All of lore.kernel.org
 help / color / mirror / Atom feed
From: Wilson Felipe Pereira <wfelipe@google.com>
To: Josh Poimboeuf <jpoimboe@kernel.org>,
	Peter Zijlstra <peterz@infradead.org>,
	 Miroslav Benes <mbenes@suse.cz>, Petr Mladek <pmladek@suse.com>,
	 Joe Lawrence <joe.lawrence@redhat.com>,
	Song Liu <song@kernel.org>
Cc: live-patching@vger.kernel.org, linux-kernel@vger.kernel.org,
	 Dylan Hatch <dylanbhatch@google.com>, Yi Zhu <chuyee@google.com>,
	 Wilson Felipe Pereira <wfelipe@google.com>
Subject: [PATCH 2/2] objtool/klp: Add test for global/local symbol name collisions
Date: Tue, 17 Nov 2026 00:08:32 +0000	[thread overview]
Message-ID: <20261117000835.1992893-3-wfelipe@google.com> (raw)
In-Reply-To: <20261117000835.1992893-1-wfelipe@google.com>

Add a test covering sympos disambiguation when a defined STB_GLOBAL symbol
shares its name with a static STB_LOCAL symbol in another translation
unit:
- A unique non-exported global continues to emit sympos 0.
- On the module path, a static in the first TU and a non-exported global
  in the second TU emit sympos 1 and 2 in .symtab order.
- On the vmlinux (.klp.symid) path, placing the global at a lower address
  than the static emits sympos 1 in linked address order, both for an
  unexported global and for an EXPORT_SYMBOL_FOR_MODULES() export.

Verified by reverting each hunk in klp-sympos.c, klp-symid.c, and
klp-diff.c individually and confirming the corresponding subtest fails.

Signed-off-by: Wilson Felipe Pereira <wfelipe@google.com>
---
 .../tests/generic/fixtures/sympos_vmlinux.c   |  6 ++
 .../tests/generic/test-sympos-global-dup.sh   | 99 +++++++++++++++++++
 2 files changed, 105 insertions(+)
 create mode 100755 tools/objtool/tests/generic/test-sympos-global-dup.sh

diff --git a/tools/objtool/tests/generic/fixtures/sympos_vmlinux.c b/tools/objtool/tests/generic/fixtures/sympos_vmlinux.c
index d5e70994c582..68d6fca414e4 100644
--- a/tools/objtool/tests/generic/fixtures/sympos_vmlinux.c
+++ b/tools/objtool/tests/generic/fixtures/sympos_vmlinux.c
@@ -26,8 +26,14 @@ static const char __modinfo[]
 	__attribute__((section(".modinfo"), used, aligned(1))) = "\0name=vmlinux";
 #endif
 
+#ifdef GLOBAL_DUP
+int dup_counter = 2;
+#elif defined(EXTERN_DUP)
+extern int dup_counter;
+#else
 /* volatile so it survives as an STT_OBJECT rather than being folded away */
 static volatile int dup_counter __attribute__((section(VARSEC))) = 1;
+#endif
 
