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/clocksource/timer-atmel-st.c | 250 +++++++++++++++++++++++++++++++++++ 1 file changed, 250 insertions(+) create mode 100644 drivers/clocksource/timer-atmel-st.c (limited to 'drivers/clocksource/timer-atmel-st.c') diff --git a/drivers/clocksource/timer-atmel-st.c b/drivers/clocksource/timer-atmel-st.c new file mode 100644 index 000000000..73e8aee44 --- /dev/null +++ b/drivers/clocksource/timer-atmel-st.c @@ -0,0 +1,250 @@ +// SPDX-License-Identifier: GPL-2.0-or-later +/* + * linux/arch/arm/mach-at91/at91rm9200_time.c + * + * Copyright (C) 2003 SAN People + * Copyright (C) 2003 ATMEL + */ + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +static unsigned long last_crtr; +static u32 irqmask; +static struct clock_event_device clkevt; +static struct regmap *regmap_st; +static int timer_latch; + +/* + * The ST_CRTR is updated asynchronously to the master clock ... but + * the updates as seen by the CPU don't seem to be strictly monotonic. + * Waiting until we read the same value twice avoids glitching. + */ +static inline unsigned long read_CRTR(void) +{ + unsigned int x1, x2; + + regmap_read(regmap_st, AT91_ST_CRTR, &x1); + do { + regmap_read(regmap_st, AT91_ST_CRTR, &x2); + if (x1 == x2) + break; + x1 = x2; + } while (1); + return x1; +} + +/* + * IRQ handler for the timer. + */ +static irqreturn_t at91rm9200_timer_interrupt(int irq, void *dev_id) +{ + u32 sr; + + regmap_read(regmap_st, AT91_ST_SR, &sr); + sr &= irqmask; + + /* + * irqs should be disabled here, but as the irq is shared they are only + * guaranteed to be off if the timer irq is registered first. + */ + WARN_ON_ONCE(!irqs_disabled()); + + /* simulate "oneshot" timer with alarm */ + if (sr & AT91_ST_ALMS) { + clkevt.event_handler(&clkevt); + return IRQ_HANDLED; + } + + /* periodic mode should handle delayed ticks */ + if (sr & AT91_ST_PITS) { + u32 crtr = read_CRTR(); + + while (((crtr - last_crtr) & AT91_ST_CRTV) >= timer_latch) { + last_crtr += timer_latch; + clkevt.event_handler(&clkevt); + } + return IRQ_HANDLED; + } + + /* this irq is shared ... */ + return IRQ_NONE; +} + +static u64 read_clk32k(struct clocksource *cs) +{ + return read_CRTR(); +} + +static struct clocksource clk32k = { + .name = "32k_counter", + .rating = 150, + .read = read_clk32k, + .mask = CLOCKSOURCE_MASK(20), + .flags = CLOCK_SOURCE_IS_CONTINUOUS, +}; + +static void clkdev32k_disable_and_flush_irq(void) +{ + unsigned int val; + + /* Disable and flush pending timer interrupts */ + regmap_write(regmap_st, AT91_ST_IDR, AT91_ST_PITS | AT91_ST_ALMS); + regmap_read(regmap_st, AT91_ST_SR, &val); + last_crtr = read_CRTR(); +} + +static int clkevt32k_shutdown(struct clock_event_device *evt) +{ + clkdev32k_disable_and_flush_irq(); + irqmask = 0; + regmap_write(regmap_st, AT91_ST_IER, irqmask); + return 0; +} + +static int clkevt32k_set_oneshot(struct clock_event_device *dev) +{ + clkdev32k_disable_and_flush_irq(); + + /* + * ALM for oneshot irqs, set by next_event() + * before 32 seconds have passed. + */ + irqmask = AT91_ST_ALMS; + regmap_write(regmap_st, AT91_ST_RTAR, last_crtr); + regmap_write(regmap_st, AT91_ST_IER, irqmask); + return 0; +} + +static int clkevt32k_set_periodic(struct clock_event_device *dev) +{ + clkdev32k_disable_and_flush_irq(); + + /* PIT for periodic irqs; fixed rate of 1/HZ */ + irqmask = AT91_ST_PITS; + regmap_write(regmap_st, AT91_ST_PIMR, timer_latch); + regmap_write(regmap_st, AT91_ST_IER, irqmask); + return 0; +} + +static int +clkevt32k_next_event(unsigned long delta, struct clock_event_device *dev) +{ + u32 alm; + unsigned int val; + + BUG_ON(delta < 2); + + /* The alarm IRQ uses absolute time (now+delta), not the relative + * time (delta) in our calling convention. Like all clockevents + * using such "match" hardware, we have a race to defend against. + * + * Our defense here is to have set up the clockevent device so the + * delta is at least two. That way we never end up writing RTAR + * with the value then held in CRTR ... which would mean the match + * wouldn't trigger until 32 seconds later, after CRTR wraps. + */ + alm = read_CRTR(); + + /* Cancel any pending alarm; flush any pending IRQ */ + regmap_write(regmap_st, AT91_ST_RTAR, alm); + regmap_read(regmap_st, AT91_ST_SR, &val); + + /* Schedule alarm by writing RTAR. */ + alm += delta; + regmap_write(regmap_st, AT91_ST_RTAR, alm); + + return 0; +} + +static struct clock_event_device clkevt = { + .name = "at91_tick", + .features = CLOCK_EVT_FEAT_PERIODIC | + CLOCK_EVT_FEAT_ONESHOT, + .rating = 150, + .set_next_event = clkevt32k_next_event, + .set_state_shutdown = clkevt32k_shutdown, + .set_state_periodic = clkevt32k_set_periodic, + .set_state_oneshot = clkevt32k_set_oneshot, + .tick_resume = clkevt32k_shutdown, +}; + +/* + * ST (system timer) module supports both clockevents and clocksource. + */ +static int __init atmel_st_timer_init(struct device_node *node) +{ + struct clk *sclk; + unsigned int sclk_rate, val; + int irq, ret; + + regmap_st = syscon_node_to_regmap(node); + if (IS_ERR(regmap_st)) { + pr_err("Unable to get regmap\n"); + return PTR_ERR(regmap_st); + } + + /* Disable all timer interrupts, and clear any pending ones */ + regmap_write(regmap_st, AT91_ST_IDR, + AT91_ST_PITS | AT91_ST_WDOVF | AT91_ST_RTTINC | AT91_ST_ALMS); + regmap_read(regmap_st, AT91_ST_SR, &val); + + /* Get the interrupts property */ + irq = irq_of_parse_and_map(node, 0); + if (!irq) { + pr_err("Unable to get IRQ from DT\n"); + return -EINVAL; + } + + /* Make IRQs happen for the system timer */ + ret = request_irq(irq, at91rm9200_timer_interrupt, + IRQF_SHARED | IRQF_TIMER | IRQF_IRQPOLL, + "at91_tick", regmap_st); + if (ret) { + pr_err("Unable to setup IRQ\n"); + return ret; + } + + sclk = of_clk_get(node, 0); + if (IS_ERR(sclk)) { + pr_err("Unable to get slow clock\n"); + return PTR_ERR(sclk); + } + + ret = clk_prepare_enable(sclk); + if (ret) { + pr_err("Could not enable slow clock\n"); + return ret; + } + + sclk_rate = clk_get_rate(sclk); + if (!sclk_rate) { + pr_err("Invalid slow clock rate\n"); + return -EINVAL; + } + timer_latch = (sclk_rate + HZ / 2) / HZ; + + /* The 32KiHz "Slow Clock" (tick every 30517.58 nanoseconds) is used + * directly for the clocksource and all clockevents, after adjusting + * its prescaler from the 1 Hz default. + */ + regmap_write(regmap_st, AT91_ST_RTMR, 1); + + /* Setup timer clockevent, with minimum of two ticks (important!!) */ + clkevt.cpumask = cpumask_of(0); + clockevents_config_and_register(&clkevt, sclk_rate, + 2, AT91_ST_ALMV); + + /* register clocksource */ + return clocksource_register_hz(&clk32k, sclk_rate); +} +TIMER_OF_DECLARE(atmel_st_timer, "atmel,at91rm9200-st", + atmel_st_timer_init); -- cgit v1.3.1