inferno hosted on Linux with arm

Re: inferno hosted on Linux with arm

iyacc needs to be of the same cpu type as your Ubuntu (as do the build
functions).
You've built the arm version of iyacc, but that is interpreted as a
shell script on x86.
(It's a bit stupid that something doesn't complain about an
incompatible architecture, and
I think it's something in the Linux environment that's at fault there,
nothing directly to do with Inferno.)

You don't need PATH=$HOME/inferno/Linux/arm/bin:...
because you're not running on the arm. Instead you need the 386 directory (if
the Ubuntu is on some variant of x86). With OBJTYPE=arm, SYSTARG=Linux,
mkfiles/mkfile-Linux-arm will expect an arm-gcc somewhere in the PATH,
as a gcc-based cross-compiler running on x86 generating code for the ARM.
(You can change the mkfile if some other cross-compiler name is appropriate.)

Re: inferno hosted on Linux with arm

Re: inferno hosted on Linux with arm

Re: inferno hosted on Linux with arm

it's a little devious, but it does look as though it should be all
right in the end.

similarly to iyacc, you need limbo as an x86 executable. possibly the
best thing is to
mk install Inferno with OBJTYPE=386, to get all the supporting commands,
including limbo (the Limbo compiler), then set OBJTYPE=arm, and mk install
to get the ARM-hosted version. I'd have written this up earlier except that:
 a. I didn't think of it
 b. to be fair, cross-compilation on non-Plan 9 systems to make other
hosted systems is relatively new
 c. it was trivial under Plan 9, because cross-compilation is there by default,
     although in retrospect, there's a similar assumption (that the
source system
     compilers have been made).

On 28 November 2011 21:02, Shubhangi Tamsekar
<shubhangi.tamsekar@...> wrote:
> Thanks Charles.
>
> I changed the path to PATH=$HOME/inferno/Linux/386/bin
> set appropriate cross compiler name in mkfile-linux-arm
> and then do mk install
>
> Now it is able to pass beyond iyacc however, it is stuck mysteriously for
> limbo. Here are the error messages that I get -
>
>
> rm -f /home/inferno/Linux/arm/bin/limbo && cp o.out
> /home/inferno/Linux/arm/bin/limbo
> (cd libinterp; mk  install)
> /home/inferno/Linux/arm/lib/libinterp.a doesn't exist: assuming it will be
> an archive
> rm -f runt.h && limbo -a -I../module ../module/runt.m > runt.h
> sh: limbo: not found
> mk: rm -f runt.h ...  : exit status=exit(127), deleting 'runt.h'
> mk: for j in ...  : exit status=exit(1)
>
>
> Now what is wrong with my Limbo? I can see mk, iyacc and limbo files in my
> /inferno/Linux/arm/bin/ folder.
>
> Any idea?
>
> On Mon, Nov 28, 2011 at 4:52 AM, Charles Forsyth <charles.forsyth <at> gmail.com>
> wrote:
>>
>> iyacc needs to be of the same cpu type as your Ubuntu (as do the build
>> functions).
>> You've built the arm version of iyacc, but that is interpreted as a
>> shell script on x86.
>> (It's a bit stupid that something doesn't complain about an
>> incompatible architecture, and
>> I think it's something in the Linux environment that's at fault there,
>> nothing directly to do with Inferno.)
>>
>> You don't need PATH=$HOME/inferno/Linux/arm/bin:...
>> because you're not running on the arm. Instead you need the 386 directory
>> (if
>> the Ubuntu is on some variant of x86). With OBJTYPE=arm, SYSTARG=Linux,
>> mkfiles/mkfile-Linux-arm will expect an arm-gcc somewhere in the PATH,
>> as a gcc-based cross-compiler running on x86 generating code for the ARM.
>> (You can change the mkfile if some other cross-compiler name is
>> appropriate.)
>
>

Re: inferno hosted on Linux with arm

check if the /home/inferno/Linux/arm/lib/ folder exists. if not create
it by hand.

On Mon, Nov 28, 2011 at 11:02 PM, Shubhangi Tamsekar
<shubhangi.tamsekar@...> wrote:
> Thanks Charles.
>
> I changed the path to PATH=$HOME/inferno/Linux/386/bin
> set appropriate cross compiler name in mkfile-linux-arm
> and then do mk install
>
> Now it is able to pass beyond iyacc however, it is stuck mysteriously for
> limbo. Here are the error messages that I get -
>
>
> rm -f /home/inferno/Linux/arm/bin/limbo && cp o.out
> /home/inferno/Linux/arm/bin/limbo
> (cd libinterp; mk  install)
> /home/inferno/Linux/arm/lib/libinterp.a doesn't exist: assuming it will be
> an archive
> rm -f runt.h && limbo -a -I../module ../module/runt.m > runt.h
> sh: limbo: not found
> mk: rm -f runt.h ...  : exit status=exit(127), deleting 'runt.h'
> mk: for j in ...  : exit status=exit(1)
>
>
> Now what is wrong with my Limbo? I can see mk, iyacc and limbo files in my
> /inferno/Linux/arm/bin/ folder.
>
> Any idea?
>
> On Mon, Nov 28, 2011 at 4:52 AM, Charles Forsyth <charles.forsyth <at> gmail.com>
> wrote:
>>
>> iyacc needs to be of the same cpu type as your Ubuntu (as do the build
>> functions).
>> You've built the arm version of iyacc, but that is interpreted as a
>> shell script on x86.
>> (It's a bit stupid that something doesn't complain about an
>> incompatible architecture, and
>> I think it's something in the Linux environment that's at fault there,
>> nothing directly to do with Inferno.)
>>
>> You don't need PATH=$HOME/inferno/Linux/arm/bin:...
>> because you're not running on the arm. Instead you need the 386 directory
>> (if
>> the Ubuntu is on some variant of x86). With OBJTYPE=arm, SYSTARG=Linux,
>> mkfiles/mkfile-Linux-arm will expect an arm-gcc somewhere in the PATH,
>> as a gcc-based cross-compiler running on x86 generating code for the ARM.
>> (You can change the mkfile if some other cross-compiler name is
>> appropriate.)
>
>

Re: inferno hosted on Linux with arm

This is what I had to do for the Hellaphone. It's a little annoying
but doesn't take that long.

