Steve and Geoff,
Thanks both of you for the quick response. I took your advice and added
support for stat64. In syscall_emul.hh at line 606 I added the following :
/// ADDED 9/6/2007
/// Target stat64() handler.
template <class OS>
SyscallReturn
stat64Func(SyscallDesc *desc, int callnum, LiveProcess *process,
ThreadContext *tc)
{
std::string path;
if (!tc->getMemPort()->tryReadString(path, tc->getSyscallArg(0)))
return -EFAULT;
// Adjust path for current working directory
path = process->fullPath(path);
struct stat64 hostBuf;
int result = stat64(path.c_str(), &hostBuf);
if (result < 0)
return -errno;
copyOutStat64Buf<OS>(tc->getMemPort(), tc->getSyscallArg(1), &hostBuf);
return 0;
}
/// END ADDED 9/6/2007
And added a pointer to that in the /arch/alpha/linx/process.cc file. It built
and ran the bechmarks giving me that error (until a core dump), but at least
that problem has been fixed so thanks a lot for that! Although, since the
benchmarks didn't actually complete, I can't say for sure that it actually
worked (lol).
In regards to the core dumps, here is the output of trying to run the SPEC06
PerlBench :
[EMAIL PROTECTED]:~/Development/M5/m5-2.0b3/build/ALPHA_SE$ ./m5.debug
../../configs/example/se.py -c
benchmarks/400.perlbench/exe/perlbench_base.amd64-m64-gcc41-nn -i
benchmarks/400.perlbench/data/ref/input/checkspam.pl -o
"-Ibenchmarks/400.perlbench/run/run_base_ref_amd64-m64-gcc41-nn.0000/lib/
benchmarks/400.perlbench/data/ref/input/checkspam.pl 2500 5 25 11 150 1 1 1 1"
M5 Simulator System
Copyright (c) 2001-2006
The Regents of The University of Michigan
All Rights Reserved
M5 compiled Sep 7 2007 13:15:11
M5 started Fri Sep 7 13:20:02 2007
M5 executing on portonaught
command line: ./m5.debug ../../configs/example/se.py -c
benchmarks/400.perlbench/exe/perlbench_base.amd64-m64-gcc41-nn -i
benchmarks/400.perlbench/data/ref/input/checkspam.pl -o
-Ibenchmarks/400.perlbench/run/run_base_ref_amd64-m64-gcc41-nn.0000/lib/
benchmarks/400.perlbench/data/ref/input/checkspam.pl 2500 5 25 11 150 1 1 1 1
Global frequency set at 1000000000000 ticks per second
warn: More than two loadable segments in ELF object.
warn: Ignoring segment @ 0x12021e000 length 0x20.
warn: More than two loadable segments in ELF object.
warn: Ignoring segment @ 0x0 length 0x0.
warn: Entering event queue @ 0. Starting simulation...
warn: Increasing stack size by one page.
warn: Increasing stack size by one page.
warn: Increasing stack size by one page.
panic: Page table fault when accessing virtual address 0
@ cycle 5278465500
[invoke:build/ALPHA_SE/sim/faults.cc, line 65]
Program aborted at cycle 5278465500
Aborted (core dumped)
Sorry about the poor formating. Thanks again for all the help.
Elliott
Steve Reinhardt <[EMAIL PROTECTED]> wrote: Hi Elliott,
I haven't heard of anyone working on this (please speak up if you have), but
it's a great idea and we'd be glad to give you a hand it getting it to work.
As Geoff pointed out, one straightforward way to do it would be to run under FS
mode; then you won't have the problem of emulating additional syscalls.
However if you want to run under SE mode it shouldn't be that hard to get
things up either.
The stat64 call should be pretty straightforward to fix... we implement
stat/lstat/fstat and lstat64/fstat64 so triangulating among those to figure out
stat64 shouldn't be too hard. I'm sure the only reason stat64 isn't there is
that no one has needed it yet. Basically you'll need to add a stat64Func in
src/sim/syscall_emul.hh that looks like statFunc but calls stat64() instead
(basically apply the same diffs as between fstatFunc and fstat64Func), then add
a pointer to that function to the Alpha Linux syscall table in
src/arch/alpha/linux/process.cc.
The other core dumps and page table faults should be fixable too (if they're
not things we've already fixed since the last release... got to get that public
repo access going!, but we'd need more details on what those are to say more.
Steve
On 9/7/07, Elliott Cooper-Balis <[EMAIL PROTECTED]> wrote: M5 Users,
Greetings. I am a new to both M5 and the list so please forgive me if I say
anything stupid. I am currently trying to get the SPEC CPU2006 benchmarks to
run on M5. I searched the list archive and found absolutely no mention of
anyone trying to do this, so if this is futile/not supported, please feel free
to ignore the rest of this email.
Under Ubuntu 7.04 (64-bit), I managed to build crosstool's (
http://www.kegel.com/crosstool/ ) Alpha gcc/g++/gfortran compilers and
subsequently build Alpha binaries for all the SPEC06 benchmarks.
Unfortunately, the only benchmark I could get to run correctly was bzip2. All
the rest either produced a page table fault/core dump or gave me an error due
to an unimplemented syscall (stat64).
Is there a trick to getting these benchmarks to work correctly? Is this even
possible at this point? Any help would be greatly appreciated. Thanks in
advance.
Elliott
---------------------------------
Yahoo! oneSearch: Finally, mobile search that gives answers, not web links.
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
---------------------------------
Yahoo! oneSearch: Finally, mobile search that gives answers, not web links. _______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users
