xref: /linux/arch/x86/include/asm/alternative.h (revision 4359a011e259a4608afc7fb3635370c9d4ba5943)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_ALTERNATIVE_H
3 #define _ASM_X86_ALTERNATIVE_H
4 
5 #include <linux/types.h>
6 #include <linux/stringify.h>
7 #include <asm/asm.h>
8 
9 #define ALTINSTR_FLAG_INV	(1 << 15)
10 #define ALT_NOT(feat)		((feat) | ALTINSTR_FLAG_INV)
11 
12 #ifndef __ASSEMBLY__
13 
14 #include <linux/stddef.h>
15 
16 /*
17  * Alternative inline assembly for SMP.
18  *
19  * The LOCK_PREFIX macro defined here replaces the LOCK and
20  * LOCK_PREFIX macros used everywhere in the source tree.
21  *
22  * SMP alternatives use the same data structures as the other
23  * alternatives and the X86_FEATURE_UP flag to indicate the case of a
24  * UP system running a SMP kernel.  The existing apply_alternatives()
25  * works fine for patching a SMP kernel for UP.
26  *
27  * The SMP alternative tables can be kept after boot and contain both
28  * UP and SMP versions of the instructions to allow switching back to
29  * SMP at runtime, when hotplugging in a new CPU, which is especially
30  * useful in virtualized environments.
31  *
32  * The very common lock prefix is handled as special case in a
33  * separate table which is a pure address list without replacement ptr
34  * and size information.  That keeps the table sizes small.
35  */
36 
37 #ifdef CONFIG_SMP
38 #define LOCK_PREFIX_HERE \
39 		".pushsection .smp_locks,\"a\"\n"	\
40 		".balign 4\n"				\
41 		".long 671f - .\n" /* offset */		\
42 		".popsection\n"				\
43 		"671:"
44 
45 #define LOCK_PREFIX LOCK_PREFIX_HERE "\n\tlock; "
46 
47 #else /* ! CONFIG_SMP */
48 #define LOCK_PREFIX_HERE ""
49 #define LOCK_PREFIX ""
50 #endif
51 
52 /*
53  * objtool annotation to ignore the alternatives and only consider the original
54  * instruction(s).
55  */
56 #define ANNOTATE_IGNORE_ALTERNATIVE				\
57 	"999:\n\t"						\
58 	".pushsection .discard.ignore_alts\n\t"			\
59 	".long 999b - .\n\t"					\
60 	".popsection\n\t"
61 
62 struct alt_instr {
63 	s32 instr_offset;	/* original instruction */
64 	s32 repl_offset;	/* offset to replacement instruction */
65 	u16 cpuid;		/* cpuid bit set for replacement */
66 	u8  instrlen;		/* length of original instruction */
67 	u8  replacementlen;	/* length of new instruction */
68 } __packed;
69 
70 /*
71  * Debug flag that can be tested to see whether alternative
72  * instructions were patched in already:
73  */
74 extern int alternatives_patched;
75 
76 extern void alternative_instructions(void);
77 extern void apply_alternatives(struct alt_instr *start, struct alt_instr *end);
78 extern void apply_retpolines(s32 *start, s32 *end);
79 extern void apply_returns(s32 *start, s32 *end);
80 extern void apply_ibt_endbr(s32 *start, s32 *end);
81 
82 struct module;
83 
84 #ifdef CONFIG_SMP
85 extern void alternatives_smp_module_add(struct module *mod, char *name,
86 					void *locks, void *locks_end,
87 					void *text, void *text_end);
88 extern void alternatives_smp_module_del(struct module *mod);
89 extern void alternatives_enable_smp(void);
90 extern int alternatives_text_reserved(void *start, void *end);
91 extern bool skip_smp_alternatives;
92 #else
93 static inline void alternatives_smp_module_add(struct module *mod, char *name,
94 					       void *locks, void *locks_end,
95 					       void *text, void *text_end) {}
96 static inline void alternatives_smp_module_del(struct module *mod) {}
97 static inline void alternatives_enable_smp(void) {}
98 static inline int alternatives_text_reserved(void *start, void *end)
99 {
100 	return 0;
101 }
102 #endif	/* CONFIG_SMP */
103 
104 #define b_replacement(num)	"664"#num
105 #define e_replacement(num)	"665"#num
106 
107 #define alt_end_marker		"663"
108 #define alt_slen		"662b-661b"
109 #define alt_total_slen		alt_end_marker"b-661b"
110 #define alt_rlen(num)		e_replacement(num)"f-"b_replacement(num)"f"
111 
112 #define OLDINSTR(oldinstr, num)						\
113 	"# ALT: oldnstr\n"						\
114 	"661:\n\t" oldinstr "\n662:\n"					\
115 	"# ALT: padding\n"						\
116 	".skip -(((" alt_rlen(num) ")-(" alt_slen ")) > 0) * "		\
117 		"((" alt_rlen(num) ")-(" alt_slen ")),0x90\n"		\
118 	alt_end_marker ":\n"
119 
120 /*
121  * gas compatible max based on the idea from:
122  * http://graphics.stanford.edu/~seander/bithacks.html#IntegerMinOrMax
123  *
124  * The additional "-" is needed because gas uses a "true" value of -1.
