Sun/Oracle C++ compiler patch

Sun/Oracle C++ compiler patch

[Chris Quenelle]

(ignore the last copy of this email, it had a terrible subject line)

Hello,

When the Sun/Oracle C++ compiler was ported to Linux, we started
bundling a patched version of gnu ld to get the necessary treatment
for our exception range sections.  I don't believe anyone has tried to
offer this patch upstream, and it would really help us out if we could
use the system linker when running on Linux.  I've included the
contents of the patch at the end of this email.

--chris



% more patch-intel-Linux-2.17.90
*** binutils-2.17.90/ld/scripttempl/elf.sc      2007-08-07 
00:00:22.000000000 +0400
--- bu-patched/ld/scripttempl/elf.sc    2008-06-06 15:08:24.602615680 +0400
***************
*** 372,377 ****
--- 372,378 ----
     .eh_frame_hdr : { *(.eh_frame_hdr) }
     .eh_frame     ${RELOCATING-0} : ONLY_IF_RO { KEEP (*(.eh_frame)) }
     .gcc_except_table ${RELOCATING-0} : ONLY_IF_RO { 
*(.gcc_except_table .gcc_except_table.*) }
+   .exception_ranges ${RELOCATING-0} : ONLY_IF_RO { *(.exception_ranges 
.exception_ranges*) }

     /* Adjust the address for the data segment.  We want to adjust up to
        the same address within the page on the next page up.  */
***************
*** 382,387 ****
--- 383,389 ----
     /* Exception handling  */
     .eh_frame     ${RELOCATING-0} : ONLY_IF_RW { KEEP (*(.eh_frame)) }
     .gcc_except_table ${RELOCATING-0} : ONLY_IF_RW { 
*(.gcc_except_table .gcc_except_table.*) }
+   .exception_ranges ${RELOCATING-0} : ONLY_IF_RW { *(.exception_ranges 
.exception_ranges*) }

     /* Thread Local Storage sections  */
     .tdata      ${RELOCATING-0} : { *(.tdata${RELOCATING+ .tdata.* 
.gnu.linkonce.td.*}) }

Re: Sun/Oracle C++ compiler patch

[Tristan Gingold]

On Feb 25, 2011, at 1:17 AM, Chris Quenelle wrote:

> (ignore the last copy of this email, it had a terrible subject line)
> 
> Hello,
> 
> When the Sun/Oracle C++ compiler was ported to Linux, we started
> bundling a patched version of gnu ld to get the necessary treatment
> for our exception range sections.  I don't believe anyone has tried to
> offer this patch upstream, and it would really help us out if we could
> use the system linker when running on Linux.  I've included the
> contents of the patch at the end of this email.

Just a suggestion: can you add a comment just before to explain that this is for Sun/Oracle C++ compiler ?  This is not
obvious from the section name.

Is the content of this section documented somewhere ?

Tristan.

Re: Sun/Oracle C++ compiler patch

[Jakub Jelinek]

On Fri, Feb 25, 2011 at 10:54:32AM +0100, Tristan Gingold wrote:
> 
> On Feb 25, 2011, at 1:17 AM, Chris Quenelle wrote:
> 
> > (ignore the last copy of this email, it had a terrible subject line)
> > 
> > Hello,
> > 
> > When the Sun/Oracle C++ compiler was ported to Linux, we started
> > bundling a patched version of gnu ld to get the necessary treatment
> > for our exception range sections.  I don't believe anyone has tried to
> > offer this patch upstream, and it would really help us out if we could
> > use the system linker when running on Linux.  I've included the
> > contents of the patch at the end of this email.
> 
> Just a suggestion: can you add a comment just before to explain that this is for Sun/Oracle C++ compiler ?  This is not
> obvious from the section name.
> 
> Is the content of this section documented somewhere ?

