Re: [RFA] New option "trust-readonly-sections"

[Andrew Cagney <ac131313@cygnus.com>]

> On Thu, Jan 24, 2002 at 09:22:09AM +0200, Eli Zaretskii wrote:
> 
>> 
>> On Thu, 24 Jan 2002, Daniel Jacobowitz wrote:
>> 
> 
>> > I'd rather see this default to on.
> 
>> 
>> That would be an incompatible change.  I think we should avoid such 
>> changes, unless we have a very good reason.
> 
> 
> Stan's reply was convincing.  i guess I've been spoiled by
> protected-memory situations.
> 
> I'd personally like to object to your objection though, Eli. 
> Performance can be a very good reason.  If it wasn't for the other
> drawbacks, I'd consider the argument.
> 
> Perhaps I'm in the minority there, though.


(Would you go near someone wearing an asbestos suit? :-)

It is really important that GDB doesn't lie.  If the tweek is safe then 
certainly enable it.  This tweek _isn't_ safe in embedded targets.

The same goes for things like breakpoints.  GDB pulls them so that the 
target is always left in a clean state.  Not pulling them would be a 
performance bost (knowing the numbers not as much as this one!).

BTW, there are other things that can also be done - for instance 
checking that the target text area hasn't changed.  There is a qCRC 
packet (but from memory it was argued that wasn't strong enough).

Andrew