Skip to content

isp2200, virtio: what mainline linux needs on the octane - #789

Merged
techomancer merged 5 commits into
techomancer:mainfrom
steigr:linux/isp2200-logins-and-virtio-access-platform
Nov 20, 2026
Merged

techomancer merged 5 commits into
techomancer:mainfrom
steigr:linux/isp2200-logins-and-virtio-access-platform

Conversation

@steigr

@steigr steigr commented Nov 19, 2026

Copy link
Copy Markdown
Contributor

Two device fixes that let Linux 7.3 (Gentoo mips3 n64, ip30_defconfig plus a few drivers) use the Octane's Fibre Channel and virtio devices.

ISP2200 port logins. Linux's qla2xxx logs in to each loop port with Login Loop Port (0x0074). From its discovery state machine it also sends mailbox commands inside a request-ring entry (type 0x39: login, logout and port database). The model refused both, so the loop came up and no disk appeared. It now answers 0x0074 for a present port and serves those commands in a ring entry, answered in the response ring; anything else sent that way is an invalid command.

virtio ACCESS_PLATFORM. Without VIRTIO_F_ACCESS_PLATFORM, Linux bypasses the DMA API and writes physical addresses into the queue registers. Behind the Bridge, a physical address below 1 GB is PCI-local, so the device never saw its rings and virtio_console spun on its first control message. The PCI transport now offers the bit (a driver that refuses it is still accepted) and Linux hands over direct-window addresses. This applies on the Tezro too.

Also two clanker notes:

  • what Linux needs on the Octane: kernel config, two kernel-side Bridge driver issues, what to expect;
  • the Octane PROM crashing when it loads an 11 MB ELF from the volume header (boot over TFTP instead).

The Linux serial console also needs virtio-serial off Bridge slot 4. That change is in a separate PR, #788.

Tests (new): port_login (loop tests), offers_access_platform.

Guest check: Linux 7.3.4 on two CPUs on the Octane with the PCI cage, ISP2200 (two loop disks), NVMe, virtio-serial and virtio-9p. 18 passing runs, each with disk readback (system, FC, NVMe) by SHA-256, a 20k flood ping and an OpenSSL build.

Tested on macOS arm64 at the tip: cargo test --workspace --no-fail-fast --release --features lightning,gpujit,jitv2 and cargo build --release --bins with the same features. No failures beyond those upstream main shows on the same host (the five port-forward tests, ip35::tests::the_jitv2_compile_setting_reaches_every_product_and_drop_joins_the_pool, iris-gui switching_from_a_populated_machine_leaves_a_valid_config). kona::back::tiles::tests::a_tiled_run_is_bit_exact_with_the_inline_one was skipped (hangs under load on main).

🤖 Generated with Claude Code

steigr and others added 4 commits November 19, 2026 22:29
Linux's qla2xxx logs in to each port on the loop with Login Loop Port
(0x0074) and, from its discovery state machine, with mailbox commands
carried in a request-ring entry (type 0x39: login, logout and the port
database for ADISC). The model refused both, so the loop came up and no
disk ever appeared. Answer 0x0074 for a present port, and serve those four
commands in a ring entry, answered in an entry of the same type in the
response ring; anything else sent that way is an invalid command.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Without VIRTIO_F_ACCESS_PLATFORM Linux bypasses the DMA API and writes
physical addresses into the queue registers. Behind the Octane's Bridge a
physical address below 1 GB is PCI-local space, so the device never read
its rings and virtio_console spun for ever on its first control message.
The PCI transport now offers the bit (a driver that does not accept it is
still taken) and Linux hands over the Bridge's direct-window addresses.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
@steigr
steigr force-pushed the linux/isp2200-logins-and-virtio-access-platform branch from 80e99b2 to e31d490 Compare November 19, 2026 21:31
@techomancer
techomancer merged commit f085576 into techomancer:main Nov 20, 2026
7 of 8 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants