]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - arch/x86/include/asm/nospec-branch.h
x86/speculation, objtool: Annotate indirect calls/jumps for objtool
[mirror_ubuntu-bionic-kernel.git] / arch / x86 / include / asm / nospec-branch.h
CommitLineData
76b04384
DW
1/* SPDX-License-Identifier: GPL-2.0 */
2
c55d51ae
BP
3#ifndef _ASM_X86_NOSPEC_BRANCH_H_
4#define _ASM_X86_NOSPEC_BRANCH_H_
76b04384
DW
5
6#include <asm/alternative.h>
7#include <asm/alternative-asm.h>
8#include <asm/cpufeatures.h>
8effdc23 9#include <asm/msr-index.h>
76b04384 10
10c16289
DW
11/*
12 * Fill the CPU return stack buffer.
13 *
14 * Each entry in the RSB, if used for a speculative 'ret', contains an
15 * infinite 'pause; lfence; jmp' loop to capture speculative execution.
16 *
17 * This is required in various cases for retpoline and IBRS-based
18 * mitigations for the Spectre variant 2 vulnerability. Sometimes to
19 * eliminate potentially bogus entries from the RSB, and sometimes
20 * purely to ensure that it doesn't get empty, which on some CPUs would
21 * allow predictions from other (unwanted!) sources to be used.
22 *
23 * We define a CPP macro such that it can be used from both .S files and
24 * inline assembly. It's possible to do a .macro and then include that
25 * from C via asm(".include <asm/nospec-branch.h>") but let's not go there.
26 */
27
28#define RSB_CLEAR_LOOPS 32 /* To forcibly overwrite all entries */
29#define RSB_FILL_LOOPS 16 /* To avoid underflow */
30
31/*
32 * Google experimented with loop-unrolling and this turned out to be
33 * the optimal version — two calls, each with their own speculation
34 * trap should their return address end up getting used, in a loop.
35 */
36#define __FILL_RETURN_BUFFER(reg, nr, sp) \
37 mov $(nr/2), reg; \
38771: \
39 call 772f; \
40773: /* speculation trap */ \
41 pause; \
42 lfence; \
43 jmp 773b; \
44772: \
45 call 774f; \
46775: /* speculation trap */ \
47 pause; \
48 lfence; \
49 jmp 775b; \
50774: \
51 dec reg; \
52 jnz 771b; \
53 add $(BITS_PER_LONG/8) * nr, sp;
54
76b04384
DW
55#ifdef __ASSEMBLY__
56
57/*
58 * This should be used immediately before a retpoline alternative. It tells
59 * objtool where the retpolines are so that it can make sense of the control
60 * flow by just reading the original instruction(s) and ignoring the
61 * alternatives.
62 */
63.macro ANNOTATE_NOSPEC_ALTERNATIVE
64 .Lannotate_\@:
65 .pushsection .discard.nospec
66 .long .Lannotate_\@ - .
67 .popsection
68.endm
69
b207cde6
PZ
70/*
71 * This should be used immediately before an indirect jump/call. It tells
72 * objtool the subsequent indirect jump/call is vouched safe for retpoline
73 * builds.
74 */
75.macro ANNOTATE_RETPOLINE_SAFE
76 .Lannotate_\@:
77 .pushsection .discard.retpoline_safe
78 _ASM_PTR .Lannotate_\@
79 .popsection
80.endm
81
76b04384
DW
82/*
83 * These are the bare retpoline primitives for indirect jmp and call.
84 * Do not use these directly; they only exist to make the ALTERNATIVE
85 * invocation below less ugly.
86 */
87.macro RETPOLINE_JMP reg:req
88 call .Ldo_rop_\@
89.Lspec_trap_\@:
90 pause
28d437d5 91 lfence
76b04384
DW
92 jmp .Lspec_trap_\@
93.Ldo_rop_\@:
94 mov \reg, (%_ASM_SP)
95 ret
96.endm
97
98/*
99 * This is a wrapper around RETPOLINE_JMP so the called function in reg
100 * returns to the instruction after the macro.
101 */
102.macro RETPOLINE_CALL reg:req
103 jmp .Ldo_call_\@
104.Ldo_retpoline_jmp_\@:
105 RETPOLINE_JMP \reg
106.Ldo_call_\@:
107 call .Ldo_retpoline_jmp_\@
108.endm
109
110/*
111 * JMP_NOSPEC and CALL_NOSPEC macros can be used instead of a simple
112 * indirect jmp/call which may be susceptible to the Spectre variant 2
113 * attack.
