Re: [linuxsh-dev] PCMCIA patch & question

[Greg B. <gb...@po...>]

NIIBE Yutaka wrote:
> 
> Greg Banks wrote:
>  >   This means that PCMCIA IO-space mappings are loaded
>  > into the TLB in exactly the same way as other mappings.
>  > In particular, wired TLB entries are *not* used.  But
>  > the downside is a small number of extra instructions in
>  > the main path for all TLB misses (PCMCIA and non-PCMCIA
>  > alike).
>  >
>  >   My question is, is this approach acceptable?  Is the
>  > performance hit in the TLB miss handler worth freeing
>  > up the TLB slots when PCMCIA is not being heavily used?
> 
> I think that wired TLB entry is better.  Some (or most) drivers don't
> expect TLB fault would occur with I/O access functions.

  Well, it doesn't seem to be hurting the IDE driver.  Also,
I've recently seen a lot of WinCE driver code and it uses
virtual mappings to access *all* IO registers, not just PCMCIA
IO-space, despite being able to do it directly for most devices.
There's no special support in the device drivers for handling
TLB misses during register accesses; it seems to just work.

  Anyway I'll look at reworking the mapping code to use wired TLB
entries.

Greg.
-- 
These are my opinions not PPIs.