6.4.2 hppa info tables question

[Joe Buehler]

In getting the registerised build up and running for
hppa 6.4.2, I have run into a bit of a road block.

I'm puzzling over the mangler and what is in the
_info tables.  Here is an example of what gcc emits:

	.SPACE $PRIVATE$
	.SUBSPA $DATA$

	.align 4
GHCziConc_lvl2_srt
	.word	DataziTypeable_lgo_closure
	.word	DataziTypeable_zdwmkTyCon_closure
	.word	GHCziConc_lvl_closure
	.word	GHCziConc_lvl1_closure

...snip...

	.SPACE $PRIVATE$
	.SUBSPA $DATA$

	.align 4
s54Q_info
	.word	GHCziConc_lvl2_srt
	.word	0
	.word	2359301

The mangler normally would replace the srt pointer in the info
table with a relative offset from the info table.  However,
the info table gets moved to the code space, in front of
its associated function, and I think the offset computation
is problematic for hppa.

>From what I surmise, the addresses in hppa are not straight
32 bit indices -- the top 2 bits are used to select a "space"
register, and this causes the linker to emits lots and lots of
warnings regarding the offset computation.  E.g.:

/afs/hekimian.com/@sys/local/hppa1.0-hp-hpux11.00/bin/ld: (Warning) Quadrant change in relocatable expression file /home/project-releases/tmp/ghc-6.4.2/libraries/base/libHSbase.a(Types.o).
/afs/hekimian.com/@sys/local/hppa1.0-hp-hpux11.00/bin/ld: (Warning) Quadrant change in relocatable expression in subspace $CODE$ at offset 0x4f44 in file /home/plibraries/base/libHSbase.a(Types.o).

A comment in the mangler indicates that GHC went to offsets in info
tables instead of pointers for PIC reasons.  This doesn't make a
whole lot of sense to me however -- the srt, srtd and btm tables
are in the data segment.  I would expect any offset to be to text
segment objects only.

Are these tables also supposed to be moved to text by the mangler?

Joe Buehler