Eric Lee / smarc-ti-linux-kernel

Download as
Browse Code ยป

Commit 61cfce3e9a5de651613a0d01aecf2d08133f4bdd

Authored by Linus Torvalds
Committed by Greg Kroah-Hartman
1 parent 299f45902a

x86: mm: move mmap_sem unlock from mm_fault_error() to caller 

commit 7fb08eca45270d0ae86e1ad9d39c40b7a55d0190 upstream.

This replaces four copies in various stages of mm_fault_error() handling
with just a single one.  It will also allow for more natural placement
of the unlocking after some further cleanup.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

Showing 1 changed file with 1 additions and 7 deletions Inline Diff

1 /* 1 /*
2 * Copyright (C) 1995 Linus Torvalds 2 * Copyright (C) 1995 Linus Torvalds
3 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs. 3 * Copyright (C) 2001, 2002 Andi Kleen, SuSE Labs.
4 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar 4 * Copyright (C) 2008-2009, Red Hat Inc., Ingo Molnar
5 */ 5 */
6 #include <linux/sched.h> /* test_thread_flag(), ... */ 6 #include <linux/sched.h> /* test_thread_flag(), ... */
7 #include <linux/kdebug.h> /* oops_begin/end, ... */ 7 #include <linux/kdebug.h> /* oops_begin/end, ... */
8 #include <linux/module.h> /* search_exception_table */ 8 #include <linux/module.h> /* search_exception_table */
9 #include <linux/bootmem.h> /* max_low_pfn */ 9 #include <linux/bootmem.h> /* max_low_pfn */
10 #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */ 10 #include <linux/kprobes.h> /* NOKPROBE_SYMBOL, ... */
11 #include <linux/mmiotrace.h> /* kmmio_handler, ... */ 11 #include <linux/mmiotrace.h> /* kmmio_handler, ... */
12 #include <linux/perf_event.h> /* perf_sw_event */ 12 #include <linux/perf_event.h> /* perf_sw_event */
13 #include <linux/hugetlb.h> /* hstate_index_to_shift */ 13 #include <linux/hugetlb.h> /* hstate_index_to_shift */
14 #include <linux/prefetch.h> /* prefetchw */ 14 #include <linux/prefetch.h> /* prefetchw */
15 #include <linux/context_tracking.h> /* exception_enter(), ... */ 15 #include <linux/context_tracking.h> /* exception_enter(), ... */
16 16
17 #include <asm/traps.h> /* dotraplinkage, ... */ 17 #include <asm/traps.h> /* dotraplinkage, ... */
18 #include <asm/pgalloc.h> /* pgd_*(), ... */ 18 #include <asm/pgalloc.h> /* pgd_*(), ... */
19 #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */ 19 #include <asm/kmemcheck.h> /* kmemcheck_*(), ... */
20 #include <asm/fixmap.h> /* VSYSCALL_ADDR */ 20 #include <asm/fixmap.h> /* VSYSCALL_ADDR */
21 #include <asm/vsyscall.h> /* emulate_vsyscall */ 21 #include <asm/vsyscall.h> /* emulate_vsyscall */
22 22
23 #define CREATE_TRACE_POINTS 23 #define CREATE_TRACE_POINTS
24 #include <asm/trace/exceptions.h> 24 #include <asm/trace/exceptions.h>
25 25
26 /* 26 /*
27 * Page fault error code bits: 27 * Page fault error code bits:
28 * 28 *
29 * bit 0 == 0: no page found 1: protection fault 29 * bit 0 == 0: no page found 1: protection fault
30 * bit 1 == 0: read access 1: write access 30 * bit 1 == 0: read access 1: write access
31 * bit 2 == 0: kernel-mode access 1: user-mode access 31 * bit 2 == 0: kernel-mode access 1: user-mode access
32 * bit 3 == 1: use of reserved bit detected 32 * bit 3 == 1: use of reserved bit detected
33 * bit 4 == 1: fault was an instruction fetch 33 * bit 4 == 1: fault was an instruction fetch
34 */ 34 */
35 enum x86_pf_error_code { 35 enum x86_pf_error_code {
36 36
37 PF_PROT = 1 << 0, 37 PF_PROT = 1 << 0,
38 PF_WRITE = 1 << 1, 38 PF_WRITE = 1 << 1,
39 PF_USER = 1 << 2, 39 PF_USER = 1 << 2,
40 PF_RSVD = 1 << 3, 40 PF_RSVD = 1 << 3,
41 PF_INSTR = 1 << 4, 41 PF_INSTR = 1 << 4,
42 }; 42 };
43 43
44 /* 44 /*
45 * Returns 0 if mmiotrace is disabled, or if the fault is not 45 * Returns 0 if mmiotrace is disabled, or if the fault is not
46 * handled by mmiotrace: 46 * handled by mmiotrace:
47 */ 47 */
48 static nokprobe_inline int 48 static nokprobe_inline int
49 kmmio_fault(struct pt_regs *regs, unsigned long addr) 49 kmmio_fault(struct pt_regs *regs, unsigned long addr)
50 { 50 {
51 if (unlikely(is_kmmio_active())) 51 if (unlikely(is_kmmio_active()))
52 if (kmmio_handler(regs, addr) == 1) 52 if (kmmio_handler(regs, addr) == 1)
53 return -1; 53 return -1;
54 return 0; 54 return 0;
55 } 55 }
56 56
57 static nokprobe_inline int kprobes_fault(struct pt_regs *regs) 57 static nokprobe_inline int kprobes_fault(struct pt_regs *regs)
58 { 58 {
59 int ret = 0; 59 int ret = 0;
60 60
61 /* kprobe_running() needs smp_processor_id() */ 61 /* kprobe_running() needs smp_processor_id() */
62 if (kprobes_built_in() && !user_mode_vm(regs)) { 62 if (kprobes_built_in() && !user_mode_vm(regs)) {
63 preempt_disable(); 63 preempt_disable();
64 if (kprobe_running() && kprobe_fault_handler(regs, 14)) 64 if (kprobe_running() && kprobe_fault_handler(regs, 14))
65 ret = 1; 65 ret = 1;
66 preempt_enable(); 66 preempt_enable();
67 } 67 }
68 68
69 return ret; 69 return ret;
70 } 70 }
71 71
72 /* 72 /*
73 * Prefetch quirks: 73 * Prefetch quirks:
74 * 74 *
75 * 32-bit mode: 75 * 32-bit mode:
76 * 76 *
77 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch. 77 * Sometimes AMD Athlon/Opteron CPUs report invalid exceptions on prefetch.
78 * Check that here and ignore it. 78 * Check that here and ignore it.
79 * 79 *
80 * 64-bit mode: 80 * 64-bit mode:
81 * 81 *
82 * Sometimes the CPU reports invalid exceptions on prefetch. 82 * Sometimes the CPU reports invalid exceptions on prefetch.
83 * Check that here and ignore it. 83 * Check that here and ignore it.
84 * 84 *
85 * Opcode checker based on code by Richard Brunner. 85 * Opcode checker based on code by Richard Brunner.
86 */ 86 */
87 static inline int 87 static inline int
88 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr, 88 check_prefetch_opcode(struct pt_regs *regs, unsigned char *instr,
89 unsigned char opcode, int *prefetch) 89 unsigned char opcode, int *prefetch)
90 { 90 {
91 unsigned char instr_hi = opcode & 0xf0; 91 unsigned char instr_hi = opcode & 0xf0;
92 unsigned char instr_lo = opcode & 0x0f; 92 unsigned char instr_lo = opcode & 0x0f;
93 93
94 switch (instr_hi) { 94 switch (instr_hi) {
95 case 0x20: 95 case 0x20:
96 case 0x30: 96 case 0x30:
97 /* 97 /*
98 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes. 98 * Values 0x26,0x2E,0x36,0x3E are valid x86 prefixes.
99 * In X86_64 long mode, the CPU will signal invalid 99 * In X86_64 long mode, the CPU will signal invalid
100 * opcode if some of these prefixes are present so 100 * opcode if some of these prefixes are present so
101 * X86_64 will never get here anyway 101 * X86_64 will never get here anyway
102 */ 102 */
103 return ((instr_lo & 7) == 0x6); 103 return ((instr_lo & 7) == 0x6);
104 #ifdef CONFIG_X86_64 104 #ifdef CONFIG_X86_64
105 case 0x40: 105 case 0x40:
106 /* 106 /*
107 * In AMD64 long mode 0x40..0x4F are valid REX prefixes 107 * In AMD64 long mode 0x40..0x4F are valid REX prefixes
108 * Need to figure out under what instruction mode the 108 * Need to figure out under what instruction mode the
109 * instruction was issued. Could check the LDT for lm, 109 * instruction was issued. Could check the LDT for lm,
110 * but for now it's good enough to assume that long 110 * but for now it's good enough to assume that long
111 * mode only uses well known segments or kernel. 111 * mode only uses well known segments or kernel.
