s390: Fix ABI issue in libstdc++.so.6
Checks
Commit Message
On Sat, Jan 10, 2026 at 05:24:15PM +0100, Stefan Schulze Frielinghaus wrote:
> libstdc++-v3/ChangeLog:
>
> * config/abi/post/s390x-linux-gnu/baseline_symbols.txt: Add
> names {,P,K}DF16.
This is wrong - an ABI issue.
You can't export new symbols in CXXABI_1.3.14 symbol version when they
weren't exported there in GCC 13.1 already.
Symbols new in GCC 16 like these should be exported in CXXABI_1.3.17.
So, I think you want (completely untested):
2026-01-10 Jakub Jelinek <jakub@redhat.com>
* config/abi/pre/gnu.ver (CXXABI_1.3.14): Don't export _ZTI*DF16_ on
s390x.
(CXXABI_1.3.17): Export _ZTI*DF16_ on s390x.
* config/abi/post/s390x-linux-gnu/baseline_symbols.txt: Remove
_ZTI{,P,K}DF16_.
Jakub
Comments
On Sat, Jan 10, 2026 at 07:04:53PM +0100, Jakub Jelinek wrote:
> On Sat, Jan 10, 2026 at 05:24:15PM +0100, Stefan Schulze Frielinghaus wrote:
> > libstdc++-v3/ChangeLog:
> >
> > * config/abi/post/s390x-linux-gnu/baseline_symbols.txt: Add
> > names {,P,K}DF16.
>
> This is wrong - an ABI issue.
>
> You can't export new symbols in CXXABI_1.3.14 symbol version when they
> weren't exported there in GCC 13.1 already.
> Symbols new in GCC 16 like these should be exported in CXXABI_1.3.17.
Yikes, you are absolutely right. It was also wrong in the first draft I
posted back in Nov. Thank you very much for double checking and
catching this! I will trigger a full test run of your patch and come
back to you.
Cheers,
Stefan
> So, I think you want (completely untested):
>
> 2026-01-10 Jakub Jelinek <jakub@redhat.com>
>
> * config/abi/pre/gnu.ver (CXXABI_1.3.14): Don't export _ZTI*DF16_ on
> s390x.
> (CXXABI_1.3.17): Export _ZTI*DF16_ on s390x.
> * config/abi/post/s390x-linux-gnu/baseline_symbols.txt: Remove
> _ZTI{,P,K}DF16_.
>
> --- libstdc++-v3/config/abi/pre/gnu.ver.jj 2026-01-02 09:56:10.583329488 +0100
> +++ libstdc++-v3/config/abi/pre/gnu.ver 2026-01-10 18:58:37.133669626 +0100
> @@ -2899,9 +2899,16 @@ CXXABI_1.3.14 {
>
> # typeinfo for _Float{16,32,64,128,32x,64x,128x} and
> # __bf16/bfloat16_t
> +#ifndef __s390x__
> _ZTIDF[0-9]*[_x];
> _ZTIPDF[0-9]*[_x];
> _ZTIPKDF[0-9]*[_x];
> +#else
> + # No _Float16 on s390x in GCC 13.
> + _ZTIDF[0-9]*[0-57-9][_x];
> + _ZTIPDF[0-9]*[0-57-9][_x];
> + _ZTIPKDF[0-9]*[0-57-9][_x];
> +#endif
> #ifndef __riscv
> _ZTIDF16b;
> _ZTIPDF16b;
> @@ -2933,6 +2940,13 @@ CXXABI_1.3.16 {
> CXXABI_1.3.17 {
> # std::exception_ptr::_M_exception_ptr_cast
> _ZNKSt15__exception_ptr13exception_ptr21_M_exception_ptr_castERKSt9type_info;
> +
> +#ifdef __s390x__
> + _ZTIDF16_;
> + _ZTIPDF16_;
> + _ZTIPKDF16_;
> +#endif
> +
> #ifdef __riscv
> } CXXABI_1.3.16;
> #else
> --- libstdc++-v3/config/abi/post/s390x-linux-gnu/baseline_symbols.txt.jj 2026-01-10 18:52:48.533640534 +0100
> +++ libstdc++-v3/config/abi/post/s390x-linux-gnu/baseline_symbols.txt 2026-01-10 18:59:02.224239870 +0100
> @@ -4951,7 +4951,6 @@ OBJECT:15:_ZTSSt8messagesIwE@@GLIBCXX_3.
