xref: /openbmc/linux/arch/riscv/mm/fault.c (revision 4e95bc26)
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 /*
22  * This routine handles page faults.  It determines the address and the
23  * problem, and then passes it off to one of the appropriate routines.
24  */
25 asmlinkage void do_page_fault(struct pt_regs *regs)
26 {
27 	struct task_struct *tsk;
28 	struct vm_area_struct *vma;
29 	struct mm_struct *mm;
30 	unsigned long addr, cause;
31 	unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
32 	int code = SEGV_MAPERR;
33 	vm_fault_t fault;
34 
35 	cause = regs->scause;
36 	addr = regs->sbadaddr;
37 
38 	tsk = current;
39 	mm = tsk->mm;
40 
41 	/*
42 	 * Fault-in kernel-space virtual memory on-demand.
43 	 * The 'reference' page table is init_mm.pgd.
44 	 *
45 	 * NOTE! We MUST NOT take any locks for this case. We may
46 	 * be in an interrupt or a critical region, and should
47 	 * only copy the information from the master page table,
48 	 * nothing more.
49 	 */
50 	if (unlikely((addr >= VMALLOC_START) && (addr <= VMALLOC_END)))
51 		goto vmalloc_fault;
52 
53 	/* Enable interrupts if they were enabled in the parent context. */
54 	if (likely(regs->sstatus & SR_SPIE))
55 		local_irq_enable();
56 
57 	/*
58 	 * If we're in an interrupt, have no user context, or are running
59 	 * in an atomic region, then we must not take the fault.
60 	 */
61 	if (unlikely(faulthandler_disabled() || !mm))
62 		goto no_context;
63 
64 	if (user_mode(regs))
65 		flags |= FAULT_FLAG_USER;
66 
67 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
68 
69 retry:
70 	down_read(&mm->mmap_sem);
71 	vma = find_vma(mm, addr);
72 	if (unlikely(!vma))
73 		goto bad_area;
74 	if (likely(vma->vm_start <= addr))
75 		goto good_area;
76 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN)))
77 		goto bad_area;
78 	if (unlikely(expand_stack(vma, addr)))
79 		goto bad_area;
80 
81 	/*
82 	 * Ok, we have a good vm_area for this memory access, so
83 	 * we can handle it.
84 	 */
85 good_area:
86 	code = SEGV_ACCERR;
87 
88 	switch (cause) {
89 	case EXC_INST_PAGE_FAULT:
90 		if (!(vma->vm_flags & VM_EXEC))
91 			goto bad_area;
92 		break;
93 	case EXC_LOAD_PAGE_FAULT:
94 		if (!(vma->vm_flags & VM_READ))
95 			goto bad_area;
96 		break;
97 	case EXC_STORE_PAGE_FAULT:
98 		if (!(vma->vm_flags & VM_WRITE))
99 			goto bad_area;
100 		flags |= FAULT_FLAG_WRITE;
101 		break;
102 	default:
103 		panic("%s: unhandled cause %lu", __func__, cause);
104 	}
105 
106 	/*
107 	 * If for any reason at all we could not handle the fault,
108 	 * make sure we exit gracefully rather than endlessly redo
109 	 * the fault.
110 	 */
111 	fault = handle_mm_fault(vma, addr, flags);
112 
113 	/*
114 	 * If we need to retry but a fatal signal is pending, handle the
115 	 * signal first. We do not need to release the mmap_sem because it
116 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
117 	 */
118 	if ((fault & VM_FAULT_RETRY) && fatal_signal_pending(tsk))
119 		return;
120 
121 	if (unlikely(fault & VM_FAULT_ERROR)) {
122 		if (fault & VM_FAULT_OOM)
123 			goto out_of_memory;
124 		else if (fault & VM_FAULT_SIGBUS)
125 			goto do_sigbus;
126 		BUG();
127 	}
128 
129 	/*
130 	 * Major/minor page fault accounting is only done on the
131 	 * initial attempt. If we go through a retry, it is extremely
132 	 * likely that the page will be found in page cache at that point.
133 	 */
134 	if (flags & FAULT_FLAG_ALLOW_RETRY) {
135 		if (fault & VM_FAULT_MAJOR) {
136 			tsk->maj_flt++;
137 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ,
138 				      1, regs, addr);
139 		} else {
140 			tsk->min_flt++;
141 			perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN,
142 				      1, regs, addr);
143 		}
144 		if (fault & VM_FAULT_RETRY) {
145 			/*
146 			 * Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
147 			 * of starvation.
148 			 */
149 			flags &= ~(FAULT_FLAG_ALLOW_RETRY);
150 			flags |= FAULT_FLAG_TRIED;
151 
152 			/*
153 			 * No need to up_read(&mm->mmap_sem) as we would
154 			 * have already released it in __lock_page_or_retry
155 			 * in mm/filemap.c.
