Refactor Vm trait #517

ludfjig · 2025-05-21T02:14:31Z

This PR introduces a minimal Vm trait. It's implemented by KvmVm, MshvVm and WhpVm. The goal is to reduce code duplication. The existing Hypervisor trait is renamed to HyperlightVm, and is now only implemented by HyperlightSandbox (and Inprocess, but not for long since we're getting rid of inproces mode). In addition, the Vm trait also reduces a bunch of code duplication related to GDB debugging.

About half of the lines of this PR is introducing common register structs used for all Vms. They Are CommonRegisters, CommonStandardRegisters, and CommonFpu, with tests. The rest should be pretty straight forward, and is mostly just plumbing.

Closes #465
Closes #312
Closes #503

There are some more cleanup to do (like cleaning up some windows surrogate process things, tests, no more need for a dyn HyperligtVM, etc.), but figured I'll do that in separate PRs as this one is quite big already.

…ll drivers. Remove duplicated GDB code. Signed-off-by: Ludvig Liljenberg <[email protected]>

danbugs

LGTM. Awesome work!

danbugs · 2025-05-21T21:08:35Z

src/hyperlight_host/src/hypervisor/hyperlight_vm.rs

+            Ok(())
+        }
+    }
+}


Maybe @dblnz wants to have a look at this?

This is copied from previous implementation

Given the scope of the changes I think we should have @dblnz review this

danbugs · 2025-05-21T21:25:55Z

src/hyperlight_host/src/hypervisor/hyperlight_vm.rs

+                Ok(HyperlightExit::MmioRead(addr)) => {
+                    #[cfg(crashdump)]
+                    crashdump::crashdump_to_tempfile(self)?;
+


Currently, on Windows, we have some very useful debug!s for different HyperlightExits that capture VCPU state. I think these were removed from whp.rs. Would be good to re-add them here, so we have consistent debug!s for all drivers.

Isn't the crashdump feature enough for this? I can add them back if you prefer though

I tend to agree with Dan, debug output might be used when crashdump feature is off, however we should probably not do this in release builds since registers might contain sensitive data

src/hyperlight_host/src/hypervisor/whp.rs

src/hyperlight_host/src/mem/layout.rs

danbugs · 2025-05-21T21:50:20Z

src/hyperlight_host/src/hypervisor/whp.rs

+unsafe impl Send for WhpVm {}
+unsafe impl Sync for WhpVm {}
+
+#[repr(C, align(16))]


Maybe add a comment here on why we do this :)

Signed-off-by: Ludvig Liljenberg <[email protected]>

simongdavies

I had to give up reviewing this, it is too big, please split this up into smaller sequential commits and a set of smaller PRs to make it reviewable. For example the changes to optimise /dev/kvm and /dev/mshv could easily be a seperate PR

simongdavies · 2025-05-22T13:06:53Z

src/hyperlight_host/src/hypervisor/gdb/arch.rs

        // Check page 19-4 Vol. 3B of Intel 64 and IA-32
        // Architectures Software Developer's Manual
-        if DR6_HW_BP_FLAGS_MASK & dr6 != 0 && hw_breakpoints.contains(&rip) {
+        if DR6_HW_BP_FLAGS_MASK & dr6 != 0 {


why did this logic change?

simongdavies · 2025-05-22T13:07:38Z

src/hyperlight_host/src/hypervisor/gdb/arch.rs


-    if BP_EX_ID == exception && sw_breakpoints.contains_key(&rip) {
-        return VcpuStopReason::SwBp;
+    if BP_EX_ID == exception {


why did this if change?

simongdavies · 2025-05-22T13:14:46Z

src/hyperlight_host/src/hypervisor/hyperlight_vm.rs

+            Ok(())
+        }
+    }
+}


Given the scope of the changes I think we should have @dblnz review this

simongdavies · 2025-05-22T13:17:20Z

src/hyperlight_host/src/hypervisor/hyperlight_vm.rs

+}
+
+#[derive(Debug)]
+pub(crate) struct HyperlightSandbox {


I am a bit confused as we now seem to have a new struct called HyperlightSandbox and we already have a Sandbox , think this needs a different name , and probably needs to be in its own module file

simongdavies · 2025-05-22T13:18:54Z

src/hyperlight_host/src/hypervisor/hyperlight_vm.rs

+    }
+}
+
+impl HyperlightVm for HyperlightSandbox {


#Why are we calling this HyperlightVm? what other Vms are we going to have? Should it be called Hypervisor?

simongdavies · 2025-05-22T13:50:39Z

src/hyperlight_host/src/hypervisor/mshv.rs

@@ -0,0 +1,470 @@
+/*


I think renaming these files is wrong, the old names were more descriptive, but, if we want to rename them then lets do it as a separate PR , it makes it even harder to review

dblnz · 2025-06-03T12:09:33Z

Do you plan on updating this after the hypervisor_handler change?
I have partially reviewed the previous draft of this #469 .
Let me know when to have a look on this

ludfjig · 2025-06-04T20:40:57Z

Do you plan on updating this after the hypervisor_handler change? I have partially reviewed the previous draft of this #469 . Let me know when to have a look on this

I won't merge yet, I'm very sorry for delaying your windows PR. Let's get your windows PR merged first if it's ready

syntactically · 2025-09-10T16:45:33Z

We still want to do this, but this is getting stale, so we will make another attempt. Desire to do this is tracked by #465

ludfjig requested review from danbugs, dblnz, devigned, jprendes, marosset, simongdavies and syntactically as code owners May 21, 2025 02:14

ludfjig added kind/refactor For PRs that restructure or remove code without adding new functionality. area/API Related to the API or public interface labels May 21, 2025

ludfjig force-pushed the vm_trait2 branch from e502597 to 10b11b9 Compare May 21, 2025 05:32

Refactor Vm trait. Introduce common representation of registers for a…

afd8964

…ll drivers. Remove duplicated GDB code. Signed-off-by: Ludvig Liljenberg <[email protected]>

ludfjig force-pushed the vm_trait2 branch 3 times, most recently from e697c0f to c18f0f1 Compare May 21, 2025 06:01

danbugs previously approved these changes May 21, 2025

View reviewed changes

ludfjig mentioned this pull request May 21, 2025

Remove hypervisor handler thread #521

Closed

ludfjig dismissed danbugs’s stale review via 37aed29 May 21, 2025 23:59

ludfjig force-pushed the vm_trait2 branch from c18f0f1 to 37aed29 Compare May 21, 2025 23:59

Fix alignment issues

ef44f60

Signed-off-by: Ludvig Liljenberg <[email protected]>

ludfjig force-pushed the vm_trait2 branch from 37aed29 to ef44f60 Compare May 22, 2025 00:00

simongdavies requested changes May 22, 2025

View reviewed changes

danbugs mentioned this pull request Jun 2, 2025

Remove hypervisor_handler thread #533

Merged

ludfjig marked this pull request as draft June 6, 2025 21:14

syntactically closed this Sep 10, 2025

ludfjig mentioned this pull request Oct 27, 2025

Reset more state when restoring snapshot #991

Draft

Refactor Vm trait #517

Refactor Vm trait #517

Uh oh!

Conversation

ludfjig commented May 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

danbugs left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

ludfjig May 21, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

simongdavies left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

dblnz commented Jun 3, 2025

Uh oh!

ludfjig commented Jun 4, 2025

Uh oh!

syntactically commented Sep 10, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

ludfjig commented May 21, 2025 •

edited

Loading

ludfjig May 21, 2025 •

edited

Loading