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 xorl %edx, %edx /* nospec dx */ 118 pushq %rcx /* pt_regs->cx */ 119 xorl %ecx, %ecx /* nospec cx */ 120 pushq \rax /* pt_regs->ax */ 121 pushq %r8 /* pt_regs->r8 */ 122 xorl %r8d, %r8d /* nospec r8 */ 123 pushq %r9 /* pt_regs->r9 */ 124 xorl %r9d, %r9d /* nospec r9 */ 125 pushq %r10 /* pt_regs->r10 */ 126 xorl %r10d, %r10d /* nospec r10 */ 127 pushq %r11 /* pt_regs->r11 */ 128 xorl %r11d, %r11d /* nospec r11*/ 129 pushq %rbx /* pt_regs->rbx */ 130 xorl %ebx, %ebx /* nospec rbx*/ 131 pushq %rbp /* pt_regs->rbp */ 132 xorl %ebp, %ebp /* nospec rbp*/ 133 pushq %r12 /* pt_regs->r12 */ 134 xorl %r12d, %r12d /* nospec r12*/ 135 pushq %r13 /* pt_regs->r13 */ 136 xorl %r13d, %r13d /* nospec r13*/ 137 pushq %r14 /* pt_regs->r14 */ 138 xorl %r14d, %r14d /* nospec r14*/ 139 pushq %r15 /* pt_regs->r15 */ 140 xorl %r15d, %r15d /* nospec r15*/ 141 UNWIND_HINT_REGS 142 .if \save_ret 143 pushq %rsi /* return address on top of stack */ 144 .endif 145 .endm 146 147 .macro POP_REGS pop_rdi=1 skip_r11rcx=0 148 popq %r15 149 popq %r14 150 popq %r13 151 popq %r12 152 popq %rbp 153 popq %rbx 154 .if \skip_r11rcx 155 popq %rsi 156 .else 157 popq %r11 158 .endif 159 popq %r10 160 popq %r9 161 popq %r8 162 popq %rax 163 .if \skip_r11rcx 164 popq %rsi 165 .else 166 popq %rcx 167 .endif 168 popq %rdx 169 popq %rsi 170 .if \pop_rdi 171 popq %rdi 172 .endif 173 .endm 174 175 /* 176 * This is a sneaky trick to help the unwinder find pt_regs on the stack. The 177 * frame pointer is replaced with an encoded pointer to pt_regs. The encoding 178 * is just setting the LSB, which makes it an invalid stack address and is also 179 * a signal to the unwinder that it's a pt_regs pointer in disguise. 180 * 181 * NOTE: This macro must be used *after* PUSH_AND_CLEAR_REGS because it corrupts 182 * the original rbp. 183 */ 184 .macro ENCODE_FRAME_POINTER ptregs_offset=0 185 #ifdef CONFIG_FRAME_POINTER 186 leaq 1+\ptregs_offset(%rsp), %rbp 187 #endif 188 .endm 189 190 #ifdef CONFIG_PAGE_TABLE_ISOLATION 191 192 /* 193 * PAGE_TABLE_ISOLATION PGDs are 8k. Flip bit 12 to switch between the two 194 * halves: 195 */ 196 #define PTI_USER_PGTABLE_BIT PAGE_SHIFT 197 #define PTI_USER_PGTABLE_MASK (1 << PTI_USER_PGTABLE_BIT) 198 #define PTI_USER_PCID_BIT X86_CR3_PTI_PCID_USER_BIT 199 #define PTI_USER_PCID_MASK (1 << PTI_USER_PCID_BIT) 200 #define PTI_USER_PGTABLE_AND_PCID_MASK (PTI_USER_PCID_MASK | PTI_USER_PGTABLE_MASK) 201 202 .macro SET_NOFLUSH_BIT reg:req 203 bts $X86_CR3_PCID_NOFLUSH_BIT, \reg 204 .endm 205 206 .macro ADJUST_KERNEL_CR3 reg:req 207 ALTERNATIVE "", "SET_NOFLUSH_BIT \reg", X86_FEATURE_PCID 208 /* Clear PCID and "PAGE_TABLE_ISOLATION bit", point CR3 at kernel pagetables: */ 209 andq $(~PTI_USER_PGTABLE_AND_PCID_MASK), \reg 210 .endm 211 212 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req 213 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 214 mov %cr3, \scratch_reg 215 ADJUST_KERNEL_CR3 \scratch_reg 216 mov \scratch_reg, %cr3 217 .Lend_\@: 218 .endm 219 220 #define THIS_CPU_user_pcid_flush_mask \ 221 PER_CPU_VAR(cpu_tlbstate) + TLB_STATE_user_pcid_flush_mask 222 223 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req 224 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 225 mov %cr3, \scratch_reg 226 227 ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID 228 229 /* 230 * Test if the ASID needs a flush. 