Little endianness

Little endianness

[Held Bier]

Hi.

Looks like PA-RISC 1.1 added bi-endian support. 
What about PA-RISC Linux? I know it's big endian. 
But does it possible to build little endian kernel? If so, how?
Is migration to little endian planned? It should be reasonable 
as it will bring more  compatibility with other Linux world.

Thanks.

Re: Little endianness

[John David Anglin]

On 2015-02-10 5:31 AM, Held Bier wrote:
> Hi.
>
> Looks like PA-RISC 1.1 added bi-endian support.
> What about PA-RISC Linux? I know it's big endian.
> But does it possible to build little endian kernel? If so, how?
> Is migration to little endian planned? It should be reasonable
> as it will bring more  compatibility with other Linux world.
Unlikely.  According to the PA 2.0 arch, implementation of this feature 
was optional.

Further the PDC/IODC firmware and software tool chains need little 
endian support.  This
is a massive amount of work.  So, unless one one decided to create a 
64-bit embedded
PA-RISC processor for mass deployment, it wouldn't be worth the effort.

Dave

-- 
John David Anglin  dave.anglin@bell.net

Re: Little endianness

[James Bottomley]

On Tue, 2015-02-10 at 10:30 -0500, John David Anglin wrote:
> On 2015-02-10 5:31 AM, Held Bier wrote:
> > Hi.
> >
> > Looks like PA-RISC 1.1 added bi-endian support.
> > What about PA-RISC Linux? I know it's big endian.
> > But does it possible to build little endian kernel? If so, how?
> > Is migration to little endian planned? It should be reasonable
> > as it will bring more  compatibility with other Linux world.

What compatibility?  x86 is LE, Arm is Schizophrenic, PPC, Sparc and PA
are BE.  We're required to support the bus standards anyway, so the SCSI
bus is BE and the PCI bus LE by spec.  There's no such thing as a
"compatible" endianness.

> Unlikely.  According to the PA 2.0 arch, implementation of this feature 
> was optional.

I think it was originally designed for some type of LASI system.
There's an endian flip in some of the LASI devices as well.  I think the
original problem was LASI was designed big endian, but the chosen SCSI
processor was designed for LE at that time, so they tried to flip the
processor.  I suspect the reason the feature became optional is that it
caused more problems than it solved.

> Further the PDC/IODC firmware and software tool chains need little 
> endian support.  This
> is a massive amount of work.  So, unless one one decided to create a 
> 64-bit embedded
> PA-RISC processor for mass deployment, it wouldn't be worth the effort.

It's a bit pointless too.  Coping with endianness isn't really an
unsolved OS problem, so it would cost a huge amount, open a large can of
worms and buy us nothing.

James

Re: Little endianness

[lausgans]

> James Bottomley <James.Bottomley@HansenPartnership.com>:
>=20
> On Tue, 2015-02-10 at 10:30 -0500, John David Anglin wrote:
>> On 2015-02-10 5:31 AM, Held Bier wrote:
>>> Hi.
>>>=20
>>> Looks like PA-RISC 1.1 added bi-endian support.
>>> What about PA-RISC Linux? I know it's big endian.
>>> But does it possible to build little endian kernel? If so, how?
>>> Is migration to little endian planned? It should be reasonable
>>> as it will bring more  compatibility with other Linux world.
>=20
> What compatibility?  x86 is LE, Arm is Schizophrenic, PPC, Sparc and =
PA
> are BE.  We're required to support the bus standards anyway, so the S=
CSI
> bus is BE and the PCI bus LE by spec.  There's no such thing as a
> "compatible" endianness.

Some software authors still assume that we're living in x86 only world,=
 or just don't care. This leads to issues when one wish to run such sof=
tware on a BE Linux.

POWER is moving to LE:
https://www.ibm.com/developerworks/community/blogs/fe313521-2e95-46f2-8=
17d-44a4f27eba32/entry/just_the_faqs_about_little_endian
"Although Power already has Linux distributions and supporting applicat=
ions that run in big endian mode, the Linux application ecosystem for x=
86 platforms is much larger and Linux on x86 uses little endian mode.  =
Numerous clients, software partners, and IBM=E2=80=99s own software dev=
elopers have told us that porting their software to Power becomes simpl=
er if the Linux environment on Power supports little endian mode, more =
closely matching the environment provided by Linux on x86.  This new le=
vel of support will lower the barrier to entry for porting Linux on x86=
 software to Linux on Power."

