xref: /openbmc/linux/arch/x86/mm/extable.c (revision 8440bb9b)
1 #include <linux/extable.h>
2 #include <linux/uaccess.h>
3 #include <linux/sched/debug.h>
4 #include <xen/xen.h>
5 
6 #include <asm/fpu/internal.h>
7 #include <asm/traps.h>
8 #include <asm/kdebug.h>
9 
10 typedef bool (*ex_handler_t)(const struct exception_table_entry *,
11 			    struct pt_regs *, int, unsigned long,
12 			    unsigned long);
13 
14 static inline unsigned long
15 ex_fixup_addr(const struct exception_table_entry *x)
16 {
17 	return (unsigned long)&x->fixup + x->fixup;
18 }
19 static inline ex_handler_t
20 ex_fixup_handler(const struct exception_table_entry *x)
21 {
22 	return (ex_handler_t)((unsigned long)&x->handler + x->handler);
23 }
24 
25 __visible bool ex_handler_default(const struct exception_table_entry *fixup,
26 				  struct pt_regs *regs, int trapnr,
27 				  unsigned long error_code,
28 				  unsigned long fault_addr)
29 {
30 	regs->ip = ex_fixup_addr(fixup);
31 	return true;
32 }
33 EXPORT_SYMBOL(ex_handler_default);
34 
35 __visible bool ex_handler_fault(const struct exception_table_entry *fixup,
36 				struct pt_regs *regs, int trapnr,
37 				unsigned long error_code,
38 				unsigned long fault_addr)
39 {
40 	regs->ip = ex_fixup_addr(fixup);
41 	regs->ax = trapnr;
42 	return true;
43 }
44 EXPORT_SYMBOL_GPL(ex_handler_fault);
45 
46 /*
47  * Handler for UD0 exception following a failed test against the
48  * result of a refcount inc/dec/add/sub.
49  */
50 __visible bool ex_handler_refcount(const struct exception_table_entry *fixup,
51 				   struct pt_regs *regs, int trapnr,
52 				   unsigned long error_code,
53 				   unsigned long fault_addr)
54 {
55 	/* First unconditionally saturate the refcount. */
56 	*(int *)regs->cx = INT_MIN / 2;
57 
58 	/*
59 	 * Strictly speaking, this reports the fixup destination, not
60 	 * the fault location, and not the actually overflowing
61 	 * instruction, which is the instruction before the "js", but
62 	 * since that instruction could be a variety of lengths, just
63 	 * report the location after the overflow, which should be close
64 	 * enough for finding the overflow, as it's at least back in
65 	 * the function, having returned from .text.unlikely.
66 	 */
67 	regs->ip = ex_fixup_addr(fixup);
68 
69 	/*
70 	 * This function has been called because either a negative refcount
71 	 * value was seen by any of the refcount functions, or a zero
72 	 * refcount value was seen by refcount_dec().
73 	 *
74 	 * If we crossed from INT_MAX to INT_MIN, OF (Overflow Flag: result
75 	 * wrapped around) will be set. Additionally, seeing the refcount
76 	 * reach 0 will set ZF (Zero Flag: result was zero). In each of
77 	 * these cases we want a report, since it's a boundary condition.
78 	 * The SF case is not reported since it indicates post-boundary
79 	 * manipulations below zero or above INT_MAX. And if none of the
80 	 * flags are set, something has gone very wrong, so report it.
81 	 */
82 	if (regs->flags & (X86_EFLAGS_OF | X86_EFLAGS_ZF)) {
83 		bool zero = regs->flags & X86_EFLAGS_ZF;
84 
85 		refcount_error_report(regs, zero ? "hit zero" : "overflow");
86 	} else if ((regs->flags & X86_EFLAGS_SF) == 0) {
87 		/* Report if none of OF, ZF, nor SF are set. */
88 		refcount_error_report(regs, "unexpected saturation");
89 	}
90 
91 	return true;
92 }
93 EXPORT_SYMBOL(ex_handler_refcount);
94 
95 /*
96  * Handler for when we fail to restore a task's FPU state.  We should never get
97  * here because the FPU state of a task using the FPU (task->thread.fpu.state)
98  * should always be valid.  However, past bugs have allowed userspace to set
99  * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn().
