/* * SPDX-License-Identifier: CDDL 1.0 * * Copyright (c) 2022 Christos Margiolis * Copyright (c) 2022 Mark Johnston * Copyright (c) 2023 The FreeBSD Foundation * * Portions of this software were developed by Christos Margiolis * under sponsorship from the FreeBSD Foundation. */ #include #include #include #include "kinst.h" DPCPU_DEFINE_STATIC(struct kinst_cpu_state, kinst_state); static int kinst_emulate(struct trapframe *frame, const struct kinst_probe *kp) { kinst_patchval_t instr = kp->kp_savedval; uint64_t imm; uint8_t cond, reg, bitpos; bool res; if (((instr >> 24) & 0x1f) == 0b10000) { /* adr/adrp */ reg = instr & 0x1f; imm = (instr >> 29) & 0x3; imm |= ((instr >> 5) & 0x0007ffff) << 2; if (((instr >> 31) & 0x1) == 0) { /* adr */ if (imm & 0x0000000000100000) imm |= 0xfffffffffff00000; frame->tf_x[reg] = frame->tf_elr + imm; } else { /* adrp */ imm <<= 12; if (imm & 0x0000000100000000) imm |= 0xffffffff00000000; frame->tf_x[reg] = (frame->tf_elr & ~0xfff) + imm; } frame->tf_elr += INSN_SIZE; } else if (((instr >> 26) & 0x3f) == 0b000101) { /* b */ imm = instr & 0x03ffffff; if (imm & 0x0000000002000000) imm |= 0xfffffffffe000000; frame->tf_elr += imm << 2; } else if (((instr >> 24) & 0xff) == 0b01010100) { /* b.cond */ imm = (instr >> 5) & 0x0007ffff; if (imm & 0x0000000000040000) imm |= 0xfffffffffffc0000; cond = instr & 0xf; switch ((cond >> 1) & 0x7) { case 0b000: /* eq/ne */ res = (frame->tf_spsr & PSR_Z) != 0; break; case 0b001: /* cs/cc */ res = (frame->tf_spsr & PSR_C) != 0; break; case 0b010: /* mi/pl */ res = (frame->tf_spsr & PSR_N) != 0; break; case 0b011: /* vs/vc */ res = (frame->tf_spsr & PSR_V) != 0; break; case 0b100: /* hi/ls */ res = ((frame->tf_spsr & PSR_C) != 0) && ((frame->tf_spsr & PSR_Z) == 0); break; case 0b101: /* ge/lt */ res = ((frame->tf_spsr & PSR_N) != 0) == ((frame->tf_spsr & PSR_V) != 0); break; case 0b110: /* gt/le */ res = ((frame->tf_spsr & PSR_Z) == 0) && (((frame->tf_spsr & PSR_N) != 0) == ((frame->tf_spsr & PSR_V) != 0)); break; case 0b111: /* al */ res = 1; break; } if ((cond & 0x1) && cond != 0b1111) res = !res; if (res) frame->tf_elr += imm << 2; else frame->tf_elr += INSN_SIZE; } else if (((instr >> 26) & 0x3f) == 0b100101) { /* bl */ imm = instr & 0x03ffffff; if (imm & 0x0000000002000000) imm |= 0xfffffffffe000000; frame->tf_lr = frame->tf_elr + INSN_SIZE; frame->tf_elr += imm << 2; } else if (((instr >> 25) & 0x3f) == 0b011010) { /* cbnz/cbz */ cond = (instr >> 24) & 0x1; reg = instr & 0x1f; imm = (instr >> 5) & 0x0007ffff; if (imm & 0x0000000000040000) imm |= 0xfffffffffffc0000; if (cond == 1 && frame->tf_x[reg] != 0) /* cbnz */ frame->tf_elr += imm << 2; else if (cond == 0 && frame->tf_x[reg] == 0) /* cbz */ frame->tf_elr += imm << 2; else frame->tf_elr += INSN_SIZE; } else if (((instr >> 25) & 0x3f) == 0b011011) { /* tbnz/tbz */ cond = (instr >> 24) & 0x1; reg = instr & 0x1f; bitpos = (instr >> 19) & 0x1f; bitpos |= ((instr >> 31) & 0x1) << 5; imm = (instr >> 5) & 0x3fff; if (imm & 0x0000000000002000) imm |= 0xffffffffffffe000; if (cond == 1 && (frame->tf_x[reg] & (1 << bitpos)) != 0) /* tbnz */ frame->tf_elr += imm << 2; else if (cond == 0 && (frame->tf_x[reg] & (1 << bitpos)) == 0) /* tbz */ frame->tf_elr += imm << 2; else frame->tf_elr += INSN_SIZE; } return (0); } static int kinst_jump_next_instr(struct trapframe *frame, const struct kinst_probe *kp) { frame->tf_elr = (register_t)((const uint8_t *)kp->kp_patchpoint + INSN_SIZE); return (0); } static void kinst_trampoline_populate(struct kinst_probe *kp) { static uint32_t bpt = KINST_PATCHVAL; kinst_memcpy(kp->kp_tramp, &kp->kp_savedval, INSN_SIZE); kinst_memcpy(&kp->kp_tramp[INSN_SIZE], &bpt, INSN_SIZE); cpu_icache_sync_range((vm_offset_t)kp->kp_tramp, (vm_size_t)KINST_TRAMP_SIZE); } /* * There are two ways by which an instruction is traced: * * - By using the trampoline. * - By emulating it in software (see kinst_emulate()). * * The trampoline is used for instructions that can be copied and executed * as-is without additional modification. However, instructions that use * PC-relative addressing have to be emulated, because ARM64 doesn't allow * encoding of large displacements in a single instruction, and since we cannot * clobber a register in order to encode the two-instruction sequence needed to * create large displacements, we cannot use the trampoline at all. * Fortunately, the instructions are simple enough to be emulated in just a few * lines of code. * * The problem discussed above also means that, unlike amd64, we cannot encode * a far-jump back from the trampoline to the next instruction. The mechanism * employed to achieve this functionality, is to use a breakpoint instead of a * jump after the copied instruction. This breakpoint is detected and handled * by kinst_invop(), which performs the jump back to the next instruction * manually (see kinst_jump_next_instr()). */ int kinst_invop(uintptr_t addr, struct trapframe *frame, uintptr_t scratch) { solaris_cpu_t *cpu; struct kinst_cpu_state *ks; const struct kinst_probe *kp; ks = DPCPU_PTR(kinst_state); /* * Detect if the breakpoint was triggered by the trampoline, and * manually set the PC to the next instruction. */ if (ks->state == KINST_PROBE_FIRED && addr == (uintptr_t)(ks->kp->kp_tramp + INSN_SIZE)) { /* * Restore interrupts if they were enabled prior to the first * breakpoint. */ if ((ks->status & PSR_I) == 0) frame->tf_spsr &= ~PSR_I; ks->state = KINST_PROBE_ARMED; return (kinst_jump_next_instr(frame, ks->kp)); } LIST_FOREACH(kp, KINST_GETPROBE(addr), kp_hashnext) { if ((uintptr_t)kp->kp_patchpoint == addr) break; } if (kp == NULL) return (0); cpu = &solaris_cpu[curcpu]; cpu->cpu_dtrace_caller = addr; dtrace_probe(kp->kp_id, 0, 0, 0, 0, 0); cpu->cpu_dtrace_caller = 0; if (kp->kp_md.emulate) return (kinst_emulate(frame, kp)); ks->state = KINST_PROBE_FIRED; ks->kp = kp; /* * Cache the current SPSR and clear interrupts for the duration * of the double breakpoint. */ ks->status = frame->tf_spsr; frame->tf_spsr |= PSR_I; frame->tf_elr = (register_t)kp->kp_tramp; return (0); } void kinst_patch_tracepoint(struct kinst_probe *kp, kinst_patchval_t val) { vm_offset_t addr; if (!arm64_get_writable_addr((vm_offset_t)kp->kp_patchpoint, &addr)) panic("%s: Unable to write new instruction", __func__); *(kinst_patchval_t *)addr = val; cpu_icache_sync_range((vm_offset_t)kp->kp_patchpoint, (vm_size_t)INSN_SIZE); } static void kinst_instr_dissect(struct kinst_probe *kp) { struct kinst_probe_md *kpmd; kinst_patchval_t instr = kp->kp_savedval; kpmd = &kp->kp_md; kpmd->emulate = false; if (((instr >> 24) & 0x1f) == 0b10000) kpmd->emulate = true; /* adr/adrp */ else if (((instr >> 26) & 0x3f) == 0b000101) kpmd->emulate = true; /* b */ else if (((instr >> 24) & 0xff) == 0b01010100) kpmd->emulate = true; /* b.cond */ else if (((instr >> 26) & 0x3f) == 0b100101) kpmd->emulate = true; /* bl */ else if (((instr >> 25) & 0x3f) == 0b011010) kpmd->emulate = true; /* cbnz/cbz */ else if (((instr >> 25) & 0x3f) == 0b011011) kpmd->emulate = true; /* tbnz/tbz */ if (!kpmd->emulate) kinst_trampoline_populate(kp); } static bool kinst_instr_ldx(kinst_patchval_t instr) { if (((instr >> 22) & 0xff) == 0b00100001) return (true); return (false); } static bool kinst_instr_stx(kinst_patchval_t instr) { if (((instr >> 22) & 0xff) == 0b00100000) return (true); return (false); } int kinst_make_probe(linker_file_t lf, int symindx, linker_symval_t *symval, void *opaque) { struct kinst_probe *kp; dtrace_kinst_probedesc_t *pd; const char *func; kinst_patchval_t *instr, *limit, *tmp; int n, off; bool ldxstx_block, found; pd = opaque; func = symval->name; if (kinst_excluded(func)) return (0); if (strcmp(func, pd->kpd_func) != 0) return (0); instr = (kinst_patchval_t *)(symval->value); limit = (kinst_patchval_t *)(symval->value + symval->size); if (instr >= limit) return (0); tmp = instr; /* * Ignore any bti instruction at the start of the function * we need to keep it there for any indirect branches calling * the function on Armv8.5+ */ if ((*tmp & BTI_MASK) == BTI_INSTR) tmp++; /* Look for stp (pre-indexed) operation */ found = false; /* * If the first instruction is a nop it's a specially marked * asm function. We only support a nop first as it's not a normal * part of the function prologue. */ if (*tmp == NOP_INSTR) found = true; for (; !found && tmp < limit; tmp++) { /* * Functions start with "stp xt1, xt2, [xn, ]!" or * "sub sp, sp, ". * * Sometimes the compiler will have a sub instruction that is * not of the above type so don't stop if we see one. */ if ((*tmp & LDP_STP_MASK) == STP_64) { /* * Assume any other store of this type means we are * past the function prolog. */ if (((*tmp >> ADDR_SHIFT) & ADDR_MASK) == 31) found = true; } else if ((*tmp & SUB_MASK) == SUB_INSTR && ((*tmp >> SUB_RD_SHIFT) & SUB_R_MASK) == 31 && ((*tmp >> SUB_RN_SHIFT) & SUB_R_MASK) == 31) found = true; } if (!found) return (0); ldxstx_block = false; for (n = 0; instr < limit; instr++) { off = (int)((uint8_t *)instr - (uint8_t *)symval->value); /* * Skip LDX/STX blocks that contain atomic operations. If a * breakpoint is placed in a LDX/STX block, we violate the * operation and the loop might fail. */ if (kinst_instr_ldx(*instr)) ldxstx_block = true; else if (kinst_instr_stx(*instr)) { ldxstx_block = false; continue; } if (ldxstx_block) continue; /* * XXX: Skip ADR and ADRP instructions. The arm64 exception * handler has a micro-optimization where it doesn't restore * callee-saved registers when returning from exceptions in * EL1. This results in a panic when the kinst emulation code * modifies one of those registers. */ if (((*instr >> 24) & 0x1f) == 0b10000) continue; if (pd->kpd_off != -1 && off != pd->kpd_off) continue; /* * Prevent separate dtrace(1) instances from creating copies of * the same probe. */ LIST_FOREACH(kp, KINST_GETPROBE(instr), kp_hashnext) { if (strcmp(kp->kp_func, func) == 0 && strtol(kp->kp_name, NULL, 10) == off) return (0); } if (++n > KINST_PROBETAB_MAX) { KINST_LOG("probe list full: %d entries", n); return (ENOMEM); } kp = malloc(sizeof(struct kinst_probe), M_KINST, M_WAITOK | M_ZERO); kp->kp_func = func; snprintf(kp->kp_name, sizeof(kp->kp_name), "%d", off); kp->kp_patchpoint = instr; kp->kp_savedval = *instr; kp->kp_patchval = KINST_PATCHVAL; if ((kp->kp_tramp = kinst_trampoline_alloc(M_WAITOK)) == NULL) { KINST_LOG("cannot allocate trampoline for %p", instr); return (ENOMEM); } kinst_instr_dissect(kp); kinst_probe_create(kp, lf); } if (ldxstx_block) KINST_LOG("warning: unterminated LDX/STX block"); return (0); } int kinst_md_init(void) { struct kinst_cpu_state *ks; int cpu; CPU_FOREACH(cpu) { ks = DPCPU_PTR(kinst_state); ks->state = KINST_PROBE_ARMED; } return (0); } void kinst_md_deinit(void) { } /* * Exclude machine-dependent functions that are not safe-to-trace. */ bool kinst_md_excluded(const char *name) { if (strcmp(name, "handle_el1h_sync") == 0 || strcmp(name, "do_el1h_sync") == 0) return (true); return (false); }