Also, do you really need ONLY_IF_R{O,W}, i.e. do some older Oracle C++ compiler versions
emit the section writable and some later compilers emit it read-only (or vice versa)?
.eh_frame with very old gcc versions used to be a writable sections that needed
runtime relocation, then gcc changed to a new format which doesn't need any relocations
and thus it is desirable to put the section into a read-only segment if all
.eh_frame input sections are read-only.

	Jakub

Re: Sun/Oracle C++ compiler patch

[Chris Quenelle]

On Friday February 25    4:19AM, Jakub Jelinek wrote:
> On Fri, Feb 25, 2011 at 10:54:32AM +0100, Tristan Gingold wrote:
>> On Feb 25, 2011, at 1:17 AM, Chris Quenelle wrote:
>>
>>> (ignore the last copy of this email, it had a terrible subject line)
>>>
>>> Hello,
>>>
>>> When the Sun/Oracle C++ compiler was ported to Linux, we started
>>> bundling a patched version of gnu ld to get the necessary treatment
>>> for our exception range sections.  I don't believe anyone has tried to
>>> offer this patch upstream, and it would really help us out if we could
>>> use the system linker when running on Linux.  I've included the
>>> contents of the patch at the end of this email.
>> Just a suggestion: can you add a comment just before to explain that this is for Sun/Oracle C++ compiler ?  This is not
>> obvious from the section name.
>>
>> Is the content of this section documented somewhere ?
> Also, do you really need ONLY_IF_R{O,W}, i.e. do some older Oracle C++ compiler versions
> emit the section writable and some later compilers emit it read-only (or vice versa)?
> .eh_frame with very old gcc versions used to be a writable sections that needed
> runtime relocation, then gcc changed to a new format which doesn't need any relocations
> and thus it is desirable to put the section into a read-only segment if all
> .eh_frame input sections are read-only.
>
> 	Jakub


Jakub: I believe our fix was made by just copying the way 
gcc_except_table is treated.
Our sections see to be writable in the a.out, I've emailed our C++ team 
to ask if we do
runtime relocations.  (ugh)

Tristan: I'll add a comment and repost the patch when I resolve Jakub's 
question.

Thanks!

--chris

Re: Sun/Oracle C++ compiler patch

[Chris Quenelle]

On Friday February 25    4:19AM, Jakub Jelinek wrote:
> On Fri, Feb 25, 2011 at 10:54:32AM +0100, Tristan Gingold wrote:
>> On Feb 25, 2011, at 1:17 AM, Chris Quenelle wrote:
>>
>>> (ignore the last copy of this email, it had a terrible subject line)
>>>
>>> Hello,
>>>
>>> When the Sun/Oracle C++ compiler was ported to Linux, we started
>>> bundling a patched version of gnu ld to get the necessary treatment
>>> for our exception range sections.  I don't believe anyone has tried to
>>> offer this patch upstream, and it would really help us out if we could
>>> use the system linker when running on Linux.  I've included the
>>> contents of the patch at the end of this email.
>> Just a suggestion: can you add a comment just before to explain that this is for Sun/Oracle C++ compiler ?  This is not
>> obvious from the section name.
>>
>> Is the content of this section documented somewhere ?
> Also, do you really need ONLY_IF_R{O,W}, i.e. do some older Oracle C++ compiler versions
> emit the section writable and some later compilers emit it read-only (or vice versa)?
> .eh_frame with very old gcc versions used to be a writable sections that needed
> runtime relocation, then gcc changed to a new format which doesn't need any relocations
> and thus it is desirable to put the section into a read-only segment if all
> .eh_frame input sections are read-only.
>
> 	Jakub

Sorry for the delay in this thread.

I'm not really sure that the "ONLY_IF_RW" in the linker script maps onto 
the sh_flags
field having the SHF_WRITE bit turned on in the section header table.  
Could you confirm that?
Currently our exception range sections are writable, but I haven't been 
able to confirm
that's necessary because of fixups by the C++ runtime system.  If it's 
not necessary or
if we change it later, it would be nice if the linker didn't need more 
modification.

