[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH 0/3] xilinx: Speed up the build
From: |
Edgar E. Iglesias |
Subject: |
Re: [Qemu-devel] [PATCH 0/3] xilinx: Speed up the build |
Date: |
Sat, 9 Jun 2012 23:23:41 +0200 |
User-agent: |
Mutt/1.5.21 (2010-09-15) |
On Sat, Jun 09, 2012 at 05:31:39PM +0200, Andreas Färber wrote:
> Am 09.06.2012 17:20, schrieb Edgar E. Iglesias:
> > On Sat, Jun 09, 2012 at 03:54:28AM +0200, Andreas Färber wrote:
> >> xilinx_ethlite.c uses tswap32(). Have you ever tested this device to work
> >> on
> >> microblazeel? I wonder if we could change the device from
> >> DEVICE_NATIVE_ENDIAN
> >> to DEVICE_BIG_ENDIAN and in place of tswap32() use a bswap32() conditional
> >> on
> >> HOST_WORDS_BIGENDIAN so that it becomes independent of the target, too?
> >
> > I don't think that will work. the swap is needed if the endianness of the
> > host
> > is different from the one of the target...
>
> My thinking was: If we can force the device endianness to a known value
> then only the host endianness (not the target endianness) matters
> because the Memory API will take care of the device endianness.
>
> The question is: Is the device LE for microblazeel or is it always BE?
It can be both LE and BE
>
> > IIRC, the issue is that the device has a built-in RAM mapped so close to
> > the regs that they end up in the same "page". With Avis memory-api maybe
> > it's
> > possible to expose this sub-page area as a memory?
>
> Me and Avi fixed some bugs for subpage areas, it should work in theory
> (for TCG/qtest). Not being aware of a MicroBlaze KVM, if it's a RAM
> region then we should definitely model it as such.
If that's possible, that would get rid of the swaps
Cheers