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/ptrace-abi.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 .macro PUSH_REGS rdx=%rdx rax=%rax save_ret=0 67 .if \save_ret 68 pushq %rsi /* pt_regs->si */ 69 movq 8(%rsp), %rsi /* temporarily store the return address in %rsi */ 70 movq %rdi, 8(%rsp) /* pt_regs->di (overwriting original return address) */ 71 .else 72 pushq %rdi /* pt_regs->di */ 73 pushq %rsi /* pt_regs->si */ 74 .endif 75 pushq \rdx /* pt_regs->dx */ 76 pushq %rcx /* pt_regs->cx */ 77 pushq \rax /* pt_regs->ax */ 78 pushq %r8 /* pt_regs->r8 */ 79 pushq %r9 /* pt_regs->r9 */ 80 pushq %r10 /* pt_regs->r10 */ 81 pushq %r11 /* pt_regs->r11 */ 82 pushq %rbx /* pt_regs->rbx */ 83 pushq %rbp /* pt_regs->rbp */ 84 pushq %r12 /* pt_regs->r12 */ 85 pushq %r13 /* pt_regs->r13 */ 86 pushq %r14 /* pt_regs->r14 */ 87 pushq %r15 /* pt_regs->r15 */ 88 UNWIND_HINT_REGS 89 90 .if \save_ret 91 pushq %rsi /* return address on top of stack */ 92 .endif 93 .endm 94 95 .macro CLEAR_REGS 96 /* 97 * Sanitize registers of values that a speculation attack might 98 * otherwise want to exploit. The lower registers are likely clobbered 99 * well before they could be put to use in a speculative execution 100 * gadget. 101 */ 102 xorl %esi, %esi /* nospec si */ 103 xorl %edx, %edx /* nospec dx */ 104 xorl %ecx, %ecx /* nospec cx */ 105 xorl %r8d, %r8d /* nospec r8 */ 106 xorl %r9d, %r9d /* nospec r9 */ 107 xorl %r10d, %r10d /* nospec r10 */ 108 xorl %r11d, %r11d /* nospec r11 */ 109 xorl %ebx, %ebx /* nospec rbx */ 110 xorl %ebp, %ebp /* nospec rbp */ 111 xorl %r12d, %r12d /* nospec r12 */ 112 xorl %r13d, %r13d /* nospec r13 */ 113 xorl %r14d, %r14d /* nospec r14 */ 114 xorl %r15d, %r15d /* nospec r15 */ 115 116 .endm 117 118 .macro PUSH_AND_CLEAR_REGS rdx=%rdx rax=%rax save_ret=0 119 PUSH_REGS rdx=\rdx, rax=\rax, save_ret=\save_ret 120 CLEAR_REGS 121 .endm 122 123 .macro POP_REGS pop_rdi=1 124 popq %r15 125 popq %r14 126 popq %r13 127 popq %r12 128 popq %rbp 129 popq %rbx 130 popq %r11 131 popq %r10 132 popq %r9 133 popq %r8 134 popq %rax 135 popq %rcx 136 popq %rdx 137 popq %rsi 138 .if \pop_rdi 139 popq %rdi 140 .endif 141 .endm 142 143 #ifdef CONFIG_PAGE_TABLE_ISOLATION 144 145 /* 146 * PAGE_TABLE_ISOLATION PGDs are 8k. Flip bit 12 to switch between the two 147 * halves: 148 */ 149 #define PTI_USER_PGTABLE_BIT PAGE_SHIFT 150 #define PTI_USER_PGTABLE_MASK (1 << PTI_USER_PGTABLE_BIT) 151 #define PTI_USER_PCID_BIT X86_CR3_PTI_PCID_USER_BIT 152 #define PTI_USER_PCID_MASK (1 << PTI_USER_PCID_BIT) 153 #define PTI_USER_PGTABLE_AND_PCID_MASK (PTI_USER_PCID_MASK | PTI_USER_PGTABLE_MASK) 154 155 .macro SET_NOFLUSH_BIT reg:req 156 bts $X86_CR3_PCID_NOFLUSH_BIT, \reg 157 .endm 158 159 .macro ADJUST_KERNEL_CR3 reg:req 160 ALTERNATIVE "", "SET_NOFLUSH_BIT \reg", X86_FEATURE_PCID 161 /* Clear PCID and "PAGE_TABLE_ISOLATION