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 #include <linux/kfence.h>
18 
19 #include <asm/ptrace.h>
20 #include <asm/tlbflush.h>
21 
22 #include "../kernel/head.h"
23 
die_kernel_fault(const char * msg,unsigned long addr,struct pt_regs * regs)24 static void die_kernel_fault(const char *msg, unsigned long addr,
25 		struct pt_regs *regs)
26 {
27 	bust_spinlocks(1);
28 
29 	pr_alert("Unable to handle kernel %s at virtual address " REG_FMT "\n", msg,
30 		addr);
31 
32 	bust_spinlocks(0);
33 	die(regs, "Oops");
34 	do_exit(SIGKILL);
35 }
36 
no_context(struct pt_regs * regs,unsigned long addr)37 static inline void no_context(struct pt_regs *regs, unsigned long addr)
38 {
39 	const char *msg;
40 
41 	/* Are we prepared to handle this kernel fault? */
42 	if (fixup_exception(regs))
43 		return;
44 
45 	/*
46 	 * Oops. The kernel tried to access some bad page. We'll have to
47 	 * terminate things with extreme prejudice.
48 	 */
49 	if (addr < PAGE_SIZE)
50 		msg = "NULL pointer dereference";
51 	else {
52 		if (kfence_handle_page_fault(addr, regs->cause == EXC_STORE_PAGE_FAULT, regs))
53 			return;
54 
55 		msg = "paging request";
56 	}
57 
58 	die_kernel_fault(msg, addr, regs);
59 }
60 
mm_fault_error(struct pt_regs * regs,unsigned long addr,vm_fault_t fault)61 static inline void mm_fault_error(struct pt_regs *regs, unsigned long addr, vm_fault_t fault)
62 {
63 	if (fault & VM_FAULT_OOM) {
64 		/*
65 		 * We ran out of memory, call the OOM killer, and return the userspace
66 		 * (which will retry the fault, or kill us if we got oom-killed).
67 		 */
68 		if (!user_mode(regs)) {
69 			no_context(regs, addr);
70 			return;
71 		}
72 		pagefault_out_of_memory();
73 		return;
74 	} else if (fault & VM_FAULT_SIGBUS) {
75 		/* Kernel mode? Handle exceptions or die */
76 		if (!user_mode(regs)) {
77 			no_context(regs, addr);
78 			return;
79 		}
80 		do_trap(regs, SIGBUS, BUS_ADRERR, addr);
81 		return;
82 	}
83 	BUG();
84 }
85 
bad_area(struct pt_regs * regs,struct mm_struct * mm,int code,unsigned long addr)86 static inline void bad_area(struct pt_regs *regs, struct mm_struct *mm, int code, unsigned long addr)
87 {
88 	/*
89 	 * Something tried to access memory that isn't in our memory map.
90 	 * Fix it, but check if it's kernel or user first.
91 	 */
92 	mmap_read_unlock(mm);
93 	/* User mode accesses just cause a SIGSEGV */
94 	if (user_mode(regs)) {
95 		do_trap(regs, SIGSEGV, code, addr);
96 		return;
97 	}
98 
99 	no_context(regs, addr);
100 }
101 
vmalloc_fault(struct pt_regs * regs,int code,unsigned long addr)102 static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr)
103 {
104 	pgd_t *pgd, *pgd_k;
105 	pud_t *pud, *pud_k;
106 	p4d_t *p4d, *p4d_k;
107 	pmd_t *pmd, *pmd_k;
108 	pte_t *pte_k;
109 	int index;
110 	unsigned long pfn;
111 
112 	/* User mode accesses just cause a SIGSEGV */
113 	if (user_mode(regs))
114 		return do_trap(regs, SIGSEGV, code, addr);
115 
116 	/*
117 	 * Synchronize this task's top level page-table
118 	 * with the 'reference' page table.
119 	 *
120 	 * Do _not_ use "tsk->active_mm->pgd" here.
121 	 * We might be inside an interrupt in the middle
122 	 * of a task switch.
