amd64/106109: amd64: si_addr is not set when sending a signal

[David Xu]

The following reply was made to PR amd64/106109; it has been noted by GNATS.

From: David Xu <davidxu at freebsd.org>
To: freebsd-amd64 at freebsd.org,
 NIIMI Satoshi <sa2c at sa2c.net>
Cc: FreeBSD-gnats-submit at freebsd.org
Subject: Re: amd64/106109: amd64: si_addr is not set when sending a signal
Date: Fri, 1 Dec 2006 13:21:09 +0800

 On Friday 01 December 2006 12:11, NIIMI Satoshi wrote:
 > >Number:         106109
 > >Category:       amd64
 > >Synopsis:       amd64: si_addr is not set when sending a signal
 > >Confidential:   no
 > >Severity:       serious
 > >Priority:       low
 > >Responsible:    freebsd-amd64
 > >State:          open
 > >Quarter:
 > >Keywords:
 > >Date-Required:
 > >Class:          sw-bug
 > >Submitter-Id:   current-users
 > >Arrival-Date:   Fri Dec 01 04:20:03 GMT 2006
 > >Closed-Date:
 > >Last-Modified:
 > >Originator:     NIIMI Satoshi
 > >Release:        FreeBSD 6.2-RC1 i386
 > >Organization:
 > >Environment:
 >
 > System: FreeBSD berkeley.l.sa2c.net 6.2-RC1 FreeBSD 6.2-RC1 #0: Thu Nov 30
 > 10:03:58 JST 2006 root at berkeley.l.sa2c.net:/usr/obj/usr/src/sys/GENERIC
 > i386
 >
 > >Description:
 >
 > POSIX style signal handers expect that the faulting address is stored
 > in si_addr member of siginfo_t.
 >
 > But the address is passed to signal handler only as non-portable
 > fourth argument on FreeBSD/amd64.
 >
 > >How-To-Repeat:
 >
 > Following program produces
 > i386: &main=0x80485e8, si_addr=0x80485e8, fourth_arg=0x0
 > amd64: &main=0x400780, si_addr=0x0, fourth_arg=0x400780
 >
 > #include <signal.h>
 > #include <stdio.h>
 > #include <stdlib.h>
 >
 > int main();
 >
 > void
 > handler(int sig, siginfo_t *siginfo, void *context, void *addr)
 > {
 > 	fprintf(stderr, "&main=%p, si_addr=%p, fourth_arg=%p\n",
 > 		&main, siginfo->si_addr, addr);
 > 	exit(1);
 > }
 >
 > int
 > main()
 > {
 > 	struct sigaction sa;
 >
 > 	sa.sa_flags = SA_SIGINFO;
 > 	sigemptyset(&sa.sa_mask);
 > 	sa.sa_sigaction = (void (*)(int, siginfo_t *, void *))handler;
 >
 > 	sigaction(SIGBUS, &sa, NULL);
 > 	sigaction(SIGSEGV, &sa, NULL);
 > 	*(int *)main = 1;
 >
 > 	return 0;
 > }
 >
 > >Fix:
 >
 > --- amd64-machdep.diff begins here ---
 > Index: machdep.c
 > ===================================================================
 > RCS file: /home/ncvs/src/sys/amd64/amd64/machdep.c,v
 > retrieving revision 1.664
 > diff -u -r1.664 machdep.c
 > --- machdep.c	19 Nov 2006 20:54:57 -0000	1.664
 > +++ machdep.c	1 Dec 2006 03:36:54 -0000
 > @@ -304,6 +304,7 @@
 >  		/* Fill in POSIX parts */
 >  		sf.sf_si = ksi->ksi_info;
 >  		sf.sf_si.si_signo = sig; /* maybe a translated signal */
 > +		sf.sf_si.si_addr = ksi->ksi_addr;
 >  		regs->tf_rcx = (register_t)ksi->ksi_addr; /* arg 4 in %rcx */
 >  	} else {
 >  		/* Old FreeBSD-style arguments. */
 > --- amd64-machdep.diff ends here ---
 
 I reviewed the sendsig() in RELENG_6 for AMD64, the si_addr is not set,
 but is set in i386 version of sendsig(), the originator's patch is wrong,
 it copied some code from HEAD which is not valid for RELENG_6, the
 HEAD has fully working siginfo_t in kernel while RELENG_6 does not, 
 the patch should be changed to:
 
 > +		sf.sf_si.si_addr = regs->tf_addr;
 
 
 David Xu