Loose ends: repin illumos-ddi, ENOMEM for failed DMA binds, page-size constant in rdrmnull #4

Merged

forge merged 6 commits from loose-ends-20260928 into main 2026-09-29 19:11:14 +00:00

Conversation 1 Commits 6 Files changed 10 +379 -40

forge commented 2026-09-28 18:37:29 +00:00

Contributor

Copy link

Loose ends after the R0, R1 and DMA-drill merges.

1. Pin

illumos-ddi moves from b24c307 to eb0150cadc, the head of illumos-ddi loose-ends-20260928 (illumos/illumos-ddi#3). I chose the branch head so this PR's CI can build now. Once illumos-ddi#3 is merged, move the pin to the merge commit on illumos-ddi main. The branch commits stay reachable from that merge commit, so the pin keeps resolving either way.

API adaptation: illumos-ddi now requires the OwnNode for minor nodes. Device::attach creates card<N> and renderD<128+N> from the node before the GEM state takes it. Opens wait until attach returns, and a later failure (GEM setup, registry) drops the nodes again. No other illumos-ddi API change affects this repository: property updates and LdiIdent::from_dip are not used here.

2. Errno for failed DMA binds

SysMem::bind mapped DDI_DMA_NOMAPPING to EFAULT. That is how every failed root-node bind on the patched kernel reached userland from CREATE_DUMB; EFAULT in a create ioctl reads as a bad user pointer. The handle allocation went through illumos-ddi's generic From<DmaError> for Errno, which also answers EFAULT for NoMapping.

The new drm::dmaerr module maps DMA failures during object creation the way Linux's DRM reports a failed object allocation:

Failure	errno
NORESOURCES, TOOBIG, NOMAPPING, PARTIAL_MAP (never requested)	ENOMEM
BADATTR, BADLIMITS, TOOSMALL, zero length, out of range	EINVAL
DmaError::NotOwnNode (illumos-ddi guard)	ENXIO
INUSE, LOCKED, STALE, DDI_FAILURE, unknown codes (not possible for a fresh handle)	EIO

The handle allocation, the bind and the drill's raw allocation all use it. The match is exhaustive over DmaError, so a new variant must be decided on. Five host tests cover the table, including NOMAPPING -> ENOMEM.

The errno was also recorded on a kernel. rdrmnull's self drill takes an optional dma-max-transfer property. tools/test/rdrmnull-dma-self-tiny.conf sets it to one page, and tools/test/drm_dma_errno_test.c found:

• A one-page dumb buffer binds.
• A four-page one fails with errno 12 (ENOMEM), through rootnex's DDI_DMA_TOOBIG, confirmed by rootnex's "coding error detected" warning.
• The gem kstat is back at zero afterwards.

DDI_DMA_NOMAPPING itself cannot be produced on a stock kernel with these flags: rootnex returns it only on the IOMMU path, or when trimming with DDI_DMA_PARTIAL, which the core never requests. That mapping is therefore covered by the host test only.

3. Advisory from the R1 review

rdrmnull now uses drm::offsets::PAGE_SIZE instead of the literal 4096 for the drill's cookie limit and for the movable-object reservation.

Verification

Local checks, targeting x86_64-unknown-illumos:

• check, clippy -D warnings with and without rdrmnull/panic-drill, fmt --check and doc -D warnings pass.
• cargo test -p drm -p drm-uapi: 47 + 1 passed. The 2 dead-code warnings in the test build are already on main.
• uapi-check: 0 differences.

Kernel runs used a fresh golden guest on Hephaistos (illumos cda5b9a49f), with rdrmnull built without panic-drill against illumos-ddi 9ebd16add0. The only difference from eb0150cadc is illumos-ddi docs.

Run	R0	R1	32-bit probes	FAIL
run-r1-test.sh, dma-mode="none"	105 PASS	153 PASS	5 and 4 PASS	none
run-r1-test.sh, dma-mode="self" under the root nexus	105 PASS	153 PASS	5 and 4 PASS	only the known /etc/path_to_inst residue
drm_dma_errno_test, dma-max-transfer=4096				4 PASS, 0 FAIL

There was no panic in any run, and rem_drv unloaded the module each time.

