[Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Karel Gardas]

Hello,

I have performed simple testing of qemu-0.5.2 (vide also my last email
about ARM emulation issues). I have used these tests:

1) C++ hello world example using iostreams
2) C++ hello world example using C stdio
3) C hello world example using C stdio
4) bogomips test

Tests 1-3 were tested in two versions, one linked dynamically and one
staticaly.

The results are (grouped by qemu-<processor>)

qemu-arm:

1) stat: failed
   dyn: failed
2) stat: run
   dyn: failed
3) stat: run
   dyn: failed
4) run (160 BogoMIPS -- very good!)

qemu-ppc:

1) stat: failed
   dyn: failed (qemu segfaults)
2) stats: run
   dyn: failed (qemu segfaults)
3) stat: run
   dyn: run
4) run (40 BogoMIPS)

qemu-sparc:

1) stat: run
   dyn: run
2) stat: run
   dyn: run
3) stat: run
   dyn: run
4) run (38 BogoMIPS)


Where it is not mentioned, binary failed with segfault, where is mentioned
``qemu segfaults'', QEMU segfaults directly.

My question is: is there anybody improving QEMU ARM support? Anyway, at
least sparc results looks very good, so I will probably try to test
qemu-sparc with some not so simple C++ application. Also is there anybody
here working on MIPS (mips32) support?

Thanks,

Karel
--
Karel Gardas                  kgardas@objectsecurity.com
ObjectSecurity Ltd.           http://www.objectsecurity.com

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Fabrice Bellard]

Interesting ! I thought that SPARC was the less stable target (I left 
some register window exception problems in it)... I think the problems 
on ARM are related to FPU emulation as it is not implemented yet (but I 
have a pending patch to merge).

ARM and x86 use direct translated block chaining while it is not 
implemented yet on SPARC and PowerPC, so it explains the performances 
you get.

Can you send me (or make available somewhere) at least the static 
executables ? If you send the dynamic ones, try to add all the relevant 
dynamic libraries.

Fabrice.

Karel Gardas wrote:
> Hello,
> 
> I have performed simple testing of qemu-0.5.2 (vide also my last email
> about ARM emulation issues). I have used these tests:
> 
> 1) C++ hello world example using iostreams
> 2) C++ hello world example using C stdio
> 3) C hello world example using C stdio
> 4) bogomips test
> 
> Tests 1-3 were tested in two versions, one linked dynamically and one
> staticaly.
> 
> The results are (grouped by qemu-<processor>)
> 
> qemu-arm:
> 
> 1) stat: failed
>    dyn: failed
> 2) stat: run
>    dyn: failed
> 3) stat: run
>    dyn: failed
> 4) run (160 BogoMIPS -- very good!)
> 
> qemu-ppc:
> 
> 1) stat: failed
>    dyn: failed (qemu segfaults)
> 2) stats: run
>    dyn: failed (qemu segfaults)
> 3) stat: run
>    dyn: run
> 4) run (40 BogoMIPS)
> 
> qemu-sparc:
> 
> 1) stat: run
>    dyn: run
> 2) stat: run
>    dyn: run
> 3) stat: run
>    dyn: run
> 4) run (38 BogoMIPS)
> 
> 
> Where it is not mentioned, binary failed with segfault, where is mentioned
> ``qemu segfaults'', QEMU segfaults directly.
> 
> My question is: is there anybody improving QEMU ARM support? Anyway, at
> least sparc results looks very good, so I will probably try to test
> qemu-sparc with some not so simple C++ application. Also is there anybody
> here working on MIPS (mips32) support?
> 
> Thanks,
> 
> Karel
> --
> Karel Gardas                  kgardas@objectsecurity.com
> ObjectSecurity Ltd.           http://www.objectsecurity.com
> 
> 
> 
> 
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://mail.nongnu.org/mailman/listinfo/qemu-devel
> 
> 

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Karel Gardas]

On Mon, 16 Feb 2004, Fabrice Bellard wrote:

> Interesting ! I thought that SPARC was the less stable target (I left
> some register window exception problems in it)... I think the problems
> on ARM are related to FPU emulation as it is not implemented yet (but I
> have a pending patch to merge).

