[MIPS] Adding binutils changes for MIPS/NaCl architecture.

[MIPS] Adding binutils changes for MIPS/NaCl architecture.

BUG=http://code.google.com/p/nativeclient/issues/detail?id=2275 
TEST=pnacl/build.sh all ;

[petarj, 2012-04-12 00:27:48]

Patch for binutils for MIPS/NaCl.

[Brad Chen, 2012-04-12 21:53:27]

Adding Roland and Derek as reviewers.

[robertm, 2012-04-24 19:04:06]

Sorry for the delay.

For PNaCl we are moving away from using binutils' ld for the final link. 
Instead we are using gold. The switch is already complete for 
static linking which is presumably all we care about for MIPS.
Moreover, there are virtually no changes we made to gold pertaining
to architecture specific things, so it may just work for MIPS as well.

gold gets build in pnacl/build.sh (binutils-gold)
and activated in pnacl/driver/pnacl-nativeld.py (USE_GOLD)

[petarj, 2012-04-24 21:42:41]

GOLD does not support MIPS yet. We are working actively on finishing MIPS port
of GOLD, the first patch is being reviewed by Ian right now, but it still has
not landed.

[robertm, 2012-04-25 20:31:09]

Does it make sense to wait until gold is ready?
Otherwise you have binutils but our driver wont use it.

Another questioned I had is: do you use binutil's gas?
PNaCl has switched over to llvm-mc and I was about to turn
off building gas. 



On 2012/04/24 21:42:41, petarj wrote:
> GOLD does not support MIPS yet. We are working actively on finishing MIPS port
> of GOLD, the first patch is being reviewed by Ian right now, but it still has
> not landed.

[petarj, 2012-04-25 20:54:31]

We should not wait for MIPS gold, unless we want to start applying the changes
for gold that have not been upstreamed and reviewed by Ian. And considering
Ian's internal schedule, reviews cycles are long, so we should still use ld. We
can make the driver use ld for MIPS until gold is ready, can't we?

As of gas, yes, we still use it. llvm-mc assembler is not working on MIPS, yet
it's already on its way, we had some issues with tblgen gen-asm-matcher but
found a way to do it, so sticking to gas instead of llvm-mc is an issue only for
the short term.

[robertm, 2012-04-25 21:40:48]

On 2012/04/25 20:54:31, petarj wrote:
> We should not wait for MIPS gold, unless we want to start applying the changes
> for gold that have not been upstreamed and reviewed by Ian. And considering

Depending on how massive these changes are, I might actually prefer 
this. Do you have a point to those changes?

> Ian's internal schedule, reviews cycles are long, so we should still use ld.
We
> can make the driver use ld for MIPS until gold is ready, can't we?

As you know first hand, our drivers are pretty complex and would benefit from
some simplifications. Having to support both 
ld (with all the linker scripts) and gold would make this worse.

> 
> As of gas, yes, we still use it. llvm-mc assembler is not working on MIPS, yet
> it's already on its way, we had some issues with tblgen gen-asm-matcher but
> found a way to do it, so sticking to gas instead of llvm-mc is an issue only
for
> the short term.

What is the ETA for the MIPS llvm-mc support? Will this make it in before llvm
3.1?
The translation component of pnacl does not really have a assembler phase, so it
would be messy to fit this into driver. How are you currently using the
assembler?

[(google.com) Derek Schuff, 2012-04-25 21:43:08]

> What is the ETA for the MIPS llvm-mc support? Will this make it in before llvm
> 3.1?
I'm pretty sure LLVM 3.1 has already branched.

[petarj, 2012-04-25 22:46:25]

On 2012/04/25 21:40:48, robertm wrote:
> Depending on how massive these changes are, I might actually prefer 
> this. Do you have a point to those changes?

