1 /* SPDX-License-Identifier: GPL-2.0 */
2 
3 #ifndef _ASM_X86_NOSPEC_BRANCH_H_
4 #define _ASM_X86_NOSPEC_BRANCH_H_
5 
6 #include <linux/static_key.h>
7 #include <linux/objtool.h>
8 #include <linux/linkage.h>
9 
10 #include <asm/alternative.h>
11 #include <asm/cpufeatures.h>
12 #include <asm/msr-index.h>
13 #include <asm/unwind_hints.h>
14 #include <asm/percpu.h>
15 
16 #define RETPOLINE_THUNK_SIZE	32
17 
18 /*
19  * Fill the CPU return stack buffer.
20  *
21  * Each entry in the RSB, if used for a speculative 'ret', contains an
22  * infinite 'pause; lfence; jmp' loop to capture speculative execution.
23  *
24  * This is required in various cases for retpoline and IBRS-based
25  * mitigations for the Spectre variant 2 vulnerability. Sometimes to
26  * eliminate potentially bogus entries from the RSB, and sometimes
27  * purely to ensure that it doesn't get empty, which on some CPUs would
28  * allow predictions from other (unwanted!) sources to be used.
29  *
30  * We define a CPP macro such that it can be used from both .S files and
31  * inline assembly. It's possible to do a .macro and then include that
32  * from C via asm(".include <asm/nospec-branch.h>") but let's not go there.
33  */
34 
35 #define RSB_CLEAR_LOOPS		32	/* To forcibly overwrite all entries */
36 
37 /*
38  * Common helper for __FILL_RETURN_BUFFER and __FILL_ONE_RETURN.
39  */
40 #define __FILL_RETURN_SLOT			\
41 	ANNOTATE_INTRA_FUNCTION_CALL;		\
42 	call	772f;				\
43 	int3;					\
44 772:
45 
46 /*
47  * Stuff the entire RSB.
48  *
49  * Google experimented with loop-unrolling and this turned out to be
50  * the optimal version — two calls, each with their own speculation
51  * trap should their return address end up getting used, in a loop.
52  */
53 #ifdef CONFIG_X86_64
54 #define __FILL_RETURN_BUFFER(reg, nr)			\
55 	mov	$(nr/2), reg;				\
56 771:							\
57 	__FILL_RETURN_SLOT				\
58 	__FILL_RETURN_SLOT				\
59 	add	$(BITS_PER_LONG/8) * 2, %_ASM_SP;	\
60 	dec	reg;					\
61 	jnz	771b;					\
62 	/* barrier for jnz misprediction */		\
63 	lfence;
64 #else
65 /*
66  * i386 doesn't unconditionally have LFENCE, as such it can't
67  * do a loop.
68  */
69 #define __FILL_RETURN_BUFFER(reg, nr)			\
70 	.rept nr;					\
71 	__FILL_RETURN_SLOT;				\
72 	.endr;						\
73 	add	$(BITS_PER_LONG/8) * nr, %_ASM_SP;
74 #endif
75 
76 /*
77  * Stuff a single RSB slot.
78  *
79  * To mitigate Post-Barrier RSB speculation, one CALL instruction must be
80  * forced to retire before letting a RET instruction execute.
81  *
82  * On PBRSB-vulnerable CPUs, it is not safe for a RET to be executed
83  * before this point.
84  */
85 #define __FILL_ONE_RETURN				\
86 	__FILL_RETURN_SLOT				\
87 	add	$(BITS_PER_LONG/8), %_ASM_SP;		\
88 	lfence;
89 
90 #ifdef __ASSEMBLY__
91 
92 /*
93  * This should be used immediately before an indirect jump/call. It tells
94  * objtool the subsequent indirect jump/call is vouched safe for retpoline
95  * builds.
96  */
97 .macro ANNOTATE_RETPOLINE_SAFE
98 	.Lannotate_\@:
99 	.pushsection .discard.retpoline_safe
100 	_ASM_PTR .Lannotate_\@
101 	.popsection
102 .endm
103 
104 /*
105  * (ab)use RETPOLINE_SAFE on RET to annotate away 'bare' RET instructions
106  * vs RETBleed validation.
107  */
108 #define ANNOTATE_UNRET_SAFE ANNOTATE_RETPOLINE_SAFE
109 
110 /*
111  * Abuse ANNOTATE_RETPOLINE_SAFE on a NOP to indicate UNRET_END, should
112  * eventually turn into it's own annotation.