125  */
126 #define alt_max_short(a, b)	"((" a ") ^ (((" a ") ^ (" b ")) & -(-((" a ") < (" b ")))))"
127 
128 /*
129  * Pad the second replacement alternative with additional NOPs if it is
130  * additionally longer than the first replacement alternative.
131  */
132 #define OLDINSTR_2(oldinstr, num1, num2) \
133 	"# ALT: oldinstr2\n"									\
134 	"661:\n\t" oldinstr "\n662:\n"								\
135 	"# ALT: padding2\n"									\
136 	".skip -((" alt_max_short(alt_rlen(num1), alt_rlen(num2)) " - (" alt_slen ")) > 0) * "	\
137 		"(" alt_max_short(alt_rlen(num1), alt_rlen(num2)) " - (" alt_slen ")), 0x90\n"	\
138 	alt_end_marker ":\n"
139 
140 #define OLDINSTR_3(oldinsn, n1, n2, n3)								\
141 	"# ALT: oldinstr3\n"									\
142 	"661:\n\t" oldinsn "\n662:\n"								\
143 	"# ALT: padding3\n"									\
144 	".skip -((" alt_max_short(alt_max_short(alt_rlen(n1), alt_rlen(n2)), alt_rlen(n3))	\
145 		" - (" alt_slen ")) > 0) * "							\
146 		"(" alt_max_short(alt_max_short(alt_rlen(n1), alt_rlen(n2)), alt_rlen(n3))	\
147 		" - (" alt_slen ")), 0x90\n"							\
148 	alt_end_marker ":\n"
149 
150 #define ALTINSTR_ENTRY(feature, num)					      \
151 	" .long 661b - .\n"				/* label           */ \
152 	" .long " b_replacement(num)"f - .\n"		/* new instruction */ \
153 	" .word " __stringify(feature) "\n"		/* feature bit     */ \
154 	" .byte " alt_total_slen "\n"			/* source len      */ \
155 	" .byte " alt_rlen(num) "\n"			/* replacement len */
156 
157 #define ALTINSTR_REPLACEMENT(newinstr, num)		/* replacement */	\
158 	"# ALT: replacement " #num "\n"						\
159 	b_replacement(num)":\n\t" newinstr "\n" e_replacement(num) ":\n"
160 
161 /* alternative assembly primitive: */
162 #define ALTERNATIVE(oldinstr, newinstr, feature)			\
163 	OLDINSTR(oldinstr, 1)						\
164 	".pushsection .altinstructions,\"a\"\n"				\
165 	ALTINSTR_ENTRY(feature, 1)					\
166 	".popsection\n"							\
167 	".pushsection .altinstr_replacement, \"ax\"\n"			\
168 	ALTINSTR_REPLACEMENT(newinstr, 1)				\
169 	".popsection\n"
170 
171 #define ALTERNATIVE_2(oldinstr, newinstr1, feature1, newinstr2, feature2)\
172 	OLDINSTR_2(oldinstr, 1, 2)					\
173 	".pushsection .altinstructions,\"a\"\n"				\
174 	ALTINSTR_ENTRY(feature1, 1)					\
175 	ALTINSTR_ENTRY(feature2, 2)					\
176 	".popsection\n"							\
177 	".pushsection .altinstr_replacement, \"ax\"\n"			\
178 	ALTINSTR_REPLACEMENT(newinstr1, 1)				\
179 	ALTINSTR_REPLACEMENT(newinstr2, 2)				\
180 	".popsection\n"
181 
182 /* If @feature is set, patch in @newinstr_yes, otherwise @newinstr_no. */
183 #define ALTERNATIVE_TERNARY(oldinstr, feature, newinstr_yes, newinstr_no) \
184 	ALTERNATIVE_2(oldinstr, newinstr_no, X86_FEATURE_ALWAYS,	\
185 		      newinstr_yes, feature)
186 
187 #define ALTERNATIVE_3(oldinsn, newinsn1, feat1, newinsn2, feat2, newinsn3, feat3) \
188 	OLDINSTR_3(oldinsn, 1, 2, 3)						\
189 	".pushsection .altinstructions,\"a\"\n"					\
190 	ALTINSTR_ENTRY(feat1, 1)						\
191 	ALTINSTR_ENTRY(feat2, 2)						\
192 	ALTINSTR_ENTRY(feat3, 3)						\
193 	".popsection\n"								\
194 	".pushsection .altinstr_replacement, \"ax\"\n"				\
195 	ALTINSTR_REPLACEMENT(newinsn1, 1)					\
196 	ALTINSTR_REPLACEMENT(newinsn2, 2)					\
197 	ALTINSTR_REPLACEMENT(newinsn3, 3)					\
198 	".popsection\n"
199 
200 /*
201  * Alternative instructions for different CPU types or capabilities.
