486681 - Updates for HPPA/Linux

Status: RESOLVED FIXED

(Core :: XPCOM, defect)

Description

[Raúl Porcel]

Attachment: hppa.patch

Hi, please apply the following patch, it's from debian.
What it does is:
-Support generic hppa CHOST.
-Compile xpcom with -O0, since more than that segfaults.

Props to Mike Hommey <glandium at debian dot org> since he did it.

Patch should apply to 1.9 as well.

Comment 1

[Mike Hommey [:glandium]]

Comment on attachment 370860 [details] [diff] [review]
hppa.patch

Please wait a bit. I'll try to ping some hppa-aware people that promised a real patch a while ago (the ifneq part still applies, though). I tackled what is happening, but lacked the hppa assembly knowledge to come up with a proper fix. The problem lies in the fact that the stubs get a data structure from the stack, but function tail call optimization in the caller function breaks the stack location expectations, which is why building with -O0 works.

Comment 2

[timeless]

Comment on attachment 370860 [details] [diff] [review]
hppa.patch

i'll gladly give this an r-.

clobbering optimizations for a large number of unrelated patches is entirely unacceptable collateral damage.

Comment 3

[timeless]

err *platforms*

Comment 4

[Mike Hommey [:glandium]]

Huh? It's ifdef'ed and wouldn't be -O0 unless compiled on hppa!

Anyways, let's wait for the real patch.

Comment 5

[timeless]

ok, so, i'd like this as a patch queue (or export series, whatever). if you're adding an optimization restriction in one patch with an explanation that something is wrong, citing a bug, and in a second patch adding an additional variant of hppa, with an explanation of what it is, then those are two separate patches that should land as distinct changesets.

Comment 6

[Raúl Porcel]

From xpcom/reflect/xptcall/src/md/unix/Makefile.in:
>   212 ifeq ($(OS_ARCH),HP-UX)
>   213 ifneq ($(CC),gcc)
>   214 ifneq ($(OS_TEST),ia64)
>   215 CPPSRCS		:= xptcinvoke_pa32.cpp xptcstubs_pa32.cpp
>   216 ASFILES		:= xptcstubs_asm_pa32.s xptcinvoke_asm_pa32.s
>   217 else
>   218 CPPSRCS		:= xptcinvoke_ipf32.cpp xptcstubs_ipf32.cpp
>   219 ASFILES		:= xptcstubs_asm_ipf32.s xptcinvoke_asm_ipf32.s
>   220 endif
>   221 
>   222 # #18875  Building the CPP's (CXX) optimized causes a crash
>   223 CXXFLAGS 	:=  $(filter-out $(MOZ_OPTIMIZE_FLAGS), $(CXXFLAGS))
>  224 endif
>  225 endif

Adding the part from 221-223 makes it work on HPPA/Linux. So, if its being done for HP-UX, we could do it on Linux, couldn't we?

What do you say Mike?

Comment 7

[Raúl Porcel]

Mike? :)

Comment 8

[Mike Hommey [:glandium]]

(In reply to comment #6)
> Adding the part from 221-223 makes it work on HPPA/Linux. So, if its being done
> for HP-UX, we could do it on Linux, couldn't we?
> 
> What do you say Mike?

I'd say this would be a fix for bug 434190, though modifying the assembly would be better, but I never got the patch from our hppa guys :(

That would leave only the generic hppa chost in the patch here. What do you think ?

Comment 9

[Raúl Porcel]

Dunno, whatever you think its better.

Comment 10

[Mike Hommey [:glandium]]

Attachment: Patch

Let's do the chost only, here, and leave the rest to bug 434190

Comment 11

[Mike Hommey [:glandium]]

http://hg.mozilla.org/mozilla-central/rev/771a0f0a295b