Date: Tue, 10 Jun 2003 12:11:13 -0700 From: Marcel Moolenaar <marcel@xcllnt.net> To: Alan Robinson <alan.robinson@fujitsu-siemens.com> Cc: freebsd-ia64@freebsd.org Subject: Re: DDB and SMP causes Unaligned Reference Message-ID: <20030610191113.GA14015@ns1.xcllnt.net> In-Reply-To: <20030610112735.A20041@fujitsu-siemens.com> References: <20030606183130.A19592@fujitsu-siemens.com> <20030606182015.GC569@dhcp01.pn.xcllnt.net> <20030610105306.A18863@fujitsu-siemens.com> <20030610112735.A20041@fujitsu-siemens.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Tue, Jun 10, 2003 at 11:27:35AM +0200, Alan Robinson wrote:
> On Tue, Jun 10, 2003 at 10:53:06AM +0200, Alan Robinson wrote:
> > On Fri, Jun 06, 2003 at 11:20:15AM -0700, Marcel Moolenaar wrote:
> > >
> > > We don't set pc_pcb on ia64. The line above should read:
> > > savectx(curthread->td_pcb);
> > >
> > Tried it.
> > >
> > > Try it and let me know if it works. I'll commit it then.
> > >
> > Works.
> > Works.
> > (It still causes the strange double entry into DDB)
> >
> > It would also be a good idea to remove the unneeded pc_pcb line with its
> > incorrect comment from src/sys/ia64/include/pcpu.h
> >
>
> I deleted the pc_pcb line from src/sys/ia64/include/pcpu.h and
> much to my surprise got a compile error in src/sys/ia64/ia64/machdep.c
> where the function cpu_pcpu_init does indeed set pc_pcb:
>
> void
> cpu_pcpu_init(struct pcpu *pcpu, int cpuid, size_t size)
> {
> KASSERT(size >= sizeof(struct pcpu) + sizeof(struct pcb),
> ("%s: too small an allocation for pcpu", __func__));
> pcpu->pc_pcb = (void*)(pcpu+1);
> }
>
> so much for it not being set!. It is set, just the alignment of
> struct pcb is wrong.
Oops. I guess I was mistaken with the i386 NPX (non-SMP) specific
pc_curpcb.
So, yes; it means we need to make sure *pc_pcb is aligned and the
place to do it is exactly the place where we initialize it :-)
> Is your suggested (and working) change
> > > savectx(curthread->td_pcb);
> still correct ?
Probably not. I suspect there's a corner case where we cannot use
the PCB of the currently running thread, even though it's otherwise
the best PCB to put it in, AFAICT. Chances are DDB expects the
current context to be saved in pc_pcb as well. The bottom line is
that I was confused, so I have to thread carefully now...
--
Marcel Moolenaar USPA: A-39004 marcel@xcllnt.net
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20030610191113.GA14015>