# $NetBSD: LAST_MINUTE,v 1.18 1998/01/20 17:12:19 jeremy Exp $ This file contains important information on the NetBSD 1.3.1 release that did not make it into the main documentation. This file is organized by architecture. Please read the "all architectures" section and the section relevant to your hardware below it. [all architectures] The "etc" set contains the file "/var/cron/tabs/root". People upgrading from an older NetBSD version should be aware that this file should be backed up before the upgrade. Unfortunately, the install may overwrite it. This affects installation using the "sysinst" program, used, for example, on the i386 and pmax. Architectures using ISA buses, in particular Ethernet cards: please note that the "ed" ethernet driver has been split up into several drivers. An ethernet interface that used to have the name "ed" will now either be called "ne", "ec" or "we". Please refer to the "Supported Hardware" section to find out which is applicable. [port-alpha] There are a few minor gotchas in the install that are documented in alpha/README.first. Please read that carefully before reading the INSTALL document. There is a serious bug in sys/stat.h that causes the alignment of struct stat to change when _POSIX_SOURCE is defined, which in turn causes some of the return values from stat/fstat/lastat to appear to be bogus. (This affects only the alpha.) We are working on a fix for this. In the meantime, if you compile a program that appears to get wrong the size or time, or possibly other attributes, of a file, you are probably being affected by this. If you don't know what to do, turn to the port-alpha mailing list for help. [port-amiga] Please note that the Ariadne Ethernet card uses now the generic lance ethernet driver. An ethernet interface that used to have the name "ae" will now be called "le". Configuration files such as "hostname.ae0" should be updated to the new name. Some users have reported kernel panics when doing lots of Ariadne i/o. We managed to analyze this in one case, and find a bug fix, but too late for the 1.3.1 cut. You'll find the source diff file in source/amiga.ariadne-fix.diff.gz and a patched kernel and miniroot in amiga/binary/kernel/netbsd-with-ariadne-fix and amiga/installation/miniroot/miniroot-with-ariadne-fix.fs. If you were one of the former victims, please report wether this works for you. [port-i386] Christian Hopps reports having some trouble with the ncr scsi driver on his machine. It manifests itself by repeated and unreported read errors on the drive -- repeated md5's of large files (like the base.tgz file in the NetBSD distribution) return different values, for example. It is not known if there is a general problem with the ncr driver; the driver appears to have been working for most users. However, there is a possibility that the latest version is more fragile than earlier ones. Most likely, the problem occurs with older PCI chipsets, as used in 486 PCI boards. A probable candidate for this problem is the UMC UM8881F chipset. For the name of your PCI chipset, check the "pchb0" line in the NetBSD boot messages. Users are advised to test their systems after booting using a method like repeated md5 runs on large files, as described above. Please report problems if you have them. [port-pmax] The released 1.3.1 source has a bug in the pmax-specific autoconfiguration code for th real-time clock. We inadvertently pulled up a post-1.3.1 change; when used with the 1.3.1 /usr/sbin/config, this caues kernels to panic early in boot. A patch for this problem is in the NetBSD-1.3.1/pmax-rtc.diff.gz. The patch has been applied to the kernels in the max binary distribution and media. Apply the patch _after_ applying the 1.3.1 patch. Alternatively, backing out the 1.3.1 changes to sys/arch/pmax/tc/mcclock_ioasic.c also works. The pmax /usr/libexec/ld.so binary in 1.3.1 has been updated to fix problems with symbols defined in more than one shared library. Source for the updated ld.so are in the normal pmax FTP directory. [port-sun3] The 1.3.1 sources create a GENERIC kernel that is too large to boot on a standard (4MB) Sun3/50. That has been corrected by providing a configuration called INSTALL and using that instead of GENERIC for the miniroot image. The source code changes for this correction are in NetBSD-1.3.1/sun3/PATCHES [port-sun3x] There is one potential pitfall in the 1.3.1 release of NetBSD/sun3x. 1. The RAMDISK kernel used during the install process (netbsd-rd.gz) is missing a section of code which allows it to automatically detect that it has an embedded filesystem. At boot, it will fail to find the ``md0'' device on its own (which contains the filesystem) and will prompt for the name of a different device to query. The default, ``le0'', is incorrect and if chosen, will cause the machine to hang. Instead, enter ``md0'' and the boot process will continue as normal. (You will be prompted for the dump device and filesystem type as well. Select the default in both cases.)