Re: Haskell wxneeded

[Martijn Rijkeboer]

Hi,

On 08/26/16 22:44, Matthias Kilian wrote:

On Thu, Aug 11, 2016 at 09:37:57AM +0200, Martijn Rijkeboer wrote:

On 2016-08-10 22:04, Matthias Kilian wrote:

But I want to move forward with ghc (because ghc-8 works much better
on OpenBSD as Karel Gardas wrote some time ago). Not necessarily
right now, but at least during this release cycle (i.e. before
OpenBSD 6.1).


Would it be possible to have multiple versions of GHC like the way it is
done with Erlang or would that be too much work?


Only if the older version of ghc is standalone (no ghc.port.mk, no
support for any hs-packages for it, and it would conflict with
ghc-8, i.e. you can't install both ghc-7 and ghc-8 packages).

And if the bulk builders don't complain about something like ghc-lts
in the ports tree.


Understood. Thanks for the explanation.

Kind regards,


Martijn Rijkeboer

Re: Haskell wxneeded

[Matthias Kilian]

hi,

On Thu, Aug 11, 2016 at 09:37:57AM +0200, Martijn Rijkeboer wrote:
> On 2016-08-10 22:04, Matthias Kilian wrote:
> > But I want to move forward with ghc (because ghc-8 works much better
> > on OpenBSD as Karel Gardas wrote some time ago). Not necessarily
> > right now, but at least during this release cycle (i.e. before
> > OpenBSD 6.1).
> 
> Would it be possible to have multiple versions of GHC like the way it is
> done with Erlang or would that be too much work?

Only if the older version of ghc is standalone (no ghc.port.mk, no
support for any hs-packages for it, and it would conflict with
ghc-8, i.e. you can't install both ghc-7 and ghc-8 packages).

And if the bulk builders don't complain about something like ghc-lts
in the ports tree.

Ciao,
Kili

Re: Haskell wxneeded

[Karel Gardas]

On Sat, Aug 13, 2016 at 10:47 PM, Matthias Kilian
 wrote:
> Hi Karel,
>
> On Sat, Aug 13, 2016 at 06:56:39PM +0200, Karel Gardas wrote:
>> > $ ./inplace/bin/ghc-stage2 --interactive
>> > GHCi, version 8.1.20160812: http://www.haskell.org/ghc/  :? for help
>> > Prelude> 1+1
>> > 2
>> > Prelude> :q
>> > Leaving GHCi.
>> > $
>> >
>> > I'll try to propagate those fixes to GHC HEAD...
>>
>> FYI: The fixes are here:
>>
>> https://phabricator.haskell.org/D2453
>> https://phabricator.haskell.org/D2454
>
> The wxneeded "fix" is really just a workaround.

It is indeed and it's clearly mentioned isn't it? But anyway it makes
things running much better here so it's improvement.

> Right now I'm building
> and then testing a ghc-7.10.3 that
>
> - doas *not* link with -z wxneeded,
>
> - uses mmap with PROT_READ | PROT_WRITE in rts/Linker.c
>
> - uses mprotect with PROT_READ | PROT_EXEC right after loading and
>   resolving an object in Linker.c (i.e., just plain W^X)
>
> Results should be available tomorrow.

Good I've been also looking into this, but now I'm investigating full
testsuite results. So far majority of failures are in external
interpreter. Anyway, I'm looking forward to see your patch(es),
although Linker.c is quite moving target I hope I'll be able to
forward-port your changes to HEAD and if working well this will be
indeed proper fix for the issue and not just mine stopgap solution.