=46or PPC the patch was made: http://lwn.net/Articles/408845/

>=20
>> Unlikely.  According to the PA 2.0 arch, implementation of this feat=
ure=20
>> was optional.
>=20
> I think it was originally designed for some type of LASI system.
> There's an endian flip in some of the LASI devices as well.  I think =
the
> original problem was LASI was designed big endian, but the chosen SCS=
I
> processor was designed for LE at that time, so they tried to flip the
> processor.  I suspect the reason the feature became optional is that =
it
> caused more problems than it solved.
>=20
>> Further the PDC/IODC firmware and software tool chains need little=20
>> endian support.  This
>> is a massive amount of work.  So, unless one one decided to create a=
=20
>> 64-bit embedded
>> PA-RISC processor for mass deployment, it wouldn't be worth the effo=
rt.
>=20
> It's a bit pointless too.  Coping with endianness isn't really an
> unsolved OS problem, so it would cost a huge amount, open a large can=
 of
> worms and buy us nothing.
>=20
> James
>=20
>=20

--
To unsubscribe from this list: send the line "unsubscribe linux-parisc"=
 in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Little endianness

[James Bottomley]

On Sun, 2015-02-15 at 09:49 +0300, lausgans@gmail.com wrote:
> > James Bottomley <James.Bottomley@HansenPartnership.com>:
> >=20
> > On Tue, 2015-02-10 at 10:30 -0500, John David Anglin wrote:
> >> On 2015-02-10 5:31 AM, Held Bier wrote:
> >>> Hi.
> >>>=20
> >>> Looks like PA-RISC 1.1 added bi-endian support.
> >>> What about PA-RISC Linux? I know it's big endian.
> >>> But does it possible to build little endian kernel? If so, how?
> >>> Is migration to little endian planned? It should be reasonable
> >>> as it will bring more  compatibility with other Linux world.
> >=20
> > What compatibility?  x86 is LE, Arm is Schizophrenic, PPC, Sparc an=
d PA
> > are BE.  We're required to support the bus standards anyway, so the=
 SCSI
> > bus is BE and the PCI bus LE by spec.  There's no such thing as a
> > "compatible" endianness.
>=20
> Some software authors still assume that we're living in x86 only
> world, or just don't care. This leads to issues when one wish to run
> such software on a BE Linux.
>=20
> POWER is moving to LE:
> https://www.ibm.com/developerworks/community/blogs/fe313521-2e95-46f2=
-817d-44a4f27eba32/entry/just_the_faqs_about_little_endian
> "Although Power already has Linux distributions and supporting
> applications that run in big endian mode, the Linux application
> ecosystem for x86 platforms is much larger and Linux on x86 uses
> little endian mode.  Numerous clients, software partners, and IBM=E2=80=
=99s
> own software developers have told us that porting their software to
> Power becomes simpler if the Linux environment on Power supports
> little endian mode, more closely matching the environment provided by
> Linux on x86.  This new level of support will lower the barrier to
> entry for porting Linux on x86 software to Linux on Power."
>=20
> For PPC the patch was made: http://lwn.net/Articles/408845/

That's not a correct inference from the facts.  Power is interested in
playing in the embedded space and some of the graphics SoCs available
there are LE only, so they're interested in enabling a LE version to ru=
n
in that space ... it won't affect the rest of the PPC boxes.

All our boxes either have no graphics or a HP special big endian card.
The only exception is the C8000 which has an ATI with special firmware
(also designed for BE).

The thing you don't seem to understand is that the endianness is
actually fixed by the firmware of the shipping systems.  You're tied to
whatever the firmware starts you in because swapping after the firmware
init is unbelievably painful, especially when you need to call the
firmware to perform OS tasks.  We don't have any new systems planned, s=
o
all the firmware is now fixed for PA and the endianness with it.