> OBJECT:15:_ZTSSt8numpunctIcE@@GLIBCXX_3.4
> OBJECT:15:_ZTSSt8numpunctIwE@@GLIBCXX_3.4
> OBJECT:16:_ZTIDF128_@@CXXABI_1.3.14
> -OBJECT:16:_ZTIDF16_@@CXXABI_1.3.14
> OBJECT:16:_ZTIDF32_@@CXXABI_1.3.14
> OBJECT:16:_ZTIDF32x@@CXXABI_1.3.14
> OBJECT:16:_ZTIDF64_@@CXXABI_1.3.14
> @@ -5616,7 +5615,6 @@ OBJECT:30:_ZTSSt7codecvtIDsDu11__mbstate
> OBJECT:32:_ZNSbIwSt11char_traitsIwESaIwEE4_Rep20_S_empty_rep_storageE@@GLIBCXX_3.4
> OBJECT:32:_ZNSs4_Rep20_S_empty_rep_storageE@@GLIBCXX_3.4
> OBJECT:32:_ZTIPDF128_@@CXXABI_1.3.14
> -OBJECT:32:_ZTIPDF16_@@CXXABI_1.3.14
> OBJECT:32:_ZTIPDF32_@@CXXABI_1.3.14
> OBJECT:32:_ZTIPDF32x@@CXXABI_1.3.14
> OBJECT:32:_ZTIPDF64_@@CXXABI_1.3.14
> @@ -5629,7 +5627,6 @@ OBJECT:32:_ZTIPDn@@CXXABI_1.3.5
> OBJECT:32:_ZTIPDs@@CXXABI_1.3.3
> OBJECT:32:_ZTIPDu@@CXXABI_1.3.12
> OBJECT:32:_ZTIPKDF128_@@CXXABI_1.3.14
> -OBJECT:32:_ZTIPKDF16_@@CXXABI_1.3.14
> OBJECT:32:_ZTIPKDF32_@@CXXABI_1.3.14
> OBJECT:32:_ZTIPKDF32x@@CXXABI_1.3.14
> OBJECT:32:_ZTIPKDF64_@@CXXABI_1.3.14
>
>
> Jakub
>
On Sat, Jan 10, 2026 at 07:40:37PM +0100, Stefan Schulze Frielinghaus wrote:
> On Sat, Jan 10, 2026 at 07:04:53PM +0100, Jakub Jelinek wrote:
> > On Sat, Jan 10, 2026 at 05:24:15PM +0100, Stefan Schulze Frielinghaus wrote:
> > > libstdc++-v3/ChangeLog:
> > >
> > > * config/abi/post/s390x-linux-gnu/baseline_symbols.txt: Add
> > > names {,P,K}DF16.
> >
> > This is wrong - an ABI issue.
> >
> > You can't export new symbols in CXXABI_1.3.14 symbol version when they
> > weren't exported there in GCC 13.1 already.
> > Symbols new in GCC 16 like these should be exported in CXXABI_1.3.17.
>
> Yikes, you are absolutely right. It was also wrong in the first draft I
> posted back in Nov. Thank you very much for double checking and
> catching this! I will trigger a full test run of your patch and come
> back to you.
Bootstrap and regtest are successful for targets z16, z14, z13, z10,
z900.
Thanks,
Stefan
On Sun, Jan 11, 2026 at 07:06:28PM +0100, Stefan Schulze Frielinghaus wrote:
> On Sat, Jan 10, 2026 at 07:40:37PM +0100, Stefan Schulze Frielinghaus wrote:
> > On Sat, Jan 10, 2026 at 07:04:53PM +0100, Jakub Jelinek wrote:
> > > On Sat, Jan 10, 2026 at 05:24:15PM +0100, Stefan Schulze Frielinghaus wrote:
> > > > libstdc++-v3/ChangeLog:
> > > >
> > > > * config/abi/post/s390x-linux-gnu/baseline_symbols.txt: Add
> > > > names {,P,K}DF16.
> > >
> > > This is wrong - an ABI issue.
BTW, if you didn't adjust the baseline_symbols.txt, the testsuite error
would tell you something wrong happened.
Guess I should verify the _Float*/__bf16 rtti symbols in all
baseline_symbols.txt from 13 to latest trunk to make sure other targets
haven't done the same thing.
> > > You can't export new symbols in CXXABI_1.3.14 symbol version when they
> > > weren't exported there in GCC 13.1 already.
> > > Symbols new in GCC 16 like these should be exported in CXXABI_1.3.17.
> >
> > Yikes, you are absolutely right. It was also wrong in the first draft I
> > posted back in Nov. Thank you very much for double checking and
> > catching this! I will trigger a full test run of your patch and come
> > back to you.
>
> Bootstrap and regtest are successful for targets z16, z14, z13, z10,
> z900.
I've also bootstrapped/regtested it successfully on x86_64-linux and
i686-linux, plus verified the abilist before/after.
Though, I'm not sure if the patch falls under target maintainership, while
the code actually doesn't affect other targets, if it was buggy it would.
So I think we need Jonathan's ack.
