Re: [PATCH] x86 CFI extensions

Re: [PATCH] x86 CFI extensions

[Jan Beulich]

>>> Mark Kettenis <mark.kettenis@xs4all.nl> 09/21/05 10:02 PM >>>
>This is ok.  However, I don't see your name listed in gdb/MAINTAINERS.
>Do you have a valid copyright assignment for GDB?

From all I can tell I do (for binutils, gcc, and gdb). I didn't care to
add myself to gdb/MAINTAINERS, as there's really not a lot I'm currently
doing with/for it specifically (binutils changes to opcodes/ certainly
affect it, though).

So, am are you OK with me going ahead and committing the patch?

Jan

Re: [PATCH] x86 CFI extensions

[Eli Zaretskii]

> Date: Fri, 23 Sep 2005 09:40:15 +0200
> From: "Jan Beulich" <jbeulich@novell.com>
> Cc: <drow@false.org>,<gdb-patches@sourceware.org>, <rth@twiddle.net>
> 
> >>> Mark Kettenis <mark.kettenis@xs4all.nl> 09/21/05 10:02 PM >>>
> >This is ok.  However, I don't see your name listed in gdb/MAINTAINERS.
> >Do you have a valid copyright assignment for GDB?
> 
> >From all I can tell I do (for binutils, gcc, and gdb).

FWIW, I don't see your name on file in the FSF list of assignments.  I
just looked in the latest version of that file.

Re: [PATCH] x86 CFI extensions

[Daniel Jacobowitz]

On Fri, Sep 23, 2005 at 11:02:26AM +0300, Eli Zaretskii wrote:
> > Date: Fri, 23 Sep 2005 09:40:15 +0200
> > From: "Jan Beulich" <jbeulich@novell.com>
> > Cc: <drow@false.org>,<gdb-patches@sourceware.org>, <rth@twiddle.net>
> > 
> > >>> Mark Kettenis <mark.kettenis@xs4all.nl> 09/21/05 10:02 PM >>>
> > >This is ok.  However, I don't see your name listed in gdb/MAINTAINERS.
> > >Do you have a valid copyright assignment for GDB?
> > 
> > >From all I can tell I do (for binutils, gcc, and gdb).
> 
> FWIW, I don't see your name on file in the FSF list of assignments.  I
> just looked in the latest version of that file.

I think it's OK: Novell has a corporate assignment on file for GDB,
GCC, and binutils.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: [PATCH] x86 CFI extensions

[Daniel Jacobowitz]

On Fri, Sep 23, 2005 at 09:40:15AM +0200, Jan Beulich wrote:
> >>> Mark Kettenis <mark.kettenis@xs4all.nl> 09/21/05 10:02 PM >>>
> >This is ok.  However, I don't see your name listed in gdb/MAINTAINERS.
> >Do you have a valid copyright assignment for GDB?
> 
> From all I can tell I do (for binutils, gcc, and gdb). I didn't care to
> add myself to gdb/MAINTAINERS, as there's really not a lot I'm currently
> doing with/for it specifically (binutils changes to opcodes/ certainly
> affect it, though).
> 
> So, am are you OK with me going ahead and committing the patch?

Could you please (A) commit the patch, and (B) commit and post a patch
adding yourself to MAINTAINERS?  I realize you don't intend to do
active GDB development; that's OK, most of the other people who've
ended up in that list don't either.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: [PATCH] x86 CFI extensions

[Jan Beulich]

>>> Daniel Jacobowitz <drow@false.org> 26.09.05 03:24:18 >>>
>On Fri, Sep 23, 2005 at 09:40:15AM +0200, Jan Beulich wrote:
>> >>> Mark Kettenis <mark.kettenis@xs4all.nl> 09/21/05 10:02 PM >>>
>> >This is ok.  However, I don't see your name listed in gdb/MAINTAINERS.
>> >Do you have a valid copyright assignment for GDB?
>> 
>> From all I can tell I do (for binutils, gcc, and gdb). I didn't care to
>> add myself to gdb/MAINTAINERS, as there's really not a lot I'm currently
>> doing with/for it specifically (binutils changes to opcodes/ certainly
>> affect it, though).
>> 
>> So, am are you OK with me going ahead and committing the patch?
>
>Could you please (A) commit the patch, and (B) commit and post a patch
>adding yourself to MAINTAINERS?  I realize you don't intend to do
>active GDB development; that's OK, most of the other people who've
>ended up in that list don't either.

