FPU saving + V86 problem - not sure how to solve it

[Anich Gregor]

Hi!

I have a problem with my FPU state saving changes and v86 mode.
In ke/i386/v86m_sup.S are the functions which enter/leave v86 mode...

Ki386RetToV86Mode enters v86 mode. It changes CurrentThread->InitialStack and 
TSS->Esp0 to %esp to make the exception handler run below the used stack and 
not overwrite it.
The exception handler (KeV86GPF) uses a PKV86M_TRAP_FRAME, which contains a 
KTRAP_FRAME at the beginning and some additional info behind. This additional 
info is the data on the stack of Ki386RetToV86Mode, since it sets TSS->Esp0 
to %esp, the CPU will push the trap-frame below when an exception is raised 
and the additional info will be after the trap-frame.

The problem is that for reserving a FX_SAVE_AREA at the top of the 
kernel-stack (without changing the KTHREAD->InitialStack value) i changed 
tskswitch.S to set TSS->Esp0 to NewThread->InitialStack - 
sizeof(FX_SAVE_AREA), not NewThread->InitialStack. Now if it's switching away 
and back to a v86 task TSS->Esp0 will be sizeof(FX_SAVE_AREA) less than 
before, and the V86 exception handler will be unable to find the additional 
info from Ki386RetToV86Mode's original stack.

I am not sure how to solve it. I think it could be done by
1) Adding another variable somewhere (or maybe there is already one somewhere 
which is meant to be used for this) for the exception-handler stack of the 
thread, which can then be used by tskswitch.S to update TSS->Esp0.
2) only substract sizeof(FX_SAVE_AREA) from InitalStack for setting up 
TSS->Esp0 in tskswitch.S if the VM bit of the EFLAGS is not set. Problem is 
that the CPU unsets the VM bit if using protected-mode exception handlers for 
v86 tasks, so we would need to get a copy (or pointer to) the eflags pushed 
by the CPU for the IRQ handler into tskswitch.S (change 
PsDispatchThread/PsDispatchThreadNoLock prototypes)

All this assumes that the FPU would never be used in V86 mode (and I hope that 
this assumption is correct ;-))

Any suggestions are welcome.

- blight

[Barubary]

I have a question about V86 mode...

Can any process set itself to V86 mode in ReactOS?  In Microsoft's NT, a 
program simply calls NtSetContextThread with the VM flag in EFLAGS to switch 
to V86 mode.

Melissa

----- Original Message ----- 
From: "Anich Gregor" [e-mail]
To: [e-mail]
Sent: Thursday, October 28, 2004 05:47
Subject: [ros-dev] FPU saving + V86 problem - not sure how to solve it
> Hi!
>
> I have a problem with my FPU state saving changes and v86 mode.
> In ke/i386/v86m_sup.S are the functions which enter/leave v86 mode...
>
> Ki386RetToV86Mode enters v86 mode. It changes CurrentThread->InitialStack 
> and
> TSS->Esp0 to %esp to make the exception handler run below the used stack 
> and
> not overwrite it.
> The exception handler (KeV86GPF) uses a PKV86M_TRAP_FRAME, which contains 
> a
> KTRAP_FRAME at the beginning and some additional info behind. This 
> additional
> info is the data on the stack of Ki386RetToV86Mode, since it sets 
> TSS->Esp0
> to %esp, the CPU will push the trap-frame below when an exception is 
> raised
> and the additional info will be after the trap-frame.
>
> The problem is that for reserving a FX_SAVE_AREA at the top of the
> kernel-stack (without changing the KTHREAD->InitialStack value) i changed
> tskswitch.S to set TSS->Esp0 to NewThread->InitialStack -
> sizeof(FX_SAVE_AREA), not NewThread->InitialStack. Now if it's switching 
> away
> and back to a v86 task TSS->Esp0 will be sizeof(FX_SAVE_AREA) less than
> before, and the V86 exception handler will be unable to find the 
> additional
> info from Ki386RetToV86Mode's original stack.
>
> I am not sure how to solve it. I think it could be done by
> Adding another variable somewhere (or maybe there is already one
>
> somewhere
> which is meant to be used for this) for the exception-handler stack of the
> thread, which can then be used by tskswitch.S to update TSS->Esp0.
> 2) only substract sizeof(FX_SAVE_AREA) from InitalStack for setting up
> TSS->Esp0 in tskswitch.S if the VM bit of the EFLAGS is not set. Problem 
> is
> that the CPU unsets the VM bit if using protected-mode exception handlers 
> for
> v86 tasks, so we would need to get a copy (or pointer to) the eflags 
> pushed
> by the CPU for the IRQ handler into tskswitch.S (change
> PsDispatchThread/PsDispatchThreadNoLock prototypes)
>
> All this assumes that the FPU would never be used in V86 mode (and I hope 
> that
> this assumption is correct ;-))
>
> Any suggestions are welcome.
> blight
>
> Ros-dev mailing list
> [e-mail]
> http://reactos.com:8080/mailman/listinfo/ros-dev