Re: [RFC PATCH glibc 1/4] glibc: Perform rseq(2) registration at C startup and thread creation (v5)

[Mathieu Desnoyers]

----- On Jan 18, 2019, at 12:41 PM, Mathieu Desnoyers mathieu.desnoyers@xxxxxxxxxxxx wrote:

> ----- On Jan 14, 2019, at 8:51 PM, Mathieu Desnoyers
> mathieu.desnoyers@xxxxxxxxxxxx wrote:
> 
> [...]
> 
>> diff --git a/sysdeps/unix/sysv/linux/rseq-sym.c
>> b/sysdeps/unix/sysv/linux/rseq-sym.c
>> new file mode 100644
>> index 0000000000..6856d0388a
> [...]
>> +/* volatile because fields can be read/updated by the kernel.  */
>> +__thread volatile struct rseq __rseq_abi = {
>> +  .cpu_id = RSEQ_CPU_ID_UNINITIALIZED,
>> +};
>> +
>> +/* volatile because refcount can be read/updated by signal handlers.  */
>> +__thread volatile uint32_t __rseq_refcount;
> 
> Back to the weak vs non-weak question about those two symbols. I understand
> that tagging them as weak symbols has little effect on the dynamic loader
> when it loads libc.so. However, I'm worried about that happens when
> libc is statically linked into an application, and there happens to
> be more than one instance of those symbols (e.g. libc and another library
> define the same symbols, and both are statically linked into the same
> application). Isn't it a situation where tagging those symbols as "weak"
> becomes useful ?

Testing shows that it seems fine to statically link two archives within an
executable in a scenario where each .a defines the same symbol, without
using "weak", so I won't worry about this further.

Thanks,

Mathieu


-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com