112 */ 112 */
113 return (!user_mode(regs) || user_64bit_mode(regs)); 113 return (!user_mode(regs) || user_64bit_mode(regs));
114 #endif 114 #endif
115 case 0x60: 115 case 0x60:
116 /* 0x64 thru 0x67 are valid prefixes in all modes. */ 116 /* 0x64 thru 0x67 are valid prefixes in all modes. */
117 return (instr_lo & 0xC) == 0x4; 117 return (instr_lo & 0xC) == 0x4;
118 case 0xF0: 118 case 0xF0:
119 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */ 119 /* 0xF0, 0xF2, 0xF3 are valid prefixes in all modes. */
120 return !instr_lo || (instr_lo>>1) == 1; 120 return !instr_lo || (instr_lo>>1) == 1;
121 case 0x00: 121 case 0x00:
122 /* Prefetch instruction is 0x0F0D or 0x0F18 */ 122 /* Prefetch instruction is 0x0F0D or 0x0F18 */
123 if (probe_kernel_address(instr, opcode)) 123 if (probe_kernel_address(instr, opcode))
124 return 0; 124 return 0;
125 125
126 *prefetch = (instr_lo == 0xF) && 126 *prefetch = (instr_lo == 0xF) &&
127 (opcode == 0x0D || opcode == 0x18); 127 (opcode == 0x0D || opcode == 0x18);
128 return 0; 128 return 0;
129 default: 129 default:
130 return 0; 130 return 0;
131 } 131 }
132 } 132 }
133 133
134 static int 134 static int
135 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr) 135 is_prefetch(struct pt_regs *regs, unsigned long error_code, unsigned long addr)
136 { 136 {
137 unsigned char *max_instr; 137 unsigned char *max_instr;
138 unsigned char *instr; 138 unsigned char *instr;
139 int prefetch = 0; 139 int prefetch = 0;
140 140
141 /* 141 /*
142 * If it was a exec (instruction fetch) fault on NX page, then 142 * If it was a exec (instruction fetch) fault on NX page, then
143 * do not ignore the fault: 143 * do not ignore the fault:
144 */ 144 */
145 if (error_code & PF_INSTR) 145 if (error_code & PF_INSTR)
146 return 0; 146 return 0;
147 147
148 instr = (void *)convert_ip_to_linear(current, regs); 148 instr = (void *)convert_ip_to_linear(current, regs);
149 max_instr = instr + 15; 149 max_instr = instr + 15;
150 150
151 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE) 151 if (user_mode(regs) && instr >= (unsigned char *)TASK_SIZE)
152 return 0; 152 return 0;
153 153
154 while (instr < max_instr) { 154 while (instr < max_instr) {
155 unsigned char opcode; 155 unsigned char opcode;
156 156
157 if (probe_kernel_address(instr, opcode)) 157 if (probe_kernel_address(instr, opcode))
158 break; 158 break;
159 159
160 instr++; 160 instr++;
161 161
162 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch)) 162 if (!check_prefetch_opcode(regs, instr, opcode, &prefetch))
163 break; 163 break;
164 } 164 }
165 return prefetch; 165 return prefetch;
166 } 166 }
167 167
168 static void 168 static void
169 force_sig_info_fault(int si_signo, int si_code, unsigned long address, 169 force_sig_info_fault(int si_signo, int si_code, unsigned long address,
170 struct task_struct *tsk, int fault) 170 struct task_struct *tsk, int fault)
171 { 171 {
172 unsigned lsb = 0; 172 unsigned lsb = 0;
173 siginfo_t info; 173 siginfo_t info;
174 174
175 info.si_signo = si_signo; 175 info.si_signo = si_signo;
176 info.si_errno = 0; 176 info.si_errno = 0;
177 info.si_code = si_code; 177 info.si_code = si_code;
178 info.si_addr = (void __user *)address; 178 info.si_addr = (void __user *)address;
179 if (fault & VM_FAULT_HWPOISON_LARGE) 179 if (fault & VM_FAULT_HWPOISON_LARGE)
180 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault)); 180 lsb = hstate_index_to_shift(VM_FAULT_GET_HINDEX(fault));
181 if (fault & VM_FAULT_HWPOISON) 181 if (fault & VM_FAULT_HWPOISON)
182 lsb = PAGE_SHIFT; 182 lsb = PAGE_SHIFT;
183 info.si_addr_lsb = lsb; 183 info.si_addr_lsb = lsb;
184 184
185 force_sig_info(si_signo, &info, tsk); 185 force_sig_info(si_signo, &info, tsk);
186 } 186 }
187 187
188 DEFINE_SPINLOCK(pgd_lock); 188 DEFINE_SPINLOCK(pgd_lock);
189 LIST_HEAD(pgd_list); 189 LIST_HEAD(pgd_list);
190 190
191 #ifdef CONFIG_X86_32 191 #ifdef CONFIG_X86_32
192 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address) 192 static inline pmd_t *vmalloc_sync_one(pgd_t *pgd, unsigned long address)
193 { 193 {
194 unsigned index = pgd_index(address); 194 unsigned index = pgd_index(address);
195 pgd_t *pgd_k; 195 pgd_t *pgd_k;
196 pud_t *pud, *pud_k; 196 pud_t *pud, *pud_k;
197 pmd_t *pmd, *pmd_k; 197 pmd_t *pmd, *pmd_k;
198 198
199 pgd += index; 199 pgd += index;
200 pgd_k = init_mm.pgd + index; 200 pgd_k = init_mm.pgd + index;
201 201
202 if (!pgd_present(*pgd_k)) 202 if (!pgd_present(*pgd_k))
203 return NULL; 203 return NULL;
204 204
205 /* 205 /*
206 * set_pgd(pgd, *pgd_k); here would be useless on PAE 206 * set_pgd(pgd, *pgd_k); here would be useless on PAE
207 * and redundant with the set_pmd() on non-PAE. As would 207 * and redundant with the set_pmd() on non-PAE. As would
208 * set_pud. 208 * set_pud.
209 */ 209 */
210 pud = pud_offset(pgd, address); 210 pud = pud_offset(pgd, address);
211 pud_k = pud_offset(pgd_k, address); 211 pud_k = pud_offset(pgd_k, address);
212 if (!pud_present(*pud_k)) 212 if (!pud_present(*pud_k))
213 return NULL; 213 return NULL;
214 214
215 pmd = pmd_offset(pud, address); 215 pmd = pmd_offset(pud, address);
216 pmd_k = pmd_offset(pud_k, address); 216 pmd_k = pmd_offset(pud_k, address);
217 if (!pmd_present(*pmd_k)) 217 if (!pmd_present(*pmd_k))
218 return NULL; 218 return NULL;
219 219
220 if (!pmd_present(*pmd)) 220 if (!pmd_present(*pmd))
221 set_pmd(pmd, *pmd_k); 221 set_pmd(pmd, *pmd_k);
222 else 222 else
223 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k)); 223 BUG_ON(pmd_page(*pmd) != pmd_page(*pmd_k));
224 224
225 return pmd_k; 225 return pmd_k;
226 } 226 }
227 227
228 void vmalloc_sync_all(void) 228 void vmalloc_sync_all(void)
229 { 229 {
230 unsigned long address; 230 unsigned long address;
231 231
232 if (SHARED_KERNEL_PMD) 232 if (SHARED_KERNEL_PMD)
233 return; 233 return;
234 234
235 for (address = VMALLOC_START & PMD_MASK; 235 for (address = VMALLOC_START & PMD_MASK;
236 address >= TASK_SIZE && address < FIXADDR_TOP; 236 address >= TASK_SIZE && address < FIXADDR_TOP;
237 address += PMD_SIZE) { 237 address += PMD_SIZE) {
238 struct page *page; 238 struct page *page;
239 239
240 spin_lock(&pgd_lock); 240 spin_lock(&pgd_lock);
241 list_for_each_entry(page, &pgd_list, lru) { 241 list_for_each_entry(page, &pgd_list, lru) {
242 spinlock_t *pgt_lock; 242 spinlock_t *pgt_lock;
243 pmd_t *ret; 243 pmd_t *ret;
244 244
245 /* the pgt_lock only for Xen */ 245 /* the pgt_lock only for Xen */
246 pgt_lock = &pgd_page_get_mm(page)->page_table_lock; 246 pgt_lock = &pgd_page_get_mm(page)->page_table_lock;
247 247
248 spin_lock(pgt_lock); 248 spin_lock(pgt_lock);
249 ret = vmalloc_sync_one(page_address(page), address); 249 ret = vmalloc_sync_one(page_address(page), address);
250 spin_unlock(pgt_lock); 250 spin_unlock(pgt_lock);
251 251
252 if (!ret) 252 if (!ret)
253 break; 253 break;
254 } 254 }
255 spin_unlock(&pgd_lock); 255 spin_unlock(&pgd_lock);
256 } 256 }
257 } 257 }
258 258
259 /* 259 /*
260 * 32-bit: 260 * 32-bit:
261 * 261 *
262 * Handle a fault on the vmalloc or module mapping area 262 * Handle a fault on the vmalloc or module mapping area
263 */ 263 */
264 static noinline int vmalloc_fault(unsigned long address) 264 static noinline int vmalloc_fault(unsigned long address)
265 { 265 {
266 unsigned long pgd_paddr; 266 unsigned long pgd_paddr;
267 pmd_t *pmd_k; 267 pmd_t *pmd_k;
268 pte_t *pte_k; 268 pte_t *pte_k;
269 269
270 /* Make sure we are in vmalloc area: */ 270 /* Make sure we are in vmalloc area: */
271 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 271 if (!(address >= VMALLOC_START && address < VMALLOC_END))
272 return -1; 272 return -1;
273 273
274 WARN_ON_ONCE(in_nmi()); 274 WARN_ON_ONCE(in_nmi());
275 275
276 /* 276 /*
277 * Synchronize this task's top level page-table 277 * Synchronize this task's top level page-table
278 * with the 'reference' page table. 278 * with the 'reference' page table.
279 * 279 *
280 * Do _not_ use "current" here. We might be inside 280 * Do _not_ use "current" here. We might be inside
281 * an interrupt in the middle of a task switch.. 281 * an interrupt in the middle of a task switch..
282 */ 282 */
283 pgd_paddr = read_cr3(); 283 pgd_paddr = read_cr3();
284 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address); 284 pmd_k = vmalloc_sync_one(__va(pgd_paddr), address);
285 if (!pmd_k) 285 if (!pmd_k)
286 return -1; 286 return -1;
287 287
288 pte_k = pte_offset_kernel(pmd_k, address); 288 pte_k = pte_offset_kernel(pmd_k, address);
289 if (!pte_present(*pte_k)) 289 if (!pte_present(*pte_k))
290 return -1; 290 return -1;
291 291
292 return 0; 292 return 0;
293 } 293 }
294 NOKPROBE_SYMBOL(vmalloc_fault); 294 NOKPROBE_SYMBOL(vmalloc_fault);
295 295
296 /* 296 /*
297 * Did it hit the DOS screen memory VA from vm86 mode? 297 * Did it hit the DOS screen memory VA from vm86 mode?
298 */ 298 */
299 static inline void 299 static inline void
300 check_v8086_mode(struct pt_regs *regs, unsigned long address, 300 check_v8086_mode(struct pt_regs *regs, unsigned long address,
301 struct task_struct *tsk) 301 struct task_struct *tsk)
302 { 302 {
303 unsigned long bit; 303 unsigned long bit;
304 304
305 if (!v8086_mode(regs)) 305 if (!v8086_mode(regs))
306 return; 306 return;
307 307
308 bit = (address - 0xA0000) >> PAGE_SHIFT; 308 bit = (address - 0xA0000) >> PAGE_SHIFT;
309 if (bit < 32) 309 if (bit < 32)
310 tsk->thread.screen_bitmap |= 1 << bit; 310 tsk->thread.screen_bitmap |= 1 << bit;
311 } 311 }
312 312
313 static bool low_pfn(unsigned long pfn) 313 static bool low_pfn(unsigned long pfn)
314 { 314 {
315 return pfn < max_low_pfn; 315 return pfn < max_low_pfn;
316 } 316 }
317 317
318 static void dump_pagetable(unsigned long address) 318 static void dump_pagetable(unsigned long address)
319 { 319 {
320 pgd_t *base = __va(read_cr3()); 320 pgd_t *base = __va(read_cr3());
321 pgd_t *pgd = &base[pgd_index(address)]; 321 pgd_t *pgd = &base[pgd_index(address)];
322 pmd_t *pmd; 322 pmd_t *pmd;
323 pte_t *pte; 323 pte_t *pte;
324 324
325 #ifdef CONFIG_X86_PAE 325 #ifdef CONFIG_X86_PAE
326 printk("*pdpt = %016Lx ", pgd_val(*pgd)); 326 printk("*pdpt = %016Lx ", pgd_val(*pgd));
327 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd)) 327 if (!low_pfn(pgd_val(*pgd) >> PAGE_SHIFT) || !pgd_present(*pgd))
328 goto out; 328 goto out;
329 #endif 329 #endif
330 pmd = pmd_offset(pud_offset(pgd, address), address); 330 pmd = pmd_offset(pud_offset(pgd, address), address);
331 printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd)); 331 printk(KERN_CONT "*pde = %0*Lx ", sizeof(*pmd) * 2, (u64)pmd_val(*pmd));
332 332
333 /* 333 /*
334 * We must not directly access the pte in the highpte 334 * We must not directly access the pte in the highpte
335 * case if the page table is located in highmem. 335 * case if the page table is located in highmem.
336 * And let's rather not kmap-atomic the pte, just in case 336 * And let's rather not kmap-atomic the pte, just in case
337 * it's allocated already: 337 * it's allocated already:
338 */ 338 */
339 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd)) 339 if (!low_pfn(pmd_pfn(*pmd)) || !pmd_present(*pmd) || pmd_large(*pmd))
340 goto out; 340 goto out;
341 341
342 pte = pte_offset_kernel(pmd, address); 342 pte = pte_offset_kernel(pmd, address);
343 printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte)); 343 printk("*pte = %0*Lx ", sizeof(*pte) * 2, (u64)pte_val(*pte));
344 out: 344 out:
345 printk("\n"); 345 printk("\n");
346 } 346 }
347 347
348 #else /* CONFIG_X86_64: */ 348 #else /* CONFIG_X86_64: */
349 349
350 void vmalloc_sync_all(void) 350 void vmalloc_sync_all(void)
351 { 351 {
352 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END, 0); 352 sync_global_pgds(VMALLOC_START & PGDIR_MASK, VMALLOC_END, 0);
353 } 353 }
354 354
355 /* 355 /*
356 * 64-bit: 356 * 64-bit:
357 * 357 *
358 * Handle a fault on the vmalloc area 358 * Handle a fault on the vmalloc area
359 * 359 *
360 * This assumes no large pages in there. 360 * This assumes no large pages in there.
361 */ 361 */
362 static noinline int vmalloc_fault(unsigned long address) 362 static noinline int vmalloc_fault(unsigned long address)
363 { 363 {
364 pgd_t *pgd, *pgd_ref; 364 pgd_t *pgd, *pgd_ref;
365 pud_t *pud, *pud_ref; 365 pud_t *pud, *pud_ref;
366 pmd_t *pmd, *pmd_ref; 366 pmd_t *pmd, *pmd_ref;
367 pte_t *pte, *pte_ref; 367 pte_t *pte, *pte_ref;
368 368
369 /* Make sure we are in vmalloc area: */ 369 /* Make sure we are in vmalloc area: */
370 if (!(address >= VMALLOC_START && address < VMALLOC_END)) 370 if (!(address >= VMALLOC_START && address < VMALLOC_END))
371 return -1; 371 return -1;
372 372
373 WARN_ON_ONCE(in_nmi()); 373 WARN_ON_ONCE(in_nmi());
374 374
375 /* 375 /*
376 * Copy kernel mappings over when needed. This can also 376 * Copy kernel mappings over when needed. This can also
377 * happen within a race in page table update. In the later 377 * happen within a race in page table update. In the later
378 * case just flush: 378 * case just flush:
379 */ 379 */
380 pgd = pgd_offset(current->active_mm, address); 380 pgd = pgd_offset(current->active_mm, address);
381 pgd_ref = pgd_offset_k(address); 381 pgd_ref = pgd_offset_k(address);
382 if (pgd_none(*pgd_ref)) 382 if (pgd_none(*pgd_ref))
383 return -1; 383 return -1;
384 384
385 if (pgd_none(*pgd)) { 385 if (pgd_none(*pgd)) {
386 set_pgd(pgd, *pgd_ref); 386 set_pgd(pgd, *pgd_ref);
387 arch_flush_lazy_mmu_mode(); 387 arch_flush_lazy_mmu_mode();
388 } else { 388 } else {
389 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref)); 389 BUG_ON(pgd_page_vaddr(*pgd) != pgd_page_vaddr(*pgd_ref));
390 } 390 }
391 391
392 /* 392 /*
393 * Below here mismatches are bugs because these lower tables 393 * Below here mismatches are bugs because these lower tables
394 * are shared: 394 * are shared:
395 */ 395 */
396 396
397 pud = pud_offset(pgd, address); 397 pud = pud_offset(pgd, address);
398 pud_ref = pud_offset(pgd_ref, address); 398 pud_ref = pud_offset(pgd_ref, address);
399 if (pud_none(*pud_ref)) 399 if (pud_none(*pud_ref))
400 return -1; 400 return -1;
401 401
402 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref)) 402 if (pud_none(*pud) || pud_page_vaddr(*pud) != pud_page_vaddr(*pud_ref))
403 BUG(); 403 BUG();
404 404
405 pmd = pmd_offset(pud, address); 405 pmd = pmd_offset(pud, address);
406 pmd_ref = pmd_offset(pud_ref, address); 406 pmd_ref = pmd_offset(pud_ref, address);
407 if (pmd_none(*pmd_ref)) 407 if (pmd_none(*pmd_ref))
408 return -1; 408 return -1;
409 409
410 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref)) 410 if (pmd_none(*pmd) || pmd_page(*pmd) != pmd_page(*pmd_ref))
411 BUG(); 411 BUG();
412 412
413 pte_ref = pte_offset_kernel(pmd_ref, address); 413 pte_ref = pte_offset_kernel(pmd_ref, address);
414 if (!pte_present(*pte_ref)) 414 if (!pte_present(*pte_ref))
415 return -1; 415 return -1;
416 416
417 pte = pte_offset_kernel(pmd, address); 417 pte = pte_offset_kernel(pmd, address);
418 418
419 /* 419 /*
420 * Don't use pte_page here, because the mappings can point 420 * Don't use pte_page here, because the mappings can point
421 * outside mem_map, and the NUMA hash lookup cannot handle 421 * outside mem_map, and the NUMA hash lookup cannot handle
422 * that: 422 * that:
423 */ 423 */
424 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref)) 424 if (!pte_present(*pte) || pte_pfn(*pte) != pte_pfn(*pte_ref))
425 BUG(); 425 BUG();
426 426
427 return 0; 427 return 0;
428 } 428 }
429 NOKPROBE_SYMBOL(vmalloc_fault); 429 NOKPROBE_SYMBOL(vmalloc_fault);
430 430
431 #ifdef CONFIG_CPU_SUP_AMD 431 #ifdef CONFIG_CPU_SUP_AMD
432 static const char errata93_warning[] = 432 static const char errata93_warning[] =
433 KERN_ERR 433 KERN_ERR
434 "******* Your BIOS seems to not contain a fix for K8 errata #93\n" 434 "******* Your BIOS seems to not contain a fix for K8 errata #93\n"
435 "******* Working around it, but it may cause SEGVs or burn power.\n" 435 "******* Working around it, but it may cause SEGVs or burn power.\n"
436 "******* Please consider a BIOS update.\n" 436 "******* Please consider a BIOS update.\n"
437 "******* Disabling USB legacy in the BIOS may also help.\n"; 437 "******* Disabling USB legacy in the BIOS may also help.\n";
438 #endif 438 #endif
439 439
440 /* 440 /*
441 * No vm86 mode in 64-bit mode: 441 * No vm86 mode in 64-bit mode:
442 */ 442 */
443 static inline void 443 static inline void
444 check_v8086_mode(struct pt_regs *regs, unsigned long address, 444 check_v8086_mode(struct pt_regs *regs, unsigned long address,
445 struct task_struct *tsk) 445 struct task_struct *tsk)
446 { 446 {
447 } 447 }
448 448
449 static int bad_address(void *p) 449 static int bad_address(void *p)
450 { 450 {
451 unsigned long dummy; 451 unsigned long dummy;
452 452
453 return probe_kernel_address((unsigned long *)p, dummy); 453 return probe_kernel_address((unsigned long *)p, dummy);
454 } 454 }
455 455
456 static void dump_pagetable(unsigned long address) 456 static void dump_pagetable(unsigned long address)
457 { 457 {
458 pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK); 458 pgd_t *base = __va(read_cr3() & PHYSICAL_PAGE_MASK);
459 pgd_t *pgd = base + pgd_index(address); 459 pgd_t *pgd = base + pgd_index(address);
460 pud_t *pud; 460 pud_t *pud;
461 pmd_t *pmd; 461 pmd_t *pmd;
462 pte_t *pte; 462 pte_t *pte;
463 463
464 if (bad_address(pgd)) 464 if (bad_address(pgd))
465 goto bad; 465 goto bad;
466 466
467 printk("PGD %lx ", pgd_val(*pgd)); 467 printk("PGD %lx ", pgd_val(*pgd));
468 468
469 if (!pgd_present(*pgd)) 469 if (!pgd_present(*pgd))
470 goto out; 470 goto out;
471 471
472 pud = pud_offset(pgd, address); 472 pud = pud_offset(pgd, address);
473 if (bad_address(pud)) 473 if (bad_address(pud))
474 goto bad; 474 goto bad;
475 475
476 printk("PUD %lx ", pud_val(*pud)); 476 printk("PUD %lx ", pud_val(*pud));
477 if (!pud_present(*pud) || pud_large(*pud)) 477 if (!pud_present(*pud) || pud_large(*pud))
478 goto out; 478 goto out;
479 479
480 pmd = pmd_offset(pud, address); 480 pmd = pmd_offset(pud, address);
481 if (bad_address(pmd)) 481 if (bad_address(pmd))
482 goto bad; 482 goto bad;
483 483
484 printk("PMD %lx ", pmd_val(*pmd)); 484 printk("PMD %lx ", pmd_val(*pmd));
485 if (!pmd_present(*pmd) || pmd_large(*pmd)) 485 if (!pmd_present(*pmd) || pmd_large(*pmd))
486 goto out; 486 goto out;
487 487
488 pte = pte_offset_kernel(pmd, address); 488 pte = pte_offset_kernel(pmd, address);
489 if (bad_address(pte)) 489 if (bad_address(pte))
490 goto bad; 490 goto bad;
491 491
492 printk("PTE %lx", pte_val(*pte)); 492 printk("PTE %lx", pte_val(*pte));
493 out: 493 out:
494 printk("\n"); 494 printk("\n");
495 return; 495 return;
496 bad: 496 bad:
497 printk("BAD\n"); 497 printk("BAD\n");
498 } 498 }
499 499
500 #endif /* CONFIG_X86_64 */ 500 #endif /* CONFIG_X86_64 */
501 501
502 /* 502 /*
503 * Workaround for K8 erratum #93 & buggy BIOS. 503 * Workaround for K8 erratum #93 & buggy BIOS.
