1// SPDX-License-Identifier: GPL-2.0-only
2#include <linux/extable.h>
3#include <linux/uaccess.h>
4#include <linux/sched/debug.h>
5#include <linux/bitfield.h>
6#include <xen/xen.h>
7
8#include <asm/fpu/api.h>
9#include <asm/fred.h>
10#include <asm/sev.h>
11#include <asm/traps.h>
12#include <asm/kdebug.h>
13#include <asm/insn-eval.h>
14#include <asm/sgx.h>
15
16static inline unsigned long *pt_regs_nr(struct pt_regs *regs, int nr)
17{
18 int reg_offset = pt_regs_offset(regs, regno: nr);
19 static unsigned long __dummy;
20
21 if (WARN_ON_ONCE(reg_offset < 0))
22 return &__dummy;
23
24 return (unsigned long *)((unsigned long)regs + reg_offset);
25}
26
27static inline unsigned long
28ex_fixup_addr(const struct exception_table_entry *x)
29{
30 return (unsigned long)&x->fixup + x->fixup;
31}
32
33static bool ex_handler_default(const struct exception_table_entry *e,
34 struct pt_regs *regs)
35{
36 if (e->data & EX_FLAG_CLEAR_AX)
37 regs->ax = 0;
38 if (e->data & EX_FLAG_CLEAR_DX)
39 regs->dx = 0;
40
41 regs->ip = ex_fixup_addr(x: e);
42 return true;
43}
44
45/*
46 * This is the *very* rare case where we do a "load_unaligned_zeropad()"
47 * and it's a page crosser into a non-existent page.
48 *
49 * This happens when we optimistically load a pathname a word-at-a-time
50 * and the name is less than the full word and the next page is not
51 * mapped. Typically that only happens for CONFIG_DEBUG_PAGEALLOC.
52 *
53 * NOTE! The faulting address is always a 'mov mem,reg' type instruction
54 * of size 'long', and the exception fixup must always point to right
55 * after the instruction.
56 */
57static bool ex_handler_zeropad(const struct exception_table_entry *e,
58 struct pt_regs *regs,
59 unsigned long fault_addr)
60{
61 struct insn insn;
62 const unsigned long mask = sizeof(long) - 1;
63 unsigned long offset, addr, next_ip, len;
64 unsigned long *reg;
65
66 next_ip = ex_fixup_addr(x: e);
67 len = next_ip - regs->ip;
68 if (len > MAX_INSN_SIZE)
69 return false;
70
71 if (insn_decode(insn: &insn, kaddr: (void *) regs->ip, buf_len: len, m: INSN_MODE_KERN))
72 return false;
73 if (insn.length != len)
74 return false;
75
76 if (insn.opcode.bytes[0] != 0x8b)
77 return false;
78 if (insn.opnd_bytes != sizeof(long))
79 return false;
80
81 addr = (unsigned long) insn_get_addr_ref(insn: &insn, regs);
82 if (addr == ~0ul)
83 return false;
84
85 offset = addr & mask;
86 addr = addr & ~mask;
87 if (fault_addr != addr + sizeof(long))
88 return false;
89
90 reg = insn_get_modrm_reg_ptr(insn: &insn, regs);
91 if (!reg)
92 return false;
93
94 *reg = *(unsigned long *)addr >> (offset * 8);
95 return ex_handler_default(e, regs);
96}
97
98static bool ex_handler_fault(const struct exception_table_entry *fixup,
99 struct pt_regs *regs, int trapnr)
100{
101 regs->ax = trapnr;
102 return ex_handler_default(e: fixup, regs);
103}
104
105static bool ex_handler_sgx(const struct exception_table_entry *fixup,
106 struct pt_regs *regs, int trapnr)
107{
108 regs->ax = trapnr | SGX_ENCLS_FAULT_FLAG;
109 return ex_handler_default(e: fixup, regs);
110}
111
112/*
113 * Handler for when we fail to restore a task's FPU state. We should never get
114 * here because the FPU state of a task using the FPU (struct fpu::fpstate)
115 * should always be valid. However, past bugs have allowed userspace to set
116 * reserved bits in the XSAVE area using PTRACE_SETREGSET or sys_rt_sigreturn().
117 * These caused XRSTOR to fail when switching to the task, leaking the FPU
118 * registers of the task previously executing on the CPU. Mitigate this class
119 * of vulnerability by restoring from the initial state (essentially, zeroing
120 * out all the FPU registers) if we can't restore from the task's FPU state.