bit", point CR3 at kernel pagetables: */ 162 andq $(~PTI_USER_PGTABLE_AND_PCID_MASK), \reg 163 .endm 164 165 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req 166 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 167 mov %cr3, \scratch_reg 168 ADJUST_KERNEL_CR3 \scratch_reg 169 mov \scratch_reg, %cr3 170 .Lend_\@: 171 .endm 172 173 #define THIS_CPU_user_pcid_flush_mask \ 174 PER_CPU_VAR(cpu_tlbstate) + TLB_STATE_user_pcid_flush_mask 175 176 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req 177 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 178 mov %cr3, \scratch_reg 179 180 ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID 181 182 /* 183 * Test if the ASID needs a flush. 184 */ 185 movq \scratch_reg, \scratch_reg2 186 andq $(0x7FF), \scratch_reg /* mask ASID */ 187 bt \scratch_reg, THIS_CPU_user_pcid_flush_mask 188 jnc .Lnoflush_\@ 189 190 /* Flush needed, clear the bit */ 191 btr \scratch_reg, THIS_CPU_user_pcid_flush_mask 192 movq \scratch_reg2, \scratch_reg 193 jmp .Lwrcr3_pcid_\@ 194 195 .Lnoflush_\@: 196 movq \scratch_reg2, \scratch_reg 197 SET_NOFLUSH_BIT \scratch_reg 198 199 .Lwrcr3_pcid_\@: 200 /* Flip the ASID to the user version */ 201 orq $(PTI_USER_PCID_MASK), \scratch_reg 202 203 .Lwrcr3_\@: 204 /* Flip the PGD to the user version */ 205 orq $(PTI_USER_PGTABLE_MASK), \scratch_reg 206 mov \scratch_reg, %cr3 207 .Lend_\@: 208 .endm 209 210 .macro SWITCH_TO_USER_CR3_STACK scratch_reg:req 211 pushq %rax 212 SWITCH_TO_USER_CR3_NOSTACK scratch_reg=\scratch_reg scratch_reg2=%rax 213 popq %rax 214 .endm 215 216 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req 217 ALTERNATIVE "jmp .Ldone_\@", "", X86_FEATURE_PTI 218 movq %cr3, \scratch_reg 219 movq \scratch_reg, \save_reg 220 /* 221 * Test the user pagetable bit. If set, then the user page tables 222 * are active. If clear CR3 already has the kernel page table 223 * active. 224 */ 225 bt $PTI_USER_PGTABLE_BIT, \scratch_reg 226 jnc .Ldone_\@ 227 228 ADJUST_KERNEL_CR3 \scratch_reg 229 movq \scratch_reg, %cr3 230 231 .Ldone_\@: 232 .endm 233 234 .macro RESTORE_CR3 scratch_reg:req save_reg:req 235 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 236 237 ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID 238 239 /* 240 * KERNEL pages can always resume with NOFLUSH as we do 241 * explicit flushes. 242 */ 243 bt $PTI_USER_PGTABLE_BIT, \save_reg 244 jnc .Lnoflush_\@ 245 246 /* 247 * Check if there's a pending flush for the user ASID we're 248 * about to set. 249 */ 250 movq \save_reg, \scratch_reg 251 andq $(0x7FF), \scratch_reg 252 bt \scratch_reg, THIS_CPU_user_pcid_flush_mask 253 jnc .Lnoflush_\@ 254 255 btr \scratch_reg, THIS_CPU_user_pcid_flush_mask 256 jmp .Lwrcr3_\@ 257 258 .Lnoflush_\@: 259 SET_NOFLUSH_BIT \save_reg 260 261 .Lwrcr3_\@: 262 /* 263 * The CR3 write could be avoided when not changing its value, 264 * but would require a CR3 read *and* a scratch register. 