On 11/12/2015 12:41 PM, Andreas Färber wrote: > Am 11.11.2015 um 09:54 schrieb Markus Armbruster: >> Peter Maydell <peter.mayd...@linaro.org> writes: >>> On 25 August 2015 at 15:17, Markus Armbruster <arm...@redhat.com> wrote: >>>> Stumbled over this while throwing away old mail. Andreas, what do you >>>> think? >>> >>> Seems right to me -- I suspect the original properties code was >>> written with the assumption that the property field would be >>> inside the device struct (and so offsets are small). The array >>> properties code breaks that assumption by allocating a separate >>> lump of memory with the properties in it; so now there's no >>> guarantee that the two pointers being subtracted will be >>> within 4G of each other. >>> >>> Reviewed-by: Peter Maydell <peter.mayd...@linaro.org> >>> >>> Arguably for consistency the 'arrayoffset' struct member should >>> also be a ptrdiff_t, though our current uses of it are such >>> that it'll always be within int range. >> >> Andreas? > > Found it archived. I honestly don't think it's necessary in practice to > have 64-bit offsets on 64-bit host, but it builds okay, queued. Testing > got stuck in ahci though, investigating. > > Thanks, > Andreas >
Did you ever reproduce this, or does it seem to just be a race?
