xref: /linux/arch/x86/include/asm/special_insns.h (revision 10a099405fdf96c3205ae775abf19fcd301b31b4)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_SPECIAL_INSNS_H
3 #define _ASM_X86_SPECIAL_INSNS_H
4 
5 
6 #ifdef __KERNEL__
7 
8 #include <asm/nops.h>
9 #include <asm/processor-flags.h>
10 #include <linux/irqflags.h>
11 #include <linux/jump_label.h>
12 
13 /*
14  * The compiler should not reorder volatile asm statements with respect to each
15  * other: they should execute in program order. However GCC 4.9.x and 5.x have
16  * a bug (which was fixed in 8.1, 7.3 and 6.5) where they might reorder
17  * volatile asm. The write functions are not affected since they have memory
18  * clobbers preventing reordering. To prevent reads from being reordered with
19  * respect to writes, use a dummy memory operand.
20  */
21 
22 #define __FORCE_ORDER "m"(*(unsigned int *)0x1000UL)
23 
24 void native_write_cr0(unsigned long val);
25 
26 static inline unsigned long native_read_cr0(void)
27 {
28 	unsigned long val;
29 	asm volatile("mov %%cr0,%0\n\t" : "=r" (val) : __FORCE_ORDER);
30 	return val;
31 }
32 
33 static __always_inline unsigned long native_read_cr2(void)
34 {
35 	unsigned long val;
36 	asm volatile("mov %%cr2,%0\n\t" : "=r" (val) : __FORCE_ORDER);
37 	return val;
38 }
39 
40 static __always_inline void native_write_cr2(unsigned long val)
41 {
42 	asm volatile("mov %0,%%cr2": : "r" (val) : "memory");
43 }
44 
45 static inline unsigned long __native_read_cr3(void)
46 {
47 	unsigned long val;
48 	asm volatile("mov %%cr3,%0\n\t" : "=r" (val) : __FORCE_ORDER);
49 	return val;
50 }
51 
52 static inline void native_write_cr3(unsigned long val)
53 {
54 	asm volatile("mov %0,%%cr3": : "r" (val) : "memory");
55 }
56 
57 static inline unsigned long native_read_cr4(void)
58 {
59 	unsigned long val;
60 #ifdef CONFIG_X86_32
61 	/*
62 	 * This could fault if CR4 does not exist.  Non-existent CR4
63 	 * is functionally equivalent to CR4 == 0.  Keep it simple and pretend
64 	 * that CR4 == 0 on CPUs that don't have CR4.
65 	 */
66 	asm volatile("1: mov %%cr4, %0\n"
67 		     "2:\n"
68 		     _ASM_EXTABLE(1b, 2b)
69 		     : "=r" (val) : "0" (0), __FORCE_ORDER);
70 #else
71 	/* CR4 always exists on x86_64. */
72 	asm volatile("mov %%cr4,%0\n\t" : "=r" (val) : __FORCE_ORDER);
73 #endif
74 	return val;
75 }
76 
77 void native_write_cr4(unsigned long val);
78 
79 #ifdef CONFIG_X86_INTEL_MEMORY_PROTECTION_KEYS
80 static inline u32 rdpkru(void)
81 {
82 	u32 ecx = 0;
83 	u32 edx, pkru;
84 
85 	/*
86 	 * "rdpkru" instruction.  Places PKRU contents in to EAX,
87 	 * clears EDX and requires that ecx=0.
88 	 */
89 	asm volatile(".byte 0x0f,0x01,0xee\n\t"
90 		     : "=a" (pkru), "=d" (edx)
91 		     : "c" (ecx));
92 	return pkru;
93 }
94 
95 static inline void wrpkru(u32 pkru)
96 {
97 	u32 ecx = 0, edx = 0;
98 
99 	/*
100 	 * "wrpkru" instruction.  Loads contents in EAX to PKRU,
101 	 * requires that ecx = edx = 0.
102 	 */
103 	asm volatile(".byte 0x0f,0x01,0xef\n\t"
104 		     : : "a" (pkru), "c"(ecx), "d"(edx));
105 }
106 
107 #else
108 static inline u32 rdpkru(void)
109 {
110 	return 0;
111 }
112 
113 static inline void wrpkru(u32 pkru)
114 {
115 }
116 #endif
117 
118 static __always_inline void native_wbinvd(void)
119 {
120 	asm volatile("wbinvd": : :"memory");
121 }
122 
123 extern asmlinkage void asm_load_gs_index(unsigned int selector);
124 
125 static inline void native_load_gs_index(unsigned int selector)
126 {
127 	unsigned long flags;
128 
129 	local_irq_save(flags);
130 	asm_load_gs_index(selector);
131 	local_irq_restore(flags);
132 }
133 
134 static inline unsigned long __read_cr4(void)
135 {
136 	return native_read_cr4();
137 }
138 
139 #ifdef CONFIG_PARAVIRT_XXL
140 #include <asm/paravirt.h>
141 #else
142 
143 static inline unsigned long read_cr0(void)
144 {
145 	return native_read_cr0();
146 }
147 
148 static inline void write_cr0(unsigned long x)
149 {
150 	native_write_cr0(x);
151 }
152 
153 static __always_inline unsigned long read_cr2(void)
154 {
155 	return native_read_cr2();
156 }
157 
158 static __always_inline void write_cr2(unsigned long x)
159 {
160 	native_write_cr2(x);
161 }
162 
163 /*
164  * Careful!  CR3 contains more than just an address.  You probably want
165  * read_cr3_pa() instead.
