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 --- drivers/cpufreq/p4-clockmod.c | 272 ++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 272 insertions(+) create mode 100644 drivers/cpufreq/p4-clockmod.c (limited to 'drivers/cpufreq/p4-clockmod.c') diff --git a/drivers/cpufreq/p4-clockmod.c b/drivers/cpufreq/p4-clockmod.c new file mode 100644 index 000000000..c1690aa48 --- /dev/null +++ b/drivers/cpufreq/p4-clockmod.c @@ -0,0 +1,272 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * Pentium 4/Xeon CPU on demand clock modulation/speed scaling + * (C) 2002 - 2003 Dominik Brodowski + * (C) 2002 Zwane Mwaikambo + * (C) 2002 Arjan van de Ven + * (C) 2002 Tora T. Engstad + * All Rights Reserved + * + * The author(s) of this software shall not be held liable for damages + * of any nature resulting due to the use of this software. This + * software is provided AS-IS with no warranties. + * + * Date Errata Description + * 20020525 N44, O17 12.5% or 25% DC causes lockup + */ + +#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt + +#include +#include +#include +#include +#include +#include +#include + +#include +#include +#include +#include + +#include "speedstep-lib.h" + +/* + * Duty Cycle (3bits), note DC_DISABLE is not specified in + * intel docs i just use it to mean disable + */ +enum { + DC_RESV, DC_DFLT, DC_25PT, DC_38PT, DC_50PT, + DC_64PT, DC_75PT, DC_88PT, DC_DISABLE +}; + +#define DC_ENTRIES 8 + + +static int has_N44_O17_errata[NR_CPUS]; +static unsigned int stock_freq; +static struct cpufreq_driver p4clockmod_driver; +static unsigned int cpufreq_p4_get(unsigned int cpu); + +static int cpufreq_p4_setdc(unsigned int cpu, unsigned int newstate) +{ + struct msr val; + + if ((newstate > DC_DISABLE) || (newstate == DC_RESV)) + return -EINVAL; + + rdmsrq_on_cpu(cpu, MSR_IA32_THERM_STATUS, &val.q); + + if (val.l & 0x01) + pr_debug("CPU#%d currently thermal throttled\n", cpu); + + if (has_N44_O17_errata[cpu] && + (newstate == DC_25PT || newstate == DC_DFLT)) + newstate = DC_38PT; + + rdmsrq_on_cpu(cpu, MSR_IA32_THERM_CONTROL, &val.q); + if (newstate == DC_DISABLE) { + pr_debug("CPU#%d disabling modulation\n", cpu); + wrmsrq_on_cpu(cpu, MSR_IA32_THERM_CONTROL, val.q & ~(1ULL << 4)); + } else { + pr_debug("CPU#%d setting duty cycle to %d%%\n", + cpu, ((125 * newstate) / 10)); + /* bits 63 - 5 : reserved + * bit 4 : enable/disable + * bits 3-1 : duty cycle + * bit 0 : reserved + */ + val.l = (val.l & ~14); + val.l = val.l | (1<<4) | ((newstate & 0x7)<<1); + wrmsrq_on_cpu(cpu, MSR_IA32_THERM_CONTROL, val.q); + } + + return 0; +} + + +static struct cpufreq_frequency_table p4clockmod_table[] = { + {0, DC_RESV, CPUFREQ_ENTRY_INVALID}, + {0, DC_DFLT, 0}, + {0, DC_25PT, 0}, + {0, DC_38PT, 0}, + {0, DC_50PT, 0}, + {0, DC_64PT, 0}, + {0, DC_75PT, 0}, + {0, DC_88PT, 0}, + {0, DC_DISABLE, 0}, + {0, DC_RESV, CPUFREQ_TABLE_END}, +}; + + +static int cpufreq_p4_target(struct cpufreq_policy *policy, unsigned int index) +{ + int i; + + /* run on each logical CPU, + * see section 13.15.3 of IA32 Intel Architecture Software + * Developer's Manual, Volume 3 + */ + for_each_cpu(i, policy->cpus) + cpufreq_p4_setdc(i, p4clockmod_table[index].driver_data); + + return 0; +} + + +static unsigned int cpufreq_p4_get_frequency(struct cpuinfo_x86 *c) +{ + if (c->x86 == 0x06) { + if (cpu_has(c, X86_FEATURE_EST)) + pr_warn_once("Warning: EST-capable CPU detected. The acpi-cpufreq module offers voltage scaling in addition to frequency scaling. You should use that instead of p4-clockmod, if possible.