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 static inline unsigned long __read_cr4(void) 124 { 125 return native_read_cr4(); 126 } 127 128 #ifdef CONFIG_PARAVIRT_XXL 129 #include <asm/paravirt.h> 130 #else 131 132 static inline unsigned long read_cr0(void) 133 { 134 return native_read_cr0(); 135 } 136 137 static inline void write_cr0(unsigned long x) 138 { 139 native_write_cr0(x); 140 } 141 142 static __always_inline unsigned long read_cr2(void) 143 { 144 return native_read_cr2(); 145 } 146 147 static __always_inline void write_cr2(unsigned long x) 148 { 149 native_write_cr2(x); 150 } 151 152 /* 153 * Careful! CR3 contains more than just an address. You probably want 154 * read_cr3_pa() instead. 155 */ 156 static inline unsigned long __read_cr3(void) 157 { 158 return __native_read_cr3(); 159 } 160 161 static inline void write_cr3(unsigned long x) 162 { 163 native_write_cr3(x); 164 } 165 166 static inline void __write_cr4(unsigned long x) 167 { 168 native_write_cr4(x); 169 } 170 171 static __always_inline void wbinvd(void) 172 { 173 native_wbinvd(); 174 } 175 176 #endif /* CONFIG_PARAVIRT_XXL */ 177 178 static __always_inline void clflush(volatile void *__p) 179 { 180 asm volatile("clflush %0" : "+m" (*(volatile char __force *)__p)); 181 } 182 183 static inline void clflushopt(volatile void *__p) 184 { 185 alternative_io(".byte 0x3e; clflush %P0", 186 ".byte 0x66; clflush %P0", 187 X86_FEATURE_CLFLUSHOPT, 188 "+m" (*(volatile char __force *)__p)); 189 } 190 191 static inline void clwb(volatile void *__p) 192 { 193 volatile struct { char x[64]; } *p = __p; 194 195 asm volatile(ALTERNATIVE_2( 196 ".byte 0x3e; clflush (%[pax])", 197 ".byte 0x66; clflush (%[pax])", /* clflushopt (%%rax) */ 198 X86_FEATURE_CLFLUSHOPT, 199 ".byte 0x66, 0x0f, 0xae, 0x30", /* clwb (%%rax) */ 200 X86_FEATURE_CLWB) 201 : [p] "+m" (*p) 202 : [pax] "a" (p)); 203 } 204 205 #define nop() asm volatile ("nop") 206 207 static inline void serialize(void) 208 { 209 /* Instruction opcode for SERIALIZE; supported in binutils >= 2.35. */ 210 asm volatile(".byte 0xf, 0x1, 0xe8" ::: "memory"); 211 } 212 213 /* The dst parameter must be 64-bytes aligned */ 214 static inline void movdir64b(void __iomem *dst, const void *src) 215 { 216 const struct { char _[64]; } *__src = src; 217 struct { char _[64]; } __iomem *__dst = dst; 218 219 /* 220 * MOVDIR64B %(rdx), rax. 221 * 222 * Both __src and __dst must be memory constraints in order to tell the 223 * compiler that no other memory accesses should be reordered around 224 * this one. 225 * 226 * Also, both must be supplied as lvalues because this tells 227 * the compiler what the object is (its size) the instruction accesses. 228 * I.e., not the pointers but what they point to, thus the deref'ing '*'. 229 */ 230 asm volatile(".byte 0x66, 0x0f, 0x38, 0xf8, 0x02" 231 : "+m" (*__dst) 232 : "m" (*__src), "a" (__dst), "d" (__src)); 233 } 234 235 /** 236 * enqcmds - Enqueue a command in supervisor (CPL0) mode 237 * @dst: destination, in MMIO space (must be 512-bit aligned) 238 * @src: 512 bits memory operand 239 * 240 * The ENQCMDS instruction allows software to write a 512-bit command to 241 * a 512-bit-aligned special MMIO region that supports the instruction. 242 * A return status is loaded into the ZF flag in the RFLAGS register. 243 * ZF = 0 equates to success, and ZF = 1 indicates retry or error. 244 * 245 * This function issues the ENQCMDS instruction to submit data from 246 * kernel space to MMIO space, in a unit of 512 bits. Order of data access 247 * is not guaranteed, nor is a memory barrier performed afterwards. It 248 * returns 0 on success and -EAGAIN on failure. 249 * 250 * Warning: Do not use this helper unless your driver has checked that the 251 * ENQCMDS instruction is supported on the platform and the device accepts 252 * ENQCMDS. 253 */ 254 static inline int enqcmds(void __iomem *dst, const void *src) 255 { 256 const struct { char _[64]; } *__src = src; 257 struct { char _[64]; } __iomem *__dst = dst; 258 bool zf; 259 260 /* 261 * ENQCMDS %(rdx), rax 262 * 263 * See movdir64b()'s comment on operand specification. 264 */ 265 asm volatile(".byte 0xf3, 0x0f, 0x38, 0xf8, 0x02, 0x66, 0x90" 266 CC_SET(z) 267 : CC_OUT(z) (zf), "+m" (*__dst) 268 : "m" (*__src), "a" (__dst), "d" (__src)); 269 270 /* Submission failure is indicated via EFLAGS.ZF=1 */ 271 if (zf) 272 return -EAGAIN; 273 274 return 0; 275 } 276 277 static __always_inline void tile_release(void) 278 { 279 /* 280 * Instruction opcode for TILERELEASE; supported in binutils 281 * version >= 2.36. 282 */ 283 asm volatile(".byte 0xc4, 0xe2, 0x78, 0x49, 0xc0"); 284 } 285 286 #endif /* __KERNEL__ */ 287 288 #endif /* _ASM_X86_SPECIAL_INSNS_H */ 289