| 1 | /* SPDX-License-Identifier: GPL-2.0-only */ |
| 2 | /* |
| 3 | * Copyright (C) 2012 Regents of the University of California |
| 4 | * Copyright (C) 2017 SiFive |
| 5 | */ |
| 6 | |
| 7 | #include <linux/init.h> |
| 8 | #include <linux/linkage.h> |
| 9 | |
| 10 | #include <asm/alternative-macros.h> |
| 11 | #include <asm/asm.h> |
| 12 | #include <asm/csr.h> |
| 13 | #include <asm/scs.h> |
| 14 | #include <asm/unistd.h> |
| 15 | #include <asm/page.h> |
| 16 | #include <asm/thread_info.h> |
| 17 | #include <asm/asm-offsets.h> |
| 18 | #include <asm/errata_list.h> |
| 19 | #include <linux/sizes.h> |
| 20 | |
| 21 | .section .irqentry.text, "ax" |
| 22 | |
| 23 | .macro new_vmalloc_check |
| 24 | REG_S a0, TASK_TI_A0(tp) |
| 25 | csrr a0, CSR_CAUSE |
| 26 | /* Exclude IRQs */ |
| 27 | blt a0, zero, .Lnew_vmalloc_restore_context_a0 |
| 28 | |
| 29 | REG_S a1, TASK_TI_A1(tp) |
| 30 | /* Only check new_vmalloc if we are in page/protection fault */ |
| 31 | li a1, EXC_LOAD_PAGE_FAULT |
| 32 | beq a0, a1, .Lnew_vmalloc_kernel_address |
| 33 | li a1, EXC_STORE_PAGE_FAULT |
| 34 | beq a0, a1, .Lnew_vmalloc_kernel_address |
| 35 | li a1, EXC_INST_PAGE_FAULT |
| 36 | bne a0, a1, .Lnew_vmalloc_restore_context_a1 |
| 37 | |
| 38 | .Lnew_vmalloc_kernel_address: |
| 39 | /* Is it a kernel address? */ |
| 40 | csrr a0, CSR_TVAL |
| 41 | bge a0, zero, .Lnew_vmalloc_restore_context_a1 |
| 42 | |
| 43 | /* Check if a new vmalloc mapping appeared that could explain the trap */ |
| 44 | REG_S a2, TASK_TI_A2(tp) |
| 45 | /* |
| 46 | * Computes: |
| 47 | * a0 = &new_vmalloc[BIT_WORD(cpu)] |
| 48 | * a1 = BIT_MASK(cpu) |
| 49 | */ |
| 50 | lw a2, TASK_TI_CPU(tp) |
| 51 | /* |
| 52 | * Compute the new_vmalloc element position: |
| 53 | * (cpu / 64) * 8 = (cpu >> 6) << 3 |
| 54 | */ |
| 55 | srli a1, a2, 6 |
| 56 | slli a1, a1, 3 |
| 57 | la a0, new_vmalloc |
| 58 | add a0, a0, a1 |
| 59 | /* |
| 60 | * Compute the bit position in the new_vmalloc element: |
| 61 | * bit_pos = cpu % 64 = cpu - (cpu / 64) * 64 = cpu - (cpu >> 6) << 6 |
| 62 | * = cpu - ((cpu >> 6) << 3) << 3 |
| 63 | */ |
| 64 | slli a1, a1, 3 |
| 65 | sub a1, a2, a1 |
| 66 | /* Compute the "get mask": 1 << bit_pos */ |
| 67 | li a2, 1 |
| 68 | sll a1, a2, a1 |
| 69 | |
| 70 | /* Check the value of new_vmalloc for this cpu */ |
| 71 | REG_L a2, 0(a0) |
| 72 | and a2, a2, a1 |
| 73 | beq a2, zero, .Lnew_vmalloc_restore_context |
| 74 | |
| 75 | /* Atomically reset the current cpu bit in new_vmalloc */ |
| 76 | amoxor.d a0, a1, (a0) |
| 77 | |
| 78 | /* Only emit a sfence.vma if the uarch caches invalid entries */ |
| 79 | ALTERNATIVE("sfence.vma" , "nop" , 0, RISCV_ISA_EXT_SVVPTC, 1) |
| 80 | |
| 81 | REG_L a0, TASK_TI_A0(tp) |
| 82 | REG_L a1, TASK_TI_A1(tp) |
