x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

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) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

1 files changed, 384 insertions, 0 deletions

diff --git a/drivers/clocksource/timer-davinci.c b/drivers/clocksource/timer-davinci.c
new file mode 100644
index 000000000..16ee0687e
--- /dev/null
+++ b/drivers/clocksource/timer-davinci.c
@@ -0,0 +1,384 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * TI DaVinci clocksource driver
+ *
+ * Copyright (C) 2019 Texas Instruments
+ * Author: Bartosz Golaszewski <bgolaszewski@baylibre.com>
+ * (with tiny parts adopted from code by Kevin Hilman <khilman@baylibre.com>)
+ */
+
+#define pr_fmt(fmt) "%s: " fmt, __func__
+
+#include <linux/clk.h>
+#include <linux/clockchips.h>
+#include <linux/interrupt.h>
+#include <linux/kernel.h>
+#include <linux/of_address.h>
+#include <linux/of_irq.h>
+#include <linux/sched_clock.h>
+
+#include <clocksource/timer-davinci.h>
+
+#define DAVINCI_TIMER_REG_TIM12			0x10
+#define DAVINCI_TIMER_REG_TIM34			0x14
+#define DAVINCI_TIMER_REG_PRD12			0x18
+#define DAVINCI_TIMER_REG_PRD34			0x1c
+#define DAVINCI_TIMER_REG_TCR			0x20
+#define DAVINCI_TIMER_REG_TGCR			0x24
+
+#define DAVINCI_TIMER_TIMMODE_MASK		GENMASK(3, 2)
+#define DAVINCI_TIMER_RESET_MASK		GENMASK(1, 0)
+#define DAVINCI_TIMER_TIMMODE_32BIT_UNCHAINED	BIT(2)
+#define DAVINCI_TIMER_UNRESET			GENMASK(1, 0)
+
+#define DAVINCI_TIMER_ENAMODE_MASK		GENMASK(1, 0)
+#define DAVINCI_TIMER_ENAMODE_DISABLED		0x00
+#define DAVINCI_TIMER_ENAMODE_ONESHOT		BIT(0)
+#define DAVINCI_TIMER_ENAMODE_PERIODIC		BIT(1)
+
+#define DAVINCI_TIMER_ENAMODE_SHIFT_TIM12	6
+#define DAVINCI_TIMER_ENAMODE_SHIFT_TIM34	22
+
+#define DAVINCI_TIMER_MIN_DELTA			0x01
+#define DAVINCI_TIMER_MAX_DELTA			0xfffffffe
+
+#define DAVINCI_TIMER_CLKSRC_BITS		32
+
+#define DAVINCI_TIMER_TGCR_DEFAULT \
+		(DAVINCI_TIMER_TIMMODE_32BIT_UNCHAINED | DAVINCI_TIMER_UNRESET)
+
+struct davinci_clockevent {
+	struct clock_event_device dev;
+	void __iomem *base;
+	unsigned int cmp_off;
+};
+
+/*
+ * This must be globally accessible by davinci_timer_read_sched_clock(), so
+ * let's keep it here.
+ */
+static struct {
+	struct clocksource dev;
+	void __iomem *base;
+	unsigned int tim_off;
+} davinci_clocksource;
+
+static struct davinci_clockevent *
+to_davinci_clockevent(struct clock_event_device *clockevent)
+{
+	return container_of(clockevent, struct davinci_clockevent, dev);
+}
+
+static unsigned int
+davinci_clockevent_read(struct davinci_clockevent *clockevent,
+			unsigned int reg)
+{
+	return readl_relaxed(clockevent->base + reg);
+}
+
+static void davinci_clockevent_write(struct davinci_clockevent *clockevent,
+				     unsigned int reg, unsigned int val)
+{
+	writel_relaxed(val, clockevent->base + reg);
+}
+
+static void davinci_tim12_shutdown(void __iomem *base)
+{
+	unsigned int tcr;
+
+	tcr = DAVINCI_TIMER_ENAMODE_DISABLED <<
+		DAVINCI_TIMER_ENAMODE_SHIFT_TIM12;
+	/*
+	 * This function is only ever called if we're using both timer
+	 * halves. In this case TIM34 runs in periodic mode and we must
+	 * not modify it.
