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 10 /* 11 12 x86 function call convention, 64-bit: 13 ------------------------------------- 14 arguments | callee-saved | extra caller-saved | return 15 [callee-clobbered] | | [callee-clobbered] | 16 --------------------------------------------------------------------------- 17 rdi rsi rdx rcx r8-9 | rbx rbp [*] r12-15 | r10-11 | rax, rdx [**] 18 19 ( rsp is obviously invariant across normal function calls. (gcc can 'merge' 20 functions when it sees tail-call optimization possibilities) rflags is 21 clobbered. Leftover arguments are passed over the stack frame.) 22 23 [*] In the frame-pointers case rbp is fixed to the stack frame. 24 25 [**] for struct return values wider than 64 bits the return convention is a 26 bit more complex: up to 128 bits width we return small structures 27 straight in rax, rdx. For structures larger than that (3 words or 28 larger) the caller puts a pointer to an on-stack return struct 29 [allocated in the caller's stack frame] into the first argument - i.e. 30 into rdi. All other arguments shift up by one in this case. 31 Fortunately this case is rare in the kernel. 32 33 For 32-bit we have the following conventions - kernel is built with 34 -mregparm=3 and -freg-struct-return: 35 36 x86 function calling convention, 32-bit: 37 ---------------------------------------- 38 arguments | callee-saved | extra caller-saved | return 39 [callee-clobbered] | | [callee-clobbered] | 40 ------------------------------------------------------------------------- 41 eax edx ecx | ebx edi esi ebp [*] | <none> | eax, edx [**] 42 43 ( here too esp is obviously invariant across normal function calls. eflags 44 is clobbered. Leftover arguments are passed over the stack frame. ) 45 46 [*] In the frame-pointers case ebp is fixed to the stack frame. 47 48 [**] We build with -freg-struct-return, which on 32-bit means similar 49 semantics as on 64-bit: edx can be used for a second return value 50 (i.e. covering integer and structure sizes up to 64 bits) - after that 51 it gets more complex and more expensive: 3-word or larger struct returns 52 get done in the caller's frame and the pointer to the return struct goes 53 into regparm0, i.e. eax - the other arguments shift up and the 54 function's register parameters degenerate to regparm=2 in essence. 55 56 */ 57 58 #ifdef CONFIG_X86_64 59 60 /* 61 * 64-bit system call stack frame layout defines and helpers, 62 * for assembly code: 63 */ 64 65 /* The layout forms the "struct pt_regs" on the stack: */ 66 /* 67 * C ABI says these regs are callee-preserved. They aren't saved on kernel entry 68 * unless syscall needs a complete, fully filled "struct pt_regs". 69 */ 70 #define R15 0*8 71 #define R14 1*8 72 #define R13 2*8 73 #define R12 3*8 74 #define RBP 4*8 75 #define RBX 5*8 76 /* These regs are callee-clobbered. Always saved on kernel entry. */ 77 #define R11 6*8 78 #define R10 7*8 79 #define R9 8*8 80 #define R8 9*8 81 #define RAX 10*8 82 #define RCX 11*8 83 #define RDX 12*8 84 #define RSI 13*8 85 #define RDI 14*8 86 /* 87 * On syscall entry, this is syscall#. On CPU exception, this is error code. 88 * On hw interrupt, it's IRQ number: 89 */ 90 #define ORIG_RAX 15*8 91 /* Return frame for iretq */ 92 #define RIP 16*8 93 #define CS 17*8 94 #define EFLAGS 18*8 95 #define RSP 19*8 96 #define SS 20*8 97 98 #define SIZEOF_PTREGS 21*8 99 100 .macro PUSH_AND_CLEAR_REGS rdx=%rdx rax=%rax save_ret=0 101 /* 102 * Push registers and sanitize registers of values that a 103 * speculation attack might otherwise want to exploit. The 104 * lower registers are likely clobbered well before they 105 * could be put to use in a speculative execution gadget. 