Re: [linus:master] BUILD REGRESSION a2e5790d841658485d642196dbb0927303d6c22f

From: Peter Zijlstra
Date: Thu Feb 08 2018 - 04:35:52 EST


On Thu, Feb 08, 2018 at 10:13:02AM +0100, Peter Zijlstra wrote:
> _The_ problem is that new binutils cannot sanely decode any function
> that has a WARN in (this very much includes perf annotate):

> new:
>
> 00000000000016a0 <copy_overflow>:
> 16a0: 48 89 f2 mov %rsi,%rdx
> 16a3: 89 fe mov %edi,%esi
> 16a5: 48 c7 c7 00 00 00 00 mov $0x0,%rdi
> 16a8: R_X86_64_32S .rodata.str1.8+0x288
> 16ac: e8 00 00 00 00 callq 16b1 <copy_overflow+0x11>
> 16ad: R_X86_64_PC32 __warn_printk-0x4
> 16b1: 0f ff c3 ud0 %ebx,%eax
> 16b4: 66 90 xchg %ax,%ax
> 16b6: 66 2e 0f 1f 84 00 00 nopw %cs:0x0(%rax,%rax,1)
> 16bd: 00 00 00

And note that this example is fairly 'good' as far as things go, it only
eats the single byte ret instruction and is on an instruction boundary
again, so the rest is good.

There are examples where what it eats results in shifting the
instruction boundary and the rest is _complete_ crap.

Look what it does to sched_cpu_dying:


7199: 48 39 c3 cmp %rax,%rbx
719c: 0f 84 cc 00 00 00 je 726e <sched_cpu_dying+0x1fe>
71a2: 0f ff 41 c6 ud0 -0x3a(%rcx),%eax
71a6: 87 3c 07 xchg %edi,(%rdi,%rax,1)
71a9: 00 00 add %al,(%rax)
71ab: 00 8b 43 04 83 f8 add %cl,-0x77cfbbd(%rbx)
71b1: 01 0f add %ecx,(%rdi)
71b3: 85 4f ff test %ecx,-0x1(%rdi)
71b6: ff (bad)
71b7: ff 48 8b decl -0x75(%rax)
71ba: 44 24 08 rex.R and $0x8,%al


7199: 48 39 c3 cmp %rax,%rbx
719c: 0f 84 cc 00 00 00 je 726e <sched_cpu_dying+0x1fe>
71a2: 0f ff (bad)
71a4: 41 c6 87 3c 07 00 00 movb $0x0,0x73c(%r15)
71ab: 00
71ac: 8b 43 04 mov 0x4(%rbx),%eax
71af: 83 f8 01 cmp $0x1,%eax
71b2: 0f 85 4f ff ff ff jne 7107 <sched_cpu_dying+0x97>
71b8: 48 8b 44 24 08 mov 0x8(%rsp),%rax
71bd: 48 8b 74 24 10 mov 0x10(%rsp),%rsi