[llvm-dev] LLD : __start_ and __end_ symbols for orphan sections

[Fangrui Song via llvm-dev]

On 2020-06-02, Petr Hosek wrote:
>In our case, we'd prefer the opposite that is for __start/__stop symbols to
>be always hidden (which is something Rafael also pointed out in
>https://reviews.llvm.org/D44566#1041984), currently this also requires a
>linker script on our side which is unfortunate. I have created a change
>https://reviews.llvm.org/D55682 to provide the -z hidden-start-stop-symbol
>but this was rejected. I'm wondering if it'd make sense to extend that
>change to instead provide -z start-stop-symbol-visibility= which could be
>used to override visibility for __start/__stop symbol (i.e. default or
>hidden) which would match the frontend -fvisibility= flag. Would that be
>acceptable?

See pcc and ruiu's objection in the review https://reviews.llvm.org/D55682
My feeling is similar: this extension looks of little value to me.

//extern char __start_f __attribute__((visibility("hidden")));
extern char __start_f; // Fuchsia wants to write code like this, but its visibility is set to hidden due to -z hidden-start-stop-symbols

If you want to persist, asking binutils at sourceware.org may help.
If they'd like to accept the extension, I think the option may get a
change passing pcc and ruiu's bar for an extension.

>On Tue, Jun 2, 2020 at 10:05 AM Moshtaghi, Alireza via llvm-dev <
>llvm-dev at lists.llvm.org> wrote:
>
>> Okay
>> Thank you
>>
>> On 6/2/20, 8:37 AM, "Fangrui Song" <maskray at google.com> wrote:
>>
>>     NetApp Security WARNING: This is an external email. Do not click links
>> or open attachments unless you recognize the sender and know the content is
>> safe.
>>
>>
>>
>>
>>     On 2020-06-02, Moshtaghi, Alireza wrote:
>>     >Sorry for the cryptic code but I had to modify stuff from original
>>     >In the following example see the difference when you comment or
>> uncomment the line in the linker script:
>>     >============ test.c ============= :
>>     >struct orphan_dummy_anno_s {
>>     > void (*func)(void);
>>     >};
>>     >
>>     >static void dummy_export_dbg_log_init_f (void) __attribute__
>> ((unused));
>>     >static struct orphan_dummy_anno_s const
>>     >  dummy_export_dbg_log_init_linker_set =
>>     >  { dummy_export_dbg_log_init_f, };
>>     >__asm__ (".globl " "__start_set_orphan_dummy_anno");
>>     >__asm__ (".globl " "__stop_set_orphan_dummy_anno");
>>     >static void const
>>     >  *__set_orphan_dummy_anno_sym_dummy_export_dbg_log_init_linker_set
>>     >  __attribute__ ((__section__ ("set_" "orphan_dummy_anno")))
>>     >  __attribute__ ((__used__)) = &dummy_export_dbg_log_init_linker_set;
>>     >============ linker_script ============= :
>>     >SECTIONS {
>>     >   data : { *(data) }
>>     ># when commented, the __start and __stop symbols will be protected.
>> How can I keep it from setting it as protected?
>>     >#   set_orphan_dummy_anno : { PROVIDE(__start_set_orphan_dummy_anno =
>> . ); *(set_orphan_dummy_anno) PROVIDE (__stop_set_orphan_dummy_anno = . );}
>>     >}
>>     >============ howto ============= :
>>     >LDFLAGS="-Bshareable -T ./linker_script"
>>     >clang -fPIC -c test.c
>>     >ld.lld $LDFLAGS test.o -o test.so
>>     >objdump -tT test.so
>>
>>     Making __start_* __stop_* STV_PROTECTED is a deliberate choice. See
>>     https://reviews.llvm.org/D44566
>>     Such symbols can still be exported to .dynsym (can be looked up from
>>     another shared object with dlopen) but prevent accidental symbol
>> preemption..
>>     >
>>     >On 6/2/20, 12:00 AM, "llvm-dev on behalf of Moshtaghi, Alireza via
>> llvm-dev" <llvm-dev-bounces at lists.llvm.org on behalf of
>> llvm-dev at lists.llvm.org> wrote:
>>     >
>>     >    NetApp Security WARNING: This is an external email. Do not click
>> links or open attachments unless you recognize the sender and know the
>> content is safe.
>>     >
>>     >
>>     >
>>     >
>>     >    You are right it creates them but sets the protected flag
>> (STV_PROTECTED) which seems to be the cause of my problem.
>>     >    How can I tell it to set the flag as STV_DEFAULT?
>>     >
>>     >    Thanks
>>     >    A
>>     >
>>     >    On 5/28/20, 11:30 PM, "Fangrui Song" <maskray at google.com> wrote:
>>     >
>>     >        NetApp Security WARNING: This is an external email. Do not
>> click links or open attachments unless you recognize the sender and know
>> the content is safe.
>>     >
>>     >
>>     >
>>     >
>>     >        On 2020-05-28, Moshtaghi, Alireza via llvm-dev wrote:
>>     >        >lld does not seem to create the __start and __end symbols
>> for orphan sections.
>>     >        >I would like to keep my linker script as generic as possible
>> so how can I tell lld to create these symbols without having to add them in
>> the linker script?
>>     >        >
>>     >        >Thanks
>>     >        >A
>>     >
>>     >        It works for me.
>>     >
>>     >        SECTIONS {
>>     >           data : { *(data) }
>>     >        }
>>     >
>>     >        .section data,"aw"
>>     >        .quad __start_orphan
>>     >        .quad __stop_orphan
>>     >
>>     >        .section orphan,"aw"
>>     >        .quad __start_data
>>     >        .quad __stop_data
>>     >
>>     >
>>     >        Can you give an example __start_ or __end_ symbols are not
>> defined for orphan sections?
>>     >
>>     >    _______________________________________________
>>     >    LLVM Developers mailing list
>>     >    llvm-dev at lists.llvm.org
>>     >    https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>     >
>>
>> _______________________________________________
>> LLVM Developers mailing list
>> llvm-dev at lists.llvm.org
>> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>>