DPMI errors on amd64, but not on i386

DPMI errors on amd64, but not on i386

[Javier]

I'm using dosemu 1.4.0.1 on Debian Lenny (compiled by myself as a 32 bit
executable). It works pretty well under a x86_32 2.6.26 kernel, running Windows
3.1 both in 286 & 386 mode, DJGPP, etc. 

However, whenever I boot a x86_64 kernel, command.com and the rest of 16-bit
applications run fine, but Windows and DJGPP do not. 

Windows only successfully boots every 100th time I launch it. Often, it just
shows "DPMI: Unhandheld Exception 0d"*. Sometimes it just hangs. After a
successful boot, it is also very buggy: Netscape (and most applications) tend to
crash due to page faults.

*Forcing 386 mode (win /d /3) results usually in DPMI Exception 06. Forcing 286
mode results in DPMI Exception 0d. 
Often, "ERROR: SMALLOC: bad pointer passed to smfree()" appears on dosemu's
stderr.

I've tried the Debian packaged 1.4.0.0 amd64 binary, but the result is the same.
I've also tried vm86dev for x86_64; "runcom" does indeed work, 16-bit
applications are faster, but windows does not boot either. 

Re: DPMI errors on amd64, but not on i386

[Julius Schwartzenberg]

Javier wrote:
> Windows only successfully boots every 100th time I launch it. Often, it just
> shows "DPMI: Unhandheld Exception 0d"*. Sometimes it just hangs. After a
> successful boot, it is also very buggy: Netscape (and most applications) tend to
> crash due to page faults.

I'm getting similar results with DPMI on AMD64. According to dosemu.conf 
DMPI on AMD64 is experimental, so I guess this is to be expected. I have 
the same problem when CPUEMU is set to full though, so I guess the CPU 
emulation is also still experimental.

Re: DPMI errors on amd64, but not on i386

[Javier]

Julius Schwartzenberg wrote:
> I'm getting similar results with DPMI on AMD64. According to dosemu.conf
> DMPI on AMD64 is experimental, so I guess this is to be expected. I have
> the same problem when CPUEMU is set to full though, so I guess the CPU
> emulation is also still experimental.

Please note that I am using the 32-bit binary on both the x86_32 and 64 
kernel -- I've no knowledge about how dosemu works, but does it follow a
diferent code path on amd64 even if compiled for i386? (apart from 
x86_64's missing vm86plus syscall). 
Because of this, I initially believed I had misconfigured my kernel 
build, but since I am also having problems with the default 64-bit Debian 
kernel I guess not.

I've also tried a 64-bit dosemu binary, but same results. 

Thanks.

Re: DPMI errors on amd64, but not on i386

[Julius Schwartzenberg]

Javier wrote:
> Julius Schwartzenberg wrote:
>> I'm getting similar results with DPMI on AMD64. According to dosemu.conf
>> DMPI on AMD64 is experimental, so I guess this is to be expected. I have
>> the same problem when CPUEMU is set to full though, so I guess the CPU
>> emulation is also still experimental.
> 
> Please note that I am using the 32-bit binary on both the x86_32 and 64 
> kernel -- I've no knowledge about how dosemu works, but does it follow a
> diferent code path on amd64 even if compiled for i386? (apart from 
> x86_64's missing vm86plus syscall). 
> Because of this, I initially believed I had misconfigured my kernel 
> build, but since I am also having problems with the default 64-bit Debian 
> kernel I guess not.

I suspect the experimental limitation doesn't come from the 32 or 64 bit 
binary, but from the 64-bits kernel. I understand certain 16-bit mode 
things aren't possible when the system runs in 64-bit mode, so that's 
why the emulation is necessary for certain parts then. This emulation 
feature is still experimental.

Re: DPMI errors on amd64, but not on i386

[Julius Schwartzenberg]

Alain M. wrote:
> Have you serached the list archives? I remember there was a message 
> saying that it was fixed post-1.4.0. This means that you will have to 
> compile from sources...

I built a package with the latest SVN tree, but I'm still having these 
problems.