PCI on Alpha issues and DS20L usability

[John Klos <john%ziaspace.com@localhost>]

After my previous adventures:

http://mail-index.netbsd.org/port-alpha/2018/08/30/msg000893.html

I've finally at a point where I ALMOST have a happy, stable, 
self-contained system that can boot and reboot on its own.

The timecounter patch still needs to be added manually:
http://mail-index.netbsd.org/port-alpha/2017/11/02/msg000879.html

For -current, MSGBUFSIZE needs to be doubled or so to see a full dmesg.


The things that don't work for my API CS20 (AlphaServer DS20L):


IDE in DMA mode for any extended amount of time seems to kill filesystems 
and/or make disks unbootable. Also, when IDE is used with DMA, rebooting 
NetBSD fails - the system hangs - but shutting down to the SRM prompt, 
then running "init" will allow an uninterrupted boot without power cycle.


arcmsr* initializes, but I can't use any volumes. I only get 1 sector 
volumes:
http://mail-index.netbsd.org/port-alpha/2018/11/02/msg000900.html


mfi* on a PCIe to PCI adapter is seen, but dmesg shows 0MB RAM and 0 
volumes after a few thousand of "stray 6600 irq 36" messages:
http://mail-index.netbsd.org/port-alpha/2018/11/10/msg000901.html


nvme* on a PCIe to PCI adapter panics after:
[   6.4580944] extent `tsp_sgmap' (0x0 - 0x7fffff), flags = 0x2
[   6.4580944]      0x1000 - 0x2fff
[   6.4590709] extent_free: start 0x0, end 0x1fff
[   6.4600475] panic: extent_free: region not found


viaide* (VIA 6421A) gives a machine check while the kernel is booting:
pci_mem_find: expected type mem, found i/o
viaide0 at pci1 dev 4 function 0: VIA Technologies VT6421 Serial ATA RAID Controller (rev. 0x50)
viaide0: using dec 6600 irq 36 for native-PCI interrupt
System Machine Check (660): Rev 0x1, Code 0x202, Flags 0x0


satalink* (Silicon Image 3512) works, but not on cold boot. If I boot the 
kernel from onboard IDE, I get:
satalink0 at pci1 dev 4 function 0: Silicon Image SATALink 3512 (rev. 0x01)
satalink0: using dec 6600 irq 36 for native-PCI interrupt
atabus2 at satalink0 channel 0
atabus3 at satalink0 channel 1
...
wd1 at atabus2 drive 0
wd1: <Samsung SSD 750 EVO 500GB>
wd1: 465 GB, 969021 cyl, 16 head, 63 sec, 512 bytes/sect x 976773168 sectors
wd1: no disk label
wd1: no disk label

disklabel /dev/wd1 shows a blank disklabel. However, if I netboot and run:
dd if=/dev/wd1c of=/dev/null bs=1m count=100

then disklabel magically shows the proper disklabel and a reboot allows 
this disk to be used as the root filesystem!


Here are the things that work:


A USB 3 card on a PCIe to PCI adapter works perfectly. I've used a USB 3 
SSD as the root filesystem for weeks at a time with lots of I/O and 
compiling and have had zero problems.

Have had no problems with PCI ethernet adapters.

Onboard SCSI (SYM 53C1000, esiop*) seems to work fine, but large disks, 
such as a 2 TB disk, cause the SRM to report a test error.

IDE without DMA (flags 0x0002) on aceride seems to work fine. It also 
allows for a reboot without SRM init. I may've had this backwards in an 
earlier post.


On my AlphaServer DS25, the PCIe to PCI adapter doesn't work. Nothing is 
seen. I have outputs of a boot on both the CS20 (DS20L) and the DS25 of a 
kernel with PCI_CONFIG_DUMP, but I haven't had time to go through them.


So... Aside from USB 3, does anyone know of a good, consistent, usable PCI 
SATA or PCI RAID card that'll definitely work with Alpha? Has anyone tried 
mvsata*, or twa*, or twe*?

Does anyone have any thoughts about what can be done to fix satalink* so 
it works on first boot?

If not, I may just get a PCIe USB 3 card with an internal USB port and 
connect the SSD internally that way to get a completely self-contained, 
self-booting system. The kernel currently boots from an IDE-flash module.

Almost there...

John Klos