Nx bit support + minor fixes by DixelU · Pull Request #1484 · copy/v86

DixelU · 2026-01-11T14:43:33Z

That's like the second attempt at cracking this thing 🐈, and there's still some things to work on left.

The good part:

Implemented the NX bit itself.
- And the most minimal test in NASM possible.
Faced the issue with RSVD Page faults being unimplemented (which prevented the boot up of linux distributions and windows 7) - implemented though I did not test it though properly yet.
Did fix some funky stuff IDE was suggesting in the NASM tests run script along the support for NX test itself ... and god knows what else (I did pull AI for that one, I'll later look into reverting most of its changes though)
Tests - all seemed to complete with return code 0, and I was able to load a small build of windows 7 (from CD rom, later bout that) with NX bit enabled:

The problems left (and noticed)

Standard windows 7 installation did not boot up being stuck at
```
16:06:52+762 [ACPI] ACPI raise irq
16:06:52+762 [CPU ] No route: level interrupt and remote IRR still set
```
repeating endlessly. Did waste an entire evening on that one, without a thing out of it. Only a cut down version did boot up properly. Not sure why.
Whilst comparing how systems boot on that branch with currently released version of v86 - I did notice that in release the system hangs seemingly randomly inside net2k implementation (might need a separate issue for that) - cannot reproduce rn, but I'll create a separate issue once I face it again.
The mess in test scripts.

…ecific to nx test ones

I have no damn clue why original v86 works fine and enabling nx bit actually uncovers this issue 👀

YD711YT · 2026-01-16T00:44:58Z

Will windows 8 work?

DixelU · 2026-01-16T05:55:42Z

Will windows 8 work?

I'll check it out this weekend

DimaThenekov · 2026-01-17T20:26:38Z

Wow, this is great! I think Windows 7+ bootloaders require something else besides the NX bit. I couldn't start Windows 8 and 10 😢...

DimaThenekov · 2026-01-17T20:45:37Z

In particular, I get a #GP exception

DixelU · 2026-03-19T15:05:52Z

Alr, I'm halfway back from the dead, hope to look into it again soon 🕊️

DixelU · 2026-05-17T20:01:04Z

An unrelated note, having this popping out from time to time: unimplemented ATA command 0xB0: ABORT [ST=0x50 ER=0x0 SC=0x1 LL=0x1 LM=0x4F LH=0xC2 FE=0xD8]

Maybe i'll look into it later
Btw successfully booted up normal Windows 7 distro
Testing out windows 8.1 rn

DixelU · 2026-05-17T21:17:35Z

Okay, 8.1 seems to boot fine. I'll check other OS later then :D

DixelU · 2026-06-03T07:40:14Z

Booting linux seems to show some signs of PAE being either broken in this exact pull request or generally ...
Really hope it's something i am doing wrong there 🤔

DixelU · 2026-06-03T08:07:51Z

Welp, i count that as a success then 💁‍♂️

Aaand, i guess i'll need to look at what parts of the CI this PR fails 🤔

- CR4 fix: covers enabling PAE on a system that already has paging active — load_pdpte fires only if CR0.PG=1 - CR3 fix: covers writing CR3 before paging is enabled — load_pdpte fires only if CR0.PG=1

DixelU · 2026-06-08T10:18:08Z

Damn it, from windows i can't properly reset the mode for rust-lld-wrapper

Revert (filter changes with mode reset) The revert did not go well

DixelU · 2026-06-16T06:32:50Z

Yep, now it should be fine - squashed the last 3 funky commits

DixelU · 2026-08-10T09:42:35Z

@copy Can you run CI job and check whether the PR is generally fine in terms of the code-style 👀

…-support-v2

copy

Thanks! This looks good overall. I have left some comments.

The nasmtest doesn't seem to fit in that testsuite very well. It could work better if added to kvm-unit-tests. tests/jit-paging would also make sense if Linux allows you to configure and catch NX faults.

