Re: kern/52616: Recent regression in t_tls_static test on i386, sparc

[Robert Elz <kre%munnari.OZ.AU@localhost>]

The following reply was made to PR kern/52616; it has been noted by GNATS.

From: Robert Elz <kre%munnari.OZ.AU@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: 
Subject: Re: kern/52616: Recent regression in t_tls_static test on i386, sparc
Date: Sat, 14 Oct 2017 09:41:26 +0700

     Date:        Fri, 13 Oct 2017 19:50:00 +0000 (UTC)
     From:        gson%gson.org@localhost (Andreas Gustafsson)
     Message-ID:  <20171013195000.CA0B27A2B0%mollari.NetBSD.org@localhost>
 
   | The amd64 one seems unaffected.
 
 There, it is reported as a "bogus test program"
 
 From the latest b5 report:
 
 tp-start: 1507889157.798129, lib/libc/tls/t_tls_static, 0
 tp-end: 1507889157.803013, lib/libc/tls/t_tls_static, Invalid format for test ca
 se list: 1: Unexpected token `<<EOF>>'; expected a header name
 
 It has been that way in all b5 test runs since the commit that was mentioned
 So, I don't think we can conclude that it works there.
 
 I don't think it is the test that is broken - I ran  a very recent userland
 with an older (amd64) kernel (a few days older - just before that commit) and
 the test worked fine there.   (I know old kernels don't always work with
 newer userland, but here, with the changes that have been made, it does...)
 
 With a current kernel, t_tls_static (same binary) dumps core, segmentation
 violation in memcpy:
 
 #0  0x000000000043383f in memcpy ()
 #1  0x00000000004147be in _rtld_tls_allocate ()
 #2  0x0000000000414851 in __libc_static_tls_setup ()
 #3  0x000000000043884f in _libc_init ()
 #4  0x00000000004002bb in ___start ()
 #5  0x0000000000000000 in ?? ()
 
 which prevents it ever writing the ATF magic to the pipe to atf-report,
 which leads to the :bogus test program: status.
 
 kre