On Wed, Oct 15, 2008 at 07:07, Ben Warren <[EMAIL PROTECTED]> wrote: > Hi Olav, > > On Fri, Oct 10, 2008 at 11:53 AM, Olav Morken <[EMAIL PROTECTED]> wrote: >> >> On Fri, Oct 10, 2008 at 7:01 PM, Ben Warren <[EMAIL PROTECTED]> >> wrote: >> > What other architectures have you tried this on? >> >> None, as we don't have any other boards to test on. I do however believe >> that this change should have no side-effects. If any architectures >> relied on this function working as some sort of memory/io barrier, they >> would have problems with other functions such as ArpRequest, which >> doesn't have anything that will work as a memory/io barrier before the >> eth_send function. >> >> Of course, I could be wrong. I would certainly not suggest including >> this change without some more testing. >> >> The bug which causes this problem is in avr32-gcc, which makes >> assumptions about the alignement of IP_t when using volatile, and this >> change shouldn't be necessary once that bug is fixed. Until that bug >> is fixed, this change is needed for anyone trying to run U-Boot on >> this microcontroller. > > I don't mean to be a pain, it's just that this code is shared by everything, > so we need to be really careful. I agree with Haavard that the volatile > keyword is probably used much more than it should be in the networking > library. > > I'll pull this into the net/testing branch in the next couple of days, and > hopefully we'll get some volunteers to try it out on different > architectures.
That is great. FWIW: I have now tested it in qemu_mips, where it appears to work. (Had to revert "qemu-mips.h: Add CFI support" before I could test it.) -- Olav Morken _______________________________________________ U-Boot mailing list U-Boot@lists.denx.de http://lists.denx.de/mailman/listinfo/u-boot