Below/Done.

Jan

gdb/
2005-09-26  Jan Beulich  <jbeulich@novell.com>

	* MAINTAINERS (Write After Approval): Add myself.

--- /home/jbeulich/src/gdb/mainline/2005-09-13/gdb/MAINTAINERS	2005-08-18 08:51:36.000000000 +0200
+++ 2005-09-13/gdb/MAINTAINERS	2005-09-26 08:59:37.421637560 +0200
@@ -301,6 +301,7 @@ David Anderson					davea@sgi.com 
 John David Anglin				dave.anglin@nrc-cnrc.gc.ca 
 Shrinivas Atre					shrinivasa@kpitcummins.com 
 Scott Bambrough					scottb@netwinder.org 
+Jan Beulich					jbeulich@novell.com 
 Jim Blandy					jimb@redhat.com 
 Philip Blundell					philb@gnu.org 
 Per Bothner					per@bothner.com 

[PATCH] x86 CFI extensions

[Jan Beulich]

[-- Attachment #1: Type: text/plain, Size: 3521 bytes --]

>>> Richard Henderson <rth@twiddle.net> 12.09.05 09:49:14 >>>
>On Mon, Sep 12, 2005 at 09:04:12AM +0200, Jan Beulich wrote:
>> But truly I think the processor-specific pieces of Dwarf's
>> frame unwind spec should provide numbering for the complete set of
>> registers.
>
>Except there is no standards body for this.  So *someone* will
>have to make it up.
>
>Make it up and put it in gas and gdb: that will make it a defacto
standard.

This adds to gdb a little more than the minimum set of registers
needed
to at least fully understand CFI annotations for Linux' pt_regs.

Built and tested on i686-pc-linux-gnu and x86_64-unknown-linux-gnu.

Jan

gdb/
2005-09-13  Jan Beulich  <jbeulich@novell.com>

	* amd64-tdep.h (AMD64_FCTRL_REGNUM, AMD64_FSTAT_REGNUM,
	AMD64_MXCSR_REGNUM): New.
	* amd64-tdep.c (amd64_dwarf_regmap): Add eflags, selector regs,
	mxcsr, fp control and status words.
	* i386-tdep.c (): Add selector regs, mxcsr, fp control and
status
	words.

---
/home/jbeulich/src/gdb/mainline/2005-09-13/gdb/amd64-tdep.c	2005-07-05
09:09:52.000000000 +0200
+++ 2005-09-13/gdb/amd64-tdep.c	2005-09-13 17:50:00.723239944
+0200
@@ -212,7 +212,35 @@ static int amd64_dwarf_regmap[] =
   AMD64_ST0_REGNUM + 0, AMD64_ST0_REGNUM + 1,
   AMD64_ST0_REGNUM + 2, AMD64_ST0_REGNUM + 3,
   AMD64_ST0_REGNUM + 4, AMD64_ST0_REGNUM + 5,
-  AMD64_ST0_REGNUM + 6, AMD64_ST0_REGNUM + 7
+  AMD64_ST0_REGNUM + 6, AMD64_ST0_REGNUM + 7,
+  
+  /* Control and Status Flags Register.  */
+  AMD64_EFLAGS_REGNUM,
+
+  /* Selector Registers.  */
+  AMD64_ES_REGNUM,
+  AMD64_CS_REGNUM,
+  AMD64_SS_REGNUM,
+  AMD64_DS_REGNUM,
+  AMD64_FS_REGNUM,
+  AMD64_GS_REGNUM,
+  -1,
+  -1,
+
+  /* Segment Base Address Registers.  */
+  -1,
+  -1,
+  -1,
+  -1,
+
+  /* Special Selector Registers.  */
+  -1,
+  -1,
+
+  /* Floating Point Control Registers.  */
+  AMD64_MXCSR_REGNUM,
+  AMD64_FCTRL_REGNUM,
+  AMD64_FSTAT_REGNUM
 };
 