106 * Interleave XOR with PUSH for better uop scheduling: 107 */ 108 .if \save_ret 109 pushq %rsi /* pt_regs->si */ 110 movq 8(%rsp), %rsi /* temporarily store the return address in %rsi */ 111 movq %rdi, 8(%rsp) /* pt_regs->di (overwriting original return address) */ 112 .else 113 pushq %rdi /* pt_regs->di */ 114 pushq %rsi /* pt_regs->si */ 115 .endif 116 pushq \rdx /* pt_regs->dx */ 117 pushq %rcx /* pt_regs->cx */ 118 pushq \rax /* pt_regs->ax */ 119 pushq %r8 /* pt_regs->r8 */ 120 xorl %r8d, %r8d /* nospec r8 */ 121 pushq %r9 /* pt_regs->r9 */ 122 xorl %r9d, %r9d /* nospec r9 */ 123 pushq %r10 /* pt_regs->r10 */ 124 xorl %r10d, %r10d /* nospec r10 */ 125 pushq %r11 /* pt_regs->r11 */ 126 xorl %r11d, %r11d /* nospec r11*/ 127 pushq %rbx /* pt_regs->rbx */ 128 xorl %ebx, %ebx /* nospec rbx*/ 129 pushq %rbp /* pt_regs->rbp */ 130 xorl %ebp, %ebp /* nospec rbp*/ 131 pushq %r12 /* pt_regs->r12 */ 132 xorl %r12d, %r12d /* nospec r12*/ 133 pushq %r13 /* pt_regs->r13 */ 134 xorl %r13d, %r13d /* nospec r13*/ 135 pushq %r14 /* pt_regs->r14 */ 136 xorl %r14d, %r14d /* nospec r14*/ 137 pushq %r15 /* pt_regs->r15 */ 138 xorl %r15d, %r15d /* nospec r15*/ 139 UNWIND_HINT_REGS 140 .if \save_ret 141 pushq %rsi /* return address on top of stack */ 142 .endif 143 .endm 144 145 .macro POP_REGS pop_rdi=1 skip_r11rcx=0 146 popq %r15 147 popq %r14 148 popq %r13 149 popq %r12 150 popq %rbp 151 popq %rbx 152 .if \skip_r11rcx 153 popq %rsi 154 .else 155 popq %r11 156 .endif 157 popq %r10 158 popq %r9 159 popq %r8 160 popq %rax 161 .if \skip_r11rcx 162 popq %rsi 163 .else 164 popq %rcx 165 .endif 166 popq %rdx 167 popq %rsi 168 .if \pop_rdi 169 popq %rdi 170 .endif 171 .endm 172 173 /* 174 * This is a sneaky trick to help the unwinder find pt_regs on the stack. The 175 * frame pointer is replaced with an encoded pointer to pt_regs. The encoding 176 * is just setting the LSB, which makes it an invalid stack address and is also 177 * a signal to the unwinder that it's a pt_regs pointer in disguise. 178 * 179 * NOTE: This macro must be used *after* PUSH_AND_CLEAR_REGS because it corrupts 180 * the original rbp. 181 */ 182 .macro ENCODE_FRAME_POINTER ptregs_offset=0 183 #ifdef CONFIG_FRAME_POINTER 184 leaq 1+\ptregs_offset(%rsp), %rbp 185 #endif 186 .endm 187 188 #ifdef CONFIG_PAGE_TABLE_ISOLATION 189 190 /* 191 * PAGE_TABLE_ISOLATION PGDs are 8k. Flip bit 12 to switch between the two 192 * halves: 193 */ 194 #define PTI_USER_PGTABLE_BIT PAGE_SHIFT 195 #define PTI_USER_PGTABLE_MASK (1 << PTI_USER_PGTABLE_BIT) 196 #define PTI_USER_PCID_BIT X86_CR3_PTI_PCID_USER_BIT 197 #define PTI_USER_PCID_MASK (1 << PTI_USER_PCID_BIT) 198 #define PTI_USER_PGTABLE_AND_PCID_MASK (PTI_USER_PCID_MASK | PTI_USER_PGTABLE_MASK) 199 200 .macro SET_NOFLUSH_BIT reg:req 201 bts $X86_CR3_PCID_NOFLUSH_BIT, \reg 202 .endm 203 204 .macro ADJUST_KERNEL_CR3 reg:req 205 ALTERNATIVE "", "SET_NOFLUSH_BIT \reg", X86_FEATURE_PCID 206 /* Clear PCID and "PAGE_TABLE_ISOLATION bit", point CR3 at kernel pagetables: */ 207 andq $(~PTI_USER_PGTABLE_AND_PCID_MASK), \reg 208 .endm 209 210 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req 211 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 212 mov %cr3, \scratch_reg 213 ADJUST_KERNEL_CR3 \scratch_reg 214 mov \scratch_reg, %cr3 215 .Lend_\@: 216 .endm 217 218 #define THIS_CPU_user_pcid_flush_mask \ 219 PER_CPU_VAR(cpu_tlbstate) + TLB_STATE_user_pcid_flush_mask 220 221 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req 222 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 223 mov %cr3, \scratch_reg 224 225 ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID 226 227 /* 228 * Test if the ASID needs a flush. 