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, 403 insertions, 0 deletions

diff --git a/drivers/cpuidle/cpuidle-tegra.c b/drivers/cpuidle/cpuidle-tegra.c
new file mode 100644
index 000000000..aca907a62
--- /dev/null
+++ b/drivers/cpuidle/cpuidle-tegra.c
@@ -0,0 +1,403 @@
+// SPDX-License-Identifier: GPL-2.0-only
+/*
+ * CPU idle driver for Tegra CPUs
+ *
+ * Copyright (c) 2010-2013, NVIDIA Corporation.
+ * Copyright (c) 2011 Google, Inc.
+ * Author: Colin Cross <ccross@android.com>
+ *         Gary King <gking@nvidia.com>
+ *
+ * Rework for 3.3 by Peter De Schrijver <pdeschrijver@nvidia.com>
+ *
+ * Tegra20/124 driver unification by Dmitry Osipenko <digetx@gmail.com>
+ */
+
+#define pr_fmt(fmt)	"tegra-cpuidle: " fmt
+
+#include <linux/atomic.h>
+#include <linux/cpuidle.h>
+#include <linux/cpumask.h>
+#include <linux/cpu_pm.h>
+#include <linux/delay.h>
+#include <linux/errno.h>
+#include <linux/platform_device.h>
+#include <linux/types.h>
+
+#include <linux/clk/tegra.h>
+#include <linux/firmware/trusted_foundations.h>
+
+#include <soc/tegra/cpuidle.h>
+#include <soc/tegra/flowctrl.h>
+#include <soc/tegra/fuse.h>
+#include <soc/tegra/irq.h>
+#include <soc/tegra/pm.h>
+#include <soc/tegra/pmc.h>
+
+#include <asm/cpuidle.h>
+#include <asm/firmware.h>
+#include <asm/smp_plat.h>
+#include <asm/suspend.h>
+
+enum tegra_state {
+	TEGRA_C1,
+	TEGRA_C7,
+	TEGRA_CC6,
+	TEGRA_STATE_COUNT,
+};
+
+static atomic_t tegra_idle_barrier;
+static atomic_t tegra_abort_flag;
+
+static void tegra_cpuidle_report_cpus_state(void)
+{
+	unsigned long cpu, lcpu, csr;
+
+	for_each_cpu(lcpu, cpu_possible_mask) {
+		cpu = cpu_logical_map(lcpu);
+		csr = flowctrl_read_cpu_csr(cpu);
+
+		pr_err("cpu%lu: online=%d flowctrl_csr=0x%08lx\n",
+		       cpu, cpu_online(lcpu), csr);
+	}
+}
+
+static int tegra_cpuidle_wait_for_secondary_cpus_parking(void)
+{
+	unsigned int retries = 3;
+
+	while (retries--) {
+		unsigned int delay_us = 10;
+		unsigned int timeout_us = 500 * 1000 / delay_us;
+
+		/*
+		 * The primary CPU0 core shall wait for the secondaries
+		 * shutdown in order to power-off CPU's cluster safely.
+		 * The timeout value depends on the current CPU frequency,
+		 * it takes about 40-150us in average and over 1000us in
+		 * a worst case scenario.
+		 */
+		do {
+			if (tegra_cpu_rail_off_ready())
+				return 0;
+
+			udelay(delay_us);
+
+		} while (timeout_us--);
+
+		pr_err("secondary CPU taking too long to park\n");
+
+		tegra_cpuidle_report_cpus_state();
+	}
+
+	pr_err("timed out waiting secondaries to park\n");
+
+	return -ETIMEDOUT;
+}
+
+static void tegra_cpuidle_unpark_secondary_cpus(void)
+{
+	unsigned int cpu, lcpu;
+
+	for_each_cpu(lcpu, cpu_online_mask) {
+		cpu = cpu_logical_map(lcpu);
+
+		if (cpu > 0) {
+			tegra_enable_cpu_clock(cpu);
+			tegra_cpu_out_of_reset(cpu);
+			flowctrl_write_cpu_halt(cpu, 0);
+		}
+	}
+}
+
+static int tegra_cpuidle_cc6_enter(unsigned int cpu)
+{
+	int ret;
+
+	if (cpu > 0) {
+		ret = cpu_suspend(cpu, tegra_pm_park_secondary_cpu);
+	} else {
+		ret = tegra_cpuidle_wait_for_secondary_cpus_parking();
+		if (!ret)
+			ret = tegra_pm_enter_lp2();
+
+		tegra_cpuidle_unpark_secondary_cpus();
+	}
+
+	return ret;
+}
+
+static int tegra_cpuidle_c7_enter(void)
+{
+	int err;
+
+	err = call_firmware_op(prepare_idle, TF_PM_MODE_LP2_NOFLUSH_L2);
+	if (err && err != -ENOSYS)
+		return err;
+
+	return cpu_suspend(0, tegra30_pm_secondary_cpu_suspend);
+}
+
+static int tegra_cpuidle_coupled_barrier(struct cpuidle_device *dev)
+{
+	if (tegra_pending_sgi()) {
+		/*
+		 * CPU got local interrupt that will be lost after GIC's
+		 * shutdown because GIC driver doesn't save/restore the
+		 * pending SGI state across CPU cluster PM.  Abort and retry
+		 * next time.