123 	 */
124 	index = pgd_index(addr);
125 	pfn = csr_read(CSR_SATP) & SATP_PPN;
126 	pgd = (pgd_t *)pfn_to_virt(pfn) + index;
127 	pgd_k = init_mm.pgd + index;
128 
129 	if (!pgd_present(*pgd_k)) {
130 		no_context(regs, addr);
131 		return;
132 	}
133 	set_pgd(pgd, *pgd_k);
134 
135 	p4d = p4d_offset(pgd, addr);
136 	p4d_k = p4d_offset(pgd_k, addr);
137 	if (!p4d_present(*p4d_k)) {
138 		no_context(regs, addr);
139 		return;
140 	}
141 
142 	pud = pud_offset(p4d, addr);
143 	pud_k = pud_offset(p4d_k, addr);
144 	if (!pud_present(*pud_k)) {
145 		no_context(regs, addr);
146 		return;
147 	}
148 
149 	/*
150 	 * Since the vmalloc area is global, it is unnecessary
151 	 * to copy individual PTEs
152 	 */
153 	pmd = pmd_offset(pud, addr);
154 	pmd_k = pmd_offset(pud_k, addr);
155 	if (!pmd_present(*pmd_k)) {
156 		no_context(regs, addr);
157 		return;
158 	}
159 	set_pmd(pmd, *pmd_k);
160 
161 	/*
162 	 * Make sure the actual PTE exists as well to
163 	 * catch kernel vmalloc-area accesses to non-mapped
164 	 * addresses. If we don't do this, this will just
165 	 * silently loop forever.
166 	 */
167 	pte_k = pte_offset_kernel(pmd_k, addr);
168 	if (!pte_present(*pte_k)) {
169 		no_context(regs, addr);
170 		return;
171 	}
172 
173 	/*
174 	 * The kernel assumes that TLBs don't cache invalid
175 	 * entries, but in RISC-V, SFENCE.VMA specifies an
176 	 * ordering constraint, not a cache flush; it is
177 	 * necessary even after writing invalid entries.
178 	 */
179 	local_flush_tlb_page(addr);
180 }
181 
access_error(unsigned long cause,struct vm_area_struct * vma)182 static inline bool access_error(unsigned long cause, struct vm_area_struct *vma)
183 {
184 	switch (cause) {
185 	case EXC_INST_PAGE_FAULT:
186 		if (!(vma->vm_flags & VM_EXEC)) {
187 			return true;
188 		}
189 		break;
190 	case EXC_LOAD_PAGE_FAULT:
191 		if (!(vma->vm_flags & VM_READ)) {
192 			return true;
193 		}
194 		break;
195 	case EXC_STORE_PAGE_FAULT:
196 		if (!(vma->vm_flags & VM_WRITE)) {
197 			return true;
198 		}
199 		break;
200 	default:
201 		panic("%s: unhandled cause %lu", __func__, cause);
202 	}
203 	return false;
204 }
205 
206 /*
207  * This routine handles page faults.  It determines the address and the
208  * problem, and then passes it off to one of the appropriate routines.
209  */
do_page_fault(struct pt_regs * regs)210 asmlinkage void do_page_fault(struct pt_regs *regs)
211 {
212 	struct task_struct *tsk;
213 	struct vm_area_struct *vma;
214 	struct mm_struct *mm;
215 	unsigned long addr, cause;
216 	unsigned int flags = FAULT_FLAG_DEFAULT;
217 	int code = SEGV_MAPERR;
218 	vm_fault_t fault;
219 
220 	cause = regs->cause;
221 	addr = regs->badaddr;
222 
223 	tsk = current;
224 	mm = tsk->mm;
225 
226 	if (kprobe_page_fault(regs, cause))
227 		return;
228 
229 	/*
230 	 * Fault-in kernel-space virtual memory on-demand.
231 	 * The 'reference' page table is init_mm.pgd.
232 	 *
233 	 * NOTE! We MUST NOT take any locks for this case. We may
234 	 * be in an interrupt or a critical region, and should
235 	 * only copy the information from the master page table,
236 	 * nothing more.
