xref: /linux/arch/csky/mm/fault.c (revision d8c6b3fc918dd7b93c66e49ff8640746b84fbfe9)
1 // SPDX-License-Identifier: GPL-2.0
2 // Copyright (C) 2018 Hangzhou C-SKY Microsystems co.,ltd.
3 
4 #include <linux/extable.h>
5 #include <linux/kprobes.h>
6 #include <linux/mmu_context.h>
7 #include <linux/perf_event.h>
8 
9 int fixup_exception(struct pt_regs *regs)
10 {
11 	const struct exception_table_entry *fixup;
12 
13 	fixup = search_exception_tables(instruction_pointer(regs));
14 	if (fixup) {
15 		regs->pc = fixup->fixup;
16 
17 		return 1;
18 	}
19 
20 	return 0;
21 }
22 
23 static inline bool is_write(struct pt_regs *regs)
24 {
25 	switch (trap_no(regs)) {
26 	case VEC_TLBINVALIDS:
27 		return true;
28 	case VEC_TLBMODIFIED:
29 		return true;
30 	}
31 
32 	return false;
33 }
34 
35 #ifdef CONFIG_CPU_HAS_LDSTEX
36 static inline void csky_cmpxchg_fixup(struct pt_regs *regs)
37 {
38 	return;
39 }
40 #else
41 extern unsigned long csky_cmpxchg_ldw;
42 extern unsigned long csky_cmpxchg_stw;
43 static inline void csky_cmpxchg_fixup(struct pt_regs *regs)
44 {
45 	if (trap_no(regs) != VEC_TLBMODIFIED)
46 		return;
47 
48 	if (instruction_pointer(regs) == csky_cmpxchg_stw)
49 		instruction_pointer_set(regs, csky_cmpxchg_ldw);
50 	return;
51 }
52 #endif
53 
54 static inline void no_context(struct pt_regs *regs, unsigned long addr)
55 {
56 	current->thread.trap_no = trap_no(regs);
57 
58 	/* Are we prepared to handle this kernel fault? */
59 	if (fixup_exception(regs))
60 		return;
61 
62 	/*
63 	 * Oops. The kernel tried to access some bad page. We'll have to
64 	 * terminate things with extreme prejudice.
65 	 */
66 	bust_spinlocks(1);
67 	pr_alert("Unable to handle kernel paging request at virtual "
68 		 "addr 0x%08lx, pc: 0x%08lx\n", addr, regs->pc);
69 	die(regs, "Oops");
70 	make_task_dead(SIGKILL);
71 }
72 
73 static inline void mm_fault_error(struct pt_regs *regs, unsigned long addr, vm_fault_t fault)
74 {
75 	current->thread.trap_no = trap_no(regs);
76 
77 	if (fault & VM_FAULT_OOM) {
78 		/*
79 		 * We ran out of memory, call the OOM killer, and return the userspace
80 		 * (which will retry the fault, or kill us if we got oom-killed).
81 		 */
82 		if (!user_mode(regs)) {
83 			no_context(regs, addr);
84 			return;
85 		}
86 		pagefault_out_of_memory();
87 		return;
88 	} else if (fault & VM_FAULT_SIGBUS) {
89 		/* Kernel mode? Handle exceptions or die */
90 		if (!user_mode(regs)) {
91 			no_context(regs, addr);
92 			return;
93 		}
94 		do_trap(regs, SIGBUS, BUS_ADRERR, addr);
95 		return;
96 	}
97 	BUG();
98 }
99 
100 static inline void bad_area_nosemaphore(struct pt_regs *regs, struct mm_struct *mm, int code, unsigned long addr)
101 {
102 	/*
103 	 * Something tried to access memory that isn't in our memory map.
104 	 * Fix it, but check if it's kernel or user first.
105 	 */
106 	/* User mode accesses just cause a SIGSEGV */
107 	if (user_mode(regs)) {
108 		do_trap(regs, SIGSEGV, code, addr);
109 		return;
110 	}
111 
112 	no_context(regs, addr);
113 }
114 
115 static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr)
116 {
117 	pgd_t *pgd, *pgd_k;
118 	pud_t *pud, *pud_k;
119 	pmd_t *pmd, *pmd_k;
120 	pte_t *pte_k;
121 	int offset;
122 
123 	/* User mode accesses just cause a SIGSEGV */
124 	if (user_mode(regs)) {
125 		do_trap(regs, SIGSEGV, code, addr);
126 		return;
127 	}
128 
129 	/*
130 	 * Synchronize this task's top level page-table
131 	 * with the 'reference' page table.
132 	 *
133 	 * Do _not_ use "tsk" here. We might be inside
134 	 * an interrupt in the middle of a task switch..
