Re: [Qemu-devel] [PATCH] 64 bit I/O support v7

[Blue Swirl <blauwirbel@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 1070 bytes --]

On 5/2/09, Paul Brook <paul@codesourcery.com> wrote:
> > Should a new intermediate bus layer also do byte swapping?
>
>
> Yes, though a bus layer isn't actually a necessary prerequisite.
>
>  As I mentioned in the recent "Smarter compilation for target devices"
>  thread[1], having individual devices do byteswapping is bogus. A while ago I
>  reworked the TLB handling, so there should now be spare bits that can be used
>  for things like byteswapping.

This version does the byte swapping and width translation in exec.c.
Sparc32 OpenBIOS/Linux does not use the wrong sized accesses, so that
part may be untested.

Some problems:

- Small writes cause a read-modify-write cycle, but I think that may
happen with real devices too. Another solution is to write zero in
other byte lanes without reading.

- The LE/BE distinction does not seem to be very obvious for devices.
I think the board should give the is_le value to the device.

- There is a small performance hit for the native access case. This
could be avoided by extending io_mem_opaque to cover all sizes.

[-- Attachment #2: Emulate_non_native_widths.patch --]
[-- Type: application/mbox, Size: 22862 bytes --]

[-- Attachment #3: Use_new_registration_functions.patch --]
[-- Type: application/mbox, Size: 4761 bytes --]