\n"); + switch (c->x86_model) { + case 0x0E: /* Core */ + case 0x0F: /* Core Duo */ + case 0x16: /* Celeron Core */ + case 0x1C: /* Atom */ + p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS; + return speedstep_get_frequency(SPEEDSTEP_CPU_PCORE); + case 0x0D: /* Pentium M (Dothan) */ + p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS; + fallthrough; + case 0x09: /* Pentium M (Banias) */ + return speedstep_get_frequency(SPEEDSTEP_CPU_PM); + } + } + + if (c->x86 != 0xF) + return 0; + + /* on P-4s, the TSC runs with constant frequency independent whether + * throttling is active or not. */ + p4clockmod_driver.flags |= CPUFREQ_CONST_LOOPS; + + if (speedstep_detect_processor() == SPEEDSTEP_CPU_P4M) { + pr_warn("Warning: Pentium 4-M detected. The speedstep-ich or acpi cpufreq modules offer voltage scaling in addition of frequency scaling. You should use either one instead of p4-clockmod, if possible.\n"); + return speedstep_get_frequency(SPEEDSTEP_CPU_P4M); + } + + return speedstep_get_frequency(SPEEDSTEP_CPU_P4D); +} + + + +static int cpufreq_p4_cpu_init(struct cpufreq_policy *policy) +{ + struct cpuinfo_x86 *c = &cpu_data(policy->cpu); + int cpuid = 0; + unsigned int i; + +#ifdef CONFIG_SMP + cpumask_copy(policy->cpus, topology_sibling_cpumask(policy->cpu)); +#endif + + /* Errata workaround */ + cpuid = (c->x86 << 8) | (c->x86_model << 4) | c->x86_stepping; + switch (cpuid) { + case 0x0f07: + case 0x0f0a: + case 0x0f11: + case 0x0f12: + has_N44_O17_errata[policy->cpu] = 1; + pr_debug("has errata -- disabling low frequencies\n"); + } + + if (speedstep_detect_processor() == SPEEDSTEP_CPU_P4D && + c->x86_model < 2) { + /* switch to maximum frequency and measure result */ + cpufreq_p4_setdc(policy->cpu, DC_DISABLE); + recalibrate_cpu_khz(); + } + /* get max frequency */ + stock_freq = cpufreq_p4_get_frequency(c); + if (!stock_freq) + return -EINVAL; + + /* table init */ + for (i = 1; (p4clockmod_table[i].frequency != CPUFREQ_TABLE_END); i++) { + if ((i < 2) && (has_N44_O17_errata[policy->cpu])) + p4clockmod_table[i].frequency = CPUFREQ_ENTRY_INVALID; + else + p4clockmod_table[i].frequency = (stock_freq * i)/8; + } + + /* cpuinfo and default policy values */ + + /* the transition latency is set to be 1 higher than the maximum + * transition latency of the ondemand governor */ + policy->cpuinfo.transition_latency = 10000001; + policy->freq_table = &p4clockmod_table[0]; + + return 0; +} + + +static unsigned int cpufreq_p4_get(unsigned int cpu) +{ + struct msr val; + + rdmsrq_on_cpu(cpu, MSR_IA32_THERM_CONTROL, &val.q); + + if (val.l & 0x10) { + val.l = val.l >> 1; + val.l &= 0x7; + } else + val.l = DC_DISABLE; + + if (val.l != DC_DISABLE) + return stock_freq * val.l / 8; + + return stock_freq; +} + +static struct cpufreq_driver p4clockmod_driver = { + .verify = cpufreq_generic_frequency_table_verify, + .target_index = cpufreq_p4_target, + .init = cpufreq_p4_cpu_init, + .get = cpufreq_p4_get, + .name = "p4-clockmod", +}; + +static const struct x86_cpu_id cpufreq_p4_id[] = { + X86_MATCH_VENDOR_FEATURE(INTEL, X86_FEATURE_ACC, NULL), + {} +}; + +/* + * Intentionally no MODULE_DEVICE_TABLE here: this driver should not + * be auto loaded. Please don't add one. + */ + +static int __init cpufreq_p4_init(void) +{ + int ret; + + /* + * THERM_CONTROL is architectural for IA32 now, so + * we can rely on the capability checks + */ + if (!x86_match_cpu(cpufreq_p4_id) || !boot_cpu_has(X86_FEATURE_ACPI)) + return -ENODEV; + + ret = cpufreq_register_driver(&p4clockmod_driver); + if (!ret) + pr_info("P4/Xeon(TM) CPU On-Demand Clock Modulation available\n"); + + return ret; +} + + +static void __exit cpufreq_p4_exit(void) +{ + cpufreq_unregister_driver(&p4clockmod_driver); +} + + +MODULE_AUTHOR("Zwane Mwaikambo "); +MODULE_DESCRIPTION("cpufreq driver for Pentium(TM) 4/Xeon(TM)"); +MODULE_LICENSE("GPL"); + +late_initcall(cpufreq_p4_init); +module_exit(cpufreq_p4_exit); -- cgit v1.3.1