Re: MMX and floating point insn suport for reversible debugging

Re: MMX and floating point insn suport for reversible debugging

[paawan oza]

Hi Hui,

I think we need to update and add floating point in following structure
for st1 to st7.

enum i386_regnum
{
  I386_EAX_REGNUM,		/* %eax */
  I386_ECX_REGNUM,		/* %ecx */
  I386_EDX_REGNUM,		/* %edx */
  I386_EBX_REGNUM,		/* %ebx */
  I386_ESP_REGNUM,		/* %esp */
  I386_EBP_REGNUM,		/* %ebp */
  I386_ESI_REGNUM,		/* %esi */
  I386_EDI_REGNUM,		/* %edi */
  I386_EIP_REGNUM,		/* %eip */
  I386_EFLAGS_REGNUM,		/* %eflags */
  I386_CS_REGNUM,		/* %cs */
  I386_SS_REGNUM,		/* %ss */
  I386_DS_REGNUM,		/* %ds */
  I386_ES_REGNUM,		/* %es */
  I386_FS_REGNUM,		/* %fs */
  I386_GS_REGNUM,		/* %gs */
  I386_ST0_REGNUM		/* %st(0) */
};

what do you say ?


Regards,
Oza.

--- On Thu, 5/7/09, Hui Zhu <teawater@gmail.com> wrote:

> From: Hui Zhu <teawater@gmail.com>
> Subject: Re: MMX and floating point insn suport for reversible debugging
> To: "paawan oza" <paawan1982@yahoo.com>
> Cc: "Thiago Jung Bauermann" <bauerman@br.ibm.com>, "gdb ml" <gdb@sourceware.org>, "Pedro Alves" <pedro@codesourcery.com>, "Marc Khouzam" <marc.khouzam@ericsson.com>, "Michael Snyder" <msnyder@vmware.com>, "Eli Zaretskii" <eliz@gnu.org>, "Mark Kettenis" <mark.kettenis@xs4all.nl>
> Date: Thursday, May 7, 2009, 8:24 AM
> On Thu, May 7, 2009 at 02:02, paawan
> oza <paawan1982@yahoo.com>
> wrote:
> >
> > Hi Hui,
> >
> > I have started understanding and working on point-1.
> > I have some questions. please let me know my
> understanding is correct ?
> >
> > 1) i386-tdep.c is the one which has been done for
> record-replay. which basically takes care of disassembling
> insns...
> > but I see some code regarding MMX and x87FPU already
> available.
> > may I know exactly, what support is missing ?
> >
> 
>     case 0x0f06:
>       break;
> 
>       /* MMX/SSE/SSE2/PNI support */
>       /* XXX */
> 
>     default:
> 
> Your mean is this part?
> 
> 
> And I suggest you can begin with float point insn
> first.  It doesn't
> record float point reg change.
> 
> 
> Thanks,
> Hui
> 

Re: MMX and floating point insn suport for reversible debugging

[Mark Kettenis]

> Date: Sat, 9 May 2009 23:37:20 -0700 (PDT)
> From: paawan oza <paawan1982@yahoo.com>
> 
> Hi Hui,
> 
> I think we need to update and add floating point in following structure
> for st1 to st7.
> 
> enum i386_regnum
> {
>   I386_EAX_REGNUM,		/* %eax */
>   I386_ECX_REGNUM,		/* %ecx */
>   I386_EDX_REGNUM,		/* %edx */
>   I386_EBX_REGNUM,		/* %ebx */
>   I386_ESP_REGNUM,		/* %esp */
>   I386_EBP_REGNUM,		/* %ebp */
>   I386_ESI_REGNUM,		/* %esi */
>   I386_EDI_REGNUM,		/* %edi */
>   I386_EIP_REGNUM,		/* %eip */
>   I386_EFLAGS_REGNUM,		/* %eflags */
>   I386_CS_REGNUM,		/* %cs */
>   I386_SS_REGNUM,		/* %ss */
>   I386_DS_REGNUM,		/* %ds */
>   I386_ES_REGNUM,		/* %es */
>   I386_FS_REGNUM,		/* %fs */
>   I386_GS_REGNUM,		/* %gs */
>   I386_ST0_REGNUM		/* %st(0) */
> };
> 
> what do you say ?

I don't think that should be necessary.

