vmem: Add stronger tlb invalidation barrier by syntactically · Pull Request #1876 · hyperlight-dev/hyperlight

syntactically · 2026-10-02T10:52:16Z

No description provided.

Copilot

Copilot review overview

🟡 Changes recommended

Unaligned ranges can leave stale permissions cached, and the new architecture-specific behavior lacks tests.

Review effort: Balanced
Findings: 1 · 1 · 2

Open (4)

Require or normalize page-aligned invalidation bounds · New
Add cross-architecture integration tests for invalidation · New
Add SAFETY comment for EL1 maintenance assembly · New
Add SAFETY comment for privileged assembly block · New

What changed in this PR

Adds a guest paging barrier for invalidating cached translations after permission downgrades.

Changes:

Exposes downgrade_in_place.
Uses invlpg on amd64.
Uses DSB, TLBI, and ISB on AArch64.

File	Description
`src/hyperlight_guest_bin/src/paging.rs`	Exposes the barrier API.
`src/hyperlight_guest_bin/src/arch/amd64/paging.rs`	Adds amd64 invalidation.
`src/hyperlight_guest_bin/src/arch/aarch64/paging.rs`	Adds AArch64 invalidation.

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

Signed-off-by: Lucy Menon <168595099+syntactically@users.noreply.github.com>

hyperlight-gh-bot · 2026-10-02T20:46:30Z

Benchmark Results

Measured commit: 7d244de79713
Baseline commit: 43cf3539bd10

kvm / amd (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`772.31 ns` (➖ 1.00x faster)
`vec_bytes`	`585.00 ns` (➖ 1.00x slower)
		`373.19 µs` (➖ 1.00x faster)

payload_allocation

	`slot_pool_segmented`
`262144`	`531.27 ns` (➖ 1.03x slower)
`65536`	`141.69 ns` (➖ 1.02x faster)

sandboxes

	`create_initialized_and_drop`
`medium`	`79.13 ms` (➖ 1.01x faster)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`7.76 ns` (➖ 1.00x faster)	`7.81 ns` (➖ 1.01x slower)	`7.73 ns` (➖ 1.00x slower)

snapshot_files

	`load_snapshot_unverified`
`small`	`92.64 µs` (➖ 1.00x faster)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`7.26 µs` (➖ 1.00x faster)	`8.00 µs` (➖ 1.10x slower)
`65536`	`2.07 µs` (➖ 1.01x slower)	`1.93 µs` (➖ 1.00x slower)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`6.22 µs` (➖ 1.02x faster)	`6.22 µs` (➖ 1.03x faster)
`8192`	`1.07 µs` (➖ 1.02x slower)	`1.08 µs` (➖ 1.02x slower)
`262144`		`26.97 µs` (➖ 1.01x faster)

kvm / intel (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`708.67 ns` (➖ 1.06x faster)
`vec_bytes`	`503.77 ns` (➖ 1.08x faster)
		`653.10 µs` (➖ 1.06x faster)

payload_allocation

	`slot_pool_segmented`
`262144`	`510.32 ns` (➖ 1.12x faster)
`65536`	`136.27 ns` (➖ 1.13x faster)

sandboxes

	`create_initialized_and_drop`
`medium`	`73.73 ms` (➖ 1.08x faster)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`7.28 ns` (➖ 1.04x faster)	`6.92 ns` (➖ 1.09x faster)	`7.11 ns` (➖ 1.08x faster)

snapshot_files

	`load_snapshot_unverified`
`small`	`44.45 µs` (➖ 1.10x faster)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`7.54 µs` (➖ 1.13x faster)	`7.59 µs` (➖ 1.13x faster)
`65536`	`2.12 µs` (➖ 1.13x faster)	`2.12 µs` (➖ 1.12x faster)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`7.13 µs` (➖ 1.10x faster)	`7.13 µs` (➖ 1.12x faster)
`8192`	`745.26 ns` (➖ 1.12x faster)	`727.91 ns` (➖ 1.08x faster)
`262144`		`29.59 µs` (➖ 1.13x faster)