Speaking of Inferno on ARM, I compiled it for my Efika Smartbook (ARM
v7l processor), but it seems to just... die, after you start a few
processes. If I do "emu -r .", I can run "ls" and "ps" and such
seemingly without limit, but if I run "man wm" a few times, it's
almost certain to kill emu. I think it has something to do with the
process handling code, I swear I ran into something like this while
working with the Hellaphone.

John

On Mon, Nov 28, 2011 at 1:31 PM, Charles Forsyth
<charles.forsyth@...> wrote:
> it's a little devious, but it does look as though it should be all
> right in the end.
>
> similarly to iyacc, you need limbo as an x86 executable. possibly the
> best thing is to
> mk install Inferno with OBJTYPE=386, to get all the supporting commands,
> including limbo (the Limbo compiler), then set OBJTYPE=arm, and mk install
> to get the ARM-hosted version. I'd have written this up earlier except that:
>  a. I didn't think of it
>  b. to be fair, cross-compilation on non-Plan 9 systems to make other
> hosted systems is relatively new
>  c. it was trivial under Plan 9, because cross-compilation is there by default,
>     although in retrospect, there's a similar assumption (that the
> source system
>     compilers have been made).
>
> On 28 November 2011 21:02, Shubhangi Tamsekar
> <shubhangi.tamsekar@...> wrote:
>> Thanks Charles.
>>
>> I changed the path to PATH=$HOME/inferno/Linux/386/bin
>> set appropriate cross compiler name in mkfile-linux-arm
>> and then do mk install
>>
>> Now it is able to pass beyond iyacc however, it is stuck mysteriously for
>> limbo. Here are the error messages that I get -
>>
>>
>> rm -f /home/inferno/Linux/arm/bin/limbo && cp o.out
>> /home/inferno/Linux/arm/bin/limbo
>> (cd libinterp; mk  install)
>> /home/inferno/Linux/arm/lib/libinterp.a doesn't exist: assuming it will be
>> an archive
>> rm -f runt.h && limbo -a -I../module ../module/runt.m > runt.h
>> sh: limbo: not found
>> mk: rm -f runt.h ...  : exit status=exit(127), deleting 'runt.h'
>> mk: for j in ...  : exit status=exit(1)
>>
>>
>> Now what is wrong with my Limbo? I can see mk, iyacc and limbo files in my
>> /inferno/Linux/arm/bin/ folder.
>>
>> Any idea?
>>
>> On Mon, Nov 28, 2011 at 4:52 AM, Charles Forsyth <charles.forsyth <at> gmail.com>
>> wrote:
>>>
>>> iyacc needs to be of the same cpu type as your Ubuntu (as do the build
>>> functions).
>>> You've built the arm version of iyacc, but that is interpreted as a
>>> shell script on x86.
>>> (It's a bit stupid that something doesn't complain about an
>>> incompatible architecture, and
>>> I think it's something in the Linux environment that's at fault there,
>>> nothing directly to do with Inferno.)
>>>
>>> You don't need PATH=$HOME/inferno/Linux/arm/bin:...
>>> because you're not running on the arm. Instead you need the 386 directory
>>> (if
>>> the Ubuntu is on some variant of x86). With OBJTYPE=arm, SYSTARG=Linux,
>>> mkfiles/mkfile-Linux-arm will expect an arm-gcc somewhere in the PATH,
>>> as a gcc-based cross-compiler running on x86 generating code for the ARM.
>>> (You can change the mkfile if some other cross-compiler name is
>>> appropriate.)
>>
>>
>

Re: inferno hosted on Linux with arm

Today is obviously "it's all my fault" day. The default installation
for Linux still uses clone,
which would be much more efficient but is less and less feasible as
glibc acquires dependencies
on pthreads. It's not trivial to substitute a different C library, so
it would be better if the default
installation instead was set up for pthreads. I'll push out the
changes to do that.

On 28 November 2011 22:10, John Floren <john@...> wrote:
> Speaking of Inferno on ARM, I compiled it for my Efika Smartbook (ARM
> v7l processor), but it seems to just... die, after you start a few
> processes. If I do "emu -r .", I can run "ls" and "ps" and such
> seemingly without limit, but if I run "man wm" a few times, it's
> almost certain to kill emu. I think it has something to do with the
> process handling code, I swear I ran into something like this while
> working with the Hellaphone.

Re: inferno hosted on Linux with arm

Re: inferno hosted on Linux with arm

That's doubly weird: first that it's using blx, second that no one's
mentioned it before.
blx is a bl (branch and link) that also forces an exchange of
instruction sets (ie, forces
Thumb mode), which is not a little odd. It looks as though the donors
of that code
were running in an environment where most code was compiled for Thumb, but
Inferno was using arm32 mode (what i'd call "normal" ARM). Phones
often liked Thumb
mode, so perhaps that's it.

Try replacing the blx by a plain bl.

On 30 November 2011 20:08, Shubhangi Tamsekar
<shubhangi.tamsekar@...> wrote:
> asm-arm.S:26: Error: selected processor does not support `blx r1'