Date: Sat, 31 Jul 2004 00:27:13 +0000 From: Alexander Kabaev <kan@freebsd.org> To: freebsd-current@freebsd.org Subject: Re: boot2 -- Round 2 Message-ID: <20040731002713.GA6709@freefall.freebsd.org> In-Reply-To: <20040730212843.GA33955@parodius.com> References: <20040730212843.GA33955@parodius.com>
next in thread | previous in thread | raw e-mail | index | archive | help
On Fri, Jul 30, 2004 at 02:28:43PM -0700, Jeremy Chadwick wrote:
> So, in regards to the commited fix:
>
> This seemed to fix the issue on one of my boxes (the one which was
> flat-out panic'ing, not the one which was reporting 0:ad(0,`) as the
> default slice to load /boot/loader from). I'll refer to the one which
> panic'd as "Box A" while the one which is doing the backtick as "Box B".
>
> After pulling cvs down last night and rebuilding world+kernel+boot
> blocks, running disklabel -B ad0s1, all on Box B, I found the machine
> once again spitting out "Invalid partition", trying to load loader(8)
> off of 0:ad(0,`) instead of 0:ad(0,a). I double-checked boot2/Makefile
> to see if -fno-unit-at-a-time was in place -- and it was.
>
> I've tried using /boot/boot off of Box A and applying it to Box B using
> disklabel -B -b /boot/box_b/boot ad0s1 to no avail.
>
> It seems almost as if the boot2 code is broken in such a way that it
> resembles an "off-by-one" error (ASCII 0x60 == `, ASCII 0x61 == a).
> Why it's picking ` is beyond me...
>
> Can someone shed some light as to how I can go about debugging this,
> as well as mention how I can temporarily work around this? Box B
> happens to run mysqld, and is suffering from some issues mentioned on
> freebsd-threads (re: machine randomly hard-locking), so it definitely
> needs to be able to boot back up on it's own without my intervention.
>
> Thanks!
Hi,
I guess I would like to get your /boot/boot. The one I got simply works
on all boxes in my home :(.
As another option, you can try an alternative patch which was proposed
by Tim Robbins. Since the problem was apparently caused by me going back to
static memcpy implementation, I am currenly working on using builtin
memcpy as it was used before. I will post it later after I've done some
more testing and if things will look good.
--
Alexander Kabaev
======== Begin quote ==============
After a few hours of head-scratching, I've tracked down the problem with
boot2 and -funit-at-a-time, and come up with a patch that makes it work:
==== //depot/user/tjr/freebsd-tjr/src/sys/boot/i386/boot2/boot2.c#7 - /home/tim/p4/src/sys/boot/i386/boot2/boot2.c ====
@@ -139,7 +139,16 @@
static int xgetc(int);
static int getc(int);
-static void memcpy(void *, const void *, int);
+/*
+ * GCC 3.4 with -funit-at-a-time (implied by -Os) may use a non-standard
+ * calling convention for static functions, using registers to pass arguments
+ * instead of the stack. However, GCC may emit calls to memcpy() when a
+ * program copies a struct with the assignment operator, and the code it
+ * emits to call memcpy() uses the standard convention, not the register
+ * convention. This means we must declare our memcpy() implementation "__used"
+ * to disable the register calling convention.
+ */
+static void memcpy(void *, const void *, int) __used;
static void
memcpy(void *dst, const void *src, int len)
{
I think this is a bug in GCC; it should emit a warning if it's about to emit
code to call memcpy(), but finds that memcpy() has a prototype that conflicts
with the assumptions it makes.
Tim
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040731002713.GA6709>