Sounds very good!

> ARM and x86 use direct translated block chaining while it is not
> implemented yet on SPARC and PowerPC, so it explains the performances
> you get.

Thanks for the info.

> Can you send me (or make available somewhere) at least the static
> executables ? If you send the dynamic ones, try to add all the relevant
> dynamic libraries.

Would you like to have just arm or arm+ppc or all three?

Thanks,

Karel
--
Karel Gardas                  kgardas@objectsecurity.com
ObjectSecurity Ltd.           http://www.objectsecurity.com

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Karel Gardas]

On Mon, 16 Feb 2004, Fabrice Bellard wrote:

> Interesting ! I thought that SPARC was the less stable target (I left
> some register window exception problems in it)... I think the problems
> on ARM are related to FPU emulation as it is not implemented yet (but I
> have a pending patch to merge).

BTW: If this patch to merge is cleanly applicable to cvs head or some
release, then I can easily give it a try here -- if you provide me it.

Thanks,

Karel
--
Karel Gardas                  kgardas@objectsecurity.com
ObjectSecurity Ltd.           http://www.objectsecurity.com

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Karel Gardas]

On Mon, 16 Feb 2004, Fabrice Bellard wrote:

> Interesting ! I thought that SPARC was the less stable target (I left
> some register window exception problems in it)...

I have just finished cross-compilation of MICO (http://www.mico.org) and
try small demo which failed with:

thinkpad:/mnt/karel/mico-sparc/demo/bench$ qemu-sparc -L /mnt/karel/cross/crosstool-0.27/result/sparc-unknown-linux-gnu/gcc-3.3.2-glibc-2.3.2/sparc-unknown-linux-gnu/ ./server
Unhandled trap: 0x2
pc: 0x0002a2f8  npc: 0x0002a2fc
General Registers:
%g0: 0x00000000 %g1: 0x0031e070 %g2: 0x0042b000 %g3: 0x0042cf70
%g4: 0x00000000 %g5: 0x00005f00 %g6: 0xff000000 %g7: 0x401f3468
Current Register Window:
%o0: 0x00215ccc %o1: 0x00000000 %o2: 0x003c3110 %o3: 0x0042b4cc
%o4: 0x3000474c %o5: 0x401db790 %o6: 0x401db6d8 %o7: 0x00303d8c
%l0: 0x0042a070 %l1: 0x402079c0 %l2: 0x42367314 %l3: 0x42374a7a
%l4: 0x00000002 %l5: 0x40207cc8 %l6: 0x00000000 %l7: 0x0042a824
%i0: 0x0042a0cc %i1: 0x402079c0 %i2: 0x00000338 %i3: 0x00000334
%i4: 0x000007a9 %i5: 0x00000005 %i6: 0x401db740 %i7: 0x00127640
psr: 0x0000001c -> ---- wim: 0x00000002
thinkpad:/mnt/karel/mico-sparc/demo/bench$

So, indeed, there are still some issues. :-)

Anyway, overall impression from QEMU is very good, I just hope that I will
be able to use it for our MICO related testing for x-linux (where x !=
x86) platforms -- I plan to do some testing of a bit larger C++ app for
you, if you are interested.

Thanks,

Karel
--
Karel Gardas                  kgardas@objectsecurity.com
ObjectSecurity Ltd.           http://www.objectsecurity.com

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Fabrice Bellard]

You usually get this exception if a FPU instruction is used (qemu does 
not support FPU on sparc yet).

Fabrice.