166  */
167 static inline unsigned long __read_cr3(void)
168 {
169 	return __native_read_cr3();
170 }
171 
172 static inline void write_cr3(unsigned long x)
173 {
174 	native_write_cr3(x);
175 }
176 
177 static inline void __write_cr4(unsigned long x)
178 {
179 	native_write_cr4(x);
180 }
181 
182 static __always_inline void wbinvd(void)
183 {
184 	native_wbinvd();
185 }
186 
187 
188 static inline void load_gs_index(unsigned int selector)
189 {
190 #ifdef CONFIG_X86_64
191 	native_load_gs_index(selector);
192 #else
193 	loadsegment(gs, selector);
194 #endif
195 }
196 
197 #endif /* CONFIG_PARAVIRT_XXL */
198 
199 static __always_inline void clflush(volatile void *__p)
200 {
201 	asm volatile("clflush %0" : "+m" (*(volatile char __force *)__p));
202 }
203 
204 static inline void clflushopt(volatile void *__p)
205 {
206 	alternative_io(".byte 0x3e; clflush %P0",
207 		       ".byte 0x66; clflush %P0",
208 		       X86_FEATURE_CLFLUSHOPT,
209 		       "+m" (*(volatile char __force *)__p));
210 }
211 
212 static inline void clwb(volatile void *__p)
213 {
214 	volatile struct { char x[64]; } *p = __p;
215 
216 	asm volatile(ALTERNATIVE_2(
217 		".byte 0x3e; clflush (%[pax])",
218 		".byte 0x66; clflush (%[pax])", /* clflushopt (%%rax) */
219 		X86_FEATURE_CLFLUSHOPT,
220 		".byte 0x66, 0x0f, 0xae, 0x30",  /* clwb (%%rax) */
221 		X86_FEATURE_CLWB)
222 		: [p] "+m" (*p)
223 		: [pax] "a" (p));
224 }
225 
226 #define nop() asm volatile ("nop")
227 
228 static inline void serialize(void)
229 {
230 	/* Instruction opcode for SERIALIZE; supported in binutils >= 2.35. */
231 	asm volatile(".byte 0xf, 0x1, 0xe8" ::: "memory");
232 }
233 
234 /* The dst parameter must be 64-bytes aligned */
235 static inline void movdir64b(void __iomem *dst, const void *src)
236 {
237 	const struct { char _[64]; } *__src = src;
238 	struct { char _[64]; } __iomem *__dst = dst;
239 
240 	/*
241 	 * MOVDIR64B %(rdx), rax.
242 	 *
243 	 * Both __src and __dst must be memory constraints in order to tell the
244 	 * compiler that no other memory accesses should be reordered around
245 	 * this one.
246 	 *
247 	 * Also, both must be supplied as lvalues because this tells
248 	 * the compiler what the object is (its size) the instruction accesses.
249 	 * I.e., not the pointers but what they point to, thus the deref'ing '*'.
250 	 */
251 	asm volatile(".byte 0x66, 0x0f, 0x38, 0xf8, 0x02"
252 		     : "+m" (*__dst)
253 		     :  "m" (*__src), "a" (__dst), "d" (__src));
254 }
255 
256 /**
257  * enqcmds - Enqueue a command in supervisor (CPL0) mode
258  * @dst: destination, in MMIO space (must be 512-bit aligned)
259  * @src: 512 bits memory operand
260  *
261  * The ENQCMDS instruction allows software to write a 512-bit command to
262  * a 512-bit-aligned special MMIO region that supports the instruction.
263  * A return status is loaded into the ZF flag in the RFLAGS register.
264  * ZF = 0 equates to success, and ZF = 1 indicates retry or error.
265  *
266  * This function issues the ENQCMDS instruction to submit data from
267  * kernel space to MMIO space, in a unit of 512 bits. Order of data access
268  * is not guaranteed, nor is a memory barrier performed afterwards. It
269  * returns 0 on success and -EAGAIN on failure.
270  *
271  * Warning: Do not use this helper unless your driver has checked that the
272  * ENQCMDS instruction is supported on the platform and the device accepts
273  * ENQCMDS.
274  */
275 static inline int enqcmds(void __iomem *dst, const void *src)
276 {
277 	const struct { char _[64]; } *__src = src;
278 	struct { char _[64]; } __iomem *__dst = dst;
279 	bool zf;
280 
281 	/*
282 	 * ENQCMDS %(rdx), rax
283 	 *
284 	 * See movdir64b()'s comment on operand specification.
285 	 */
286 	asm volatile(".byte 0xf3, 0x0f, 0x38, 0xf8, 0x02, 0x66, 0x90"
287 		     CC_SET(z)
288 		     : CC_OUT(z) (zf), "+m" (*__dst)
289 		     : "m" (*__src), "a" (__dst), "d" (__src));
290 
291 	/* Submission failure is indicated via EFLAGS.ZF=1 */
292 	if (zf)
293 		return -EAGAIN;
294 
295 	return 0;
296 }
297 
298 static __always_inline void tile_release(void)
299 {
300 	/*
301 	 * Instruction opcode for TILERELEASE; supported in binutils
302 	 * version >= 2.36.
303 	 */
304 	asm volatile(".byte 0xc4, 0xe2, 0x78, 0x49, 0xc0");
305 }
306 
307 #endif /* __KERNEL__ */
308 
309 #endif /* _ASM_X86_SPECIAL_INSNS_H */
310