504 * 504 *
505 * BIOS SMM functions are required to use a specific workaround 505 * BIOS SMM functions are required to use a specific workaround
506 * to avoid corruption of the 64bit RIP register on C stepping K8. 506 * to avoid corruption of the 64bit RIP register on C stepping K8.
507 * 507 *
508 * A lot of BIOS that didn't get tested properly miss this. 508 * A lot of BIOS that didn't get tested properly miss this.
509 * 509 *
510 * The OS sees this as a page fault with the upper 32bits of RIP cleared. 510 * The OS sees this as a page fault with the upper 32bits of RIP cleared.
511 * Try to work around it here. 511 * Try to work around it here.
512 * 512 *
513 * Note we only handle faults in kernel here. 513 * Note we only handle faults in kernel here.
514 * Does nothing on 32-bit. 514 * Does nothing on 32-bit.
515 */ 515 */
516 static int is_errata93(struct pt_regs *regs, unsigned long address) 516 static int is_errata93(struct pt_regs *regs, unsigned long address)
517 { 517 {
518 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD) 518 #if defined(CONFIG_X86_64) && defined(CONFIG_CPU_SUP_AMD)
519 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD 519 if (boot_cpu_data.x86_vendor != X86_VENDOR_AMD
520 || boot_cpu_data.x86 != 0xf) 520 || boot_cpu_data.x86 != 0xf)
521 return 0; 521 return 0;
522 522
523 if (address != regs->ip) 523 if (address != regs->ip)
524 return 0; 524 return 0;
525 525
526 if ((address >> 32) != 0) 526 if ((address >> 32) != 0)
527 return 0; 527 return 0;
528 528
529 address |= 0xffffffffUL << 32; 529 address |= 0xffffffffUL << 32;
530 if ((address >= (u64)_stext && address <= (u64)_etext) || 530 if ((address >= (u64)_stext && address <= (u64)_etext) ||
531 (address >= MODULES_VADDR && address <= MODULES_END)) { 531 (address >= MODULES_VADDR && address <= MODULES_END)) {
532 printk_once(errata93_warning); 532 printk_once(errata93_warning);
533 regs->ip = address; 533 regs->ip = address;
534 return 1; 534 return 1;
535 } 535 }
536 #endif 536 #endif
537 return 0; 537 return 0;
538 } 538 }
539 539
540 /* 540 /*
541 * Work around K8 erratum #100 K8 in compat mode occasionally jumps 541 * Work around K8 erratum #100 K8 in compat mode occasionally jumps
542 * to illegal addresses >4GB. 542 * to illegal addresses >4GB.
543 * 543 *
544 * We catch this in the page fault handler because these addresses 544 * We catch this in the page fault handler because these addresses
545 * are not reachable. Just detect this case and return. Any code 545 * are not reachable. Just detect this case and return. Any code
546 * segment in LDT is compatibility mode. 546 * segment in LDT is compatibility mode.
547 */ 547 */
548 static int is_errata100(struct pt_regs *regs, unsigned long address) 548 static int is_errata100(struct pt_regs *regs, unsigned long address)
549 { 549 {
550 #ifdef CONFIG_X86_64 550 #ifdef CONFIG_X86_64
551 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32)) 551 if ((regs->cs == __USER32_CS || (regs->cs & (1<<2))) && (address >> 32))
552 return 1; 552 return 1;
553 #endif 553 #endif
554 return 0; 554 return 0;
555 } 555 }
556 556
557 static int is_f00f_bug(struct pt_regs *regs, unsigned long address) 557 static int is_f00f_bug(struct pt_regs *regs, unsigned long address)
558 { 558 {
559 #ifdef CONFIG_X86_F00F_BUG 559 #ifdef CONFIG_X86_F00F_BUG
560 unsigned long nr; 560 unsigned long nr;
561 561
562 /* 562 /*
563 * Pentium F0 0F C7 C8 bug workaround: 563 * Pentium F0 0F C7 C8 bug workaround:
564 */ 564 */
565 if (boot_cpu_has_bug(X86_BUG_F00F)) { 565 if (boot_cpu_has_bug(X86_BUG_F00F)) {
566 nr = (address - idt_descr.address) >> 3; 566 nr = (address - idt_descr.address) >> 3;
567 567
568 if (nr == 6) { 568 if (nr == 6) {
569 do_invalid_op(regs, 0); 569 do_invalid_op(regs, 0);
570 return 1; 570 return 1;
571 } 571 }
572 } 572 }
573 #endif 573 #endif
574 return 0; 574 return 0;
575 } 575 }
576 576
577 static const char nx_warning[] = KERN_CRIT 577 static const char nx_warning[] = KERN_CRIT
578 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n"; 578 "kernel tried to execute NX-protected page - exploit attempt? (uid: %d)\n";
579 static const char smep_warning[] = KERN_CRIT 579 static const char smep_warning[] = KERN_CRIT
580 "unable to execute userspace code (SMEP?) (uid: %d)\n"; 580 "unable to execute userspace code (SMEP?) (uid: %d)\n";
581 581
582 static void 582 static void
583 show_fault_oops(struct pt_regs *regs, unsigned long error_code, 583 show_fault_oops(struct pt_regs *regs, unsigned long error_code,
584 unsigned long address) 584 unsigned long address)
585 { 585 {
586 if (!oops_may_print()) 586 if (!oops_may_print())
587 return; 587 return;
588 588
589 if (error_code & PF_INSTR) { 589 if (error_code & PF_INSTR) {
590 unsigned int level; 590 unsigned int level;
591 pgd_t *pgd; 591 pgd_t *pgd;
592 pte_t *pte; 592 pte_t *pte;
593 593
594 pgd = __va(read_cr3() & PHYSICAL_PAGE_MASK); 594 pgd = __va(read_cr3() & PHYSICAL_PAGE_MASK);
595 pgd += pgd_index(address); 595 pgd += pgd_index(address);
596 596
597 pte = lookup_address_in_pgd(pgd, address, &level); 597 pte = lookup_address_in_pgd(pgd, address, &level);
598 598
599 if (pte && pte_present(*pte) && !pte_exec(*pte)) 599 if (pte && pte_present(*pte) && !pte_exec(*pte))
600 printk(nx_warning, from_kuid(&init_user_ns, current_uid())); 600 printk(nx_warning, from_kuid(&init_user_ns, current_uid()));
601 if (pte && pte_present(*pte) && pte_exec(*pte) && 601 if (pte && pte_present(*pte) && pte_exec(*pte) &&
602 (pgd_flags(*pgd) & _PAGE_USER) && 602 (pgd_flags(*pgd) & _PAGE_USER) &&
603 (read_cr4() & X86_CR4_SMEP)) 603 (read_cr4() & X86_CR4_SMEP))
604 printk(smep_warning, from_kuid(&init_user_ns, current_uid())); 604 printk(smep_warning, from_kuid(&init_user_ns, current_uid()));
605 } 605 }
606 606
607 printk(KERN_ALERT "BUG: unable to handle kernel "); 607 printk(KERN_ALERT "BUG: unable to handle kernel ");
608 if (address < PAGE_SIZE) 608 if (address < PAGE_SIZE)
609 printk(KERN_CONT "NULL pointer dereference"); 609 printk(KERN_CONT "NULL pointer dereference");
610 else 610 else
611 printk(KERN_CONT "paging request"); 611 printk(KERN_CONT "paging request");
612 612
613 printk(KERN_CONT " at %p\n", (void *) address); 613 printk(KERN_CONT " at %p\n", (void *) address);
614 printk(KERN_ALERT "IP:"); 614 printk(KERN_ALERT "IP:");
615 printk_address(regs->ip); 615 printk_address(regs->ip);
616 616
617 dump_pagetable(address); 617 dump_pagetable(address);
618 } 618 }
619 619
620 static noinline void 620 static noinline void
621 pgtable_bad(struct pt_regs *regs, unsigned long error_code, 621 pgtable_bad(struct pt_regs *regs, unsigned long error_code,
622 unsigned long address) 622 unsigned long address)
623 { 623 {
624 struct task_struct *tsk; 624 struct task_struct *tsk;
625 unsigned long flags; 625 unsigned long flags;
626 int sig; 626 int sig;
627 627
628 flags = oops_begin(); 628 flags = oops_begin();
629 tsk = current; 629 tsk = current;
630 sig = SIGKILL; 630 sig = SIGKILL;
631 631
632 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n", 632 printk(KERN_ALERT "%s: Corrupted page table at address %lx\n",
633 tsk->comm, address); 633 tsk->comm, address);
634 dump_pagetable(address); 634 dump_pagetable(address);
635 635
636 tsk->thread.cr2 = address; 636 tsk->thread.cr2 = address;
637 tsk->thread.trap_nr = X86_TRAP_PF; 637 tsk->thread.trap_nr = X86_TRAP_PF;
638 tsk->thread.error_code = error_code; 638 tsk->thread.error_code = error_code;
639 639
640 if (__die("Bad pagetable", regs, error_code)) 640 if (__die("Bad pagetable", regs, error_code))
641 sig = 0; 641 sig = 0;
642 642
643 oops_end(flags, regs, sig); 643 oops_end(flags, regs, sig);
644 } 644 }
645 645
646 static noinline void 646 static noinline void
647 no_context(struct pt_regs *regs, unsigned long error_code, 647 no_context(struct pt_regs *regs, unsigned long error_code,
648 unsigned long address, int signal, int si_code) 648 unsigned long address, int signal, int si_code)
649 { 649 {
650 struct task_struct *tsk = current; 650 struct task_struct *tsk = current;
651 unsigned long flags; 651 unsigned long flags;
652 int sig; 652 int sig;
653 653
654 /* Are we prepared to handle this kernel fault? */ 654 /* Are we prepared to handle this kernel fault? */
655 if (fixup_exception(regs)) { 655 if (fixup_exception(regs)) {
656 /* 656 /*
657 * Any interrupt that takes a fault gets the fixup. This makes 657 * Any interrupt that takes a fault gets the fixup. This makes
658 * the below recursive fault logic only apply to a faults from 658 * the below recursive fault logic only apply to a faults from
659 * task context. 659 * task context.
660 */ 660 */
661 if (in_interrupt()) 661 if (in_interrupt())
662 return; 662 return;
663 663
664 /* 664 /*
665 * Per the above we're !in_interrupt(), aka. task context. 665 * Per the above we're !in_interrupt(), aka. task context.
666 * 666 *
667 * In this case we need to make sure we're not recursively 667 * In this case we need to make sure we're not recursively
668 * faulting through the emulate_vsyscall() logic. 668 * faulting through the emulate_vsyscall() logic.
