RE: [RFC 0/2, gdbserver] Set linux target in async mode in default

[Marc Khouzam <marc dot khouzam at ericsson dot com>]

> -----Original Message-----
> From: Yao Qi [mailto:yao@codesourcery.com] 
> Sent: Tuesday, September 18, 2012 10:35 AM
> To: Marc Khouzam
> Cc: 'gdb-patches@sourceware.org'
> Subject: Re: [RFC 0/2, gdbserver] Set linux target in async 
> mode in default
> 
> On 09/18/2012 10:12 PM, Marc Khouzam wrote:
> > To clarify, after this patch we can still use "-gdb-set 
> target-async off"
> > to turn off async mode right?
> >
> 
> This patch doesn't change the behaviour from the GDB's and Eclipse's 
> perspective.  We can still set target-async on or off in GDB 
> side as needed.
> 
> IIUC, the term "async mode" in GDBserver is different from its 
> counterpart in GDB.  In GDBserver, "async mode" means an "async event 
> loop" which is used for non-stop and notification.  If we look at RSP,
> there is no command to turn "async mode" on or off directly 
> in GDBserver 
> (QNonStop can start async mode on linux target), the state of "async 
> mode" of GDBserver is unknown to GDB, and vice versa.

Sorry, I missed that this was for gdbserver only.


> 
> In current GDBserver/linux, we have only two modes, 'non-stop with 
> async-event-loop' and 'all-stop without async-event-loop'.  
> This patch 
> only changes the latter to 'all-stop with async-event-loop', so that 
> async notification can work on top it.  At the same time, GDB still 
> works in both "target-async on" and "target-async off" with 
> this patched 
> GDbserver.
> 
> > Eclipse currently uses non-async mode for all-stop and some code
> > is based on that assumption (e.g., interrupting the target)
> > so we need to keep non-async available.
> 
> Forgot to mention in my mail that I run testsuite with 
> {native-gdbserver, native-extended-gdbserver} x {async, sync} (in GDB 
> side), and no regression.
> 
> -- 
> Yao
>