mshv3 / amd (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`953.48 ns` (➖ 1.00x slower)
`vec_bytes`	`704.25 ns` (➖ 1.01x faster)
		`283.74 µs` (➖ 1.26x faster)

payload_allocation

	`slot_pool_segmented`
`262144`	`739.96 ns` (➖ 1.01x slower)
`65536`	`193.49 ns` (➖ 1.02x faster)

sandboxes

	`create_initialized_and_drop`
`medium`	`57.00 ms` (➖ 1.02x slower)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`10.59 ns` (➖ 1.04x slower)	`9.94 ns` (➖ 1.03x slower)	`9.69 ns` (➖ 1.04x faster)

snapshot_files

	`load_snapshot_unverified`
`small`	`86.00 µs` (➖ 1.02x slower)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`10.16 µs` (➖ 1.01x slower)	`8.82 µs` (➖ 1.01x faster)
`65536`	`2.26 µs` (➖ 1.00x slower)	`2.48 µs` (➖ 1.01x faster)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`9.18 µs` (➖ 1.11x slower)	`8.11 µs` (➖ 1.01x slower)
`8192`	`1.31 µs` (➖ 1.03x slower)	`1.30 µs` (➖ 1.03x faster)
`262144`		`35.69 µs` (➖ 1.02x faster)

mshv3 / intel (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`941.95 ns` (➖ 1.01x slower)
`vec_bytes`	`655.65 ns` (➖ 1.01x faster)
		`765.30 µs` (➖ 1.14x slower)

payload_allocation

	`slot_pool_segmented`
`262144`	`629.73 ns` (➖ 1.02x faster)
`65536`	`165.23 ns` (➖ 1.00x slower)

sandboxes

	`create_initialized_and_drop`
`medium`	`61.52 ms` (➖ 1.01x faster)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`8.76 ns` (➖ 1.00x faster)	`8.75 ns` (➖ 1.00x faster)	`10.00 ns` (➖ 1.05x slower)

snapshot_files

	`load_snapshot_unverified`
`small`	`43.85 µs` (➖ 1.00x faster)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`7.77 µs` (➖ 1.00x slower)	`7.74 µs` (➖ 1.04x faster)
`65536`	`2.22 µs` (➖ 1.02x slower)	`2.26 µs` (➖ 1.01x faster)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`7.50 µs` (➖ 1.00x faster)	`7.57 µs` (➖ 1.00x slower)
`8192`	`875.03 ns` (➖ 1.02x slower)	`877.32 ns` (➖ 1.01x faster)
`262144`		`35.25 µs` (➖ 1.08x faster)

hyperv-ws2025 / amd (Windows) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`1.17 µs` (➖ 1.04x slower)
`vec_bytes`	`787.36 ns` (➖ 1.21x faster)
		`2.26 ms` (➖ 1.11x faster)

payload_allocation

	`slot_pool_segmented`
`262144`	`790.83 ns` (➖ 1.02x faster)
`65536`	`234.16 ns` (➖ 1.02x slower)

sandboxes

	`create_initialized_and_drop`
`medium`	`88.23 ms` (➖ 1.25x faster)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`9.94 ns` (➖ 1.39x faster)	`10.50 ns` (➖ 1.03x faster)	`10.12 ns` (➖ 1.04x faster)

snapshot_files

	`load_snapshot_unverified`
`small`	`802.40 µs` (➖ 1.01x faster)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`9.23 µs` (➖ 1.06x slower)	`9.16 µs` (➖ 1.02x faster)
`65536`	`2.41 µs` (➖ 1.05x faster)	`2.40 µs` (➖ 1.13x faster)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`9.22 µs` (➖ 1.06x slower)	`9.10 µs` (➖ 1.05x slower)
`8192`	`1.29 µs` (➖ 1.11x faster)	`1.31 µs` (➖ 1.07x faster)
`262144`		`46.09 µs` (➖ 1.27x slower)

