Re: [PATCH v5 03/15] gdb, gdbserver: support dlmopen()

[Kevin Buettner <kevinb@redhat.com>]

Hi Markus,

This patch looks pretty good.  Just a couple of comments/questions...

On Thu,  2 Jun 2022 15:25:02 +0200
Markus Metzger via Gdb-patches <gdb-patches@sourceware.org> wrote:

> diff --git a/gdb/solib-svr4.c b/gdb/solib-svr4.c
> index 6222f79a7a0..1f2abcf16ef 100644
[...]
> @@ -1258,49 +1338,110 @@ svr4_current_sos_direct (struct svr4_info *info)
>        if (library_list.main_lm)
>  	info->main_lm_addr = library_list.main_lm;
>  
> -      return library_list.head ? library_list.head : svr4_default_sos (info);
> +      /* Remove an empty special zero namespace so we know that when there
> +	 is one, it is actually used, and we have a flat list without
> +	 namespace information.  */
> +      if ((library_list.solib_lists.find (0)
> +	   != library_list.solib_lists.end ())
> +	  && (library_list.solib_lists[0] == nullptr))
> +	library_list.solib_lists.erase (0);
> +
> +      std::swap (info->solib_lists, library_list.solib_lists);
> +      return;
>      }
[...]
> @@ -1697,17 +1872,34 @@ solist_update_incremental (struct svr4_info *info, CORE_ADDR lm)
>        if (!svr4_current_sos_via_xfer_libraries (&library_list, annex))
>  	return 0;
>  
> -      tail->next = library_list.head;
> +      /* We expect gdbserver to provide updates for the namespace that
> +	 contains LM, which whould be this namespace...  */
> +      so_list *sos = nullptr;
> +      if (library_list.solib_lists.find (debug_base)
> +	  != library_list.solib_lists.end ())
> +	std::swap (sos, library_list.solib_lists[debug_base]);
> +      if (sos == nullptr)
> +	{
> +	  /* ...or for the special zero namespace for earlier versions...  */
> +	  if (library_list.solib_lists.find (0)
> +	      != library_list.solib_lists.end ())
> +	    std::swap (sos, library_list.solib_lists[0]);
[...]

I was puzzled for a while by the uses of std:swap() in the above
snippets of code.  I checked the rest of the GDB sources and found
that (at least in GDB) this is not a common idiom.  I'd appreciate
it if you could add a few lines of explanation for at least one
of them.

> diff --git a/gdb/testsuite/lib/gdb.exp b/gdb/testsuite/lib/gdb.exp
> index 6aef39e5e02..98715fdc87e 100644
> --- a/gdb/testsuite/lib/gdb.exp
> +++ b/gdb/testsuite/lib/gdb.exp
> @@ -2451,6 +2451,102 @@ proc skip_shlib_tests {} {
>      return 1
>  }
>  
> +# Return 1 if we should skip dlmopen tests, 0 if we should not.
> +
> +proc skip_dlmopen_tests {} {
> +    global srcdir subdir gdb_prompt inferior_exited_re
> +
> +    # We need shared library support.
> +    if { [skip_shlib_tests] } {
> +	return 1
> +    }
> +
> +    set me "skip_dlmopen_tests"
> +    set lib {
> +	int foo (void) {
> +	    return 42;
> +	}
> +    }
> +    set src {
> +	#define _GNU_SOURCE
> +	#include <dlfcn.h>
> +	#include <link.h>
> +	#include <stdio.h>
> +	#include <errno.h>
> +
> +	int  main (void) {
> +	    struct r_debug *r_debug;
> +	    ElfW(Dyn) *dyn;
> +	    void *handle;
> +
> +	    /* The version is kept at 1 until we create a new namespace.  */
> +	    handle = dlmopen (LM_ID_NEWLM, DSO_NAME, RTLD_LAZY | RTLD_LOCAL);
> +	    if (!handle) {
> +		printf ("dlmopen failed: %s.\n", dlerror ());
> +		return 1;
> +	    }
> +
> +	    r_debug = 0;
> +	    /* Taken from /usr/include/link.h.  */
> +	    for (dyn = _DYNAMIC; dyn->d_tag != DT_NULL; ++dyn)
> +	        if (dyn->d_tag == DT_DEBUG)
> +	            r_debug = (struct r_debug *) dyn->d_un.d_ptr;
> +
> +	    if (!r_debug) {
> +	        printf ("r_debug not found.\n");
> +		return 1;
> +	    }
> +	    if (r_debug->r_version < 2) {
> +	        printf ("dlmopen debug not supported.\n");
> +		return 1;
> +	    }
> +	    printf ("dlmopen debug supported.\n");
> +	    return 0;
> +	}
> +    }
> +
> +    set libsrc [standard_temp_file "libfoo.c"]
> +    set libout [standard_temp_file "libfoo.so"]
> +    gdb_produce_source $libsrc $lib
> +
> +    if { [gdb_compile_shlib $libsrc $libout {debug}] != "" } {
> +	verbose -log "failed to build library"
> +	return 1
> +    }
> +    if { ![gdb_simple_compile $me $src executable \
> +	       [list shlib_load debug \
> +		    additional_flags=-DDSO_NAME=\"$libout\"]] } {
> +	verbose -log "failed to build executable"
> +        return 1
> +    }
> +
> +    gdb_exit
> +    gdb_start
> +    gdb_reinitialize_dir $srcdir/$subdir
> +    gdb_load $obj
> +
> +    if { [gdb_run_cmd] != 0 } {
> +	verbose -log "failed to start skip test"
> +	return 1
> +    }
> +    gdb_expect {
> +        -re "$inferior_exited_re normally.*${gdb_prompt} $" {
> +            set skip_dlmopen_tests 0
> +        }
> +        -re "$inferior_exited_re with code.*${gdb_prompt} $" {
> +            set skip_dlmopen_tests 1
> +        }
> +        default {
> +	    warning "\n$me: default case taken"
> +            set skip_dlmopen_tests 1
> +        }
> +    }
> +    gdb_exit
> +
> +    verbose "$me:  returning $skip_dlmopen_tests" 2
> +    return $skip_dlmopen_tests
> +}
> +
>  # Return 1 if we should skip tui related tests.

I'm wondering if this can be a gdb_caching_proc?  At the moment, there's
probably not much point since I think it's only used by the new test.
But if we end up with more tests which use it, it might be worthwhile.

Kevin