This is the mail archive of the gdb-patches@sources.redhat.com mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [patch] gdbserver fails on 32-bit ppc rfs running in a-64 bit 2.6 linux kernel


The strtoul change in my patch was already present. Sorry about that.

You have changed the data type of thread_resume::thread as well as cont_thread 
to unsigned long. "cont_thread = -1" and "(cont_thread > 0)" are still in 
place. How does this work?

Can we use a thread value of "0" to indicate all threads or no-thread instead 
of "-1"? The condition "cont_thread > 0" becomes valid if we do that.

-Amit

On Monday 14 Mar 2005 9:52 pm, Daniel Jacobowitz wrote:
> On Mon, Mar 14, 2005 at 09:37:35PM +0530, Amit S. Kale wrote:
> > Hi,
> >
> > I found that gdbserver fails on a 32-bit ppc rfs running in a 64-bit 2.6
> > linux
>
> "rfs"?  It'd be nice to explain one's acronyms...
>
> > kernel. The thread id's fetched from thread-db library are above
> > 0x7fffffff. gdbserver has implicit assumptions that thread id's are
> > "signed integers". This is true when running under a 32-bit kernel, but
> > not true here. Attached patch fixes this problem by changing the "strtol"
> > that scans the thread-id to "strtoul" and by changing the "cont_thread >
> > 0" comparison to "cont_thread != -1".
> >
> > Is there any likelyhood of this assumption being made in some other
> > places in gdbserver or gdb?
>
> Please see:
>
> 2005-03-03  Daniel Jacobowitz  <dan@codesourcery.com>
>
>         * inferiors.c (change_inferior_id, add_thread, find_inferior_id):
>         Take unsigned long arguments for PIDs.
>         * linux-low.c (add_process, linux_attach_lwp, linux_attach)
>         (linux_thread_alive, linux_wait_for_event, kill_lwp, send_sigstop)
>         (wait_for_sigstop, linux_resume_one_process)
>         (regsets_fetch_inferior_registers, linux_send_signal)
>         (linux_read_auxv): Likewise.  Update the types of variables holding
>         PIDs.  Update format string specifiers.
>         * linux-low.h (struct process_info, linux_attach_lwp): Likewise.
>         * remote-utils.c (prepare_resume_reply): Likewise.
>         * server.c (cont_thread, general_thread, step_thread)
>         (thread_from_wait, old_thread_from_wait, signal_pid): Change type
> to unsigned long.
>         (handle_query): Update format specifiers.
>         (handle_v_cont, main): Use strtoul for thread IDs.
>         * server.h (struct inferior_list_entry): Use unsigned long for ID.
>         (add_thread, find_inferior_id, change_inferior_id, cont_thread)
>         (general_thread, step_thread, thread_from_wait)
>         (old_thread_from_wait): Update.
>         * target.h (struct thread_resume): Use unsigned long for THREAD.
>         (struct target_ops): Use unsigned long for arguments to attach and
>         thread_alive.
>
> Your sources are dated after this but don't seem to include this fix.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]