1 /* SPDX-License-Identifier: GPL-2.0 */
3 #include <linux/stringify.h>
4 #include <linux/linkage.h>
5 #include <asm/dwarf2.h>
6 #include <asm/cpufeatures.h>
7 #include <asm/alternative.h>
8 #include <asm/asm-offsets.h>
9 #include <asm/export.h>
10 #include <asm/nospec-branch.h>
11 #include <asm/unwind_hints.h>
12 #include <asm/percpu.h>
13 #include <asm/frame.h>
16 .section .text.__x86.indirect_thunk
20 ANNOTATE_INTRA_FUNCTION_CALL
35 .align RETPOLINE_THUNK_SIZE
36 SYM_INNER_LABEL(__x86_indirect_thunk_\reg, SYM_L_GLOBAL)
40 ALTERNATIVE_2 __stringify(RETPOLINE \reg), \
41 __stringify(lfence; ANNOTATE_RETPOLINE_SAFE; jmp *%\reg; int3), X86_FEATURE_RETPOLINE_LFENCE, \
42 __stringify(ANNOTATE_RETPOLINE_SAFE; jmp *%\reg), ALT_NOT(X86_FEATURE_RETPOLINE)
47 * Despite being an assembler file we can't just use .irp here
48 * because __KSYM_DEPS__ only uses the C preprocessor and would
49 * only see one instance of "__x86_indirect_thunk_\reg" rather
50 * than one per register with the correct names. So we do it
51 * the simple and nasty way...
53 * Worse, you can only have a single EXPORT_SYMBOL per line,
54 * and CPP can't insert newlines, so we have to repeat everything
58 #define __EXPORT_THUNK(sym) _ASM_NOKPROBE(sym); EXPORT_SYMBOL(sym)
60 .align RETPOLINE_THUNK_SIZE
61 SYM_CODE_START(__x86_indirect_thunk_array)
63 #define GEN(reg) THUNK reg
64 #include <asm/GEN-for-each-reg.h>
67 .align RETPOLINE_THUNK_SIZE
68 SYM_CODE_END(__x86_indirect_thunk_array)
70 #define GEN(reg) __EXPORT_THUNK(__x86_indirect_thunk_ ## reg)
71 #include <asm/GEN-for-each-reg.h>
74 #ifdef CONFIG_CALL_DEPTH_TRACKING
76 .align RETPOLINE_THUNK_SIZE
78 SYM_INNER_LABEL(__x86_indirect_call_thunk_\reg, SYM_L_GLOBAL)
89 .align RETPOLINE_THUNK_SIZE
90 SYM_CODE_START(__x86_indirect_call_thunk_array)
92 #define GEN(reg) CALL_THUNK reg
93 #include <asm/GEN-for-each-reg.h>
96 .align RETPOLINE_THUNK_SIZE
97 SYM_CODE_END(__x86_indirect_call_thunk_array)
99 #define GEN(reg) __EXPORT_THUNK(__x86_indirect_call_thunk_ ## reg)
100 #include <asm/GEN-for-each-reg.h>
103 .macro JUMP_THUNK reg
104 .align RETPOLINE_THUNK_SIZE
106 SYM_INNER_LABEL(__x86_indirect_jump_thunk_\reg, SYM_L_GLOBAL)
107 UNWIND_HINT_UNDEFINED
115 .align RETPOLINE_THUNK_SIZE
116 SYM_CODE_START(__x86_indirect_jump_thunk_array)
118 #define GEN(reg) JUMP_THUNK reg
119 #include <asm/GEN-for-each-reg.h>
122 .align RETPOLINE_THUNK_SIZE
123 SYM_CODE_END(__x86_indirect_jump_thunk_array)
125 #define GEN(reg) __EXPORT_THUNK(__x86_indirect_jump_thunk_ ## reg)
126 #include <asm/GEN-for-each-reg.h>
130 * This function name is magical and is used by -mfunction-return=thunk-extern
131 * for the compiler to generate JMPs to it.
133 #ifdef CONFIG_RETHUNK
136 * srso_untrain_ret_alias() and srso_safe_ret_alias() are placed at
139 * - srso_untrain_ret_alias() is 2M aligned
140 * - srso_safe_ret_alias() is also in the same 2M page but bits 2, 8, 14
141 * and 20 in its virtual address are set (while those bits in the
142 * srso_untrain_ret_alias() function are cleared).
144 * This guarantees that those two addresses will alias in the branch
145 * target buffer of Zen3/4 generations, leading to any potential
146 * poisoned entries at that BTB slot to get evicted.
148 * As a result, srso_safe_ret_alias() becomes a safe return.
