feat(aarch64): add GDB debugging for Windows ARM64 WHP by cshung · Pull Request #1907 · hyperlight-dev/hyperlight

cshung · 2026-10-08T17:36:47Z

Summary

Adds GDB stub support for guests running on the ARM64 WHP backend.

Changes

Debug core

Split the x86_64 debug code out of hyperlight_vm/x86_64.rs into a shared hyperlight_vm/debug.rs.
Rename gdb/x86_64_target.rs to gdb/target.rs and add per-architecture register handling in gdb/arch/.
Give DebuggableVm default methods so backends without hardware breakpoints or single step report Unsupported.

ARM64 WHP

Implement DebuggableVm for the aarch64 WHP VM, using the batched register helpers.
Flush the host instruction cache after guest memory writes so software breakpoints take effect.

Docs

Document ARM64 debugging in docs/how-to-debug-a-hyperlight-guest.md and CHANGELOG.md.

Verification

On Windows ARM64 (WHP):

just build, just clippy, just guests, just test and just fmt-check pass.

Copilot

🟡 Changes recommended

Breakpoint state can survive crash or cancellation recovery, and the hidden cache-sync mapping conflicts with the public mapping contract.

4 open findings

Clean up debugger breakpoints after crash detach · New
Cancel ARM64 software steps for all cancellation sources · New
Reserve hidden GPA 0x1000 in the mapping contract · New
Restore corrupted mathematical symbols in comments · New

What changed in this PR

Adds Windows ARM64 WHP guest debugging and generalizes the GDB implementation across architectures.

Changes:

Adds ARM64 registers, software breakpoints, stepping, and cache synchronization.
Extracts shared debugger logic from x86_64 code.
Updates tests, documentation, and platform configuration.

File	Description
`CHANGELOG.md`	Records ARM64 GDB support.
`docs/how-to-debug-a-hyperlight-guest.md`	Documents architecture-specific capabilities.
`src/hyperlight_host/build.rs`	Enables GDB on Windows ARM64.
`src/hyperlight_host/examples/guest-debugging/main.rs`	Adds ARM64 debugging tests.
`src/hyperlight_host/src/hypervisor/gdb/arch.rs`	Dispatches architecture-specific logic.
`src/hyperlight_host/src/hypervisor/gdb/arch/aarch64.rs`	Implements ARM64 breakpoint and step decoding.
`src/hyperlight_host/src/hypervisor/gdb/arch/x86_64.rs`	Houses extracted x86_64 logic.
`src/hyperlight_host/src/hypervisor/gdb/event_loop.rs`	Maps ARM64 stop reasons.
`src/hyperlight_host/src/hypervisor/gdb/mod.rs`	Extends shared debugging interfaces.
`src/hyperlight_host/src/hypervisor/gdb/target.rs`	Adds architecture-specific GDB registers.
`src/hyperlight_host/src/hypervisor/hyperlight_vm/aarch64.rs`	Connects GDB to ARM64 VMs.
`src/hyperlight_host/src/hypervisor/hyperlight_vm/debug.rs`	Centralizes breakpoint and stepping state.
`src/hyperlight_host/src/hypervisor/hyperlight_vm/mod.rs`	Handles ARM64 debug exits.
`src/hyperlight_host/src/hypervisor/hyperlight_vm/x86_64.rs`	Removes extracted debugger logic.
`src/hyperlight_host/src/hypervisor/virtual_machine/mod.rs`	Extends VM exits and errors.
`src/hyperlight_host/src/hypervisor/virtual_machine/whp/aarch64.rs`	Implements WHP debugging and cache synchronization.
`src/hyperlight_host/src/sandbox/initialized.rs`	Guards snapshots with active breakpoints.

🧠 Review effort: Balanced

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

syntactically · 2026-10-12T15:50:39Z

@cshung I guess you are using software breakpoints here because WHPX aarch64 does not yet support setting MDCR_EL2.TDE. I think this is very divergent from the other implementations and especially the emulation of software step looks very fragile in its current state.

I wonder if we can make things a little bit more congruent with the other implementations by using the same hardware debug approach as other targets, but on WHP set MDSCR_EL1.KDE and have little assembly stubs catch this case and make hvcs (or, even make an hvc on every exception entry, when in gdb mode)? Admittedly this will not work very well if VBAR_EL1 or virtual memory configuration for the exception handlers are messed up, but in that case I think given the current guest architecture it is extremely unlikely that guest code would continue to execute and produce any useful results.