Karel Gardas wrote:
> On Mon, 16 Feb 2004, Fabrice Bellard wrote:
> 
> 
>>Interesting ! I thought that SPARC was the less stable target (I left
>>some register window exception problems in it)...
> 
> 
> I have just finished cross-compilation of MICO (http://www.mico.org) and
> try small demo which failed with:
> 
> thinkpad:/mnt/karel/mico-sparc/demo/bench$ qemu-sparc -L /mnt/karel/cross/crosstool-0.27/result/sparc-unknown-linux-gnu/gcc-3.3.2-glibc-2.3.2/sparc-unknown-linux-gnu/ ./server
> Unhandled trap: 0x2
> pc: 0x0002a2f8  npc: 0x0002a2fc
> General Registers:
> %g0: 0x00000000 %g1: 0x0031e070 %g2: 0x0042b000 %g3: 0x0042cf70
> %g4: 0x00000000 %g5: 0x00005f00 %g6: 0xff000000 %g7: 0x401f3468
> Current Register Window:
> %o0: 0x00215ccc %o1: 0x00000000 %o2: 0x003c3110 %o3: 0x0042b4cc
> %o4: 0x3000474c %o5: 0x401db790 %o6: 0x401db6d8 %o7: 0x00303d8c
> %l0: 0x0042a070 %l1: 0x402079c0 %l2: 0x42367314 %l3: 0x42374a7a
> %l4: 0x00000002 %l5: 0x40207cc8 %l6: 0x00000000 %l7: 0x0042a824
> %i0: 0x0042a0cc %i1: 0x402079c0 %i2: 0x00000338 %i3: 0x00000334
> %i4: 0x000007a9 %i5: 0x00000005 %i6: 0x401db740 %i7: 0x00127640
> psr: 0x0000001c -> ---- wim: 0x00000002
> thinkpad:/mnt/karel/mico-sparc/demo/bench$
> 
> So, indeed, there are still some issues. :-)
> 
> Anyway, overall impression from QEMU is very good, I just hope that I will
> be able to use it for our MICO related testing for x-linux (where x !=
> x86) platforms -- I plan to do some testing of a bit larger C++ app for
> you, if you are interested.
> 
> Thanks,
> 
> Karel
> --
> Karel Gardas                  kgardas@objectsecurity.com
> ObjectSecurity Ltd.           http://www.objectsecurity.com
> 
> 
> 
> _______________________________________________
> Qemu-devel mailing list
> Qemu-devel@nongnu.org
> http://mail.nongnu.org/mailman/listinfo/qemu-devel
> 
> 

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Fabrice Bellard]

Karel Gardas wrote:

> qemu-sparc:
> 
> 1) stat: run
>    dyn: run
> 2) stat: run
>    dyn: run
> 3) stat: run
>    dyn: run
> 4) run (38 BogoMIPS)

I just commited direct chaining support for SPARC: the SPARC target is 
about 3 times faster on BogoMips. It should be possible to go faster by 
using a register for the register window pointer, but I cannot do it now.

$ ./qemu-sparc /tmp/package/sparc/sparc-bogomips
Calibrating delay loop.. ok - 360.00 BogoMips

$ qemu-sparc-0.5.2 /tmp/package/sparc/sparc-bogomips
Calibrating delay loop.. ok - 108.00 BogoMips

Fabrice.

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Karel Gardas]

On Mon, 16 Feb 2004, Fabrice Bellard wrote:

> Karel Gardas wrote:
>
> > qemu-sparc:
> >
> > 1) stat: run
> >    dyn: run
> > 2) stat: run
> >    dyn: run
> > 3) stat: run
> >    dyn: run
> > 4) run (38 BogoMIPS)
>
> I just commited direct chaining support for SPARC: the SPARC target is
> about 3 times faster on BogoMips. It should be possible to go faster by
> using a register for the register window pointer, but I cannot do it now.
>
> $ ./qemu-sparc /tmp/package/sparc/sparc-bogomips
> Calibrating delay loop.. ok - 360.00 BogoMips
>
> $ qemu-sparc-0.5.2 /tmp/package/sparc/sparc-bogomips
> Calibrating delay loop.. ok - 108.00 BogoMips
>

Wow! That's pretty impressive. I'm looking forward to seeing your FPU
support either for ARM or SPARC -- just to continue with my C++ testing.
:-)

Thanks,

Karel
--
Karel Gardas                  kgardas@objectsecurity.com
ObjectSecurity Ltd.           http://www.objectsecurity.com

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Fabrice Bellard]

Karel Gardas wrote:
> I'm looking forward to seeing your FPU
> support either for ARM or SPARC -- just to continue with my C++ testing.
> :-)

I commited the ARM FPU. It helps for one static ARM program. The dynamic 
programs still have problems.

