Different problem with fastinterlck_asm.S ;-(

[Alex Buell]

[GAS]      ntoskrnl/ex/i386/fastinterlck_asm.S
ntoskrnl/ex/i386/fastinterlck_asm.S: Assembler messages:
ntoskrnl/ex/i386/fastinterlck_asm.S:408: Error: suffix or operands
invalid for `cmpxchg8b' ntoskrnl/ex/i386/fastinterlck_asm.S:452: Error:
suffix or operands invalid for `cmpxchg8b'
ntoskrnl/ex/i386/fastinterlck_asm.S:492: Error: suffix or operands
invalid for `cmpxchg8b' ntoskrnl/ex/i386/fastinterlck_asm.S:626: Error:
suffix or operands invalid for `cmpxchg8b'
ntoskrnl/ex/i386/fastinterlck_asm.S:658: Error: suffix or operands
invalid for `cmpxchg8b' make[1]: ***
[obj-i386/ntoskrnl/ex/i386/fastinterlck_asm.o] Error 1
-- 
http://www.munted.org.uk

Anyone that thinks an imaginary deity is going to protect them against 
earthquakes and hurricanes needs psychiatric help.

[Ge van Geldorp]

> From: Alex Buell
>
> [GAS]      ntoskrnl/ex/i386/fastinterlck_asm.S
> ntoskrnl/ex/i386/fastinterlck_asm.S: Assembler messages:
> ntoskrnl/ex/i386/fastinterlck_asm.S:408: Error: suffix or operands invalid

for `cmpxchg8b'

I believe cmpxchg8b is a Pentium-or-higher instruction, so we shouldn't be
using it (at least not without testing if it's available).

GvG

[Alex Buell]

On Sun, 1 Jan 2006 20:22:38 +0100 "Ge van Geldorp" [e-mail]
waved a wand and this message magically appeared:
> > From: Alex Buell
> >
> > [GAS]      ntoskrnl/ex/i386/fastinterlck_asm.S
> > ntoskrnl/ex/i386/fastinterlck_asm.S: Assembler messages:
> > ntoskrnl/ex/i386/fastinterlck_asm.S:408: Error: suffix or operands invalid
>
> for `cmpxchg8b'
>
> I believe cmpxchg8b is a Pentium-or-higher instruction, so we shouldn't be
> using it (at least not without testing if it's available).

FYI I use the following versions: binutils 2.15.90.0.2, gcc 3.4.2 to
build ReactOS.
-- 
http://www.munted.org.uk

Anyone that thinks an imaginary deity is going to protect them against 
earthquakes and hurricanes needs psychiatric help.

[Ged Murphy]

Alex Buell wrote:
> FYI I use the following versions: binutils 2.15.90.0.2, gcc 3.4.2 to
> build ReactOS.

Why don't you try upgrading. That's a pretty old version.

I'm using 2.16.91 and that works fine.

[Ge van Geldorp]

> From: Ged Murphy
>
> Why don't you try upgrading. That's a pretty old version.

That doesn't solve the problem that we seem to be using Pentium-only
instructions (killing i486 support).

GvG

[Alex Ionescu]

Ge van Geldorp wrote:
> > From: Alex Buell
> >
> > [GAS]      ntoskrnl/ex/i386/fastinterlck_asm.S
> > ntoskrnl/ex/i386/fastinterlck_asm.S: Assembler messages:
> > ntoskrnl/ex/i386/fastinterlck_asm.S:408: Error: suffix or operands invalid
>
> for `cmpxchg8b'
>
> I believe cmpxchg8b is a Pentium-or-higher instruction, so we shouldn't be
> using it (at least not without testing if it's available).
>
> GvG

Hi,

We already did, so I simply re-used that assumption. Does/should ReactOS 
still work on 486? If so, I can add code to test and modify the 
assumption (but once again, I note that some of the original code 
already used cmpxhcg8b and nobody has complained ever.

Best regards,
Alex Ionescu

[Hartmut Birr]

Alex Ionescu wrote:
> Ge van Geldorp wrote:
> > > From: Alex Buell
> > >
> > > [GAS]      ntoskrnl/ex/i386/fastinterlck_asm.S
> > > ntoskrnl/ex/i386/fastinterlck_asm.S: Assembler messages:
> > > ntoskrnl/ex/i386/fastinterlck_asm.S:408: Error: suffix or operands
> > > invalid
> >
> > for `cmpxchg8b'
> > I believe cmpxchg8b is a Pentium-or-higher instruction, so we
> > shouldn't be
> > using it (at least not without testing if it's available).
> >
> > GvG
>
> Hi,
>
> We already did, so I simply re-used that assumption. Does/should
> ReactOS still work on 486? If so, I can add code to test and modify
> the assumption (but once again, I note that some of the original code
> already used cmpxhcg8b and nobody has complained ever.
>
> Best regards,
> Alex Ionescu

cmpxchg8b was introduced by me for ExfpInterlockedExchange64 and
ExfInterlockedCompareExchange64. This two functions are used in page.c
for pae mode functions. Pae mode works only on pentium or newer cpus.
The kernel works also on a 486, because pae mode would be never enabled
on a 486.

- Hartmut

[Alex Ionescu]

Hartmut Birr wrote:
> cmpxchg8b was introduced by me for ExfpInterlockedExchange64 and
> ExfInterlockedCompareExchange64. This two functions are used in page.c
> for pae mode functions. Pae mode works only on pentium or newer cpus.
> The kernel works also on a 486, because pae mode would be never enabled
> on a 486.
> Hartmut

Good point! I'll fix this up ASAP.

Best regards,
Alex Ionescu

[Alex Buell]

On Mon, 02 Jan 2006 10:27:28 -0500 Alex Ionescu [e-mail]
waved a wand and this message magically appeared:
> We already did, so I simply re-used that assumption. Does/should ReactOS 
> still work on 486? If so, I can add code to test and modify the 
> assumption (but once again, I note that some of the original code 
> already used cmpxhcg8b and nobody has complained ever.

I think you'll find that it worked in the past because it used different
opcodes/suffixes.
-- 
http://www.munted.org.uk

Anyone that thinks an imaginary deity is going to protect them against 
earthquakes and hurricanes needs psychiatric help.