1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #include <linux/jump_label.h> 3 #include <asm/unwind_hints.h> 4 #include <asm/cpufeatures.h> 5 #include <asm/page_types.h> 6 #include <asm/percpu.h> 7 #include <asm/asm-offsets.h> 8 #include <asm/processor-flags.h> 9 #include <asm/inst.h> 10 11 /* 12 13 x86 function call convention, 64-bit: 14 ------------------------------------- 15 arguments | callee-saved | extra caller-saved | return 16 [callee-clobbered] | | [callee-clobbered] | 17 --------------------------------------------------------------------------- 18 rdi rsi rdx rcx r8-9 | rbx rbp [*] r12-15 | r10-11 | rax, rdx [**] 19 20 ( rsp is obviously invariant across normal function calls. (gcc can 'merge' 21 functions when it sees tail-call optimization possibilities) rflags is 22 clobbered. Leftover arguments are passed over the stack frame.) 23 24 [*] In the frame-pointers case rbp is fixed to the stack frame. 25 26 [**] for struct return values wider than 64 bits the return convention is a 27 bit more complex: up to 128 bits width we return small structures 28 straight in rax, rdx. For structures larger than that (3 words or 29 larger) the caller puts a pointer to an on-stack return struct 30 [allocated in the caller's stack frame] into the first argument - i.e. 31 into rdi. All other arguments shift up by one in this case. 32 Fortunately this case is rare in the kernel. 33 34 For 32-bit we have the following conventions - kernel is built with 35 -mregparm=3 and -freg-struct-return: 36 37 x86 function calling convention, 32-bit: 38 ---------------------------------------- 39 arguments | callee-saved | extra caller-saved | return 40 [callee-clobbered] | | [callee-clobbered] | 41 ------------------------------------------------------------------------- 42 eax edx ecx | ebx edi esi ebp [*] | <none> | eax, edx [**] 43 44 ( here too esp is obviously invariant across normal function calls. eflags 45 is clobbered. Leftover arguments are passed over the stack frame. ) 46 47 [*] In the frame-pointers case ebp is fixed to the stack frame. 48 49 [**] We build with -freg-struct-return, which on 32-bit means similar 50 semantics as on 64-bit: edx can be used for a second return value 51 (i.e. covering integer and structure sizes up to 64 bits) - after that 52 it gets more complex and more expensive: 3-word or larger struct returns 53 get done in the caller's frame and the pointer to the return struct goes 54 into regparm0, i.e. eax - the other arguments shift up and the 55 function's register parameters degenerate to regparm=2 in essence. 56 57 */ 58 59 #ifdef CONFIG_X86_64 60 61 /* 62 * 64-bit system call stack frame layout defines and helpers, 63 * for assembly code: 64 */ 65 66 /* The layout forms the "struct pt_regs" on the stack: */ 67 /* 68 * C ABI says these regs are callee-preserved. They aren't saved on kernel entry 69 * unless syscall needs a complete, fully filled "struct pt_regs". 70 */ 71 #define R15 0*8 72 #define R14 1*8 73 #define R13 2*8 74 #define R12 3*8 75 #define RBP 4*8 76 #define RBX 5*8 77 /* These regs are callee-clobbered. Always saved on kernel entry. */ 78 #define R11 6*8 79 #define R10 7*8 80 #define R9 8*8 81 #define R8 9*8 82 #define RAX 10*8 83 #define RCX 11*8 84 #define RDX 12*8 85 #define RSI 13*8 86 #define RDI 14*8 87 /* 88 * On syscall entry, this is syscall#. On CPU exception, this is error code. 89 * On hw interrupt, it's IRQ number: 90 */ 91 #define ORIG_RAX 15*8 92 /* Return frame for iretq */ 93 #define RIP 16*8 94 #define CS 17*8 95 #define EFLAGS 18*8 96 #define RSP 19*8 97 #define SS 20*8 98 99 #define SIZEOF_PTREGS 21*8 100 101 .macro PUSH_AND_CLEAR_REGS rdx=%rdx rax=%rax save_ret=0 102 .if \save_ret 103 pushq %rsi /* pt_regs->si */ 104 movq 8(%rsp), %rsi /* temporarily store the return address in %rsi */ 105 movq %rdi, 8(%rsp) /* pt_regs->di (overwriting original return address) */ 106 .else 107 pushq %rdi /* pt_regs->di */ 108 pushq %rsi /* pt_regs->si */ 109 .endif 110 pushq \rdx /* pt_regs->dx */ 111 pushq %rcx /* pt_regs->cx */ 112 pushq \rax /* pt_regs->ax */ 113 pushq %r8 /* pt_regs->r8 */ 114 pushq %r9 /* pt_regs->r9 */ 115 pushq %r10 /* pt_regs->r10 */ 116 pushq %r11 /* pt_regs->r11 */ 117 pushq %rbx /* pt_regs->rbx */ 118 pushq %rbp /* pt_regs->rbp */ 119 pushq %r12 /* pt_regs->r12 */ 120 pushq %r13 /* pt_regs->r13 */ 121 pushq %r14 /* pt_regs->r14 */ 122 pushq %r15 /* pt_regs->r15 */ 123 UNWIND_HINT_REGS 124 125 .if \save_ret 126 pushq %rsi /* return address on top of stack */ 127 .endif 128 129 /* 130 * Sanitize registers of values that a speculation attack might 131 * otherwise want to exploit. The lower registers are likely clobbered 132 * well before they could be put to use in a speculative execution 133 * gadget. 134 */ 135 xorl %edx, %edx /* nospec dx */ 136 xorl %ecx, %ecx /* nospec cx */ 137 xorl %r8d, %r8d /* nospec r8 */ 138 xorl %r9d, %r9d /* nospec r9 */ 139 xorl %r10d, %r10d /* nospec r10 */ 140 xorl %r11d, %r11d /* nospec r11 */ 141 xorl %ebx, %ebx /* nospec rbx */ 142 xorl %ebp, %ebp /* nospec rbp */ 143 xorl %r12d, %r12d /* nospec r12 */ 144 xorl %r13d, %r13d /* nospec r13 */ 145 xorl %r14d, %r14d /* nospec r14 */ 146 xorl %r15d, %r15d /* nospec r15 */ 147 148 .endm 149 150 .macro POP_REGS pop_rdi=1 skip_r11rcx=0 151 popq %r15 152 popq %r14 153 popq %r13 154 popq %r12 155 popq %rbp 156 popq %rbx 157 .if \skip_r11rcx 158 popq %rsi 159 .else 160 popq %r11 161 .endif 162 popq %r10 163 popq %r9 164 popq %r8 165 popq %rax 166 .if \skip_r11rcx 167 popq %rsi 168 .else 169 popq %rcx 170 .endif 171 popq %rdx 172 popq %rsi 173 .if \pop_rdi 174 popq %rdi 175 .endif 176 .endm 177 178 #ifdef CONFIG_PAGE_TABLE_ISOLATION 179 180 /* 181 * PAGE_TABLE_ISOLATION PGDs are 8k. Flip bit 12 to switch between the two 182 * halves: 183 */ 184 #define PTI_USER_PGTABLE_BIT PAGE_SHIFT 185 #define PTI_USER_PGTABLE_MASK (1 << PTI_USER_PGTABLE_BIT) 186 #define PTI_USER_PCID_BIT X86_CR3_PTI_PCID_USER_BIT 187 #define PTI_USER_PCID_MASK (1 << PTI_USER_PCID_BIT) 188 #define PTI_USER_PGTABLE_AND_PCID_MASK (PTI_USER_PCID_MASK | PTI_USER_PGTABLE_MASK) 189 190 .macro SET_NOFLUSH_BIT reg:req 191 bts $X86_CR3_PCID_NOFLUSH_BIT, \reg 192 .endm 193 194 .macro ADJUST_KERNEL_CR3 reg:req 195 ALTERNATIVE "", "SET_NOFLUSH_BIT \reg", X86_FEATURE_PCID 196 /* Clear PCID and "PAGE_TABLE_ISOLATION bit", point CR3 at kernel pagetables: */ 197 andq $(~PTI_USER_PGTABLE_AND_PCID_MASK), \reg 198 .endm 199 200 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req 201 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 202 mov %cr3, \scratch_reg 203 ADJUST_KERNEL_CR3 \scratch_reg 204 mov \scratch_reg, %cr3 205 .Lend_\@: 206 .endm 207 208 #define THIS_CPU_user_pcid_flush_mask \ 209 PER_CPU_VAR(cpu_tlbstate) + TLB_STATE_user_pcid_flush_mask 210 211 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req 212 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 213 mov %cr3, \scratch_reg 214 215 ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID 216 217 /* 218 * Test if the ASID needs a flush. 219 */ 220 movq \scratch_reg, \scratch_reg2 221 andq $(0x7FF), \scratch_reg /* mask ASID */ 222 bt \scratch_reg, THIS_CPU_user_pcid_flush_mask 223 jnc .Lnoflush_\@ 224 225 /* Flush needed, clear the bit */ 226 btr \scratch_reg, THIS_CPU_user_pcid_flush_mask 227 movq \scratch_reg2, \scratch_reg 228 jmp .Lwrcr3_pcid_\@ 229 230 .Lnoflush_\@: 231 movq \scratch_reg2, \scratch_reg 232 SET_NOFLUSH_BIT \scratch_reg 233 234 .Lwrcr3_pcid_\@: 235 /* Flip the ASID to the user version */ 236 orq $(PTI_USER_PCID_MASK), \scratch_reg 237 238 .Lwrcr3_\@: 239 /* Flip the PGD to the user version */ 240 orq $(PTI_USER_PGTABLE_MASK), \scratch_reg 241 mov \scratch_reg, %cr3 242 .Lend_\@: 243 .endm 244 245 .macro SWITCH_TO_USER_CR3_STACK scratch_reg:req 246 pushq %rax 247 SWITCH_TO_USER_CR3_NOSTACK scratch_reg=\scratch_reg scratch_reg2=%rax 248 popq %rax 249 .endm 250 251 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req 252 ALTERNATIVE "jmp .Ldone_\@", "", X86_FEATURE_PTI 253 movq %cr3, \scratch_reg 254 movq \scratch_reg, \save_reg 255 /* 256 * Test the user pagetable bit. If set, then the user page tables 257 * are active. If clear CR3 already has the kernel page table 258 * active. 