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, 361 insertions, 0 deletions

diff --git a/scripts/dtc/dtx_diff b/scripts/dtc/dtx_diff
new file mode 100755
index 000000000..f2bbde4bb
--- /dev/null
+++ b/scripts/dtc/dtx_diff
@@ -0,0 +1,361 @@
+#! /bin/bash
+# SPDX-License-Identifier: GPL-2.0-only
+
+# Copyright (C) 2015 Frank Rowand
+#
+
+
+usage() {
+
+	# use spaces instead of tabs in the usage message
+	cat >&2 <<eod
+
+Usage:
+
+   `basename $0` DTx
+        decompile DTx
+
+   `basename $0` DTx_1 DTx_2
+        diff DTx_1 and DTx_2
+
+
+      --annotate    synonym for -T
+      --color       synonym for -c (requires diff with --color support)
+       -c           enable colored output
+       -f           print full dts in diff (--unified=99999)
+       -h           synonym for --help
+       -help        synonym for --help
+      --help        print this message and exit
+       -s SRCTREE   linux kernel source tree is at path SRCTREE
+                        (default is current directory)
+       -S           linux kernel source tree is at root of current git repo
+       -T           annotate output .dts with input source file and line
+                        (-T -T for more details)
+       -u           unsorted, do not sort DTx
+
+
+Each DTx is processed by the dtc compiler to produce a sorted dts source
+file.  If DTx is a dts source file then it is pre-processed in the same
+manner as done for the compile of the dts source file in the Linux kernel
+build system ('#include' and '/include/' directives are processed).
+
+If two DTx are provided, the resulting dts source files are diffed.
+
+If DTx is a directory, it is treated as a DT subtree, such as
+  /proc/device-tree.
+
+If DTx contains the binary blob magic value in the first four bytes,
+  it is treated as a binary blob (aka .dtb or FDT).
+
+Otherwise DTx is treated as a dts source file (aka .dts).
+
+   If this script is not run from the root of the linux source tree,
+   and DTx utilizes '#include' or '/include/' then the path of the
+   linux source tree can be provided by '-s SRCTREE' or '-S' so that
+   include paths will be set properly.
+
+   The shell variable \${ARCH} must provide the architecture containing
+   the dts source file for include paths to be set properly for '#include'
+   or '/include/' to be processed.
+
+   If DTx_1 and DTx_2 are in different architectures, then this script
+   may not work since \${ARCH} is part of the include path.  The following
+   workaround can be used:
+
+      `basename $0` ARCH=arch_of_dtx_1 DTx_1 >tmp_dtx_1.dts
+      `basename $0` ARCH=arch_of_dtx_2 DTx_2 >tmp_dtx_2.dts
+      `basename $0` tmp_dtx_1.dts tmp_dtx_2.dts
+      rm tmp_dtx_1.dts tmp_dtx_2.dts
+
+   If DTx_1 and DTx_2 are in different directories, then this script will
+   add the path of DTx_1 and DTx_2 to the include paths.  If DTx_2 includes
+   a local file that exists in both the path of DTx_1 and DTx_2 then the
+   file in the path of DTx_1 will incorrectly be included.  Possible
+   workaround:
+
+      `basename $0` DTx_1 >tmp_dtx_1.dts
+      `basename $0` DTx_2 >tmp_dtx_2.dts
+      `basename $0` tmp_dtx_1.dts tmp_dtx_2.dts
+      rm tmp_dtx_1.dts tmp_dtx_2.dts
+
+eod
+}
+
+
+compile_to_dts() {
+
+	dtx="$1"
+	dtc_include="$2"
+
+	if [ -d "${dtx}" ] ; then
+
+		# -----  input is file tree
+
+		if ( ! ${DTC} -I fs ${dtx} ) ; then