114 */
115.macro JMP_NOSPEC reg:req
116#ifdef CONFIG_RETPOLINE
117 ANNOTATE_NOSPEC_ALTERNATIVE
b207cde6 118 ALTERNATIVE_2 __stringify(ANNOTATE_RETPOLINE_SAFE; jmp *\reg), \
76b04384 119 __stringify(RETPOLINE_JMP \reg), X86_FEATURE_RETPOLINE, \
b207cde6 120 __stringify(lfence; ANNOTATE_RETPOLINE_SAFE; jmp *\reg), X86_FEATURE_RETPOLINE_AMD
76b04384
DW
121#else
122 jmp *\reg
123#endif
124.endm
125
126.macro CALL_NOSPEC reg:req
127#ifdef CONFIG_RETPOLINE
128 ANNOTATE_NOSPEC_ALTERNATIVE
b207cde6 129 ALTERNATIVE_2 __stringify(ANNOTATE_RETPOLINE_SAFE; call *\reg), \
76b04384 130 __stringify(RETPOLINE_CALL \reg), X86_FEATURE_RETPOLINE,\
b207cde6 131 __stringify(lfence; ANNOTATE_RETPOLINE_SAFE; call *\reg), X86_FEATURE_RETPOLINE_AMD
76b04384
DW
132#else
133 call *\reg
134#endif
117cc7a9
DW
135.endm
136
10c16289
DW
137 /*
138 * A simpler FILL_RETURN_BUFFER macro. Don't make people use the CPP
139 * monstrosity above, manually.
140 */
141.macro FILL_RETURN_BUFFER reg:req nr:req ftr:req
117cc7a9 142#ifdef CONFIG_RETPOLINE
10c16289
DW
143 ANNOTATE_NOSPEC_ALTERNATIVE
144 ALTERNATIVE "jmp .Lskip_rsb_\@", \
145 __stringify(__FILL_RETURN_BUFFER(\reg,\nr,%_ASM_SP)) \
146 \ftr
147.Lskip_rsb_\@:
117cc7a9 148#endif
76b04384
DW
149.endm
150
151#else /* __ASSEMBLY__ */
152
153#define ANNOTATE_NOSPEC_ALTERNATIVE \
154 "999:\n\t" \
155 ".pushsection .discard.nospec\n\t" \
156 ".long 999b - .\n\t" \
157 ".popsection\n\t"
158
b207cde6
PZ
159#define ANNOTATE_RETPOLINE_SAFE \
160 "999:\n\t" \
161 ".pushsection .discard.retpoline_safe\n\t" \
162 _ASM_PTR " 999b\n\t" \
163 ".popsection\n\t"
164
76b04384
DW
165#if defined(CONFIG_X86_64) && defined(RETPOLINE)
166
167/*
168 * Since the inline asm uses the %V modifier which is only in newer GCC,
169 * the 64-bit one is dependent on RETPOLINE not CONFIG_RETPOLINE.
170 */
171# define CALL_NOSPEC \
172 ANNOTATE_NOSPEC_ALTERNATIVE \
173 ALTERNATIVE( \
b207cde6 174 ANNOTATE_RETPOLINE_SAFE \
76b04384
DW
175 "call *%[thunk_target]\n", \
176 "call __x86_indirect_thunk_%V[thunk_target]\n", \
177 X86_FEATURE_RETPOLINE)
178# define THUNK_TARGET(addr) [thunk_target] "r" (addr)
179
180#elif defined(CONFIG_X86_32) && defined(CONFIG_RETPOLINE)
181/*
182 * For i386 we use the original ret-equivalent retpoline, because
183 * otherwise we'll run out of registers. We don't care about CET
184 * here, anyway.
185 */
186# define CALL_NOSPEC ALTERNATIVE("call *%[thunk_target]\n", \
187 " jmp 904f;\n" \
188 " .align 16\n" \
189 "901: call 903f;\n" \
190 "902: pause;\n" \
28d437d5 191 " lfence;\n" \
76b04384
DW
192 " jmp 902b;\n" \
193 " .align 16\n" \
194 "903: addl $4, %%esp;\n" \
195 " pushl %[thunk_target];\n" \
196 " ret;\n" \
197 " .align 16\n" \
198 "904: call 901b;\n", \
199 X86_FEATURE_RETPOLINE)
200
201# define THUNK_TARGET(addr) [thunk_target] "rm" (addr)
117cc7a9 202#else /* No retpoline for C / inline asm */
76b04384
DW
203# define CALL_NOSPEC "call *%[thunk_target]\n"
204# define THUNK_TARGET(addr) [thunk_target] "rm" (addr)
205#endif
206
da285121
DW
207/* The Spectre V2 mitigation variants */
208enum spectre_v2_mitigation {
209 SPECTRE_V2_NONE,
210 SPECTRE_V2_RETPOLINE_MINIMAL,
211 SPECTRE_V2_RETPOLINE_MINIMAL_AMD,
212 SPECTRE_V2_RETPOLINE_GENERIC,
213 SPECTRE_V2_RETPOLINE_AMD,
214 SPECTRE_V2_IBRS,
215};
216
736e80a4
MH
217extern char __indirect_thunk_start[];
218extern char __indirect_thunk_end[];
219
117cc7a9
DW
220/*
221 * On VMEXIT we must ensure that no RSB predictions learned in the guest
222 * can be followed in the host, by overwriting the RSB completely. Both
223 * retpoline and IBRS mitigations for Spectre v2 need this; only on future
894655b1 224 * CPUs with IBRS_ALL *might* it be avoided.
