userlib: Introduce `Arch` trait by jamesmunns · Pull Request #2705 · oxidecomputer/hubris

jamesmunns · 2026-09-21T14:20:42Z

This is the third PR in a series to be able to potentially run the kernal and tasks on a 64-bit host for testing purposes. This is currently based on #2704 (and #2703).

This PR introduces an Arch trait for userlib, similar to the Arch trait introduced for the kernel in #2704.

This follows the pattern we've started using for BSP impls lately: We define a trait, each BSP (or Arch in this case) implements it, and we expose the One Selected Impl using cfgs as a commonly named type (here: Current) which the rest of the crate uses.

Like #2703, this one might be worth review commit-by-commit, as the total diff is a bit large, but the individual commits are each pretty straightforward.

jamesmunns · 2026-09-21T15:27:03Z

For reviewers, I've added some "big picture" notes in #2706 on how this PR fits into the bigger picture.

jamesmunns · 2026-10-09T16:03:03Z

About to do the renames, but I repeated the analysis I did in #2704 (comment), but between james/arch-trait and james/userlib-trait. This time the only diff was in the caboose, no difference in the kernel or any of the task regions.

james@fool:~/oxide ls -al *.bin
-rw-r--r--  1 james  staff  748288  9 Oct 16:26 arch-trait.bin
-rw-r--r--  1 james  staff  748288  9 Oct 16:24 master.bin
-rw-r--r--  1 james  staff  748288  9 Oct 17:58 userlib-trait.bin
james@fool:~/oxide xxd ./arch-trait.bin > ./arch-trait.hex
james@fool:~/oxide xxd ./userlib-trait.bin > ./userlib-trait.hex
james@fool:~/oxide diff -u ./arch-trait.hex ./userlib-trait.hex
--- ./arch-trait.hex	2026-10-09 17:59:59
+++ ./userlib-trait.hex	2026-10-09 18:00:12
@@ -46751,9 +46751,9 @@
 000b69e0: ffff ffff ffff ffff ffff ffff ffff ffff  ................
 000b69f0: ffff ffff ffff ffff ffff ffff ffff ffff  ................
 000b6a00: 5e00 b0ca 4749 5443 2800 0000 b8d8 0c0c  ^...GITC(.......
-000b6a10: 3732 3338 3235 6262 3230 6365 3830 3537  723825bb20ce8057
-000b6a20: 3035 3261 3137 3432 3034 6135 6132 6534  052a174204a5a2e4
-000b6a30: 3365 3933 3565 6238 b302 eb84 424f 5244  3e935eb8....BORD
+000b6a10: 3833 6336 6338 3334 3264 3563 3964 3065  83c6c8342d5c9d0e
+000b6a20: 3239 3538 6432 3830 3131 3630 3862 6635  2958d28011608bf5
+000b6a30: 3961 3565 6237 3534 5ea8 00f2 424f 5244  9a5eb754^...BORD
 000b6a40: 0700 0000 e67f 1fca 636f 736d 6f2d 6200  ........cosmo-b.
 000b6a50: dc94 eabb 4e41 4d45 0700 0000 fac4 843c  ....NAME.......<
 000b6a60: 636f 736d 6f2d 6200 dc94 eabb ffff ffff  cosmo-b.........
james@fool:~/oxide

edit, and post-renames:

james@fool:~/oxide xxd ./userlib-trait-2.bin > ./userlib-trait-2.hex
james@fool:~/oxide diff -u ./arch-trait.hex ./userlib-trait-2.hex                 
--- ./arch-trait.hex	2026-10-09 17:59:59
+++ ./userlib-trait-2.hex	2026-10-09 18:28:00
@@ -46751,9 +46751,9 @@
 000b69e0: ffff ffff ffff ffff ffff ffff ffff ffff  ................
 000b69f0: ffff ffff ffff ffff ffff ffff ffff ffff  ................
 000b6a00: 5e00 b0ca 4749 5443 2800 0000 b8d8 0c0c  ^...GITC(.......
-000b6a10: 3732 3338 3235 6262 3230 6365 3830 3537  723825bb20ce8057
-000b6a20: 3035 3261 3137 3432 3034 6135 6132 6534  052a174204a5a2e4
-000b6a30: 3365 3933 3565 6238 b302 eb84 424f 5244  3e935eb8....BORD
+000b6a10: 3864 3164 3836 3035 3638 3364 3838 6265  8d1d8605683d88be
+000b6a20: 3331 6663 6432 3564 3062 6264 3563 3939  31fcd25d0bbd5c99
+000b6a30: 3964 6335 3363 6538 35b8 864b 424f 5244  9dc53ce85..KBORD
 000b6a40: 0700 0000 e67f 1fca 636f 736d 6f2d 6200  ........cosmo-b.
 000b6a50: dc94 eabb 4e41 4d45 0700 0000 fac4 843c  ....NAME.......<
 000b6a60: 636f 736d 6f2d 6200 dc94 eabb ffff ffff  cosmo-b.........

