On 26/10/2026 21:04, Michael Kelly wrote:
Having removed a duplicate call to devsw_attach:
[ 3.6100050] nvme0: interrupting at pausebreak
[ 3.6100050] nvme0: QEMU NVMe Ctrl, firmware 10.0.13, serial nvme-1
[ 3.6100050] ld at nvme0 nsid 1 not configured
[ 3.6100050] vendor 8086 product 293e (mixed mode multimedia, HD
Audio 1.0, revision 0x03) at pci0 dev 27 function 0 not configured
[ 3.6100050] vendor 8086 product 2918 (ISA bridge, revision 0x02) at
pci0 dev 31 function 0 not configured
[ 3.6100050] vendor 8086 product 2930 (SMBus serial bus, revision
0x02) at pci0 dev 31 function 3 not configured
[ 3.6100050] ld0 at nvme0 nsid 1
[ 3.6100050] ld0: 1024 MB, 1040 cyl, 32 head, 63 sec, 512 bytes/sect
x 2097152 sectors
I'm having a bit of a nightmare with this one. I had this all working at
lunchtime so following my normal development method I cloned my
reference virtual machine to create a clean environment for validating
the patch set. In the new virtual machine, attempts to open and read the
'ld0' device hang, using for example:
# od /dev/ld0 -N 512
Running rumpdisk.static from the new VM on the old one works correctly
suggesting the problem is related to the VM. I've finally spotted that
the 2 have different memory configurations after wasting lots of time on
dead ends. It works only when the machine memory is lower than 4G which
indicates to me that there is a problem with the DMA (again).
I'm not sure that it's worth issuing this patch yet if you can only run
it on small machines but I can do so if requested.
Anyway, the patch will be ready when it's ready, given that I don't know
how difficult this new issue will be to resolve.
Cheers,
Mike.