| 83 | REG_L a2, TASK_TI_A2(tp) |
| 84 | csrw CSR_SCRATCH, x0 |
| 85 | sret |
| 86 | |
| 87 | .Lnew_vmalloc_restore_context: |
| 88 | REG_L a2, TASK_TI_A2(tp) |
| 89 | .Lnew_vmalloc_restore_context_a1: |
| 90 | REG_L a1, TASK_TI_A1(tp) |
| 91 | .Lnew_vmalloc_restore_context_a0: |
| 92 | REG_L a0, TASK_TI_A0(tp) |
| 93 | .endm |
| 94 | |
| 95 | |
| 96 | SYM_CODE_START(handle_exception) |
| 97 | /* |
| 98 | * If coming from userspace, preserve the user thread pointer and load |
| 99 | * the kernel thread pointer. If we came from the kernel, the scratch |
| 100 | * register will contain 0, and we should continue on the current TP. |
| 101 | */ |
| 102 | csrrw tp, CSR_SCRATCH, tp |
| 103 | bnez tp, .Lsave_context |
| 104 | |
| 105 | .Lrestore_kernel_tpsp: |
| 106 | csrr tp, CSR_SCRATCH |
| 107 | |
| 108 | #ifdef CONFIG_64BIT |
| 109 | /* |
| 110 | * The RISC-V kernel does not eagerly emit a sfence.vma after each |
| 111 | * new vmalloc mapping, which may result in exceptions: |
| 112 | * - if the uarch caches invalid entries, the new mapping would not be |
| 113 | * observed by the page table walker and an invalidation is needed. |
| 114 | * - if the uarch does not cache invalid entries, a reordered access |
| 115 | * could "miss" the new mapping and traps: in that case, we only need |
| 116 | * to retry the access, no sfence.vma is required. |
| 117 | */ |
| 118 | new_vmalloc_check |
| 119 | #endif |
| 120 | |
| 121 | REG_S sp, TASK_TI_KERNEL_SP(tp) |
| 122 | |
| 123 | #ifdef CONFIG_VMAP_STACK |
| 124 | addi sp, sp, -(PT_SIZE_ON_STACK) |
| 125 | srli sp, sp, THREAD_SHIFT |
| 126 | andi sp, sp, 0x1 |
| 127 | bnez sp, handle_kernel_stack_overflow |
| 128 | REG_L sp, TASK_TI_KERNEL_SP(tp) |
| 129 | #endif |
| 130 | |
| 131 | .Lsave_context: |
| 132 | REG_S sp, TASK_TI_USER_SP(tp) |
| 133 | REG_L sp, TASK_TI_KERNEL_SP(tp) |
| 134 | addi sp, sp, -(PT_SIZE_ON_STACK) |
| 135 | REG_S x1, PT_RA(sp) |
| 136 | REG_S x3, PT_GP(sp) |
| 137 | REG_S x5, PT_T0(sp) |
| 138 | save_from_x6_to_x31 |
| 139 | |
| 140 | /* |
| 141 | * Disable user-mode memory access as it should only be set in the |
| 142 | * actual user copy routines. |
| 143 | * |
| 144 | * Disable the FPU/Vector to detect illegal usage of floating point |
| 145 | * or vector in kernel space. |
| 146 | */ |
| 147 | li t0, SR_SUM | SR_FS_VS |
| 148 | |
| 149 | REG_L s0, TASK_TI_USER_SP(tp) |
| 150 | csrrc s1, CSR_STATUS, t0 |
| 151 | csrr s2, CSR_EPC |
| 152 | csrr s3, CSR_TVAL |
| 153 | csrr s4, CSR_CAUSE |
| 154 | csrr s5, CSR_SCRATCH |
| 155 | REG_S s0, PT_SP(sp) |
| 156 | REG_S s1, PT_STATUS(sp) |
| 157 | REG_S s2, PT_EPC(sp) |
| 158 | REG_S s3, PT_BADADDR(sp) |
| 159 | REG_S s4, PT_CAUSE(sp) |
| 160 | REG_S s5, PT_TP(sp) |
| 161 | |
| 162 | /* |