117cc7a9
DW
225 */
226static inline void vmexit_fill_RSB(void)
227{
228#ifdef CONFIG_RETPOLINE
10c16289
DW
229 unsigned long loops;
230
231 asm volatile (ANNOTATE_NOSPEC_ALTERNATIVE
232 ALTERNATIVE("jmp 910f",
233 __stringify(__FILL_RETURN_BUFFER(%0, RSB_CLEAR_LOOPS, %1)),
234 X86_FEATURE_RETPOLINE)
235 "910:"
236 : "=r" (loops), ASM_CALL_CONSTRAINT
237 : : "memory" );
117cc7a9
DW
238#endif
239}
3f7d8755 240
390b99c3
DW
241#define alternative_msr_write(_msr, _val, _feature) \
242 asm volatile(ALTERNATIVE("", \
243 "movl %[msr], %%ecx\n\t" \
244 "movl %[val], %%eax\n\t" \
245 "movl $0, %%edx\n\t" \
246 "wrmsr", \
247 _feature) \
248 : : [msr] "i" (_msr), [val] "i" (_val) \
249 : "eax", "ecx", "edx", "memory")
250
bd12e896
DW
251static inline void indirect_branch_prediction_barrier(void)
252{
390b99c3
DW
253 alternative_msr_write(MSR_IA32_PRED_CMD, PRED_CMD_IBPB,
254 X86_FEATURE_USE_IBPB);
255}
256
257/*
258 * With retpoline, we must use IBRS to restrict branch prediction
259 * before calling into firmware.
260 */
261static inline void firmware_restrict_branch_speculation_start(void)
262{
263 preempt_disable();
264 alternative_msr_write(MSR_IA32_SPEC_CTRL, SPEC_CTRL_IBRS,
265 X86_FEATURE_USE_IBRS_FW);
266}
267
268static inline void firmware_restrict_branch_speculation_end(void)
269{
270 alternative_msr_write(MSR_IA32_SPEC_CTRL, 0,
271 X86_FEATURE_USE_IBRS_FW);
272 preempt_enable();
bd12e896
DW
273}
274
76b04384 275#endif /* __ASSEMBLY__ */
29fed5fa
DB
276
277/*
278 * Below is used in the eBPF JIT compiler and emits the byte sequence
279 * for the following assembly:
280 *
281 * With retpolines configured:
282 *
283 * callq do_rop
284 * spec_trap:
285 * pause
286 * lfence
287 * jmp spec_trap
288 * do_rop:
289 * mov %rax,(%rsp)
290 * retq
291 *
292 * Without retpolines configured:
293 *
294 * jmp *%rax
295 */
296#ifdef CONFIG_RETPOLINE
297# define RETPOLINE_RAX_BPF_JIT_SIZE 17
298# define RETPOLINE_RAX_BPF_JIT() \
299 EMIT1_off32(0xE8, 7); /* callq do_rop */ \
300 /* spec_trap: */ \
301 EMIT2(0xF3, 0x90); /* pause */ \
302 EMIT3(0x0F, 0xAE, 0xE8); /* lfence */ \
303 EMIT2(0xEB, 0xF9); /* jmp spec_trap */ \
304 /* do_rop: */ \
305 EMIT4(0x48, 0x89, 0x04, 0x24); /* mov %rax,(%rsp) */ \
306 EMIT1(0xC3); /* retq */
307#else
308# define RETPOLINE_RAX_BPF_JIT_SIZE 2
309# define RETPOLINE_RAX_BPF_JIT() \
310 EMIT2(0xFF, 0xE0); /* jmp *%rax */
311#endif
312
c55d51ae 313#endif /* _ASM_X86_NOSPEC_BRANCH_H_ */