[Ltrace-devel] [PATCH 4/8] Add LS_TOPLT_GOTONLY

[Petr Machata]

edgar.iglesias at gmail.com writes:

> From: "Edgar E. Iglesias" <edgar at axis.com>
>
> to describe symbols with a GOT entry but without PLT
> entry.
>
> Signed-off-by: Edgar E. Iglesias <edgar at axis.com>
> ---
>  library.h |    1 +
>  1 files changed, 1 insertions(+), 0 deletions(-)
>
> diff --git a/library.h b/library.h
> index f207502..d764da9 100644
> --- a/library.h
> +++ b/library.h
> @@ -31,6 +31,7 @@ struct library;
>  
>  enum toplt {
>  	LS_TOPLT_NONE = 0,	/* PLT not used for this symbol. */
> +	LS_TOPLT_GOTONLY,	/* Has a GOT entry but no PLT.  */
>  	LS_TOPLT_EXEC,		/* PLT for this symbol is executable. */
>  };

This is fine for now, but consider the following.

I see you use that value to decide whether to enable the PLT entry right
away, or wait for a bit.  I have a patch sketched, for support of -l.
For -l, we need to keep around "latent" PLT entries: those that nobody
requested directly (via -e), but that we keep around (disabled) in case
a library is mapped matching the -l criterion.  If that happens, some of
those latent entries may be enabled.

This seems to fit your bill.  In arch_library_symbol_init, you would
mark those symbols as latent (and sym->arch.gotonly = 1 or some such, so
that you can tell them from the rest).  Later, in arch_dynlink_done, you
loop through libraries like you do now, looking for gotonly symbols, and
enabling them.

Thanks,
PM