[RFC,go] : Port to ALPHA arch

On Sat, Apr 2, 2011 at 1:09 AM, Ian Lance Taylor <iant@google.com> wrote:
> On Wed, Mar 30, 2011 at 12:58 PM, Uros Bizjak <ubizjak@gmail.com> wrote:
>>
>> Attached ports go to ALPHA architecture.
>
> Thanks!
>
> Committed.
>
>
>> b) alpha doesn't define "struct user_regs_struct" from which "type
>> PtraceRegs" is derived. I have manually created PtraceRegs from
>> pt_regs structure and patched generated libgo/sysinfo.go in build
>> directory after the build broke. However - the comment from sys/user.h
>> says that this file is for GDB and GDB only...
>
> libgo uses it to support ptrace, which in effect is the same as what
> gdb does.  If mksysinfo.sh is unable to provide any definition for the
> structure, then the choices are either to patch up mksysinfo.sh so
> that it works, or to simply define the structure in
> libgo/syscalls/syscall_linux_alpha.go.  it is unlikely to change so
> the latter seems acceptable if patching mksysinfo is too hard.

Attached patch implements this suggestion and builds libgo without problems.

>> c) some weird issue with double definition of "const _SOCK_NONBLOCK"
>> in gen-sysinfo.go and consequently sysinfo.go.
>
> The code in gcc/godump.c uses a hash table to keep this from
> happening; I'm not sure why that is not working in this case.  Is
> SOCK_NONBLOCK both a #define macro and an enum value?

Yes, SOCK_NONBLOCK is defined in bits/sock.h as

--cut here--
enum __socket_type
{
#undef SOCK_NONBLOCK
  SOCK_NONBLOCK = 0x40000000    /* Atomically mark descriptor(s) as
                                   non-blocking.  */
#define SOCK_NONBLOCK SOCK_NONBLOCK
};

/* Get socket manipulation related informations from kernel headers.  */
#include "asm/socket.h"
--cut here--

asm/socket.h:

--cut here--
/* O_NONBLOCK clashes with the bits used for socket types.  Therefore we
 *  have to define SOCK_NONBLOCK to a different value here.
 */
#define SOCK_NONBLOCK   0x40000000
--cut here--

Processing bits/socket.h with -fdumo-go-spec= , we get:

// unknowndefine SOCK_NONBLOCK SOCK_NONBLOCK
const _SOCK_NONBLOCK = 0x40000000
const _SOCK_NONBLOCK = 1073741824

>
>> The test results from "make -k check" in libgo directory are quite encouraging:
>
> Yes.
>
>> FAIL: cmath
>> panic: runtime error: integer divide by zero or floating point error
>
> This is peculiar--it looks like there is something systematically
> causing a division by zero.  I wonder what it could be.

I will try to analyse these failures. Perhaps these failures are due
to the fact that alpha needs -mieee to handle NaN and Inf values (a
FPE is generated otherwise).

Thanks,
Uros.

Uros Bizjak <ubizjak@gmail.com> writes:

> On Sat, Apr 2, 2011 at 1:09 AM, Ian Lance Taylor <iant@google.com> wrote:
>> On Wed, Mar 30, 2011 at 12:58 PM, Uros Bizjak <ubizjak@gmail.com> wrote:
>>>
>>> Attached ports go to ALPHA architecture.
>>
>> Thanks!
>>
>> Committed.
>>
>>
>>> b) alpha doesn't define "struct user_regs_struct" from which "type
>>> PtraceRegs" is derived. I have manually created PtraceRegs from
>>> pt_regs structure and patched generated libgo/sysinfo.go in build
>>> directory after the build broke. However - the comment from sys/user.h
>>> says that this file is for GDB and GDB only...
>>
>> libgo uses it to support ptrace, which in effect is the same as what
>> gdb does.  If mksysinfo.sh is unable to provide any definition for the
>> structure, then the choices are either to patch up mksysinfo.sh so
>> that it works, or to simply define the structure in
>> libgo/syscalls/syscall_linux_alpha.go.  it is unlikely to change so
>> the latter seems acceptable if patching mksysinfo is too hard.
>
> Attached patch implements this suggestion and builds libgo without problems.

Committed.  Thanks.

Ian