Loading...
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
24static 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 make_task_dead(SIGKILL);
35}
36
37static 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
61static 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
86static 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
102static inline void vmalloc_fault(struct pt_regs *regs, int code, unsigned long addr)
103{
104 pgd_t *pgd, *pgd_k;
105 pud_t *pud_k;
106 p4d_t *p4d_k;
107 pmd_t *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_k = p4d_offset(pgd_k, addr);
136 if (!p4d_present(*p4d_k)) {
137 no_context(regs, addr);
138 return;
139 }
140
141 pud_k = pud_offset(p4d_k, addr);
142 if (!pud_present(*pud_k)) {
143 no_context(regs, addr);
144 return;
145 }
146
147 /*
148 * Since the vmalloc area is global, it is unnecessary
149 * to copy individual PTEs
150 */
151 pmd_k = pmd_offset(pud_k, addr);
152 if (!pmd_present(*pmd_k)) {
153 no_context(regs, addr);
154 return;
155 }
156
157 /*
158 * Make sure the actual PTE exists as well to
159 * catch kernel vmalloc-area accesses to non-mapped
160 * addresses. If we don't do this, this will just
161 * silently loop forever.
162 */
163 pte_k = pte_offset_kernel(pmd_k, addr);
164 if (!pte_present(*pte_k)) {
165 no_context(regs, addr);
166 return;
167 }
168
169 /*
170 * The kernel assumes that TLBs don't cache invalid
171 * entries, but in RISC-V, SFENCE.VMA specifies an
172 * ordering constraint, not a cache flush; it is
173 * necessary even after writing invalid entries.
174 */
175 local_flush_tlb_page(addr);
176}
177
178static inline bool access_error(unsigned long cause, struct vm_area_struct *vma)
179{
180 switch (cause) {
181 case EXC_INST_PAGE_FAULT:
182 if (!(vma->vm_flags & VM_EXEC)) {
183 return true;
184 }
185 break;
186 case EXC_LOAD_PAGE_FAULT:
187 /* Write implies read */
188 if (!(vma->vm_flags & (VM_READ | VM_WRITE))) {
189 return true;
190 }
191 break;
192 case EXC_STORE_PAGE_FAULT:
193 if (!(vma->vm_flags & VM_WRITE)) {
194 return true;
195 }
196 break;
197 default:
198 panic("%s: unhandled cause %lu", __func__, cause);
199 }
200 return false;
201}
202
203/*
204 * This routine handles page faults. It determines the address and the
205 * problem, and then passes it off to one of the appropriate routines.
206 */
207asmlinkage void do_page_fault(struct pt_regs *regs)
208{
209 struct task_struct *tsk;
210 struct vm_area_struct *vma;
211 struct mm_struct *mm;
212 unsigned long addr, cause;
213 unsigned int flags = FAULT_FLAG_DEFAULT;
214 int code = SEGV_MAPERR;
215 vm_fault_t fault;
216
217 cause = regs->cause;
218 addr = regs->badaddr;
219
220 tsk = current;
221 mm = tsk->mm;
222
223 if (kprobe_page_fault(regs, cause))
224 return;
225
226 /*
227 * Fault-in kernel-space virtual memory on-demand.
228 * The 'reference' page table is init_mm.pgd.
229 *
230 * NOTE! We MUST NOT take any locks for this case. We may
231 * be in an interrupt or a critical region, and should
232 * only copy the information from the master page table,
233 * nothing more.
234 */
235 if (unlikely((addr >= VMALLOC_START) && (addr < VMALLOC_END))) {
236 vmalloc_fault(regs, code, addr);
237 return;
238 }
239
240#ifdef CONFIG_64BIT
241 /*
242 * Modules in 64bit kernels lie in their own virtual region which is not
243 * in the vmalloc region, but dealing with page faults in this region
244 * or the vmalloc region amounts to doing the same thing: checking that
245 * the mapping exists in init_mm.pgd and updating user page table, so
246 * just use vmalloc_fault.
247 */
248 if (unlikely(addr >= MODULES_VADDR && addr < MODULES_END)) {
249 vmalloc_fault(regs, code, addr);
250 return;
251 }
252#endif
253 /* Enable interrupts if they were enabled in the parent context. */
254 if (likely(regs->status & SR_PIE))
255 local_irq_enable();
256
257 /*
258 * If we're in an interrupt, have no user context, or are running
259 * in an atomic region, then we must not take the fault.
260 */
261 if (unlikely(faulthandler_disabled() || !mm)) {
262 tsk->thread.bad_cause = cause;
263 no_context(regs, addr);
264 return;
265 }
266
267 if (user_mode(regs))
268 flags |= FAULT_FLAG_USER;
269
270 if (!user_mode(regs) && addr < TASK_SIZE &&
271 unlikely(!(regs->status & SR_SUM)))
272 die_kernel_fault("access to user memory without uaccess routines",
273 addr, regs);
274
275 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, addr);
276
277 if (cause == EXC_STORE_PAGE_FAULT)
278 flags |= FAULT_FLAG_WRITE;
279 else if (cause == EXC_INST_PAGE_FAULT)
280 flags |= FAULT_FLAG_INSTRUCTION;
281retry:
282 mmap_read_lock(mm);
283 vma = find_vma(mm, addr);
284 if (unlikely(!vma)) {
285 tsk->thread.bad_cause = cause;
286 bad_area(regs, mm, code, addr);
287 return;
288 }
289 if (likely(vma->vm_start <= addr))
290 goto good_area;
291 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
292 tsk->thread.bad_cause = cause;
293 bad_area(regs, mm, code, addr);
294 return;
295 }
296 if (unlikely(expand_stack(vma, addr))) {
297 tsk->thread.bad_cause = cause;
298 bad_area(regs, mm, code, addr);
299 return;
300 }
301
302 /*
303 * Ok, we have a good vm_area for this memory access, so
304 * we can handle it.
305 */
306good_area:
307 code = SEGV_ACCERR;
308
309 if (unlikely(access_error(cause, vma))) {
310 tsk->thread.bad_cause = cause;
311 bad_area(regs, mm, code, addr);
312 return;
313 }
314
315 /*
316 * If for any reason at all we could not handle the fault,
317 * make sure we exit gracefully rather than endlessly redo
318 * the fault.
319 */
320 fault = handle_mm_fault(vma, addr, flags, regs);
321
322 /*
323 * If we need to retry but a fatal signal is pending, handle the
324 * signal first. We do not need to release the mmap_lock because it
325 * would already be released in __lock_page_or_retry in mm/filemap.c.
326 */
327 if (fault_signal_pending(fault, regs))
328 return;
329
330 /* The fault is fully completed (including releasing mmap lock) */
331 if (fault & VM_FAULT_COMPLETED)
332 return;
333
334 if (unlikely(fault & VM_FAULT_RETRY)) {
335 flags |= FAULT_FLAG_TRIED;
336
337 /*
338 * No need to mmap_read_unlock(mm) as we would
339 * have already released it in __lock_page_or_retry
340 * in mm/filemap.c.
341 */
342 goto retry;
343 }
344
345 mmap_read_unlock(mm);
346
347 if (unlikely(fault & VM_FAULT_ERROR)) {
348 tsk->thread.bad_cause = cause;
349 mm_fault_error(regs, addr, fault);
350 return;
351 }
352 return;
353}
354NOKPROBE_SYMBOL(do_page_fault);
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 */
27asmlinkage 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->scause;
38 addr = regs->sbadaddr;
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->sstatus & SR_SPIE))
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
71retry:
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 */
87good_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 */
170bad_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
178no_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 */
198out_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
205do_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
213vmalloc_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}