On Fri, May 22, 2020 at 23:36:18 +0100, Peter Maydell wrote: > So is this: > (a) a TSan false positive, because we've analysed the use > of this struct field and know it's not a race because > [details], but which we're choosing to silence in this way > (b) an actual race for which the correct fix is to make the > accesses atomic because [details] > > ?
It is (b), as shown in the per-cpu lock series. In particular, see these two patches: - [PATCH v9 33/74] cpu: define cpu_interrupt_request helpers https://lists.nongnu.org/archive/html/qemu-devel/2020-05/msg06322.html - [PATCH v9 39/74] arm: convert to cpu_interrupt_request https://lists.nongnu.org/archive/html/qemu-devel/2020-05/msg06328.html Since a more thorough fix is included in that other series, I think this patch should be dropped from this series -- I'll post a reply to patch 00/19 with more details. Thanks, Emilio
