On Sun, 17 Aug 2025 09:16:43 +0200,
Martin Pieuchot <[email protected]> wrote:
>
> > Reproducing it on octeon is quite simple: I'm building lang/gcc/15 from
> > ports. On very small amount of RAM.
>
> Well the machine is out of "low" pages and all I/O are blocked.
>
> I added some workarounds to ensure swap on a disk makes progress in such
> case. Unfortunately swapping on a file requires even more low pages and
> is completely broken in such situation.
>
> > So, I'll try, as the next step.
> >
> > > > Next attempt with swap on the local eMMC leads to somehow frozen device.
>
> Which driver attached to this eMMC? I saw only a USB disk in your
> dmesg.
>
> How looked your fstab in such case?
>
This is Ubiquiti Networks UniFi Security Gateway and full used fstab is:
1f20a4f764a59247.a / ffs rw,wxallowed 1 1
/swap.local none swap sw
#172.31.2.23:/volume1/octeon/swap none swap sw,nfsmntpt=/swap
#172.31.2.23:/volume1/octeon /mnt nfs rw,nodev,soft,intr,tcp,-x=2 0 0
#/mnt/swap none swap sw
172.31.2.23:/volume1/octeon/src /usr/src nfs rw,nodev,soft,intr,tcp,-x=2 0 0
172.31.2.23:/volume1/octeon/obj /usr/obj nfs rw,nodev,soft,intr,tcp,-x=2 0 0
172.31.2.23:/volume1/octeon/ports /usr/ports nfs rw,nodev,soft,intr,tcp,-x=2 0 0
the device has very small eMMC (4Gb) and here no USB slot.
On U-Boot this eMMC reads as:
USB: (port 0) scanning bus for devices... 1 USB Devices found
scanning bus for storage devices...
Device 0: Vendor: Prod.: UDinfo UF2 4GB Rev: PMAP
Type: Removable Hard Disk
Capacity: 3824.0 MB = 3.7 GB (7831552 x 512)
which turns into:
umass0 at uhub0 port 1 configuration 1 interface 0 " UDinfo UF2 4GB" rev
2.00/1.00 addr 2
umass0: using SCSI over Bulk-Only
scsibus0 at umass0: 2 targets, initiator 0
sd0 at scsibus0 targ 1 lun 0: <, UDinfo UF2 4GB, PMAP> removable
serial.13fe420086F821804937
sd0: 3824MB, 512 bytes/sector, 7831552 sectors
>
> I'm sorry there's no easy fix to this problem. I have been advocating
> moving away from flipping buffer and using bounce buffer instead in OOM
> situation.
>
> This kind of bugs are a result of the current design. I'm sorry but I
> don't believe they can be fixed.
>
Well, I not sure that is the right design here. Deadlock defently not good
solution. Not sure that panic is better.
Anyway, shall it be documented that running OpenBSD on device with less than
1Gb RAM turns into this, and here no way to use swap?
--
wbr, Kirill