Jakub
On Sun, 11 Jan 2026 at 18:08, Stefan Schulze Frielinghaus
<stefansf@linux.ibm.com> wrote:
>
> On Sat, Jan 10, 2026 at 07:40:37PM +0100, Stefan Schulze Frielinghaus wrote:
> > On Sat, Jan 10, 2026 at 07:04:53PM +0100, Jakub Jelinek wrote:
> > > On Sat, Jan 10, 2026 at 05:24:15PM +0100, Stefan Schulze Frielinghaus wrote:
> > > > libstdc++-v3/ChangeLog:
> > > >
> > > > * config/abi/post/s390x-linux-gnu/baseline_symbols.txt: Add
> > > > names {,P,K}DF16.
> > >
> > > This is wrong - an ABI issue.
> > >
> > > You can't export new symbols in CXXABI_1.3.14 symbol version when they
> > > weren't exported there in GCC 13.1 already.
> > > Symbols new in GCC 16 like these should be exported in CXXABI_1.3.17.
> >
> > Yikes, you are absolutely right. It was also wrong in the first draft I
> > posted back in Nov. Thank you very much for double checking and
> > catching this! I will trigger a full test run of your patch and come
> > back to you.
>
> Bootstrap and regtest are successful for targets z16, z14, z13, z10,
> z900.
The patch is OK then, thanks
@@ -2899,9 +2899,16 @@ CXXABI_1.3.14 {
# typeinfo for _Float{16,32,64,128,32x,64x,128x} and
# __bf16/bfloat16_t
+#ifndef __s390x__
_ZTIDF[0-9]*[_x];
_ZTIPDF[0-9]*[_x];
_ZTIPKDF[0-9]*[_x];
+#else
+ # No _Float16 on s390x in GCC 13.
+ _ZTIDF[0-9]*[0-57-9][_x];
+ _ZTIPDF[0-9]*[0-57-9][_x];
+ _ZTIPKDF[0-9]*[0-57-9][_x];
+#endif
#ifndef __riscv
_ZTIDF16b;
_ZTIPDF16b;
@@ -2933,6 +2940,13 @@ CXXABI_1.3.16 {
CXXABI_1.3.17 {
# std::exception_ptr::_M_exception_ptr_cast
_ZNKSt15__exception_ptr13exception_ptr21_M_exception_ptr_castERKSt9type_info;
+
+#ifdef __s390x__
+ _ZTIDF16_;
+ _ZTIPDF16_;
+ _ZTIPKDF16_;
+#endif
+
#ifdef __riscv
} CXXABI_1.3.16;
#else
@@ -4951,7 +4951,6 @@ OBJECT:15:_ZTSSt8messagesIwE@@GLIBCXX_3.
OBJECT:15:_ZTSSt8numpunctIcE@@GLIBCXX_3.4
OBJECT:15:_ZTSSt8numpunctIwE@@GLIBCXX_3.4
OBJECT:16:_ZTIDF128_@@CXXABI_1.3.14
-OBJECT:16:_ZTIDF16_@@CXXABI_1.3.14
OBJECT:16:_ZTIDF32_@@CXXABI_1.3.14
OBJECT:16:_ZTIDF32x@@CXXABI_1.3.14
OBJECT:16:_ZTIDF64_@@CXXABI_1.3.14
@@ -5616,7 +5615,6 @@ OBJECT:30:_ZTSSt7codecvtIDsDu11__mbstate
OBJECT:32:_ZNSbIwSt11char_traitsIwESaIwEE4_Rep20_S_empty_rep_storageE@@GLIBCXX_3.4
OBJECT:32:_ZNSs4_Rep20_S_empty_rep_storageE@@GLIBCXX_3.4
OBJECT:32:_ZTIPDF128_@@CXXABI_1.3.14
-OBJECT:32:_ZTIPDF16_@@CXXABI_1.3.14
OBJECT:32:_ZTIPDF32_@@CXXABI_1.3.14
OBJECT:32:_ZTIPDF32x@@CXXABI_1.3.14
OBJECT:32:_ZTIPDF64_@@CXXABI_1.3.14
@@ -5629,7 +5627,6 @@ OBJECT:32:_ZTIPDn@@CXXABI_1.3.5
OBJECT:32:_ZTIPDs@@CXXABI_1.3.3
OBJECT:32:_ZTIPDu@@CXXABI_1.3.12
OBJECT:32:_ZTIPKDF128_@@CXXABI_1.3.14
-OBJECT:32:_ZTIPKDF16_@@CXXABI_1.3.14
OBJECT:32:_ZTIPKDF32_@@CXXABI_1.3.14
OBJECT:32:_ZTIPKDF32x@@CXXABI_1.3.14
OBJECT:32:_ZTIPKDF64_@@CXXABI_1.3.14