 * (which will retry the fault, or kill us if we got oom-killed).
197 	 */
198 out_of_memory:
199 	up_read(&mm->mmap_sem);
200 	if (!user_mode(regs))
201 		goto no_context;
202 	pagefault_out_of_memory();
203 	return;
204 
205 do_sigbus:
206 	up_read(&mm->mmap_sem);
207 	/* Kernel mode? Handle exceptions or die */
208 	if (!user_mode(regs))
209 		goto no_context;
210 	do_trap(regs, SIGBUS, BUS_ADRERR, addr);
211 	return;
212 
213 vmalloc_fault:
214 	{
215 		pgd_t *pgd, *pgd_k;
216 		pud_t *pud, *pud_k;
217 		p4d_t *p4d, *p4d_k;
218 		pmd_t *pmd, *pmd_k;
219 		pte_t *pte_k;
220 		int index;
221 
222 		/* User mode accesses just cause a SIGSEGV */
223 		if (user_mode(regs))
224 			return do_trap(regs, SIGSEGV, code, addr);
225 
226 		/*
227 		 * Synchronize this task's top level page-table
228 		 * with the 'reference' page table.
229 		 *
230 		 * Do _not_ use "tsk->active_mm->pgd" here.
231 		 * We might be inside an interrupt in the middle
232 		 * of a task switch.
233 		 */
234 		index = pgd_index(addr);
235 		pgd = (pgd_t *)pfn_to_virt(csr_read(CSR_SATP)) + index;
236 		pgd_k = init_mm.pgd + index;
237 
238 		if (!pgd_present(*pgd_k))
239 			goto no_context;
240 		set_pgd(pgd, *pgd_k);
241 
242 		p4d = p4d_offset(pgd, addr);
243 		p4d_k = p4d_offset(pgd_k, addr);
244 		if (!p4d_present(*p4d_k))
245 			goto no_context;
246 
247 		pud = pud_offset(p4d, addr);
248 		pud_k = pud_offset(p4d_k, addr);
249 		if (!pud_present(*pud_k))
250 			goto no_context;
251 
252 		/*
253 		 * Since the vmalloc area is global, it is unnecessary
254 		 * to copy individual PTEs
255 		 */
256 		pmd = pmd_offset(pud, addr);
257 		pmd_k = pmd_offset(pud_k, addr);
258 		if (!pmd_present(*pmd_k))
259 			goto no_context;
260 		set_pmd(pmd, *pmd_k);
261 
262 		/*
263 		 * Make sure the actual PTE exists as well to
264 		 * catch kernel vmalloc-area accesses to non-mapped
265 		 * addresses. If we don't do this, this will just
266 		 * silently loop forever.
267 		 */
268 		pte_k = pte_offset_kernel(pmd_k, addr);
269 		if (!pte_present(*pte_k))
270 			goto no_context;
271 
272 		/*
273 		 * The kernel assumes that TLBs don't cache invalid
274 		 * entries, but in RISC-V, SFENCE.VMA specifies an
275 		 * ordering constraint, not a cache flush; it is
276 		 * necessary even after writing invalid entries.
277 		 */
278 		local_flush_tlb_page(addr);
279 
280 		return;
281 	}
282 }
283