mips 32 24kec eabi. . .

mips 32 24kec eabi. . .

[ANDY KENNEDY]

All,

I derived a patch that I stole from the ARM folks for their eabi
settings.  I have the ability to make an elf toolchain that is eabi.
When attempting to build a Linux glibc toolchain, I get errors on the
tuple for building the base glibc as it doesn't understand
mipsel-linux-gnueabi.  Digging into the goings on I see the error
-mabicalls conflicts with -mabi=eabi.  I have searched every way I can
think of to locate the -mabicalls (which is the same as -mnoabicalls if
you didn't already know that) but it is not physically located in the
source of ct-ng.  Where is this coming from?

The comments around the selection for eabi state that this is not
applicable for mips, however, for my embedded processor, it is.  In
fact, this is a valid configuration according to the information from
gcc and (I think) also glibc. . .

Anyone have any ideas on where we are getting the -mabicalls?

Thanks,
Andy

--
For unsubscribe information see http://sourceware.org/lists.html#faq

Re: mips 32 24kec eabi. . .

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 1230 bytes --]

On 02 Jul 2015 18:14, ANDY KENNEDY wrote:
> I derived a patch that I stole from the ARM folks for their eabi
> settings.  I have the ability to make an elf toolchain that is eabi.
> When attempting to build a Linux glibc toolchain, I get errors on the
> tuple for building the base glibc as it doesn't understand
> mipsel-linux-gnueabi.  Digging into the goings on I see the error
> -mabicalls conflicts with -mabi=eabi.  I have searched every way I can
> think of to locate the -mabicalls (which is the same as -mnoabicalls if
> you didn't already know that) but it is not physically located in the
> source of ct-ng.  Where is this coming from?
> 
> The comments around the selection for eabi state that this is not
> applicable for mips, however, for my embedded processor, it is.  In
> fact, this is a valid configuration according to the information from
> gcc and (I think) also glibc. . .
> 
> Anyone have any ideas on where we are getting the -mabicalls?

glibc itself enforces one of three -mabi settings:
sysdeps/mips/mips32/Makefile:CC += -mabi=32
sysdeps/mips/mips64/n32/Makefile:CC += -mabi=n32
sysdeps/mips/mips64/n64/Makefile:CC += -mabi=64

that's going to conflict with -mabi=eabi.
-mike

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

RE: mips 32 24kec eabi. . .

[ANDY KENNEDY]

> -----Original Message-----
> From: crossgcc-owner@sourceware.org [mailto:crossgcc-owner@sourceware.org] On Behalf Of Mike
> Frysinger
> Sent: Monday, July 06, 2015 5:07 AM
> To: ANDY KENNEDY
> Cc: crossgcc@sourceware.org
> Subject: Re: mips 32 24kec eabi. . .
> 
> On 02 Jul 2015 18:14, ANDY KENNEDY wrote:
> > I derived a patch that I stole from the ARM folks for their eabi
> > settings.  I have the ability to make an elf toolchain that is eabi.
> > When attempting to build a Linux glibc toolchain, I get errors on the
> > tuple for building the base glibc as it doesn't understand
> > mipsel-linux-gnueabi.  Digging into the goings on I see the error
> > -mabicalls conflicts with -mabi=eabi.  I have searched every way I can
> > think of to locate the -mabicalls (which is the same as -mnoabicalls if
> > you didn't already know that) but it is not physically located in the
> > source of ct-ng.  Where is this coming from?
> >
> > The comments around the selection for eabi state that this is not
> > applicable for mips, however, for my embedded processor, it is.  In
> > fact, this is a valid configuration according to the information from
> > gcc and (I think) also glibc. . .
> >
> > Anyone have any ideas on where we are getting the -mabicalls?
> 
> glibc itself enforces one of three -mabi settings:
> sysdeps/mips/mips32/Makefile:CC += -mabi=32
> sysdeps/mips/mips64/n32/Makefile:CC += -mabi=n32
> sysdeps/mips/mips64/n64/Makefile:CC += -mabi=64
> 
> that's going to conflict with -mabi=eabi.
> -mike

Agreed.

