svn commit: r326203 - head/sys/conf

[Justin Hibbits]

On Tue, Nov 28, 2017 at 10:30 AM, Nathan Whitehorn
<nwhitehorn at freebsd.org> wrote:
>
>
> On 11/28/17 08:23, Justin Hibbits wrote:
>>
>> On Tue, Nov 28, 2017 at 10:13 AM, Nathan Whitehorn
>> <nwhitehorn at freebsd.org> wrote:
>>>
>>>
>>> On 11/28/17 07:27, Justin Hibbits wrote:
>>>>
>>>> On Sat, Nov 25, 2017 at 3:45 PM, Nathan Whitehorn
>>>> <nwhitehorn at freebsd.org> wrote:
>>>>>
>>>>> Author: nwhitehorn
>>>>> Date: Sat Nov 25 21:45:51 2017
>>>>> New Revision: 326203
>>>>> URL: https://svnweb.freebsd.org/changeset/base/326203
>>>>>
>>>>> Log:
>>>>>     Avoid emitting a PT_INTERP section for powerpc64 kernels and
>>>>> arrange
>>>>> for
>>>>>     the first instruction to be at the start of the text segment. This
>>>>> allows
>>>>>     the kernel to be booted correctly by stock kexec-lite.
>>>>>
>>>>>     MFC after:    2 weeks
>>>>>
>>>>> Modified:
>>>>>     head/sys/conf/ldscript.powerpc64
>>>>>
>>>>> Modified: head/sys/conf/ldscript.powerpc64
>>>>>
>>>>>
>>>>> ==============================================================================
>>>>> --- head/sys/conf/ldscript.powerpc64    Sat Nov 25 21:44:23 2017
>>>>> (r326202)
>>>>> +++ head/sys/conf/ldscript.powerpc64    Sat Nov 25 21:45:51 2017
>>>>> (r326203)
>>>>> @@ -10,7 +10,7 @@ SECTIONS
>>>>>    {
>>>>>      /* Read-only sections, merged into text segment: */
>>>>>
>>>>> -  . = kernbase + SIZEOF_HEADERS;
>>>>> +  . = kernbase;
>>>>>      PROVIDE (begin = . - SIZEOF_HEADERS);
>>>>>
>>>>>      .text      :
>>>>> @@ -24,7 +24,10 @@ SECTIONS
>>>>>      _etext = .;
>>>>>      PROVIDE (etext = .);
>>>>>
>>>>> -  .interp     : { *(.interp)   }
>>>>> +  /* Do not emit PT_INTERP section, which confuses some loaders
>>>>> (kexec-lite) */
>>>>> +  .interpX    : { *(.interp)   } : NONE
>>>>> +  /DISCARD/   : { *(.interp)   }
>>>>> +
>>>>>      .hash          : { *(.hash)          }
>>>>>      .dynsym        : { *(.dynsym)                }
>>>>>      .dynstr        : { *(.dynstr)                }
>>>>>
>>>> This broke powerpc64 Book-E kernels.  It now puts a 1MB blank space
>>>> ahead of the kernel data (ELF header + 1MB - sizeof(header) of 0's),
>>>> meaning that now the kernel needs to be loaded by uboot 1MB earlier in
>>>> memory, rather than straight on the 64MB boundary as it has been.
>>>>
>>>> - Justin
>>>>
>>> How on Earth? It doesn't do that on my system. What binutils are you
>>> using?
>>> -Nathan
>>>
>> This is using base binutils (2.17.50...)  I don't know why it's doing
>> this, but readelf shows that file offset 0x0000000000100000 maps to
>> 0xc000000000000000, and it goes from there.
>>
>> - Justin
>>
>
> Bizarre. Why don't you just revert for now (I need to run) and I can figure
> out what went wrong later?
> -Nathan

I reverted privately already.  I doubt there are many/any other users
of Book-E powerpc64, so I can wait it out for now.

- Justin