On 3/10/22 01:43, Dmitry Osipenko wrote: >> I think that, given virgl uses host storage, guest shrinker should be >> still useful.. so I think continue with this series. > Guest shrinker indeed will be useful for virgl today. I was already > questioning why virgl needs both host and guest storages, maybe it will > move to a host-only storage approach in the future. > > I think the variant with the timer expiration actually could be > interesting to try because it should allow host to purge its VM BOs by > itself, preventing host OOMs. While I was working on shrinker v2, I noticed that host-side allocation may take a significant time. So I decided to defer implementation of my idea of using timer-based expiration for host-only BOs. I'll need to examine it more. >> For host shrinker, I'll have to look more at when crosvm triggers >> balloon inflation. I could still go the MADV:WILLNEED_REPLACE >> approach instead, which does have the advantage of host kernel not >> relying on host userspace or vm having a chance to run in order to >> release pages. The downside (perhaps?) is it would be more specific >> to virtgpu-native-context and less so to virgl or venus (but I guess >> there doesn't currently exist a way for madvise to be useful for vk >> drivers). > I'll also take a look at what CrOS does, maybe it has some interesting > ideas. I looked at CrOS kernel and crosvm, and haven't noticed anything special there in regards to purging GPU's memory of VM on host-side memory pressure. If you'll find something, then please let me know. I sent out v2 of the shrinker series, but missed to CC you on it by accident, please find it here [1]. [1] https://lore.kernel.org/dri-devel/20220314224253.236359-1-dmitry.osipenko@xxxxxxxxxxxxx/T/#t