+		 */
+		atomic_set(&tegra_abort_flag, 1);
+	}
+
+	cpuidle_coupled_parallel_barrier(dev, &tegra_idle_barrier);
+
+	if (atomic_read(&tegra_abort_flag)) {
+		cpuidle_coupled_parallel_barrier(dev, &tegra_idle_barrier);
+		atomic_set(&tegra_abort_flag, 0);
+		return -EINTR;
+	}
+
+	return 0;
+}
+
+static __cpuidle int tegra_cpuidle_state_enter(struct cpuidle_device *dev,
+					       int index, unsigned int cpu)
+{
+	int err;
+
+	/*
+	 * CC6 state is the "CPU cluster power-off" state.  In order to
+	 * enter this state, at first the secondary CPU cores need to be
+	 * parked into offline mode, then the last CPU should clean out
+	 * remaining dirty cache lines into DRAM and trigger Flow Controller
+	 * logic that turns off the cluster's power domain (which includes
+	 * CPU cores, GIC and L2 cache).
+	 */
+	if (index == TEGRA_CC6) {
+		err = tegra_cpuidle_coupled_barrier(dev);
+		if (err)
+			return err;
+	}
+
+	local_fiq_disable();
+	tegra_pm_set_cpu_in_lp2();
+	cpu_pm_enter();
+
+	ct_cpuidle_enter();
+
+	switch (index) {
+	case TEGRA_C7:
+		err = tegra_cpuidle_c7_enter();
+		break;
+
+	case TEGRA_CC6:
+		err = tegra_cpuidle_cc6_enter(cpu);
+		break;
+
+	default:
+		err = -EINVAL;
+		break;
+	}
+
+	ct_cpuidle_exit();
+
+	cpu_pm_exit();
+	tegra_pm_clear_cpu_in_lp2();
+	local_fiq_enable();
+
+	return err ?: index;
+}
+
+static int tegra_cpuidle_adjust_state_index(int index, unsigned int cpu)
+{
+	/*
+	 * On Tegra30 CPU0 can't be power-gated separately from secondary
+	 * cores because it gates the whole CPU cluster.
+	 */
+	if (cpu > 0 || index != TEGRA_C7 || tegra_get_chip_id() != TEGRA30)
+		return index;
+
+	/* put CPU0 into C1 if C7 is requested and secondaries are online */
+	if (!IS_ENABLED(CONFIG_PM_SLEEP) || num_online_cpus() > 1)
+		index = TEGRA_C1;
+	else
+		index = TEGRA_CC6;
+
+	return index;
+}
+
+static __cpuidle int tegra_cpuidle_enter(struct cpuidle_device *dev,
+					 struct cpuidle_driver *drv,
+					 int index)
+{
+	bool do_rcu = drv->states[index].flags & CPUIDLE_FLAG_RCU_IDLE;
+	unsigned int cpu = cpu_logical_map(dev->cpu);
+	int ret;
+
+	index = tegra_cpuidle_adjust_state_index(index, cpu);
+	if (dev->states_usage[index].disable)
+		return -1;
+
+	if (index == TEGRA_C1) {
+		if (do_rcu)
+			ct_cpuidle_enter();
+		ret = arm_cpuidle_simple_enter(dev, drv, index);
+		if (do_rcu)
+			ct_cpuidle_exit();
+	} else
+		ret = tegra_cpuidle_state_enter(dev, index, cpu);
+
+	if (ret < 0) {
+		if (ret != -EINTR || index != TEGRA_CC6)
+			pr_err_once("failed to enter state %d err: %d\n",
+				    index, ret);
+		index = -1;
+	} else {
+		index = ret;
+	}
+
+	return index;
+}
+
+static int tegra114_enter_s2idle(struct cpuidle_device *dev,
+				 struct cpuidle_driver *drv,
+				 int index)
+{
+	tegra_cpuidle_enter(dev, drv, index);
+
+	return 0;
+}
+
+/*
+ * The previous versions of Tegra CPUIDLE driver used a different "legacy"
+ * terminology for naming of the idling states, while this driver uses the
+ * new terminology.
+ *
+ * Mapping of the old terms into the new ones:
+ *
+ * Old | New
+ * ---------
+ * LP3 | C1	(CPU core clock gating)
+ * LP2 | C7	(CPU core power gating)
+ * LP2 | CC6	(CPU cluster power gating)
+ *
+ * Note that that the older CPUIDLE driver versions didn't explicitly
+ * differentiate the LP2 states because these states either used the same
+ * code path or because CC6 wasn't supported.
