https://bugs.kde.org/show_bug.cgi?id=415136
--- Comment #5 from ahashmi <assad.has...@linaro.org> --- Thanks for reviewing in detail. Good point about splitting up the large patch. I've uploaded the implementation and tests as 2 separate patches: arm64_v81_cas_impl.patch arm64_v81_cas_test.patch > I am going to need some context to make sense of this... The loop in this context is not looping for the same reason guest code using a CAS instruction would loop. In this case the cbne w8, loop instruction is checking to see if the previous store exclusive worked. The spec says that store exclusives return 0 in the w register if the store updated memory, 1 if it failed to update. This way we always make sure that memory has been updated before returning from the CAS. Previously it would have returned even if the store failed. > Also, what is the reason for changing the hardwired register numbers? Adding the store check loop means we need to compare the store's status value in a w register and since w8 is the scratch register we use that, leaving w1/x1 untouched after the load. -- You are receiving this mail because: You are watching all bug changes.