Cheers,
Karel

Re: Haskell wxneeded

[Matthias Kilian]

Hi Karel,

On Sat, Aug 13, 2016 at 06:56:39PM +0200, Karel Gardas wrote:
> > $ ./inplace/bin/ghc-stage2 --interactive
> > GHCi, version 8.1.20160812: http://www.haskell.org/ghc/  :? for help
> > Prelude> 1+1
> > 2
> > Prelude> :q
> > Leaving GHCi.
> > $
> >
> > I'll try to propagate those fixes to GHC HEAD...
> 
> FYI: The fixes are here:
> 
> https://phabricator.haskell.org/D2453
> https://phabricator.haskell.org/D2454

The wxneeded "fix" is really just a workaround. Right now I'm building
and then testing a ghc-7.10.3 that

- doas *not* link with -z wxneeded,

- uses mmap with PROT_READ | PROT_WRITE in rts/Linker.c

- uses mprotect with PROT_READ | PROT_EXEC right after loading and
  resolving an object in Linker.c (i.e., just plain W^X)

Results should be available tomorrow.

> if you do have any objections feel free to comment on phabricator...

If the testing goes well, I'll add a comment there.

Ciao,
Kili

Re: Haskell wxneeded

[Karel Gardas]

> With two small fixes GHC HEAD's GHCi runs with enforced wxneeded on
> wxallowed fs:
>
> $ ./inplace/bin/ghc-stage2 --interactive
> GHCi, version 8.1.20160812: http://www.haskell.org/ghc/  :? for help
> Prelude> 1+1
> 2
> Prelude> :q
> Leaving GHCi.
> $
>
> I'll try to propagate those fixes to GHC HEAD...

FYI: The fixes are here:

https://phabricator.haskell.org/D2453
https://phabricator.haskell.org/D2454

if you do have any objections feel free to comment on phabricator...

Re: Haskell wxneeded

[Karel Gardas]

On Sat, Aug 13, 2016 at 7:47 AM, Karel Gardas  wrote:
>>> Perhaps this is just intended behavior somehow as kernel complains on
>>> console with:
>>>
>>> /home/karel/src/ghc-head-openbsd-wxneeded/inplace/lib/bin/ghc-stage2(94953):
>>> W^X binary outside wxallowed mountpoint
>>>
>>> and obviously my current is w/o any wxallowed fs...
>>
>> Your port build directory and /usr/local need wxallowed, failures are 
>> expected
>> otherwise. (At present, there are some failures anyway even with this).
>
> Thanks for clarification, so I'll need to create some wxallowed space
> on my -current setup for hacking on GHC.

With two small fixes GHC HEAD's GHCi runs with enforced wxneeded on
wxallowed fs:

$ ./inplace/bin/ghc-stage2 --interactive
GHCi, version 8.1.20160812: http://www.haskell.org/ghc/  :? for help
Prelude> 1+1
2
Prelude> :q
Leaving GHCi.
$

I'll try to propagate those fixes to GHC HEAD...

Re: Haskell wxneeded

[Karel Gardas]

>> Perhaps this is just intended behavior somehow as kernel complains on
>> console with:
>>
>> /home/karel/src/ghc-head-openbsd-wxneeded/inplace/lib/bin/ghc-stage2(94953):
>> W^X binary outside wxallowed mountpoint
>>
>> and obviously my current is w/o any wxallowed fs...
>
> Your port build directory and /usr/local need wxallowed, failures are expected
> otherwise. (At present, there are some failures anyway even with this).

Thanks for clarification, so I'll need to create some wxallowed space
on my -current setup for hacking on GHC.

Re: Haskell wxneeded

[Stuart Henderson]

On 2016/08/13 01:03, Karel Gardas wrote:
> On Sat, Aug 13, 2016 at 12:56 AM, Karel Gardas  wrote:
> [...]
> > $ ./ghc/stage2/build/tmp/ghc-stage2
> > ./ghc/stage2/build/tmp/ghc-stage2[1]: syntax error: `)' unexpected
> > $ file ./ghc/stage2/build/tmp/ghc-sa   tage2
> > ./ghc/stage2/build/tmp/ghc-stage2: ELF 64-bit LSB shared object,
> > x86-64, version 1
> >
> >
> > Have anybody here seen this already? The system is:
> >
> 
> Perhaps this is just intended behavior somehow as kernel complains on
> console with:
> 
> /home/karel/src/ghc-head-openbsd-wxneeded/inplace/lib/bin/ghc-stage2(94953):
> W^X binary outside wxallowed mountpoint
> 
> and obviously my current is w/o any wxallowed fs...

