From a5f7a5bb3b7f28ba7e4fa246775b29a0e5537255 Mon Sep 17 00:00:00 2001 From: Jinke Han Date: Tue, 8 Sep 2026 15:37:42 +0800 Subject: x86/kprobes: Fix crash when probing CS CALL instructions When using eBPF to probe CS CALL instructions within a function, a crash can be triggered. The eBPF tool probes offset 257 of the __hrtimer_run_queues() function: <__hrtimer_run_queues+249>: nopl 0x0(%rax,%rax,1) <__hrtimer_run_queues+254>: mov %r14,%rdi <__hrtimer_run_queues+257>: cs call <__x86_indirect_thunk_r12> <__hrtimer_run_queues+263>: mov %eax,%r12d <__hrtimer_run_queues+266>: xchg %ax,%ax <__hrtimer_run_queues+268>: mov %r13,%rdi Which triggers this crash: BUG: unable to handle page fault for address: 00000000000f41c9 #PF: supervisor write access in kernel mode #PF: error_code(0x0002) - not-present page PGD 0 P4D 0 Oops: 0002 [#1] SMP NOPTI CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P RIP: 0010:__hrtimer_run_queues+0x106/0x230 Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is at the 6th byte of the above CS CALL instruction. Since the CS CALL instruction occupies 6 bytes, the exception occurred in the middle of that call instruction. The root cause is that when using eBPF tools to probe in the middle of a function, a kprobe with INT3 is used as the underlying implementation. During single-step emulation of the original CALL instruction, int3_emulate_call() assumes that the probed CALL instruction is 5 bytes long. However, the actual CS-prefixed CALL instruction occupies 6 bytes, so it constructs an incorrect exception return address. When the CPU returns from the kprobe handler, the next instruction to be executed is at the address of the last byte of that CS CALL instruction. Coincidentally, starting from that address, the CPU fetches and decodes a completely different instruction, which ultimately triggers a kernel crash. Fix the issue by using the actual instruction length obtained from the instruction decoder when constructing the exception return address, rather than relying on the hardcoded CALL_INSN_SIZE macro. [ mingo: Refined the changelog ] Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step") Suggested-by: Masami Hiramatsu (Google) Signed-off-by: Jinke Han Signed-off-by: Ingo Molnar Reviewed-by: Masami Hiramatsu (Google) Acked-by: Yafang Shao Acked-by: Borislav Petkov Cc: Peter Zijlstra Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000 --- arch/riscv/lib/delay.c | 116 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 116 insertions(+) create mode 100644 arch/riscv/lib/delay.c (limited to 'arch/riscv/lib/delay.c') diff --git a/arch/riscv/lib/delay.c b/arch/riscv/lib/delay.c new file mode 100644 index 000000000..bd9b38256 --- /dev/null +++ b/arch/riscv/lib/delay.c @@ -0,0 +1,116 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Copyright (C) 2012 Regents of the University of California + */ + +#include +#include +#include +#include +#include + +#include + +/* + * This is copies from arch/arm/include/asm/delay.h + * + * Loop (or tick) based delay: + * + * loops = loops_per_jiffy * jiffies_per_sec * delay_us / us_per_sec + * + * where: + * + * jiffies_per_sec = HZ + * us_per_sec = 1000000 + * + * Therefore the constant part is HZ / 1000000 which is a small + * fractional number. To make this usable with integer math, we + * scale up this constant by 2^31, perform the actual multiplication, + * and scale the result back down by 2^31 with a simple shift: + * + * loops = (loops_per_jiffy * delay_us * UDELAY_MULT) >> 31 + * + * where: + * + * UDELAY_MULT = 2^31 * HZ / 1000000 + * = (2^31 / 1000000) * HZ + * = 2147.483648 * HZ + * = 2147 * HZ + 483648 * HZ / 1000000 + * + * 31 is the biggest scale shift value that won't overflow 32 bits for + * delay_us * UDELAY_MULT assuming HZ <= 1000 and delay_us <= 2000. + */ +#define MAX_UDELAY_US 2000 +#define MAX_UDELAY_HZ 1000 +#define UDELAY_MULT (2147UL * HZ + 483648UL * HZ / 1000000UL) +#define UDELAY_SHIFT 31 + +#if HZ > MAX_UDELAY_HZ +#error "HZ > MAX_UDELAY_HZ" +#endif + +/* + * RISC-V supports both UDELAY and NDELAY. This is largely the same as above, + * but with different constants. I added 10 bits to the shift to get this, but + * the result is that I need a 64-bit multiply, which is slow on 32-bit + * platforms. + * + * NDELAY_MULT = 2^41 * HZ / 1000000000 + * = (2^41 / 1000000000) * HZ + * = 2199.02325555 * HZ + * = 2199 * HZ + 23255550 * HZ / 1000000000 + * + * The maximum here is to avoid 64-bit overflow, but it isn't checked as it + * won't happen. + */ +#define MAX_NDELAY_NS (1ULL << 42) +#define MAX_NDELAY_HZ MAX_UDELAY_HZ +#define NDELAY_MULT ((unsigned long long)(2199ULL * HZ + 23255550ULL * HZ / 1000000000ULL)) +#define NDELAY_SHIFT 41 + +#if HZ > MAX_NDELAY_HZ +#error "HZ > MAX_NDELAY_HZ" +#endif + +void __delay(unsigned long cycles) +{ + u64 t0 = get_cycles(); + + while ((unsigned long)(get_cycles() - t0) < cycles) + cpu_relax(); +} +EXPORT_SYMBOL(__delay); + +void udelay(unsigned long usecs) +{ + u64 ucycles = (u64)usecs * lpj_fine * UDELAY_MULT; + u64 n; + + if (unlikely(usecs > MAX_UDELAY_US)) { + n = (u64)usecs * riscv_timebase; + do_div(n, 1000000); + + __delay(n); + return; + } + + __delay(ucycles >> UDELAY_SHIFT); +} +EXPORT_SYMBOL(udelay); + +void ndelay(unsigned long nsecs) +{ + /* + * This doesn't bother checking for overflow, as it won't happen (it's + * an hour) of delay. + */ + unsigned long long ncycles = nsecs * lpj_fine * NDELAY_MULT; + __delay(ncycles >> NDELAY_SHIFT); +} +EXPORT_SYMBOL(ndelay); + +bool delay_read_timer(unsigned long *timer_val) +{ + *timer_val = get_cycles(); + return true; +} -- cgit v1.3.1