229 */ 230 movq \scratch_reg, \scratch_reg2 231 andq $(0x7FF), \scratch_reg /* mask ASID */ 232 bt \scratch_reg, THIS_CPU_user_pcid_flush_mask 233 jnc .Lnoflush_\@ 234 235 /* Flush needed, clear the bit */ 236 btr \scratch_reg, THIS_CPU_user_pcid_flush_mask 237 movq \scratch_reg2, \scratch_reg 238 jmp .Lwrcr3_pcid_\@ 239 240 .Lnoflush_\@: 241 movq \scratch_reg2, \scratch_reg 242 SET_NOFLUSH_BIT \scratch_reg 243 244 .Lwrcr3_pcid_\@: 245 /* Flip the ASID to the user version */ 246 orq $(PTI_USER_PCID_MASK), \scratch_reg 247 248 .Lwrcr3_\@: 249 /* Flip the PGD to the user version */ 250 orq $(PTI_USER_PGTABLE_MASK), \scratch_reg 251 mov \scratch_reg, %cr3 252 .Lend_\@: 253 .endm 254 255 .macro SWITCH_TO_USER_CR3_STACK scratch_reg:req 256 pushq %rax 257 SWITCH_TO_USER_CR3_NOSTACK scratch_reg=\scratch_reg scratch_reg2=%rax 258 popq %rax 259 .endm 260 261 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req 262 ALTERNATIVE "jmp .Ldone_\@", "", X86_FEATURE_PTI 263 movq %cr3, \scratch_reg 264 movq \scratch_reg, \save_reg 265 /* 266 * Test the user pagetable bit. If set, then the user page tables 267 * are active. If clear CR3 already has the kernel page table 268 * active. 269 */ 270 bt $PTI_USER_PGTABLE_BIT, \scratch_reg 271 jnc .Ldone_\@ 272 273 ADJUST_KERNEL_CR3 \scratch_reg 274 movq \scratch_reg, %cr3 275 276 .Ldone_\@: 277 .endm 278 279 .macro RESTORE_CR3 scratch_reg:req save_reg:req 280 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 281 282 ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID 283 284 /* 285 * KERNEL pages can always resume with NOFLUSH as we do 286 * explicit flushes. 287 */ 288 bt $PTI_USER_PGTABLE_BIT, \save_reg 289 jnc .Lnoflush_\@ 290 291 /* 292 * Check if there's a pending flush for the user ASID we're 293 * about to set. 294 */ 295 movq \save_reg, \scratch_reg 296 andq $(0x7FF), \scratch_reg 297 bt \scratch_reg, THIS_CPU_user_pcid_flush_mask 298 jnc .Lnoflush_\@ 299 300 btr \scratch_reg, THIS_CPU_user_pcid_flush_mask 301 jmp .Lwrcr3_\@ 302 303 .Lnoflush_\@: 304 SET_NOFLUSH_BIT \save_reg 305 306 .Lwrcr3_\@: 307 /* 308 * The CR3 write could be avoided when not changing its value, 309 * but would require a CR3 read *and* a scratch register. 310 */ 311 movq \save_reg, %cr3 312 .Lend_\@: 313 .endm 314 315 #else /* CONFIG_PAGE_TABLE_ISOLATION=n: */ 316 317 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req 318 .endm 319 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req 320 .endm 321 .macro SWITCH_TO_USER_CR3_STACK scratch_reg:req 322 .endm 323 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req 324 .endm 325 .macro RESTORE_CR3 scratch_reg:req save_reg:req 326 .endm 327 328 #endif 329 330 #endif /* CONFIG_X86_64 */ 331 332 /* 333 * This does 'call enter_from_user_mode' unless we can avoid it based on 334 * kernel config or using the static jump infrastructure. 335 */ 336 .macro CALL_enter_from_user_mode 337 #ifdef CONFIG_CONTEXT_TRACKING 338 #ifdef HAVE_JUMP_LABEL 339 STATIC_JUMP_IF_FALSE .Lafter_call_\@, context_tracking_enabled, def=0 340 #endif 341 call enter_from_user_mode 342 .Lafter_call_\@: 343 #endif 344 .endm 345