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/base/power/runtime-test.c | 249 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 249 insertions(+) create mode 100644 drivers/base/power/runtime-test.c (limited to 'drivers/base/power/runtime-test.c') diff --git a/drivers/base/power/runtime-test.c b/drivers/base/power/runtime-test.c new file mode 100644 index 000000000..1535ad2b0 --- /dev/null +++ b/drivers/base/power/runtime-test.c @@ -0,0 +1,249 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Copyright 2025 Google, Inc. + */ + +#include +#include +#include +#include + +#define DEVICE_NAME "pm_runtime_test_device" + +static void pm_runtime_depth_test(struct kunit *test) +{ + struct device *dev = kunit_device_register(test, DEVICE_NAME); + + KUNIT_ASSERT_NOT_ERR_OR_NULL(test, dev); + + pm_runtime_enable(dev); + + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); + KUNIT_EXPECT_EQ(test, 0, pm_runtime_get_sync(dev)); + KUNIT_EXPECT_TRUE(test, pm_runtime_active(dev)); + KUNIT_EXPECT_EQ(test, 1, pm_runtime_get_sync(dev)); /* "already active" */ + KUNIT_EXPECT_EQ(test, 0, pm_runtime_put_sync(dev)); + KUNIT_EXPECT_EQ(test, 0, pm_runtime_put_sync(dev)); + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); +} + +/* Test pm_runtime_put() and friends when already suspended. */ +static void pm_runtime_already_suspended_test(struct kunit *test) +{ + struct device *dev = kunit_device_register(test, DEVICE_NAME); + + KUNIT_ASSERT_NOT_ERR_OR_NULL(test, dev); + + pm_runtime_enable(dev); + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); + + pm_runtime_get_noresume(dev); + KUNIT_EXPECT_EQ(test, 1, pm_runtime_put_sync(dev)); + + KUNIT_EXPECT_EQ(test, 1, pm_runtime_suspend(dev)); + KUNIT_EXPECT_EQ(test, 1, pm_runtime_autosuspend(dev)); + KUNIT_EXPECT_EQ(test, 1, pm_request_autosuspend(dev)); + + pm_runtime_get_noresume(dev); + KUNIT_EXPECT_EQ(test, 1, pm_runtime_put_sync_autosuspend(dev)); + + pm_runtime_get_noresume(dev); + pm_runtime_put_autosuspend(dev); + + /* Grab 2 refcounts */ + pm_runtime_get_noresume(dev); + pm_runtime_get_noresume(dev); + /* The first put() sees usage_count 1 */ + KUNIT_EXPECT_EQ(test, 0, pm_runtime_put_sync_autosuspend(dev)); + /* The second put() sees usage_count 0 but tells us "already suspended". */ + KUNIT_EXPECT_EQ(test, 1, pm_runtime_put_sync_autosuspend(dev)); + + /* Should have remained suspended the whole time. */ + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); +} + +static void pm_runtime_idle_test(struct kunit *test) +{ + struct device *dev = kunit_device_register(test, DEVICE_NAME); + + KUNIT_ASSERT_NOT_ERR_OR_NULL(test, dev); + + pm_runtime_enable(dev); + + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); + KUNIT_EXPECT_EQ(test, 0, pm_runtime_get_sync(dev)); + KUNIT_EXPECT_TRUE(test, pm_runtime_active(dev)); + KUNIT_EXPECT_EQ(test, -EAGAIN, pm_runtime_idle(dev)); + KUNIT_EXPECT_TRUE(test, pm_runtime_active(dev)); + pm_runtime_put_noidle(dev); + KUNIT_EXPECT_TRUE(test, pm_runtime_active(dev)); + KUNIT_EXPECT_EQ(test, 0, pm_runtime_idle(dev)); + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); + KUNIT_EXPECT_EQ(test, -EAGAIN, pm_runtime_idle(dev)); + KUNIT_EXPECT_EQ(test, -EAGAIN, pm_request_idle(dev)); +} + +static void pm_runtime_disabled_test(struct kunit *test) +{ + struct device *dev = kunit_device_register(test, DEVICE_NAME); + + KUNIT_ASSERT_NOT_ERR_OR_NULL(test, dev); + + /* Never called pm_runtime_enable() */ + KUNIT_EXPECT_FALSE(test, pm_runtime_enabled(dev)); + + /* "disabled" is treated as "active" */ + KUNIT_EXPECT_TRUE(test, pm_runtime_active(dev)); + KUNIT_EXPECT_FALSE(test, pm_runtime_suspended(dev)); + + /* + * Note: these "fail", but they still acquire/release refcounts, so + * keep them balanced. + */ + KUNIT_EXPECT_EQ(test, -EACCES, pm_runtime_get(dev)); + pm_runtime_put(dev); + + KUNIT_EXPECT_EQ(test, -EACCES, pm_runtime_get_sync(dev)); + KUNIT_EXPECT_EQ(test, -EACCES, pm_runtime_put_sync(dev)); + + KUNIT_EXPECT_EQ(test, -EACCES, pm_runtime_get(dev)); + pm_runtime_put_autosuspend(dev); + + KUNIT_EXPECT_EQ(test, -EACCES, pm_runtime_resume_and_get(dev)); + KUNIT_EXPECT_EQ(test, -EACCES, pm_runtime_idle(dev)); + KUNIT_EXPECT_EQ(test, -EACCES, pm_request_idle(dev)); + KUNIT_EXPECT_EQ(test, -EACCES, pm_request_resume(dev)); + KUNIT_EXPECT_EQ(test, -EACCES, pm_request_autosuspend(dev)); + KUNIT_EXPECT_EQ(test, -EACCES, pm_runtime_suspend(dev)); + KUNIT_EXPECT_EQ(test, -EACCES, pm_runtime_resume(dev)); + KUNIT_EXPECT_EQ(test, -EACCES, pm_runtime_autosuspend(dev)); + + /* Still