113  */
114 .macro ANNOTATE_UNRET_END
115 #if (defined(CONFIG_CPU_UNRET_ENTRY) || defined(CONFIG_CPU_SRSO))
116 	ANNOTATE_RETPOLINE_SAFE
117 	nop
118 #endif
119 .endm
120 
121 /*
122  * JMP_NOSPEC and CALL_NOSPEC macros can be used instead of a simple
123  * indirect jmp/call which may be susceptible to the Spectre variant 2
124  * attack.
125  */
126 .macro JMP_NOSPEC reg:req
127 #ifdef CONFIG_RETPOLINE
128 	ALTERNATIVE_2 __stringify(ANNOTATE_RETPOLINE_SAFE; jmp *%\reg), \
129 		      __stringify(jmp __x86_indirect_thunk_\reg), X86_FEATURE_RETPOLINE, \
130 		      __stringify(lfence; ANNOTATE_RETPOLINE_SAFE; jmp *%\reg), X86_FEATURE_RETPOLINE_LFENCE
131 #else
132 	jmp	*%\reg
133 #endif
134 .endm
135 
136 .macro CALL_NOSPEC reg:req
137 #ifdef CONFIG_RETPOLINE
138 	ALTERNATIVE_2 __stringify(ANNOTATE_RETPOLINE_SAFE; call *%\reg), \
139 		      __stringify(call __x86_indirect_thunk_\reg), X86_FEATURE_RETPOLINE, \
140 		      __stringify(lfence; ANNOTATE_RETPOLINE_SAFE; call *%\reg), X86_FEATURE_RETPOLINE_LFENCE
141 #else
142 	call	*%\reg
143 #endif
144 .endm
145 
146  /*
147   * A simpler FILL_RETURN_BUFFER macro. Don't make people use the CPP
148   * monstrosity above, manually.
149   */
150 .macro FILL_RETURN_BUFFER reg:req nr:req ftr:req ftr2=ALT_NOT(X86_FEATURE_ALWAYS)
151 	ALTERNATIVE_2 "jmp .Lskip_rsb_\@", \
152 		__stringify(__FILL_RETURN_BUFFER(\reg,\nr)), \ftr, \
153 		__stringify(__FILL_ONE_RETURN), \ftr2
154 
155 .Lskip_rsb_\@:
156 .endm
157 
158 #ifdef CONFIG_CPU_UNRET_ENTRY
159 #define CALL_UNTRAIN_RET	"call entry_untrain_ret"
160 #else
161 #define CALL_UNTRAIN_RET	""
162 #endif
163 
164 /*
165  * Mitigate RETBleed for AMD/Hygon Zen uarch. Requires KERNEL CR3 because the
166  * return thunk isn't mapped into the userspace tables (then again, AMD
167  * typically has NO_MELTDOWN).
168  *
169  * While retbleed_untrain_ret() doesn't clobber anything but requires stack,
170  * entry_ibpb() will clobber AX, CX, DX.
171  *
172  * As such, this must be placed after every *SWITCH_TO_KERNEL_CR3 at a point
173  * where we have a stack but before any RET instruction.
174  */
175 .macro UNTRAIN_RET
176 #if defined(CONFIG_CPU_UNRET_ENTRY) || defined(CONFIG_CPU_IBPB_ENTRY) || \
177 	defined(CONFIG_CPU_SRSO)
178 	ANNOTATE_UNRET_END
179 	ALTERNATIVE_2 "",						\
180 		      CALL_UNTRAIN_RET, X86_FEATURE_UNRET,		\
181 		      "call entry_ibpb", X86_FEATURE_ENTRY_IBPB
182 #endif
183 .endm
184 
185 /*
186  * Macro to execute VERW instruction that mitigate transient data sampling
187  * attacks such as MDS. On affected systems a microcode update overloaded VERW
188  * instruction to also clear the CPU buffers. VERW clobbers CFLAGS.ZF.
189  *
190  * Note: Only the memory operand variant of VERW clears the CPU buffers.
