[PATCH] D58102: Support X86 Control-flow Enforcement Technology (CET) in LLD

[Peter Smith via Phabricator via llvm-commits]

peter.smith added a comment.

In D58102#1437364 <https://reviews.llvm.org/D58102#1437364>, @xiangzhangllvm wrote:

> In D58102#1436187 <https://reviews.llvm.org/D58102#1436187>, @peter.smith wrote:
>
> > I've checked that it does work, although at the moment it does leave a redundant .splt section in the binary. I think that this can be removed by accounting for this in the SPltSection::empty(), .........
>
>
> It shouldn't have **redundant** .splt section, there should not have any .splt section if IBT is disable or non-dynamic link. Could you show me your test please?


My test is a simple modification of x86-64-feature-1-and.s and Inputs/x86-64-feature-1-and1.s 
Step 1: add func3 and func4 to Inputs/x86-64-feature-1-and1.s so that we can resolve the symbols.

   .text
   .globl func3
   .type func3, at function
  func3:
   retq
  
  .text
   .globl func4
   .type func4, at function
  func4:
   retq

For x86-64-feature-1-and.s 
change.text to

   .text
  
  .type foo STT_GNU_IFUNC
  .globl foo
  foo:
   ret
  
   .globl func1
   .type func1, at function
  func1:
          callq func2
          callq func3
          callq foo        
          callq func4
   retq

I've added an ifunc, a call to an ifunc and a call to a couple more functions. Notably the ifunc is after the calls to func1 and func2, so we turn off the feature after having added entries to the .splt.
Disassembly from gnu objdump

  0000000000201000 <foo>:
    201000:	c3                   	retq   
  
  0000000000201001 <func1>:
    201001:	e8 2a 00 00 00       	callq  201030 <func2 at plt>
    201006:	e8 35 00 00 00       	callq  201040 <func3 at plt>
    20100b:	e8 50 00 00 00       	callq  201060 <*ABS*+0x201000 at plt>
    201010:	e8 3b 00 00 00       	callq  201050 <func4 at plt>
    201015:	c3                   	retq   
  
  Disassembly of section .plt:
  
  0000000000201020 <func2 at plt-0x10>:
    201020:	ff 35 e2 1f 00 00    	pushq  0x1fe2(%rip)        # 203008 <_DYNAMIC+0x1008>
    201026:	ff 25 e4 1f 00 00    	jmpq   *0x1fe4(%rip)        # 203010 <_DYNAMIC+0x1010>
    20102c:	0f 1f 40 00          	nopl   0x0(%rax)
  
  0000000000201030 <func2 at plt>:
    201030:	ff 25 e2 1f 00 00    	jmpq   *0x1fe2(%rip)        # 203018 <func2>
    201036:	68 00 00 00 00       	pushq  $0x0
    20103b:	e9 e0 ff ff ff       	jmpq   201020 <func1+0x1f>
  
  0000000000201040 <func3 at plt>:
    201040:	ff 25 da 1f 00 00    	jmpq   *0x1fda(%rip)        # 203020 <func3>
    201046:	68 01 00 00 00       	pushq  $0x1
    20104b:	e9 d0 ff ff ff       	jmpq   201020 <func1+0x1f>
  
  0000000000201050 <func4 at plt>:
    201050:	ff 25 d2 1f 00 00    	jmpq   *0x1fd2(%rip)        # 203028 <func4>
    201056:	68 02 00 00 00       	pushq  $0x2
    20105b:	e9 c0 ff ff ff       	jmpq   201020 <func1+0x1f>
  
  0000000000201060 <*ABS*+0x201000 at plt>:
    201060:	ff 25 ca 1f 00 00    	jmpq   *0x1fca(%rip)        # 203030 <_DYNAMIC+0x1030>
    201066:	68 00 00 00 00       	pushq  $0x0
    20106b:	e9 e0 ff ff ff       	jmpq   201050 <func4 at plt>
  
  Disassembly of section .splt:
  
  0000000000201070 <.splt>:
    201070:	f3 0f 1e fa          	endbr64 
    201074:	ff 25 9e 1f 00 00    	jmpq   *0x1f9e(%rip)        # 203018 <func2>
    20107a:	66 0f 1f 44 00 00    	nopw   0x0(%rax,%rax,1)
    201080:	f3 0f 1e fa          	endbr64 
    201084:	ff 25 96 1f 00 00    	jmpq   *0x1f96(%rip)        # 203020 <func3>
    20108a:	66 0f 1f 44 00 00    	nopw   0x0(%rax,%rax,1)

Note the redundant splt. I'm using the previous version of the patch as the current one didn't compile for me. I will double check the most recent patch with a more recent compiler.


Repository:
  rLLD LLVM Linker

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D58102/new/

https://reviews.llvm.org/D58102