Re: [RFC] 00/10 non-stop support (and linux native support)

[Pedro Alves <pedro at codesourcery dot com>]

A Tuesday 06 May 2008 16:45:44, Pedro Alves wrote:
> Hi,
>

Hi again,

When I posted this, I forgot to say a couple of things.
First, that the series was tested withtout regressions on 
x86-pc-linux-gnu and arm-linux-gnueabi; second, that this series
depends on a couple of patches that aren't in HEAD yet:

Those were:

 [ Use observers to report stop events. ]
 http://sourceware.org/ml/gdb-patches/2008-05/msg00160.html

 [ Implement *running. ]
 http://sourceware.org/ml/gdb-patches/2008-05/msg00170.html

Since posting I noticed that stop_bpstat should be context switched
in non-stop, and breakpoint.c needed fixing:

 [ remove global stop_bpstat dependency from breakpoints module ]
 http://sourceware.org/ml/gdb-patches/2008-05/msg00266.html

I'm about to post a few updated patches to reflect that, and also
fix a couple of additional bugs.

> This series adds inferior control support for non-stop mode, and
> then adds linux native support on top.
>
> There are still ongoing discussions on how to activate and
> expose to frontends the non-stop support.  MI seems to getting
> stable enough now.
>
> This series adds no docs yet, as they are yet to be writen, but
> we'll get to it.
>
> In the meantime, I'd like to get comments on the rest of the
> code, and maybe even a couple of testers.
>
> One patch I'm not adding in this series , is the support
> for GDB to *not* switch threads automatically on events
> (in non-stop mode), and what to do when the current thread exits.
> I'll post those separatelly.
>
> MI/non-stop is our main focus, but CLI works a bit too, although
> there are a lot of rough edges:
>
> Debugging a multi-threaded app has these looks:
>
> (gdb) r&
> Starting program: /home/pedro/gdb/tests/threads32
> (gdb) [Thread debugging using libthread_db enabled]
> [New LWP 23453]
> [New LWP 23454]
> info threads
>   3 Thread 0xf7603b90 (LWP 23454)  (running)
>   2 Thread 0xf7e04b90 (LWP 23453)  (running)
> * 1 LWP 23450  (running)
> (gdb) thread 2
> [Switching to thread 2 (Thread 0xf7e04b90 (LWP 23453))] (running)
> (gdb) interrupt
> (gdb)
> Program received signal SIGINT, Interrupt.
> [Switching to Thread 0xf7e04b90 (LWP 23453)]
> 0xffffe410 in __kernel_vsyscall ()
> b 63
> Breakpoint 1 at 0x80485a6: file threads.c, line 63.
> (gdb) info threads
> During symbol reading, incomplete CFI data; unspecified registers (e.g.,
> eax) at 0xffffe411.
>   3 Thread 0xf7603b90 (LWP 23454)  (running)
> * 2 Thread 0xf7e04b90 (LWP 23453)  0xffffe410 in __kernel_vsyscall ()
>   1 Thread 0xf7e056b0 (LWP 23450)  (running)
> (gdb) c&
> Continuing.
> (gdb)
> Breakpoint 1, thread_function0 (arg=0x0) at threads.c:63
> 63              (*myp) ++;
> info threads
>   3 Thread 0xf7603b90 (LWP 23454)  (running)
> * 2 Thread 0xf7e04b90 (LWP 23453)  thread_function0 (arg=0x0) at
> threads.c:63 1 Thread 0xf7e056b0 (LWP 23450)  (running)
>
> (gdb) b 80
> Breakpoint 2 at 0x80485f0: file threads.c, line 80.
> (gdb) [Switching to Thread 0xf7603b90 (LWP 23454)]
>
> Breakpoint 2, thread_function1 (arg=0x1) at threads.c:80
> 80              (*myp) ++;
> info threads
> * 3 Thread 0xf7603b90 (LWP 23454)  thread_function1 (arg=0x1) at
> threads.c:80 2 Thread 0xf7e04b90 (LWP 23453)  thread_function0 (arg=0x0) at
> threads.c:63 1 Thread 0xf7e056b0 (LWP 23450)  (running)
> (gdb)
>
>
> The series is layed out as:
>
> 001 Add "executing" property
>
>  Adding an "executing" property to GDB's view of threads.
>
> 002 Don't trim ptids on fork/exec
>
>  Don't trim ptids on follow fork/vfork/exec, so the previous
>  patch works in those cases.
>
> 003 Add non-stop global
>
>  Add a non_stop global, and associated "set/show non-stop"
>  commands.  Split out or core non-stop, because the following
>  patch uses it.
>
> 004 per-thread commands in non-stop mode
>
>  In non-stop mode, we can have one exec command active per-thread,
>  and they're independant, while in all-stop, there can only be one,
>  and it's global.  This adds the necessary context-switching glue.
>
> 005 Expose context_switch
>
>  In non-stop mode, whenever we switch threads, and because each
>  is handled independently of the others, we need to
>  context-switch outside of infrun.c (e.g., "thread n" command,
>  thread apply all, "continue --all".)
>
> 006 Don't rely on ecs->wait_for_more before fetching the event
>
>  ecs->wait_for_more is being checked before fetching the pending
>  event from the target, and if not, then we're clearing some state.
>  This patch avoids the need to context switching that variable, by
>  moving that code in questiong to proceed.
>
> 007 non-stop inferior control
>
>  This adds the inferior control in non-stop mode.
>
> 008 Enable thread commands
>
>  Now that "info threads" and "thread" are "running" thread aware,
>  enable them.
>
> 009 Add "continue --all"
>
>  This adds "continue --all" for non-stop mode.
>
> 010 non-stop for linux native
>
>  Adds non-stop support for linux native.



-- 
Pedro Alves