191  */
192 .macro CLEAR_CPU_BUFFERS
193 	ALTERNATIVE "jmp .Lskip_verw_\@", "", X86_FEATURE_CLEAR_CPU_BUF
194 	verw _ASM_RIP(mds_verw_sel)
195 .Lskip_verw_\@:
196 .endm
197 
198 #else /* __ASSEMBLY__ */
199 
200 #define ANNOTATE_RETPOLINE_SAFE					\
201 	"999:\n\t"						\
202 	".pushsection .discard.retpoline_safe\n\t"		\
203 	_ASM_PTR " 999b\n\t"					\
204 	".popsection\n\t"
205 
206 #ifdef CONFIG_RETHUNK
207 extern void __x86_return_thunk(void);
208 #else
209 static inline void __x86_return_thunk(void) {}
210 #endif
211 
212 extern void retbleed_return_thunk(void);
213 extern void srso_return_thunk(void);
214 extern void srso_alias_return_thunk(void);
215 
216 extern void retbleed_untrain_ret(void);
217 extern void srso_untrain_ret(void);
218 extern void srso_alias_untrain_ret(void);
219 
220 extern void entry_untrain_ret(void);
221 extern void entry_ibpb(void);
222 
223 #ifdef CONFIG_RETPOLINE
224 
225 typedef u8 retpoline_thunk_t[RETPOLINE_THUNK_SIZE];
226 
227 #define GEN(reg) \
228 	extern retpoline_thunk_t __x86_indirect_thunk_ ## reg;
229 #include <asm/GEN-for-each-reg.h>
230 #undef GEN
231 
232 extern retpoline_thunk_t __x86_indirect_thunk_array[];
233 
234 #ifdef CONFIG_X86_64
235 
236 /*
237  * Inline asm uses the %V modifier which is only in newer GCC
238  * which is ensured when CONFIG_RETPOLINE is defined.
239  */
240 # define CALL_NOSPEC						\
241 	ALTERNATIVE_2(						\
242 	ANNOTATE_RETPOLINE_SAFE					\
243 	"call *%[thunk_target]\n",				\
244 	"call __x86_indirect_thunk_%V[thunk_target]\n",		\
245 	X86_FEATURE_RETPOLINE,					\
246 	"lfence;\n"						\
247 	ANNOTATE_RETPOLINE_SAFE					\
248 	"call *%[thunk_target]\n",				\
249 	X86_FEATURE_RETPOLINE_LFENCE)
250 
251 # define THUNK_TARGET(addr) [thunk_target] "r" (addr)
252 
253 #else /* CONFIG_X86_32 */
254 /*
255  * For i386 we use the original ret-equivalent retpoline, because
256  * otherwise we'll run out of registers. We don't care about CET
257  * here, anyway.
258  */
259 # define CALL_NOSPEC						\
260 	ALTERNATIVE_2(						\
261 	ANNOTATE_RETPOLINE_SAFE					\
262 	"call *%[thunk_target]\n",				\
263 	"       jmp    904f;\n"					\
264 	"       .align 16\n"					\
265 	"901:	call   903f;\n"					\
266 	"902:	pause;\n"					\
267 	"    	lfence;\n"					\
268 	"       jmp    902b;\n"					\
269 	"       .align 16\n"					\
270 	"903:	lea    4(%%esp), %%esp;\n"			\
271 	"       pushl  %[thunk_target];\n"			\
272 	"       ret;\n"						\
273 	"       .align 16\n"					\
274 	"904:	call   901b;\n",				\
275 	X86_FEATURE_RETPOLINE,					\
276 	"lfence;\n"						\
277 	ANNOTATE_RETPOLINE_SAFE					\
278 	"call *%[thunk_target]\n",				\
279 	X86_FEATURE_RETPOLINE_LFENCE)
280 
281 # define THUNK_TARGET(addr) [thunk_target] "rm" (addr)
282 #endif
283 #else /* No retpoline for C / inline asm */
284 # define CALL_NOSPEC "call *%[thunk_target]\n"
285 # define THUNK_TARGET(addr) [thunk_target] "rm" (addr)
286 #endif
287 
288 /* The Spectre V2 mitigation variants */
289 enum spectre_v2_mitigation {
290 	SPECTRE_V2_NONE,
291 	SPECTRE_V2_RETPOLINE,
292 	SPECTRE_V2_LFENCE,
293 	SPECTRE_V2_EIBRS,
294 	SPECTRE_V2_EIBRS_RETPOLINE,
295 	SPECTRE_V2_EIBRS_LFENCE,
296 	SPECTRE_V2_IBRS,
297 };
298 
299 /* The indirect branch speculation control variants */
300 enum spectre_v2_user_mitigation {
301 	SPECTRE_V2_USER_NONE,
302 	SPECTRE_V2_USER_STRICT,
303 	SPECTRE_V2_USER_STRICT_PREFERRED,
304 	SPECTRE_V2_USER_PRCTL,
305 	SPECTRE_V2_USER_SECCOMP,
306 };