Adds GDB stub support for guests running on the ARM64 WHP backend. * Split the x86_64 debug code out of `hyperlight_vm/x86_64.rs` into a shared `hyperlight_vm/debug.rs`. * Rename `gdb/x86_64_target.rs` to `gdb/target.rs` and add per-architecture register handling in `gdb/arch/`. * Give `DebuggableVm` default methods so backends without hardware breakpoints or single step report `Unsupported`. * Implement `DebuggableVm` for the aarch64 WHP VM, using the batched register helpers. * Flush the host instruction cache after guest memory writes so software breakpoints take effect. * Document ARM64 debugging in `docs/how-to-debug-a-hyperlight-guest.md` and `CHANGELOG.md`. On Windows ARM64 (WHP): * `just build`, `just clippy`, `just guests`, `just test` and `just fmt-check` pass. --------- Signed-off-by: cshung <3410332+cshung@users.noreply.github.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

## Changes * Run the full disable path when the debugger detaches after a crash, so software breakpoints and pending steps are cleaned up. * Cancel pending software steps for any cancellation source, not only debugger interrupts. * Reject `map_region` requests that overlap the hidden cache sync page at GPA 0x1000, and add a test. * Restore the `≡` and `×` characters in `hyperlight_vm/x86_64.rs` comments. ## Verification On Windows ARM64 (WHP): * `just clippy`, `just fmt-check` and the `hypervisor::` and `map_region` lib tests with the `gdb` feature pass. --------- Signed-off-by: cshung <3410332+cshung@users.noreply.github.com> Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Guest synchronous exception vectors forward debug stops through HVC. Hardware single-step supports branches, returns, and self-loops. Two hardware breakpoint slots and BRK software breakpoints are available. Interrupted register context and breakpoint configuration survive cache synchronization and vCPU reset. Asynchronous vectors retain their normal exception paths. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: cshung <3410332+cshung@users.noreply.github.com>

hyperlight-gh-bot · 2026-10-20T18:46:58Z

Benchmark Results

Measured commit: 698b85108794
Baseline commit: f0934723ebf0

kvm / amd (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`749.69 ns` (➖ 1.00x faster)
`vec_bytes`	`581.49 ns` (➖ 1.02x faster)
		`374.57 µs` (➖ 1.00x slower)

payload_allocation

	`slot_pool_segmented`
`65536`	`142.86 ns` (➖ 1.00x faster)
`262144`	`524.20 ns` (➖ 1.00x faster)

sandboxes

	`create_initialized_and_drop`
`medium`	`83.75 ms` (➖ 1.04x slower)

slot_pool

	`alloc_dealloc_4096`	`alloc_dealloc_128`	`alloc_dealloc_1500`
	`7.72 ns` (➖ 1.00x faster)	`8.66 ns` (➖ 1.12x slower)	`7.88 ns` (➖ 1.02x slower)

snapshot_files

	`load_snapshot_unverified`
`small`	`94.58 µs` (➖ 1.02x slower)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`2.05 µs` (➖ 1.04x slower)	`1.92 µs` (➖ 1.09x faster)
`262144`	`7.04 µs` (➖ 1.13x faster)	`7.54 µs` (➖ 1.07x slower)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`6.29 µs` (➖ 1.01x faster)	`6.22 µs` (➖ 1.10x faster)
`8192`	`1.06 µs` (➖ 1.07x faster)	`1.05 µs` (➖ 1.01x faster)
`262144`		`27.31 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`	`788.95 ns` (➖ 1.01x faster)
`vec_bytes`	`587.19 ns` (➖ 1.02x slower)
		`708.92 µs` (➖ 1.01x faster)

payload_allocation

	`slot_pool_segmented`
`65536`	`149.32 ns` (➖ 1.01x faster)
`262144`	`576.67 ns` (➖ 1.01x faster)

sandboxes

	`create_initialized_and_drop`
`medium`	`79.11 ms` (➖ 1.02x faster)

