Re: [PATCH] xen/riscv: fix build issue for bullseye-riscv64 container

[Jan Beulich <jbeulich@xxxxxxxx>]

On 31.07.2024 17:37, Andrew Cooper wrote:
> On 31/07/2024 4:07 pm, Oleksii Kurochko wrote:
>> Address compilation error on bullseye-riscv64 container:
>>    undefined reference to `guest_physmap_remove_page`
>>
>> Since there is no current implementation of `guest_physmap_remove_page()`,
>> a stub function has been added.
>>
>> Signed-off-by: Oleksii Kurochko <oleksii.kurochko@xxxxxxxxx>
>> Reported-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> 
> Usually we do Fixes/Reported-by tags first.  (We try to do tags in
> chronological/logical order).
> 
> I've confirmed this fixes the issue, so Acked-by: Andrew Cooper
> <andrew.cooper3@xxxxxxxxxx>
> 
> Having tried a different approach,
> 
> user@3c393ef3b4c0:/build/xen$ make
>   UPD     include/xen/compile.h
> Xen 4.20-unstable
> make[1]: Nothing to be done for 'include'.
> make[1]: 'arch/riscv/include/asm/asm-offsets.h' is up to date.
>   CC      common/version.o
>   LD      common/built_in.o
>   CC      arch/riscv/setup.o
>   LD      arch/riscv/built_in.o
>   LD      prelink.o
>   LDS     arch/riscv/xen.lds
> riscv64-linux-gnu-ld      -T arch/riscv/xen.lds -N prelink.o \
>     ./common/symbols-dummy.o -o ./.xen-syms.0
> riscv64-linux-gnu-ld: prelink.o: in function `register_keyhandler':
> /build/xen/common/keyhandler.c:106: undefined reference to
> `guest_physmap_remove_page'
> riscv64-linux-gnu-ld: ./.xen-syms.0: hidden symbol
> `guest_physmap_remove_page' isn't defined
> riscv64-linux-gnu-ld: final link failed: bad value
> make[2]: *** [arch/riscv/Makefile:15: xen-syms] Error 1
> make[1]: *** [build.mk:90: xen] Error 2
> make: *** [Makefile:610: xen] Error 2
> user@3c393ef3b4c0:/build/xen$ find . -name \*.o | while read F; do nm $F
> | grep guest_physmap_remove_page && echo $F; done
>                  U guest_physmap_remove_page
> ./common/memory.o
>                  U guest_physmap_remove_page
> ./common/built_in.o
>                  U guest_physmap_remove_page
> ./prelink.o
> 
> 
> despite the linker fingering keyhandler.c (very likely a toolchain bug),
> it's memory.o which has the undefined reference.
> 
> Looking at memory.i, there are calls in guest_remove_page(),
> memory_exchange(), and do_memory_op(), and no obvious way to DCE any of
> them.
> 
> However, with bookworm, while memory.i does still have the same calls,
> the resulting object file has no undefined references, so clearly
> something has DCE'd them.

The many stubs that have just BUG() in them result in a lot of DCE
according to my checking (many functions or basic blocks end - somewhat
confusingly - in a function call leading to such a stub). But apparently
not enough with the older compiler version.

Jan