Re: MMX and floating point insn suport for reversible debugging

[paawan oza]

Hi Hui,

dont we need to save FPU flags also, while saving floating point registers?

for e.g. instruction
FMUL/FMULP/FIMUL—Multiply ; aafter the result,
FPU flags may get affected 
as follows.

FPU Flags Affected
C1 Set to 0 if stack underflow occurred.
Set if result was rounded up; cleared otherwise.
C0, C2, C3 Undefined.

what is you opinion ?

Regards,
Oza.

Re: MMX and floating point insn suport for reversible debugging

[paawan oza]

I think, 
record_arch_list_add_reg (ir.regcache, I386_ST0_REGNUM) will be sufficient 
to fetch all floating point registers.
no need to update the structure.
Regards,
Oza.


--- On Sun, 5/10/09, paawan oza <paawan1982@yahoo.com> wrote:

> From: paawan oza <paawan1982@yahoo.com>
> Subject: Re: MMX and floating point insn suport for reversible debugging
> To: "Hui Zhu" <teawater@gmail.com>
> Cc: "Thiago Jung Bauermann" <bauerman@br.ibm.com>, "gdb ml" <gdb@sourceware.org>, "Pedro Alves" <pedro@codesourcery.com>, "Marc Khouzam" <marc.khouzam@ericsson.com>, "Michael Snyder" <msnyder@vmware.com>, "Eli Zaretskii" <eliz@gnu.org>, "Mark Kettenis" <mark.kettenis@xs4all.nl>
> Date: Sunday, May 10, 2009, 12:07 PM
> 
> Hi Hui,
> 
> I think we need to update and add floating point in
> following structure
> for st1 to st7.
> 
> enum i386_regnum
> {
>   I386_EAX_REGNUM,   
>     /* %eax */
>   I386_ECX_REGNUM,   
>     /* %ecx */
>   I386_EDX_REGNUM,   
>     /* %edx */
>   I386_EBX_REGNUM,   
>     /* %ebx */
>   I386_ESP_REGNUM,   
>     /* %esp */
>   I386_EBP_REGNUM,   
>     /* %ebp */
>   I386_ESI_REGNUM,   
>     /* %esi */
>   I386_EDI_REGNUM,   
>     /* %edi */
>   I386_EIP_REGNUM,   
>     /* %eip */
>   I386_EFLAGS_REGNUM,   
>     /* %eflags */
>   I386_CS_REGNUM,       
> /* %cs */
>   I386_SS_REGNUM,       
> /* %ss */
>   I386_DS_REGNUM,       
> /* %ds */
>   I386_ES_REGNUM,       
> /* %es */
>   I386_FS_REGNUM,       
> /* %fs */
>   I386_GS_REGNUM,       
> /* %gs */
>   I386_ST0_REGNUM       
> /* %st(0) */
> };
> 
> what do you say ?
> 
> 
> Regards,
> Oza.
> 
> --- On Thu, 5/7/09, Hui Zhu <teawater@gmail.com>
> wrote:
> 
> > From: Hui Zhu <teawater@gmail.com>
> > Subject: Re: MMX and floating point insn suport for
> reversible debugging
> > To: "paawan oza" <paawan1982@yahoo.com>
> > Cc: "Thiago Jung Bauermann" <bauerman@br.ibm.com>,
> "gdb ml" <gdb@sourceware.org>,
> "Pedro Alves" <pedro@codesourcery.com>,
> "Marc Khouzam" <marc.khouzam@ericsson.com>,
> "Michael Snyder" <msnyder@vmware.com>,
> "Eli Zaretskii" <eliz@gnu.org>,
> "Mark Kettenis" <mark.kettenis@xs4all.nl>
> > Date: Thursday, May 7, 2009, 8:24 AM
> > On Thu, May 7, 2009 at 02:02, paawan
> > oza <paawan1982@yahoo.com>
> > wrote:
> > >
> > > Hi Hui,
> > >
> > > I have started understanding and working on
> point-1.
> > > I have some questions. please let me know my
> > understanding is correct ?
> > >
> > > 1) i386-tdep.c is the one which has been done
> for
> > record-replay. which basically takes care of
> disassembling
> > insns...
> > > but I see some code regarding MMX and x87FPU
> already
> > available.
> > > may I know exactly, what support is missing ?
> > >
> > 
> >     case 0x0f06:
> >       break;
> > 
> >       /* MMX/SSE/SSE2/PNI support */
> >       /* XXX */
> > 
> >     default:
> > 
> > Your mean is this part?
> > 
> > 
> > And I suggest you can begin with float point insn
> > first.  It doesn't
> > record float point reg change.
> > 
> > 
> > Thanks,
> > Hui
> > 
> 
> 
> 
> 

Re: MMX and floating point insn suport for reversible debugging

[paawan oza]

Hi,

yes.
I guess, MMX support needs to be added here, 

case 0x0f06:
       break;
       /* MMX/SSE/SSE2/PNI support */
       /* XXX */ 

but before I would start adding floating point support.

