Dsb(sy) carries a huge amount of overhead.. This seems like overkill for
coherent DMA. Replace it with dmb(oshld)/dmb(oshst)
This saves about 1.5% CPU on my tiny N1 setup serving 80Gb/s of Netflix traffic.
Note this depends on D60098 to apply.
Differential D60099
arm64: Use only required ordering for coherent dma Authored by gallatin on Mon, Sep 28, 5:48 PM. Tags None Referenced Files
Details
Diff Detail
Event Timeline |