| 163 | * Set the scratch register to 0, so that if a recursive exception |
| 164 | * occurs, the exception vector knows it came from the kernel |
| 165 | */ |
| 166 | csrw CSR_SCRATCH, x0 |
| 167 | |
| 168 | /* Load the global pointer */ |
| 169 | load_global_pointer |
| 170 | |
| 171 | /* Load the kernel shadow call stack pointer if coming from userspace */ |
| 172 | scs_load_current_if_task_changed s5 |
| 173 | |
| 174 | #ifdef CONFIG_RISCV_ISA_V_PREEMPTIVE |
| 175 | move a0, sp |
| 176 | call riscv_v_context_nesting_start |
| 177 | #endif |
| 178 | move a0, sp /* pt_regs */ |
| 179 | |
| 180 | /* |
| 181 | * MSB of cause differentiates between |
| 182 | * interrupts and exceptions |
| 183 | */ |
| 184 | bge s4, zero, 1f |
| 185 | |
| 186 | /* Handle interrupts */ |
| 187 | call do_irq |
| 188 | j ret_from_exception |
| 189 | 1: |
| 190 | /* Handle other exceptions */ |
| 191 | slli t0, s4, RISCV_LGPTR |
| 192 | la t1, excp_vect_table |
| 193 | la t2, excp_vect_table_end |
| 194 | add t0, t1, t0 |
| 195 | /* Check if exception code lies within bounds */ |
| 196 | bgeu t0, t2, 3f |
| 197 | REG_L t1, 0(t0) |
| 198 | 2: jalr t1 |
| 199 | j ret_from_exception |
| 200 | 3: |
| 201 | |
| 202 | la t1, do_trap_unknown |
| 203 | j 2b |
| 204 | SYM_CODE_END(handle_exception) |
| 205 | ASM_NOKPROBE(handle_exception) |
| 206 | |
| 207 | /* |
| 208 | * The ret_from_exception must be called with interrupt disabled. Here is the |
| 209 | * caller list: |
| 210 | * - handle_exception |
| 211 | * - ret_from_fork |
| 212 | */ |
| 213 | SYM_CODE_START_NOALIGN(ret_from_exception) |
| 214 | REG_L s0, PT_STATUS(sp) |
| 215 | #ifdef CONFIG_RISCV_M_MODE |
| 216 | /* the MPP value is too large to be used as an immediate arg for addi */ |
| 217 | li t0, SR_MPP |
| 218 | and s0, s0, t0 |
| 219 | #else |
| 220 | andi s0, s0, SR_SPP |
| 221 | #endif |
| 222 | bnez s0, 1f |
| 223 | |
| 224 | #ifdef CONFIG_KSTACK_ERASE |
| 225 | call stackleak_erase_on_task_stack |
| 226 | #endif |
| 227 | |
| 228 | /* Save unwound kernel stack pointer in thread_info */ |
| 229 | addi s0, sp, PT_SIZE_ON_STACK |
| 230 | REG_S s0, TASK_TI_KERNEL_SP(tp) |
| 231 | |
| 232 | /* Save the kernel shadow call stack pointer */ |
| 233 | scs_save_current |
| 234 | |
| 235 | /* |
| 236 | * Save TP into the scratch register , so we can find the kernel data |
| 237 | * structures again. |
| 238 | */ |
| 239 | csrw CSR_SCRATCH, tp |
| 240 | 1: |
| 241 | #ifdef CONFIG_RISCV_ISA_V_PREEMPTIVE |
| 242 | move a0, sp |
| 243 | call riscv_v_context_nesting_end |
| 244 | #endif |
| 245 | REG_L a0, PT_STATUS(sp) |
| 246 | /* |
| 247 | * The current load reservation is effectively part of the processor's |
| 248 | * state, in the sense that load reservations cannot be shared between |
