Propolis panics with requests for VMs with small amounts of guest memory

[bnaecker]

I've been running some experiments using the new Oxide CLI, mostly to test behaviors in Nexus. At some point, I got lazy and started creating requests with small amounts of guest memory. If one requests a VM with 1MiB of memory, Propolis panics. Below is the server's log file:

bnaecker@feldspar : ~/omicron $ pfexec tail -f $(pfexec svcs -z oxz_propolis-server_1254f5f1-7ed6-449f-8ab3-b1ee3a006c6d -L propolis-server)
Mar 30 17:23:12.283 INFO Starting server...
Mar 30 17:23:12.454 INFO accepted connection, remote_addr: 192.168.1.145:62667, local_addr: 192.168.1.216:12400
Mar 30 17:23:12.454 INFO request completed, error_message_external: Internal Server Error, error_message_internal: Server not initialized (no instance), response_code: 500, uri: /instances/00000000-0000-0000-0000-000000000000, method: GET, req_id: a47f8763-7025-48f8-9322-72436d3d0462, remote_addr: 192.168.1.145:62667, local_addr: 192.168.1.216:12400
thread 'tokio-runtime-worker' panicked at 'assertion failed: lowmem >= MEM_BASE', propolis/src/hw/bhyve/rtc.rs:49:9
note: run with `RUST_BACKTRACE=1` environment variable to display a backtrace

It seems like Propolis should provide some minimal amount of verification of the requested guest memory sizes, and return a 400-level error where those expectations are violated. Currently, this returns a 500, with no further details. I have in fact never seen any error other than a 500 from Propolis server, which makes sense I think looking at how the client handles errors.

[smklein]

Followed-up on this a bit in the Hypervisor sync:

• The input to Propolis' HTTP server is already in MB
• Sled Agent is performing a byte -> MB conversion
• Most of the rest of usage in Omicron is in bytes

So, although Propolis should definitely consider imposing minimum sizes, the "byte granularity" responsibility is now up to Omicron to propagate.