disabled */ + KUNIT_EXPECT_TRUE(test, pm_runtime_active(dev)); + KUNIT_EXPECT_FALSE(test, pm_runtime_enabled(dev)); +} + +static void pm_runtime_error_test(struct kunit *test) +{ + struct device *dev = kunit_device_register(test, DEVICE_NAME); + + KUNIT_ASSERT_NOT_ERR_OR_NULL(test, dev); + + pm_runtime_enable(dev); + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); + + /* Fake a .runtime_resume() error */ + dev->power.runtime_error = -EIO; + + /* + * Note: these "fail", but they still acquire/release refcounts, so + * keep them balanced. + */ + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_get(dev)); + pm_runtime_put(dev); + + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_get_sync(dev)); + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_put_sync(dev)); + + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_get(dev)); + pm_runtime_put_autosuspend(dev); + + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_get(dev)); + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_put_sync_autosuspend(dev)); + + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_resume_and_get(dev)); + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_idle(dev)); + KUNIT_EXPECT_EQ(test, -EINVAL, pm_request_idle(dev)); + KUNIT_EXPECT_EQ(test, -EINVAL, pm_request_resume(dev)); + KUNIT_EXPECT_EQ(test, -EINVAL, pm_request_autosuspend(dev)); + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_suspend(dev)); + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_resume(dev)); + KUNIT_EXPECT_EQ(test, -EINVAL, pm_runtime_autosuspend(dev)); + + /* Error is still pending */ + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); + KUNIT_EXPECT_EQ(test, -EIO, dev->power.runtime_error); + /* Clear error */ + KUNIT_EXPECT_EQ(test, 0, pm_runtime_set_suspended(dev)); + KUNIT_EXPECT_EQ(test, 0, dev->power.runtime_error); + /* Still suspended */ + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); + + KUNIT_EXPECT_EQ(test, 0, pm_runtime_get(dev)); + pm_runtime_barrier(dev); + pm_runtime_put(dev); + pm_runtime_suspend(dev); /* flush the put(), to suspend */ + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); + + KUNIT_EXPECT_EQ(test, 0, pm_runtime_get_sync(dev)); + KUNIT_EXPECT_EQ(test, 0, pm_runtime_put_sync(dev)); + + KUNIT_EXPECT_EQ(test, 0, pm_runtime_get_sync(dev)); + pm_runtime_put_autosuspend(dev); + + KUNIT_EXPECT_EQ(test, 0, pm_runtime_resume_and_get(dev)); + + /* + * The following should all return -EAGAIN (usage is non-zero) or 1 + * (already resumed). + */ + KUNIT_EXPECT_EQ(test, -EAGAIN, pm_runtime_idle(dev)); + KUNIT_EXPECT_EQ(test, -EAGAIN, pm_request_idle(dev)); + KUNIT_EXPECT_EQ(test, 1, pm_request_resume(dev)); + KUNIT_EXPECT_EQ(test, -EAGAIN, pm_request_autosuspend(dev)); + KUNIT_EXPECT_EQ(test, -EAGAIN, pm_runtime_suspend(dev)); + KUNIT_EXPECT_EQ(test, 1, pm_runtime_resume(dev)); + KUNIT_EXPECT_EQ(test, -EAGAIN, pm_runtime_autosuspend(dev)); + + KUNIT_EXPECT_EQ(test, 0, pm_runtime_put_sync(dev)); + + /* Suspended again */ + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); +} + +/* + * Explore a typical probe() sequence in which a device marks itself powered, + * but doesn't hold any runtime PM reference, so it suspends as soon as it goes + * idle. + */ +static void pm_runtime_probe_active_test(struct kunit *test) +{ + struct device *dev = kunit_device_register(test, DEVICE_NAME); + + KUNIT_ASSERT_NOT_ERR_OR_NULL(test, dev); + + KUNIT_EXPECT_TRUE(test, pm_runtime_status_suspended(dev)); + + KUNIT_EXPECT_EQ(test, 0, pm_runtime_set_active(dev)); + KUNIT_EXPECT_TRUE(test, pm_runtime_active(dev)); + + pm_runtime_enable(dev); + KUNIT_EXPECT_TRUE(test, pm_runtime_active(dev)); + + /* Nothing to flush. We stay active. */ + pm_runtime_barrier(dev); + KUNIT_EXPECT_TRUE(test, pm_runtime_active(dev)); + + /* Ask for idle? Now we suspend. */ + KUNIT_EXPECT_EQ(test, 0, pm_runtime_idle(dev)); + KUNIT_EXPECT_TRUE(test, pm_runtime_suspended(dev)); +} + +static struct kunit_case pm_runtime_test_cases[] = { + KUNIT_CASE(pm_runtime_depth_test), + KUNIT_CASE(pm_runtime_already_suspended_test), + KUNIT_CASE(pm_runtime_idle_test), + KUNIT_CASE(pm_runtime_disabled_test), + KUNIT_CASE(pm_runtime_error_test), + KUNIT_CASE(pm_runtime_probe_active_test), + {} +}; + +static struct kunit_suite pm_runtime_test_suite = { + .name = "pm_runtime_test_cases", + .test_cases = pm_runtime_test_cases, +}; + +kunit_test_suite(pm_runtime_test_suite); +MODULE_DESCRIPTION("Runtime power management unit test suite"); +MODULE_LICENSE("GPL"); -- cgit v1.3.1