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, 451 insertions, 0 deletions

diff --git a/scripts/bash-completion/make b/scripts/bash-completion/make
new file mode 100644
index 000000000..42e8dcead
--- /dev/null
+++ b/scripts/bash-completion/make
@@ -0,0 +1,451 @@
+# SPDX-License-Identifier: GPL-2.0-only
+# bash completion for GNU make with kbuild extension       -*- shell-script -*-
+
+# Load the default completion script for make. It is typically located at
+# /usr/share/bash-completion/completions/make, but we do not rely on it.
+__kbuild_load_default_make_completion()
+{
+	local -a dirs=("${BASH_COMPLETION_USER_DIR:-${XDG_DATA_HOME:-$HOME/.local/share}/bash-completion}/completions")
+	local ifs=$IFS IFS=: dir compfile this_dir
+
+	for dir in ${XDG_DATA_DIRS:-/usr/local/share:/usr/share}; do
+	        dirs+=("$dir"/bash-completion/completions)
+	done
+	IFS=$ifs
+
+	this_dir="$(realpath "$(dirname "${BASH_SOURCE[0]}")")"
+
+	for dir in "${dirs[@]}"; do
+		if [[ ! -d ${dir} || ${dir} = "${this_dir}" ]]; then
+			continue
+		fi
+
+		for compfile in make make.bash _make; do
+			compfile=$dir/$compfile
+			# Avoid trying to source dirs; https://bugzilla.redhat.com/903540
+			if [[ -f ${compfile} ]] && . "${compfile}" &>/dev/null; then
+
+				__kbuild_default_make_completion=$(
+					# shellcheck disable=SC2046 # word splitting is the point here
+					set -- $(complete -p make)
+
+					while [[ $# -gt 1 && "$1" != -F ]]; do
+						shift
+					done
+
+					if [[ "$1" = -F ]]; then
+						echo "$2"
+					fi
+				)
+
+				return
+			fi
+		done
+	done
+}
+
+__kbuild_load_default_make_completion
+
+__kbuild_handle_variable()
+{
+	local var=${1%%=*}
+	local cur=${cur#"${var}"=}
+	local srctree=$2
+	local keywords=()
+
+	case $var in
+	ARCH)
+		# sub-directories under arch/
+		keywords+=($(find "${srctree}/arch" -mindepth 1 -maxdepth 1 -type d -printf '%P\n'))
+		# architectures hard-coded in the top Makefile
+		keywords+=(i386 x86_64 sparc32 sparc64 parisc64)
+		;;
+	CROSS_COMPILE)
+		# toolchains with a full path
+		local cross_compile=()
+		local c c2
+		_filedir
+
+		for c in "${COMPREPLY[@]}"; do
+			# eval for tilde expansion
+			# suppress error, as this fails when it contains a space
+			eval "c2=${c}" 2>/dev/null || continue
+			if [[ ${c} == *-elfedit && ! -d ${c2} && -x ${c2} ]]; then
+				cross_compile+=("${c%elfedit}")
+			fi
+		done
+
+		# toolchains in the PATH environment
+		while read -r c; do
+			if [[ ${c} == *-elfedit ]]; then
+				keywords+=("${c%elfedit}")
+			fi
+		done < <(compgen -c)
+
+		COMPREPLY=()
+		_filedir -d
+
+		# Add cross_compile directly without passing it to compgen.
+		# Otherwise, toolchain paths with a tilde do not work.
+		# e.g.)
+		#   CROSS_COMPILE=~/0day/gcc-14.2.0-nolibc/aarch64-linux/bin/aarch64-linux-
+		COMPREPLY+=("${cross_compile[@]}")
+		;;
+	LLVM)
+		# LLVM=1 uses the default 'clang' etc.
+		keywords+=(1)
+
+		# suffix for a particular version. LLVM=-18 uses 'clang-18' etc.
+		while read -r c; do
+			if [[ ${c} == clang-[0-9]* ]]; then
+				keywords+=("${c#clang}")
+			fi
+		done < <(compgen -c)
+
+		# directory path to LLVM toolchains
+		_filedir -d
+		;;
+	KCONFIG_ALLCONFIG)
+		# KCONFIG_ALLCONFIG=1 selects the default fragment
+		keywords+=(1)
+		# or the path to a fragment file
+		_filedir
+		;;
+	C | KBUILD_CHECKSRC)
+		keywords+=(1 2)
+		;;
+	V | KBUILD_VERBOSE)
+		keywords+=({,1}{,2})
+		;;
+	W | KBUILD_EXTRA_WARN)
+		keywords+=({,1}{,2}{,3}{,c}{,e})
+		;;
+	KBUILD_ABS_SRCTREE | KBUILD_MODPOST_NOFINAL | KBUILD_MODPOST_WARN | \
+		CLIPPY | KBUILD_CLIPPY | KCONFIG_NOSILENTUPDATE | \
+		KCONFIG_OVERWRITECONFIG | KCONFIG_WARN_UNKNOWN_SYMBOL | \
+		KCONFIG_WERROR )
+		keywords+=(1)
+		;;
+	INSTALL_MOD_STRIP)
+		keywords+=(1 --strip-debug --strip-unneeded)
+		;;
+	O | KBUILD_OUTPUT | M | KBUILD_EXTMOD | MO | KBUILD_EXTMOD_OUTPUT | *_PATH)
+		# variables that take a directory.
+		_filedir -d
+		return
+		;;
+	KBUILD_EXTRA_SYMBOL | KBUILD_KCONFIG | KCONFIG_CONFIG)
+		# variables that take a file.
+		_filedir
+		return
+	esac
+
+	COMPREPLY+=($(compgen -W "${keywords[*]}" -- "${cur}"))
+}
+
+# Check the -C, -f options and 'source' symlink. Return the source tree we are
+# working in.
+__kbuild_get_srctree()
+{
+	local words=("$@")
+	local cwd makef_dir
+
+	# see if a path was specified with -C/--directory
+	for ((i = 1; i < ${#words[@]}; i++)); do
+		if [[ ${words[i]} == -@(C|-directory) ]]; then
+			# eval for tilde expansion.
+			# suppress error, as this fails when it contains a space
+			eval "cwd=${words[i + 1]}" 2>/dev/null
+			break
+		fi
+	done
+
+	if [[ -z ${cwd} ]]; then
+		cwd=.
+	fi
+
+	# see if a Makefile was specified with -f/--file/--makefile
+	for ((i = 1; i < ${#words[@]}; i++)); do
+		if [[ ${words[i]} == -@(f|-?(make)file) ]]; then
+			# eval for tilde expansion
+			# suppress error, as this fails when it contains a space
+			eval "makef_dir=${words[i + 1]%/*}" 2>/dev/null
+			break
+		fi
+	done
+
+	if [ -z "${makef_dir}" ]; then
+		makef_dir=${cwd}
+	elif [[ ${makef_dir} != /* ]]; then
+		makef_dir=${cwd}/${makef_dir}
+	fi
+
+	# If ${makef_dir} is a build directory created by the O= option, there
+	# is a symbolic link 'source', which points to the kernel source tree.
+	if [[ -L ${makef_dir}/source ]]; then
+		makef_dir=$(readlink "${makef_dir}/source")
+	fi
+
+	echo "${makef_dir}"
+}
+
+# Get SRCARCH to do a little more clever things
+__kbuild_get_srcarch()
+{
+	local words=("$@")
+	local arch srcarch uname_m
+
+	# see if ARCH= is explicitly specified
+	for ((i = 1; i < ${#words[@]}; i++)); do
+		if [[ ${words[i]} == ARCH=* ]]; then
+			arch=${words[i]#ARCH=}
+			break
+		fi
+	done
+
+	# If ARCH= is not specified, check the build marchine's architecture
+	if [[ -z ${arch} ]]; then
+		uname_m=$(uname -m)
+
+		# shellcheck disable=SC2209 # 'sh' is SuperH, not a shell command
+		case ${uname_m} in
+		arm64 | aarch64*) arch=arm64 ;;
+		arm* | sa110)     arch=arm ;;
+		i?86 | x86_64)    arch=x86 ;;
+		loongarch*)       arch=loongarch ;;
+		mips*)            arch=mips ;;
+		ppc*)             arch=powerpc ;;
+		riscv*)           arch=riscv ;;
+		s390x)            arch=s390 ;;
+		sh[234]*)         arch=sh ;;
+		sun4u)            arch=sparc64 ;;
+		*)                arch=${uname_m} ;;
+		esac
+	fi
+
+	case ${arch} in
+		parisc64)          srcarch=parisc ;;
+		sparc32 | sparc64) srcarch=sparc ;;
+		i386 | x86_64)     srcarch=x86 ;;
+		*)                 srcarch=${arch} ;;
+	esac
+
+	echo "$srcarch"
+}
+
+# small Makefile to parse obj-* syntax
+__kbuild_tmp_makefile()
+{
+cat <<'EOF'
+.PHONY: __default
+__default:
+	$(foreach m,$(obj-y) $(obj-m) $(obj-),$(foreach s, -objs -y -m -,$($(m:%.o=%$s))) $(m))
+EOF
+echo "include ${1}"
+}
+
+_make_for_kbuild ()
+{
+	# shellcheck disable=SC2034 # these are set by _init_completion
+	local cur prev words cword split
+	_init_completion -s || return
+
+	local srctree
+	srctree=$(__kbuild_get_srctree "${words[@]}")
+
+	# If 'kernel' and 'Documentation' directories are found, we assume this
+	# is a kernel tree. Otherwise, we fall back to the generic rule provided
+	# by the bash-completion project.
