We found an issue with null pointer access due to kprobe debug exception error handling on 5.10, and I proposed a separate fix patch for 5.10, see [1]. But as Greg gave advice, we always choose to backport relevant patches from upstream to fix issues with stable kernels, so I made this patch set.
The main one we need to backport is patch 5, which uses int3 instead of debug trap for single-stepping, thus avoiding the problems we encountered with kprobe debug exception error handling. Patches 1-4 are pre-patches, and patches 6-9 are fixes for patch 5. The major modifications are patch 2 and patch 5. Patch 2 optimizes resume_execution() to avoid repeated instruction decoding, and patch 5 uses int3 instead of debug trap, and as Masami said in the commit message this patch will change some behavior of kprobe, but it has almost no effect on the actual usage.
Please let me know if there are any problems, thanks!
[1] https://lore.kernel.org/lkml/20230630020845.227939-1-lihuafei1@huawei.com/
Gustavo A. R. Silva (1): kprobes/x86: Fix fall-through warnings for Clang
Masami Hiramatsu (5): x86/kprobes: Do not decode opcode in resume_execution() x86/kprobes: Retrieve correct opcode for group instruction x86/kprobes: Identify far indirect JMP correctly x86/kprobes: Use int3 instead of debug trap for single-step x86/kprobes: Fix to identify indirect jmp and others using range case
Masami Hiramatsu (Google) (1): x86/kprobes: Update kcb status flag after singlestepping
Nadav Amit (1): x86/kprobes: Fix JNG/JNLE emulation
Wei Yongjun (1): x86/kprobes: Move 'inline' to the beginning of the kprobe_is_ss() declaration
arch/x86/include/asm/kprobes.h | 24 +- arch/x86/kernel/kprobes/core.c | 639 ++++++++++++++++++++------------- arch/x86/kernel/traps.c | 3 - 3 files changed, 409 insertions(+), 257 deletions(-)