 static const int amd64_dwarf_regmap_len =
---
/home/jbeulich/src/gdb/mainline/2005-09-13/gdb/amd64-tdep.h	2004-04-12
18:17:20.000000000 +0200
+++ 2005-09-13/gdb/amd64-tdep.h	2005-09-13 15:41:55.000000000
+0200
@@ -52,8 +52,11 @@ enum amd64_regnum
   AMD64_FS_REGNUM,		/* %fs */
   AMD64_GS_REGNUM,		/* %gs */
   AMD64_ST0_REGNUM = 24,	/* %st0 */
+  AMD64_FCTRL_REGNUM = AMD64_ST0_REGNUM + 8,
+  AMD64_FSTAT_REGNUM = AMD64_ST0_REGNUM + 9,
   AMD64_XMM0_REGNUM = 40,	/* %xmm0 */
-  AMD64_XMM1_REGNUM		/* %xmm1 */
+  AMD64_XMM1_REGNUM,		/* %xmm1 */
+  AMD64_MXCSR_REGNUM = AMD64_XMM0_REGNUM + 16
 };
 
 /* Number of general purpose registers.  */
---
/home/jbeulich/src/gdb/mainline/2005-09-13/gdb/i386-tdep.c	2005-07-26
08:26:49.000000000 +0200
+++ 2005-09-13/gdb/i386-tdep.c	2005-09-13 11:46:18.000000000 +0200
@@ -225,12 +225,25 @@ i386_svr4_reg_to_regnum (int reg)
       /* Floating-point registers.  */
       return reg - 11 + I387_ST0_REGNUM;
     }
-  else if (reg >= 21)
+  else if (reg >= 21 && reg <= 36)
     {
       /* The SSE and MMX registers have the same numbers as with dbx. 
*/
       return i386_dbx_reg_to_regnum (reg);
     }
 
+  switch (reg)
+    {
+    case 37: return I387_FCTRL_REGNUM;
+    case 38: return I387_FSTAT_REGNUM;
+    case 39: return I387_MXCSR_REGNUM;
+    case 40: return I386_ES_REGNUM;
+    case 41: return I386_CS_REGNUM;
+    case 42: return I386_SS_REGNUM;
+    case 43: return I386_DS_REGNUM;
+    case 44: return I386_FS_REGNUM;
+    case 45: return I386_GS_REGNUM;
+    }
+
   /* This will hopefully provoke a warning.  */
   return NUM_REGS + NUM_PSEUDO_REGS;
 }


[-- Attachment #2: gdb-mainline-x86-cfi-extensions.patch --]
[-- Type: application/octet-stream, Size: 3083 bytes --]

This adds to gdb a little more than the minimum set of registers needed
to at least fully understand CFI annotations for Linux' pt_regs.

Built and tested on i686-pc-linux-gnu and x86_64-unknown-linux-gnu.

Jan

gdb/
2005-09-13  Jan Beulich  <jbeulich@novell.com>

	* amd64-tdep.h (AMD64_FCTRL_REGNUM, AMD64_FSTAT_REGNUM,
	AMD64_MXCSR_REGNUM): New.
	* amd64-tdep.c (amd64_dwarf_regmap): Add eflags, selector regs,
	mxcsr, fp control and status words.
	* i386-tdep.c (): Add selector regs, mxcsr, fp control and status
	words.