Loose ends after the R0, R1 and DMA-drill merges. ## 1. Pin illumos-ddi moves from `b24c307` to `eb0150cadc`, the head of illumos-ddi `loose-ends-20260928` (illumos/illumos-ddi#3). I chose the branch head so this PR's CI can build now. **Once illumos-ddi#3 is merged, move the pin to the merge commit on illumos-ddi `main`.** The branch commits stay reachable from that merge commit, so the pin keeps resolving either way. API adaptation: illumos-ddi now requires the `OwnNode` for minor nodes. `Device::attach` creates `card<N>` and `renderD<128+N>` from the node before the GEM state takes it. Opens wait until attach returns, and a later failure (GEM setup, registry) drops the nodes again. No other illumos-ddi API change affects this repository: property updates and `LdiIdent::from_dip` are not used here. ## 2. Errno for failed DMA binds `SysMem::bind` mapped `DDI_DMA_NOMAPPING` to `EFAULT`. That is how every failed root-node bind on the patched kernel reached userland from `CREATE_DUMB`; `EFAULT` in a create ioctl reads as a bad user pointer. The handle allocation went through illumos-ddi's generic `From<DmaError> for Errno`, which also answers `EFAULT` for `NoMapping`. The new `drm::dmaerr` module maps DMA failures during object creation the way Linux's DRM reports a failed object allocation: | Failure | errno | | --- | --- | | `NORESOURCES`, `TOOBIG`, `NOMAPPING`, `PARTIAL_MAP` (never requested) | `ENOMEM` | | `BADATTR`, `BADLIMITS`, `TOOSMALL`, zero length, out of range | `EINVAL` | | `DmaError::NotOwnNode` (illumos-ddi guard) | `ENXIO` | | `INUSE`, `LOCKED`, `STALE`, `DDI_FAILURE`, unknown codes (not possible for a fresh handle) | `EIO` | The handle allocation, the bind and the drill's raw allocation all use it. The match is exhaustive over `DmaError`, so a new variant must be decided on. Five host tests cover the table, including NOMAPPING -> ENOMEM. The errno was also recorded on a kernel. `rdrmnull`'s self drill takes an optional `dma-max-transfer` property. `tools/test/rdrmnull-dma-self-tiny.conf` sets it to one page, and `tools/test/drm_dma_errno_test.c` found: - A one-page dumb buffer binds. - A four-page one fails with errno 12 (`ENOMEM`), through rootnex's `DDI_DMA_TOOBIG`, confirmed by rootnex's "coding error detected" warning. - The `gem` kstat is back at zero afterwards. `DDI_DMA_NOMAPPING` itself cannot be produced on a stock kernel with these flags: rootnex returns it only on the IOMMU path, or when trimming with `DDI_DMA_PARTIAL`, which the core never requests. That mapping is therefore covered by the host test only. ## 3. Advisory from the R1 review `rdrmnull` now uses `drm::offsets::PAGE_SIZE` instead of the literal `4096` for the drill's cookie limit and for the movable-object reservation. ## Verification Local checks, targeting `x86_64-unknown-illumos`: - `check`, `clippy -D warnings` with and without `rdrmnull/panic-drill`, `fmt --check` and `doc -D warnings` pass. - `cargo test -p drm -p drm-uapi`: 47 + 1 passed. The 2 dead-code warnings in the test build are already on `main`. - `uapi-check`: 0 differences. Kernel runs used a fresh golden guest on Hephaistos (illumos `cda5b9a49f`), with `rdrmnull` built without `panic-drill` against illumos-ddi `9ebd16add0`. The only difference from `eb0150cadc` is illumos-ddi docs. | Run | R0 | R1 | 32-bit probes | FAIL | | --- | --- | --- | --- | --- | | `run-r1-test.sh`, `dma-mode="none"` | 105 PASS | 153 PASS | 5 and 4 PASS | none | | `run-r1-test.sh`, `dma-mode="self"` under the root nexus | 105 PASS | 153 PASS | 5 and 4 PASS | only the known `/etc/path_to_inst` residue | | `drm_dma_errno_test`, `dma-max-transfer=4096` | | | | 4 PASS, 0 FAIL | There was no panic in any run, and `rem_drv` unloaded the module each time.

forge added 3 commits 2026-09-28 18:37:30 +00:00

Pin illumos-ddi at its loose-ends branch and create minor nodes from OwnNode ... a6b65711f1

illumos-ddi now takes the driver's OwnNode for minor nodes, property
updates and LDI identifiers. Device::attach creates the card and render
nodes from the node before the GEM state takes it; opens wait until
attach returns, and a later failure drops the nodes again.

