IBM 440GP 36-bit addressing

IBM 440GP 36-bit addressing

[Vishwanath]

Hi,

I am new to PPC Linux. I am working on IBM 440 GP board. I am using
MontaVista Linux source code. I didn't understand how 36-bit addressing is
taken care in this code. If an application has to access the a memory
location in SDRAM of 440 GP or its internal peripheral registers, whether it
has to pass 36-bit address or normal 32-bit address.

I am afraid my doubt is silly one. But still I am totally confused.

Hope to get reply soon.

Thanks and regards,
Vishwanath


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: IBM 440GP 36-bit addressing

[Hollis Blanchard]

On Fri, 2002-08-23 at 10:00, Vishwanath wrote:
>
> I am new to PPC Linux. I am working on IBM 440 GP board. I am using
> MontaVista Linux source code. I didn't understand how 36-bit addressing is
> taken care in this code. If an application has to access the a memory
> location in SDRAM of 440 GP or its internal peripheral registers, whether it
> has to pass 36-bit address or normal 32-bit address.

It's a 36-bit real address space, but 32-bit effective (aka "virtual").
Applications, which run entirely in effective address space, have no
reason to think about such things and require no modification.

The kernel has to care about the real address space, mostly when you
want to ioremap (all the on-chip peripheral addresses are higher than
4GB as I recall) a 36-bit real address to a 32-bit effective address so
you can use it in the kernel (which also runs in effective mode).

-Hollis


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: IBM 440GP 36-bit addressing

[Vishwanath]

Hi All,

Thanks for your information. I have few more doubts.

By going through the code, I understood that ioremap64() is used for mapping
the 36-bit real address onto 32-bit effective address space. How this
mapping information is conveyed to user applications ?

If I am writing assembly instructions to access 440GP registers, do I have
to pass 36-bit real address or 32-bit effective address ?

Physical address map of 440GP provides information only about Real
addresses.  How can I convert these real addresses onto effective addresses
? Do I have to write a TLB entry regarding my addresses ?

Thanks and Regards,
Vishwa


----- Original Message -----
From: "Hollis Blanchard" <hollis@austin.ibm.com>
To: <linuxppc-embedded@lists.linuxppc.org>
Sent: Friday, August 23, 2002 8:30 PM
Subject: Re: IBM 440GP 36-bit addressing


>
> On Fri, 2002-08-23 at 10:00, Vishwanath wrote:
> >
> > I am new to PPC Linux. I am working on IBM 440 GP board. I am using
> > MontaVista Linux source code. I didn't understand how 36-bit addressing
is
> > taken care in this code. If an application has to access the a memory
> > location in SDRAM of 440 GP or its internal peripheral registers,
whether it
> > has to pass 36-bit address or normal 32-bit address.
>
> It's a 36-bit real address space, but 32-bit effective (aka "virtual").
> Applications, which run entirely in effective address space, have no
> reason to think about such things and require no modification.
>
> The kernel has to care about the real address space, mostly when you
> want to ioremap (all the on-chip peripheral addresses are higher than
> 4GB as I recall) a 36-bit real address to a 32-bit effective address so
> you can use it in the kernel (which also runs in effective mode).
>
> -Hollis
>
>
>
>


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: IBM 440GP 36-bit addressing

[Hollis Blanchard]

On Sat, 2002-08-24 at 02:05, Vishwanath wrote:
>
> By going through the code, I understood that ioremap64() is used for mapping
> the 36-bit real address onto 32-bit effective address space. How this
> mapping information is conveyed to user applications ?

I don't understand the question. Do you mean "how does userspace know
what the kernel has ioremapped"? I don't think it does.

Do you mean "how can userspace map in a 36-bit real address"? The usual
(32-bit) way is to mmap(/dev/mem). I don't know if that will work with
36-bit addresses.

> If I am writing assembly instructions to access 440GP registers, do I have
> to pass 36-bit real address or 32-bit effective address ?

Do you mean a memory-mapped register, i.e. an on-chip peripheral like
the UART? If so, that's what we've been talking about. You must map the
36-bit real address into your 32-bit effective address space, then
access the effective address.

> Physical address map of 440GP provides information only about Real
> addresses.  How can I convert these real addresses onto effective addresses
> ? Do I have to write a TLB entry regarding my addresses ?

Exactly, that's what I mean by "mapping".

Of course, some TLB entries may already have been set up for you by the
kernel or firmware. Whatever the case, you need to make sure there is a
TLB entry in place.

-Hollis


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/

Re: IBM 440GP 36-bit addressing

[Eugene Surovegin]

Hollis,

At 08:48 AM 8/26/2002, you wrote:

>The usual(32-bit) way is to mmap(/dev/mem). I don't know if that will work
>with
>36-bit addresses.

It will work only for first 4G of physical address space.
/dev/mem cannot be used to access EPB/OCP or PCI address space on 440.

For this purpose "/dev/mem-like" device can be easily written that takes
into account 36-bit phys addresses.

Thanks,

  Eugene Surovegin <mailto:ebs@innocent.com>


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/