slot_pool

	`alloc_dealloc_4096`	`alloc_dealloc_128`	`alloc_dealloc_1500`
	`7.84 ns` (➖ 1.01x faster)	`7.71 ns` (➖ 1.01x faster)	`7.68 ns` (➖ 1.01x faster)

snapshot_files

	`load_snapshot_unverified`
`small`	`46.56 µs` (➖ 1.05x faster)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`2.33 µs` (➖ 1.06x slower)	`2.31 µs` (➖ 1.06x slower)
`262144`	`8.29 µs` (➖ 1.04x slower)	`8.34 µs` (➖ 1.07x slower)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`8.02 µs` (➖ 1.00x faster)	`8.10 µs` (➖ 1.02x slower)
`8192`	`949.69 ns` (➖ 1.12x slower)	`843.54 ns` (➖ 1.01x slower)
`262144`		`33.20 µs` (➖ 1.01x faster)

mshv3 / amd (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`919.09 ns` (➖ 1.01x slower)
`vec_bytes`	`700.80 ns` (➖ 1.01x faster)
		`318.51 µs` (➖ 1.00x slower)

payload_allocation

	`slot_pool_segmented`
`65536`	`187.48 ns` (➖ 1.01x slower)
`262144`	`684.49 ns` (➖ 1.02x slower)

sandboxes

	`create_initialized_and_drop`
`medium`	`65.08 ms` (➖ 1.07x slower)

slot_pool

	`alloc_dealloc_4096`	`alloc_dealloc_128`	`alloc_dealloc_1500`
	`9.91 ns` (➖ 1.00x faster)	`9.93 ns` (➖ 1.00x faster)	`9.95 ns` (➖ 1.00x slower)

snapshot_files

	`load_snapshot_unverified`
`small`	`86.07 µs` (➖ 1.00x slower)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`2.68 µs` (➖ 1.00x faster)	`2.47 µs` (➖ 1.00x slower)
`262144`	`9.15 µs` (➖ 1.02x faster)	`9.00 µs` (➖ 1.01x faster)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`8.04 µs` (➖ 1.03x faster)	`8.04 µs` (➖ 1.03x faster)
`8192`	`1.38 µs` (➖ 1.00x slower)	`1.36 µs` (➖ 1.02x slower)
`262144`		`34.80 µs` (➖ 1.03x faster)

mshv3 / intel (Linux) (➖ stable)

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`946.40 ns` (➖ 1.04x slower)
`vec_bytes`	`654.72 ns` (➖ 1.01x slower)
		`717.81 µs` (➖ 1.08x faster)

payload_allocation

	`slot_pool_segmented`
`65536`	`173.27 ns` (➖ 1.06x faster)
`262144`	`664.66 ns` (➖ 1.02x slower)

sandboxes

	`create_initialized_and_drop`
`medium`	`66.28 ms` (➖ 1.01x faster)

slot_pool

	`alloc_dealloc_4096`	`alloc_dealloc_128`	`alloc_dealloc_1500`
	`9.40 ns` (➖ 1.01x faster)	`9.33 ns` (➖ 1.00x faster)	`9.65 ns` (➖ 1.01x faster)

snapshot_files

	`load_snapshot_unverified`
`small`	`46.14 µs` (➖ 1.02x slower)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`2.28 µs` (➖ 1.01x slower)	`2.28 µs` (➖ 1.01x slower)
`262144`	`8.21 µs` (➖ 1.03x slower)	`7.93 µs` (➖ 1.00x slower)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`7.56 µs` (➖ 1.00x faster)	`7.51 µs` (➖ 1.03x faster)
`8192`	`921.88 ns` (➖ 1.02x faster)	`942.39 ns` (➖ 1.04x slower)
`262144`		`41.45 µs` (➖ 1.06x faster)