+	if [[ ! -d ${srctree}/kernel || ! -d ${srctree}/Documentation ]]; then
+		if [ -n "${__kbuild_default_make_completion}" ]; then
+			"${__kbuild_default_make_completion}" "$@"
+		fi
+		return
+	fi
+
+	# make options with a parameter (copied from the bash-completion project)
+	case ${prev} in
+	--file | --makefile | --old-file | --assume-old | --what-if | --new-file | \
+		--assume-new | -!(-*)[foW])
+		_filedir
+		return
+		;;
+	--include-dir | --directory | -!(-*)[ICm])
+		_filedir -d
+		return
+		;;
+	-!(-*)E)
+		COMPREPLY=($(compgen -v -- "$cur"))
+		return
+		;;
+	--eval | -!(-*)[DVx])
+		return
+		;;
+	--jobs | -!(-*)j)
+		COMPREPLY=($(compgen -W "{1..$(($(_ncpus) * 2))}" -- "$cur"))
+		return
+		;;
+	esac
+
+	local keywords=()
+
+	case ${cur} in
+	-*)
+		# make options (copied from the bash-completion project)
+		local opts
+		opts="$(_parse_help "$1")"
+		COMPREPLY=($(compgen -W "${opts:-$(_parse_usage "$1")}" -- "$cur"))
+		if [[ ${COMPREPLY-} == *= ]]; then
+			compopt -o nospace
+		fi
+		return
+		;;
+	*=*)
+		__kbuild_handle_variable "${cur}" "${srctree}"
+		return
+		;;
+	KBUILD_*)
+		# There are many variables prefixed with 'KBUILD_'.
+		# Display them only when 'KBUILD_' is entered.
+		# shellcheck disable=SC2191 # '=' is appended for variables
+		keywords+=(
+			KBUILD_{CHECKSRC,EXTMOD,EXTMOD_OUTPUT,OUTPUT,VERBOSE,EXTRA_WARN,CLIPPY}=
+			KBUILD_BUILD_{USER,HOST,TIMESTAMP}=
+			KBUILD_MODPOST_{NOFINAL,WARN}=
+			KBUILD_{ABS_SRCTREE,EXTRA_SYMBOLS,KCONFIG}=
+		)
+		;;
+	KCONFIG_*)
+		# There are many variables prefixed with 'KCONFIG_'.
+		# Display them only when 'KCONFIG_' is entered.
+		# shellcheck disable=SC2191 # '=' is appended for variables
+		keywords+=(
+			KCONFIG_{CONFIG,ALLCONFIG,NOSILENTUPDATE,OVERWRITECONFIG}=
+			KCONFIG_{SEED,PROBABILITY}=
+			KCONFIG_WARN_UNKNOWN_SYMBOL=
+			KCONFIG_WERROR=
+		)
+		;;
+	*)
+		# By default, hide KBUILD_* and KCONFIG_* variables.
+		# Instead, display only the prefix parts.
