diff mbox series

x86_64/cet: fix tst-shstk-legacy-1{f,g} tests [BZ #31877]

Message ID 20260912113420.3723703-1-aurelien@aurel32.net (mailing list archive)
State Under Review
Delegated to: Arjun Shankar
Headers
Series x86_64/cet: fix tst-shstk-legacy-1{f,g} tests [BZ #31877] |

Checks

Context Check Description
redhat-pt-bot/TryBot-apply_patch success Patch applied to master at the time it was sent
redhat-pt-bot/TryBot-32bit success Build for i686
linaro-tcwg-bot/tcwg_glibc_build--master-aarch64 success Build passed
linaro-tcwg-bot/tcwg_glibc_build--master-arm success Build passed
linaro-tcwg-bot/tcwg_glibc_check--master-aarch64 fail Test failed
linaro-tcwg-bot/tcwg_glibc_check--master-arm success Test passed

Commit Message

Aurelien Jarno Sept. 12, 2026, 11:34 a.m. UTC 
  The tst-shstk-legacy-1f and tst-shstk-legacy-1g tests depend on the
tst-shstk-legacy-mod-1.so module to test that legacy shadow stack code
in init_array does not trigger a segfault in tst-shstk-legacy-1f, while
it does trigger a segfault in tst-shstk-legacy-1g. However, the module
is built but never loaded by the tests.

Load it using LD_PRELOAD. When glibc is configured with --enable-cet,
this allows tst-shstk-legacy-1g to pass on systems with kernel and
hardware support for shadow stacks.
---
 sysdeps/x86_64/Makefile               | 2 +-
 sysdeps/x86_64/tst-shstk-legacy-1g.sh | 1 +
 2 files changed, 2 insertions(+), 1 deletion(-)

Comments

H.J. Lu Sept. 15, 2026, 11:41 p.m. UTC | #1 
On Sat, Sep 12, 2026 at 7:34 PM Aurelien Jarno <aurelien@aurel32.net> wrote:
>
> The tst-shstk-legacy-1f and tst-shstk-legacy-1g tests depend on the
> tst-shstk-legacy-mod-1.so module to test that legacy shadow stack code
> in init_array does not trigger a segfault in tst-shstk-legacy-1f, while
> it does trigger a segfault in tst-shstk-legacy-1g. However, the module
> is built but never loaded by the tests.
>
> Load it using LD_PRELOAD. When glibc is configured with --enable-cet,
> this allows tst-shstk-legacy-1g to pass on systems with kernel and
> hardware support for shadow stacks.

With the current glibc, I got

[hjl@gnu-zen4-1 build-x86_64-linux]$
GLIBC_TUNABLES=glibc.cpu.hwcaps=SHSTK elf/tst-shstk-legacy-1g ; echo
$?
Segmentation fault         (core dumped)
GLIBC_TUNABLES=glibc.cpu.hwcaps=SHSTK elf/tst-shstk-legacy-1g
139
[hjl@gnu-zen4-1 build-x86_64-linux]$

on AMD Ryzen 7 8845HS under kernel 7.2.5.  Please show the output of

$ ldd elf/tst-shstk-legacy-1g.  I got

[hjl@gnu-zen4-1 build-x86_64-linux]$ ldd  elf/tst-shstk-legacy-1g
linux-vdso.so.1 (0x00007f6e2cd84000)
/export/build/gnu/tools-build/glibc-cet-gitlab/build-x86_64-linux/elf/tst-shstk-legacy-mod-1.so
(0x00007f6e2cd71000)
libc.so.6 => /export/build/gnu/tools-build/glibc-cet-gitlab/build-x86_64-linux/libc.so.6
(0x00007f6e2ca00000)
/export/build/gnu/tools-build/glibc-cet-gitlab/build-x86_64-linux/elf/ld.so
=> /lib64/ld-linux-x86-64.so.2 (0x00007f6e2cd86000)
[hjl@gnu-zen4-1 build-x86_64-linux]$
Yury Khrustalev Sept. 16, 2026, 3:30 p.m. UTC | #2 
On Sat, Sep 12, 2026 at 01:34:20PM +0200, Aurelien Jarno wrote:
> The tst-shstk-legacy-1f and tst-shstk-legacy-1g tests depend on the
> tst-shstk-legacy-mod-1.so module to test that legacy shadow stack code
> in init_array does not trigger a segfault in tst-shstk-legacy-1f, while
> it does trigger a segfault in tst-shstk-legacy-1g. However, the module
> is built but never loaded by the tests.
> 
> Load it using LD_PRELOAD. When glibc is configured with --enable-cet,
> this allows tst-shstk-legacy-1g to pass on systems with kernel and
> hardware support for shadow stacks.