The pin is the head of illumos-ddi's loose-ends-20260928 branch
(eb0150cadc); move it to the merge commit on illumos-ddi main once that
branch is merged.

Report failed DMA binds of new objects as ENOMEM, not EFAULT ... 931552866e

On a patched kernel the root-node drill saw every failed bind reach
userland as EFAULT from CREATE_DUMB: SysMem::bind mapped
DDI_DMA_NOMAPPING to EFAULT, which in a create ioctl reads as a bad user
pointer. drm::dmaerr now maps DMA failures during object creation the
way Linux's DRM reports a failed object allocation: ENOMEM for a
shortage or no mapping (NORESOURCES, TOOBIG, NOMAPPING, PARTIAL_MAP),
EINVAL for attributes that can never be satisfied, ENXIO for the
own-node guard, EIO for faults a fresh handle cannot produce. The handle
allocation, the bind and the drill's raw allocation all use it; host
tests cover the table.

rdrmnull's self drill takes an optional dma-max-transfer property;
tools/test/rdrmnull-dma-self-tiny.conf sets it to one page and
drm_dma_errno_test.c checks that a four-page dumb buffer then fails with
ENOMEM.

Use the core's page size instead of a literal 4096 in rdrmnull ... bef21aa92e

The drill's cookie limit and the movable objects' reservation now use
drm::offsets::PAGE_SIZE, as the R1 review advised.

forge added 3 commits 2026-09-29 19:04:13 +00:00

Decode every DDI_DMA_* code with illumos-ddi's DmaError::check ... 56a2f7366c

The local decoder knew six of the ten failure codes, so a raw BADLIMITS or
TOOSMALL became EIO although the module documentation promises EINVAL.
illumos-ddi now exports its complete decoder; use it through code_errno and
test all ten codes against the documented errno.

Scope dma-max-transfer to the self drill and check the gem kstat ... 2ca6065c46

The property is read, applied and logged only for dma-mode="self", and a
bad value reports the actual lookup error. The errno test now checks that a
failed creation leaves the gem kstat unchanged, and the tiny conf names the
rootnex warning that the refusal logs.

Pin illumos-ddi at 449bfe73e9, which exports DmaError::check 450b7850b9

forge commented 2026-09-29 19:10:54 +00:00

Author

Contributor

Copy link

CI note: the kmod-illumos job at 450b785 never ran. Solstice runs 11 and 12 failed in "Set up job" with solstice-ci: failed to enqueue job(s) to message broker (a broker outage; the hosted job passed on this head in run 11). Equivalent check done by hand: cargo +nightly-2026-04-16 build --locked -p rdrmnull --release for the kmod target, then on oibldofgate (illumos-cda5b9a49f) ld -ztype=kmod -z allextract succeeded and elfdump -d shows SUNW_KMOD; archive sha256 0776c4b6… identical on both sides.

CI note: the kmod-illumos job at `450b785` never ran. Solstice runs 11 and 12 failed in "Set up job" with `solstice-ci: failed to enqueue job(s) to message broker` (a broker outage; the hosted job passed on this head in run 11). Equivalent check done by hand: `cargo +nightly-2026-04-16 build --locked -p rdrmnull --release` for the kmod target, then on oibldofgate (illumos-cda5b9a49f) `ld -ztype=kmod -z allextract` succeeded and `elfdump -d` shows `SUNW_KMOD`; archive sha256 `0776c4b6…` identical on both sides.

forge merged commit 931c290942 into main 2026-09-29 19:11:14 +00:00

forge referenced this pull request from a commit 2026-09-29 19:11:16 +00:00

Merge pull request #4 from loose-ends-20260928: ENOMEM for failed DMA binds, full DMA code decoding, page-size constant

Sign in to join this conversation.

Reviewers

No reviewers

Labels

Clear labels

No items

No labels

Milestone

Clear milestone

No items

No milestone

Projects

Clear projects

No items

No project

Assignees

Clear assignees

No assignees

1 participant

Notifications

Subscribe

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference

illumos/illumos-drm!4

No description provided.