156 			 */
157 			goto retry;
158 		}
159 	}
160 
161 	up_read(&mm->mmap_sem);
162 	return;
163 
164 	/*
165 	 * Something tried to access memory that isn't in our memory map.
166 	 * Fix it, but check if it's kernel or user first.
167 	 */
168 bad_area:
169 	up_read(&mm->mmap_sem);
170 	/* User mode accesses just cause a SIGSEGV */
171 	if (user_mode(regs)) {
172 		do_trap(regs, SIGSEGV, code, addr, tsk);
173 		return;
174 	}
175 
176 no_context:
177 	/* Are we prepared to handle this kernel fault? */
178 	if (fixup_exception(regs))
179 		return;
180 
181 	/*
182 	 * Oops. The kernel tried to access some bad page. We'll have to
183 	 * terminate things with extreme prejudice.
184 	 */
185 	bust_spinlocks(1);
186 	pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n",
187 		(addr < PAGE_SIZE) ? "NULL pointer dereference" :
188 		"paging request", addr);
189 	die(regs, "Oops");
190 	do_exit(SIGKILL);
191 
192 	/*
193 	 * We ran out of memory, call the OOM killer, and return the userspace
194 	 * (which will retry the fault, or kill us if we got oom-killed).
195 	 */
196 out_of_memory:
197 	up_read(&mm->mmap_sem);
198 	if (!user_mode(regs))
199 		goto no_context;
200 	pagefault_out_of_memory();
201 	return;
202 
203 do_sigbus:
204 	up_read(&mm->mmap_sem);
205 	/* Kernel mode? Handle exceptions or die */
206 	if (!user_mode(regs))
207 		goto no_context;
208 	do_trap(regs, SIGBUS, BUS_ADRERR, addr, tsk);
209 	return;
210 
211 vmalloc_fault:
212 	{
213 		pgd_t *pgd, *pgd_k;
214 		pud_t *pud, *pud_k;
215 		p4d_t *p4d, *p4d_k;
216 		pmd_t *pmd, *pmd_k;
217 		pte_t *pte_k;
218 		int index;
219 
220 		/* User mode accesses just cause a SIGSEGV */
221 		if (user_mode(regs))
222 			return do_trap(regs, SIGSEGV, code, addr, tsk);
223 
224 		/*
225 		 * Synchronize this task's top level page-table
226 		 * with the 'reference' page table.
227 		 *
228 		 * Do _not_ use "tsk->active_mm->pgd" here.
229 		 * We might be inside an interrupt in the middle
230 		 * of a task switch.
231 		 */
232 		index = pgd_index(addr);
233 		pgd = (pgd_t *)pfn_to_virt(csr_read(CSR_SATP)) + index;
234 		pgd_k = init_mm.pgd + index;
235 
236 		if (!pgd_present(*pgd_k))
237 			goto no_context;
238 		set_pgd(pgd, *pgd_k);
239 
240 		p4d = p4d_offset(pgd, addr);
241 		p4d_k = p4d_offset(pgd_k, addr);
242 		if (!p4d_present(*p4d_k))
243 			goto no_context;
244 
245 		pud = pud_offset(p4d, addr);
246 		pud_k = pud_offset(p4d_k, addr);
247 		if (!pud_present(*pud_k))
248 			goto no_context;
249 
250 		/*
251 		 * Since the vmalloc area is global, it is unnecessary
252 		 * to copy individual PTEs
253 		 */
254 		pmd = pmd_offset(pud, addr);
255 		pmd_k = pmd_offset(pud_k, addr);
256 		if (!pmd_present(*pmd_k))
257 			goto no_context;
258 		set_pmd(pmd, *pmd_k);
259 
260 		/*
261 		 * Make sure the actual PTE exists as well to
262 		 * catch kernel vmalloc-area accesses to non-mapped
263 		 * addresses. If we don't do this, this will just
264 		 * silently loop forever.
265 		 */
266 		pte_k = pte_offset_kernel(pmd_k, addr);
267 		if (!pte_present(*pte_k))
268 			goto no_context;
269 
270 		/*
271 		 * The kernel assumes that TLBs don't cache invalid
272 		 * entries, but in RISC-V, SFENCE.VMA specifies an
273 		 * ordering constraint, not a cache flush; it is
274 		 * necessary even after writing invalid entries.
275 		 * Relying on flush_tlb_fix_spurious_fault would
276 		 * suffice, but the extra traps reduce
277 		 * performance. So, eagerly SFENCE.VMA.
278 		 */
279 		local_flush_tlb_page(addr);
280 
281 		return;
282 	}
283 }
284