Your port build directory and /usr/local need wxallowed, failures are expected
otherwise. (At present, there are some failures anyway even with this).

Re: Haskell wxneeded

[Karel Gardas]

On Sat, Aug 13, 2016 at 12:56 AM, Karel Gardas  wrote:
[...]
> $ ./ghc/stage2/build/tmp/ghc-stage2
> ./ghc/stage2/build/tmp/ghc-stage2[1]: syntax error: `)' unexpected
> $ file ./ghc/stage2/build/tmp/ghc-sa   tage2
> ./ghc/stage2/build/tmp/ghc-stage2: ELF 64-bit LSB shared object,
> x86-64, version 1
>
>
> Have anybody here seen this already? The system is:
>

Perhaps this is just intended behavior somehow as kernel complains on
console with:

/home/karel/src/ghc-head-openbsd-wxneeded/inplace/lib/bin/ghc-stage2(94953):
W^X binary outside wxallowed mountpoint

and obviously my current is w/o any wxallowed fs...

Karel

Re: Haskell wxneeded

[Karel Gardas]

> @Karel Gardas and anyone else who is playing with the newest upstream
> ghc (8.x): If you have some time, could you please try to build
> some hs-programs or libraries for which naddy@ reported w^x violations
> recently with ghc-8 (and probably cabal-install, because I don't
> think this will work with the hs-libraries we currently have in the
> ports tree)? I'd owe you unlimited amounts of beer for such a test,
> of course ;-)

:-) Beer is nothing for me, but anyway test done. First of all what
Martijn is seeing here is not ghc failure perse, but ghci failure. I
smell TH usage here probably. Anyway, GHC head is still buildable by
GHC 7.10.3, the obvious problem is that GHCi fails:

$ ./inplace/bin/ghc-stage2 --interactive
GHCi, version 8.1.20160812: http://www.haskell.org/ghc/  :? for help
Abort trap (core dumped)

if I try to build it with -Wl,-zwxneeded, then I get strange binary:

$ readelf -l ./ghc/stage2/build/tmp/ghc-stage2

Elf file type is DYN (Shared object file)
Entry point 0x440
There are 11 program headers, starting at offset 64

Program Headers:
  Type   Offset VirtAddr   PhysAddr
 FileSizMemSiz  Flags  Align
  PHDR   0x0040 0x0040 0x0040
 0x0348 0x0348  R E8
  INTERP 0x000c325e 0x001c325e 0x001c325e
 0x0013 0x0013  R  1
  [Requesting program interpreter: /usr/libexec/ld.so]
  LOAD   0x 0x 0x
 0x000c325e 0x000c325e  R E10
  LOAD   0x000c325e 0x001c325e 0x001c325e
 0x0003e0ae 0x0003e0ae  R  10
  LOAD   0x00102180 0x00302180 0x00302180
 0x00036998 0x000369e0  RW 10
  DYNAMIC0x00132bf0 0x00332bf0 0x00332bf0
 0x0360 0x0360  RW 8
  NOTE   0x000c3274 0x001c3274 0x001c3274
 0x0018 0x0018  R  4
  GNU_EH_FRAME   0x0010114c 0x0020114c 0x0020114c
 0x005c 0x005c  R  4
  GNU_STACK  0x 0x 0x
 0x 0x  RWE8
  OPENBSD_WXNEED 0x 0x 0x
 0x 0xE8
  GNU_RELRO  0x00102180 0x00302180 0x00302180
 0x00032e80 0x00032e80  R  1