669 */ 669 */
670 if (current_thread_info()->sig_on_uaccess_error && signal) { 670 if (current_thread_info()->sig_on_uaccess_error && signal) {
671 tsk->thread.trap_nr = X86_TRAP_PF; 671 tsk->thread.trap_nr = X86_TRAP_PF;
672 tsk->thread.error_code = error_code | PF_USER; 672 tsk->thread.error_code = error_code | PF_USER;
673 tsk->thread.cr2 = address; 673 tsk->thread.cr2 = address;
674 674
675 /* XXX: hwpoison faults will set the wrong code. */ 675 /* XXX: hwpoison faults will set the wrong code. */
676 force_sig_info_fault(signal, si_code, address, tsk, 0); 676 force_sig_info_fault(signal, si_code, address, tsk, 0);
677 } 677 }
678 678
679 /* 679 /*
680 * Barring that, we can do the fixup and be happy. 680 * Barring that, we can do the fixup and be happy.
681 */ 681 */
682 return; 682 return;
683 } 683 }
684 684
685 /* 685 /*
686 * 32-bit: 686 * 32-bit:
687 * 687 *
688 * Valid to do another page fault here, because if this fault 688 * Valid to do another page fault here, because if this fault
689 * had been triggered by is_prefetch fixup_exception would have 689 * had been triggered by is_prefetch fixup_exception would have
690 * handled it. 690 * handled it.
691 * 691 *
692 * 64-bit: 692 * 64-bit:
693 * 693 *
694 * Hall of shame of CPU/BIOS bugs. 694 * Hall of shame of CPU/BIOS bugs.
695 */ 695 */
696 if (is_prefetch(regs, error_code, address)) 696 if (is_prefetch(regs, error_code, address))
697 return; 697 return;
698 698
699 if (is_errata93(regs, address)) 699 if (is_errata93(regs, address))
700 return; 700 return;
701 701
702 /* 702 /*
703 * Oops. The kernel tried to access some bad page. We'll have to 703 * Oops. The kernel tried to access some bad page. We'll have to
704 * terminate things with extreme prejudice: 704 * terminate things with extreme prejudice:
705 */ 705 */
706 flags = oops_begin(); 706 flags = oops_begin();
707 707
708 show_fault_oops(regs, error_code, address); 708 show_fault_oops(regs, error_code, address);
709 709
710 if (task_stack_end_corrupted(tsk)) 710 if (task_stack_end_corrupted(tsk))
711 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n"); 711 printk(KERN_EMERG "Thread overran stack, or stack corrupted\n");
712 712
713 tsk->thread.cr2 = address; 713 tsk->thread.cr2 = address;
714 tsk->thread.trap_nr = X86_TRAP_PF; 714 tsk->thread.trap_nr = X86_TRAP_PF;
715 tsk->thread.error_code = error_code; 715 tsk->thread.error_code = error_code;
716 716
717 sig = SIGKILL; 717 sig = SIGKILL;
718 if (__die("Oops", regs, error_code)) 718 if (__die("Oops", regs, error_code))
719 sig = 0; 719 sig = 0;
720 720
721 /* Executive summary in case the body of the oops scrolled away */ 721 /* Executive summary in case the body of the oops scrolled away */
722 printk(KERN_DEFAULT "CR2: %016lx\n", address); 722 printk(KERN_DEFAULT "CR2: %016lx\n", address);
723 723
724 oops_end(flags, regs, sig); 724 oops_end(flags, regs, sig);
725 } 725 }
726 726
727 /* 727 /*
728 * Print out info about fatal segfaults, if the show_unhandled_signals 728 * Print out info about fatal segfaults, if the show_unhandled_signals
729 * sysctl is set: 729 * sysctl is set:
730 */ 730 */
731 static inline void 731 static inline void
732 show_signal_msg(struct pt_regs *regs, unsigned long error_code, 732 show_signal_msg(struct pt_regs *regs, unsigned long error_code,
733 unsigned long address, struct task_struct *tsk) 733 unsigned long address, struct task_struct *tsk)
734 { 734 {
735 if (!unhandled_signal(tsk, SIGSEGV)) 735 if (!unhandled_signal(tsk, SIGSEGV))
736 return; 736 return;
737 737
738 if (!printk_ratelimit()) 738 if (!printk_ratelimit())
739 return; 739 return;
740 740
741 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx", 741 printk("%s%s[%d]: segfault at %lx ip %p sp %p error %lx",
742 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG, 742 task_pid_nr(tsk) > 1 ? KERN_INFO : KERN_EMERG,
743 tsk->comm, task_pid_nr(tsk), address, 743 tsk->comm, task_pid_nr(tsk), address,
744 (void *)regs->ip, (void *)regs->sp, error_code); 744 (void *)regs->ip, (void *)regs->sp, error_code);
745 745
746 print_vma_addr(KERN_CONT " in ", regs->ip); 746 print_vma_addr(KERN_CONT " in ", regs->ip);
747 747
748 printk(KERN_CONT "\n"); 748 printk(KERN_CONT "\n");
749 } 749 }
750 750
751 static void 751 static void
752 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 752 __bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
753 unsigned long address, int si_code) 753 unsigned long address, int si_code)
754 { 754 {
755 struct task_struct *tsk = current; 755 struct task_struct *tsk = current;
756 756
757 /* User mode accesses just cause a SIGSEGV */ 757 /* User mode accesses just cause a SIGSEGV */
758 if (error_code & PF_USER) { 758 if (error_code & PF_USER) {
759 /* 759 /*
760 * It's possible to have interrupts off here: 760 * It's possible to have interrupts off here:
761 */ 761 */
762 local_irq_enable(); 762 local_irq_enable();
763 763
764 /* 764 /*
765 * Valid to do another page fault here because this one came 765 * Valid to do another page fault here because this one came
766 * from user space: 766 * from user space:
767 */ 767 */
768 if (is_prefetch(regs, error_code, address)) 768 if (is_prefetch(regs, error_code, address))
769 return; 769 return;
770 770
771 if (is_errata100(regs, address)) 771 if (is_errata100(regs, address))
772 return; 772 return;
773 773
774 #ifdef CONFIG_X86_64 774 #ifdef CONFIG_X86_64
775 /* 775 /*
776 * Instruction fetch faults in the vsyscall page might need 776 * Instruction fetch faults in the vsyscall page might need
777 * emulation. 777 * emulation.
778 */ 778 */
779 if (unlikely((error_code & PF_INSTR) && 779 if (unlikely((error_code & PF_INSTR) &&
780 ((address & ~0xfff) == VSYSCALL_ADDR))) { 780 ((address & ~0xfff) == VSYSCALL_ADDR))) {
781 if (emulate_vsyscall(regs, address)) 781 if (emulate_vsyscall(regs, address))
782 return; 782 return;
783 } 783 }
784 #endif 784 #endif
785 /* Kernel addresses are always protection faults: */ 785 /* Kernel addresses are always protection faults: */
786 if (address >= TASK_SIZE) 786 if (address >= TASK_SIZE)
787 error_code |= PF_PROT; 787 error_code |= PF_PROT;
788 788
789 if (likely(show_unhandled_signals)) 789 if (likely(show_unhandled_signals))
790 show_signal_msg(regs, error_code, address, tsk); 790 show_signal_msg(regs, error_code, address, tsk);
791 791
792 tsk->thread.cr2 = address; 792 tsk->thread.cr2 = address;
793 tsk->thread.error_code = error_code; 793 tsk->thread.error_code = error_code;
794 tsk->thread.trap_nr = X86_TRAP_PF; 794 tsk->thread.trap_nr = X86_TRAP_PF;
795 795
796 force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0); 796 force_sig_info_fault(SIGSEGV, si_code, address, tsk, 0);
797 797
798 return; 798 return;
799 } 799 }
800 800
801 if (is_f00f_bug(regs, address)) 801 if (is_f00f_bug(regs, address))
802 return; 802 return;
803 803
804 no_context(regs, error_code, address, SIGSEGV, si_code); 804 no_context(regs, error_code, address, SIGSEGV, si_code);
805 } 805 }
806 806
807 static noinline void 807 static noinline void
808 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code, 808 bad_area_nosemaphore(struct pt_regs *regs, unsigned long error_code,
809 unsigned long address) 809 unsigned long address)
810 { 810 {
811 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR); 811 __bad_area_nosemaphore(regs, error_code, address, SEGV_MAPERR);
812 } 812 }
813 813
814 static void 814 static void
815 __bad_area(struct pt_regs *regs, unsigned long error_code, 815 __bad_area(struct pt_regs *regs, unsigned long error_code,
816 unsigned long address, int si_code) 816 unsigned long address, int si_code)
817 { 817 {
818 struct mm_struct *mm = current->mm; 818 struct mm_struct *mm = current->mm;
819 819
820 /* 820 /*
821 * Something tried to access memory that isn't in our memory map.. 821 * Something tried to access memory that isn't in our memory map..
822 * Fix it, but check if it's kernel or user first.. 822 * Fix it, but check if it's kernel or user first..