--- /home/jbeulich/src/gdb/mainline/2005-09-13/gdb/amd64-tdep.c	2005-07-05 09:09:52.000000000 +0200
+++ 2005-09-13/gdb/amd64-tdep.c	2005-09-13 17:50:00.723239944 +0200
@@ -212,7 +212,35 @@ static int amd64_dwarf_regmap[] =
   AMD64_ST0_REGNUM + 0, AMD64_ST0_REGNUM + 1,
   AMD64_ST0_REGNUM + 2, AMD64_ST0_REGNUM + 3,
   AMD64_ST0_REGNUM + 4, AMD64_ST0_REGNUM + 5,
-  AMD64_ST0_REGNUM + 6, AMD64_ST0_REGNUM + 7
+  AMD64_ST0_REGNUM + 6, AMD64_ST0_REGNUM + 7,
+  
+  /* Control and Status Flags Register.  */
+  AMD64_EFLAGS_REGNUM,
+
+  /* Selector Registers.  */
+  AMD64_ES_REGNUM,
+  AMD64_CS_REGNUM,
+  AMD64_SS_REGNUM,
+  AMD64_DS_REGNUM,
+  AMD64_FS_REGNUM,
+  AMD64_GS_REGNUM,
+  -1,
+  -1,
+
+  /* Segment Base Address Registers.  */
+  -1,
+  -1,
+  -1,
+  -1,
+
+  /* Special Selector Registers.  */
+  -1,
+  -1,
+
+  /* Floating Point Control Registers.  */
+  AMD64_MXCSR_REGNUM,
+  AMD64_FCTRL_REGNUM,
+  AMD64_FSTAT_REGNUM
 };
 
 static const int amd64_dwarf_regmap_len =
--- /home/jbeulich/src/gdb/mainline/2005-09-13/gdb/amd64-tdep.h	2004-04-12 18:17:20.000000000 +0200
+++ 2005-09-13/gdb/amd64-tdep.h	2005-09-13 15:41:55.000000000 +0200
@@ -52,8 +52,11 @@ enum amd64_regnum
   AMD64_FS_REGNUM,		/* %fs */
   AMD64_GS_REGNUM,		/* %gs */
   AMD64_ST0_REGNUM = 24,	/* %st0 */
+  AMD64_FCTRL_REGNUM = AMD64_ST0_REGNUM + 8,
+  AMD64_FSTAT_REGNUM = AMD64_ST0_REGNUM + 9,
   AMD64_XMM0_REGNUM = 40,	/* %xmm0 */
-  AMD64_XMM1_REGNUM		/* %xmm1 */
+  AMD64_XMM1_REGNUM,		/* %xmm1 */
+  AMD64_MXCSR_REGNUM = AMD64_XMM0_REGNUM + 16
 };
 
 /* Number of general purpose registers.  */
--- /home/jbeulich/src/gdb/mainline/2005-09-13/gdb/i386-tdep.c	2005-07-26 08:26:49.000000000 +0200
+++ 2005-09-13/gdb/i386-tdep.c	2005-09-13 11:46:18.000000000 +0200
@@ -225,12 +225,25 @@ i386_svr4_reg_to_regnum (int reg)
       /* Floating-point registers.  */
       return reg - 11 + I387_ST0_REGNUM;
     }
-  else if (reg >= 21)
+  else if (reg >= 21 && reg <= 36)
     {
       /* The SSE and MMX registers have the same numbers as with dbx.  */
       return i386_dbx_reg_to_regnum (reg);
     }
 
+  switch (reg)
+    {
+    case 37: return I387_FCTRL_REGNUM;
+    case 38: return I387_FSTAT_REGNUM;
+    case 39: return I387_MXCSR_REGNUM;
+    case 40: return I386_ES_REGNUM;
+    case 41: return I386_CS_REGNUM;
+    case 42: return I386_SS_REGNUM;
+    case 43: return I386_DS_REGNUM;
+    case 44: return I386_FS_REGNUM;
+    case 45: return I386_GS_REGNUM;
+    }
+
   /* This will hopefully provoke a warning.  */
   return NUM_REGS + NUM_PSEUDO_REGS;
 }

Re: [PATCH] x86 CFI extensions

[Daniel Jacobowitz]