Could you upload one or more disk images (small if possible) that exercise this feature? E.g. OSes that didn't boot before and now do.

copy · 2026-09-17T18:13:34Z

+            this.push_irq();
+            break;
+        }
+


Could you send this as a separate PR?

copy · 2026-09-17T18:22:55Z

        },

+        0x80000008 => {
+            eax = 32; // physical address width


Should probably also set virtual address bits

copy · 2026-09-17T18:24:46Z

    // cmpss xmm, xmm/m32
    let destination = read_xmm_f32(r);
-    let source: f32 = f32::from_bits(i32::cast_unsigned(source));
+    let source: f32 = f32::from_bits(source as u32);


What's the purpose of this change? I try to avoid as because it tends to do have unintended side-effects if you got the types wrong.

copy · 2026-09-17T18:28:57Z

+        {
+            const array = new Array(8 + 1 + 8 + 16 + 32 + (STACK_TOP - BSS >> 2) + 3).fill(0);
+            array[8] = 0x1000;
+            return { img_name, fixture: { array, exception: "PF" } };


It's unclear to me why this special case for the nx image is necessary.

copy · 2026-09-17T18:29:54Z

+    mov eax, 0x80000001
+    cpuid
+    test edx, 1 << 20
+    jz .no_nx


I don't think this NX check is necessary. It's gonna be enabled forever once this is merged.

copy · 2026-09-17T21:28:29Z

    clear_tlb_code(page);
    tlb_data[page as usize] = 0;
-    let error_code = (user as i32) << 2 | (write as i32) << 1 | present as i32;
+    let error_code = (is_instruction_fetch as i32) << 4


This bit should only be set under certain conditions: https://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-system-programming-manual-325384.pdf#page=137

copy · 2026-09-17T22:24:56Z


 pub unsafe fn safe_read_f32(addr: i32) -> OrPageFault<f32> {
-    Ok(f32::from_bits(i32::cast_unsigned(safe_read32s(addr)?)))
+    Ok(f32::from_bits(safe_read32s(addr)? as u32))


See my other comment regarding casts

copy · 2026-09-17T22:41:21Z

    for_writing: bool,
    user: bool,
    jit: bool,
+    is_instruction_fetch: bool,


Make the argument order consistent between this and translate_address.

copy · 2026-09-17T22:44:58Z

+pub unsafe fn translate_address_fetch(address: i32) -> OrPageFault<u32> {
+    translate_address(address, false, *cpl == 3, false, true, true)
 }
 pub unsafe fn translate_address_read_jit(address: i32) -> OrPageFault<u32> {


This name is confusing, because historically and in the non-jit case, it means read without instruction access. The naming should be translate_address_{read,write,fetch}{,_jit}

copy · 2026-09-17T22:51:22Z

+                        if side_effects {
+                            trigger_pagefault(addr, true, for_writing, user, jit, true, false);
+                        }
+                        return Err(());


Check if this is correct: From my understanding, if a lower table is not present, we should fault with not-present error code. Also check how the reserved bits error code should be handled.

- KVM test with and w/o JIT - Separate out the ATA_CMD_READ_LOG_EXT implementation patch - Consistent argument ordering (historically had a little issue with amount of arguments passed, as it did change dramatically in a different local branch) - Reverted cast changes proposed by IDE - Integrate NX into codegen, several renaming changes - Drop test name changes introduced before tests refactoring - And tons of other things ; - ;

DixelU · 2026-09-24T08:26:25Z

It's kind of funny how to prove that NX bit indeed is supported by booting previously unsupported OS-es
You basically have to add like 3 other patches on top 🐈

DixelU · 2026-09-24T13:35:21Z

Make it 4 patches;
And Windows 10 2303 - https://archive.org/details/tiny-10_202301
Can be installed and booted from within v86;
I'll be polishing the code later and probably put final disk image somewhere on the cloud

Several fixes, including:

DixelU · 2026-10-01T12:33:04Z

W10 Image (though i am not sure it's okay to distribute it like that -): https://drive.google.com/file/d/1wQ4vIXJPmZTuiHGXedxGWNVdJTdp5Cda/view?usp=sharing

copy · 2026-10-04T07:35:32Z

Squashed and merged in #1647

Thanks a lot @DixelU, nice work!

