260899 - 386 assembly implementations of PR_StackPush / PR_StackPop should be removed

Status: RESOLVED FIXED

(NSPR :: NSPR, defect, P2)

Description

[Julien Pierre]

These implementations have been known to cause problems. The 386 assembly code
should be purged from the builds for all x86 operating systems, notably Solaris x86.

Comment 1

[Wan-Teh Chang]

Please see bug 113740 for the problem that Julien
referred to.

Comment 2

[Wan-Teh Chang]

It would be good to have the Solaris thread team
look at the problem described in bug 113740.  We
didn't have the connection to do that back then.
Even if the response is "our thread library is
unaware of your home-grown spin lock and therefore
the thread scheduler can't possibly take that
into account", that's fine.

Comment 3

[Julien Pierre]

Wan-Teh, it would be good to have a test case for this problem to see if it is
still reproduce with current versions of Solaris, and what action to take on it
if so.

I know I wrote this bug after talking to you, but I don't recall what the issue
was under the other x86 operating systems. Was it the same one as under solaris ?

Comment 4

[Wan-Teh Chang]

Julien, I don't have a test case.  Sorry.

Comment 5

[Julien Pierre]

Attachment: remove assembly code that Solaris libthreads can't deal with

Note that os_SunOS_32.s is dead code and should also be removed from the tip.

Comment 6

[Wan-Teh Chang]

Comment on attachment 174539 [details] [diff] [review]
remove assembly code that Solaris libthreads can't deal with

r=wtc.	I checked in the patch on the NSPR
trunk (NSPR 4.6) and NSPRPUB_PRE_4_2_CLIENT_BRANCH
(Mozilla 1.8 Beta 2).

Comment 7

[Wan-Teh Chang]

Julien:

In comment 3, you asked about "the issue
under the other x86 operating systems".

The only such issue reported to me by AOL's
NPE team was under Solaris SPARC.

They didn't run NPE on Solaris x86, and
they didn't report this problem under the
other x86 operating systems.

A problem with spinlock based PR_StackPush
and PR_StackPop was reported under OSF1 before.
(see _osf1.h, rev. 3.8 and osf1.c, rev. 3.4)
That's the only similar problem I know of under
the other operating systems.

I will cvs remove os_SunOS_32.s.  Thanks for
pointing that out.