ioremap() vs. ioremap_nocache()

ioremap() vs. ioremap_nocache()

[David Mosberger]

Jes Sorensen was kind enough to point out a longstanding
mis-conception that I had about ioremap(): I thought that ioremap()
was being deprecated in favor of ioremap_nocache() because the former
does not clearly define what kind of memory attribute will be used to
access the mapped memory (cached, write-through cached,
write-coalescing, etc).  But I seem to have been wrong about that.
Now, as far as I know, on x86, ioremap() will give write-through
cached mappings (in the absence of mtrr games).  If this is true, how
can this work?  There are many drivers out there that use ioremap() on
memory mapped I/O regions that do NOT have the "Prefetchable" bit set
in the PCI BAR.  For example, the eepro100 driver does this and it has
a routine called wait_for_cmd_done(), which spins on an ioremapped
read.  On an x86, what prevents these reads from being cached?

Thanks,

	--david

Re: ioremap() vs. ioremap_nocache()

[Alan Cox]

> Now, as far as I know, on x86, ioremap() will give write-through
> cached mappings (in the absence of mtrr games).  If this is true, how

On x86 ioremap will give mappings appropriate to the object you map - which
means by default it wil give uncached mappings. The PCI hardware will do
intelligent things in certain cases such as write merging

Alan

Re: ioremap() vs. ioremap_nocache()

[Jes Sorensen]

>>>>> "Alan" == Alan Cox <alan@lxorguk.ukuu.org.uk> writes:

>> Now, as far as I know, on x86, ioremap() will give write-through
>> cached mappings (in the absence of mtrr games).  If this is true,
>> how

Alan> On x86 ioremap will give mappings appropriate to the object you
Alan> map - which means by default it wil give uncached mappings. The
Alan> PCI hardware will do intelligent things in certain cases such as
Alan> write merging

Are you thereby saying that ioremap() and ioremap_nocache() are
identical on the x86?

Cheers,
Jes

Re: ioremap() vs. ioremap_nocache()

[Alan Cox]

> Alan> On x86 ioremap will give mappings appropriate to the object you
> Alan> map - which means by default it wil give uncached mappings. The
> Alan> PCI hardware will do intelligent things in certain cases such as
> Alan> write merging
> 
> Are you thereby saying that ioremap() and ioremap_nocache() are
> identical on the x86?

In certain peculiar cases - no. Think about an ioremap of an object mapped
onto the system bus as RAM.

Re: ioremap() vs. ioremap_nocache()

[David Mosberger]

>>>>> On Fri, 5 Oct 2001 16:51:45 +0100 (BST), Alan Cox <alan@lxorguk.ukuu.org.uk> said:

  >> >>>>> On Fri, 5 Oct 2001 12:18:07 +0100 (BST), Alan Cox
  >> <alan@lxorguk.ukuu.org.uk> said:
  >> 
  >> Can you tell me what the answer is to this question?

  >> When is a programmer supposed to use ioremap()
  >> vs. ioremap_nocache().

  Alan> You should never need ioremap_nocache for anything
  Alan> non-ultraweird.

Then who added it and why?

	--david