307 
308 /* The Speculative Store Bypass disable variants */
309 enum ssb_mitigation {
310 	SPEC_STORE_BYPASS_NONE,
311 	SPEC_STORE_BYPASS_DISABLE,
312 	SPEC_STORE_BYPASS_PRCTL,
313 	SPEC_STORE_BYPASS_SECCOMP,
314 };
315 
316 extern char __indirect_thunk_start[];
317 extern char __indirect_thunk_end[];
318 
319 static __always_inline
320 void alternative_msr_write(unsigned int msr, u64 val, unsigned int feature)
321 {
322 	asm volatile(ALTERNATIVE("", "wrmsr", %c[feature])
323 		: : "c" (msr),
324 		    "a" ((u32)val),
325 		    "d" ((u32)(val >> 32)),
326 		    [feature] "i" (feature)
327 		: "memory");
328 }
329 
330 extern u64 x86_pred_cmd;
331 
332 static inline void indirect_branch_prediction_barrier(void)
333 {
334 	alternative_msr_write(MSR_IA32_PRED_CMD, x86_pred_cmd, X86_FEATURE_USE_IBPB);
335 }
336 
337 /* The Intel SPEC CTRL MSR base value cache */
338 extern u64 x86_spec_ctrl_base;
339 DECLARE_PER_CPU(u64, x86_spec_ctrl_current);
340 extern void update_spec_ctrl_cond(u64 val);
341 extern u64 spec_ctrl_current(void);
342 
343 /*
344  * With retpoline, we must use IBRS to restrict branch prediction
345  * before calling into firmware.
346  *
347  * (Implemented as CPP macros due to header hell.)
348  */
349 #define firmware_restrict_branch_speculation_start()			\
350 do {									\
351 	preempt_disable();						\
352 	alternative_msr_write(MSR_IA32_SPEC_CTRL,			\
353 			      spec_ctrl_current() | SPEC_CTRL_IBRS,	\
354 			      X86_FEATURE_USE_IBRS_FW);			\
355 	alternative_msr_write(MSR_IA32_PRED_CMD, PRED_CMD_IBPB,		\
356 			      X86_FEATURE_USE_IBPB_FW);			\
357 } while (0)
358 
359 #define firmware_restrict_branch_speculation_end()			\
360 do {									\
361 	alternative_msr_write(MSR_IA32_SPEC_CTRL,			\
362 			      spec_ctrl_current(),			\
363 			      X86_FEATURE_USE_IBRS_FW);			\
364 	preempt_enable();						\
365 } while (0)
366 
367 DECLARE_STATIC_KEY_FALSE(switch_to_cond_stibp);
368 DECLARE_STATIC_KEY_FALSE(switch_mm_cond_ibpb);
369 DECLARE_STATIC_KEY_FALSE(switch_mm_always_ibpb);
370 
371 DECLARE_STATIC_KEY_FALSE(mds_idle_clear);
372 
373 DECLARE_STATIC_KEY_FALSE(mmio_stale_data_clear);
374 
375 extern u16 mds_verw_sel;
376 
377 #include <asm/segment.h>
378 
379 /**
380  * mds_clear_cpu_buffers - Mitigation for MDS and TAA vulnerability
381  *
382  * This uses the otherwise unused and obsolete VERW instruction in
383  * combination with microcode which triggers a CPU buffer flush when the
384  * instruction is executed.
385  */
386 static __always_inline void mds_clear_cpu_buffers(void)
387 {
388 	static const u16 ds = __KERNEL_DS;
389 
390 	/*
391 	 * Has to be the memory-operand variant because only that
392 	 * guarantees the CPU buffer flush functionality according to
393 	 * documentation. The register-operand variant does not.
394 	 * Works with any segment selector, but a valid writable
395 	 * data segment is the fastest variant.
396 	 *
397 	 * "cc" clobber is required because VERW modifies ZF.
398 	 */
399 	asm volatile("verw %[ds]" : : [ds] "m" (ds) : "cc");
400 }
401 
402 /**
403  * mds_idle_clear_cpu_buffers - Mitigation for MDS vulnerability
404  *
405  * Clear CPU buffers if the corresponding static key is enabled
406  */
407 static inline void mds_idle_clear_cpu_buffers(void)
408 {
409 	if (static_branch_likely(&mds_idle_clear))
410 		mds_clear_cpu_buffers();
411 }
412 
413 #endif /* __ASSEMBLY__ */
414 
415 #endif /* _ASM_X86_NOSPEC_BRANCH_H_ */
416