Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Windows 8 BSOD #1

Open
nepx opened this issue Feb 8, 2020 · 2 comments
Open

Windows 8 BSOD #1

nepx opened this issue Feb 8, 2020 · 2 comments

Comments

@nepx
Copy link
Owner

nepx commented Feb 8, 2020

Windows 8 doesn't boot. It BSODs (or rather GSODs -- gray screen of death) right after the Windows logo disappears.

Screenshot from 2020-01-01 17-57-54

It's because of a Kernel Data Inpage Error, which, according to this source indicates that Windows 8 couldn't read from the pagefile.

I think it's a hardware issue, specifically pertaining to how hard disk information is read/written. It's not a IDE DMA issue -- I created a Windows 8 disk image that uses the READ/WRITE MULTIPLE command, and it still BSODs this way.

Multi-sector reads/writes to the disk image haven't been extensively tested. I think there's a subtle bug where cross-block writes don't work properly. The only operating system that uses them is OS/2 (which works fine).

It's possible that the BSOD is due to a cross-block bug. Windows XP had trouble booting in the Emscripten version because single sector reads at the end of the block were corrupting the next one. OS/2 sometimes has trouble booting on my Android phone because of multi-block reads.

More investigation is required.
@nepx
Copy link
Owner Author

nepx commented Mar 21, 2020

It definitely looks like a hard drive issue. During normal boot, VESA is switched off after a while, a cursor renders, and the Gray Screen of Death appears.

Setting MAX_MULTIPLE_SECTORS to zero, effectively disabling READ/WRITE MULTIPLE, keeps VESA on. The crash still happens, except that VESA isn't disabled.

Interestingly, despite forcing READ/WRITE MULTIPLE to fail, Windows 8 still uses the command.

The problem starts around here:

[I/O] writeb: port=0x01f6 data=0xa0
[IDE] Chose master drive on primary
[I/O] writeb: port=0x01f6 data=0xa0
[IDE] Chose master drive on primary
[I/O] writeb: port=0x01f1 data=0x00
[I/O] writeb: port=0x01f2 data=0x10
[IDE] -1
[I/O] writeb: port=0x01f3 data=0x00
[I/O] writeb: port=0x01f4 data=0x00
[I/O] writeb: port=0x01f5 data=0x00
[I/O] writeb: port=0x01f7 data=0xc6
[IDE] Command: SET MULTIPLE MODE (16)
[IDE] SET MULTIPLE MODE command failed <-- key line
[PIC] Raising IRQ 14
[APIC] Received bus message: vector=91 type=3 trigger=0
[APIC] Sending interrupt 91
[I/O] readb: port=0xc002 res=0x26
[I/O] writeb: port=0xc002 data=0x04
[APIC] EOI'ed: 91 Next highest: ffffffff
[I/O] writeb: port=0x01f6 data=0xb0
[IDE] Chose slave drive on primary
[I/O] writeb: port=0x01f6 data=0xb0
[IDE] Chose slave drive on primary
[I/O] writeb: port=0x01f1 data=0x03
[I/O] writeb: port=0x01f2 data=0x0b
[IDE] -1
[I/O] writeb: port=0x01f3 data=0x00
[I/O] writeb: port=0x01f4 data=0x00
[I/O] writeb: port=0x01f5 data=0x00
[I/O] writeb: port=0x01f7 data=0xef
[IDE] Command: SET FEATURES [idx=03]
[PIC] Raising IRQ 14
[APIC] Received bus message: vector=91 type=3 trigger=0
[APIC] Sending interrupt 91
[I/O] readb: port=0xc002 res=0x26
[I/O] writeb: port=0xc002 data=0x04
[APIC] EOI'ed: 91 Next highest: ffffffff
[I/O] writeb: port=0x0176 data=0xa0
[I/O] writeb: port=0x0376 data=0x06
[IDE] Device Control Register: 06
[I/O] writeb: port=0x0376 data=0x00
[IDE] Device Control Register: 00
[IDE] Reset controller id=1
[PIC] Raising IRQ 0
[I/O] readb: port=0x0177 res=0x00
[I/O] writeb: port=0x0176 data=0xa0
[I/O] readb: port=0x0177 res=0x00

After a few hundred page faults and #NM exceptions, the following READ MULTIPLE command is executed:

[I/O] writeb: port=0x01f6 data=0xa0
[IDE] Chose master drive on primary
[I/O] writeb: port=0x01f6 data=0xe0
[IDE] Chose master drive on primary
[I/O] writeb: port=0x01f1 data=0x00
[I/O] writeb: port=0x01f2 data=0x10
[IDE] -1
[I/O] writeb: port=0x01f3 data=0xa0
[I/O] writeb: port=0x01f4 data=0x28
[I/O] writeb: port=0x01f5 data=0xb7
[I/O] writeb: port=0x01f7 data=0xc4
[IDE] Command: READ MULTIPLE [w/o LBA]
[IDE] READ MULTIPLE failed
[PIC] Raising IRQ 14
[APIC] Received bus message: vector=91 type=3 trigger=0
[APIC] Sending interrupt 91
[I/O] readb: port=0xc002 res=0x26
[I/O] writeb: port=0xc002 data=0x04
[I/O] readb: port=0x01f1 res=0x04
[APIC] EOI'ed: 91 Next highest: ffffffff

Note that this command is never retried, despite the IDE controller aborting the command. All the other reads are done correctly.

A few failed disk reads later, it crashes with a BSOD (this time, the screen is blue), further indicating that the READ MULTIPLE command is probably the reason behind this crash.

BSOD

So far, I have determined the following:

  • Disabling the READ MULTIPLE command results in different behavior than with it enabled (this alone should be a red flag)
  • Most of the early reads are of 128-sector blocks, read 16 sectors at a time. Based on how the Windows boot screen loads, most invocations of the READ MULTIPLE work fine
  • The READ MULTIPLE invocations that fail read anywhere from 8 to 32 sectors at a time
  • I believe the problem is that we're reading multiple sectors at a time because the same problem happened with DMA reads/writes too
  • I still don't know which read in particular is upsetting Windows
  • Windows 8 has a very strange hard disk driver

@nepx
Copy link
Owner Author

nepx commented May 19, 2020

Windows 8 doesn't seem to do any kind of error checking on multi-sector reads. It assumes the read succeeded, even if it returned garbage, transferred zero bytes, or did something else entirely.

I created a build of Bochs that disabled multi-sector reads completely and Windows 8 crashed with a similar BSOD, indicating that READ MULTIPLE is most likely at fault. Or maybe it's a problem with WRITE MULTIPLE. I tried installing Windows without multiple sector support, but it hung during installation.

Why do multi-sector reads work in OS/2 but not on Windows 8?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@nepx