121 */
122static bool ex_handler_fprestore(const struct exception_table_entry *fixup,
123 struct pt_regs *regs)
124{
125 WARN_ONCE(1, "Bad FPU state detected at %pB, reinitializing FPU registers.",
126 (void *)instruction_pointer(regs));
127
128 fpu_reset_from_exception_fixup();
129
130 return ex_handler_default(e: fixup, regs);
131}
132
133/*
134 * On x86-64, we end up being imprecise with 'access_ok()', and allow
135 * non-canonical user addresses to make the range comparisons simpler,
136 * and to not have to worry about LAM being enabled.
137 *
138 * In fact, we allow up to one page of "slop" at the sign boundary,
139 * which means that we can do access_ok() by just checking the sign
140 * of the pointer for the common case of having a small access size.
141 */
142static bool gp_fault_address_ok(unsigned long fault_address)
143{
144#ifdef CONFIG_X86_64
145 /* Is it in the "user space" part of the non-canonical space? */
146 if (valid_user_address(fault_address))
147 return true;
148
149 /* .. or just above it? */
150 fault_address -= PAGE_SIZE;
151 if (valid_user_address(fault_address))
152 return true;
153#endif
154 return false;
155}
156
157static bool ex_handler_uaccess(const struct exception_table_entry *fixup,
158 struct pt_regs *regs, int trapnr,
159 unsigned long fault_address)
160{
161 WARN_ONCE(trapnr == X86_TRAP_GP && !gp_fault_address_ok(fault_address),
162 "General protection fault in user access. Non-canonical address?");
163 return ex_handler_default(e: fixup, regs);
164}
165
166static bool ex_handler_msr(const struct exception_table_entry *fixup,
167 struct pt_regs *regs, bool wrmsr, bool safe, int reg)
168{
169 if (__ONCE_LITE_IF(!safe && wrmsr)) {
170 pr_warn("unchecked MSR access error: WRMSR to 0x%x (tried to write 0x%08x%08x) at rIP: 0x%lx (%pS)\n",
171 (unsigned int)regs->cx, (unsigned int)regs->dx,
172 (unsigned int)regs->ax, regs->ip, (void *)regs->ip);
173 show_stack_regs(regs);
174 }
175
176 if (__ONCE_LITE_IF(!safe && !wrmsr)) {
177 pr_warn("unchecked MSR access error: RDMSR from 0x%x at rIP: 0x%lx (%pS)\n",
178 (unsigned int)regs->cx, regs->ip, (void *)regs->ip);
179 show_stack_regs(regs);
180 }
181
182 if (!wrmsr) {
183 /* Pretend that the read succeeded and returned 0. */
184 regs->ax = 0;
185 regs->dx = 0;
186 }
187
188 if (safe)
189 *pt_regs_nr(regs, nr: reg) = -EIO;
190
191 return ex_handler_default(e: fixup, regs);
192}
193
194static bool ex_handler_clear_fs(const struct exception_table_entry *fixup,
195 struct pt_regs *regs)
196{
197 if (static_cpu_has(X86_BUG_NULL_SEG))
198 asm volatile ("mov %0, %%fs" : : "rm" (__USER_DS));
199 asm volatile ("mov %0, %%fs" : : "rm" (0));
200 return ex_handler_default(e: fixup, regs);
201}
202
203static bool ex_handler_imm_reg(const struct exception_table_entry *fixup,
204 struct pt_regs *regs, int reg, int imm)
205{
206 *pt_regs_nr(regs, nr: reg) = (long)imm;
207 return ex_handler_default(e: fixup, regs);
208}
209
210static bool ex_handler_ucopy_len(const struct exception_table_entry *fixup,
211 struct pt_regs *regs, int trapnr,
212 unsigned long fault_address,
213 int reg, int imm)
214{
215 regs->cx = imm * regs->cx + *pt_regs_nr(regs, nr: reg);
216 return ex_handler_uaccess(fixup, regs, trapnr, fault_address);
217}
218
219#ifdef CONFIG_X86_FRED
220static bool ex_handler_eretu(const struct exception_table_entry *fixup,
221 struct pt_regs *regs, unsigned long error_code)
222{
223 struct pt_regs *uregs = (struct pt_regs *)(regs->sp - offsetof(struct pt_regs, orig_ax));
224 unsigned short ss = uregs->ss;
225 unsigned short cs = uregs->cs;
226
227 /*
228 * Move the NMI bit from the invalid stack frame, which caused ERETU
229 * to fault, to the fault handler's stack frame, thus to unblock NMI
230 * with the fault handler's ERETS instruction ASAP if NMI is blocked.