823 */ 823 */
824 up_read(&mm->mmap_sem); 824 up_read(&mm->mmap_sem);
825 825
826 __bad_area_nosemaphore(regs, error_code, address, si_code); 826 __bad_area_nosemaphore(regs, error_code, address, si_code);
827 } 827 }
828 828
829 static noinline void 829 static noinline void
830 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address) 830 bad_area(struct pt_regs *regs, unsigned long error_code, unsigned long address)
831 { 831 {
832 __bad_area(regs, error_code, address, SEGV_MAPERR); 832 __bad_area(regs, error_code, address, SEGV_MAPERR);
833 } 833 }
834 834
835 static noinline void 835 static noinline void
836 bad_area_access_error(struct pt_regs *regs, unsigned long error_code, 836 bad_area_access_error(struct pt_regs *regs, unsigned long error_code,
837 unsigned long address) 837 unsigned long address)
838 { 838 {
839 __bad_area(regs, error_code, address, SEGV_ACCERR); 839 __bad_area(regs, error_code, address, SEGV_ACCERR);
840 } 840 }
841 841
842 static void 842 static void
843 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address, 843 do_sigbus(struct pt_regs *regs, unsigned long error_code, unsigned long address,
844 unsigned int fault) 844 unsigned int fault)
845 { 845 {
846 struct task_struct *tsk = current; 846 struct task_struct *tsk = current;
847 struct mm_struct *mm = tsk->mm;
848 int code = BUS_ADRERR; 847 int code = BUS_ADRERR;
849 848
850 up_read(&mm->mmap_sem);
851
852 /* Kernel mode? Handle exceptions or die: */ 849 /* Kernel mode? Handle exceptions or die: */
853 if (!(error_code & PF_USER)) { 850 if (!(error_code & PF_USER)) {
854 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR); 851 no_context(regs, error_code, address, SIGBUS, BUS_ADRERR);
855 return; 852 return;
856 } 853 }
857 854
858 /* User-space => ok to do another page fault: */ 855 /* User-space => ok to do another page fault: */
859 if (is_prefetch(regs, error_code, address)) 856 if (is_prefetch(regs, error_code, address))
860 return; 857 return;
861 858
862 tsk->thread.cr2 = address; 859 tsk->thread.cr2 = address;
863 tsk->thread.error_code = error_code; 860 tsk->thread.error_code = error_code;
864 tsk->thread.trap_nr = X86_TRAP_PF; 861 tsk->thread.trap_nr = X86_TRAP_PF;
865 862
866 #ifdef CONFIG_MEMORY_FAILURE 863 #ifdef CONFIG_MEMORY_FAILURE
867 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) { 864 if (fault & (VM_FAULT_HWPOISON|VM_FAULT_HWPOISON_LARGE)) {
868 printk(KERN_ERR 865 printk(KERN_ERR
869 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n", 866 "MCE: Killing %s:%d due to hardware memory corruption fault at %lx\n",
870 tsk->comm, tsk->pid, address); 867 tsk->comm, tsk->pid, address);
871 code = BUS_MCEERR_AR; 868 code = BUS_MCEERR_AR;
872 } 869 }
873 #endif 870 #endif
874 force_sig_info_fault(SIGBUS, code, address, tsk, fault); 871 force_sig_info_fault(SIGBUS, code, address, tsk, fault);
875 } 872 }
876 873
877 static noinline void 874 static noinline void
878 mm_fault_error(struct pt_regs *regs, unsigned long error_code, 875 mm_fault_error(struct pt_regs *regs, unsigned long error_code,
879 unsigned long address, unsigned int fault) 876 unsigned long address, unsigned int fault)
880 { 877 {
881 if (fatal_signal_pending(current) && !(error_code & PF_USER)) { 878 if (fatal_signal_pending(current) && !(error_code & PF_USER)) {
882 up_read(&current->mm->mmap_sem);
883 no_context(regs, error_code, address, 0, 0); 879 no_context(regs, error_code, address, 0, 0);
884 return; 880 return;
885 } 881 }
886 882
887 if (fault & VM_FAULT_OOM) { 883 if (fault & VM_FAULT_OOM) {
888 /* Kernel mode? Handle exceptions or die: */ 884 /* Kernel mode? Handle exceptions or die: */
889 if (!(error_code & PF_USER)) { 885 if (!(error_code & PF_USER)) {
890 up_read(&current->mm->mmap_sem);
891 no_context(regs, error_code, address, 886 no_context(regs, error_code, address,
892 SIGSEGV, SEGV_MAPERR); 887 SIGSEGV, SEGV_MAPERR);
893 return; 888 return;
894 } 889 }
895 890
896 up_read(&current->mm->mmap_sem);
897
898 /* 891 /*
899 * We ran out of memory, call the OOM killer, and return the 892 * We ran out of memory, call the OOM killer, and return the
900 * userspace (which will retry the fault, or kill us if we got 893 * userspace (which will retry the fault, or kill us if we got
901 * oom-killed): 894 * oom-killed):
902 */ 895 */
903 pagefault_out_of_memory(); 896 pagefault_out_of_memory();
904 } else { 897 } else {
905 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON| 898 if (fault & (VM_FAULT_SIGBUS|VM_FAULT_HWPOISON|
906 VM_FAULT_HWPOISON_LARGE)) 899 VM_FAULT_HWPOISON_LARGE))
907 do_sigbus(regs, error_code, address, fault); 900 do_sigbus(regs, error_code, address, fault);
908 else if (fault & VM_FAULT_SIGSEGV) 901 else if (fault & VM_FAULT_SIGSEGV)
909 bad_area_nosemaphore(regs, error_code, address); 902 bad_area_nosemaphore(regs, error_code, address);
910 else 903 else
911 BUG(); 904 BUG();
912 } 905 }
913 } 906 }
914 907
915 static int spurious_fault_check(unsigned long error_code, pte_t *pte) 908 static int spurious_fault_check(unsigned long error_code, pte_t *pte)
916 { 909 {
917 if ((error_code & PF_WRITE) && !pte_write(*pte)) 910 if ((error_code & PF_WRITE) && !pte_write(*pte))
918 return 0; 911 return 0;
919 912
920 if ((error_code & PF_INSTR) && !pte_exec(*pte)) 913 if ((error_code & PF_INSTR) && !pte_exec(*pte))
921 return 0; 914 return 0;
922 915
923 return 1; 916 return 1;
924 } 917 }
925 918
926 /* 919 /*
927 * Handle a spurious fault caused by a stale TLB entry. 920 * Handle a spurious fault caused by a stale TLB entry.
928 * 921 *
929 * This allows us to lazily refresh the TLB when increasing the 922 * This allows us to lazily refresh the TLB when increasing the
930 * permissions of a kernel page (RO -> RW or NX -> X). Doing it 923 * permissions of a kernel page (RO -> RW or NX -> X). Doing it
931 * eagerly is very expensive since that implies doing a full 924 * eagerly is very expensive since that implies doing a full
932 * cross-processor TLB flush, even if no stale TLB entries exist 925 * cross-processor TLB flush, even if no stale TLB entries exist
933 * on other processors. 926 * on other processors.
934 * 927 *
935 * Spurious faults may only occur if the TLB contains an entry with 928 * Spurious faults may only occur if the TLB contains an entry with
936 * fewer permission than the page table entry. Non-present (P = 0) 929 * fewer permission than the page table entry. Non-present (P = 0)
937 * and reserved bit (R = 1) faults are never spurious. 930 * and reserved bit (R = 1) faults are never spurious.
938 * 931 *
939 * There are no security implications to leaving a stale TLB when 932 * There are no security implications to leaving a stale TLB when
940 * increasing the permissions on a page. 933 * increasing the permissions on a page.
941 * 934 *
942 * Returns non-zero if a spurious fault was handled, zero otherwise. 935 * Returns non-zero if a spurious fault was handled, zero otherwise.
943 * 936 *
944 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3 937 * See Intel Developer's Manual Vol 3 Section 4.10.4.3, bullet 3
945 * (Optional Invalidation). 938 * (Optional Invalidation).
946 */ 939 */
947 static noinline int 940 static noinline int
948 spurious_fault(unsigned long error_code, unsigned long address) 941 spurious_fault(unsigned long error_code, unsigned long address)
949 { 942 {
950 pgd_t *pgd; 943 pgd_t *pgd;
951 pud_t *pud; 944 pud_t *pud;
952 pmd_t *pmd; 945 pmd_t *pmd;
953 pte_t *pte; 946 pte_t *pte;
954 int ret; 947 int ret;
955 948
956 /* 949 /*
957 * Only writes to RO or instruction fetches from NX may cause 950 * Only writes to RO or instruction fetches from NX may cause
958 * spurious faults. 951 * spurious faults.
959 * 952 *
960 * These could be from user or supervisor accesses but the TLB 953 * These could be from user or supervisor accesses but the TLB
961 * is only lazily flushed after a kernel mapping protection 954 * is only lazily flushed after a kernel mapping protection
962 * change, so user accesses are not expected to cause spurious 955 * change, so user accesses are not expected to cause spurious
963 * faults. 956 * faults.
964 */ 957 */
965 if (error_code != (PF_WRITE | PF_PROT) 958 if (error_code != (PF_WRITE | PF_PROT)
966 && error_code != (PF_INSTR | PF_PROT)) 959 && error_code != (PF_INSTR | PF_PROT))
967 return 0; 960 return 0;
968 961
969 pgd = init_mm.pgd + pgd_index(address); 962 pgd = init_mm.pgd + pgd_index(address);
970 if (!pgd_present(*pgd)) 963 if (!pgd_present(*pgd))
971 return 0; 964 return 0;
972 965
973 pud = pud_offset(pgd, address); 966 pud = pud_offset(pgd, address);
974 if (!pud_present(*pud)) 967 if (!pud_present(*pud))
975 return 0; 968 return 0;
976 969
977 if (pud_large(*pud)) 970 if (pud_large(*pud))
978 return spurious_fault_check(error_code, (pte_t *) pud); 971 return spurious_fault_check(error_code, (pte_t *) pud);
979 972
980 pmd = pmd_offset(pud, address); 973 pmd = pmd_offset(pud, address);
981 if (!pmd_present(*pmd)) 974 if (!pmd_present(*pmd))
982 return 0; 975 return 0;
983 976
984 if (pmd_large(*pmd)) 977 if (pmd_large(*pmd))
985 return spurious_fault_check(error_code, (pte_t *) pmd); 978 return spurious_fault_check(error_code, (pte_t *) pmd);
986 979
987 pte = pte_offset_kernel(pmd, address); 980 pte = pte_offset_kernel(pmd, address);
988 if (!pte_present(*pte)) 981 if (!pte_present(*pte))
989 return 0; 982 return 0;
990 983
991 ret = spurious_fault_check(error_code, pte); 984 ret = spurious_fault_check(error_code, pte);
992 if (!ret) 985 if (!ret)
993 return 0; 986 return 0;
994 987
995 /* 988 /*
996 * Make sure we have permissions in PMD. 989 * Make sure we have permissions in PMD.
