207803 - [PATCH] Add FreeBSD Sparc64 support to Mozilla

Status: RESOLVED WORKSFORME

(Core :: XPConnect, defect)

Description

[Joe Marcus Clarke]

User-Agent:       Mozilla/5.0 (X11; U; FreeBSD i386; en-US; rv:1.4b) Gecko/20030531 Mozilla Firebird/0.6
Build Identifier: Mozilla/5.0 (X11; U; FreeBSD i386; en-US; rv:1.4b) Gecko/20030531 Mozilla Firebird/0.6

The attached patches created by Thomas Moestl <[e-mail]>, Kris Kennaway
<[e-mail]>, and myself add 64-bit UltraSPARC FreeBSD support to Mozilla.

Reproducible: Always

Steps to Reproduce:
1.
2.
3.

Comment 1

[Joe Marcus Clarke]

Attachment: Patch to add 64-bit UltraSPARC FreeBSD to Mozilla (NSPR portion checked in)

Comment 2

[Joe Marcus Clarke]

Attachment: Better patch to teach nspr we have socklen_t (checked in)

This patch replaces the patch to unix.c in the previous attachment.  It uses
the autoconf system to teach nspr that FreeBSD has socklen_t.

Comment 3

[Joe Marcus Clarke]

Note, previous patch was alsmo submitted by Thomas Moestl.

Comment 4

[Joe Marcus Clarke]

Attachment: Correct transformiix build on FreeBSD Sparc64

Comment 5

[Christian :Biesinger (don't email me, ping me on IRC)]

-> nspr and confirming

Comment 6

[Wan-Teh Chang]

Comment on attachment 124646 [details] [diff] [review]
Patch to add 64-bit UltraSPARC FreeBSD to Mozilla (NSPR portion checked in)

r=wtc on the NSPR portion of this patch.  I understand that
the patch for unix.c should be ignored and replaced by the next
patch.

Someone else needs to review the xpcom/reflect/xptcall portion
of this patch.

Comment 7

[Wan-Teh Chang]

Comment on attachment 124684 [details] [diff] [review]
Better patch to teach nspr we have socklen_t (checked in)

This patch and the NSPR portion of the previous patch have
been checked in on the NSPR trunk (NSPR 4.5) and
NSPRPUB_PRE_4_2_CLIENT_BRANCH (Mozilla 1.6b).

Comment 8

[Wan-Teh Chang]

Joe, do all recent FreeBSD versions have socklen_t?

Reassigned to the owner of xpcom/reflect/xptcall to review
the xpcom/reflect/xptcall portion of the first patch (attachment
124646 [details] [diff] [review]).

Comment 9

[Christian :Biesinger (don't email me, ping me on IRC)]

cc'ing some people about that transformiix patch

Comment 10

[Joe Marcus Clarke]

Yes, all recent (supported) versions of FreeBSD have socklen_t.  The type was
added over three years ago, and first appeared in FreeBSD 3.5-RELEASE.

Comment 11

[Joe Marcus Clarke]

Attachment: New fix for transformiix on FreeBSD

This patch corrects recent crashes that crept back in on Sparc64.  This was
done by Thomas Moestl <[e-mail]>.  Here is his description:

The problem was that values stored in arrays of 32-bit integers were used as
doubles via pointer magic. Since doubles have stricter alignment requirements
(8 bytes) than the compiler guarantees for these arrays,this can lead to
problems on platforms which do not allow unaligned memory accesses, like
sparc64 or alpha (while there's fixup code forsome instructions on sparc64, the
one used for doubles is not supported by that code currently). This is of
course only triggered if those variables are not on an 8-byte boundary by
chance, which is why it didn't occur earlier with the same version of
Double.cpp.  The patch changes that code to use unions for the type conversion,
which is the standard way of doing such things.

Comment 12

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Comment on attachment 135668 [details] [diff] [review]
New fix for transformiix on FreeBSD

Looks good to me, r=sicking

Comment 13

[Jim Rees]

Attachment: Similar fix from OpenBSD ports

This fix is from the OpenBSD mozilla port. I think it does the same thing in a
different way.

Comment 14

[Jonas Sicking (:sicking) No longer reading bugmail consistently]

Comment on attachment 141327 [details] [diff] [review]
Similar fix from OpenBSD ports

Could you explain what problem this patch solves? Also, shouldn't the
STRIC_ALIGN define live inside some sort of #ifdef for this platform only, i
doubt that compiles across all compilers.

Comment 15

[timeless]

Comment on attachment 135668 [details] [diff] [review]
New fix for transformiix on FreeBSD

this patch is better, although brendan mentioned there's more that should be
swiped from jsnum.[ch]

Comment 16

[Brendan Eich [:brendan]]

Comment on attachment 135668 [details] [diff] [review]
New fix for transformiix on FreeBSD

This rips off js/src/jsnum.[ch] incompletely -- besides needing a union to
align the uint32 pairs as this patch does, you need to consider ARM's
"PDP-endian-ness".

/be

Comment 17

[Wan-Teh Chang]

Brendan, what is "PDP-endian-ness"?

In prdtoa.c, we added a new macro IEEE_ARM to
handle ARM's double.  See
http://lxr.mozilla.org/nspr/source/nsprpub/pr/src/misc/prdtoa.c#143.

Is that what you were referring to?

Comment 18

[Phoenix]

According to this link - http://freebsd.1045724.n5.nabble.com/Does-Firefox-run-on-the-SPARC64-port-of-FreeBSD-tp3749610p3749611.html at least Firefox is working with FreeBSD starting from 6.2, so closing bug as WFM.