James




--
To unsubscribe from this list: send the line "unsubscribe linux-parisc"=
 in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: Little endianness

[⁣ ⁣]

2015-02-15 19:50 GMT+00:00 James Bottomley
<James.Bottomley@hansenpartnership.com>:
> On Sun, 2015-02-15 at 09:49 +0300, lausgans@gmail.com wrote:
>> > James Bottomley <James.Bottomley@HansenPartnership.com>:
>> >
>> > On Tue, 2015-02-10 at 10:30 -0500, John David Anglin wrote:
>> >> On 2015-02-10 5:31 AM, Held Bier wrote:
>> >>> Hi.
>> >>>
>> >>> Looks like PA-RISC 1.1 added bi-endian support.
>> >>> What about PA-RISC Linux? I know it's big endian.
>> >>> But does it possible to build little endian kernel? If so, how?
>> >>> Is migration to little endian planned? It should be reasonable
>> >>> as it will bring more  compatibility with other Linux world.
>> >
>> > What compatibility?  x86 is LE, Arm is Schizophrenic, PPC, Sparc a=
nd PA
>> > are BE.  We're required to support the bus standards anyway, so th=
e SCSI
>> > bus is BE and the PCI bus LE by spec.  There's no such thing as a
>> > "compatible" endianness.
>>
>> Some software authors still assume that we're living in x86 only
>> world, or just don't care. This leads to issues when one wish to run
>> such software on a BE Linux.
>>
>> POWER is moving to LE:
>> https://www.ibm.com/developerworks/community/blogs/fe313521-2e95-46f=
2-817d-44a4f27eba32/entry/just_the_faqs_about_little_endian
>> "Although Power already has Linux distributions and supporting
>> applications that run in big endian mode, the Linux application
>> ecosystem for x86 platforms is much larger and Linux on x86 uses
>> little endian mode.  Numerous clients, software partners, and IBM=E2=
=80=99s
>> own software developers have told us that porting their software to
>> Power becomes simpler if the Linux environment on Power supports
>> little endian mode, more closely matching the environment provided b=
y
>> Linux on x86.  This new level of support will lower the barrier to
>> entry for porting Linux on x86 software to Linux on Power."
>>
>> For PPC the patch was made: http://lwn.net/Articles/408845/
>
> That's not a correct inference from the facts.  Power is interested i=
n
> playing in the embedded space and some of the graphics SoCs available
> there are LE only, so they're interested in enabling a LE version to =
run
> in that space ... it won't affect the rest of the PPC boxes.
>
> All our boxes either have no graphics or a HP special big endian card=
=2E
> The only exception is the C8000 which has an ATI with special firmwar=
e
> (also designed for BE).
>
> The thing you don't seem to understand is that the endianness is
> actually fixed by the firmware of the shipping systems.  You're tied =
to
> whatever the firmware starts you in because swapping after the firmwa=
re
> init is unbelievably painful, especially when you need to call the
> firmware to perform OS tasks.  We don't have any new systems planned,=
 so
> all the firmware is now fixed for PA and the endianness with it.
>
> James

James,

This was a general question. In case if you're connecting dots with my
later topic "Video option for a big endian machine?" - I should rather
name it "... for a non-x86 arch", as there may be more platform
specific issues, than just byte order. By the links i've provided
there are thoughts that some graphics driver / stack (as well as many
other non-related programs) may still be not fixed, e.g. they may
don't use the host byte order at the CPU side operations, and so don't
work. Of course the proper solution is to fix the actual code, but
having a LE OS option for that hardware platform should be also an
ultimate so-so solution.
Sorry for my question then, that was just a curiosity, I'm in no event
asking anyone to sacrifice their forces and time just to bring LE to
PA Linux.

> The only exception is the C8000 which has an ATI with special firmwar=
e
> (also designed for BE).

This doesn't seem to be true:
- Once I've put ATI designed for (and working in, under HP-UX) C8000
into IA64 machine with LE Linux and it worked fine;
- If HP asked ATI to produce a custom firmware, what's the purpose of
putting x86 emulator into PDC? Actually both IA64 and PA-RISC ZX1
machines can POST most of relatively modern cards with unmodified
VBIOS.

To add, Nvidia cards fail similarly to ATI with ring buffer issue (GPU
locks up). If it's not the Nouveau breakage, than some PA Linux code
(possibly DMA related) does something wrong.

> The thing you don't seem to understand is that the endianness is
> actually fixed by the firmware of the shipping systems.  You're tied =
to
> whatever the firmware starts you in because swapping after the firmwa=
re
> init is unbelievably painful, especially when you need to call the
> firmware to perform OS tasks.

I'm sure you're right, i'm not expert there. I'm wondering why IA64
(or mentioned POWER) doesn't suffer from such obstacles?
https://forums.gentoo.org/viewtopic-p-2432985.html?sid=3D890fd4a4113673=
0c64d573192c9e3b57#2432985
--
To unsubscribe from this list: send the line "unsubscribe linux-parisc"=
 in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html