| 249 | * different hart contexts. We can't actually save and restore a load |
| 250 | * reservation, so instead here we clear any existing reservation -- |
| 251 | * it's always legal for implementations to clear load reservations at |
| 252 | * any point (as long as the forward progress guarantee is kept, but |
| 253 | * we'll ignore that here). |
| 254 | * |
| 255 | * Dangling load reservations can be the result of taking a trap in the |
| 256 | * middle of an LR/SC sequence, but can also be the result of a taken |
| 257 | * forward branch around an SC -- which is how we implement CAS. As a |
| 258 | * result we need to clear reservations between the last CAS and the |
| 259 | * jump back to the new context. While it is unlikely the store |
| 260 | * completes, implementations are allowed to expand reservations to be |
| 261 | * arbitrarily large. |
| 262 | */ |
| 263 | REG_L a2, PT_EPC(sp) |
| 264 | REG_SC x0, a2, PT_EPC(sp) |
| 265 | |
| 266 | csrw CSR_STATUS, a0 |
| 267 | csrw CSR_EPC, a2 |
| 268 | |
| 269 | REG_L x1, PT_RA(sp) |
| 270 | REG_L x3, PT_GP(sp) |
| 271 | REG_L x4, PT_TP(sp) |
| 272 | REG_L x5, PT_T0(sp) |
| 273 | restore_from_x6_to_x31 |
| 274 | |
| 275 | REG_L x2, PT_SP(sp) |
| 276 | |
| 277 | #ifdef CONFIG_RISCV_M_MODE |
| 278 | mret |
| 279 | #else |
| 280 | sret |
| 281 | #endif |
| 282 | SYM_INNER_LABEL(ret_from_exception_end, SYM_L_GLOBAL) |
| 283 | SYM_CODE_END(ret_from_exception) |
| 284 | ASM_NOKPROBE(ret_from_exception) |
| 285 | |
| 286 | #ifdef CONFIG_VMAP_STACK |
| 287 | SYM_CODE_START_LOCAL(handle_kernel_stack_overflow) |
| 288 | /* we reach here from kernel context, sscratch must be 0 */ |
| 289 | csrrw x31, CSR_SCRATCH, x31 |
| 290 | asm_per_cpu sp, overflow_stack, x31 |
| 291 | li x31, OVERFLOW_STACK_SIZE |
| 292 | add sp, sp, x31 |
| 293 | /* zero out x31 again and restore x31 */ |
| 294 | xor x31, x31, x31 |
| 295 | csrrw x31, CSR_SCRATCH, x31 |
| 296 | |
| 297 | addi sp, sp, -(PT_SIZE_ON_STACK) |
| 298 | |
| 299 | //save context to overflow stack |
| 300 | REG_S x1, PT_RA(sp) |
| 301 | REG_S x3, PT_GP(sp) |
| 302 | REG_S x5, PT_T0(sp) |
| 303 | save_from_x6_to_x31 |
| 304 | |
| 305 | REG_L s0, TASK_TI_KERNEL_SP(tp) |
| 306 | csrr s1, CSR_STATUS |
| 307 | csrr s2, CSR_EPC |
| 308 | csrr s3, CSR_TVAL |
| 309 | csrr s4, CSR_CAUSE |
| 310 | csrr s5, CSR_SCRATCH |
| 311 | REG_S s0, PT_SP(sp) |
| 312 | REG_S s1, PT_STATUS(sp) |
| 313 | REG_S s2, PT_EPC(sp) |
| 314 | REG_S s3, PT_BADADDR(sp) |
| 315 | REG_S s4, PT_CAUSE(sp) |
| 316 | REG_S s5, PT_TP(sp) |
| 317 | move a0, sp |
| 318 | tail handle_bad_stack |
| 319 | SYM_CODE_END(handle_kernel_stack_overflow) |
| 320 | ASM_NOKPROBE(handle_kernel_stack_overflow) |
| 321 | #endif |
| 322 | |