231 */
232 regs->fred_ss.nmi = uregs->fred_ss.nmi;
233
234 /*
235 * Sync event information to uregs, i.e., the ERETU return frame, but
236 * is it safe to write to the ERETU return frame which is just above
237 * current event stack frame?
238 *
239 * The RSP used by FRED to push a stack frame is not the value in %rsp,
240 * it is calculated from %rsp with the following 2 steps:
241 * 1) RSP = %rsp - (IA32_FRED_CONFIG & 0x1c0) // Reserve N*64 bytes
242 * 2) RSP = RSP & ~0x3f // Align to a 64-byte cache line
243 * when an event delivery doesn't trigger a stack level change.
244 *
245 * Here is an example with N*64 (N=1) bytes reserved:
246 *
247 * 64-byte cache line ==> ______________
248 * |___Reserved___|
249 * |__Event_data__|
250 * |_____SS_______|
251 * |_____RSP______|
252 * |_____FLAGS____|
253 * |_____CS_______|
254 * |_____IP_______|
255 * 64-byte cache line ==> |__Error_code__| <== ERETU return frame
256 * |______________|
257 * |______________|
258 * |______________|
259 * |______________|
260 * |______________|
261 * |______________|
262 * |______________|
263 * 64-byte cache line ==> |______________| <== RSP after step 1) and 2)
264 * |___Reserved___|
265 * |__Event_data__|
266 * |_____SS_______|
267 * |_____RSP______|
268 * |_____FLAGS____|
269 * |_____CS_______|
270 * |_____IP_______|
271 * 64-byte cache line ==> |__Error_code__| <== ERETS return frame
272 *
273 * Thus a new FRED stack frame will always be pushed below a previous
274 * FRED stack frame ((N*64) bytes may be reserved between), and it is
275 * safe to write to a previous FRED stack frame as they never overlap.
276 */
277 fred_info(uregs)->edata = fred_event_data(regs);
278 uregs->ssx = regs->ssx;
279 uregs->fred_ss.ss = ss;
280 /* The NMI bit was moved away above */
281 uregs->fred_ss.nmi = 0;
282 uregs->csx = regs->csx;
283 uregs->fred_cs.sl = 0;
284 uregs->fred_cs.wfe = 0;
285 uregs->cs = cs;
286 uregs->orig_ax = error_code;
287
288 return ex_handler_default(fixup, regs);
289}
290#endif
291
292int ex_get_fixup_type(unsigned long ip)
293{
294 const struct exception_table_entry *e = search_exception_tables(add: ip);
295
296 return e ? FIELD_GET(EX_DATA_TYPE_MASK, e->data) : EX_TYPE_NONE;
297}
298
299int fixup_exception(struct pt_regs *regs, int trapnr, unsigned long error_code,
300 unsigned long fault_addr)
301{
302 const struct exception_table_entry *e;
303 int type, reg, imm;
304
305#ifdef CONFIG_PNPBIOS
306 if (unlikely(SEGMENT_IS_PNP_CODE(regs->cs))) {
307 extern u32 pnp_bios_fault_eip, pnp_bios_fault_esp;
308 extern u32 pnp_bios_is_utter_crap;
309 pnp_bios_is_utter_crap = 1;
310 printk(KERN_CRIT "PNPBIOS fault.. attempting recovery.\n");
311 __asm__ volatile(
312 "movl %0, %%esp\n\t"
313 "jmp *%1\n\t"
314 : : "g" (pnp_bios_fault_esp), "g" (pnp_bios_fault_eip));
315 panic("do_trap: can't hit this");
316 }
317#endif
318
319 e = search_exception_tables(add: regs->ip);
320 if (!e)
321 return 0;
322
323 type = FIELD_GET(EX_DATA_TYPE_MASK, e->data);
324 reg = FIELD_GET(EX_DATA_REG_MASK, e->data);
325 imm = FIELD_GET(EX_DATA_IMM_MASK, e->data);
326
327 switch (type) {
328 case EX_TYPE_DEFAULT:
329 case EX_TYPE_DEFAULT_MCE_SAFE:
330 return ex_handler_default(e, regs);
331 case EX_TYPE_FAULT:
332 case EX_TYPE_FAULT_MCE_SAFE:
333 return ex_handler_fault(fixup: e, regs, trapnr);
334 case EX_TYPE_UACCESS:
335 return ex_handler_uaccess(fixup: e, regs, trapnr, fault_address: fault_addr);
336 case EX_TYPE_CLEAR_FS:
337 return ex_handler_clear_fs(fixup: e, regs);
338 case EX_TYPE_FPU_RESTORE:
339 return ex_handler_fprestore(fixup: e, regs);
340 case EX_TYPE_BPF:
341 return ex_handler_bpf(x: e, regs);
342 case EX_TYPE_WRMSR:
343 return ex_handler_msr(fixup: e, regs, wrmsr: true, safe: false, reg);
344 case EX_TYPE_RDMSR:
345 return ex_handler_msr(fixup: e, regs, wrmsr: false, safe: false, reg);
346 case EX_TYPE_WRMSR_SAFE:
347 return ex_handler_msr(fixup: e, regs, wrmsr: true, safe: true, reg);
348 case EX_TYPE_RDMSR_SAFE:
349 return ex_handler_msr(fixup: e, regs, wrmsr: false, safe: true, reg);
350 case EX_TYPE_WRMSR_IN_MCE:
351 ex_handler_msr_mce(regs, wrmsr: true);
352 break;
353 case EX_TYPE_RDMSR_IN_MCE:
354 ex_handler_msr_mce(regs, wrmsr: false);
355 break;
356 case EX_TYPE_POP_REG:
357 regs->sp += sizeof(long);
358 fallthrough;
359 case EX_TYPE_IMM_REG:
360 return ex_handler_imm_reg(fixup: e, regs, reg, imm);
361 case EX_TYPE_FAULT_SGX:
362 return ex_handler_sgx(fixup: e, regs, trapnr);
363 case EX_TYPE_UCOPY_LEN:
364 return ex_handler_ucopy_len(fixup: e, regs, trapnr, fault_address: fault_addr, reg, imm);
365 case EX_TYPE_ZEROPAD:
366 return ex_handler_zeropad(e, regs, fault_addr);
367#ifdef CONFIG_X86_FRED
368 case EX_TYPE_ERETU:
369 return ex_handler_eretu(e, regs, error_code);
370#endif
371 }
372 BUG();
373}
374
375extern unsigned int early_recursion_flag;
376
377/* Restricted version used during very early boot */
378void __init early_fixup_exception(struct pt_regs *regs, int trapnr)
379{
380 /* Ignore early NMIs. */
381 if (trapnr == X86_TRAP_NMI)
382 return;
383
384 if (early_recursion_flag > 2)
385 goto halt_loop;
386
387 /*
388 * Old CPUs leave the high bits of CS on the stack
389 * undefined. I'm not sure which CPUs do this, but at least
390 * the 486 DX works this way.
391 * Xen pv domains are not using the default __KERNEL_CS.
392 */
393 if (!xen_pv_domain() && regs->cs != __KERNEL_CS)
394 goto fail;
395
396 /*
397 * The full exception fixup machinery is available as soon as
398 * the early IDT is loaded. This means that it is the
399 * responsibility of extable users to either function correctly
400 * when handlers are invoked early or to simply avoid causing
401 * exceptions before they're ready to handle them.
402 *
403 * This is better than filtering which handlers can be used,
404 * because refusing to call a handler here is guaranteed to
405 * result in a hard-to-debug panic.
406 *
407 * Keep in mind that not all vectors actually get here. Early
408 * page faults, for example, are special.
409 */
410 if (fixup_exception(regs, trapnr, error_code: regs->orig_ax, fault_addr: 0))
411 return;
412
413 if (trapnr == X86_TRAP_UD) {
414 if (report_bug(bug_addr: regs->ip, regs) == BUG_TRAP_TYPE_WARN) {
415 /* Skip the ud2. */
416 regs->ip += LEN_UD2;
417 return;
418 }
419
420 /*
421 * If this was a BUG and report_bug returns or if this
422 * was just a normal #UD, we want to continue onward and
423 * crash.
424 */
425 }
426
427fail:
428 early_printk(fmt: "PANIC: early exception 0x%02x IP %lx:%lx error %lx cr2 0x%lx\n",
429 (unsigned)trapnr, (unsigned long)regs->cs, regs->ip,
430 regs->orig_ax, read_cr2());
431
432 show_regs(regs);
433
434halt_loop:
435 while (true)
436 halt();
437}
438