Same is with the adjacent patch, there is a failure report in Patchwork
for aarch64 but I cannot reproduce it, must be false positive.

Thanks,
Yury
Aurelien Jarno Sept. 16, 2026, 5:23 p.m. UTC | #3 
On 2026-09-16 16:30, Yury Khrustalev wrote:
> On Sat, Sep 12, 2026 at 01:34:20PM +0200, Aurelien Jarno wrote:
> > The tst-shstk-legacy-1f and tst-shstk-legacy-1g tests depend on the
> > tst-shstk-legacy-mod-1.so module to test that legacy shadow stack code
> > in init_array does not trigger a segfault in tst-shstk-legacy-1f, while
> > it does trigger a segfault in tst-shstk-legacy-1g. However, the module
> > is built but never loaded by the tests.
> > 
> > Load it using LD_PRELOAD. When glibc is configured with --enable-cet,
> > this allows tst-shstk-legacy-1g to pass on systems with kernel and
> > hardware support for shadow stacks.
> 
> Same is with the adjacent patch, there is a failure report in Patchwork
> for aarch64 but I cannot reproduce it, must be false positive.

Yes, I already mentioned that to linaro-toolchain@lists.linaro.org. This 
patch only changes sysdeps/x86_64, so I really think it is a fault 
positive.

Anyway it is superseded by:
https://inbox.sourceware.org/libc-alpha/CAMe9rOqpDK6S+73_rUSa3NoT=eqrXBckpHfn-GQ4RG0DJSMb9Q@mail.gmail.com

Regards
Aurelien
diff mbox series

Patch

diff --git a/sysdeps/x86_64/Makefile b/sysdeps/x86_64/Makefile
index ff0463d7ca..688b537a48 100644
--- a/sysdeps/x86_64/Makefile
+++ b/sysdeps/x86_64/Makefile
@@ -466,7 +466,7 @@  $(objpfx)tst-shstk-legacy-1e-static.out: \
   $(objpfx)tst-shstk-legacy-1e-static
 	$(SHELL) $< $(common-objpfx) 2> $@; \
 	$(evaluate-test)
-tst-shstk-legacy-1f-ENV = GLIBC_TUNABLES=glibc.cpu.hwcaps=SHSTK
+tst-shstk-legacy-1f-ENV = GLIBC_TUNABLES=glibc.cpu.hwcaps=SHSTK LD_PRELOAD=$(objpfx)tst-shstk-legacy-mod-1.so
 $(objpfx)tst-shstk-legacy-1f: $(objpfx)tst-shstk-legacy-mod-1.so
 $(objpfx)tst-shstk-legacy-mod-1.so: \
   $(objpfx)tst-shstk-legacy-mod-1.os \
diff --git a/sysdeps/x86_64/tst-shstk-legacy-1g.sh b/sysdeps/x86_64/tst-shstk-legacy-1g.sh
index 449bf7493b..0453e3d3af 100755
--- a/sysdeps/x86_64/tst-shstk-legacy-1g.sh
+++ b/sysdeps/x86_64/tst-shstk-legacy-1g.sh
@@ -22,6 +22,7 @@  common_objpfx=$1; shift
 test_program_prefix=$1; shift
 
 GLIBC_TUNABLES=glibc.cpu.hwcaps=SHSTK \
+LD_PRELOAD=${common_objpfx}elf/tst-shstk-legacy-mod-1.so \
 ${test_program_prefix} \
   ${common_objpfx}elf/tst-shstk-legacy-1g
 # The exit status should only be unsupported (77) or segfault (139).