Re: [PATCH 7/8] x86/EFI: keep debug info in xen.efi

[Jan Beulich <jbeulich@xxxxxxxx>]

On 22.04.2021 17:53, Roger Pau Monné wrote:
> On Thu, Apr 22, 2021 at 05:46:28PM +0200, Jan Beulich wrote:
>> On 22.04.2021 16:56, Roger Pau Monné wrote:
>>> On Thu, Apr 22, 2021 at 01:03:13PM +0200, Jan Beulich wrote:
>>>> On 22.04.2021 10:14, Roger Pau Monné wrote:
>>>>> On Wed, Apr 21, 2021 at 05:38:42PM +0200, Jan Beulich wrote:
>>>>>> On 21.04.2021 17:30, Roger Pau Monné wrote:
>>>>>>> On Wed, Apr 21, 2021 at 03:06:36PM +0200, Jan Beulich wrote:
>>>>>>>> On 21.04.2021 13:15, Roger Pau Monné wrote:
>>>>>>>>> On Thu, Apr 01, 2021 at 11:47:03AM +0200, Jan Beulich wrote:
>>>>>>>>>> --- a/xen/arch/x86/xen.lds.S
>>>>>>>>>> +++ b/xen/arch/x86/xen.lds.S
>>>>>>>>>> @@ -312,10 +312,60 @@ SECTIONS
>>>>>>>>>>      *(.reloc)
>>>>>>>>>>      __base_relocs_end = .;
>>>>>>>>>>    }
>>>>>>>>>> -  /* Trick the linker into setting the image size to exactly 16Mb. 
>>>>>>>>>> */
>>>>>>>>>> -  . = ALIGN(__section_alignment__);
>>>>>>>>>> -  DECL_SECTION(.pad) {
>>>>>>>>>> -    . = ALIGN(MB(16));
>>>>>>>>>> +  .debug_abbrev ALIGN(1) (NOLOAD) : {
>>>>>>>>>> +     *(.debug_abbrev)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_info ALIGN(1) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_info)
>>>>>>>>>> +    *(.gnu.linkonce.wi.*)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_types ALIGN(1) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_types)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_str ALIGN(1) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_str)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_line ALIGN(1) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_line)
>>>>>>>>>> +    *(.debug_line.*)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_line_str ALIGN(1) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_line_str)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_names ALIGN(4) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_names)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_frame ALIGN(4) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_frame)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_loc ALIGN(1) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_loc)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_loclists ALIGN(4) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_loclists)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_ranges ALIGN(8) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_ranges)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_rnglists ALIGN(4) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_rnglists)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_addr ALIGN(8) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_addr)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_aranges ALIGN(1) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_aranges)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_pubnames ALIGN(1) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_pubnames)
>>>>>>>>>> +  }
>>>>>>>>>> +  .debug_pubtypes ALIGN(1) (NOLOAD) : {
>>>>>>>>>> +    *(.debug_pubtypes)
>>>>>>>>>> +  }
>>>>>>>>>> +  /* Trick the linker into setting the image size to no less than 
>>>>>>>>>> 16Mb. */
>>>>>>>>>> +  __image_end__ = .;
>>>>>>>>>> +  .pad ALIGN(__section_alignment__) : {
>>>>>>>>>> +    . = __image_end__ < __image_base__ + MB(16) ? ALIGN(MB(16)) : .;
>>>>>>>>>
>>>>>>>>> I think this is inside an ifdef EFI region, since this is DWARF info
>>>>>>>>> couldn't it also be present when building with EFI disabled?
>>>>>>>>
>>>>>>>> Of course (and it's not just "could" but "will"), yet the linker will
>>>>>>>> do fine (and perhaps even better) without when building ELF. Also
>>>>>>>> note that we'll be responsible for keeping the list of sections up-to-
>>>>>>>> date. The linker will recognize Dwarf sections by looking for a
>>>>>>>> .debug_ prefix. We can't use such here (or at least I'm not aware of
>>>>>>>> a suitable mechanism); .debug_* would mean munging together all the
>>>>>>>> different kinds of Dwarf sections. Hence by limiting the explicit
>>>>>>>> enumeration to PE, I'm trying to avoid anomalies in ELF down the road.
>>>>>>>
>>>>>>> Right, so we will have to keep this list of debug_ sections updated
>>>>>>> manually if/when more of those appear as part of DWARF updates?
>>>>>>
>>>>>> Yes.
>>>>>>
>>>>>>> Do we have a way to get some kind of warning or error when a new
>>>>>>> section not explicitly handled here appears?
>>>>>>
>>>>>> ld 2.37 will start warning about such sections, as they'd land at
>>>>>> VA 0 and hence below image base.
>>>>>
>>>>> That seems like a bug in ld?
>>>>>
>>>>> The '--image-base' option description mentions: "This is the lowest
>>>>> memory location that will be used when your program or dll is
>>>>> loaded.", so I would expect that if the option is used the default VA
>>>>> should be >= image-base, or else the description of the option is not
>>>>> consistent, as ld will still place sections at addresses below
>>>>> image-base.
>>>>
>>>> ld's "general" logic is pretty ELF-centric. Hence debugging sections
>>>> get placed at VA 0 by default, not matter what the (PE-specific)
>>>> --image-base says. Whether that's a bug though I'm not sure: There
>>>> are no really good alternatives that could be used by default. Doing
>>>> what we do here (and what e.g. Cygwin does) via linker script may not
>>>> be appropriate in the common case. In particular it is not generally
>>>> correct for debug info to be part of what gets loaded into memory.
>>>
>>> So with this change here you placate the warnings from newer ld about
>>> having a VA < image base,
>>
>> It's not just about silencing the warnings. The resulting image is
>> unusable when the sections don't get placed at a suitable VA.
> 
> And this wasn't an issue before because the linker won't even attempt
> to place DWARF sections into a PE output.

No, this wasn't an issue before since, for things to work, we
simply had to uniformly strip debug info when linking xen.efi. And
this is what Andrew said should change. I was initially opposed,
until I saw that Cygwin does just this as well.

>>> but the end result is that now the debug
>>> sections will also get loaded when booted from the EFI loader?
>>> (because the NOLOAD doesn't have any effect with PE)
>>
>> Yes. I currently see no other way to retain debug info in xen.efi.
>> But to be clear, the memory debug info occupies isn't lost - we
>> still free space from _end (or alike) onwards. .reloc, for example,
>> also lives there. And I was wondering whether we shouldn't keep
>> .comment this way as well.
> 
> Yes, I already realized all this is past _end.
> 
> I wonder however if the use of (NOLOAD) makes all this more confusing,
> such sections should only be present on the linker script used for the
> PE output, and then the (NOLOAD) doesn't make sense there?
> 
> If so, I think the (NOLOAD) directive should be dropped, and a comment
> noting that the debug sections need to be manually added to the PE
> output in order to avoid them being placed at VA 0 would be helpful
> IMO, likely also mentioning that they would be loaded but discarded
> afterwards by Xen because they are all past _end.

Earlier on (another sub-thread, maybe) I think I've already said that
I'd like to keep (NOLOAD) both for documentation purposes and just in
case the linker develops some smarts to actually translate it into
anything sensible when linking PE. This is quite different from
.reloc, after all - that section has to be loaded for our re-
relocation to work correctly. Hence that section not having (NOLOAD)
and the debugging sections having it points out the difference.

Jan