Setting up an old HP8000 desktop small-form-factor system as a crash-and-burn test system, with dual-boot Fedora 29 and Windows 7, on separate drives.

First slight gotcha with the Fedora install:
The installation utility shows the two disks, the first of which has Windows 7 (sda) and the second – additional – was for the Fedora Linux install (sdb). One might assume that only the second disk should be selected as this is the one that will have Fedora on it, but in fact both drives need to be selected, otherwise the installation will not create a proper dual-boot system, but just one that boots back into Win7.. Not sure if this is a bug, but certainly non-intuitive..

On this type of dual-boot system, it is necessary to make sure that ‘device mapper’ ( under ‘RAID’ etc..) is selected, when compiling the kernel from scratch. This is needed for GRUB2 to find the ‘other’ Windows 7 drive, and add it to the boot menu.

An additional problem I encountered was that the onboard Intel Gigabit Ethernet (e1000e driver) was extremely sluggish, despite ethtool etc, showing all was OK.. It became so bad that even a local file transfer on the same LAN segment would stall.. I had found complaints about this NIC going back some years, but nothing very specific. I had a spare USB3 Gigabit Ethernet adapter, and had already added a USB3 i/o card, and found that throughput was far better with this combination. The system includes a single spare PCI slot (as well as the PCIe ones..) and I will probably find a RealTek chipset PCI Gigabit Ethernet card to put in this.

The only other slight frustration was with the serial console attachment, which worked OK with the GRUB menu, but showed nothing after this. I had already checked all the config options – including the usual kernel boot command string, and the RS323 null-modem cable was known/good. I then decided to check the dmesg output when booting, and found that there were two 16550A serial connections showing up:

..................
[ 0.750508] 00:03: ttyS2 at I/O 0x3e8 (irq = 4, base_baud = 115200) is a 16550A
[ 0.777549] 0000:00:03.3: ttyS0 at I/O 0x1248 (irq = 17, base_baud = 115200) is a 16550A
..................

From this, the usual ttyS0 connection was not the external RS232 port, which was in fact ttyS2. The corresponding part of the kernel command line was changed to ‘..console=ttyS2,115200n8…’ and then everything worked correctly.

Robert Gadsdon.  February 18, 2019.