[Lldb-commits] [PATCH] Properly handle the DYLD attach step for GDB remotes.

[Todd Fiala]

> but I work in a sync so that the exe doesn't exec until after llgs
attaches

That's a bit misleading - what I mean is I have to add a separate
fork/sync/exec launch mechanism to emulate what Darwin provides in a
non-standard posix launch extension not available on Linux.  In both cases,
at the point where llgs/debugserver attach, the inferior is stopped (or
made to stop) prior to beginning of execution, ready to go.


On Thu, Sep 4, 2014 at 8:34 PM, Todd Fiala <tfiala at google.com> wrote:

> Yep - see my comments here that came in about 30 minutes before you posted
> your patch:
>
> https://github.com/tfiala/lldb/issues/57
>
> For using llgs (lldb-gdbserver) for local Linux debugging (i.e. using it
> all the time, instead of ProcessLinux and ProcessMonitor), I follow Apple's
> approach to launching the exe from the lldb side, but I work in a sync so
> that the exe doesn't exec until after llgs attaches.  So, even though we're
> "attaching" on the llgs side, it definitely is hitting the case where the
> rendezvous isn't set up right.  You're working right in the code that I am
> modifying as well.
>
> There's one more check I think I'll need to add to what you're doing.  In
> my scenario, the rendezvous partially resolves but the memory isn't set
> yet, so the resolve step isn't sufficient for a check.  I'm going to ferret
> that out hopefully tonight.
>
> Thanks for digging into this!
>
> -Todd
>
>
> On Thu, Sep 4, 2014 at 3:44 PM, Stephane Sezer <sas at fb.com> wrote:
>
>> No problem. Thanks :)
>>
>> This is basically a bug I found when working on remote debugging; it
>> happens when the remote debugger starts the inferior itself. The inferior
>> is blocked before hitting the entry point (before the rendezvous structure
>> has been setup), which means we can’t load any shared object yet. AFAICT,
>> this can only happen during remote debugging, as attaching to a live
>> process with lldb directly would necessarily attach after the entry point.
>>
>> On Sep 4, 2014, at 3:32 PM, Todd Fiala <tfiala at google.com> wrote:
>>
>> > (I've also added a ton of logging on the 'enable log lldb dyld' flag).
>> >
>> > I'm heading out now but will look at this tonight since I'm all in that
>> code too.
>> >
>> >
>> > On Thu, Sep 4, 2014 at 3:28 PM, Todd Fiala <tfiala at google.com> wrote:
>> > That's bizarre - I'm actually working on this exact code right now...
>> >
>> >
>> > On Thu, Sep 4, 2014 at 3:24 PM, Stephane Sezer <sas at fb.com> wrote:
>> >
>> > _______________________________________________
>> > lldb-commits mailing list
>> > lldb-commits at cs.uiuc.edu
>> > http://lists.cs.uiuc.edu/mailman/listinfo/lldb-commits
>> >
>> >
>> >
>> >
>> > --
>> > Todd Fiala |   Software Engineer |     tfiala at google.com |
>> 650-943-3180
>> >
>> >
>> >
>> >
>> > --
>> > Todd Fiala |   Software Engineer |     tfiala at google.com |
>> 650-943-3180
>> >
>>
>>
>
>
> --
> Todd Fiala | Software Engineer |  tfiala at google.com |  650-943-3180
>



-- 
Todd Fiala | Software Engineer | tfiala at google.com | 650-943-3180
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/lldb-commits/attachments/20140904/efbe9cfe/attachment.html>