Fabrice.

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Karel Gardas]

On Mon, 16 Feb 2004, Fabrice Bellard wrote:

> I commited the ARM FPU. It helps for one static ARM program. The dynamic
> programs still have problems.

I have tried to build MICO statically for ARM, but demo bench still fails.
If you are curious to see binaries, please download and unpack
http://mico.org/~karel/arm-corba.tar.bz2 -- I have provided simplified run
script just for your idea how to start things up -- the result should be
that the client will print how long time (in ms) takes one two-way CORBA
message sent from client to server.

BTW: Is there any way how to debug segfault(s) reported by Qemu?

Thanks,

Karel
--
Karel Gardas                  kgardas@objectsecurity.com
ObjectSecurity Ltd.           http://www.objectsecurity.com

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Fabrice Bellard]

I commited the last fixes of Jocelyn and the 'lwarx' fix makes more 
dynamic examples work.

Fabrice.

Re: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[J. Mayer]

On Mon, 2004-02-16 at 13:58, Karel Gardas wrote:
> Hello,
> 
> I have performed simple testing of qemu-0.5.2 (vide also my last email
> about ARM emulation issues). I have used these tests:
> 
> 1) C++ hello world example using iostreams
> 2) C++ hello world example using C stdio
> 3) C hello world example using C stdio
> 4) bogomips test
> 
...

> qemu-ppc:
> 
> 1) stat: failed
>    dyn: failed (qemu segfaults)
> 2) stats: run
>    dyn: failed (qemu segfaults)
> 3) stat: run
>    dyn: run
> 4) run (40 BogoMIPS)

I don't know if this fixes all issues, but I just found some bugs,
trying kernel-mode emulation, which will also affect user-mode:

1/ lwarx does all right, but doesn't keep the loaded value:

Index: target-ppc/op_mem.h
===================================================================
RCS file: /cvsroot/qemu/qemu/target-ppc/op_mem.h,v
retrieving revision 1.3
diff -u -d -w -B -b -d -p -r1.3 op_mem.h
--- target-ppc/op_mem.h 18 Jan 2004 22:49:57 -0000      1.3
+++ target-ppc/op_mem.h 20 Feb 2004 23:07:16 -0000
@@ -141,8 +141,8 @@ PPC_OP(glue(lwarx, MEMSUFFIX))
         do_queue_exception(EXCP_ALIGN);
         do_process_exceptions();
     } else {
-       glue(ldl, MEMSUFFIX)((void *)T0);
-       regs->reserve = T0 & ~0x03;
+       T1 = glue(ldl, MEMSUFFIX)((void *)T0);
+       regs->reserve = T0;
     }
     RETURN();
 }

2/ Always have to update nip when processing exceptions:
Index: op.c
===================================================================
RCS file: /cvsroot/qemu/qemu/target-ppc/op.c,v
retrieving revision 1.5
diff -u -d -w -B -b -d -p -r1.5 op.c
--- op.c        18 Jan 2004 22:49:57 -0000      1.5
+++ op.c        20 Feb 2004 23:10:40 -0000
@@ -220,8 +220,8 @@ PPC_OP(queue_exception)
 
 PPC_OP(process_exceptions)
 {
-    if (env->exceptions != 0) {
         env->nip = PARAM(1);
+    if (env->exceptions != 0) {
         do_check_exception_state();
     }
 }