+			exit 3
+		fi
+
+	elif [ -f "${dtx}" ] && [ -r "${dtx}" ] ; then
+
+		magic=`hexdump -n 4 -e '/1 "%02x"' ${dtx}`
+		if [ "${magic}" = "d00dfeed" ] ; then
+
+			# -----  input is FDT (binary blob)
+
+			if ( ! ${DTC} -I dtb ${dtx} ) ; then
+				exit 3
+			fi
+
+			return
+
+		fi
+
+		# -----  input is DTS (source)
+
+		if ( cpp ${cpp_flags} -x assembler-with-cpp ${dtx} \
+			| ${DTC} ${dtc_include} -I dts ) ; then
+			return
+		fi
+
+		echo ""                                                      >&2
+		echo "Possible hints to resolve the above error:"            >&2
+		echo "  (hints might not fix the problem)"                   >&2
+
+		hint_given=0
+
+		if [ "${ARCH}" = "" ] ; then
+			hint_given=1
+			echo ""                                              >&2
+			echo "  shell variable \$ARCH not set"               >&2
+		fi
+
+		dtx_arch=`echo "/${dtx}" | sed -e 's|.*/arch/||' -e 's|/.*||'`
+
+		if [ "${dtx_arch}" != ""  -a "${dtx_arch}" != "${ARCH}" ] ; then
+			hint_given=1
+			echo ""                                              >&2
+			echo "  architecture ${dtx_arch} is in file path,"   >&2
+			echo "  but does not match shell variable \$ARCH"    >&2
+			echo "  >>\$ARCH<< is: >>${ARCH}<<"                  >&2
+		fi
+
+		if [ ! -d ${srctree}/arch/${ARCH} ] ; then
+			hint_given=1
+			echo ""                                              >&2
+			echo "  ${srctree}/arch/${ARCH}/ does not exist"     >&2
+			echo "  Is \$ARCH='${ARCH}' correct?"                >&2
+			echo "  Possible fix: use '-s' option"               >&2
+
+			git_root=`git rev-parse --show-toplevel 2>/dev/null`
+			if [ -d ${git_root}/arch/ ] ; then
+				echo "  Possible fix: use '-S' option"       >&2
+			fi
+		fi
+
+		if [ $hint_given = 0 ] ; then
+			echo ""                                              >&2
+			echo "  No hints available."                         >&2
+		fi
+
+		echo ""                                                      >&2
+
+		exit 3
+
+	else
+		echo ""                                                     >&2
+		echo "ERROR: ${dtx} does not exist or is not readable"      >&2
+		echo ""                                                     >&2
+		exit 2
+	fi
+
+}
+
+
+# -----  start of script
+
+annotate=""
+cmd_diff=0
+diff_flags="-u"
+diff_color=""
+dtx_file_1=""
+dtx_file_2=""
+dtc_sort="-s"
+help=0
+srctree=""
+
+
+while [ $# -gt 0 ] ; do
+
+	case $1 in
+
+	-c | --color )
+		if diff --color /dev/null /dev/null 2>/dev/null ; then
+			diff_color="--color=always"
+		fi
+		shift
+		;;
+
+	-f )
+		diff_flags="--unified=999999"
+		shift
+		;;
+
+	-h | -help | --help )
+		help=1
+		shift
+		;;
+
+	-s )
+		srctree="$2"
+		shift 2
+		;;
+
+	-S )
+		git_root=`git rev-parse --show-toplevel 2>/dev/null`
+		srctree="${git_root}"
+		shift
+		;;
+
+	-T | --annotate )
+		if [ "${annotate}"  = "" ] ; then
+			annotate="-T"
+		elif [ "${annotate}"  = "-T" ] ; then
+			annotate="-T -T"
+		fi
+		shift
+		;;
+	-u )
+		dtc_sort=""
+		shift
+		;;
+
+	*)
+		if [ "${dtx_file_1}"  = "" ] ; then
+			dtx_file_1="$1"
+		elif [ "${dtx_file_2}" = "" ] ; then
+			dtx_file_2="$1"
+		else
+			echo ""                                             >&2
+			echo "ERROR: Unexpected parameter: $1"              >&2
+			echo ""                                             >&2
+			exit 2
+		fi
+		shift
+		;;
+
+	esac
+
+done
+
+if [ "${srctree}" = "" ] ; then
+	srctree="."
