early_ioremap for MIPS

early_ioremap for MIPS

[Florian Fainelli]

[-- Attachment #1: Type: text/plain, Size: 211 bytes --]

Hi list,

Is there any need for early_ioremap on MIPS ? Seems like only x86_64 is 
implementing it for now.

Thank you for your answer.
-- 
Cordialement, Florian Fainelli
------------------------------

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: early_ioremap for MIPS

[Maciej W. Rozycki]

On Thu, 7 Feb 2008, Florian Fainelli wrote:

> Is there any need for early_ioremap on MIPS ? Seems like only x86_64 is 
> implementing it for now.

 There is hardly any need as generally KSEG0/KSEG1 and XPHYS mappings 
fulfil the need and are always available, even before paging has been 
initialised.  Some 32-bit systems with devices outside low 512MB of 
physical address space could potentially benefit though.  I recall some 
Alchemy systems may fall into this category.

  Maciej

Re: early_ioremap for MIPS

[Thomas Bogendoerfer]

On Fri, Feb 08, 2008 at 11:05:12AM +0000, Maciej W. Rozycki wrote:
> On Thu, 7 Feb 2008, Florian Fainelli wrote:
> 
> > Is there any need for early_ioremap on MIPS ? Seems like only x86_64 is 
> > implementing it for now.
> 
> physical address space could potentially benefit though.  I recall some 
> Alchemy systems may fall into this category.

Jazz has the same problem. Right now it's solved by using wired tlb
entries. Which is sort of an early_ioremap.

Thomas.

-- 
Crap can work. Given enough thrust pigs will fly, but it's not necessary a
good idea.                                                [ RFC1925, 2.3 ]

Re: early_ioremap for MIPS

[Ralf Baechle]

On Fri, Feb 08, 2008 at 01:28:59PM +0100, Thomas Bogendoerfer wrote:

> Jazz has the same problem. Right now it's solved by using wired tlb
> entries. Which is sort of an early_ioremap.

One with a totally awkward API requiring the user having to know about
the underlying TLB organization.  A better implementation wouldn't be
hard to do, for anything that's outside of KSEG1's address range grab
a new TLB entry if needed and wire an entry into it.  Use the same
API as good old ioremap() and call the result early_ioremap().

  Ralf

Re: early_ioremap for MIPS

[Maciej W. Rozycki]

On Fri, 8 Feb 2008, Ralf Baechle wrote:

> > Jazz has the same problem. Right now it's solved by using wired tlb
> > entries. Which is sort of an early_ioremap.
> 
> One with a totally awkward API requiring the user having to know about
> the underlying TLB organization.  A better implementation wouldn't be
> hard to do, for anything that's outside of KSEG1's address range grab
> a new TLB entry if needed and wire an entry into it.  Use the same
> API as good old ioremap() and call the result early_ioremap().

 And presumably by the time paging has been set up for real, all the early 
allocations could get automatically graduated to ordinary ones, freeing up 
all the wired TLB entries set up so far and keeping the values of cookies 
obtained intact.

  Maciej

Re: early_ioremap for MIPS

[M. Warner Losh]

In message: <Pine.LNX.4.64N.0802081058350.7017@blysk.ds.pg.gda.pl>
            "Maciej W. Rozycki" <macro@linux-mips.org> writes:
: On Thu, 7 Feb 2008, Florian Fainelli wrote:
: 
: > Is there any need for early_ioremap on MIPS ? Seems like only x86_64 is 
: > implementing it for now.
: 
:  There is hardly any need as generally KSEG0/KSEG1 and XPHYS mappings 
: fulfil the need and are always available, even before paging has been 
: initialised.  Some 32-bit systems with devices outside low 512MB of 
: physical address space could potentially benefit though.  I recall some 
: Alchemy systems may fall into this category.

The Acer Pica machines, as well as the Deskstation Tynes, had devices
mapped outside of this range...  Of course Ralf will be able to say
more, if he chooses to jump into the way-back machine...

Warner

Re: early_ioremap for MIPS

[Maciej W. Rozycki]

On Sun, 10 Feb 2008, M. Warner Losh wrote:

> :  There is hardly any need as generally KSEG0/KSEG1 and XPHYS mappings 
> : fulfil the need and are always available, even before paging has been 
> : initialised.  Some 32-bit systems with devices outside low 512MB of 
> : physical address space could potentially benefit though.  I recall some 
> : Alchemy systems may fall into this category.
> 
> The Acer Pica machines, as well as the Deskstation Tynes, had devices
> mapped outside of this range...  Of course Ralf will be able to say
> more, if he chooses to jump into the way-back machine...

 Please note that it is not the sole existence of some hardware in the 
physical address range inaccessible through the mapped space that matters 
here.  It is the need to use such a piece of hardware before paging has 
been initialised that is of concern for early ioremap() and for most 
devices it means there is no issue.

  Maciej

Re: early_ioremap for MIPS

[Ralf Baechle]

On Sun, Feb 10, 2008 at 03:44:01PM -0700, M. Warner Losh wrote:

> The Acer Pica machines, as well as the Deskstation Tynes, had devices
> mapped outside of this range...  Of course Ralf will be able to say
> more, if he chooses to jump into the way-back machine...

Yes, I recall.  Unfortunately I don't have my PICA anymore; it's been a
hell of a machine by the standards of its days.  Anyway, designs which
just like the PICA need some ioremap - and preferably available early
during bootup - have been developed after the PICA and are still being
developped.

  Ralf