On Thu, Jul 18, 2019 at 01:57:06PM +0800, Wei Wang wrote: > On 07/18/2019 12:13 PM, Michael S. Tsirkin wrote: > > > > It makes sense for pages in the balloon (requested by hypervisor). > > However free page hinting can freeze up lots of memory for its own > > internal reasons. It does not make sense to ask hypervisor > > to set flags in order to fix internal guest issues. > > Sounds reasonable to me. Probably we could move the flag check to > shrinker_count and shrinker_scan as a reclaiming condition for > ballooning pages only? I think so, yes. I also wonder whether we should stop reporting at that point - otherwise we'll just allocate the freed pages again. > > > > > Right. But that does not include the pages in the hint vq, > > which could be a significant amount of memory. > > I think it includes, as vb->num_free_page_blocks records the total number > of free page blocks that balloon has taken from mm. Oh - you are right. Thanks! > For shrink_free_pages, it calls return_free_pages_to_mm, which pops pages > from vb->free_page_list (this is the list where pages get enlisted after > they > are put to the hint vq, see get_free_page_and_send). > > > > > > > > > > - if free pages are being reported, pages freed > > > > by shrinker will just get re-allocated again > > > fill_balloon will re-try the allocation after sleeping 200ms once allocation fails. > > Even if ballon was never inflated, if shrinker frees some memory while > > we are hinting, hint vq will keep going and allocate it back without > > sleeping. > > Still see get_free_page_and_send. -EINTR is returned when page allocation > fails, > and reporting ends then. what if it does not fail? > > Shrinker is called on system memory pressure. On memory pressure > get_free_page_and_send will fail memory allocation, so it stops allocating > more. Memory pressure could be triggered by an unrelated allocation e.g. from another driver. > > > Best, > Wei