3/ Fix check for lswi:
Index: translate.c
===================================================================
RCS file: /cvsroot/qemu/qemu/target-ppc/translate.c,v
retrieving revision 1.6
diff -u -d -w -B -b -d -p -r1.6 translate.c
--- translate.c 3 Feb 2004 23:39:42 -0000       1.6
+++ translate.c 20 Feb 2004 23:12:40 -0000
@@ -1229,8 +1225,9 @@ GEN_HANDLER(lswi, 0x1F, 0x15, 0x12, 0x00
     if (nb == 0)
         nb = 32;
     nr = nb / 4;
-    if (((start + nr) > 32  && start <= ra && (start + nr - 32) >= ra)
||
-        ((start + nr) <= 32 && start <= ra && (start + nr) >= ra)) {
+    if (((start + nr) > 32  && start <= ra && (start + nr - 32) > ra)
||
+        ((start + nr) <= 32 && start <= ra && (start + nr) > ra)) {
+       printf("Invalid lswi: start=%d nr=%d ra=%d\n", start, nr, ra);
         RET_EXCP(EXCP_PROGRAM, EXCP_INVAL | EXCP_INVAL_LSWX);
         }
     if (ra == 0) {

Hope this helps...

-- 
J. Mayer <l_indien@magic.fr>
Never organized

RE: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)

[Yelich, Scott D.]

Unfortunately, I'm not a coder... but I would love to see qemu on the zaurus (arm).
I'm monitoring this list for any news on the ARM (and sparc?) front.

Scott





-----Original Message-----
From: Karel Gardas [mailto:kgardas@objectsecurity.com]
Sent: Monday, February 16, 2004 7:59 AM
To: QEMU Development Mailing List
Subject: [Qemu-devel] qemu simple user mode testing (arm, ppc, sparc)


Hello,

I have performed simple testing of qemu-0.5.2 (vide also my last email
about ARM emulation issues). I have used these tests:

1) C++ hello world example using iostreams
2) C++ hello world example using C stdio
3) C hello world example using C stdio
4) bogomips test

Tests 1-3 were tested in two versions, one linked dynamically and one
staticaly.

The results are (grouped by qemu-<processor>)

qemu-arm:

1) stat: failed
   dyn: failed
2) stat: run
   dyn: failed
3) stat: run
   dyn: failed
4) run (160 BogoMIPS -- very good!)

qemu-ppc:

1) stat: failed
   dyn: failed (qemu segfaults)
2) stats: run
   dyn: failed (qemu segfaults)
3) stat: run
   dyn: run
4) run (40 BogoMIPS)

qemu-sparc:

1) stat: run
   dyn: run
2) stat: run
   dyn: run
3) stat: run
   dyn: run
4) run (38 BogoMIPS)


Where it is not mentioned, binary failed with segfault, where is mentioned
``qemu segfaults'', QEMU segfaults directly.

My question is: is there anybody improving QEMU ARM support? Anyway, at
least sparc results looks very good, so I will probably try to test
qemu-sparc with some not so simple C++ application. Also is there anybody
here working on MIPS (mips32) support?

Thanks,

Karel
--
Karel Gardas                  kgardas@objectsecurity.com
ObjectSecurity Ltd.           http://www.objectsecurity.com




_______________________________________________
Qemu-devel mailing list
Qemu-devel@nongnu.org
http://mail.nongnu.org/mailman/listinfo/qemu-devel

________________________________________________________________________
This email has been scanned for all viruses by the MessageLabs Email Security System.
________________________________________________________________________


This mail message originated outside Commerzbank via the Internet. As a result, the sender's address is not verifiable.


**********************************************************************
This communication is confidential and is intended only for the person to whom it is addressed.  If you are not that person you are not permitted to make use of the information and you are requested to notify Commerzbank Aktiengesellschaft, New York Branch immediately that you have received it and then to destroy the copy in your possession.  Views expressed in this e-mail do not necessarily reflect the views of Commerzbank AG.
**********************************************************************