100  * These caused XRSTOR to fail when switching to the task, leaking the FPU
101  * registers of the task previously executing on the CPU.  Mitigate this class
102  * of vulnerability by restoring from the initial state (essentially, zeroing
103  * out all the FPU registers) if we can't restore from the task's FPU state.
104  */
105 __visible bool ex_handler_fprestore(const struct exception_table_entry *fixup,
106 				    struct pt_regs *regs, int trapnr,
107 				    unsigned long error_code,
108 				    unsigned long fault_addr)
109 {
110 	regs->ip = ex_fixup_addr(fixup);
111 
112 	WARN_ONCE(1, "Bad FPU state detected at %pB, reinitializing FPU registers.",
113 		  (void *)instruction_pointer(regs));
114 
115 	__copy_kernel_to_fpregs(&init_fpstate, -1);
116 	return true;
117 }
118 EXPORT_SYMBOL_GPL(ex_handler_fprestore);
119 
120 __visible bool ex_handler_uaccess(const struct exception_table_entry *fixup,
121 				  struct pt_regs *regs, int trapnr,
122 				  unsigned long error_code,
123 				  unsigned long fault_addr)
124 {
125 	WARN_ONCE(trapnr == X86_TRAP_GP, "General protection fault in user access. Non-canonical address?");
126 	regs->ip = ex_fixup_addr(fixup);
127 	return true;
128 }
129 EXPORT_SYMBOL(ex_handler_uaccess);
130 
131 __visible bool ex_handler_ext(const struct exception_table_entry *fixup,
132 			      struct pt_regs *regs, int trapnr,
133 			      unsigned long error_code,
134 			      unsigned long fault_addr)
135 {
136 	/* Special hack for uaccess_err */
137 	current->thread.uaccess_err = 1;
138 	regs->ip = ex_fixup_addr(fixup);
139 	return true;
140 }
141 EXPORT_SYMBOL(ex_handler_ext);
142 
143 __visible bool ex_handler_rdmsr_unsafe(const struct exception_table_entry *fixup,
144 				       struct pt_regs *regs, int trapnr,
145 				       unsigned long error_code,
146 				       unsigned long fault_addr)
147 {
148 	if (pr_warn_once("unchecked MSR access error: RDMSR from 0x%x at rIP: 0x%lx (%pF)\n",
149 			 (unsigned int)regs->cx, regs->ip, (void *)regs->ip))
150 		show_stack_regs(regs);
151 
152 	/* Pretend that the read succeeded and returned 0. */
153 	regs->ip = ex_fixup_addr(fixup);
154 	regs->ax = 0;
155 	regs->dx = 0;
156 	return true;
157 }
158 EXPORT_SYMBOL(ex_handler_rdmsr_unsafe);
159 
160 __visible bool ex_handler_wrmsr_unsafe(const struct exception_table_entry *fixup,
161 				       struct pt_regs *regs, int trapnr,
162 				       unsigned long error_code,
163 				       unsigned long fault_addr)
164 {
165 	if (pr_warn_once("unchecked MSR access error: WRMSR to 0x%x (tried to write 0x%08x%08x) at rIP: 0x%lx (%pF)\n",
166 			 (unsigned int)regs->cx, (unsigned int)regs->dx,
167 			 (unsigned int)regs->ax,  regs->ip, (void *)regs->ip))
168 		show_stack_regs(regs);
169 
170 	/* Pretend that the write succeeded. */
171 	regs->ip = ex_fixup_addr(fixup);
172 	return true;
173 }
174 EXPORT_SYMBOL(ex_handler_wrmsr_unsafe);
175 
176 __visible bool ex_handler_clear_fs(const struct exception_table_entry *fixup,
177 				   struct pt_regs *regs, int trapnr,
178 				   unsigned long error_code,
179 				   unsigned long fault_addr)
180 {
181 	if (static_cpu_has(X86_BUG_NULL_SEG))
182 		asm volatile ("mov %0, %%fs" : : "rm" (__USER_DS));
183 	asm volatile ("mov %0, %%fs" : : "rm" (0));
184 	return ex_handler_default(fixup, regs, trapnr, error_code, fault_addr);
185 }
186 EXPORT_SYMBOL(ex_handler_clear_fs);
187 
188 __visible bool ex_has_fault_handler(unsigned long ip)
189 {
190 	const struct exception_table_entry *e;
191 	ex_handler_t handler;
192 
193 	e = search_exception_tables(ip);
194 	if (!e)
195 		return false;
196 	handler = ex_fixup_handler(e);
197 
198 	return handler == ex_handler_fault;
199 }
200 
201 int fixup_exception(struct pt_regs *regs, int trapnr, unsigned long error_code,
202 		    unsigned long fault_addr)
203 {
204 	const struct exception_table_entry *e;
205 	ex_handler_t handler;
206 
207 #ifdef CONFIG_PNPBIOS
208 	if (unlikely(SEGMENT_IS_PNP_CODE(regs->cs))) {
209 		extern u32 pnp_bios_fault_eip, pnp_bios_fault_esp;
210 		extern u32 pnp_bios_is_utter_crap;
211 		pnp_bios_is_utter_crap = 1;
212 		printk(KERN_CRIT "PNPBIOS fault.. attempting recovery.\n");
213 		__asm__ volatile(
214 			"movl %0, %%esp\n\t"
215 			"jmp *%1\n\t"
216 			: : "g" (pnp_bios_fault_esp), "g" (pnp_bios_fault_eip));
217 		panic("do_trap: can't hit this");
218 	}
219 #endif
220 
221 	e = search_exception_tables(regs->ip);
222 	if (!e)
223 		return 0;
224 
225 	handler = ex_fixup_handler(e);
226 	return handler(e, regs, trapnr, error_code, fault_addr);
227 }
228 
229 extern unsigned int early_recursion_flag;
230 
231 /* Restricted version used during very early boot */
232 void __init early_fixup_exception(struct pt_regs *regs, int trapnr)
233 {
234 	/* Ignore early NMIs. */
235 	if (trapnr == X86_TRAP_NMI)
236 		return;
237 
238 	if (early_recursion_flag > 2)
239 		goto halt_loop;
240 
241 	/*
242 	 * Old CPUs leave the high bits of CS on the stack
243 	 * undefined.  I'm not sure which CPUs do this, but at least
244 	 * the 486 DX works this way.
245 	 * Xen pv domains are not using the default __KERNEL_CS.
246 	 */
247 	if (!xen_pv_domain() && regs->cs != __KERNEL_CS)
248 		goto fail;
249 
250 	/*
251 	 * The full exception fixup machinery is available as soon as
252 	 * the early IDT is loaded.  This means that it is the
253 	 * responsibility of extable users to either function correctly
254 	 * when handlers are invoked early or to simply avoid causing
255 	 * exceptions before they're ready to handle them.
256 	 *
257 	 * This is better than filtering which handlers can be used,
258 	 * because refusing to call a handler here is guaranteed to
259 	 * result in a hard-to-debug panic.
260 	 *
261 	 * Keep in mind that not all vectors actually get here.  Early
262 	 * page faults, for example, are special.
263 	 */
264 	if (fixup_exception(regs, trapnr, regs->orig_ax, 0))
265 		return;
266 
267 	if (fixup_bug(regs, trapnr))
268 		return;
269 
270 fail:
271 	early_printk("PANIC: early exception 0x%02x IP %lx:%lx error %lx cr2 0x%lx\n",
272 		     (unsigned)trapnr, (unsigned long)regs->cs, regs->ip,
273 		     regs->orig_ax, read_cr2());
274 
275 	show_regs(regs);
276 
277 halt_loop:
278 	while (true)
279 		halt();
280 }
281