I have 2 doubts:
-> in function record_arch_list_add_reg
we are doing:
rec->u.reg.val = (gdb_byte *) xmalloc (MAX_REGISTER_SIZE);
we are allocating 128 bytes.....
but I suppose only SSE extensions can fully utilize the memory...
for rest of the registers, are not we allocating more memory than required for record ?

-> SSE support is not there in linux krenel (ptrace), I also read in some piece of comments also.
but SSE support (PT_GETXMMREGS) is suported in openbsd.

with that sense do we need to add SSE/SSE2/SSE3/SSE4 
support in i386-tdep.c generically ?  (of coursse our first priority is to add floating point suport, but I am querying out of my curiousity)

Regards,
Oza.


--- On Thu, 5/7/09, Hui Zhu <teawater@gmail.com> wrote:

> From: Hui Zhu <teawater@gmail.com>
> Subject: Re: MMX and floating point insn suport for reversible debugging
> To: "paawan oza" <paawan1982@yahoo.com>
> Cc: "Thiago Jung Bauermann" <bauerman@br.ibm.com>, "gdb ml" <gdb@sourceware.org>, "Pedro Alves" <pedro@codesourcery.com>, "Marc Khouzam" <marc.khouzam@ericsson.com>, "Michael Snyder" <msnyder@vmware.com>, "Eli Zaretskii" <eliz@gnu.org>, "Mark Kettenis" <mark.kettenis@xs4all.nl>
> Date: Thursday, May 7, 2009, 8:24 AM
> On Thu, May 7, 2009 at 02:02, paawan
> oza <paawan1982@yahoo.com>
> wrote:
> >
> > Hi Hui,
> >
> > I have started understanding and working on point-1.
> > I have some questions. please let me know my
> understanding is correct ?
> >
> > 1) i386-tdep.c is the one which has been done for
> record-replay. which basically takes care of disassembling
> insns...
> > but I see some code regarding MMX and x87FPU already
> available.
> > may I know exactly, what support is missing ?
> >
> 
>     case 0x0f06:
>       break;
> 
>       /* MMX/SSE/SSE2/PNI support */
>       /* XXX */
> 
>     default:
> 
> Your mean is this part?
> 
> 
> And I suggest you can begin with float point insn
> first.  It doesn't
> record float point reg change.
> 
> 
> Thanks,
> Hui
> 

MMX and floating point insn suport for reversible debugging

[paawan oza]

Hi Hui,

I have started understanding and working on point-1. 
I have some questions. please let me know my understanding is correct ?

1) i386-tdep.c is the one which has been done for record-replay. which basically takes care of disassembling insns...
but I see some code regarding MMX and x87FPU already available. 
may I know exactly, what support is missing ?

2) I also see SSE XMM0-XMM7 register stuufs in teh code too.
we need to support XMM (SSE) or it is already done.

please clarify above two points. please clarify to what support is missing and what is expected ?

Anyway, I am going through i386-tdep.c thoroughly to understand it fully.

Regards,
Oza.













      

Re: MMX and floating point insn suport for reversible debugging

[Hui Zhu]

On Thu, May 7, 2009 at 02:02, paawan oza <paawan1982@yahoo.com> wrote:
>
> Hi Hui,
>
> I have started understanding and working on point-1.
> I have some questions. please let me know my understanding is correct ?
>
> 1) i386-tdep.c is the one which has been done for record-replay. which basically takes care of disassembling insns...
> but I see some code regarding MMX and x87FPU already available.
> may I know exactly, what support is missing ?
>

    case 0x0f06:
      break;

      /* MMX/SSE/SSE2/PNI support */
      /* XXX */

    default:

Your mean is this part?


And I suggest you can begin with float point insn first.  It doesn't
record float point reg change.


Thanks,
Hui