265 */ 266 movq \save_reg, %cr3 267 .Lend_\@: 268 .endm 269 270 #else /* CONFIG_PAGE_TABLE_ISOLATION=n: */ 271 272 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req 273 .endm 274 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req 275 .endm 276 .macro SWITCH_TO_USER_CR3_STACK scratch_reg:req 277 .endm 278 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req 279 .endm 280 .macro RESTORE_CR3 scratch_reg:req save_reg:req 281 .endm 282 283 #endif 284 285 /* 286 * Mitigate Spectre v1 for conditional swapgs code paths. 287 * 288 * FENCE_SWAPGS_USER_ENTRY is used in the user entry swapgs code path, to 289 * prevent a speculative swapgs when coming from kernel space. 290 * 291 * FENCE_SWAPGS_KERNEL_ENTRY is used in the kernel entry non-swapgs code path, 292 * to prevent the swapgs from getting speculatively skipped when coming from 293 * user space. 294 */ 295 .macro FENCE_SWAPGS_USER_ENTRY 296 ALTERNATIVE "", "lfence", X86_FEATURE_FENCE_SWAPGS_USER 297 .endm 298 .macro FENCE_SWAPGS_KERNEL_ENTRY 299 ALTERNATIVE "", "lfence", X86_FEATURE_FENCE_SWAPGS_KERNEL 300 .endm 301 302 .macro STACKLEAK_ERASE_NOCLOBBER 303 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK 304 PUSH_AND_CLEAR_REGS 305 call stackleak_erase 306 POP_REGS 307 #endif 308 .endm 309 310 .macro SAVE_AND_SET_GSBASE scratch_reg:req save_reg:req 311 rdgsbase \save_reg 312 GET_PERCPU_BASE \scratch_reg 313 wrgsbase \scratch_reg 314 .endm 315 316 #else /* CONFIG_X86_64 */ 317 # undef UNWIND_HINT_IRET_REGS 318 # define UNWIND_HINT_IRET_REGS 319 #endif /* !CONFIG_X86_64 */ 320 321 .macro STACKLEAK_ERASE 322 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK 323 call stackleak_erase 324 #endif 325 .endm 326 327 #ifdef CONFIG_SMP 328 329 /* 330 * CPU/node NR is loaded from the limit (size) field of a special segment 331 * descriptor entry in GDT. 332 */ 333 .macro LOAD_CPU_AND_NODE_SEG_LIMIT reg:req 334 movq $__CPUNODE_SEG, \reg 335 lsl \reg, \reg 336 .endm 337 338 /* 339 * Fetch the per-CPU GSBASE value for this processor and put it in @reg. 340 * We normally use %gs for accessing per-CPU data, but we are setting up 341 * %gs here and obviously can not use %gs itself to access per-CPU data. 342 * 343 * Do not use RDPID, because KVM loads guest's TSC_AUX on vm-entry and 344 * may not restore the host's value until the CPU returns to userspace. 345 * Thus the kernel would consume a guest's TSC_AUX if an NMI arrives 346 * while running KVM's run loop. 347 */ 348 .macro GET_PERCPU_BASE reg:req 349 LOAD_CPU_AND_NODE_SEG_LIMIT \reg 350 andq $VDSO_CPUNODE_MASK, \reg 351 movq __per_cpu_offset(, \reg, 8), \reg 352 .endm 353 354 #else 355 356 .macro GET_PERCPU_BASE reg:req 357 movq pcpu_unit_offsets(%rip), \reg 358 .endm 359 360 #endif /* CONFIG_SMP */ 361