Re: [PATCH glibc 0/3] Restartable Sequences enablement

[Mathieu Desnoyers via Libc-alpha <libc-alpha@sourceware.org>]

----- On Jun 3, 2020, at 9:40 AM, Florian Weimer fweimer@redhat.com wrote:

> * Florian Weimer via Libc-alpha:
> 
>> This series unfortunately needs to be rebased (with conflicts) on
>> current master.
>>
>> There is a new elf/tst-auditmany failure, showing bug 26076, but
>> actually caused by bug 26075 (I think).  I will attempt to work around
>> these two issues, which are independent of the rseq changes.
> 
> Mathieu, it turns out that a change like this is needed:
> 
> diff --git a/elf/dl-tls.c b/elf/dl-tls.c
> index fa03234610..817bcbbf59 100644
> --- a/elf/dl-tls.c
> +++ b/elf/dl-tls.c
> @@ -31,7 +31,7 @@
> 
> /* Amount of excess space to allocate in the static TLS area
>    to allow dynamic loading of modules defining IE-model TLS data.  */
> -#define TLS_STATIC_SURPLUS     64 + DL_NNS * 100
> +#define TLS_STATIC_SURPLUS     64 + DL_NNS * 176
> 
> 
> /* Out-of-memory handler.  */
> 
> 
> I think you can just put that into the main rseq patch.  This change
> needs to be mentioned in the commit message, of course.
> 
> With this change, elf/tst-auditmany passes again for me.
> 
> The increase (76 bytes) is larger than 32 bytes because we have not done
> this in quite a while.  The cost in terms of additional TLS storage is
> quite significant, but it will also obscure some initial-exec-related
> dlopen failures.

OK, done!

Thanks,

Mathieu

> 
> Conceptually, we could also switch to 64 + (DL_NNS - 1) * 176, an
> increase of 1040 bytes over what we have today.
> 
> Thanks,
> Florian

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