[PATCH] Async signal safe TLS accesses

[Paul Pluzhnikov]

On 1/9/14, 7:46 AM, Joseph S. Myers wrote:

>> Could you switch to the other thread and print *imap?
>
> $1 = {l_addr = 266268672,
>    l_name = 0x100140a8
> "/scratch/jmyers/eglibc/p/obj/glibc-4.7-0-powerpc-linux-gnu-i686-pc-linux-gnu/default/nptl/tst-tls7mod.so",

266268672 == 0xfdef000

Crash is at 0x0fdef838.

So that confirms the "action() from tst-tls7mod.so is still running when 
dlclose unmaps the library" theory.

What about the race I posited as possible explanation? Do you agree that 
there is a race? Does sleep(1) before dlclose() "fix" it?

Also, is the crash you are seeing intermittent (how frequent?) or reliable?

Thanks,