202  *
203  * This allows to use optimized instructions even on generic binary
204  * kernels.
205  *
206  * length of oldinstr must be longer or equal the length of newinstr
207  * It can be padded with nops as needed.
208  *
209  * For non barrier like inlines please define new variants
210  * without volatile and memory clobber.
211  */
212 #define alternative(oldinstr, newinstr, feature)			\
213 	asm_inline volatile (ALTERNATIVE(oldinstr, newinstr, feature) : : : "memory")
214 
215 #define alternative_2(oldinstr, newinstr1, feature1, newinstr2, feature2) \
216 	asm_inline volatile(ALTERNATIVE_2(oldinstr, newinstr1, feature1, newinstr2, feature2) ::: "memory")
217 
218 #define alternative_ternary(oldinstr, feature, newinstr_yes, newinstr_no) \
219 	asm_inline volatile(ALTERNATIVE_TERNARY(oldinstr, feature, newinstr_yes, newinstr_no) ::: "memory")
220 
221 /*
222  * Alternative inline assembly with input.
223  *
224  * Peculiarities:
225  * No memory clobber here.
226  * Argument numbers start with 1.
227  * Leaving an unused argument 0 to keep API compatibility.
228  */
229 #define alternative_input(oldinstr, newinstr, feature, input...)	\
230 	asm_inline volatile (ALTERNATIVE(oldinstr, newinstr, feature)	\
231 		: : "i" (0), ## input)
232 
233 /*
234  * This is similar to alternative_input. But it has two features and
235  * respective instructions.
236  *
237  * If CPU has feature2, newinstr2 is used.
238  * Otherwise, if CPU has feature1, newinstr1 is used.
239  * Otherwise, oldinstr is used.
240  */
241 #define alternative_input_2(oldinstr, newinstr1, feature1, newinstr2,	     \
242 			   feature2, input...)				     \
243 	asm_inline volatile(ALTERNATIVE_2(oldinstr, newinstr1, feature1,     \
244 		newinstr2, feature2)					     \
245 		: : "i" (0), ## input)
246 
247 /* Like alternative_input, but with a single output argument */
248 #define alternative_io(oldinstr, newinstr, feature, output, input...)	\
249 	asm_inline volatile (ALTERNATIVE(oldinstr, newinstr, feature)	\
250 		: output : "i" (0), ## input)
251 
252 /* Like alternative_io, but for replacing a direct call with another one. */
253 #define alternative_call(oldfunc, newfunc, feature, output, input...)	\
254 	asm_inline volatile (ALTERNATIVE("call %P[old]", "call %P[new]", feature) \
255 		: output : [old] "i" (oldfunc), [new] "i" (newfunc), ## input)
256 
257 /*
258  * Like alternative_call, but there are two features and respective functions.
259  * If CPU has feature2, function2 is used.
260  * Otherwise, if CPU has feature1, function1 is used.
261  * Otherwise, old function is used.