150 #ifdef CONFIG_CPU_SRSO
151 .section .text.__x86.rethunk_untrain
153 SYM_START(srso_untrain_ret_alias, SYM_L_GLOBAL, SYM_A_NONE)
156 jmp __x86_return_thunk
157 SYM_FUNC_END(srso_untrain_ret_alias)
158 __EXPORT_THUNK(srso_untrain_ret_alias)
160 .section .text.__x86.rethunk_safe
163 /* Needs a definition for the __x86_return_thunk alternative below. */
164 SYM_START(srso_safe_ret_alias, SYM_L_GLOBAL, SYM_A_NONE)
165 #ifdef CONFIG_CPU_SRSO
172 SYM_FUNC_END(srso_safe_ret_alias)
174 .section .text.__x86.return_thunk
177 * Safety details here pertain to the AMD Zen{1,2} microarchitecture:
178 * 1) The RET at __x86_return_thunk must be on a 64 byte boundary, for
179 * alignment within the BTB.
180 * 2) The instruction at zen_untrain_ret must contain, and not
181 * end with, the 0xc3 byte of the RET.
182 * 3) STIBP must be enabled, or SMT disabled, to prevent the sibling thread
183 * from re-poisioning the BTB prediction.
186 .skip 64 - (__ret - zen_untrain_ret), 0xcc
187 SYM_START(zen_untrain_ret, SYM_L_GLOBAL, SYM_A_NONE)
190 * As executed from zen_untrain_ret, this is:
194 * JMP __x86_return_thunk
196 * Executing the TEST instruction has a side effect of evicting any BTB
197 * prediction (potentially attacker controlled) attached to the RET, as
198 * __x86_return_thunk + 1 isn't an instruction boundary at the moment.
203 * As executed from __x86_return_thunk, this is a plain RET.
205 * As part of the TEST above, RET is the ModRM byte, and INT3 the imm8.
207 * We subsequently jump backwards and architecturally execute the RET.
208 * This creates a correct BTB prediction (type=ret), but in the
209 * meantime we suffer Straight Line Speculation (because the type was
210 * no branch) which is halted by the INT3.
212 * With SMT enabled and STIBP active, a sibling thread cannot poison
213 * RET's prediction to a type of its choice, but can evict the
214 * prediction due to competitive sharing. If the prediction is
215 * evicted, __x86_return_thunk will suffer Straight Line Speculation
216 * which will be contained safely by the INT3.
218 SYM_INNER_LABEL(__ret, SYM_L_GLOBAL)
224 * Ensure the TEST decoding / BTB invalidation is complete.
229 * Jump back and execute the RET in the middle of the TEST instruction.
230 * INT3 is for SLS protection.
234 SYM_FUNC_END(zen_untrain_ret)
235 __EXPORT_THUNK(zen_untrain_ret)
238 * SRSO untraining sequence for Zen1/2, similar to zen_untrain_ret()
239 * above. On kernel entry, srso_untrain_ret() is executed which is a
241 * movabs $0xccccccc308c48348,%rax
243 * and when the return thunk executes the inner label srso_safe_ret()
244 * later, it is a stack manipulation and a RET which is mispredicted and
245 * thus a "safe" one to use.
248 .skip 64 - (srso_safe_ret - srso_untrain_ret), 0xcc
249 SYM_START(srso_untrain_ret, SYM_L_GLOBAL, SYM_A_NONE)
253 SYM_INNER_LABEL(srso_safe_ret, SYM_L_GLOBAL)
262 SYM_CODE_END(srso_safe_ret)
263 SYM_FUNC_END(srso_untrain_ret)
264 __EXPORT_THUNK(srso_untrain_ret)
266 SYM_FUNC_START(__x86_return_thunk)
267 ALTERNATIVE_2 "jmp __ret", "call srso_safe_ret", X86_FEATURE_SRSO, \
268 "call srso_safe_ret_alias", X86_FEATURE_SRSO_ALIAS
270 SYM_CODE_END(__x86_return_thunk)
271 EXPORT_SYMBOL(__x86_return_thunk)
273 #endif /* CONFIG_RETHUNK */
275 #ifdef CONFIG_CALL_DEPTH_TRACKING
278 SYM_FUNC_START(__x86_return_skl)
281 * Keep the hotpath in a 16byte I-fetch for the non-debug
284 CALL_THUNKS_DEBUG_INC_RETS
285 shlq $5, PER_CPU_VAR(pcpu_hot + X86_call_depth)
291 CALL_THUNKS_DEBUG_INC_STUFFS
293 ANNOTATE_INTRA_FUNCTION_CALL
305 SYM_FUNC_END(__x86_return_skl)
307 #endif /* CONFIG_CALL_DEPTH_TRACKING */