Re: [PATCH v8 3/8] nptl: Add public __rseq_feature_size symbol

[Michael Jeanson <mjeanson@efficios.com>]

On 2024-02-16 17:07, DJ Delorie wrote:
> 
> * Suggestion for manual entry
> * attributes on externs
> * comment about _rseq_feature_size default
> 

>> diff --git a/manual/threads.texi b/manual/threads.texi
> 
>> +@deftypevar {unsigned int} __rseq_feature_size
>> +@standards{Linux, sys/rseq.h}
>> +This variable is either zero (if restartable sequence registration
>> +failed or has been disabled) or the size of the restartable sequence
>> +features.  If registration is successful, @code{__rseq_feature_size}
>> +is at least 20 (the initial feature size of @code{struct rseq}).
>> +@end deftypevar
> 
> Perhaps a note about where the magic value '20' comes from, and how to
> change it?  (Yes, *I* know it comes from the kernel, but the reader
> might not)

I'll add an explaination on where the magic value comes from.

Not sure about the 'change it' part, this minimum value is in the ABI and 
should never change. Do you mean explaining why the 'feature_size' would change?


>> diff --git a/sysdeps/nptl/dl-tls_init_tp.c b/sysdeps/nptl/dl-tls_init_tp.c
>> +extern unsigned int _rseq_feature_size attribute_relro attribute_hidden;
> 
> Again with attributes on externs.

Ack.


>> @@ -108,6 +109,7 @@ __tls_init_tp (void)
>>       if (rseq_register_current_thread (pd, do_rseq))
>>         {
>>           _rseq_size = GLRO (dl_tls_rseq_size);
>> +        _rseq_feature_size = GLRO (dl_tls_rseq_feature_size);
>>         }
> 
> Ok.  Note "comment needed here" from our other mail thread ;-)

Ack.