262  */
263 #define alternative_call_2(oldfunc, newfunc1, feature1, newfunc2, feature2,   \
264 			   output, input...)				      \
265 	asm_inline volatile (ALTERNATIVE_2("call %P[old]", "call %P[new1]", feature1,\
266 		"call %P[new2]", feature2)				      \
267 		: output, ASM_CALL_CONSTRAINT				      \
268 		: [old] "i" (oldfunc), [new1] "i" (newfunc1),		      \
269 		  [new2] "i" (newfunc2), ## input)
270 
271 /*
272  * use this macro(s) if you need more than one output parameter
273  * in alternative_io
274  */
275 #define ASM_OUTPUT2(a...) a
276 
277 /*
278  * use this macro if you need clobbers but no inputs in
279  * alternative_{input,io,call}()
280  */
281 #define ASM_NO_INPUT_CLOBBER(clbr...) "i" (0) : clbr
282 
283 #else /* __ASSEMBLY__ */
284 
285 #ifdef CONFIG_SMP
286 	.macro LOCK_PREFIX
287 672:	lock
288 	.pushsection .smp_locks,"a"
289 	.balign 4
290 	.long 672b - .
291 	.popsection
292 	.endm
293 #else
294 	.macro LOCK_PREFIX
295 	.endm
296 #endif
297 
298 /*
299  * objtool annotation to ignore the alternatives and only consider the original
300  * instruction(s).
301  */
302 .macro ANNOTATE_IGNORE_ALTERNATIVE
303 	.Lannotate_\@:
304 	.pushsection .discard.ignore_alts
305 	.long .Lannotate_\@ - .
306 	.popsection
307 .endm
308 
309 /*
310  * Issue one struct alt_instr descriptor entry (need to put it into
311  * the section .altinstructions, see below). This entry contains
312  * enough information for the alternatives patching code to patch an
313  * instruction. See apply_alternatives().
314  */
315 .macro altinstruction_entry orig alt feature orig_len alt_len
316 	.long \orig - .
317 	.long \alt - .
318 	.word \feature
319 	.byte \orig_len
320 	.byte \alt_len
321 .endm
322 
323 /*
324  * Define an alternative between two instructions. If @feature is
325  * present, early code in apply_alternatives() replaces @oldinstr with
326  * @newinstr. ".skip" directive takes care of proper instruction padding
327  * in case @newinstr is longer than @oldinstr.
328  */
329 .macro ALTERNATIVE oldinstr, newinstr, feature
330 140:
331 	\oldinstr
332 141:
333 	.skip -(((144f-143f)-(141b-140b)) > 0) * ((144f-143f)-(141b-140b)),0x90
334 142:
335 
336 	.pushsection .altinstructions,"a"
337 	altinstruction_entry 140b,143f,\feature,142b-140b,144f-143f
338 	.popsection
339 
340 	.pushsection .altinstr_replacement,"ax"
341 143:
342 	\newinstr
343 144:
344 	.popsection
345 .endm
346 
347 #define old_len			141b-140b
348 #define new_len1		144f-143f
349 #define new_len2		145f-144f
350 
351 /*
352  * gas compatible max based on the idea from:
353  * http://graphics.stanford.edu/~seander/bithacks.html#IntegerMinOrMax
354  *
355  * The additional "-" is needed because gas uses a "true" value of -1.
356  */
357 #define alt_max_short(a, b)	((a) ^ (((a) ^ (b)) & -(-((a) < (b)))))
358 
359 
360 /*
361  * Same as ALTERNATIVE macro above but for two alternatives. If CPU
362  * has @feature1, it replaces @oldinstr with @newinstr1. If CPU has
363  * @feature2, it replaces @oldinstr with @feature2.
364  */
365 .macro ALTERNATIVE_2 oldinstr, newinstr1, feature1, newinstr2, feature2
366 140:
367 	\oldinstr
368 141:
369 	.skip -((alt_max_short(new_len1, new_len2) - (old_len)) > 0) * \
370 		(alt_max_short(new_len1, new_len2) - (old_len)),0x90
371 142:
372 
373 	.pushsection .altinstructions,"a"
374 	altinstruction_entry 140b,143f,\feature1,142b-140b,144f-143f
375 	altinstruction_entry 140b,144f,\feature2,142b-140b,145f-144f
376 	.popsection
377 
378 	.pushsection .altinstr_replacement,"ax"
379 143:
380 	\newinstr1
381 144:
382 	\newinstr2
383 145:
384 	.popsection
385 .endm
386 
387 /* If @feature is set, patch in @newinstr_yes, otherwise @newinstr_no. */
388 #define ALTERNATIVE_TERNARY(oldinstr, feature, newinstr_yes, newinstr_no) \
389 	ALTERNATIVE_2 oldinstr, newinstr_no, X86_FEATURE_ALWAYS,	\
390 	newinstr_yes, feature
391 
392 #endif /* __ASSEMBLY__ */
393 
394 #endif /* _ASM_X86_ALTERNATIVE_H */
395