Date: Tue, 22 Oct 2002 19:51:26 +0200 From: marius@alchemy.franken.de To: Poul-Henning Kamp <phk@FreeBSD.ORG> Cc: hackers@FreeBSD.ORG, current@FreeBSD.ORG Subject: Re: 5.0-RUSH: -current install testers wanted! Message-ID: <20021022175126.GA89712@newtrinity.zeist.de> In-Reply-To: <15328.1035268433@critter.freebsd.dk> References: <15328.1035268433@critter.freebsd.dk>
next in thread | previous in thread | raw e-mail | index | archive | help
Last week I replace a broken mainboard with a dual-Athlon one (Tyan
Tiger s2466n-4m) and decided to upgrade that box from 4-stable to
-current by installing the 0917-jpsnap via the floppies and passive
ftp. I hit several sysinstall-problems some of which my already be
fixed:
- The hd I install -current onto previously had 4-stable on it, I
deleted slice one (was the only one) and created a new one and
selected the standard MBR. I decided to give UFS2 a try and created
the filesystems with '-O 2 -U' (there was some problem toggling
Softupdates und just adding '-O 2' to the newfs-options). But after
rebooting the 4-stable (!) bootloader came claiming it wasn't able
to load /kernel, a `ls` at the bootloader-prompt showed the contents
of the former root-fs, even the former contents of some of the sub-
direcroties , e.g. /etc could be displayed.
- After a `dd if=/dev/zero of=/dev/da0 count=16` (booted with another
hd containing 4-stable) I repeated the above procedure and ended up
with the -current bootloader yelling "No UFS" several times. For now
I ended up having the root-fs UFS1 and var, usr and tmp UFS2.
This problem seems to be on the todo-list.
- During the 6-7 sysinstall-runs (it hung and crashed unreproduceable
4-5 times) I always had the problem that after configuring the nic
with a ipv4-adress it took some random time between ~15 seconds up
to several minutes to look up the hostname of the jpsnap-server.
This definitely wasn't a network- or dns-problem, another box
connected via the same line and using the same nameserver didn't
have problems looking up the hostname. I ran tcpdump on the other
box and the reason for this seems to be sysinstall doing ipv6
neigbhourhood-detection and ping6 the ipv6-address of the jpsnap-
server also the interface wasn't configured for ipv6. I also saw
some stuff that I don't know of what it is:
fe80::2e0:81ff:fe22:d7cf > ff02::2:f8c7:7880: HBH icmp6: multicast listener report max resp delay: 0 addr: ff02::2:f8c7:7880 [hlim 1]
- The 3 or 4 times I got to the point to set the root password the
prompt to enter it popped up in ttyv1 and not in tty0 like the
rest of sysinstall. I wasn't actually able to set one but sysinstall
returned to the "post-install-configuration"-menue when hitting
ctrl-c in ttyv0.
Some problems I have with the installed -current:
- The bios of the board offers ACPI-support and as I thought FreeBSD's
support of this is advanced enough I decided to turn it but it turned
out to not be SMP-"safe". I can't remeber a panic while running an
UP-kernel for the short time to update to latest -current and build
a SMP-kernel. When both ACPI- (via kld) and SMP-support are enabled
the box is fscking unstable, I get about 3 lock-order-reversal- and
locking-against-myself-panics per hour and occasionally spontaneous
reboots. After turing of ACPI-support in the bios the ACPI-kld no
longer gets loaded and the box runs stable for 3 days (no more
panics or spontaneous reboots), still with the same kernel built
of sources as of Oct 17.
The mainboard has 2 pci-bridges, they and all devices behind them
successfully get probed when running with ACPI enabled so this
doesn't sound like the problem described in the todo-list.
acpiconf doesn't work except for `acpiconf -s 1` (after an
`acpiconf -e`), `acpiconf -s 1` turns off the output of the gfx-
card for the fraction of second (once also the hd sounded as it
would spin-down) and as soon it returns 2 resume-messages get
displayed. Doing this causes a panic (pagefaults iirc) in about
1 of 5 times.
- After the tons of panics I got the background fsck always cleaned
up an alarming number of files and directories on the UFS2-
filesystems, much more than I've ever seen after a panic of a
4-stable box. As I got most panics while extracting tarballs or
building ports this could be ok but once also file I successfully
downloaded some 30-60 seconds before a panic got deleted during
the fsck-run, imho this shouldn't happen.
Last but not least it would be fine if sysinstall would also support
ATAPI-floppies (/dev/afd0) for mounting the fixit-floppy. I didn't
check recently but I think support for mounting the live-cdrom in
SCSI-cdroms is also broken, last time I tried it also wasn't
possible to install from a SCSI-cdrom as sysinstall didn't detect
/dev/cd0c. Hrm, the minor looks wrong in devices.c, could this be the
reason ?
static struct _devname {
DeviceType type;
char *name;
char *description;
int major, minor, delta, max;
} device_names[] = {
{ DEVICE_TYPE_CDROM, "cd%dc", "SCSI CDROM drive", 15, 2, 8, 4
but:
ls -la /dev/cd0c
crw-r----- 1 root operator 15, 0 Oct 22 19:02 /dev/cd0c
At least the minor matches for acd0c.
Marius
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20021022175126.GA89712>