On Tue, Sep 13, 2005 at 06:18:03PM +0200, Jan Beulich wrote:
> >>> Richard Henderson <rth@twiddle.net> 12.09.05 09:49:14 >>>
> >On Mon, Sep 12, 2005 at 09:04:12AM +0200, Jan Beulich wrote:
> >> But truly I think the processor-specific pieces of Dwarf's
> >> frame unwind spec should provide numbering for the complete set of
> >> registers.
> >
> >Except there is no standards body for this.  So *someone* will
> >have to make it up.
> >
> >Make it up and put it in gas and gdb: that will make it a defacto
> standard.
> 
> This adds to gdb a little more than the minimum set of registers
> needed
> to at least fully understand CFI annotations for Linux' pt_regs.
> 
> Built and tested on i686-pc-linux-gnu and x86_64-unknown-linux-gnu.
> 
> Jan
> 
> gdb/
> 2005-09-13  Jan Beulich  <jbeulich@novell.com>
> 
> 	* amd64-tdep.h (AMD64_FCTRL_REGNUM, AMD64_FSTAT_REGNUM,
> 	AMD64_MXCSR_REGNUM): New.
> 	* amd64-tdep.c (amd64_dwarf_regmap): Add eflags, selector regs,
> 	mxcsr, fp control and status words.
> 	* i386-tdep.c (): Add selector regs, mxcsr, fp control and
> status
> 	words.

Mark, did you want to take a look at this?  It seems fine to me.

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: [PATCH] x86 CFI extensions

[Mark Kettenis]

> Date: Sat, 17 Sep 2005 16:59:39 -0400
> From: Daniel Jacobowitz <drow@false.org>
> 
> > 2005-09-13  Jan Beulich  <jbeulich@novell.com>
> > 
> > 	* amd64-tdep.h (AMD64_FCTRL_REGNUM, AMD64_FSTAT_REGNUM,
> > 	AMD64_MXCSR_REGNUM): New.
> > 	* amd64-tdep.c (amd64_dwarf_regmap): Add eflags, selector regs,
> > 	mxcsr, fp control and status words.
> > 	* i386-tdep.c (): Add selector regs, mxcsr, fp control and
> > status
> > 	words.
> 
> Mark, did you want to take a look at this?  It seems fine to me.

Sorry, slipped through.  Thanks for reminding me Daniel!

Jan, what's the status new DWARF register numbers?  At least the amd64
ones look very Linux-specific to me.  They're not listed in the AMD64
ABI (at least not version 0.95), and AFAICT gcc won't produce them
either.

Mark

Re: [PATCH] x86 CFI extensions

[Daniel Jacobowitz]

On Sun, Sep 18, 2005 at 12:31:44AM +0200, Mark Kettenis wrote:
> > Date: Sat, 17 Sep 2005 16:59:39 -0400
> > From: Daniel Jacobowitz <drow@false.org>
> > 
> > > 2005-09-13  Jan Beulich  <jbeulich@novell.com>
> > > 
> > > 	* amd64-tdep.h (AMD64_FCTRL_REGNUM, AMD64_FSTAT_REGNUM,
> > > 	AMD64_MXCSR_REGNUM): New.
> > > 	* amd64-tdep.c (amd64_dwarf_regmap): Add eflags, selector regs,
> > > 	mxcsr, fp control and status words.
> > > 	* i386-tdep.c (): Add selector regs, mxcsr, fp control and
> > > status
> > > 	words.
> > 
> > Mark, did you want to take a look at this?  It seems fine to me.
> 
> Sorry, slipped through.  Thanks for reminding me Daniel!
> 
> Jan, what's the status new DWARF register numbers?  At least the amd64
> ones look very Linux-specific to me.  They're not listed in the AMD64
> ABI (at least not version 0.95), and AFAICT gcc won't produce them
> either.

They _weren't_ listed in the AMD64 ABI: they are now :-)  Michael Matz
committed a patch to do this on Tuesday.  I don't think a new PDF's
been made since, but you can find the CVS archive on x86-64.org.

GCC won't use them because GCC never puts anything in those registers
that it emits debug or unwind data for, to the best of my knowledge.

Is there any active authority for the DWARF numbering on i386?