+	 */
+	tcr |= DAVINCI_TIMER_ENAMODE_PERIODIC <<
+		DAVINCI_TIMER_ENAMODE_SHIFT_TIM34;
+
+	writel_relaxed(tcr, base + DAVINCI_TIMER_REG_TCR);
+}
+
+static void davinci_tim12_set_oneshot(void __iomem *base)
+{
+	unsigned int tcr;
+
+	tcr = DAVINCI_TIMER_ENAMODE_ONESHOT <<
+		DAVINCI_TIMER_ENAMODE_SHIFT_TIM12;
+	/* Same as above. */
+	tcr |= DAVINCI_TIMER_ENAMODE_PERIODIC <<
+		DAVINCI_TIMER_ENAMODE_SHIFT_TIM34;
+
+	writel_relaxed(tcr, base + DAVINCI_TIMER_REG_TCR);
+}
+
+static int davinci_clockevent_shutdown(struct clock_event_device *dev)
+{
+	struct davinci_clockevent *clockevent;
+
+	clockevent = to_davinci_clockevent(dev);
+
+	davinci_tim12_shutdown(clockevent->base);
+
+	return 0;
+}
+
+static int davinci_clockevent_set_oneshot(struct clock_event_device *dev)
+{
+	struct davinci_clockevent *clockevent = to_davinci_clockevent(dev);
+
+	davinci_clockevent_write(clockevent, DAVINCI_TIMER_REG_TIM12, 0x0);
+
+	davinci_tim12_set_oneshot(clockevent->base);
+
+	return 0;
+}
+
+static int
+davinci_clockevent_set_next_event_std(unsigned long cycles,
+				      struct clock_event_device *dev)
+{
+	struct davinci_clockevent *clockevent = to_davinci_clockevent(dev);
+
+	davinci_clockevent_shutdown(dev);
+
+	davinci_clockevent_write(clockevent, DAVINCI_TIMER_REG_TIM12, 0x0);
+	davinci_clockevent_write(clockevent, DAVINCI_TIMER_REG_PRD12, cycles);
+
+	davinci_clockevent_set_oneshot(dev);
+
+	return 0;
+}
+
+static int
+davinci_clockevent_set_next_event_cmp(unsigned long cycles,
+				      struct clock_event_device *dev)
+{
+	struct davinci_clockevent *clockevent = to_davinci_clockevent(dev);
+	unsigned int curr_time;
+
+	curr_time = davinci_clockevent_read(clockevent,
+					    DAVINCI_TIMER_REG_TIM12);
+	davinci_clockevent_write(clockevent,
+				 clockevent->cmp_off, curr_time + cycles);
+
+	return 0;
+}
+
+static irqreturn_t davinci_timer_irq_timer(int irq, void *data)
+{
+	struct davinci_clockevent *clockevent = data;
+
+	if (!clockevent_state_oneshot(&clockevent->dev))
+		davinci_tim12_shutdown(clockevent->base);
+
+	clockevent->dev.event_handler(&clockevent->dev);
+
+	return IRQ_HANDLED;
+}
+
+static u64 notrace davinci_timer_read_sched_clock(void)
+{
+	return readl_relaxed(davinci_clocksource.base +
+			     davinci_clocksource.tim_off);
+}
+
+static u64 davinci_clocksource_read(struct clocksource *dev)
+{
+	return davinci_timer_read_sched_clock();
+}
+
+/*
+ * Standard use-case: we're using tim12 for clockevent and tim34 for
+ * clocksource. The default is making the former run in oneshot mode
+ * and the latter in periodic mode.
