Re: glibc build failure due to Systemtap probe change

[Stefan Liebler <stli@linux.ibm.com>]

On 16/11/2021 10:29, Florian Weimer wrote:
> On Fedora rawhide, glibc no longer builds:
> 
> In file included from ../include/stap-probe.h:25,
>                  from ../sysdeps/s390/s390-64/__longjmp.c:24:
> ../sysdeps/s390/s390-64/__longjmp.c: In function '__longjmp':
> ../sysdeps/s390/s390-64/__longjmp.c:52:50: error: expected ':' or ')' before numeric constant
>    52 |                         LIBC_PROBE_ASM (longjmp, 8@%1 -4@%0 8@%%r4)
>       |                                                  ^
> ../sysdeps/s390/s390-64/__longjmp.c:52:25: note: in expansion of macro 'LIBC_PROBE_ASM'
>    52 |                         LIBC_PROBE_ASM (longjmp, 8@%1 -4@%0 8@%%r4)
>       |                         ^~~~~~~~~~~~~~
> 
> This is because of a Systemtap probe:
> 
> | /* Jump to the position specified by ENV, causing the
> |    setjmp call there to return VAL, or 1 if VAL is 0.  */
> | void
> | __longjmp (__jmp_buf env, int val)
> | {
> | #ifdef PTR_DEMANGLE
> |   uintptr_t guard = THREAD_GET_POINTER_GUARD ();
> | # ifdef CHECK_SP
> |   CHECK_SP (env, guard);
> | # endif
> | #elif defined CHECK_SP
> |   CHECK_SP (env, 0);
> | #endif
> |   register long int r2 __asm__ ("%r2") = val == 0 ? 1 : val;
> | #ifdef PTR_DEMANGLE
> |   register uintptr_t r3 __asm__ ("%r3") = guard;
> |   register void *r1 __asm__ ("%r1") = (void *) env;
> | #endif
> |   /* Restore registers and jump back.  */
> |   __asm__ __volatile__ (
> | 			/* longjmp probe expects longjmp first argument, second
> | 			   argument and target address.  */
> | #ifdef PTR_DEMANGLE
> | 			"lmg  %%r4,%%r5,64(%1)\n\t"
> | 			"xgr  %%r4,%2\n\t"
> | 			"xgr  %%r5,%2\n\t"
> | 			LIBC_PROBE_ASM (longjmp, 8@%1 -4@%0 8@%%r4)
> | #else
> | 			LIBC_PROBE_ASM (longjmp, 8@%1 -4@%0 8@%%r14)
> | #endif
> 
> My guess this is related to the macro changes to support floating point
> values in probes added in this commit:
> 
> commit eaa15b047688175a94e3ae796529785a3a0af208
> Author: Stan Cox <scox@redhat.com>
> Date:   Thu Sep 30 16:11:29 2021 -0400
> 
>     PR27829: Support floating point values passed through sdt.h markers
>     
>     Add the type to the individual arg entries in the .notes.stapsdt section;
>     currently SP@A, where S is optional '-' sign, P is precision of type and A is
>     address.  Revised format is SPT@A where T is optional 'f' for float variables.
>     Add x8664 float registers xmm8 - xmm15 and aarch64 float registers v8 - v31.
>     Parse the type field; result is currently ignored.  asm statements are
>     restricted to 30 arguments; sdt probes can have up to 12 arguments.  To fit
>     this into a single asm statement, precision and type are encoded into a single
>     field: 0xSSTT where SS is the precision and TT is the type as encoded by
>     __builtin_classify_type.  The sign S, precision P, and type T are decoded by
>     _SDT_SIGN, _SDT_SIZE, and _SDT_TYPE.  Test that the revised
>     .notes.stapsdt section interacts correctly with eu-elfutils and gdb.
> 
> The macro changes in this commit seem to trigger additional macro
> expansion in the preprocessor, and therefore require syntax closer to C.
> 
> LIBC_PROBE_ASM was added to longjmp to support GDB, but if GDB really
> needs this, I'm not sure if Systemtap probes (via <sys/sdt.h>) are the
> right mechanism because building with Systemtap support is optional.
> Maybe we should copy <sys/sdt.h> into the glibc sources, and then we
> could drop the floating point support again. 8-/
> 
> Or we could rewrite s390/s390x __longjmp in assembler.  It's mostly
> assembler anyway.  Stefan, what do you think?
I would prefer to keep the c-code with inline-assembly as is. Then we
don't need to reimplement the macros in assembler.
But let's wait for Stan's experiments with the preprocessed files.

> 
> Thanks,
> Florian
>