259 */ 260 bt $PTI_USER_PGTABLE_BIT, \scratch_reg 261 jnc .Ldone_\@ 262 263 ADJUST_KERNEL_CR3 \scratch_reg 264 movq \scratch_reg, %cr3 265 266 .Ldone_\@: 267 .endm 268 269 .macro RESTORE_CR3 scratch_reg:req save_reg:req 270 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 271 272 ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID 273 274 /* 275 * KERNEL pages can always resume with NOFLUSH as we do 276 * explicit flushes. 277 */ 278 bt $PTI_USER_PGTABLE_BIT, \save_reg 279 jnc .Lnoflush_\@ 280 281 /* 282 * Check if there's a pending flush for the user ASID we're 283 * about to set. 284 */ 285 movq \save_reg, \scratch_reg 286 andq $(0x7FF), \scratch_reg 287 bt \scratch_reg, THIS_CPU_user_pcid_flush_mask 288 jnc .Lnoflush_\@ 289 290 btr \scratch_reg, THIS_CPU_user_pcid_flush_mask 291 jmp .Lwrcr3_\@ 292 293 .Lnoflush_\@: 294 SET_NOFLUSH_BIT \save_reg 295 296 .Lwrcr3_\@: 297 /* 298 * The CR3 write could be avoided when not changing its value, 299 * but would require a CR3 read *and* a scratch register. 300 */ 301 movq \save_reg, %cr3 302 .Lend_\@: 303 .endm 304 305 #else /* CONFIG_PAGE_TABLE_ISOLATION=n: */ 306 307 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req 308 .endm 309 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req 310 .endm 311 .macro SWITCH_TO_USER_CR3_STACK scratch_reg:req 312 .endm 313 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req 314 .endm 315 .macro RESTORE_CR3 scratch_reg:req save_reg:req 316 .endm 317 318 #endif 319 320 /* 321 * Mitigate Spectre v1 for conditional swapgs code paths. 322 * 323 * FENCE_SWAPGS_USER_ENTRY is used in the user entry swapgs code path, to 324 * prevent a speculative swapgs when coming from kernel space. 325 * 326 * FENCE_SWAPGS_KERNEL_ENTRY is used in the kernel entry non-swapgs code path, 327 * to prevent the swapgs from getting speculatively skipped when coming from 328 * user space. 329 */ 330 .macro FENCE_SWAPGS_USER_ENTRY 331 ALTERNATIVE "", "lfence", X86_FEATURE_FENCE_SWAPGS_USER 332 .endm 333 .macro FENCE_SWAPGS_KERNEL_ENTRY 334 ALTERNATIVE "", "lfence", X86_FEATURE_FENCE_SWAPGS_KERNEL 335 .endm 336 337 .macro STACKLEAK_ERASE_NOCLOBBER 338 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK 339 PUSH_AND_CLEAR_REGS 340 call stackleak_erase 341 POP_REGS 342 #endif 343 .endm 344 345 .macro SAVE_AND_SET_GSBASE scratch_reg:req save_reg:req 346 rdgsbase \save_reg 347 GET_PERCPU_BASE \scratch_reg 348 wrgsbase \scratch_reg 349 .endm 350 351 #else /* CONFIG_X86_64 */ 352 # undef UNWIND_HINT_IRET_REGS 353 # define UNWIND_HINT_IRET_REGS 354 #endif /* !CONFIG_X86_64 */ 355 356 .macro STACKLEAK_ERASE 357 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK 358 call stackleak_erase 359 #endif 360 .endm 361 362 #ifdef CONFIG_SMP 363 364 /* 365 * CPU/node NR is loaded from the limit (size) field of a special segment 366 * descriptor entry in GDT. 367 */ 368 .macro LOAD_CPU_AND_NODE_SEG_LIMIT reg:req 369 movq $__CPUNODE_SEG, \reg 370 lsl \reg, \reg 371 .endm 372 373 /* 374 * Fetch the per-CPU GSBASE value for this processor and put it in @reg. 375 * We normally use %gs for accessing per-CPU data, but we are setting up 376 * %gs here and obviously can not use %gs itself to access per-CPU data. 377 * 378 * Do not use RDPID, because KVM loads guest's TSC_AUX on vm-entry and 379 * may not restore the host's value until the CPU returns to userspace. 380 * Thus the kernel would consume a guest's TSC_AUX if an NMI arrives 381 * while running KVM's run loop. 382 */ 383 .macro GET_PERCPU_BASE reg:req 384 LOAD_CPU_AND_NODE_SEG_LIMIT \reg 385 andq $VDSO_CPUNODE_MASK, \reg 386 movq __per_cpu_offset(, \reg, 8), \reg 387 .endm 388 389 #else 390 391 .macro GET_PERCPU_BASE reg:req 392 movq pcpu_unit_offsets(%rip), \reg 393 .endm 394 395 #endif /* CONFIG_SMP */ 396