Re: NetBSD_10.0_BETA boot issues

[Robert Nestor <rnestor%mac.com@localhost>]

Martin,

Yes it is, but getting a dmesg during the failure seems to be difficult since I haven’t seen a failure when booting with serial console and most of the details of the failure scroll off the screen before the boot crashes.

However I may have stumbled across something that could be causing this.  The performance improvements between 9.2 and 10.0 are impressive and very noticeable, especially the boot up time.  The HP6200 has a fairly fast CPU and all the disks I’m seeing the boot problems on are 5,400 RPM disks.  Some are inexpensive 3.5” and some are 2.5” pulls from old laptops installed into 3.5” sleds.  All have identical creatures - 16-sector PIO, LBA48 addressing, PIO Mode 4, DMA Mode 2, Ultra DMA Mode 6, Write DMA FUA NCQ 32.  On a hunch I dug up a 7,200 RPM disk and installed 10.0 onto it.  That disk has never failed to boot and runs 10.0 just fine.  So far I’ve booted it up about a dozen times using both BIOS and UEFI and have yet to see any failure.  With the 5,400 disks I’d probably be lucky to see one successful boot in a dozen tries, so something is certainly different here!

My hunch is that with the performance improvements the code runs a tad bit faster (my guess is something like 10-20% faster) and on a fast enough CPU with slow ATA attached disks sometimes the code gets ahead of the disk in completing a disk request.  That might also explain why booting with a serial console never seems to fail as well, as outputting the console log at 9,600 Baud vs dumping it to the attached monitor slows the boot down just enough to succeed.  I’m going to try setting the speed on the serial port to the highest I can to see if that tickles the boot issue.

-bob

On Jan 5, 2023, at 1:35 AM, Martin Husemann <martin%duskware.de@localhost> wrote:

> On Wed, Jan 04, 2023 at 02:07:04PM -0600, Robert Nestor wrote:
>> Be happy to try and provide more info if someone has suggestions.
> 
> Is this the same machine as in https://gnats.netbsd.org/56737 ?
> As Rin asked there: we need the full dmesg output of the machine.
> 
> Martin