+ */
+static struct cpuidle_driver tegra_idle_driver = {
+	.name = "tegra_idle",
+	.states = {
+		[TEGRA_C1] = ARM_CPUIDLE_WFI_STATE_PWR(600),
+		[TEGRA_C7] = {
+			.enter			= tegra_cpuidle_enter,
+			.exit_latency		= 2000,
+			.target_residency	= 2200,
+			.power_usage		= 100,
+			.flags			= CPUIDLE_FLAG_TIMER_STOP |
+						  CPUIDLE_FLAG_RCU_IDLE,
+			.name			= "C7",
+			.desc			= "CPU core powered off",
+		},
+		[TEGRA_CC6] = {
+			.enter			= tegra_cpuidle_enter,
+			.exit_latency		= 5000,
+			.target_residency	= 10000,
+			.power_usage		= 0,
+			.flags			= CPUIDLE_FLAG_TIMER_STOP |
+						  CPUIDLE_FLAG_RCU_IDLE   |
+						  CPUIDLE_FLAG_COUPLED,
+			.name			= "CC6",
+			.desc			= "CPU cluster powered off",
+		},
+	},
+	.state_count = TEGRA_STATE_COUNT,
+	.safe_state_index = TEGRA_C1,
+};
+
+static inline void tegra_cpuidle_disable_state(enum tegra_state state)
+{
+	cpuidle_driver_state_disabled(&tegra_idle_driver, state, true);
+}
+
+/*
+ * Tegra20 HW appears to have a bug such that PCIe device interrupts, whether
+ * they are legacy IRQs or MSI, are lost when CC6 is enabled.  To work around
+ * this, simply disable CC6 if the PCI driver and DT node are both enabled.
+ */
+void tegra_cpuidle_pcie_irqs_in_use(void)
+{
+	struct cpuidle_state *state_cc6 = &tegra_idle_driver.states[TEGRA_CC6];
+
+	if ((state_cc6->flags & CPUIDLE_FLAG_UNUSABLE) ||
+	    tegra_get_chip_id() != TEGRA20)
+		return;
+
+	pr_info("disabling CC6 state, since PCIe IRQs are in use\n");
+	tegra_cpuidle_disable_state(TEGRA_CC6);
+}
+EXPORT_SYMBOL_GPL(tegra_cpuidle_pcie_irqs_in_use);
+
+static void tegra_cpuidle_setup_tegra114_c7_state(void)
+{
+	struct cpuidle_state *s = &tegra_idle_driver.states[TEGRA_C7];
+
+	s->enter_s2idle = tegra114_enter_s2idle;
+	s->target_residency = 1000;
+	s->exit_latency = 500;
+}
+
+static int tegra_cpuidle_probe(struct platform_device *pdev)
+{
+	if (tegra_pmc_get_suspend_mode() == TEGRA_SUSPEND_NOT_READY)
+		return -EPROBE_DEFER;
+
+	/* LP2 could be disabled in device-tree */
+	if (tegra_pmc_get_suspend_mode() < TEGRA_SUSPEND_LP2)
+		tegra_cpuidle_disable_state(TEGRA_CC6);
+
+	/*
+	 * Required suspend-resume functionality, which is provided by the
+	 * Tegra-arch core and PMC driver, is unavailable if PM-sleep option
+	 * is disabled.
+	 */
+	if (!IS_ENABLED(CONFIG_PM_SLEEP)) {
+		tegra_cpuidle_disable_state(TEGRA_C7);
+		tegra_cpuidle_disable_state(TEGRA_CC6);
+	}
+
+	/*
+	 * Generic WFI state (also known as C1 or LP3) and the coupled CPU
+	 * cluster power-off (CC6 or LP2) states are common for all Tegra SoCs.
+	 */
+	switch (tegra_get_chip_id()) {
+	case TEGRA20:
+		/* Tegra20 isn't capable to power-off individual CPU cores */
+		tegra_cpuidle_disable_state(TEGRA_C7);
+		break;
+
+	case TEGRA30:
+		break;
+
+	case TEGRA114:
+	case TEGRA124:
+		tegra_cpuidle_setup_tegra114_c7_state();
+
+		/* coupled CC6 (LP2) state isn't implemented yet */
+		tegra_cpuidle_disable_state(TEGRA_CC6);
+		break;
+
+	default:
+		return -EINVAL;
+	}
+
+	return cpuidle_register(&tegra_idle_driver, cpu_possible_mask);
+}
+
+static struct platform_driver tegra_cpuidle_driver = {
+	.probe = tegra_cpuidle_probe,
+	.driver = {
+		.name = "tegra-cpuidle",
+	},
+};
+builtin_platform_driver(tegra_cpuidle_driver);