+		keywords+=(KBUILD_ KCONFIG_)
+		;;
+	esac
+
+	if [[ ${cur} != /* && ${cur} != *//* ]]; then
+		local dir srcarch kbuild_file tmp
+		srcarch=$(__kbuild_get_srcarch "${words[@]}")
+
+		# single build
+		dir=${cur}
+		while true; do
+			if [[ ${dir} == */* ]]; then
+				dir=${dir%/*}
+			else
+				dir=.
+			fi
+
+			# Search for 'Kbuild' or 'Makefile' in the parent
+			# directories (may not be a direct parent)
+			if [[ -f ${srctree}/${dir}/Kbuild ]]; then
+				kbuild_file=${srctree}/${dir}/Kbuild
+				break
+			fi
+			if [[ -f ${srctree}/${dir}/Makefile ]]; then
+				kbuild_file=${srctree}/${dir}/Makefile
+				break
+			fi
+
+			if [[ ${dir} == . ]]; then
+				break
+			fi
+		done
+
+		if [[ -n ${kbuild_file} ]]; then
+			tmp=($(__kbuild_tmp_makefile "${kbuild_file}" |
+			       SRCARCH=${srcarch} obj=${dir} src=${srctree}/${dir} \
+			       "${1}" -n -f - 2>/dev/null))
+
+			# Add $(obj)/ prefix
+			if [[ ${dir} != . ]]; then
+				tmp=("${tmp[@]/#/${dir}\/}")
+			fi
+
+			keywords+=("${tmp[@]}")
+		fi
+
+		# *_defconfig and *.config files. These might be grouped into
+		# subdirectories, e.g., arch/powerpc/configs/*/*_defconfig.
+		if [[ ${cur} == */* ]]; then
+			dir=${cur%/*}
+		else
+			dir=.
+		fi
+
+		tmp=($(find "${srctree}/arch/${srcarch}/configs/${dir}" \
+		       "${srctree}/kernel/configs/${dir}" \
+		       -mindepth 1 -maxdepth 1 -type d -printf '%P/\n' \
+		       -o -printf '%P\n' 2>/dev/null))
+
+		if [[ ${dir} != . ]]; then
+			tmp=("${tmp[@]/#/${dir}\/}")
+		fi
+
+		keywords+=("${tmp[@]}")
+	fi
+
+	# shellcheck disable=SC2191 # '=' is appended for variables
+	keywords+=(
+		#
+		# variables (append =)
+		#
+		ARCH=
+		CROSS_COMPILE=
+		LLVM=
+		C= M= MO= O= V= W=
+		INSTALL{,_MOD,_HDR,_DTBS}_PATH=
+		KERNELRELEASE=
+
+		#
+		# targets
+		#
+		all help
+		clean mrproper distclean
+		clang-{tidy,analyzer} compile_commands.json
+		coccicheck
+		dtbs{,_check,_install} dt_binding_{check,schemas}
+		headers{,_install}
+		vmlinux install
+		modules{,_prepare,_install,_sign}
+		vdso_install
+		tags TAGS cscope gtags
+		rust{available,fmt,fmtcheck}
+		kernel{version,release} image_name
+		kselftest{,-all,-install,-clean,-merge}
+
+		# configuration
+		{,old,olddef,sync,def,savedef,rand,listnew,helpnew,test,tiny}config
+		{,build_}{menu,n,g,x}config
+		local{mod,yes}config
+		all{no,yes,mod,def}config
+		{yes2mod,mod2yes,mod2no}config
+
+		# docs
+		{html,textinfo,info,latex,pdf,epub,xml,linkcheck,refcheck,clean}docs
+
+		# package
+		{,bin,src}{rpm,deb}-pkg
+		{pacman,dir,tar}-pkg
+		tar{,gz,bz2,xz,zst}-pkg
+		perf-tar{,gz,bz2,xz,zst}-src-pkg
+	)
+
+	COMPREPLY=($(compgen -W "${keywords[*]}" -- "${cur}"))
+
+	# Do not append a space for variables, subdirs, "KBUILD_", "KCONFIG_".
+	if [[ ${COMPREPLY-} == *[=/] || ${COMPREPLY-} =~ ^(KBUILD|KCONFIG)_$ ]]; then
+		compopt -o nospace
+	fi
+
+} && complete -F _make_for_kbuild make