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

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`1.15 µs` (➖ 1.01x faster)
`vec_bytes`	`801.45 ns` (➖ 1.01x slower)
		`1.99 ms` (➖ 1.00x slower)

payload_allocation

	`slot_pool_segmented`
`65536`	`240.03 ns` (➖ 1.03x slower)
`262144`	`788.28 ns` (➖ 1.00x slower)

sandboxes

	`create_initialized_and_drop`
`medium`	`83.22 ms` (➖ 1.04x faster)

slot_pool

	`alloc_dealloc_4096`	`alloc_dealloc_128`	`alloc_dealloc_1500`
	`10.54 ns` (➖ 1.01x slower)	`10.16 ns` (➖ 1.01x slower)	`11.09 ns` (➖ 1.08x slower)

snapshot_files

	`load_snapshot_unverified`
`small`	`1.12 ms` (➖ 1.44x slower)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`2.35 µs` (➖ 1.03x faster)	`2.34 µs` (➖ 1.00x faster)
`262144`	`9.00 µs` (➖ 1.05x faster)	`9.16 µs` (➖ 1.00x slower)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`9.21 µs` (➖ 1.03x slower)	`8.91 µs` (➖ 1.05x faster)
`8192`	`1.31 µs` (➖ 1.07x faster)	`1.29 µs` (➖ 1.03x faster)
`262144`		`40.64 µs` (➖ 1.03x faster)

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

No benchmark improved or regressed.

Benchmark Results

function_call_codec

	`encode_control`	`decode_vec_bytes_copy`
`byte_chunks`	`1.24 µs` (➖ 1.04x slower)
`vec_bytes`	`784.27 ns` (➖ 1.01x slower)
		`2.98 ms` (➖ 1.05x faster)

payload_allocation

	`slot_pool_segmented`
`65536`	`208.55 ns` (➖ 1.02x faster)
`262144`	`740.62 ns` (➖ 1.08x faster)

sandboxes

	`create_initialized_and_drop`
`medium`	`103.25 ms` (➖ 1.13x faster)

slot_pool

	`alloc_dealloc_4096`	`alloc_dealloc_128`	`alloc_dealloc_1500`
	`9.99 ns` (➖ 1.00x slower)	`9.97 ns` (➖ 1.00x slower)	`9.97 ns` (➖ 1.00x slower)

snapshot_files

	`load_snapshot_unverified`
`small`	`539.28 µs` (➖ 1.09x faster)

virtq_readonly

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`2.34 µs` (➖ 1.01x slower)	`2.34 µs` (➖ 1.01x faster)
`262144`	`7.79 µs` (➖ 1.02x slower)	`7.79 µs` (➖ 1.00x slower)

virtq_readwrite

	`slot_pool_segmented_fragmented`	`slot_pool_segmented`
`65536`	`8.54 µs` (➖ 1.02x faster)	`8.54 µs` (➖ 1.04x faster)
`8192`	`1.12 µs` (➖ 1.12x faster)	`1.09 µs` (➖ 1.01x faster)
`262144`		`42.84 µs` (➖ 1.01x faster)

_{Reported by cargo ci bench-report --candidate run:39470905052 --config-file bench_report.toml.}

cshung added the kind/enhancement For PRs adding features, improving functionality, docs, tests, etc. label Oct 8, 2026

This comment has been minimized.

Sign in to view

cshung marked this pull request as ready for review October 8, 2026 19:33

Copilot AI balanced review requested due to automatic review settings October 8, 2026 19:33

cshung requested review from andreiltd, danbugs, dblnz, devigned, jprendes, jsturtevant, ludfjig, simongdavies, squillace and syntactically as code owners October 8, 2026 19:33

Copilot started reviewing on behalf of cshung October 8, 2026 19:37

View session

Copilot AI reviewed Oct 8, 2026

View reviewed changes

cshung force-pushed the cshung/whp-aarch64-gdb branch from 8c0c245 to 39ea4cb Compare October 9, 2026 16:54

This comment has been minimized.

Sign in to view

cshung force-pushed the cshung/whp-aarch64-gdb branch from 7238539 to 6c35092 Compare October 19, 2026 14:56

This comment has been minimized.

Sign in to view

cshung and others added 3 commits October 20, 2026 10:58

cshung force-pushed the cshung/whp-aarch64-gdb branch from dd86b49 to 698b851 Compare October 20, 2026 17:59

Conversation

cshung commented Oct 8, 2026

Summary

Changes

Debug core

ARM64 WHP

Docs

Verification

Uh oh!

This comment has been minimized.

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

This comment has been minimized.

syntactically commented Oct 12, 2026

Uh oh!

This comment has been minimized.

This comment has been minimized.

This comment has been minimized.

hyperlight-gh-bot Bot commented Oct 20, 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!

This branch has not been deployed

Reviewers

Assignees

Labels

Projects