Bug 25051 - aarch64, powerpc64 uses surplus static tls for dynamically loaded dsos
Summary: aarch64, powerpc64 uses surplus static tls for dynamically loaded dsos
Status: RESOLVED FIXED
Alias: None
Product: glibc
Classification: Unclassified
Component: dynamic-link (show other bugs)
Version: 2.28
: P2 normal
Target Milestone: 2.32
Assignee: Szabolcs Nagy
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2019-09-30 17:24 UTC by Szabolcs Nagy
Modified: 2020-07-08 16:41 UTC (History)
6 users (show)

See Also:
Host:
Target:
Build:
Last reconfirmed:
fweimer: security-


Attachments

Note You need to log in before you can comment on or make changes to this bug.
Description Szabolcs Nagy 2019-09-30 17:24:48 UTC
tlsdesc (default on aarch64) and the powerpc tls optimization
allows dsos with general tls access to be more efficient if
static tls (with fixed tp offset) is used (instead of dynamic
tls via dtv indirections).

however in case of dynamically loaded dsos glibc even uses
the static tls surplus area for this optimization (i.e. when
no static tls was allocated for the dso specifically).

this means the surplus tls runs out more quickly compared
to other targets where that is only used for dynamically
loaded dsos with initial-exec tls access.

since there are libraries that rely on initial-exec tls, glibc
should reserve the surplus for that usage only.
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91938
(this bug depends on the resolution of that gcc bug)

i.e. TRY_STATIC_TLS should be replaced with HAVE_STATIC_TLS
in tls reloc processing code.

(initial-exec tls can continue to use CHECK_STATIC_TLS, but
the macros may need documentation changes)
Comment 1 Alan Modra 2019-10-01 05:22:32 UTC
(In reply to Szabolcs Nagy from comment #0)
> i.e. TRY_STATIC_TLS should be replaced with HAVE_STATIC_TLS
> in tls reloc processing code.

That does sound quite reasonable.  Shared libraries loaded at startup will still have the benefit of the tlsdesc optimisation.
Comment 2 Szabolcs Nagy 2020-01-07 12:02:50 UTC
posted a patch for discussion
https://sourceware.org/ml/libc-alpha/2020-01/msg00099.html
Comment 3 Dmitry Shachnev 2020-07-04 10:36:04 UTC
We got bug reports about a similar issue in Debian [1] (which affects PyQt5 and PySide packages on aarch64).

The latest version (v5) of the patch from [2] works, thanks!

[1]: https://bugs.debian.org/cgi-bin/bugreport.cgi?bug=964141
[2]: https://sourceware.org/pipermail/libc-alpha/2020-June/115284.html
Comment 4 Sourceware Commits 2020-07-08 16:34:19 UTC
The master branch has been updated by Szabolcs Nagy <nsz@sourceware.org>:

https://sourceware.org/git/gitweb.cgi?p=glibc.git;h=ffb17e7ba3a5ba9632cee97330b325072fbe41dd

commit ffb17e7ba3a5ba9632cee97330b325072fbe41dd
Author: Szabolcs Nagy <szabolcs.nagy@arm.com>
Date:   Wed Jun 10 13:40:40 2020 +0100

    rtld: Avoid using up static TLS surplus for optimizations [BZ #25051]
    
    On some targets static TLS surplus area can be used opportunistically
    for dynamically loaded modules such that the TLS access then becomes
    faster (TLSDESC and powerpc TLS optimization). However we don't want
    all surplus TLS to be used for this optimization because dynamically
    loaded modules with initial-exec model TLS can only use surplus TLS.
    
    The new contract for surplus static TLS use is:
    
    - libc.so can have up to 192 bytes of IE TLS,
    - other system libraries together can have up to 144 bytes of IE TLS.
    - Some "optional" static TLS is available for opportunistic use.
    
    The optional TLS is now tunable: rtld.optional_static_tls, so users
    can directly affect the allocated static TLS size. (Note that module
    unloading with dlclose does not reclaim static TLS. After the optional
    TLS runs out, TLS access is no longer optimized to use static TLS.)
    
    The default setting of rtld.optional_static_tls is 512 so the surplus
    TLS is 3*192 + 4*144 + 512 = 1664 by default, the same as before.
    
    Fixes BZ #25051.
    
    Tested on aarch64-linux-gnu and x86_64-linux-gnu.
    
    Reviewed-by: Carlos O'Donell <carlos@redhat.com>
Comment 5 Szabolcs Nagy 2020-07-08 16:41:55 UTC
fixed for 2.32.