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 --- scripts/markup_oops.pl | 366 +++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 366 insertions(+) create mode 100755 scripts/markup_oops.pl (limited to 'scripts/markup_oops.pl') diff --git a/scripts/markup_oops.pl b/scripts/markup_oops.pl new file mode 100755 index 000000000..e476caf52 --- /dev/null +++ b/scripts/markup_oops.pl @@ -0,0 +1,366 @@ +#!/usr/bin/env perl +# SPDX-License-Identifier: GPL-2.0-only + +use File::Basename; +use Math::BigInt; +use Getopt::Long; + +# Copyright 2008, Intel Corporation +# +# This file is part of the Linux kernel +# +# Authors: +# Arjan van de Ven + + +my $cross_compile = ""; +my $vmlinux_name = ""; +my $modulefile = ""; + +# Get options +Getopt::Long::GetOptions( + 'cross-compile|c=s' => \$cross_compile, + 'module|m=s' => \$modulefile, + 'help|h' => \&usage, +) || usage (); +my $vmlinux_name = $ARGV[0]; +if (!defined($vmlinux_name)) { + my $kerver = `uname -r`; + chomp($kerver); + $vmlinux_name = "/lib/modules/$kerver/build/vmlinux"; + print "No vmlinux specified, assuming $vmlinux_name\n"; +} +my $filename = $vmlinux_name; + +# Parse the oops to find the EIP value + +my $target = "0"; +my $function; +my $module = ""; +my $func_offset = 0; +my $vmaoffset = 0; + +my %regs; + + +sub parse_x86_regs +{ + my ($line) = @_; + if ($line =~ /EAX: ([0-9a-f]+) EBX: ([0-9a-f]+) ECX: ([0-9a-f]+) EDX: ([0-9a-f]+)/) { + $regs{"%eax"} = $1; + $regs{"%ebx"} = $2; + $regs{"%ecx"} = $3; + $regs{"%edx"} = $4; + } + if ($line =~ /ESI: ([0-9a-f]+) EDI: ([0-9a-f]+) EBP: ([0-9a-f]+) ESP: ([0-9a-f]+)/) { + $regs{"%esi"} = $1; + $regs{"%edi"} = $2; + $regs{"%esp"} = $4; + } + if ($line =~ /RAX: ([0-9a-f]+) RBX: ([0-9a-f]+) RCX: ([0-9a-f]+)/) { + $regs{"%eax"} = $1; + $regs{"%ebx"} = $2; + $regs{"%ecx"} = $3; + } + if ($line =~ /RDX: ([0-9a-f]+) RSI: ([0-9a-f]+) RDI: ([0-9a-f]+)/) { + $regs{"%edx"} = $1; + $regs{"%esi"} = $2; + $regs{"%edi"} = $3; + } + if ($line =~ /RBP: ([0-9a-f]+) R08: ([0-9a-f]+) R09: ([0-9a-f]+)/) { + $regs{"%r08"} = $2; + $regs{"%r09"} = $3; + } + if ($line =~ /R10: ([0-9a-f]+) R11: ([0-9a-f]+) R12: ([0-9a-f]+)/) { + $regs{"%r10"} = $1; + $regs{"%r11"} = $2; + $regs{"%r12"} = $3; + } + if ($line =~ /R13: ([0-9a-f]+) R14: ([0-9a-f]+) R15: ([0-9a-f]+)/) { + $regs{"%r13"} = $1; + $regs{"%r14"} = $2; + $regs{"%r15"} = $3; + } +} + +sub reg_name +{ + my ($reg) = @_; + $reg =~ s/r(.)x/e\1x/; + $reg =~ s/r(.)i/e\1i/; + $reg =~ s/r(.)p/e\1p/; + return $reg; +} + +sub process_x86_regs +{ + my ($line, $cntr) = @_; + my $str = ""; + if (length($line) < 40) { + return ""; # not an asm istruction + } + + # find the arguments to the instruction + if ($line =~ /([0-9a-zA-Z\,\%\(\)\-\+]+)$/) { + $lastword = $1; + } else { + return ""; + } + + # we need to find the registers that get clobbered, + # since their value is no longer relevant for previous + # instructions in the stream. + + $clobber = $lastword; + # first, remove all memory operands, they're read only + $clobber =~ s/\([a-z0-9\%\,]+\)//g; + # then, remove everything before the comma, thats the read part + $clobber =~ s/.*\,//g; + + # if this is the instruction that faulted, we haven't actually done + # the write yet... nothing is clobbered. + if ($cntr == 0) { + $clobber = ""; + } + + foreach $reg (keys(%regs)) { + my $clobberprime = reg_name($clobber); + my $lastwordprime = reg_name($lastword); + my $val = $regs{$reg}; + if ($val =~ /^[0]+$/) { + $val = "0"; + } else { + $val =~ s/^0*//; + } + + # first check if we're clobbering this register; if we do + # we print it with a =>, and then delete its value + if ($clobber =~ /$reg/ || $clobberprime =~ /$reg/) { + if (length($val) > 0) { + $str = $str . " $reg => $val "; + } + $regs{$reg} = ""; + $val = ""; + } + # now check if we're reading this register + if ($lastword =~ /$reg/ || $lastwordprime =~ /$reg/) { + if (length($val) > 0) { + $str = $str . " $reg = $val "; + } + } + } + return $str; +} + +# parse the oops +while () { + my $line = $_; + if ($line =~ /EIP: 0060:\[\<([a-z0-9]+)\>\]/) { + $target = $1; + } + if ($line =~ /RIP: 0010:\[\<([a-z0-9]+)\>\]/) { + $target = $1; + } + if ($line =~ /EIP is at ([a-zA-Z0-9\_]+)\+0x([0-9a-f]+)\/0x[a-f0-9]/) { + $function = $1; + $func_offset = $2; + } + if ($line =~ /RIP: 0010:\[\<[0-9a-f]+\>\] \[\<[0-9a-f]+\>\] ([a-zA-Z0-9\_]+)\+0x([0-9a-f]+)\/0x[a-f0-9]/) { + $function = $1; + $func_offset = $2; + } + + # check if it's a module + if ($line =~ /EIP is at ([a-zA-Z0-9\_]+)\+(0x[0-9a-f]+)\/0x[a-f0-9]+\W\[([a-zA-Z0-9\_\-]+)\]/) { + $module = $3; + } + if ($line =~ /RIP: 0010:\[\<[0-9a-f]+\>\] \[\<[0-9a-f]+\>\] ([a-zA-Z0-9\_]+)\+(0x[0-9a-f]+)\/0x[a-f0-9]+\W\[([a-zA-Z0-9\_\-]+)\]/) { + $module = $3; + } + parse_x86_regs($line); +} + +my $decodestart = Math::BigInt->from_hex("0x$target") - Math::BigInt->from_hex("0x$func_offset"); +my $decodestop = Math::BigInt->from_hex("0x$target") + 8192; +if ($target eq "0") { + print "No oops found!\n"; + usage(); +} + +# if it's a module, we need to find the .ko file and calculate a load offset +if ($module ne "") { + if ($modulefile eq "") { + $modulefile = `modinfo -F filename $module`; + chomp($modulefile); + } + $filename = $modulefile; + if ($filename eq "") { + print "Module .ko file for $module not found. Aborting\n"; + exit; + } + # ok so we found the module, now we need to calculate the vma offset + open(FILE, $cross_compile."objdump -dS $filename |") || die "Cannot start objdump"; + while () { + if ($_ =~ /^([0-9a-f]+) \<$function\>\:/) { + my $fu = $1; + $vmaoffset = Math::BigInt->from_hex("0x$target") - Math::BigInt->from_hex("0x$fu") - Math::BigInt->from_hex("0x$func_offset"); + } + } + close(FILE); +} + +my $counter = 0; +my $state = 0; +my $center = -1; +my @lines; +my @reglines; + +sub InRange { + my ($address, $target) = @_; + my $ad = "0x".$address; + my $ta = "0x".$target; + my $delta = Math::BigInt->from_hex($ad) - Math::BigInt->from_hex($ta); + + if (($delta > -4096) && ($delta < 4096)) { + return 1; + } + return 0; +} + + + +# first, parse the input into the lines array, but to keep size down, +# we only do this for 4Kb around the sweet spot + +open(FILE, $cross_compile."objdump -dS --adjust-vma=$vmaoffset --start-address=$decodestart --stop-address=$decodestop $filename |") || die "Cannot start objdump"; + +while () { + my $line = $_; + chomp($line); + if ($state == 0) { + if ($line =~ /^([a-f0-9]+)\:/) { + if (InRange($1, $target)) { + $state = 1; + } + } + } + if ($state == 1) { + if ($line =~ /^([a-f0-9][a-f0-9][a-f0-9][a-f0-9][a-f0-9][a-f0-9]+)\:/) { + my $val = $1; + if (!InRange($val, $target)) { + last; + } + if ($val eq $target) { + $center = $counter; + } + } + $lines[$counter] = $line; + + $counter = $counter + 1; + } +} + +close(FILE); + +if ($counter == 0) { + print "No matching code found \n"; + exit; +} + +if ($center == -1) { + print "No matching code found \n"; + exit; +} + +my $start; +my $finish; +my $codelines = 0; +my $binarylines = 0; +# now we go up and down in the array to find how much we want to print + +$start = $center; + +while ($start > 1) { + $start = $start - 1; + my $line = $lines[$start]; + if ($line =~ /^([a-f0-9]+)\:/) { + $binarylines = $binarylines + 1; + } else { + $codelines = $codelines + 1; + } + if ($codelines > 10) { + last; + } + if ($binarylines > 20) { + last; + } +} + + +$finish = $center; +$codelines = 0; +$binarylines = 0; +while ($finish < $counter) { + $finish = $finish + 1; + my $line = $lines[$finish]; + if ($line =~ /^([a-f0-9]+)\:/) { + $binarylines = $binarylines + 1; + } else { + $codelines = $codelines + 1; + } + if ($codelines > 10) { + last; + } + if ($binarylines > 20) { + last; + } +} + + +my $i; + + +# start annotating the registers in the asm. +# this goes from the oopsing point back, so that the annotator +# can track (opportunistically) which registers got written and +# whos value no longer is relevant. + +$i = $center; +while ($i >= $start) { + $reglines[$i] = process_x86_regs($lines[$i], $center - $i); + $i = $i - 1; +} + +$i = $start; +while ($i < $finish) { + my $line; + if ($i == $center) { + $line = "*$lines[$i] "; + } else { + $line = " $lines[$i] "; + } + print $line; + if (defined($reglines[$i]) && length($reglines[$i]) > 0) { + my $c = 60 - length($line); + while ($c > 0) { print " "; $c = $c - 1; }; + print "| $reglines[$i]"; + } + if ($i == $center) { + print "<--- faulting instruction"; + } + print "\n"; + $i = $i +1; +} + +sub usage { + print <