MIPS gold can be seen at:

https://dmz-portal.mips.com/git/?p=gold.git

Support for MIPS includes changes to arch-independent module in gold as well.
The first patch is already on review at binutils mailing list. If we have to
apply all the changes to nacl repository and maintain the branch as we go, fine.
Though having ld for a slightly more time would help though. It's up to you.

> What is the ETA for the MIPS llvm-mc support? Will this make it in before llvm
> 3.1?
> The translation component of pnacl does not really have a assembler phase, so
it
> would be messy to fit this into driver. How are you currently using the
> assembler?

The 3.1 was branched 10 days ago 

http://llvm.org/svn/llvm-project/llvm/branches/release_31

So, the assembler for MIPS will not be in this release.

As of gas, it is used for the validator test data
(e.g. http://codereview.chromium.org/9979025/patch/5001/6024)

We also need assembler for compiling *.S files for secure loader.

[petarj, 2012-04-26 13:55:37]

On 2012/04/25 21:40:48, robertm wrote:
> How are you currently using the
> assembler?

Just to make it clear on the assembler question, PNaCl for MIPS currently uses
two binutils assemblers: 

a) nacl binutils gas which is used for:

- validator test data
- pnacl/support files (setjmp_mipsel.S, tls_get_addr.S) [we used $AS_MIPS in
driver_env.py, similar to $AS_ARM to switch to gas instead of llvm-mc)

b) codesourcery binutils gas which is used for: 

- service runtime (i.e. sel_ldr)

[robertm, 2012-04-26 14:18:30]

On Thu, Apr 26, 2012 at 9:55 AM, <petarj@mips.com> wrote:

> On 2012/04/25 21:40:48, robertm wrote:
>
>> How are you currently using the
>> assembler?
>>
>
> Just to make it clear on the assembler question, PNaCl for MIPS currently
> uses
> two binutils assemblers:
>
> a) nacl binutils gas which is used for:
>
> - validator test data
> - pnacl/support files (setjmp_mipsel.S, tls_get_addr.S) [we used $AS_MIPS
> in
> driver_env.py, similar to $AS_ARM to switch to gas instead of llvm-mc)
>

This is the one we care about. We would probably call this the "untrusted"
assembler.
Is this assembler identical to the "trusted assembler" from codesourcery
(functionality wise) or does it include
additional patches? (I did not see anything in this CL)
What I am really curious about is: is this assembler used in the "path"
from .bc -> .o, i.e. does your llc emit .s files or does it go straight to
.o.



> b) codesourcery binutils gas which is used for:
>
> - service runtime (i.e. sel_ldr)
>
>
>
>
http://codereview.chromium.**org/10020059/<http://codereview.chromium.org/100...
>
> --
> You received this message because you are subscribed to the Google Groups
> "Native-Client-Reviews" group.
> To post to this group, send email to native-client-reviews@**
> googlegroups.com <native-client-reviews@googlegroups.com>.
> To unsubscribe from this group, send email to native-client-reviews+**
>
unsubscribe@googlegroups.com<native-client-reviews%2Bunsubscribe@googlegroups...
> .
> For more options, visit this group at http://groups.google.com/**
>
group/native-client-reviews?**hl=en<http://groups.google.com/group/native-client-reviews?hl=en>
> .
>
>

-- 
You received this message because you are subscribed to the Google Groups
"Native-Client-Reviews" group.
To post to this group, send email to native-client-reviews@googlegroups.com.
To unsubscribe from this group, send email to
native-client-reviews+unsubscribe@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/native-client-reviews?hl=en.

[petarj, 2012-04-26 18:10:51]

On 2012/04/26 14:18:30, robertm wrote:
> What I am really curious about is: is this assembler used in the "path"
> from .bc -> .o, i.e. does your llc emit .s files or does it go straight to
> .o.

We do not emit .s files, we emit directly object files.