hyperv-ws2025 / intel (Windows) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`1.22 µs` (➖ 1.01x slower)
`vec_bytes`	`819.82 ns` (➖ 1.05x slower)
		`3.13 ms` (➖ 1.00x slower)

payload_allocation

	`slot_pool_segmented`
`262144`	`734.12 ns` (➖ 1.00x faster)
`65536`	`210.43 ns` (➖ 1.01x faster)

sandboxes

	`create_initialized_and_drop`
`medium`	`108.38 ms` (➖ 1.11x faster)

slot_pool

	`alloc_dealloc_1500`	`alloc_dealloc_4096`	`alloc_dealloc_128`
	`10.03 ns` (➖ 1.00x slower)	`10.12 ns` (➖ 1.00x slower)	`10.59 ns` (➖ 1.00x faster)

snapshot_files

	`load_snapshot_unverified`
`small`	`614.64 µs` (➖ 1.02x slower)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`262144`	`7.75 µs` (➖ 1.02x slower)	`7.69 µs` (➖ 1.01x faster)
`65536`	`2.33 µs` (➖ 1.03x slower)	`2.31 µs` (➖ 1.01x faster)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`8.50 µs` (➖ 1.02x slower)	`8.41 µs` (➖ 1.00x slower)
`8192`	`1.14 µs` (➖ 1.11x faster)	`1.23 µs` (➖ 1.10x slower)
`262144`		`50.18 µs` (➖ 1.17x slower)

_{Reported by cargo ci bench-report --candidate run:37059659974 --baseline run:36945564806 --config-file bench_report.toml.}

syntactically added this pull request to stack #1875 October 2, 2026 10:52

syntactically requested review from andreiltd, danbugs, dblnz, devigned, jprendes, jsturtevant, ludfjig, simongdavies and squillace as code owners October 2, 2026 10:52

This comment has been minimized.

Sign in to view

syntactically added kind/enhancement For PRs adding features, improving functionality, docs, tests, etc. ready-for-review PR is ready for (re-)review labels Oct 2, 2026

Copilot AI balanced review requested due to automatic review settings October 2, 2026 12:29

syntactically force-pushed the lm/extra-tlbi branch from 9717eee to e18ba90 Compare October 2, 2026 12:29

Copilot started reviewing on behalf of syntactically October 2, 2026 12:33 View session

Copilot AI reviewed Oct 2, 2026

View reviewed changes

Comment thread src/hyperlight_guest_bin/src/paging.rs

Comment thread src/hyperlight_guest_bin/src/paging.rs

Comment thread src/hyperlight_guest_bin/src/arch/aarch64/paging.rs

Comment thread src/hyperlight_guest_bin/src/arch/amd64/paging.rs

This comment has been minimized.

Sign in to view

Base automatically changed from lm/modify-mappings to main October 2, 2026 15:45

syntactically force-pushed the lm/extra-tlbi branch from e18ba90 to 3423312 Compare October 2, 2026 15:48

This comment has been minimized.

Sign in to view

jprendes approved these changes Oct 2, 2026

View reviewed changes

ludfjig approved these changes Oct 2, 2026

View reviewed changes

github-actions Bot removed the ready-for-review PR is ready for (re-)review label Oct 2, 2026

vmem: Add stronger tlb invalidation barrier

7d244de

Signed-off-by: Lucy Menon <168595099+syntactically@users.noreply.github.com>

syntactically force-pushed the lm/extra-tlbi branch from 3423312 to 7d244de Compare October 2, 2026 20:17

syntactically merged commit 7d244de into main Oct 3, 2026
60 checks passed

syntactically deleted the lm/extra-tlbi branch October 3, 2026 00:24

Conversation

syntactically commented Oct 2, 2026

Uh oh!

This comment has been minimized.

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

This comment has been minimized.

hyperlight-gh-bot Bot commented Oct 2, 2026

Benchmark Results

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Benchmark Results

function_call_codec

payload_allocation

sandboxes

slot_pool

snapshot_files

virtq_readonly

virtq_readwrite

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants