I can't reproduce it here using ubuntu 9.10.

Can you attach the config.log and main/php_config.h (from the build directory if you compile PHP outside the root src directory).

Since I do not know how to attach files here, there goes links:

php_config.h: http://pastebin.com/xxKxuBS2
config.h: http://pastebin.com/8KcBe0FF

Also, added strace output: http://pastebin.com/sP4hT4LB

"Click Here to Submit a Patch" sounds like what you need

It works just fine here.

Can you try to run it through gdb, stop the app when it reached the endless loop, and get a backtrace (bt)?

Added GDB backtrace as a patch (what a system:).

It does not look like a crypt issue at all. Is the src file you call exactly what you pasted here? Or can you attach it as zip please?

Stupid me, you did not call crypt php. Please do it again using:

gdb --args php crypt.php

or

gdb php
> run crypt.php
<break>
> bt

Updated backtrace (patch:).

Can you try using the attached patch please? It should fix the dead lock.

Thank you. Provided fixes this exact bug.

Same problem on windows (5.3.2 binary, php5apache2_2.dll):

Run 2 requests in parallel:
for ($i=0; $i<500000; $i++) {
  crypt('a', '_');
}

ab -n 1 -c 1 http://localhost/test.php
=> ~1 second, ok

ab -n 2 -c 2 http://localhost/test.php
=> hangs

Also confirmed third call hang on Win7/Apache2.2/PHP5.3.2.

Can anyone suggest a workaround until the next release?

hi  laacz at laacz dot lv
 can u please try with this patch and see if this addresses u r issue

to apply this patch, u will need to save the below contents into a file and 
run

cd php-5.3.2
patch -d . < <saved-file-name>

gmake clean
gmake

 
--- ext/standard/php_crypt_r.c.orig     Wed Jun 16 05:59:16 2010
+++ ext/standard/php_crypt_r.c  Wed Jun 16 06:00:17 2010
@@ -81,9 +81,7 @@
        tsrm_mutex_lock(php_crypt_extended_init_lock);
 #endif
 
-       if (initialized) {
-               return;
-       } else {
+       if (!initialized) {
                _crypt_extended_init();
                initialized = 1;
        }

This patch was what I proposed initially, it only reduces the risk but does not fix all cases.

What I committed is over safe as we could remove the tsrm lock. However I do need to know how we can do the membar on solaris.

i have attached a patch to add membar functionality for solaris. af course, this would be more relevant if we want to remove the tsrm lock around this.