997 * If not, then there's a bug in the page tables: 990 * If not, then there's a bug in the page tables:
998 */ 991 */
999 ret = spurious_fault_check(error_code, (pte_t *) pmd); 992 ret = spurious_fault_check(error_code, (pte_t *) pmd);
1000 WARN_ONCE(!ret, "PMD has incorrect permission bits\n"); 993 WARN_ONCE(!ret, "PMD has incorrect permission bits\n");
1001 994
1002 return ret; 995 return ret;
1003 } 996 }
1004 NOKPROBE_SYMBOL(spurious_fault); 997 NOKPROBE_SYMBOL(spurious_fault);
1005 998
1006 int show_unhandled_signals = 1; 999 int show_unhandled_signals = 1;
1007 1000
1008 static inline int 1001 static inline int
1009 access_error(unsigned long error_code, struct vm_area_struct *vma) 1002 access_error(unsigned long error_code, struct vm_area_struct *vma)
1010 { 1003 {
1011 if (error_code & PF_WRITE) { 1004 if (error_code & PF_WRITE) {
1012 /* write, present and write, not present: */ 1005 /* write, present and write, not present: */
1013 if (unlikely(!(vma->vm_flags & VM_WRITE))) 1006 if (unlikely(!(vma->vm_flags & VM_WRITE)))
1014 return 1; 1007 return 1;
1015 return 0; 1008 return 0;
1016 } 1009 }
1017 1010
1018 /* read, present: */ 1011 /* read, present: */
1019 if (unlikely(error_code & PF_PROT)) 1012 if (unlikely(error_code & PF_PROT))
1020 return 1; 1013 return 1;
1021 1014
1022 /* read, not present: */ 1015 /* read, not present: */
1023 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))) 1016 if (unlikely(!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE))))
1024 return 1; 1017 return 1;
1025 1018
1026 return 0; 1019 return 0;
1027 } 1020 }
1028 1021
1029 static int fault_in_kernel_space(unsigned long address) 1022 static int fault_in_kernel_space(unsigned long address)
1030 { 1023 {
1031 return address >= TASK_SIZE_MAX; 1024 return address >= TASK_SIZE_MAX;
1032 } 1025 }
1033 1026
1034 static inline bool smap_violation(int error_code, struct pt_regs *regs) 1027 static inline bool smap_violation(int error_code, struct pt_regs *regs)
1035 { 1028 {
1036 if (!IS_ENABLED(CONFIG_X86_SMAP)) 1029 if (!IS_ENABLED(CONFIG_X86_SMAP))
1037 return false; 1030 return false;
1038 1031
1039 if (!static_cpu_has(X86_FEATURE_SMAP)) 1032 if (!static_cpu_has(X86_FEATURE_SMAP))
1040 return false; 1033 return false;
1041 1034
1042 if (error_code & PF_USER) 1035 if (error_code & PF_USER)
1043 return false; 1036 return false;
1044 1037
1045 if (!user_mode_vm(regs) && (regs->flags & X86_EFLAGS_AC)) 1038 if (!user_mode_vm(regs) && (regs->flags & X86_EFLAGS_AC))
1046 return false; 1039 return false;
1047 1040
1048 return true; 1041 return true;
1049 } 1042 }
1050 1043
1051 /* 1044 /*
1052 * This routine handles page faults. It determines the address, 1045 * This routine handles page faults. It determines the address,
1053 * and the problem, and then passes it off to one of the appropriate 1046 * and the problem, and then passes it off to one of the appropriate
1054 * routines. 1047 * routines.
1055 * 1048 *
1056 * This function must have noinline because both callers 1049 * This function must have noinline because both callers
1057 * {,trace_}do_page_fault() have notrace on. Having this an actual function 1050 * {,trace_}do_page_fault() have notrace on. Having this an actual function
1058 * guarantees there's a function trace entry. 1051 * guarantees there's a function trace entry.
1059 */ 1052 */
1060 static noinline void 1053 static noinline void
1061 __do_page_fault(struct pt_regs *regs, unsigned long error_code, 1054 __do_page_fault(struct pt_regs *regs, unsigned long error_code,
1062 unsigned long address) 1055 unsigned long address)
1063 { 1056 {
1064 struct vm_area_struct *vma; 1057 struct vm_area_struct *vma;
1065 struct task_struct *tsk; 1058 struct task_struct *tsk;
1066 struct mm_struct *mm; 1059 struct mm_struct *mm;
1067 int fault; 1060 int fault;
1068 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE; 1061 unsigned int flags = FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_KILLABLE;
1069 1062
1070 tsk = current; 1063 tsk = current;
1071 mm = tsk->mm; 1064 mm = tsk->mm;
1072 1065
1073 /* 1066 /*
1074 * Detect and handle instructions that would cause a page fault for 1067 * Detect and handle instructions that would cause a page fault for
1075 * both a tracked kernel page and a userspace page. 1068 * both a tracked kernel page and a userspace page.
1076 */ 1069 */
1077 if (kmemcheck_active(regs)) 1070 if (kmemcheck_active(regs))
1078 kmemcheck_hide(regs); 1071 kmemcheck_hide(regs);
1079 prefetchw(&mm->mmap_sem); 1072 prefetchw(&mm->mmap_sem);
1080 1073
1081 if (unlikely(kmmio_fault(regs, address))) 1074 if (unlikely(kmmio_fault(regs, address)))
1082 return; 1075 return;
1083 1076
1084 /* 1077 /*
1085 * We fault-in kernel-space virtual memory on-demand. The 1078 * We fault-in kernel-space virtual memory on-demand. The
1086 * 'reference' page table is init_mm.pgd. 1079 * 'reference' page table is init_mm.pgd.
1087 * 1080 *
1088 * NOTE! We MUST NOT take any locks for this case. We may 1081 * NOTE! We MUST NOT take any locks for this case. We may
1089 * be in an interrupt or a critical region, and should 1082 * be in an interrupt or a critical region, and should
1090 * only copy the information from the master page table, 1083 * only copy the information from the master page table,
1091 * nothing more. 1084 * nothing more.
1092 * 1085 *
1093 * This verifies that the fault happens in kernel space 1086 * This verifies that the fault happens in kernel space
1094 * (error_code & 4) == 0, and that the fault was not a 1087 * (error_code & 4) == 0, and that the fault was not a
1095 * protection error (error_code & 9) == 0. 1088 * protection error (error_code & 9) == 0.
1096 */ 1089 */
1097 if (unlikely(fault_in_kernel_space(address))) { 1090 if (unlikely(fault_in_kernel_space(address))) {
1098 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) { 1091 if (!(error_code & (PF_RSVD | PF_USER | PF_PROT))) {
1099 if (vmalloc_fault(address) >= 0) 1092 if (vmalloc_fault(address) >= 0)
1100 return; 1093 return;
1101 1094
1102 if (kmemcheck_fault(regs, address, error_code)) 1095 if (kmemcheck_fault(regs, address, error_code))
1103 return; 1096 return;
1104 } 1097 }
1105 1098
1106 /* Can handle a stale RO->RW TLB: */ 1099 /* Can handle a stale RO->RW TLB: */
1107 if (spurious_fault(error_code, address)) 1100 if (spurious_fault(error_code, address))
1108 return; 1101 return;
1109 1102
1110 /* kprobes don't want to hook the spurious faults: */ 1103 /* kprobes don't want to hook the spurious faults: */
1111 if (kprobes_fault(regs)) 1104 if (kprobes_fault(regs))
1112 return; 1105 return;
1113 /* 1106 /*
1114 * Don't take the mm semaphore here. If we fixup a prefetch 1107 * Don't take the mm semaphore here. If we fixup a prefetch
1115 * fault we could otherwise deadlock: 1108 * fault we could otherwise deadlock:
1116 */ 1109 */
1117 bad_area_nosemaphore(regs, error_code, address); 1110 bad_area_nosemaphore(regs, error_code, address);
1118 1111
1119 return; 1112 return;
1120 } 1113 }
1121 1114
1122 /* kprobes don't want to hook the spurious faults: */ 1115 /* kprobes don't want to hook the spurious faults: */
1123 if (unlikely(kprobes_fault(regs))) 1116 if (unlikely(kprobes_fault(regs)))
1124 return; 1117 return;
1125 1118
1126 if (unlikely(error_code & PF_RSVD)) 1119 if (unlikely(error_code & PF_RSVD))
1127 pgtable_bad(regs, error_code, address); 1120 pgtable_bad(regs, error_code, address);
1128 1121
1129 if (unlikely(smap_violation(error_code, regs))) { 1122 if (unlikely(smap_violation(error_code, regs))) {
1130 bad_area_nosemaphore(regs, error_code, address); 1123 bad_area_nosemaphore(regs, error_code, address);
1131 return; 1124 return;
1132 } 1125 }
1133 1126
1134 /* 1127 /*
1135 * If we're in an interrupt, have no user context or are running 1128 * If we're in an interrupt, have no user context or are running
1136 * in an atomic region then we must not take the fault: 1129 * in an atomic region then we must not take the fault:
1137 */ 1130 */
1138 if (unlikely(in_atomic() || !mm)) { 1131 if (unlikely(in_atomic() || !mm)) {
1139 bad_area_nosemaphore(regs, error_code, address); 1132 bad_area_nosemaphore(regs, error_code, address);
1140 return; 1133 return;
1141 } 1134 }
1142 1135
1143 /* 1136 /*
1144 * It's safe to allow irq's after cr2 has been saved and the 1137 * It's safe to allow irq's after cr2 has been saved and the
1145 * vmalloc fault has been handled. 1138 * vmalloc fault has been handled.
1146 * 1139 *
1147 * User-mode registers count as a user access even for any 1140 * User-mode registers count as a user access even for any
1148 * potential system fault or CPU buglet: 1141 * potential system fault or CPU buglet:
1149 */ 1142 */
1150 if (user_mode_vm(regs)) { 1143 if (user_mode_vm(regs)) {
1151 local_irq_enable(); 1144 local_irq_enable();
1152 error_code |= PF_USER; 1145 error_code |= PF_USER;
1153 flags |= FAULT_FLAG_USER; 1146 flags |= FAULT_FLAG_USER;
1154 } else { 1147 } else {
1155 if (regs->flags & X86_EFLAGS_IF) 1148 if (regs->flags & X86_EFLAGS_IF)
1156 local_irq_enable(); 1149 local_irq_enable();
1157 } 1150 }
1158 1151
1159 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address); 1152 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, regs, address);
1160 1153
1161 if (error_code & PF_WRITE) 1154 if (error_code & PF_WRITE)
1162 flags |= FAULT_FLAG_WRITE; 1155 flags |= FAULT_FLAG_WRITE;
1163 1156
1164 /* 1157 /*
1165 * When running in the kernel we expect faults to occur only to 1158 * When running in the kernel we expect faults to occur only to
1166 * addresses in user space. All other faults represent errors in 1159 * addresses in user space. All other faults represent errors in
1167 * the kernel and should generate an OOPS. Unfortunately, in the 1160 * the kernel and should generate an OOPS. Unfortunately, in the
1168 * case of an erroneous fault occurring in a code path which already 1161 * case of an erroneous fault occurring in a code path which already
1169 * holds mmap_sem we will deadlock attempting to validate the fault 1162 * holds mmap_sem we will deadlock attempting to validate the fault
1170 * against the address space. Luckily the kernel only validly 1163 * against the address space. Luckily the kernel only validly
1171 * references user space from well defined areas of code, which are 1164 * references user space from well defined areas of code, which are
1172 * listed in the exceptions table. 1165 * listed in the exceptions table.