237 	 */
238 	if (unlikely((addr >= VMALLOC_START) && (addr <= VMALLOC_END))) {
239 		vmalloc_fault(regs, code, addr);
240 		return;
241 	}
242 
243 #ifdef CONFIG_64BIT
244 	/*
245 	 * Modules in 64bit kernels lie in their own virtual region which is not
246 	 * in the vmalloc region, but dealing with page faults in this region
247 	 * or the vmalloc region amounts to doing the same thing: checking that
248 	 * the mapping exists in init_mm.pgd and updating user page table, so
249 	 * just use vmalloc_fault.
250 	 */
251 	if (unlikely(addr >= MODULES_VADDR && addr < MODULES_END)) {
252 		vmalloc_fault(regs, code, addr);
253 		return;
254 	}
255 #endif
256 	/* Enable interrupts if they were enabled in the parent context. */
257 	if (likely(regs->status & SR_PIE))
258 		local_irq_enable();
259 
260 	/*
261 	 * If we're in an interrupt, have no user context, or are running
262 	 * in an atomic region, then we must not take the fault.
263 	 */
264 	if (unlikely(faulthandler_disabled() || !mm)) {
265 		tsk->thread.bad_cause = cause;
266 		no_context(regs, addr);
267 		return;
268 	}
269 
270 	if (user_mode(regs))
271 		flags |= FAULT_FLAG_USER;
272 
273 	if (!user_mode(regs) && addr < TASK_SIZE &&
274 			unlikely(!(regs->status & SR_SUM)))
275 		die_kernel_fault("access to user memory without uaccess routines",
276 				addr, regs);
277 
278 	perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
279 
280 	if (cause == EXC_STORE_PAGE_FAULT)
281 		flags |= FAULT_FLAG_WRITE;
282 	else if (cause == EXC_INST_PAGE_FAULT)
283 		flags |= FAULT_FLAG_INSTRUCTION;
284 retry:
285 	mmap_read_lock(mm);
286 	vma = find_vma(mm, addr);
287 	if (unlikely(!vma)) {
288 		tsk->thread.bad_cause = cause;
289 		bad_area(regs, mm, code, addr);
290 		return;
291 	}
292 	if (likely(vma->vm_start <= addr))
293 		goto good_area;
294 	if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
295 		tsk->thread.bad_cause = cause;
296 		bad_area(regs, mm, code, addr);
297 		return;
298 	}
299 	if (unlikely(expand_stack(vma, addr))) {
300 		tsk->thread.bad_cause = cause;
301 		bad_area(regs, mm, code, addr);
302 		return;
303 	}
304 
305 	/*
306 	 * Ok, we have a good vm_area for this memory access, so
307 	 * we can handle it.
308 	 */
309 good_area:
310 	code = SEGV_ACCERR;
311 
312 	if (unlikely(access_error(cause, vma))) {
313 		tsk->thread.bad_cause = cause;
314 		bad_area(regs, mm, code, addr);
315 		return;
316 	}
317 
318 	/*
319 	 * If for any reason at all we could not handle the fault,
320 	 * make sure we exit gracefully rather than endlessly redo
321 	 * the fault.
322 	 */
323 	fault = handle_mm_fault(vma, addr, flags, regs);
324 
325 	/*
326 	 * If we need to retry but a fatal signal is pending, handle the
327 	 * signal first. We do not need to release the mmap_lock because it
328 	 * would already be released in __lock_page_or_retry in mm/filemap.c.
329 	 */
330 	if (fault_signal_pending(fault, regs))
331 		return;
332 
333 	if (unlikely((fault & VM_FAULT_RETRY) && (flags & FAULT_FLAG_ALLOW_RETRY))) {
334 		flags |= FAULT_FLAG_TRIED;
335 
336 		/*
337 		 * No need to mmap_read_unlock(mm) as we would
338 		 * have already released it in __lock_page_or_retry
339 		 * in mm/filemap.c.
340 		 */
341 		goto retry;
342 	}
343 
344 	mmap_read_unlock(mm);
345 
346 	if (unlikely(fault & VM_FAULT_ERROR)) {
347 		tsk->thread.bad_cause = cause;
348 		mm_fault_error(regs, addr, fault);
349 		return;
350 	}
351 	return;
352 }
353 NOKPROBE_SYMBOL(do_page_fault);
354