jamesmunns · 2026-10-09T19:57:35Z

One more check post-rebase:

james@fool:~/oxide/hubris git:(james/userlib-trait) cp target/cosmo-b/dist/default/final.bin ../userlib-trait-3.bin
james@fool:~/oxide/hubris git:(james/userlib-trait) cd ..
james@fool:~/oxide xxd ./userlib-trait-3.bin > ./userlib-trait-3.hex
james@fool:~/oxide diff -u ./arch-trait.hex ./userlib-trait-3.hex                 
--- ./arch-trait.hex	2026-10-09 17:59:59
+++ ./userlib-trait-3.hex	2026-10-09 21:56:34
@@ -46751,9 +46751,9 @@
 000b69e0: ffff ffff ffff ffff ffff ffff ffff ffff  ................
 000b69f0: ffff ffff ffff ffff ffff ffff ffff ffff  ................
 000b6a00: 5e00 b0ca 4749 5443 2800 0000 b8d8 0c0c  ^...GITC(.......
-000b6a10: 3732 3338 3235 6262 3230 6365 3830 3537  723825bb20ce8057
-000b6a20: 3035 3261 3137 3432 3034 6135 6132 6534  052a174204a5a2e4
-000b6a30: 3365 3933 3565 6238 b302 eb84 424f 5244  3e935eb8....BORD
+000b6a10: 3131 6236 6163 6230 6433 3966 6264 3035  11b6acb0d39fbd05
+000b6a20: 3264 3666 6335 3239 6139 3434 3939 6633  2d6fc529a94499f3
+000b6a30: 3965 3463 3539 6362 d491 3ee1 424f 5244  9e4c59cb..>.BORD
 000b6a40: 0700 0000 e67f 1fca 636f 736d 6f2d 6200  ........cosmo-b.
 000b6a50: dc94 eabb 4e41 4d45 0700 0000 fac4 843c  ....NAME.......<
 000b6a60: 636f 736d 6f2d 6200 dc94 eabb ffff ffff  cosmo-b.........

hawkw · 2026-10-12T18:22:41Z

+    cfg_if::cfg_if! {
+        if #[cfg(armv6m)] {


part of me now wonders if it might make sense to just have separate Arch impls for v6m and v7m/v8m, and select between them based on cfgs, so that we don't have to duplicate the cfg_if! in every single stub...but, on the other hand, this way we can just copy the previous code more or less unchanged, so it may not be worth it, I dunno...

Oh, that's actually a pretty good idea. I might want to do that in a follow-up PR, just to keep the diff reasonable? But I think overall it's much more of the "vibe", moving towards something like:

- arch/ - arm_m/ - mod.rs (common parts) - armv6m.rs - armv7m.rs - armv8m.rs

Yeah, I think doing it in a follow up PR is entirely reasonable, just figured it was worth mentioning. AFAICT it's just the syscall stubs that currently differ here, and
v7m and v8m use the same assembly. I'd kind of prefer to have one version of the code for v[78]m, rather than duplicating it, until we actually need to add code that differs between those versions... I'm always a bit leery about introducing opportunities for identical code to drift apart unless the drift is on purpose.

hawkw · 2026-10-12T18:26:33Z

+    let msg = unsafe { pw.buf.get_unchecked(..pw.pos) };
+
+    // Pass it to kernel.
+    ArmM::panic(msg)


most of the code here seems to not actually be arch-specific...i wonder if there's some way to have the panic handler not be defined in the arch specific module and just call into the architecture-specific panic code via the ArchImpl renaming import in lib.rs?

