[Qemu-devel] /* XXX: suppress this hack */

[Qemu-devel] /* XXX: suppress this hack */

[Magnus Damm]

Hi,

Better performance for i386-softmmu under PowerPC? Yes, please!
At line 96 in "dyngen-exec.h" sits the following code:

/* XXX: suppress this hack */
#if defined(CONFIG_USER_ONLY)
#define AREG4 "r16"
#define AREG5 "r17"
#define AREG6 "r18"
#define AREG7 "r19"
#define AREG8 "r20"
#define AREG9 "r21"
#define AREG10 "r22"
#define AREG11 "r23"
#endif

I would like to "suppress that hack", ie make sure that AREG4 to AREG11
gets defined for i386-softmmu. Someone that knows the code - what needs
to be done? Just removing the ifdef does not work for sure. Is it some
issue with setjmp()/longjmp()? I'm willing to set aside some hours to
fix this if someone who knows points me in the right direction.

Thanks!

/ magnus

Re: [Qemu-devel] /* XXX: suppress this hack */

[Johannes Schindelin]

Hi,

On Sun, 10 Oct 2004, Magnus Damm wrote:

> Hi,
>
> Better performance for i386-softmmu under PowerPC? Yes, please!
> At line 96 in "dyngen-exec.h" sits the following code:
>
> /* XXX: suppress this hack */
> #if defined(CONFIG_USER_ONLY)
> #define AREG4 "r16"
> #define AREG5 "r17"
> #define AREG6 "r18"
> #define AREG7 "r19"
> #define AREG8 "r20"
> #define AREG9 "r21"
> #define AREG10 "r22"
> #define AREG11 "r23"
> #endif
>
> I would like to "suppress that hack", ie make sure that AREG4 to AREG11
> gets defined for i386-softmmu. Someone that knows the code - what needs
> to be done? Just removing the ifdef does not work for sure. Is it some
> issue with setjmp()/longjmp()? I'm willing to set aside some hours to
> fix this if someone who knows points me in the right direction.

It's all about ABIs: see

http://www-106.ibm.com/developerworks/linux/library/l-ppc/#h4

You want all those registers you use to be volatile, i.e. every function
which wants to use one of those has to make sure that their original
contents gets restored before returning.

Hth,
Dscho

Re: [Qemu-devel] /* XXX: suppress this hack */

[Magnus Damm]

Hello there, thanks for the reply!

[snip]

> > I would like to "suppress that hack", ie make sure that AREG4 to AREG11
> > gets defined for i386-softmmu. Someone that knows the code - what needs
> > to be done? Just removing the ifdef does not work for sure. Is it some
> > issue with setjmp()/longjmp()? I'm willing to set aside some hours to
> > fix this if someone who knows points me in the right direction.
> 
> It's all about ABIs: see
> 
> http://www-106.ibm.com/developerworks/linux/library/l-ppc/#h4
> 
> You want all those registers you use to be volatile, i.e. every function
> which wants to use one of those has to make sure that their original
> contents gets restored before returning.

Yeah, registers "r3" to "r12" are "caller save", ie these registers must
be saved before a function is called. But what about "r16" to "r23",
these are currently only used when CONFIG_USER_ONLY is set - I thought
that they were supposed to be saved by the called function prior use,
"callee save". But how does that affect the softmmu code in a bad way
but the user code works? Any ideas?

Thanks!

/ magnus

Re: [Qemu-devel] /* XXX: suppress this hack */

[Johannes Schindelin]

Hi,

On Sun, 10 Oct 2004, Magnus Damm wrote:

> Yeah, registers "r3" to "r12" are "caller save", ie these registers must
> be saved before a function is called. But what about "r16" to "r23",
> these are currently only used when CONFIG_USER_ONLY is set - I thought
> that they were supposed to be saved by the called function prior use,
> "callee save". But how does that affect the softmmu code in a bad way
> but the user code works? Any ideas?

*scratcheshishead* I cannot think of one way that code should work in user
mode... IMHO there is no protection to those registers when calling libc
or a syscall.

Ciao,
Dscho