[robertm, 2012-04-26 18:27:37]

On Thu, Apr 26, 2012 at 2:10 PM, <petarj@mips.com> wrote:

> On 2012/04/26 14:18:30, robertm wrote:
>
>> What I am really curious about is: is this assembler used in the "path"
>> from .bc -> .o, i.e. does your llc emit .s files or does it go straight to
>> .o.
>>
>
> We do not emit .s files, we emit directly object files.
>


That is great news.
So could you use the trusted gas (from the code sourcery toolchain) for
untrusted purposes as well
until llvm-mc for mips is ready?


>
>
http://codereview.chromium.**org/10020059/<http://codereview.chromium.org/100...
>

-- 
You received this message because you are subscribed to the Google Groups
"Native-Client-Reviews" group.
To post to this group, send email to native-client-reviews@googlegroups.com.
To unsubscribe from this group, send email to
native-client-reviews+unsubscribe@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/native-client-reviews?hl=en.

[robertm, 2012-04-26 18:34:31]

On Wed, Apr 25, 2012 at 6:46 PM, <petarj@mips.com> wrote:

> On 2012/04/25 21:40:48, robertm wrote:
>
>> Depending on how massive these changes are, I might actually prefer
>> this. Do you have a point to those changes?
>>
>
> MIPS gold can be seen at:
>
>
https://dmz-portal.mips.com/**git/?p=gold.git<https://dmz-portal.mips.com/git...



Those changes look isolated enough that I would be fine with them going
into
our local binutils/gold repo.


>
>
> Support for MIPS includes changes to arch-independent module in gold as
> well.
> The first patch is already on review at binutils mailing list. If we have
> to
> apply all the changes to nacl repository and maintain the branch as we go,
> fine.
> Though having ld for a slightly more time would help though. It's up to
> you.
>
>  What is the ETA for the MIPS llvm-mc support? Will this make it in before
>> llvm
>> 3.1?
>> The translation component of pnacl does not really have a assembler
>> phase, so
>>
> it
>
>> would be messy to fit this into driver. How are you currently using the
>> assembler?
>>
>
> The 3.1 was branched 10 days ago
>
>
http://llvm.org/svn/llvm-**project/llvm/branches/release_**31<http://llvm.org...
>
> So, the assembler for MIPS will not be in this release.
>
> As of gas, it is used for the validator test data
> (e.g.
http://codereview.chromium.**org/9979025/patch/5001/6024<http://codereview.ch...
> )
>
> We also need assembler for compiling *.S files for secure loader.
>
>
>
http://codereview.chromium.**org/10020059/<http://codereview.chromium.org/100...
>
> --
> You received this message because you are subscribed to the Google Groups
> "Native-Client-Reviews" group.
> To post to this group, send email to native-client-reviews@**
> googlegroups.com <native-client-reviews@googlegroups.com>.
> To unsubscribe from this group, send email to native-client-reviews+**
>
unsubscribe@googlegroups.com<native-client-reviews%2Bunsubscribe@googlegroups...
> .
> For more options, visit this group at http://groups.google.com/**
>
group/native-client-reviews?**hl=en<http://groups.google.com/group/native-client-reviews?hl=en>
> .
>
>

-- 
You received this message because you are subscribed to the Google Groups
"Native-Client-Reviews" group.
To post to this group, send email to native-client-reviews@googlegroups.com.
To unsubscribe from this group, send email to
native-client-reviews+unsubscribe@googlegroups.com.
For more options, visit this group at
http://groups.google.com/group/native-client-reviews?hl=en.

[petarj, 2012-04-27 19:08:41]

On 2012/04/26 18:34:31, robertm wrote:
> Those changes look isolated enough that I would be fine with them going
> into
> our local binutils/gold repo.

Excellent. I have just uploaded them at:

http://codereview.chromium.org/10252012/

[petarj, 2012-04-27 23:55:44]

On 2012/04/26 18:27:37, robertm wrote:
> That is great news.
> So could you use the trusted gas (from the code sourcery toolchain) for
> untrusted purposes as well
> until llvm-mc for mips is ready?

I guess we could.