So, yes and no. I think the reason this ended up like this is mostly that on std/native-hosted we may not want a custom panic handler at all, so it made more sense to make this at least no_std+cortex-m specific.

That being said, if we end up supporting rv32 or something else in the future, I think we could definitely refactor this out.

Mm, yeah, that's fair. I guess seeing this kinda makes me question why we even have a separate function on the Arch trait that is distinct from the rest of the panic handling machinery, but I suppose that allows you to have code that calls the panic syscall directly without going through the actual panic handler (though I don't know if that exists or if it's even something we would want). I think it's fine to leave this as-is today and factor out the non-arch-specific parts if and when it becomes necessary to add another architecture.

Co-authored-by: Eliza Weisman <eliza@elizas.website>

hawkw

Okay, this seems good to me, though perhaps you'll also want to wait for an ack from Laura?

jamesmunns · 2026-10-13T08:28:19Z

Talked to Laura, she's fine with this going through and took a pass over this earlier. Thank you both for the reviews!

jamesmunns requested review from cbiffle, hawkw and labbott September 21, 2026 14:20

jamesmunns mentioned this pull request Sep 21, 2026

Tracking Issue: Running Tasks on Host/Native #2706

Open

4 tasks

labbott reviewed Oct 5, 2026

View reviewed changes

Comment thread sys/userlib/src/arch/mod.rs Outdated

Comment thread sys/userlib/src/arch/arm_m.rs

Comment thread sys/userlib/src/arch/mod.rs Outdated

jamesmunns force-pushed the james/arch-trait branch from ec52c70 to a7136fd Compare October 9, 2026 14:02

jamesmunns requested a review from flihp as a code owner October 9, 2026 14:02

jamesmunns force-pushed the james/userlib-trait branch from 6ba5f69 to 83c6c83 Compare October 9, 2026 15:55

jamesmunns force-pushed the james/userlib-trait branch from 8d1d860 to 55ce3b1 Compare October 9, 2026 17:14

Base automatically changed from james/arch-trait to master October 9, 2026 18:13

jamesmunns added 7 commits October 9, 2026 21:54

Move userlib stub functions into arch/thumb.rs

324ae8f

Introduce (but don't implement) userlib arch trait

d9d23a0

Implement top level sys_ functions in terms of the Arch functions

1be9ce5

Rename to be consistent with the kernel

5c1841d

Current -> ArchImpl

5d9feb7

Thumb -> ArmM

9b7eecf

Change userlib arch detection to match kernel

11b6acb

jamesmunns force-pushed the james/userlib-trait branch from 55ce3b1 to 11b6acb Compare October 9, 2026 19:56

jamesmunns added 2 commits October 12, 2026 17:11

Improve arch trait docs, link to reference

418761f

Fix punctuation, add an impl not for ArmM.

b2c7201

hawkw reviewed Oct 12, 2026

View reviewed changes

Update sys/userlib/src/arch/arm_m.rs

c3868c3

Co-authored-by: Eliza Weisman <eliza@elizas.website>

hawkw approved these changes Oct 12, 2026

View reviewed changes

jamesmunns merged commit b9378fc into master Oct 13, 2026
198 checks passed

jamesmunns deleted the james/userlib-trait branch October 13, 2026 08:28

jamesmunns mentioned this pull request Oct 14, 2026

host-native: Initial support for host-native tasks #2736

Open

Conversation

jamesmunns commented Sep 21, 2026

Uh oh!

jamesmunns commented Sep 21, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jamesmunns commented Oct 9, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jamesmunns commented Oct 9, 2026

Uh oh!

Uh oh!

hawkw Oct 12, 2026

Choose a reason for hiding this comment

Uh oh!

jamesmunns Oct 12, 2026

Choose a reason for hiding this comment

Uh oh!

hawkw Oct 12, 2026

Choose a reason for hiding this comment

Uh oh!

hawkw Oct 12, 2026

Choose a reason for hiding this comment

Uh oh!

jamesmunns Oct 12, 2026

Choose a reason for hiding this comment

Uh oh!

hawkw Oct 12, 2026

Choose a reason for hiding this comment

Uh oh!

hawkw left a comment

Choose a reason for hiding this comment

Uh oh!

jamesmunns commented Oct 13, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

jamesmunns commented Oct 9, 2026 •

edited

Loading