1173 * 1166 *
1174 * As the vast majority of faults will be valid we will only perform 1167 * As the vast majority of faults will be valid we will only perform
1175 * the source reference check when there is a possibility of a 1168 * the source reference check when there is a possibility of a
1176 * deadlock. Attempt to lock the address space, if we cannot we then 1169 * deadlock. Attempt to lock the address space, if we cannot we then
1177 * validate the source. If this is invalid we can skip the address 1170 * validate the source. If this is invalid we can skip the address
1178 * space check, thus avoiding the deadlock: 1171 * space check, thus avoiding the deadlock:
1179 */ 1172 */
1180 if (unlikely(!down_read_trylock(&mm->mmap_sem))) { 1173 if (unlikely(!down_read_trylock(&mm->mmap_sem))) {
1181 if ((error_code & PF_USER) == 0 && 1174 if ((error_code & PF_USER) == 0 &&
1182 !search_exception_tables(regs->ip)) { 1175 !search_exception_tables(regs->ip)) {
1183 bad_area_nosemaphore(regs, error_code, address); 1176 bad_area_nosemaphore(regs, error_code, address);
1184 return; 1177 return;
1185 } 1178 }
1186 retry: 1179 retry:
1187 down_read(&mm->mmap_sem); 1180 down_read(&mm->mmap_sem);
1188 } else { 1181 } else {
1189 /* 1182 /*
1190 * The above down_read_trylock() might have succeeded in 1183 * The above down_read_trylock() might have succeeded in
1191 * which case we'll have missed the might_sleep() from 1184 * which case we'll have missed the might_sleep() from
1192 * down_read(): 1185 * down_read():
1193 */ 1186 */
1194 might_sleep(); 1187 might_sleep();
1195 } 1188 }
1196 1189
1197 vma = find_vma(mm, address); 1190 vma = find_vma(mm, address);
1198 if (unlikely(!vma)) { 1191 if (unlikely(!vma)) {
1199 bad_area(regs, error_code, address); 1192 bad_area(regs, error_code, address);
1200 return; 1193 return;
1201 } 1194 }
1202 if (likely(vma->vm_start <= address)) 1195 if (likely(vma->vm_start <= address))
1203 goto good_area; 1196 goto good_area;
1204 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) { 1197 if (unlikely(!(vma->vm_flags & VM_GROWSDOWN))) {
1205 bad_area(regs, error_code, address); 1198 bad_area(regs, error_code, address);
1206 return; 1199 return;
1207 } 1200 }
1208 if (error_code & PF_USER) { 1201 if (error_code & PF_USER) {
1209 /* 1202 /*
1210 * Accessing the stack below %sp is always a bug. 1203 * Accessing the stack below %sp is always a bug.
1211 * The large cushion allows instructions like enter 1204 * The large cushion allows instructions like enter
1212 * and pusha to work. ("enter $65535, $31" pushes 1205 * and pusha to work. ("enter $65535, $31" pushes
1213 * 32 pointers and then decrements %sp by 65535.) 1206 * 32 pointers and then decrements %sp by 65535.)
1214 */ 1207 */
1215 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) { 1208 if (unlikely(address + 65536 + 32 * sizeof(unsigned long) < regs->sp)) {
1216 bad_area(regs, error_code, address); 1209 bad_area(regs, error_code, address);
1217 return; 1210 return;
1218 } 1211 }
1219 } 1212 }
1220 if (unlikely(expand_stack(vma, address))) { 1213 if (unlikely(expand_stack(vma, address))) {
1221 bad_area(regs, error_code, address); 1214 bad_area(regs, error_code, address);
1222 return; 1215 return;
1223 } 1216 }
1224 1217
1225 /* 1218 /*
1226 * Ok, we have a good vm_area for this memory access, so 1219 * Ok, we have a good vm_area for this memory access, so
1227 * we can handle it.. 1220 * we can handle it..
1228 */ 1221 */
1229 good_area: 1222 good_area:
1230 if (unlikely(access_error(error_code, vma))) { 1223 if (unlikely(access_error(error_code, vma))) {
1231 bad_area_access_error(regs, error_code, address); 1224 bad_area_access_error(regs, error_code, address);
1232 return; 1225 return;
1233 } 1226 }
1234 1227
1235 /* 1228 /*
1236 * If for any reason at all we couldn't handle the fault, 1229 * If for any reason at all we couldn't handle the fault,
1237 * make sure we exit gracefully rather than endlessly redo 1230 * make sure we exit gracefully rather than endlessly redo
1238 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if 1231 * the fault. Since we never set FAULT_FLAG_RETRY_NOWAIT, if
1239 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked. 1232 * we get VM_FAULT_RETRY back, the mmap_sem has been unlocked.
1240 */ 1233 */
1241 fault = handle_mm_fault(mm, vma, address, flags); 1234 fault = handle_mm_fault(mm, vma, address, flags);
1242 1235
1243 /* 1236 /*
1244 * If we need to retry but a fatal signal is pending, handle the 1237 * If we need to retry but a fatal signal is pending, handle the
1245 * signal first. We do not need to release the mmap_sem because it 1238 * signal first. We do not need to release the mmap_sem because it
1246 * would already be released in __lock_page_or_retry in mm/filemap.c. 1239 * would already be released in __lock_page_or_retry in mm/filemap.c.
1247 */ 1240 */
1248 if (unlikely((fault & VM_FAULT_RETRY) && fatal_signal_pending(current))) 1241 if (unlikely((fault & VM_FAULT_RETRY) && fatal_signal_pending(current)))
1249 return; 1242 return;
1250 1243
1251 if (unlikely(fault & VM_FAULT_ERROR)) { 1244 if (unlikely(fault & VM_FAULT_ERROR)) {
1245 up_read(&mm->mmap_sem);
1252 mm_fault_error(regs, error_code, address, fault); 1246 mm_fault_error(regs, error_code, address, fault);
1253 return; 1247 return;
1254 } 1248 }
1255 1249
1256 /* 1250 /*
1257 * Major/minor page fault accounting is only done on the 1251 * Major/minor page fault accounting is only done on the
1258 * initial attempt. If we go through a retry, it is extremely 1252 * initial attempt. If we go through a retry, it is extremely
1259 * likely that the page will be found in page cache at that point. 1253 * likely that the page will be found in page cache at that point.
1260 */ 1254 */
1261 if (flags & FAULT_FLAG_ALLOW_RETRY) { 1255 if (flags & FAULT_FLAG_ALLOW_RETRY) {
1262 if (fault & VM_FAULT_MAJOR) { 1256 if (fault & VM_FAULT_MAJOR) {
1263 tsk->maj_flt++; 1257 tsk->maj_flt++;
1264 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 1258 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1,
1265 regs, address); 1259 regs, address);
1266 } else { 1260 } else {
1267 tsk->min_flt++; 1261 tsk->min_flt++;
1268 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 1262 perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1,
1269 regs, address); 1263 regs, address);
1270 } 1264 }
1271 if (fault & VM_FAULT_RETRY) { 1265 if (fault & VM_FAULT_RETRY) {
1272 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk 1266 /* Clear FAULT_FLAG_ALLOW_RETRY to avoid any risk
1273 * of starvation. */ 1267 * of starvation. */
1274 flags &= ~FAULT_FLAG_ALLOW_RETRY; 1268 flags &= ~FAULT_FLAG_ALLOW_RETRY;
1275 flags |= FAULT_FLAG_TRIED; 1269 flags |= FAULT_FLAG_TRIED;
1276 goto retry; 1270 goto retry;
1277 } 1271 }
1278 } 1272 }
1279 1273
1280 check_v8086_mode(regs, address, tsk); 1274 check_v8086_mode(regs, address, tsk);
1281 1275
1282 up_read(&mm->mmap_sem); 1276 up_read(&mm->mmap_sem);
1283 } 1277 }
1284 NOKPROBE_SYMBOL(__do_page_fault); 1278 NOKPROBE_SYMBOL(__do_page_fault);
1285 1279
1286 dotraplinkage void notrace 1280 dotraplinkage void notrace
1287 do_page_fault(struct pt_regs *regs, unsigned long error_code) 1281 do_page_fault(struct pt_regs *regs, unsigned long error_code)
1288 { 1282 {
1289 unsigned long address = read_cr2(); /* Get the faulting address */ 1283 unsigned long address = read_cr2(); /* Get the faulting address */
1290 enum ctx_state prev_state; 1284 enum ctx_state prev_state;
1291 1285
1292 /* 1286 /*
1293 * We must have this function tagged with __kprobes, notrace and call 1287 * We must have this function tagged with __kprobes, notrace and call
1294 * read_cr2() before calling anything else. To avoid calling any kind 1288 * read_cr2() before calling anything else. To avoid calling any kind
1295 * of tracing machinery before we've observed the CR2 value. 1289 * of tracing machinery before we've observed the CR2 value.
1296 * 1290 *
1297 * exception_{enter,exit}() contain all sorts of tracepoints. 1291 * exception_{enter,exit}() contain all sorts of tracepoints.
1298 */ 1292 */
1299 1293
1300 prev_state = exception_enter(); 1294 prev_state = exception_enter();
1301 __do_page_fault(regs, error_code, address); 1295 __do_page_fault(regs, error_code, address);
1302 exception_exit(prev_state); 1296 exception_exit(prev_state);
1303 } 1297 }
1304 NOKPROBE_SYMBOL(do_page_fault); 1298 NOKPROBE_SYMBOL(do_page_fault);
1305 1299
1306 #ifdef CONFIG_TRACING 1300 #ifdef CONFIG_TRACING
1307 static nokprobe_inline void 1301 static nokprobe_inline void
1308 trace_page_fault_entries(unsigned long address, struct pt_regs *regs, 1302 trace_page_fault_entries(unsigned long address, struct pt_regs *regs,
1309 unsigned long error_code) 1303 unsigned long error_code)
1310 { 1304 {
1311 if (user_mode(regs)) 1305 if (user_mode(regs))
1312 trace_page_fault_user(address, regs, error_code); 1306 trace_page_fault_user(address, regs, error_code);
1313 else 1307 else
1314 trace_page_fault_kernel(address, regs, error_code); 1308 trace_page_fault_kernel(address, regs, error_code);
1315 } 1309 }
1316 1310
1317 dotraplinkage void notrace 1311 dotraplinkage void notrace
1318 trace_do_page_fault(struct pt_regs *regs, unsigned long error_code) 1312 trace_do_page_fault(struct pt_regs *regs, unsigned long error_code)
1319 { 1313 {
1320 /* 1314 /*
1321 * The exception_enter and tracepoint processing could 1315 * The exception_enter and tracepoint processing could
1322 * trigger another page faults (user space callchain 1316 * trigger another page faults (user space callchain
1323 * reading) and destroy the original cr2 value, so read 1317 * reading) and destroy the original cr2 value, so read
1324 * the faulting address now. 1318 * the faulting address now.
1325 */ 1319 */
1326 unsigned long address = read_cr2(); 1320 unsigned long address = read_cr2();
1327 enum ctx_state prev_state; 1321 enum ctx_state prev_state;
1328 1322
1329 prev_state = exception_enter(); 1323 prev_state = exception_enter();
1330 trace_page_fault_entries(address, regs, error_code); 1324 trace_page_fault_entries(address, regs, error_code);
1331 __do_page_fault(regs, error_code, address); 1325 __do_page_fault(regs, error_code, address);
1332 exception_exit(prev_state); 1326 exception_exit(prev_state);
1333 } 1327 }
1334 NOKPROBE_SYMBOL(trace_do_page_fault); 1328 NOKPROBE_SYMBOL(trace_do_page_fault);
1335 #endif /* CONFIG_TRACING */ 1329 #endif /* CONFIG_TRACING */