LOADADDR and low physical addresses?

LOADADDR and low physical addresses?

[Matthew Dharm]

So, I've got an interesting problem... which has forced me to look at
the use of the LOADADDR variable in the Makefile, and try (quickly) to
brush up on my linker scripting...

Basically I've got a processor with on-chip registers that need to be
located in the first 512MByte of _physical_ address.  To make things
difficult, they cannot be re-located once placed (configuration is
done by a hardware config stream at reset).  It's only 16KBytes of
address, but since I recall that linux on mips can't (well, probably
can't) handle discontiguous memory maps (we discussed this about a
year ago, I think), I was looking for a good place to put them.

Now, I think my problems are solved if the LOADADDR variable works the
way I think it does -- that the kernel loads at that address, and only
uses memory from that point upwards.  So, if my LOADADDR is
0x80100000, then the first 0x100000 won't get used.  Of course, the
exception vectors are there, but that doesn't take up that much space.
So there should be a chunk of address space I can use for other
things, like this register bank.

Yes? No?  Is my understanding even close?

Matt

P.S. Of course, if this is right, then I need to figure out the
proper/best way to use the add_memory_region() function....

--
Matthew D. Dharm                            Senior Software Designer
Momentum Computer Inc.                      1815 Aston Ave.  Suite 107
(760) 431-8663 X-115                        Carlsbad, CA 92008-7310
Momentum Works For You                      www.momenco.com

Re: LOADADDR and low physical addresses?

[Jun Sun]

On Fri, Sep 06, 2002 at 01:04:28PM -0700, Matthew Dharm wrote:
> So, I've got an interesting problem... which has forced me to look at
> the use of the LOADADDR variable in the Makefile, and try (quickly) to
> brush up on my linker scripting...
> 
> Basically I've got a processor with on-chip registers that need to be
> located in the first 512MByte of _physical_ address.  To make things
> difficult, they cannot be re-located once placed (configuration is
> done by a hardware config stream at reset).  It's only 16KBytes of
> address, but since I recall that linux on mips can't (well, probably
> can't) handle discontiguous memory maps (we discussed this about a
> year ago, I think), I was looking for a good place to put them.
> 
> Now, I think my problems are solved if the LOADADDR variable works the
> way I think it does -- that the kernel loads at that address, and only
> uses memory from that point upwards.  So, if my LOADADDR is
> 0x80100000, then the first 0x100000 won't get used.  Of course, the
> exception vectors are there, but that doesn't take up that much space.
> So there should be a chunk of address space I can use for other
> things, like this register bank.
> 
> Yes? No?  Is my understanding even close?
> 

That is right.

The only catch is that if you make LOADADDR very high (as in the case
system ram starts at a high address), the kernel page
table will be very high too.  It starts from phys address 0.

Also if you map your control registers at near-zero phy address, don't you 
also have system RAM there too?  Normally it is not ok to have two
devices decoded at the same phys address.

> P.S. Of course, if this is right, then I need to figure out the
> proper/best way to use the add_memory_region() function....

Unless I misunderstand something here, I don't think you need 
to mess with add_memory_region().

Jun

RE: LOADADDR and low physical addresses?

[Matthew Dharm]

Yes, the having two devices at the same physical address might be a
problem, but one I _might_ be able to work around.  Not only do I have
a large bank of SDRAM, but I also have a small bank of on-chip SRAM.

So I'm thinking that the map will go (starting from 0) like this:
on-chip SRAM, control registers, main memory

And this is where I think the add_memory_region() magic might need to
happen.  Do I need to add the on-chip SRAM and control registers using
add_memory_region()?  Is it going to be okay to have a large and
mis-aligned bank of SDRAM?

Questions abound.  Needless to say, I'm going to have some choice
words with the chip designers over this...