+ */
+static void davinci_clocksource_init_tim34(void __iomem *base)
+{
+	int tcr;
+
+	tcr = DAVINCI_TIMER_ENAMODE_PERIODIC <<
+		DAVINCI_TIMER_ENAMODE_SHIFT_TIM34;
+	tcr |= DAVINCI_TIMER_ENAMODE_ONESHOT <<
+		DAVINCI_TIMER_ENAMODE_SHIFT_TIM12;
+
+	writel_relaxed(0x0, base + DAVINCI_TIMER_REG_TIM34);
+	writel_relaxed(UINT_MAX, base + DAVINCI_TIMER_REG_PRD34);
+	writel_relaxed(tcr, base + DAVINCI_TIMER_REG_TCR);
+}
+
+/*
+ * Special use-case on da830: the DSP may use tim34. We're using tim12 for
+ * both clocksource and clockevent. We set tim12 to periodic and don't touch
+ * tim34.
+ */
+static void davinci_clocksource_init_tim12(void __iomem *base)
+{
+	unsigned int tcr;
+
+	tcr = DAVINCI_TIMER_ENAMODE_PERIODIC <<
+		DAVINCI_TIMER_ENAMODE_SHIFT_TIM12;
+
+	writel_relaxed(0x0, base + DAVINCI_TIMER_REG_TIM12);
+	writel_relaxed(UINT_MAX, base + DAVINCI_TIMER_REG_PRD12);
+	writel_relaxed(tcr, base + DAVINCI_TIMER_REG_TCR);
+}
+
+static void davinci_timer_init(void __iomem *base)
+{
+	/* Set clock to internal mode and disable it. */
+	writel_relaxed(0x0, base + DAVINCI_TIMER_REG_TCR);
+	/*
+	 * Reset both 32-bit timers, set no prescaler for timer 34, set the
+	 * timer to dual 32-bit unchained mode, unreset both 32-bit timers.
+	 */
+	writel_relaxed(DAVINCI_TIMER_TGCR_DEFAULT,
+		       base + DAVINCI_TIMER_REG_TGCR);
+	/* Init both counters to zero. */
+	writel_relaxed(0x0, base + DAVINCI_TIMER_REG_TIM12);
+	writel_relaxed(0x0, base + DAVINCI_TIMER_REG_TIM34);
+}
+
+int __init davinci_timer_register(struct clk *clk,
+				  const struct davinci_timer_cfg *timer_cfg)
+{
+	struct davinci_clockevent *clockevent;
+	unsigned int tick_rate;
+	void __iomem *base;
+	int rv;
+
+	rv = clk_prepare_enable(clk);
+	if (rv) {
+		pr_err("Unable to prepare and enable the timer clock\n");
+		return rv;
+	}
+
+	if (!request_mem_region(timer_cfg->reg.start,
+				resource_size(&timer_cfg->reg),
+				"davinci-timer")) {
+		pr_err("Unable to request memory region\n");
+		rv = -EBUSY;
+		goto exit_clk_disable;
+	}
+
+	base = ioremap(timer_cfg->reg.start, resource_size(&timer_cfg->reg));
+	if (!base) {
+		pr_err("Unable to map the register range\n");
+		rv = -ENOMEM;
+		goto exit_mem_region;
+	}
+
+	davinci_timer_init(base);
+	tick_rate = clk_get_rate(clk);
+
+	clockevent = kzalloc_obj(*clockevent);
+	if (!clockevent) {
+		rv = -ENOMEM;
+		goto exit_iounmap_base;
+	}
+
+	clockevent->dev.name = "tim12";
+	clockevent->dev.features = CLOCK_EVT_FEAT_ONESHOT;
+	clockevent->dev.cpumask = cpumask_of(0);
+	clockevent->base = base;
+
+	if (timer_cfg->cmp_off) {
+		clockevent->cmp_off = timer_cfg->cmp_off;
+		clockevent->dev.set_next_event =