231 */ 232 movq \scratch_reg, \scratch_reg2 233 andq $(0x7FF), \scratch_reg /* mask ASID */ 234 bt \scratch_reg, THIS_CPU_user_pcid_flush_mask 235 jnc .Lnoflush_\@ 236 237 /* Flush needed, clear the bit */ 238 btr \scratch_reg, THIS_CPU_user_pcid_flush_mask 239 movq \scratch_reg2, \scratch_reg 240 jmp .Lwrcr3_pcid_\@ 241 242 .Lnoflush_\@: 243 movq \scratch_reg2, \scratch_reg 244 SET_NOFLUSH_BIT \scratch_reg 245 246 .Lwrcr3_pcid_\@: 247 /* Flip the ASID to the user version */ 248 orq $(PTI_USER_PCID_MASK), \scratch_reg 249 250 .Lwrcr3_\@: 251 /* Flip the PGD to the user version */ 252 orq $(PTI_USER_PGTABLE_MASK), \scratch_reg 253 mov \scratch_reg, %cr3 254 .Lend_\@: 255 .endm 256 257 .macro SWITCH_TO_USER_CR3_STACK scratch_reg:req 258 pushq %rax 259 SWITCH_TO_USER_CR3_NOSTACK scratch_reg=\scratch_reg scratch_reg2=%rax 260 popq %rax 261 .endm 262 263 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req 264 ALTERNATIVE "jmp .Ldone_\@", "", X86_FEATURE_PTI 265 movq %cr3, \scratch_reg 266 movq \scratch_reg, \save_reg 267 /* 268 * Test the user pagetable bit. If set, then the user page tables 269 * are active. If clear CR3 already has the kernel page table 270 * active. 271 */ 272 bt $PTI_USER_PGTABLE_BIT, \scratch_reg 273 jnc .Ldone_\@ 274 275 ADJUST_KERNEL_CR3 \scratch_reg 276 movq \scratch_reg, %cr3 277 278 .Ldone_\@: 279 .endm 280 281 .macro RESTORE_CR3 scratch_reg:req save_reg:req 282 ALTERNATIVE "jmp .Lend_\@", "", X86_FEATURE_PTI 283 284 ALTERNATIVE "jmp .Lwrcr3_\@", "", X86_FEATURE_PCID 285 286 /* 287 * KERNEL pages can always resume with NOFLUSH as we do 288 * explicit flushes. 289 */ 290 bt $PTI_USER_PGTABLE_BIT, \save_reg 291 jnc .Lnoflush_\@ 292 293 /* 294 * Check if there's a pending flush for the user ASID we're 295 * about to set. 296 */ 297 movq \save_reg, \scratch_reg 298 andq $(0x7FF), \scratch_reg 299 bt \scratch_reg, THIS_CPU_user_pcid_flush_mask 300 jnc .Lnoflush_\@ 301 302 btr \scratch_reg, THIS_CPU_user_pcid_flush_mask 303 jmp .Lwrcr3_\@ 304 305 .Lnoflush_\@: 306 SET_NOFLUSH_BIT \save_reg 307 308 .Lwrcr3_\@: 309 /* 310 * The CR3 write could be avoided when not changing its value, 311 * but would require a CR3 read *and* a scratch register. 312 */ 313 movq \save_reg, %cr3 314 .Lend_\@: 315 .endm 316 317 #else /* CONFIG_PAGE_TABLE_ISOLATION=n: */ 318 319 .macro SWITCH_TO_KERNEL_CR3 scratch_reg:req 320 .endm 321 .macro SWITCH_TO_USER_CR3_NOSTACK scratch_reg:req scratch_reg2:req 322 .endm 323 .macro SWITCH_TO_USER_CR3_STACK scratch_reg:req 324 .endm 325 .macro SAVE_AND_SWITCH_TO_KERNEL_CR3 scratch_reg:req save_reg:req 326 .endm 327 .macro RESTORE_CR3 scratch_reg:req save_reg:req 328 .endm 329 330 #endif 331 332 #endif /* CONFIG_X86_64 */ 333 334 /* 335 * This does 'call enter_from_user_mode' unless we can avoid it based on 336 * kernel config or using the static jump infrastructure. 337 */ 338 .macro CALL_enter_from_user_mode 339 #ifdef CONFIG_CONTEXT_TRACKING 340 #ifdef HAVE_JUMP_LABEL 341 STATIC_JUMP_IF_FALSE .Lafter_call_\@, context_tracking_enabled, def=0 342 #endif 343 call enter_from_user_mode 344 .Lafter_call_\@: 345 #endif 346 .endm 347