From mboxrd@z Thu Jan  1 00:00:00 1970
From: "William A. Gatliff" <bgat@saturn.billgatliff.com>
To: cygwin@cygwin.com, gdb@sources.redhat.com
Subject: Re: [1.3.3] breaks serial i/o?
Date: Fri, 19 Oct 2001 14:55:00 -0000
Message-id: <20011019165550.A1767@saturn.billgatliff.com>
References: <20011018161003.A3059@saturn.billgatliff.com> <20011018222406.C11830@redhat.com> <20011019085618.A5013@saturn.billgatliff.com> <20011019114712.A23101@visi.com> <20011019130301.A26885@redhat.com> <20011019133042.J1140@saturn.billgatliff.com> <20011019141204.L1140@saturn.billgatliff.com>
X-SW-Source: 2001-10/msg00202.html

Guys (and gals too!):


> Here's some of the (apparently) relevant code in gdb's RDI support:

NOnonono, that was all wrong.  For starters, under Cygwin it doesn't
compile with COMPILING_FOR_WINDOWS at all.

I whipped out the oscilloscope, and confirmed that, in fact, I do get
serialio on com1 under Cygwin 1.3.3 whether CTS/RTS are tied together
or not.  This is news to me, as the problem seemed to follow
CTS/RTS...

I'm now running arm-elf-gdb under gdb under cygwin-1.3.3.  And from
what I've seen so far, there are just general instabilities everywhere
in the serial i/o related stuff.

It seems to frequently hang at line 307 of unixcomm.c, in the
Unix_ReadSerial() function.  It's as if it goes into read(), and never
returns.  Serpfd==3, err==1 in most cases.  It doesn't always hang
there, but it has hung there several times.

Here's the code:

gdb/rdi-share/unixcomm.c

extern int Unix_ReadSerial(unsigned char *buf, int n, bool block)
{
    fd_set fdset;
    struct timeval tv;
    int err;
 
    FD_ZERO(&fdset);
    FD_SET(serpfd, &fdset);
 
    tv.tv_sec = 0;
    tv.tv_usec = (block ? 10000 : 0);
 
    err = select(serpfd + 1, &fdset, NULL, NULL, &tv);
 
    if (err < 0 && errno != EINTR)
    {
#ifdef DEBUG
        perror("select");
#endif
        panic("select failure");
        return -1;
    }
    else if (err > 0 && FD_ISSET(serpfd, &fdset))
      {
        int s;
 
        s = read(serpfd, buf, n);
        if (s < 0)
          perror("read:");
        return s;
...


I'll try it under 1.3.2 and see if I can get any insights (pardon the pun).

Ideas?



b.g.
-- 
Bill Gatliff
bgat@billgatliff.com