| 323 | SYM_CODE_START(ret_from_fork_kernel_asm) |
| 324 | call schedule_tail |
| 325 | move a0, s1 /* fn_arg */ |
| 326 | move a1, s0 /* fn */ |
| 327 | move a2, sp /* pt_regs */ |
| 328 | call ret_from_fork_kernel |
| 329 | j ret_from_exception |
| 330 | SYM_CODE_END(ret_from_fork_kernel_asm) |
| 331 | |
| 332 | SYM_CODE_START(ret_from_fork_user_asm) |
| 333 | call schedule_tail |
| 334 | move a0, sp /* pt_regs */ |
| 335 | call ret_from_fork_user |
| 336 | j ret_from_exception |
| 337 | SYM_CODE_END(ret_from_fork_user_asm) |
| 338 | |
| 339 | #ifdef CONFIG_IRQ_STACKS |
| 340 | /* |
| 341 | * void call_on_irq_stack(struct pt_regs *regs, |
| 342 | * void (*func)(struct pt_regs *)); |
| 343 | * |
| 344 | * Calls func(regs) using the per-CPU IRQ stack. |
| 345 | */ |
| 346 | SYM_FUNC_START(call_on_irq_stack) |
| 347 | /* Create a frame record to save ra and s0 (fp) */ |
| 348 | addi sp, sp, -STACKFRAME_SIZE_ON_STACK |
| 349 | REG_S ra, STACKFRAME_RA(sp) |
| 350 | REG_S s0, STACKFRAME_FP(sp) |
| 351 | addi s0, sp, STACKFRAME_SIZE_ON_STACK |
| 352 | |
| 353 | /* Switch to the per-CPU shadow call stack */ |
| 354 | scs_save_current |
| 355 | scs_load_irq_stack t0 |
| 356 | |
| 357 | /* Switch to the per-CPU IRQ stack and call the handler */ |
| 358 | load_per_cpu t0, irq_stack_ptr, t1 |
| 359 | li t1, IRQ_STACK_SIZE |
| 360 | add sp, t0, t1 |
| 361 | jalr a1 |
| 362 | |
| 363 | /* Switch back to the thread shadow call stack */ |
| 364 | scs_load_current |
| 365 | |
| 366 | /* Switch back to the thread stack and restore ra and s0 */ |
| 367 | addi sp, s0, -STACKFRAME_SIZE_ON_STACK |
| 368 | REG_L ra, STACKFRAME_RA(sp) |
| 369 | REG_L s0, STACKFRAME_FP(sp) |
| 370 | addi sp, sp, STACKFRAME_SIZE_ON_STACK |
| 371 | |
| 372 | ret |
| 373 | SYM_FUNC_END(call_on_irq_stack) |
| 374 | #endif /* CONFIG_IRQ_STACKS */ |
| 375 | |
| 376 | /* |
| 377 | * Integer register context switch |
| 378 | * The callee-saved registers must be saved and restored. |
| 379 | * |
| 380 | * a0: previous task_struct (must be preserved across the switch) |
| 381 | * a1: next task_struct |
| 382 | * |
| 383 | * The value of a0 and a1 must be preserved by this function, as that's how |
| 384 | * arguments are passed to schedule_tail. |
| 385 | */ |
| 386 | SYM_FUNC_START(__switch_to) |
| 387 | /* Save context into prev->thread */ |
| 388 | li a4, TASK_THREAD_RA |
| 389 | add a3, a0, a4 |
| 390 | add a4, a1, a4 |
| 391 | REG_S ra, TASK_THREAD_RA_RA(a3) |
| 392 | REG_S sp, TASK_THREAD_SP_RA(a3) |
| 393 | REG_S s0, TASK_THREAD_S0_RA(a3) |
| 394 | REG_S s1, TASK_THREAD_S1_RA(a3) |
| 395 | REG_S s2, TASK_THREAD_S2_RA(a3) |
| 396 | REG_S s3, TASK_THREAD_S3_RA(a3) |
| 397 | REG_S s4, TASK_THREAD_S4_RA(a3) |
| 398 | REG_S s5, TASK_THREAD_S5_RA(a3) |