Is there any harm in allowing the same behavior if our sections later 
become read-only?

Because of the time-lapse here, I'll include the patch again so that 
people have context
for the thread.

--chris

% more patch-intel-Linux-2.17.90
*** binutils-2.17.90/ld/scripttempl/elf.sc      2007-08-07 
00:00:22.000000000 +0400
--- bu-patched/ld/scripttempl/elf.sc    2008-06-06 15:08:24.602615680 +0400
***************
*** 372,377 ****
--- 372,378 ----
     .eh_frame_hdr : { *(.eh_frame_hdr) }
     .eh_frame     ${RELOCATING-0} : ONLY_IF_RO { KEEP (*(.eh_frame)) }
     .gcc_except_table ${RELOCATING-0} : ONLY_IF_RO { 
*(.gcc_except_table .gcc_except_table.*) }
+   .exception_ranges ${RELOCATING-0} : ONLY_IF_RO { *(.exception_ranges 
.exception_ranges*) }

     /* Adjust the address for the data segment.  We want to adjust up to
        the same address within the page on the next page up.  */
***************
*** 382,387 ****
--- 383,389 ----
     /* Exception handling  */
     .eh_frame     ${RELOCATING-0} : ONLY_IF_RW { KEEP (*(.eh_frame)) }
     .gcc_except_table ${RELOCATING-0} : ONLY_IF_RW { 
*(.gcc_except_table .gcc_except_table.*) }
+   .exception_ranges ${RELOCATING-0} : ONLY_IF_RW { *(.exception_ranges 
.exception_ranges*) }

     /* Thread Local Storage sections  */
     .tdata      ${RELOCATING-0} : { *(.tdata${RELOCATING+ .tdata.* 
.gnu.linkonce.td.*}) }

Re: Sun/Oracle C++ compiler patch

[Nick Clifton]

Hi Chris,

> I'm not really sure that the "ONLY_IF_RW" in the linker script maps onto
> the sh_flags field having the SHF_WRITE bit turned on in the section
> header table.  Could you confirm that?

It does.

> Currently our exception range sections are writable, but I haven't been
> able to confirm that's necessary because of fixups by the C++ runtime
> system. If it's not necessary or if we change it later, it would be
> nice if the linker didn't need more modification.

Agreed.

> Is there any harm in allowing the same behavior if our sections later
> become read-only?

None.

> .gcc_except_table ${RELOCATING-0} : ONLY_IF_RO { *(.gcc_except_table
> .gcc_except_table.*) }
> + .exception_ranges ${RELOCATING-0} : ONLY_IF_RO { *(.exception_ranges
> .exception_ranges*) }

> .gcc_except_table ${RELOCATING-0} : ONLY_IF_RW { *(.gcc_except_table
> .gcc_except_table.*) }
> + .exception_ranges ${RELOCATING-0} : ONLY_IF_RW { *(.exception_ranges
> .exception_ranges*) }

The patch looks OK to me, and so I have checked it in along with this 
changelog entry.  I did take Tristan's suggestion however and so I added 
a comment explaining where the .exception_ranges section originates.

Cheers
   Nick

ld/ChangeLog
2011-04-11  Chris Quenelle  <chris.quenelle@oracle.com>

	* scripttempl/elf.sc (.exception_ranges): Add new section.

Re: Sun/Oracle C++ compiler patch

[Chris Quenelle]

Thank you very much for your help, I appreciate it.


On Monday April 11    8:40AM, Nick Clifton wrote:
>
> The patch looks OK to me, and so I have checked it in along with this 
> changelog entry.  I did take Tristan's suggestion however and so I 
> added a comment explaining where the .exception_ranges section 
> originates.
>
> Cheers
>   Nick
>
> ld/ChangeLog
> 2011-04-11  Chris Quenelle <chris.quenelle@oracle.com>
>
>     * scripttempl/elf.sc (.exception_ranges): Add new section.
>