135 	 */
136 	offset = pgd_index(addr);
137 
138 	pgd = get_pgd() + offset;
139 	pgd_k = init_mm.pgd + offset;
140 
141 	if (!pgd_present(*pgd_k)) {
142 		no_context(regs, addr);
143 		return;
144 	}
145 	set_pgd(pgd, *pgd_k);
146 
147 	pud = (pud_t *)pgd;
148 	pud_k = (pud_t *)pgd_k;
149 	if (!pud_present(*pud_k)) {
150 		no_context(regs, addr);
151 		return;
152 	}
153 
154 	pmd = pmd_offset(pud, addr);
155 	pmd_k = pmd_offset(pud_k, addr);
156 	if (!pmd_present(*pmd_k)) {
157 		no_context(regs, addr);
158 		return;
159 	}
160 	set_pmd(pmd, *pmd_k);
161 
162 	pte_k = pte_offset_kernel(pmd_k, addr);
163 	if (!pte_present(*pte_k)) {
164 		no_context(regs, addr);
165 		return;
166 	}
167 
168 	flush_tlb_one(addr);
169 }
170 
171 static inline bool access_error(struct pt_regs *regs, struct vm_area_struct *vma)
172 {
173 	if (is_write(regs)) {
174 		if (!(vma->vm_flags & VM_WRITE))
175 			return true;
176 	} else {
177 		if (unlikely(!vma_is_accessible(vma)))
178 			return true;
179 	}
180 	return false;
181 }
182 
183 /*
184  * This routine handles page faults.  It determines the address and the
185  * problem, and then passes it off to one of the appropriate routines.
186  */
187 asmlinkage void do_page_fault(struct pt_regs *regs)
188 {
189 	struct task_struct *tsk;
190 	struct vm_area_struct *vma;
191 	struct mm_struct *mm;
192 	unsigned long addr = read_mmu_entryhi() & PAGE_MASK;
193 	unsigned int flags = FAULT_FLAG_DEFAULT;
194 	int code = SEGV_MAPERR;
195 	vm_fault_t fault;
196 
197 	tsk = current;
198 	mm = tsk->mm;
199 
200 	csky_cmpxchg_fixup(regs);
201 
202 	if (kprobe_page_fault(regs, tsk->thread.trap_no))
203 		return;
204 
205 	/*
206 	 * Fault-in kernel-space virtual memory on-demand.
207 	 * The 'reference' page table is init_mm.pgd.
208 	 *
209 	 * NOTE! We MUST NOT take any locks for this case. We may
210 	 * be in an interrupt or a critical region, and should
211 	 * only copy the information from the master page table,
212 	 * nothing more.
213 	 */
214 	if (unlikely((addr >= VMALLOC_START) && (addr <= VMALLOC_END))) {
215 		vmalloc_fault(regs, code, addr);
216 		return;
217 	}
218 
219 	/* Enable interrupts if they were enabled in the parent context. */
220 	if (likely(regs->sr & BIT(6)))
221 		local_irq_enable();
222 
223 	/*
224 	 * If we're in an interrupt, have no user context, or are running
225 	 * in an atomic region, then we must not take the fault.
226 	 */
227 	if (unlikely(faulthandler_disabled() || !mm)) {
228 		no_context(regs, addr);
229 		return;
230 	}
231 
232 	if (user_mode(regs))
233 		flags |= FAULT_FLAG_USER;
234 
235 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
236 
237 	if (is_write(regs))
238 		flags |= FAULT_FLAG_WRITE;
239 retry:
240 	vma = lock_mm_and_find_vma(mm, addr, regs);
241 	if (unlikely(!vma)) {
242 		bad_area_nosemaphore(regs, mm, code, addr);
243 		return;
244 	}
245 
246 	/*
247 	 * Ok, we have a good vm_area for this memory access, so
248 	 * we can handle it.
249 	 */
250 	code = SEGV_ACCERR;
251 
252 	if (unlikely(access_error(regs, vma))) {
253 		mmap_read_unlock(mm);
254 		bad_area_nosemaphore(regs, mm, code, addr);
255 		return;
256 	}
257 
258 	/*
259 	 * If for any reason at all we could not handle the fault,
260 	 * make sure we exit gracefully rather than endlessly redo
261 	 * the fault.
262 	 */
263 	fault = handle_mm_fault(vma, addr, flags, regs);
264 
265 	/*
266 	 * If we need to retry but a fatal signal is pending, handle the
267 	 * signal first. We do not need to release the mmap_lock because it
268 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
269 	 */
270 	if (fault_signal_pending(fault, regs)) {
271 		if (!user_mode(regs))
272 			no_context(regs, addr);
273 		return;
274 	}
275 
276 	/* The fault is fully completed (including releasing mmap lock) */
277 	if (fault & VM_FAULT_COMPLETED)
278 		return;
279 
280 	if (unlikely((fault & VM_FAULT_RETRY) && (flags & FAULT_FLAG_ALLOW_RETRY))) {
281 		flags |= FAULT_FLAG_TRIED;
282 
283 		/*
284 		 * No need to mmap_read_unlock(mm) as we would
285 		 * have already released it in __lock_page_or_retry
286 		 * in mm/filemap.c.
287 		 */
288 		goto retry;
289 	}
290 
291 	mmap_read_unlock(mm);
292 
293 	if (unlikely(fault & VM_FAULT_ERROR)) {
294 		mm_fault_error(regs, addr, fault);
295 		return;
296 	}
297 	return;
298 }
299