| 399 | REG_S s6, TASK_THREAD_S6_RA(a3) |
| 400 | REG_S s7, TASK_THREAD_S7_RA(a3) |
| 401 | REG_S s8, TASK_THREAD_S8_RA(a3) |
| 402 | REG_S s9, TASK_THREAD_S9_RA(a3) |
| 403 | REG_S s10, TASK_THREAD_S10_RA(a3) |
| 404 | REG_S s11, TASK_THREAD_S11_RA(a3) |
| 405 | |
| 406 | /* save the user space access flag */ |
| 407 | csrr s0, CSR_STATUS |
| 408 | REG_S s0, TASK_THREAD_SUM_RA(a3) |
| 409 | |
| 410 | /* Save the kernel shadow call stack pointer */ |
| 411 | scs_save_current |
| 412 | /* Restore context from next->thread */ |
| 413 | REG_L s0, TASK_THREAD_SUM_RA(a4) |
| 414 | li s1, SR_SUM |
| 415 | and s0, s0, s1 |
| 416 | csrs CSR_STATUS, s0 |
| 417 | REG_L ra, TASK_THREAD_RA_RA(a4) |
| 418 | REG_L sp, TASK_THREAD_SP_RA(a4) |
| 419 | REG_L s0, TASK_THREAD_S0_RA(a4) |
| 420 | REG_L s1, TASK_THREAD_S1_RA(a4) |
| 421 | REG_L s2, TASK_THREAD_S2_RA(a4) |
| 422 | REG_L s3, TASK_THREAD_S3_RA(a4) |
| 423 | REG_L s4, TASK_THREAD_S4_RA(a4) |
| 424 | REG_L s5, TASK_THREAD_S5_RA(a4) |
| 425 | REG_L s6, TASK_THREAD_S6_RA(a4) |
| 426 | REG_L s7, TASK_THREAD_S7_RA(a4) |
| 427 | REG_L s8, TASK_THREAD_S8_RA(a4) |
| 428 | REG_L s9, TASK_THREAD_S9_RA(a4) |
| 429 | REG_L s10, TASK_THREAD_S10_RA(a4) |
| 430 | REG_L s11, TASK_THREAD_S11_RA(a4) |
| 431 | /* The offset of thread_info in task_struct is zero. */ |
| 432 | move tp, a1 |
| 433 | /* Switch to the next shadow call stack */ |
| 434 | scs_load_current |
| 435 | ret |
| 436 | SYM_FUNC_END(__switch_to) |
| 437 | |
| 438 | #ifndef CONFIG_MMU |
| 439 | #define do_page_fault do_trap_unknown |
| 440 | #endif |
| 441 | |
| 442 | .section ".rodata" |
| 443 | .align LGREG |
| 444 | /* Exception vector table */ |
| 445 | SYM_DATA_START_LOCAL(excp_vect_table) |
| 446 | RISCV_PTR do_trap_insn_misaligned |
| 447 | ALT_INSN_FAULT(RISCV_PTR do_trap_insn_fault) |
| 448 | RISCV_PTR do_trap_insn_illegal |
| 449 | RISCV_PTR do_trap_break |
| 450 | RISCV_PTR do_trap_load_misaligned |
| 451 | RISCV_PTR do_trap_load_fault |
| 452 | RISCV_PTR do_trap_store_misaligned |
| 453 | RISCV_PTR do_trap_store_fault |
| 454 | RISCV_PTR do_trap_ecall_u /* system call */ |
| 455 | RISCV_PTR do_trap_ecall_s |
| 456 | RISCV_PTR do_trap_unknown |
| 457 | RISCV_PTR do_trap_ecall_m |
| 458 | /* instruction page fault */ |
| 459 | ALT_PAGE_FAULT(RISCV_PTR do_page_fault) |
| 460 | RISCV_PTR do_page_fault /* load page fault */ |
| 461 | RISCV_PTR do_trap_unknown |
| 462 | RISCV_PTR do_page_fault /* store page fault */ |
| 463 | SYM_DATA_END_LABEL(excp_vect_table, SYM_L_LOCAL, excp_vect_table_end) |
| 464 | |
| 465 | #ifndef CONFIG_MMU |
| 466 | SYM_DATA_START(__user_rt_sigreturn) |
| 467 | li a7, __NR_rt_sigreturn |
| 468 | ecall |
| 469 | SYM_DATA_END(__user_rt_sigreturn) |
| 470 | #endif |
| 471 | |