Re: [PATCH 5/5] x86: don't build unused entry code when !PV32

[Jan Beulich <jbeulich@xxxxxxxx>]

On 06.04.2021 19:34, Andrew Cooper wrote:
> On 01/04/2021 15:37, Jan Beulich wrote:
>> On 01.04.2021 16:31, Andrew Cooper wrote:
>>> On 25/11/2020 08:51, Jan Beulich wrote:
>>>> @@ -102,19 +102,21 @@ void __dummy__(void)
>>>>      BLANK();
>>>>  #endif
>>>>  
>>>> -#ifdef CONFIG_PV
>>>> +#ifdef CONFIG_PV32
>>>>      OFFSET(DOMAIN_is_32bit_pv, struct domain, arch.pv.is_32bit);
>>> Even if PV32 is compiled out, the is_32bit field still exists, and is
>>> still necessary for crash analysis.  XCA parses this offset information
>>> as part of dissecting /proc/vmcore.
>>>
>>> It's one single bool in a fixed size allocation which we've got plenty
>>> of room in.  It can and should stay to avoid impacting the existing
>>> diagnostic tools.
>> I'm afraid I don't understand at all: I'm not removing the field.
> 
> You talked about removing it in the commit message.

Oh, in a post-commit message TBD remark, yes. Is your objection here
then merely to this possible further plan of removing that field, but
not against the changes in this patch?

>> All I'm removing is the entry for it in asm-offsets.h.
> 
> Yes, and that will break XCA, which is used by several downstreams, not
> just XenServer.
> 
> For RPM package reasons, you can't use debuginfo packages, because
> what's on disk doesn't match what's in memory until you've rebooted. 
> Livepatching adds an extra dimension of fun here.  There's not enough
> space in the vmcoreinfo page to pass enough structure information, so
> asm offsets is appended to the symbol table.  Yes its a gross hack, but
> its how things currently work.

Would you mind pointing me at where this appending is happening? You
can't mean the ELF symbol table in xen-syms - there's no entry for
DOMAIN_is_32bit_pv there. Plus I don't see the problem here, as the
asm-offsets entry already is conditional - it's merely the condition
which gets changed. Or if you mean the is_32bit field - there's no
representation there either in xen-syms (except, when enabled, in
debug info).

Jan