Repository navigation
arm64: T6050 at EL2: broken vGIC SEIS, WFxT treated as absent (K116s, KVM) - #1724
Merged
RyanTheTide merged 2 commits intoNov 13, 2026
Merged
RyanTheTide merged 2 commits into
RyanTheTide merged 2 commits into
Conversation
RyanTheTide
approved these changes
Nov 12, 2026
RyanTheTide
left a comment
RyanTheTide
left a comment
Contributor
There was a problem hiding this comment.
Both lists only gain the two T6050 parts, and the EL2 KVM results cover the SEIS entry. Looks good.
Contributor
|
@sithglan this no longer merges cleanly into integration/aurora (cpu_errata.c/cpufeature.c moved under it). Please rebase it and we'll merge it. |
ICH_VTR_EL2 on the K116s (MacBook Pro 16-inch, M5 Pro, firmware 27.2) reads 0x90580007 at EL2: eight list registers and SEIS set, on the boot CPU, part 0x064. The other Apple cores with this value are listed in apple_broken_seis_cpus so that KVM does not trust the locally generated SError the vGIC claims to support; without the entry SMP guests livelock. Add the two T6050 core parts. The register was read through the boot loader's proxy at EL2 before Linux ran. The K116s now runs Linux at EL2 on every CPU (padded Stage 1, 27.2) with this entry, which logs "Known broken GICv3 SEIS implementation", and KVM works: qemu -accel kvm -M virt,gic-version=3 -cpu host guests with 1, 2, 6 and 18 vCPUs boot and exit cleanly. Signed-off-by: Jens Axboe <axboe@kernel.dk>
has_wfxt() hides WFxT on the T8142 and T8152 cores, which clear architectural state across WFIT. Add the T6050 (M5 Pro and Max) core parts, 0x064 and 0x065. This is a precaution, not a measured fix: the T6050 was not measured for the WFIT state loss. Under the boot loader's EL1 monitor ID_AA64ISAR2_EL1.WFxT read 0. The K116s now runs Linux at EL2 on every CPU (padded Stage 1, 27.2), where the field reads 2, so ARM64_HAS_WFXT would be set without these entries. With them, a sha256 loop under schedutil on all 18 cores ran 67k times without a wrong result, and KVM guests run. The kernel was not run at EL2 without them. __delay() already skips WFIT on these cores by MIDR; the entries keep the capability clear for its other users as well. The HWCAP2_WFXT bit userspace sees comes from the ID register field, not from this capability, and is not changed. Signed-off-by: Jens Axboe <axboe@kernel.dk>
sithglan
force-pushed
the
thomas/t6050-el2-errata
branch
from
November 13, 2026 06:14
d751204 to
af0dcdf
Compare
Contributor
Author
|
@RyanTheTide rebased onto integration 17903f5, now af0dcdf. The only conflict was in |
RyanTheTide
merged commit Nov 13, 2026
2b7d391
into
aurora-silicon:integration/aurora
1 check passed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Base: integration/aurora (6ea76a5). Two commits for the M5 Pro and M5 Max cores (parts 0x064 and 0x065) when Linux runs at EL2.
apple_broken_seis_cpus, so KVM doesn't trust the locally generated SError the vGIC claims.has_wfxt()'s list, as T8142 and T8152 did. This is a precaution: the WFIT state loss itself was not measured on T6050.Tested by Jens on the K116s, running Linux at EL2 on all 18 CPUs from a padded Stage 1 (CTXR window) on the 27.2 stub. The kernel was integration/aurora 2b4523c plus his series:
qemu -accel kvm -M virt,gic-version=3 -cpu hostguests with 1, 2, 6 and 18 vCPUs boot and exit cleanly.Build: arm64, LLVM 22,
Image.gz modules dtbswith Jens's K116s config, in a test merge with the three other K116s branches sent today. Every commit builds on its own, and the test merge passes the host tests (tools/asahi/run-host-tests.py, 292 of 292).Authorship. Jens Axboe wrote both commits and runs them on his K116s (16-inch MacBook Pro, M5 Pro, OLED; Mac17,12; stub macOS 27.2 26B8125). They come from his K116s bundle of 2026-11-12. I cherry-picked them onto integration/aurora 6ea76a5. I also updated the paragraphs that still described the machine at EL1 ("KVM itself has not yet run on this machine", "runs at EL1 under the boot loader's monitor"), using the EL2 results from his status mail of 2026-11-12. The code is unchanged. The commits carry his Signed-off-by.
cc @Acelogic @DjDeveloperr @RyanTheTide @LiamRay10