Linux-PPC egcs-1.1b g++ still not thread safe

Linux-PPC egcs-1.1b g++ still not thread safe

[Troy Benjegerdes]

I saw a web page regarding egcs 1.1, which says c++ exception handling is
now thread safe on x86.

I can verify that it still doesn't work on PPC.

I am getting segfaults in a particular function epilogue in which it
appears to be attempting to change or load some exception-related
variable, which some other thread has already conveniently trashed.

A snip of the offending assembly follows:

0x18211e0 <func__5cpmani+888>:  bl      0x19dacfc <IOMGR_Select>
0x18211e4 <func__5cpmani+892>:  addi    r9,r27,4
0x18211e8 <func__5cpmani+896>:  lwz     r11,0(r9)
0x18211ec <func__5cpmani+900>:  addi    r9,r11,4
0x18211f0 <func__5cpmani+904>:  lwz     r11,0(r9)
0x18211f4 <func__5cpmani+908>:  lwz     r0,0(r11)  

It crashes at 0x18211f4 with r11=0. The IOMGR_Select function is the last
function in the source, and in this case is a 60 second delay for the
current thread. This is from the Coda filesystem package, which uses it's
own user level threads library. 

I also just found out that if I use the '-fno-exceptions' flag, everything
appears to work.

--------------------------------------------------------------------------
| Troy Benjegerdes    |    troybenj@iastate.edu    |    hozer@drgw.net   |
|    Unix is user friendly... You just have to be friendly to it first.  |
| This message composed with 100% free software.    http://www.linux.org |
--------------------------------------------------------------------------

Re: Linux-PPC egcs-1.1b g++ still not thread safe

[Martin von Loewis]

> It crashes at 0x18211f4 with r11=0. The IOMGR_Select function is the last
> function in the source, and in this case is a 60 second delay for the
> current thread. This is from the Coda filesystem package, which uses it's
> own user level threads library. 

Ahh! Of course, the thread-safe exception support requires knowledge
about the thread package being used. For example, the current
exception is stored in thread-local storage. Different packages
provide different APIs to access this storage, if they provide API at
all.

So in order to use it, you'd have to adapt egcs to that thread
package. Please look at gcc/gthr-posix.h how the POSIX threads
adaptation works. If you don't want to adapt it, you might be better
of using -fno-exceptions, indeed. Please note that you might need to
recompile all the libraries.

Regards,
Martin