On 14 November 2011 15:18, Avi Kivity <a...@redhat.com> wrote: > On 11/14/2011 05:15 PM, Peter Maydell wrote: >> One of the failure cases I was thinking of is that if PIE means >> the platform's loader puts things in a different bit of the >> address space this might break TCG's assumptions about maximum >> distances between the codegen buffer and host C code. (That's >> a bug in TCG really but it would still be near-to-release >> breakage.)
And indeed testing this on an ARM host running i386 TCG system mode, applying this patch causes qemu to fail at startup with a tcg abort due to an out of range jump. I've already said that ARM is going to be broken for 1.0 so that's not inherently a problem but it does indicate that we definitely need to test the other TCG target systems (and not just a "does it compile" test) if we want to put this change in. -- PMM