-- 
Daniel Jacobowitz
CodeSourcery, LLC

Re: [PATCH] x86 CFI extensions

[Jan Beulich]

>>> Mark Kettenis <mark.kettenis@xs4all.nl> 18.09.05 00:31:44 >>>
>> Date: Sat, 17 Sep 2005 16:59:39 -0400
>> From: Daniel Jacobowitz <drow@false.org>
>> 
>> > 2005-09-13  Jan Beulich  <jbeulich@novell.com>
>> > 
>> > 	* amd64-tdep.h (AMD64_FCTRL_REGNUM, AMD64_FSTAT_REGNUM,
>> > 	AMD64_MXCSR_REGNUM): New.
>> > 	* amd64-tdep.c (amd64_dwarf_regmap): Add eflags, selector regs,
>> > 	mxcsr, fp control and status words.
>> > 	* i386-tdep.c (): Add selector regs, mxcsr, fp control and
>> > status
>> > 	words.
>> 
>> Mark, did you want to take a look at this?  It seems fine to me.
>
>Sorry, slipped through.  Thanks for reminding me Daniel!
>
>Jan, what's the status new DWARF register numbers?  At least the
amd64
>ones look very Linux-specific to me.  They're not listed in the AMD64
>ABI (at least not version 0.95), and AFAICT gcc won't produce them
>either.

In addition to Daniel's reply: while gcc will probably never care about
them (and thus never produce them), the gas now can (patch went in last
week), and while initially this indeed is for Linux, any OS should want
them for describing exception and interrupt frames...

Jan

Re: [PATCH] x86 CFI extensions

[Mark Kettenis]

> Date: Mon, 19 Sep 2005 09:27:47 +0200
> From: "Jan Beulich" <JBeulich@novell.com>
> 
> >>> Mark Kettenis <mark.kettenis@xs4all.nl> 18.09.05 00:31:44 >>>
> >> Date: Sat, 17 Sep 2005 16:59:39 -0400
> >> From: Daniel Jacobowitz <drow@false.org>
> >> 
> >> > 2005-09-13  Jan Beulich  <jbeulich@novell.com>
> >> > 
> >> > 	* amd64-tdep.h (AMD64_FCTRL_REGNUM, AMD64_FSTAT_REGNUM,
> >> > 	AMD64_MXCSR_REGNUM): New.
> >> > 	* amd64-tdep.c (amd64_dwarf_regmap): Add eflags, selector regs,
> >> > 	mxcsr, fp control and status words.
> >> > 	* i386-tdep.c (): Add selector regs, mxcsr, fp control and
> >> > status
> >> > 	words.
> >> 
> >> Mark, did you want to take a look at this?  It seems fine to me.
> >
> >Sorry, slipped through.  Thanks for reminding me Daniel!
> >
> >Jan, what's the status new DWARF register numbers?  At least the
> amd64
> >ones look very Linux-specific to me.  They're not listed in the AMD64
> >ABI (at least not version 0.95), and AFAICT gcc won't produce them
> >either.
> 
> In addition to Daniel's reply: while gcc will probably never care about
> them (and thus never produce them), the gas now can (patch went in last
> week), and while initially this indeed is for Linux, any OS should want
> them for describing exception and interrupt frames...
> 

This is ok.  However, I don't see your name listed in gdb/MAINTAINERS.
Do you have a valid copyright assignment for GDB?

Mark

Re: [PATCH] x86 CFI extensions

[Richard Henderson]

On Wed, Sep 21, 2005 at 10:02:02PM +0200, Mark Kettenis wrote:
> This is ok.  However, I don't see your name listed in gdb/MAINTAINERS.

Line 337.

> Do you have a valid copyright assignment for GDB?

Of course.  Blanket assignment from Red Hat.


r~

Re: [PATCH] x86 CFI extensions

[Daniel Jacobowitz]

On Thu, Sep 22, 2005 at 01:21:41AM -0700, Richard Henderson wrote:
> On Wed, Sep 21, 2005 at 10:02:02PM +0200, Mark Kettenis wrote:
> > This is ok.  However, I don't see your name listed in gdb/MAINTAINERS.
> 
> Line 337.
> 
> > Do you have a valid copyright assignment for GDB?
> 
> Of course.  Blanket assignment from Red Hat.

Wrong thread :-)  He was talking to Jan...

-- 
Daniel Jacobowitz
CodeSourcery, LLC