It seems that we are required to use the eabi system calls at boot for
this processor.  Is that real or have we fallen off into the weeds on
this?  We are not (by far) mips experts.

Thanks for the clarification,
Andy

Re: mips 32 24kec eabi. . .

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 532 bytes --]

On 06 Jul 2015 16:00, ANDY KENNEDY wrote:
> It seems that we are required to use the eabi system calls at boot for
> this processor.  Is that real or have we fallen off into the weeds on
> this?  We are not (by far) mips experts.

if you only need it for the boot loader (e.g. u-boot), then you don't want a
linux/glibc toolchain.  people use something like mips-elf (*) instead and that
can either use newlib or forego the C library entirely.

(*) i haven't done mips/eabi myself, so the exact tuple here might be different.
-mike

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

RE: mips 32 24kec eabi. . .

[ANDY KENNEDY]

> -----Original Message-----
> From: crossgcc-owner@sourceware.org [mailto:crossgcc-owner@sourceware.org] On Behalf Of Mike
> Frysinger
> Sent: Monday, July 06, 2015 11:43 AM
> To: ANDY KENNEDY
> Cc: crossgcc@sourceware.org
> Subject: Re: mips 32 24kec eabi. . .
> 
> On 06 Jul 2015 16:00, ANDY KENNEDY wrote:
> > It seems that we are required to use the eabi system calls at boot for
> > this processor.  Is that real or have we fallen off into the weeds on
> > this?  We are not (by far) mips experts.
> 
> if you only need it for the boot loader (e.g. u-boot), then you don't want a
> linux/glibc toolchain.  people use something like mips-elf (*) instead and that
> can either use newlib or forego the C library entirely.

In the end, I'll need both Linux and u-Boot.  At the moment, I'd settle
for u-Boot only.  I cannot seem to make the eabi stuff work correctly.
I'm missing "qp" from the instruction set -- not that I have a clue
what that means :).  I have made a toolchain for Linux that worked after
booting through RedBoot -- but that same toolchain didn't seem to do the
right thing for building u-Boot.

> 
> (*) i haven't done mips/eabi myself, so the exact tuple here might be different.
> -mike

Yeah, even after making my changes, the guy using the toolchain informed
me that it still wasn't eabi, though, that is what I thought I had.

Thanks for your help!

Andy

Re: mips 32 24kec eabi. . .

[Mike Frysinger]

[-- Attachment #1: Type: text/plain, Size: 1264 bytes --]

On 06 Jul 2015 17:39, ANDY KENNEDY wrote:
> From: Mike Frysinger
> > On 06 Jul 2015 16:00, ANDY KENNEDY wrote:
> > > It seems that we are required to use the eabi system calls at boot for
> > > this processor.  Is that real or have we fallen off into the weeds on
> > > this?  We are not (by far) mips experts.
> > 
> > if you only need it for the boot loader (e.g. u-boot), then you don't want a
> > linux/glibc toolchain.  people use something like mips-elf (*) instead and that
> > can either use newlib or forego the C library entirely.
> 
> In the end, I'll need both Linux and u-Boot.  At the moment, I'd settle
> for u-Boot only.  I cannot seem to make the eabi stuff work correctly.
> I'm missing "qp" from the instruction set -- not that I have a clue
> what that means :).  I have made a toolchain for Linux that worked after
> booting through RedBoot -- but that same toolchain didn't seem to do the
> right thing for building u-Boot.

Linux (the kernel) can be built using the same bare metal toolchain as the boot 
loader, but it could probably also be built using a linux/glibc toolchain while 
also targetting EABI.  but i'm not sure what the status is of userland EABI 
MIPS.  you're probably in untested waters there.
-mike

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 819 bytes --]