 int FUNC_NAME(int x)
 {
diff --git a/tools/objtool/tests/generic/test-sympos-global-dup.sh b/tools/objtool/tests/generic/test-sympos-global-dup.sh
new file mode 100755
index 000000000000..b309b1f3f7f8
--- /dev/null
+++ b/tools/objtool/tests/generic/test-sympos-global-dup.sh
@@ -0,0 +1,99 @@
+#!/bin/bash
+# SPDX-License-Identifier: GPL-2.0
+#
+# A defined STB_GLOBAL symbol in one translation unit can share its name with a
+# static (STB_LOCAL) symbol in another translation unit (for example,
+# cgroup_storage_lookup in kernel/bpf/local_storage.c vs
+# kernel/bpf/bpf_cgrp_storage.c).
+#
+# At runtime, kallsyms_on_each_match_symbol() and
+# module_kallsyms_on_each_symbol() count both local and global symbols of that
+# name, so emitting sympos 0 for the global reference fails module load with
+# "unresolvable ambiguity for symbol". Both the module path (symtab order) and
+# the vmlinux path (.klp.symid address order) must disambiguate the global
+# symbol with its 1-based position while still emitting 0 when the global name
+# is unique.
+
+. "$(dirname "$0")/../lib.sh"
+
+# 1. Unique non-exported global: sympos must remain 0.
+setup
+build_one sympos_vmlinux.c def_g.o     -DFUNC_NAME=def_g -DGLOBAL_DUP
+build_one sympos_vmlinux.c use_orig.o  -DFUNC_NAME=caller -DEXTERN_DUP -DNO_MODINFO
+build_one sympos_vmlinux.c use_patch.o -DFUNC_NAME=caller -DEXTERN_DUP -DNO_MODINFO \
+	-DPATCHED
+
+partial_link "$workdir/orig.o"    "$workdir/def_g.o" "$workdir/use_orig.o" ||
+	probe_skip "partial link unavailable"
+partial_link "$workdir/patched.o" "$workdir/def_g.o" "$workdir/use_patch.o" ||
+	probe_skip "partial link unavailable"
+
+run_diff
+assert_klp_sympos dup_counter 0
+cleanup
+
+# 2. Module path: STB_LOCAL in first TU + non-exported STB_GLOBAL in second TU.
+#    In ELF .symtab, STB_LOCAL symbols precede STB_GLOBAL symbols, so the local
+#    is position 1 and the global is position 2.
+setup
+for p in "" "-DPATCHED"; do
+	# shellcheck disable=SC2086
+	build_one sympos_vmlinux.c "loc$p.o" -DFUNC_NAME=use_local $p
+	# shellcheck disable=SC2086
+	build_one sympos_vmlinux.c "glob$p.o" -DFUNC_NAME=use_global \
+		-DGLOBAL_DUP -DNO_MODINFO $p
+done
+
+partial_link "$workdir/orig.o"    "$workdir/loc.o"         "$workdir/glob.o" ||
+	probe_skip "partial link unavailable"
+partial_link "$workdir/patched.o" "$workdir/loc-DPATCHED.o" "$workdir/glob-DPATCHED.o" ||
+	probe_skip "partial link unavailable"
+
+[ "$(count_input_symbols orig.o dup_counter)" = 2 ] ||
+	fail "fixture did not produce two dup_counter symbols"
+
+run_diff
+assert_klp_sympos dup_counter 1
+assert_klp_sympos dup_counter 2
+out_symbols | grep -qE '\.klp\.sym\.[^.]+\.dup_counter,0([[:space:]]|$)' &&
+	fail "sympos 0 emitted for global symbol colliding with local symbol"
+cleanup
+
+# 3. vmlinux path (.klp.symid): with -fdata-sections the STB_GLOBAL lands in
+#    .data.dup_counter (lower address) while the STB_LOCAL lands in .data.zzz
+#    (higher address), opposite to ELF .symtab order where locals come first.
+#    Patching only the global caller must emit sympos 1 (address order).
+setup
+build_one sympos_vmlinux.c loc.o -DFUNC_NAME=use_local -DVARSEC='".data.zzz"'
+build_one sympos_vmlinux.c glob_orig.o -DFUNC_NAME=use_global -DGLOBAL_DUP \
+	-DNO_MODINFO
+build_one sympos_vmlinux.c glob_patch.o -DFUNC_NAME=use_global -DGLOBAL_DUP \
+	-DNO_MODINFO -DPATCHED
+
+make_vmlinux_pair "$workdir/loc.o" "$workdir/glob_orig.o" \
+	       -- "$workdir/loc.o" "$workdir/glob_patch.o"
+
+[ "$(count_input_symbols vmlinux.o dup_counter)" = 2 ] ||
+	fail "vmlinux fixture did not produce two dup_counter symbols"
+
+# Premise: in ELF .symtab the local comes first (higher address .data.zzz),
+# while in address order the global comes first (.data.aaa).
+first_addr="$(in_symbols vmlinux | awk '$8 == "dup_counter" { print $2; exit }')"
+low_addr="$(in_symbols vmlinux | awk '$8 == "dup_counter" { print $2 }' | sort | head -1)"
+[ "$first_addr" != "$low_addr" ] ||
+	probe_skip "linker did not place global before local in address order"
+
+run_diff
+assert_klp_sympos dup_counter 1
+out_symbols | grep -qE '\.klp\.sym\.[^.]+\.dup_counter,(0|2)([[:space:]]|$)' &&
+	fail "wrong sympos emitted for vmlinux global colliding with higher-addr local"
+
+# 4. Same vmlinux collision when the global uses EXPORT_SYMBOL_FOR_MODULES(),
+#    which takes the export branch in clone_reloc_klp().
+add_exports_ns vmlinux "module:other" dup_counter
+run_diff
+assert_klp_sympos dup_counter 1
+out_symbols | grep -qE '\.klp\.sym\.[^.]+\.dup_counter,(0|2)([[:space:]]|$)' &&
+	fail "wrong sympos emitted for module-restricted export colliding with local"
+
+pass "global symbol colliding with static local gets disambiguated sympos"
-- 
2.56.1.703.gf848ed807c-goog


      parent reply	other threads:[~2026-11-17  0:08 UTC|newest]

Thread overview: 4+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-17  0:08 [PATCH 0/2] objtool/klp: Fix sympos for global/local symbol collisions Wilson Felipe Pereira
2026-11-17  0:08 ` [PATCH 1/2] objtool/klp: Disambiguate global symbols colliding with local symbols Wilson Felipe Pereira
2026-11-17  0:25   ` sashiko-bot
2026-11-17  0:08 ` Wilson Felipe Pereira [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20261117000835.1992893-3-wfelipe@google.com \
    --to=wfelipe@google.com \
    --cc=chuyee@google.com \
    --cc=dylanbhatch@google.com \
    --cc=joe.lawrence@redhat.com \
    --cc=jpoimboe@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=live-patching@vger.kernel.org \
    --cc=mbenes@suse.cz \
    --cc=peterz@infradead.org \
    --cc=pmladek@suse.com \
    --cc=song@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.