Matt

--
Matthew D. Dharm                            Senior Software Designer
Momentum Computer Inc.                      1815 Aston Ave.  Suite 107
(760) 431-8663 X-115                        Carlsbad, CA 92008-7310
Momentum Works For You                      www.momenco.com

> -----Original Message-----
> From: Jun Sun [mailto:jsun@mvista.com]
> Sent: Friday, September 06, 2002 1:53 PM
> To: Matthew Dharm
> Cc: Linux-MIPS; jsun@mvista.com
> Subject: Re: LOADADDR and low physical addresses?
>
>
> On Fri, Sep 06, 2002 at 01:04:28PM -0700, Matthew Dharm wrote:
> > So, I've got an interesting problem... which has forced
> me to look at
> > the use of the LOADADDR variable in the Makefile, and try
> (quickly) to
> > brush up on my linker scripting...
> >
> > Basically I've got a processor with on-chip registers
> that need to be
> > located in the first 512MByte of _physical_ address.  To
> make things
> > difficult, they cannot be re-located once placed (configuration is
> > done by a hardware config stream at reset).  It's only 16KBytes of
> > address, but since I recall that linux on mips can't
> (well, probably
> > can't) handle discontiguous memory maps (we discussed this about a
> > year ago, I think), I was looking for a good place to put them.
> >
> > Now, I think my problems are solved if the LOADADDR
> variable works the
> > way I think it does -- that the kernel loads at that
> address, and only
> > uses memory from that point upwards.  So, if my LOADADDR is
> > 0x80100000, then the first 0x100000 won't get used.  Of
> course, the
> > exception vectors are there, but that doesn't take up
> that much space.
> > So there should be a chunk of address space I can use for other
> > things, like this register bank.
> >
> > Yes? No?  Is my understanding even close?
> >
>
> That is right.
>
> The only catch is that if you make LOADADDR very high (as
> in the case
> system ram starts at a high address), the kernel page
> table will be very high too.  It starts from phys address 0.
>
> Also if you map your control registers at near-zero phy
> address, don't you
> also have system RAM there too?  Normally it is not ok to have two
> devices decoded at the same phys address.
>
> > P.S. Of course, if this is right, then I need to figure out the
> > proper/best way to use the add_memory_region() function....
>
> Unless I misunderstand something here, I don't think you need
> to mess with add_memory_region().
>
> Jun
>

Re: LOADADDR and low physical addresses?

[Jun Sun]

On Fri, Sep 06, 2002 at 02:13:08PM -0700, Matthew Dharm wrote:
> Yes, the having two devices at the same physical address might be a
> problem, but one I _might_ be able to work around.  Not only do I have
> a large bank of SDRAM, but I also have a small bank of on-chip SRAM.
> 
> So I'm thinking that the map will go (starting from 0) like this:
> on-chip SRAM, control registers, main memory
> 
> And this is where I think the add_memory_region() magic might need to
> happen.  Do I need to add the on-chip SRAM and control registers using
> add_memory_region()?  

I don't think you have to.  I *think* it works if you don't.  Not sure
know if you actuall do add.

> Is it going to be okay to have a large and
> mis-aligned bank of SDRAM?

That should be ok.

Jun

Re: LOADADDR and low physical addresses?

[Maciej W. Rozycki]

On Fri, 6 Sep 2002, Jun Sun wrote:

> > And this is where I think the add_memory_region() magic might need to
> > happen.  Do I need to add the on-chip SRAM and control registers using
> > add_memory_region()?  
> 
> I don't think you have to.  I *think* it works if you don't.  Not sure
> know if you actuall do add.

 Well, you have to register usable RAM areas with add_memory_region(), if
you want to make them available to Linux.  Adding other ranges is optional
but you have to consider a part of the kernel may want to know if the
range is usable or occupied (consider e.g. the PCI setup code modifying
BARs).  It's also nice and sometimes useful to a user to let him see what
space is occupied.

 What you register with add_memory_region() is printed upon boot and
available from /proc/iomem.

-- 
+  Maciej W. Rozycki, Technical University of Gdansk, Poland   +
+--------------------------------------------------------------+
+        e-mail: macro@ds2.pg.gda.pl, PGP key available        +

RE: LOADADDR and low physical addresses?

[Manoj Ekbote]

-----Original Message-----
From: Matthew Dharm [mailto:mdharm@momenco.com]
Sent: Friday, September 06, 2002 2:13 PM
To: Jun Sun
Cc: Linux-MIPS
Subject: RE: LOADADDR and low physical addresses?


Yes, the having two devices at the same physical address might be a
problem, but one I _might_ be able to work around.  Not only do I have
a large bank of SDRAM, but I also have a small bank of on-chip SRAM.

So I'm thinking that the map will go (starting from 0) like this:
on-chip SRAM, control registers, main memory

And this is where I think the add_memory_region() magic might need to
happen.  Do I need to add the on-chip SRAM and control registers using
add_memory_region()?  


--I am pretty sure you don't have to do that.You just need to tell Linux 
--where the main memory starts at.

Manoj