Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!cs.utexas.edu!uunet!sco!seanf
From: seanf@sco.COM (Sean Fagan)
Newsgroups: comp.arch
Subject: Re: break on data reference, was Re: More mips
Message-ID: <3461@scolex.sco.COM>
Date: 20 Oct 89 03:03:29 GMT
References: <771301127@8909291517.AA00260@maxwell.ece.c> <130800001@peg> <6535@pt.cs.cmu.edu> <1989Oct16.225206.22386@paris.ics.uci.edu> <1989Oct17.045410.399@esegue.segue.boston.ma.us>
Reply-To: seanf@sco.COM (Sean Fagan)
Organization: The Santa Cruz Operation, Inc.
Lines: 35

In article <1989Oct17.045410.399@esegue.segue.boston.ma.us> johnl@esegue.segue.boston.ma.us (John R. Levine) writes:
>>>"breakpoint at data address". ...
>Some of us low-budget types feel pretty lucky.  The 80386 can trap up to four
>memory addresses, separately monitoring each for fetch, read, or write.
>When this feature is in use your program runs slower, but it's a heck of
>a lot faster than instruction at a time tracing.
>Astonishingly, AT&T System V/386 ptrace() lets you read and write the debug
>registers (take that, you Berkeloids) that control this swell feature.
>I keep meaning to teach gdb how to use them.  Maybe somebody else already
>has.

This is a brief pseudo-plug, followed by a real question...

Anyway:  CodeView(tm) under SCO UNIX 3.2 uses the debug registers to set
'Watchpoints' and 'tracepoints.'  I think it also uses other debug registers
to set breakpoints.  A problem with this is that there are only 4 register
(of each kind, I believe:  4 code breakpoint and 4 data breakpoint, but I'm
not sure), thus you can only set 4 breakpoints...  BTW, it is a very nice
feature.  I first saw something like it in dbx on a Sun (trace <var>), but
using it slowed it down incredibly (it had to single step, I believe).  With
codeview, it runs along at full speed until it accesses the variable.
*Very* nice.

Lots of other processors have these, of course, including, say, the NS32xxx
series.  My chief complaint is that they only catch a single address; I
think they would have been much more useful if it had been <address,
length>.  For a real-world example, I could then tell codeview to stop
whenever any element of a structure has changed.  Anybody got a reason why
it never seems to be implemented that way, other than not slowing it down (I
would not be surprised if it slowed down memory accesses by 1 or 2 cycles)?

-- 
Sean Eric Fagan  | "Time has little to do with infinity and jelly donuts."
seanf@sco.COM    |    -- Thomas Magnum (Tom Selleck), _Magnum, P.I._
(408) 458-1422   | Any opinions expressed are my own, not my employers'.