Re: [PATCH 4/5] gdb: add program_space parameter to clear_solib

[Simon Marchi <simon.marchi@efficios.com>]

On 2/6/24 12:14, Simon Marchi wrote:
> Make the current_program_space reference bubble up one level.
> 
> Remove one unnecessary declaration of clear_solib.
> 
> Change-Id: I234e2c8c0b71713364fc7b76cee2bee2b026bd6d
> ---
>  gdb/corelow.c |  2 +-
>  gdb/solib.c   | 21 ++++++++++++---------
>  gdb/solib.h   |  6 +++---
>  gdb/symtab.h  |  4 ----
>  4 files changed, 16 insertions(+), 17 deletions(-)
> 
> diff --git a/gdb/corelow.c b/gdb/corelow.c
> index 43c1f69b1317..f291b2aba191 100644
> --- a/gdb/corelow.c
> +++ b/gdb/corelow.c
> @@ -335,7 +335,7 @@ core_target::clear_core ()
>  
>        /* Clear out solib state while the bfd is still open.  See
>  	 comments in clear_solib in solib.c.  */
> -      clear_solib ();
> +      clear_solib (current_program_space);
>  
>        current_program_space->cbfd.reset (nullptr);
>      }
> diff --git a/gdb/solib.c b/gdb/solib.c
> index 98cda039a833..bd69c549b8e8 100644
> --- a/gdb/solib.c
> +++ b/gdb/solib.c
> @@ -1180,23 +1180,26 @@ solib_keep_data_in_core (CORE_ADDR vaddr, unsigned long size)
>      return false;
>  }
>  
> -/* Called by free_all_symtabs */
> +/* See solib.h.  */
>  
>  void
> -clear_solib (void)
> +clear_solib (program_space *pspace)
>  {
> -  const solib_ops *ops = gdbarch_so_ops (current_inferior ()->arch ());
> +  inferior *inf = find_inferior_for_program_space (pspace);
> +  gdb_assert (inf != nullptr);

Thanks to the Linaro CI for pointing it out, this particular change
causes some regressions, for instance in gdb.base/catch-fork-kill.exp.
The assertion fails because inf is nullptr.

We're here:

    #0  internal_error_loc (file=0x55adc73f51c0 "/home/smarchi/src/binutils-gdb/gdb/solib.c", line=1189,
        fmt=0x55adc73f4b00 "%s: Assertion `%s' failed.") at /home/smarchi/src/binutils-gdb/gdbsupport/errors.cc:57
    #1  0x000055adcb297e6d in clear_solib (pspace=0x613000011d80) at /home/smarchi/src/binutils-gdb/gdb/solib.c:1189
    #2  0x000055adcb29822e in no_shared_libraries (ignored=0x0, from_tty=0) at /home/smarchi/src/binutils-gdb/gdb/solib.c:1250
    #3  0x000055adca9f26d4 in program_space::~program_space (this=0x613000011d80, __in_chrg=<optimized out>)
        at /home/smarchi/src/binutils-gdb/gdb/progspace.c:120
    #4  0x000055adca0c570c in delete_inferior (inf=0x618000225480) at /home/smarchi/src/binutils-gdb/gdb/inferior.c:302
    #5  0x000055adca0c801d in prune_inferiors () at /home/smarchi/src/binutils-gdb/gdb/inferior.c:492

What I observe is that the current inferior is #1, bound to pspace #1,
and we're deleting inferior #2, bound to pspace #2.  My change doesn't
work, because at this point inferior #2 has been removed from the
inferior list, so find_inferior_for_program_space returns nullptr.

However, the current code seems wrong too, because we're using inferior
#1's arch to get hold of an solib_ops to call ops->clear_solib on,
despite deleting program space #2, which was bound to inferior #2.  It
works here, because both inferiors have the same gdbarch with the same
solib_ops.  But it would be possible for the two inferiors to have
different gdbarches and solib_ops.

My upcoming solib series should improve the situation, by having a
backlink from program_space to solib_ops.  So we won't need an inferior
here in order to get the solib_ops.  In the mean time, I'll try to
revert using current_inferior to get the arch and solib_ops, even though
I know it's wrong.

Simon