[PATCH] Handle Initial-Exec to Local-Exec for x32

[PATCH] Handle Initial-Exec to Local-Exec for x32

[H.J. Lu]

Hi,

This patch handls Handle Initial-Exec to Local-Exec for x32.  OK
to install?

Thanks.


H.J.
---
	PR gold/17809
	* x86_64.cc (Target_x86_64<size>::Relocate::tls_ie_to_le): Handle
	x32.
---
 gold/ChangeLog | 6 ++++++
 gold/x86_64.cc | 6 ++++++
 2 files changed, 12 insertions(+)

diff --git a/gold/ChangeLog b/gold/ChangeLog
index 496d9a7..af56066 100644
--- a/gold/ChangeLog
+++ b/gold/ChangeLog
@@ -1,3 +1,9 @@
+2015-01-06  H.J. Lu  <hongjiu.lu@intel.com>
+
+	PR gold/17809
+	* x86_64.cc (Target_x86_64<size>::Relocate::tls_ie_to_le): Handle
+	x32.
+
 2015-01-02  Alan Modra  <amodra@gmail.com>
 
 	* version.cc (print_version): Just print current year.
diff --git a/gold/x86_64.cc b/gold/x86_64.cc
index b45eb3b..c368b03 100644
--- a/gold/x86_64.cc
+++ b/gold/x86_64.cc
@@ -4100,6 +4100,8 @@ Target_x86_64<size>::Relocate::tls_ie_to_le(
       // movq
       if (op1 == 0x4c)
 	view[-3] = 0x49;
+      else if (size == 32 && op1 == 0x44)
+	view[-3] = 0x41;
       view[-2] = 0xc7;
       view[-1] = 0xc0 | reg;
     }
@@ -4108,6 +4110,8 @@ Target_x86_64<size>::Relocate::tls_ie_to_le(
       // Special handling for %rsp.
       if (op1 == 0x4c)
 	view[-3] = 0x49;
+      else if (size == 32 && op1 == 0x44)
+	view[-3] = 0x41;
       view[-2] = 0x81;
       view[-1] = 0xc0 | reg;
     }
@@ -4116,6 +4120,8 @@ Target_x86_64<size>::Relocate::tls_ie_to_le(
       // addq
       if (op1 == 0x4c)
 	view[-3] = 0x4d;
+      else if (size == 32 && op1 == 0x44)
+	view[-3] = 0x45;
       view[-2] = 0x8d;
       view[-1] = 0x80 | reg | (reg << 3);
     }
-- 
1.9.3

Re: [PATCH] Handle Initial-Exec to Local-Exec for x32

[Cary Coutant]

> This patch handls Handle Initial-Exec to Local-Exec for x32.  OK
> to install?
>
>         PR gold/17809
>         * x86_64.cc (Target_x86_64<size>::Relocate::tls_ie_to_le): Handle
>         x32.

This is OK. Thanks!

When I run the test case from the PR through gold, I get a "TLS
relocation out of range" error for the first instruction, since it
doesn't have a REX prefix, and we'd otherwise try to look at .text -
1. How did you get around that? Do we need to support TLS relocations
on instructions without a REX prefix?

Where is the TLS ABI for x32 documented?

-cary

Re: [PATCH] Handle Initial-Exec to Local-Exec for x32

[H.J. Lu]

On Tue, Jan 6, 2015 at 1:18 PM, Cary Coutant <ccoutant@google.com> wrote:
>> This patch handls Handle Initial-Exec to Local-Exec for x32.  OK
>> to install?
>>
>>         PR gold/17809
>>         * x86_64.cc (Target_x86_64<size>::Relocate::tls_ie_to_le): Handle
>>         x32.
>
> This is OK. Thanks!

Is it OK for 2.25 branch?

> When I run the test case from the PR through gold, I get a "TLS
> relocation out of range" error for the first instruction, since it
> doesn't have a REX prefix, and we'd otherwise try to look at .text -
> 1. How did you get around that? Do we need to support TLS relocations

You need to install binutils with fix for

https://sourceware.org/bugzilla/show_bug.cgi?id=17482

> on instructions without a REX prefix?

Well, it is

https://sourceware.org/bugzilla/show_bug.cgi?id=17795

>
> -cary
> Where is the TLS ABI for x32 documented?

It is at

https://github.com/hjl-tools/x86-64-psABI/tree/hjl/x32/master


-- 
H.J.

Re: [PATCH] Handle Initial-Exec to Local-Exec for x32

[Cary Coutant]

>>>         PR gold/17809
>>>         * x86_64.cc (Target_x86_64<size>::Relocate::tls_ie_to_le): Handle
>>>         x32.
>>
>> This is OK. Thanks!
>
> Is it OK for 2.25 branch?

OK with me.

-cary

Re: [PATCH] Handle Initial-Exec to Local-Exec for x32

[Cary Coutant]

>> Do we need to support TLS relocations
>> on instructions without a REX prefix?
>
> Well, it is
>
> https://sourceware.org/bugzilla/show_bug.cgi?id=17795

Yeah -- I just wasn't yet sure whether you agreed with me.

-cary

Re: [PATCH] Handle Initial-Exec to Local-Exec for x32

[H.J. Lu]

On Wed, Jan 7, 2015 at 2:16 PM, Cary Coutant <ccoutant@google.com> wrote:
>>> Do we need to support TLS relocations
>>> on instructions without a REX prefix?
>>
>> Well, it is
>>
>> https://sourceware.org/bugzilla/show_bug.cgi?id=17795
>
> Yeah -- I just wasn't yet sure whether you agreed with me.
>

I agree with you on this.  Sorry that I didn't make it clear before.


-- 
H.J.