+				davinci_clockevent_set_next_event_cmp;
+	} else {
+		clockevent->dev.set_next_event =
+				davinci_clockevent_set_next_event_std;
+		clockevent->dev.set_state_oneshot =
+				davinci_clockevent_set_oneshot;
+		clockevent->dev.set_state_shutdown =
+				davinci_clockevent_shutdown;
+	}
+
+	rv = request_irq(timer_cfg->irq[DAVINCI_TIMER_CLOCKEVENT_IRQ].start,
+			 davinci_timer_irq_timer, IRQF_TIMER,
+			 "clockevent/tim12", clockevent);
+	if (rv) {
+		pr_err("Unable to request the clockevent interrupt\n");
+		goto exit_free_clockevent;
+	}
+
+	davinci_clocksource.dev.rating = 300;
+	davinci_clocksource.dev.read = davinci_clocksource_read;
+	davinci_clocksource.dev.mask =
+			CLOCKSOURCE_MASK(DAVINCI_TIMER_CLKSRC_BITS);
+	davinci_clocksource.dev.flags = CLOCK_SOURCE_IS_CONTINUOUS;
+	davinci_clocksource.base = base;
+
+	if (timer_cfg->cmp_off) {
+		davinci_clocksource.dev.name = "tim12";
+		davinci_clocksource.tim_off = DAVINCI_TIMER_REG_TIM12;
+		davinci_clocksource_init_tim12(base);
+	} else {
+		davinci_clocksource.dev.name = "tim34";
+		davinci_clocksource.tim_off = DAVINCI_TIMER_REG_TIM34;
+		davinci_clocksource_init_tim34(base);
+	}
+
+	clockevents_config_and_register(&clockevent->dev, tick_rate,
+					DAVINCI_TIMER_MIN_DELTA,
+					DAVINCI_TIMER_MAX_DELTA);
+
+	rv = clocksource_register_hz(&davinci_clocksource.dev, tick_rate);
+	if (rv) {
+		pr_err("Unable to register clocksource\n");
+		goto exit_free_irq;
+	}
+
+	sched_clock_register(davinci_timer_read_sched_clock,
+			     DAVINCI_TIMER_CLKSRC_BITS, tick_rate);
+
+	return 0;
+
+exit_free_irq:
+	free_irq(timer_cfg->irq[DAVINCI_TIMER_CLOCKEVENT_IRQ].start,
+			clockevent);
+exit_free_clockevent:
+	kfree(clockevent);
+exit_iounmap_base:
+	iounmap(base);
+exit_mem_region:
+	release_mem_region(timer_cfg->reg.start,
+			   resource_size(&timer_cfg->reg));
+exit_clk_disable:
+	clk_disable_unprepare(clk);
+	return rv;
+}
+
+static int __init of_davinci_timer_register(struct device_node *np)
+{
+	struct davinci_timer_cfg timer_cfg = { };
+	struct clk *clk;
+	int rv;
+
+	rv = of_address_to_resource(np, 0, &timer_cfg.reg);
+	if (rv) {
+		pr_err("Unable to get the register range for timer\n");
+		return rv;
+	}
+
+	rv = of_irq_to_resource_table(np, timer_cfg.irq,
+				      DAVINCI_TIMER_NUM_IRQS);
+	if (rv != DAVINCI_TIMER_NUM_IRQS) {
+		pr_err("Unable to get the interrupts for timer\n");
+		return rv;
+	}
+
+	clk = of_clk_get(np, 0);
+	if (IS_ERR(clk)) {
+		pr_err("Unable to get the timer clock\n");
+		return PTR_ERR(clk);
+	}
+
+	rv = davinci_timer_register(clk, &timer_cfg);
+	if (rv)
+		clk_put(clk);
+
+	return rv;
+}
+TIMER_OF_DECLARE(davinci_timer, "ti,da830-timer", of_davinci_timer_register);