Date: Fri, 10 Sep 1999 18:02:26 +0200 From: Juergen Lock <nox@jelal.kn-bremen.de> To: Bruce Evans <bde@zeta.org.au> Cc: Juergen Lock <nox@jelal.kn-bremen.de>, Luoqi Chen <luoqi@watermarkgroup.com>, marcus@jet.franken.de, FreeBSD-emulation@FreeBSD.ORG, wine-devel@winehq.com Subject: Re: debugger, stepi, c(ontinue) from breakpoint... Message-ID: <19990910180225.A22248@saturn.kn-bremen.de> In-Reply-To: <Pine.BSF.4.10.9909091500320.14184-100000@alphplex.bde.org> References: <19990908225333.C1394@saturn.kn-bremen.de> <Pine.BSF.4.10.9909091500320.14184-100000@alphplex.bde.org>
next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, Sep 09, 1999 at 03:36:50PM +1000, Bruce Evans wrote:
> On Wed, 8 Sep 1999, Juergen Lock wrote:
>
> > On Wed, Sep 08, 1999 at 02:33:57PM -0400, Luoqi Chen wrote:
> > > >
> > > > - if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) {
> > > > + if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM) &&
> > > > + *callp->sy_call != sigreturn && *callp->sy_call != linux_sigreturn) {
> > > > /* Traced syscall. */
> > > > frame.tf_eflags &= ~PSL_T;
> > > > + /* tell the signal handler this is a trace trap */
> > > > + frame.tf_trapno = T_TRCTRAP;
> > > > trapsignal(p, SIGTRAP, 0);
>
> This should probably be trapsignal(p, SIGTRAP, T_TRCTRAP). Signal handlers
> should examine the signal code (the 3rd arg to trapsignal()) before examining
> tf_trapno. It may be a bug to even look at tf_trapno. The signal code is
> supposed to be a properly translated version of tf_trapno.
Hmm, currently code is often 0, for example in SIGTRAP... (for both trap
and breakpoint.)
>
> I think the purpose of (original version of the) above code is only to
> make a traced syscall stop on return from the syscall and not one
> instruction later.
Yep.
> However, this is wrong if the trace flag was set
> by sigreturn(). Your changes seem to fix this case but break the case
> where sigreturn() is being traced but is not responsible for setting the
> trace flag.
True...
> This can probably be fixed by calling trapsignal() if the
> trace flag was set on entry to syscall(), not if it set on exit from
> syscall().
Indeed! If i do that,
Index: i386/i386/trap.c
===================================================================
RCS file: /home/cvs/cvs/src/sys/i386/i386/trap.c,v
retrieving revision 1.133.2.1
diff -u -u -r1.133.2.1 trap.c
--- trap.c 1999/08/29 16:05:56 1.133.2.1
+++ trap.c 1999/09/10 14:42:21
@@ -1041,6 +1084,7 @@
int error;
int args[8];
u_int code;
+ int tracedsyscall = ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM));
#ifdef DIAGNOSTIC
if (ISPL(frame.tf_cs) != SEL_UPL)
@@ -1135,10 +1179,12 @@
break;
}
- if ((frame.tf_eflags & PSL_T) && !(frame.tf_eflags & PSL_VM)) {
+ if (tracedsyscall) {
/* Traced syscall. */
frame.tf_eflags &= ~PSL_T;
- trapsignal(p, SIGTRAP, 0);
+ /* tell the signal handler this is a trace trap */
+ frame.tf_trapno = T_TRCTRAP;
+ trapsignal(p, SIGTRAP, T_TRCTRAP);
}
userret(p, &frame, sticks);
I can single-step in wine's debugger and single-step over the sigreturn
syscall from gdb like it should be. (at least in a simple test program, i
first tried to test this on wine itself but there i got strange signals when
i tried to set breakpoints at the end of its signal handlers. this may have
to do something with the fact that wine uses sigaltstack, i don't know...)
>
> > > > }
> > > >
> > > > (so wine must be the first program that handles SIGTAP for itself and
> > > > sets the trace bit from a signal handler? :) I Cc'd this to -emulation,
>
> Quite possibly.
Heh.
Regards,
--
Juergen Lock <nox.foo@jelal.kn-bremen.de>
(remove dot foo from address to reply)
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-emulation" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?19990910180225.A22248>