+fi
+
+if [ "${dtx_file_2}" != "" ]; then
+	cmd_diff=1
+fi
+
+if (( ${help} )) ; then
+	usage
+	exit 1
+fi
+
+# this must follow check for ${help}
+if [ "${dtx_file_1}" = "" ]; then
+	echo ""                                                             >&2
+	echo "ERROR: parameter DTx required"                                >&2
+	echo ""                                                             >&2
+	exit 2
+fi
+
+
+# -----  prefer dtc from linux kernel, allow fallback to dtc in $PATH
+
+if [ "${KBUILD_OUTPUT:0:2}" = ".." ] ; then
+	__KBUILD_OUTPUT="${srctree}/${KBUILD_OUTPUT}"
+elif [ "${KBUILD_OUTPUT}" = "" ] ; then
+	__KBUILD_OUTPUT="."
+else
+	__KBUILD_OUTPUT="${KBUILD_OUTPUT}"
+fi
+
+DTC="${__KBUILD_OUTPUT}/scripts/dtc/dtc"
+
+if [ ! -x ${DTC} ] ; then
+	__DTC="dtc"
+	if grep -q "^CONFIG_DTC=y" ${__KBUILD_OUTPUT}/.config 2>/dev/null; then
+		make_command='
+         make scripts'
+	else
+		make_command='
+         Enable CONFIG_DTC in the kernel configuration
+         make scripts'
+	fi
+	if ( ! which ${__DTC} >/dev/null ) ; then
+
+		# use spaces instead of tabs in the error message
+		cat >&2 <<eod
+
+ERROR: unable to find a 'dtc' program
+
+   Preferred 'dtc' (built from Linux kernel source tree) was not found or
+   is not executable.
+
+      'dtc' is: ${DTC}
+
+      If it does not exist, create it from the root of the Linux source tree:
+${make_command}
+
+      If not at the root of the Linux kernel source tree -s SRCTREE or -S
+      may need to be specified to find 'dtc'.
+
+      If 'O=\${dir}' is specified in your Linux builds, this script requires
+      'export KBUILD_OUTPUT=\${dir}' or add \${dir}/scripts/dtc to \$PATH
+      before running.
+
+      If \${KBUILD_OUTPUT} is a relative path, then '-s SRCDIR', -S, or run
+      this script from the root of the Linux kernel source tree is required.
+
+   Fallback '${__DTC}' was also not in \${PATH} or is not executable.
+
+eod
+		exit 2
+	fi
+	DTC=${__DTC}
+fi
+
+
+# -----  cpp and dtc flags same as for linux source tree build of .dtb files,
+#        plus directories of the dtx file(s)
+
+dtx_path_1_dtc_include="-i `dirname ${dtx_file_1}`"
+
+dtx_path_2_dtc_include=""
+if (( ${cmd_diff} )) ; then
+	dtx_path_2_dtc_include="-i `dirname ${dtx_file_2}`"
+fi
+
+cpp_flags="\
+	-nostdinc                                  \
+	-I${srctree}/scripts/dtc/include-prefixes  \
+	-undef -D__DTS__"
+
+DTC="\
+	${DTC}                                     \
+	-i ${srctree}/scripts/dtc/include-prefixes \
+	-O dts -qq -f ${dtc_sort} ${annotate} -o -"
+
+
+# -----  do the diff or decompile
+
+if (( ${cmd_diff} )) ; then
+
+	diff ${diff_flags} ${diff_color} --label "${dtx_file_1}" --label "${dtx_file_2}" \
+		<(compile_to_dts "${dtx_file_1}" "${dtx_path_1_dtc_include}") \
+		<(compile_to_dts "${dtx_file_2}" "${dtx_path_2_dtc_include}")
+
+else
+
+	compile_to_dts "${dtx_file_1}" "${dtx_path_1_dtc_include}"
+
+fi