 Section to Segment mapping:
  Segment Sections...
   00
   01 .interp
   02 .init .plt .text .fini
   03 .interp .note.openbsd.ident .hash .dynsym .dynstr .rela.dyn
.rela.plt .rodata .eh_frame_hdr .eh_frame
   04 .jcr .data.rel.ro .dynamic .ctors .dtors .got .data .bss
   05 .dynamic
   06 .note.openbsd.ident
   07 .eh_frame_hdr
   08
   09
   10 .jcr .data.rel.ro .dynamic .ctors .dtors .got
$ ./ghc/stage2/build/tmp/ghc-stage2
./ghc/stage2/build/tmp/ghc-stage2[1]: syntax error: `)' unexpected
$ file ./ghc/stage2/build/tmp/ghc-sa   tage2
./ghc/stage2/build/tmp/ghc-stage2: ELF 64-bit LSB shared object,
x86-64, version 1


Have anybody here seen this already? The system is:

OpenBSD 6.0-current (GENERIC.MP) #0: Fri Aug 12 22:41:37 CEST 2016

Cheers,
Karel

Re: Haskell wxneeded

[Martijn Rijkeboer]

Hi,

On 2016-08-10 22:04, Matthias Kilian wrote:
> But I want to move forward with ghc (because ghc-8 works much better
> on OpenBSD as Karel Gardas wrote some time ago). Not necessarily
> right now, but at least during this release cycle (i.e. before
> OpenBSD 6.1).

Would it be possible to have multiple versions of GHC like the way it is
done with Erlang or would that be too much work? In case that having
multiple versions is not viable, I'll switch to Stackage nightly till
GHC 8 becomes LTS. Anyway, thanks for your work on this port.

Kind regards,


Martijn Rijkeboer

Re: Haskell wxneeded

[Matthias Kilian]

Hi,

On Tue, Aug 09, 2016 at 10:20:47AM +0200, Martijn Rijkeboer wrote:
> > A real fix would be either hacking on ghcs dynamic loader or trying
> > to update to a newer ghc.
> 
> For now I would like to stick to 7.10.3 because I'm using Stackage LTS
> which targets that version.

But I want to move forward with ghc (because ghc-8 works much better
on OpenBSD as Karel Gardas wrote some time ago). Not necessarily
right now, but at least during this release cycle (i.e. before
OpenBSD 6.1).

Ciao,
Kili

Re: Haskell wxneeded

[Martijn Rijkeboer]

Hi,

On 2016-08-08 20:53, Matthias Kilian wrote:
> Looks like it's ghc itself failing here while trying to dynamically
> load ghc-prim. I'll try to write a patch for ghc which marks it as
> wxneeded as a workaround.

That would be nice.


> A real fix would be either hacking on ghcs dynamic loader or trying
> to update to a newer ghc.

For now I would like to stick to 7.10.3 because I'm using Stackage LTS
which targets that version.


Kind regards,


Martijn Rijkeboer

Haskell wxneeded

[Martijn Rijkeboer]

Hi,

How do I set the wxneeded flag for Haskell? I tried the following in my
cabal file, but to no avail:

  ghc-options: -optl -Wl,-z,wxneeded
  ld-options: -optl -Wl,-z,wxneeded

I have also tried the ld-options flag without the '-optl'. When I try
to compile my yesod app I get the following:

  ghc: mmap 630784 bytes at 0x4000: Not supported
  ghc: Try specifying an address with +RTS -xm -RTS
  ghc: panic! (the 'impossible' happened)
(GHC version 7.10.3 for x86_64-unknown-openbsd):
  loadObj 
"/usr/local/lib/ghc/ghcpr_8TmvWUcS1U1IKHT0levwg3/HSghc-prim-0.4.0.0-8TmvWUcS1U1IKHT0levwg3.o": 
failed



For the record, my partition is mounted wxallowed.

$ sysctl kern.version
kern.version=OpenBSD 6.0-current (GENERIC.MP) #2335: Sun Aug  7 20:50:41 
MDT 2016

dera...@amd64.openbsd.org:/usr/src/sys/arch/amd64/compile/GENERIC.MP


Kind regards,


Martijn Rijkeboer