DixelU added 10 commits December 29, 2025 11:43

Foundational stuff

3acc83e

rust+asm only

e95b34c

And test stuff - am not sure these changes are really needed aside sp…

64d0508

…ecific to nx test ones

...

2a5edcf

...

f51b4da

Update cpu.rs

54eaa7f

RSVD Faults

347be5a

I have no damn clue why original v86 works fine and enabling nx bit actually uncovers this issue 👀

Update instructions_0f.rs

c1b70e4

clear tlb only upon efer change

793ce63

add efer reset

32bf805

DixelU added 3 commits March 19, 2026 20:18

Merge branch 'master' into nx-support-v2

c459d9b

Build fix with updated dev container + possible PAE misuse

5826d86

Funny mistake

cc21f68

DixelU added 3 commits May 18, 2026 01:05

Merge branch 'master' into nx-support-v2

52693a3

Revery nasm run

6c7866a

READ_LOG_EXT for windows 8.1 boot up

2e56c4d

Fixes for linux with PAE support enabled?

ce9f183

- CR4 fix: covers enabling PAE on a system that already has paging active — load_pdpte fires only if CR0.PG=1 - CR3 fix: covers writing CR3 before paging is enabled — load_pdpte fires only if CR0.PG=1

DixelU marked this pull request as ready for review June 3, 2026 08:19

Minor style issues

fe55a80

copy reviewed Jun 8, 2026

View reviewed changes

Comment thread tools/rust-lld-wrapper Outdated

Fix unintnended changes (squash)

c21844d

Revert (filter changes with mode reset) The revert did not go well

DixelU force-pushed the nx-support-v2 branch from c47e7ee to c21844d Compare June 16, 2026 06:31

DixelU added 5 commits September 4, 2026 17:04

Merge remote-tracking branch 'upstream/master' into nx-support-v2

9e5be1c

Seems fine now;

6ecba73

Merge branch 'master' into nx-support-v2

a199564

Merge branch 'nx-support-v2' of https://github.com/DixelU/v86 into nx…

d019f7e

…-support-v2

restore skip parameters, reset nasmtest cache upon source change

13d2544

copy reviewed Sep 17, 2026

View reviewed changes

DixelU mentioned this pull request Sep 24, 2026

Implement ATA_CMD_READ_LOG_EXT #1639

Merged

DixelU added 4 commits September 30, 2026 12:04

rustfmt rerun

1607852

The NX branch adds TLB_NO_EXEC (bit 6, 0x40)

2f6a9ac

Merge pull request #2 from DixelU/nx-support-post-review-changes

c1c9fe3

Several fixes, including:

Merge branch 'master' into nx-support-v2

ff57958

copy closed this Oct 4, 2026

Conversation

DixelU commented Jan 11, 2026

The good part:

The problems left (and noticed)

Uh oh!

YD711YT commented Jan 16, 2026

Uh oh!

DixelU commented Jan 16, 2026

Uh oh!

DimaThenekov commented Jan 17, 2026

Uh oh!

DimaThenekov commented Jan 17, 2026

Uh oh!

DixelU commented Mar 19, 2026

Uh oh!

DixelU commented May 17, 2026

Uh oh!

DixelU commented May 17, 2026

Uh oh!

DixelU commented Jun 3, 2026

Uh oh!

DixelU commented Jun 3, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

DixelU commented Jun 8, 2026

Uh oh!

DixelU commented Jun 16, 2026

Uh oh!

DixelU commented Aug 10, 2026

Uh oh!

copy 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!

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!

DixelU commented Sep 24, 2026

Uh oh!

DixelU commented Sep 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

DixelU commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

copy commented Oct 4, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

DixelU commented Jun 3, 2026 •

edited

Loading

DixelU commented Sep 24, 2026 •

edited

Loading

DixelU commented Oct 1, 2026 •

edited

Loading