On Mon, Nov 27, 2023 at 1:00 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote: > > On Mon, 27 Nov 2023 11:37:03 -0800 Nhat Pham <nphamcs@xxxxxxxxx> wrote: > > > Currently, we only shrink the zswap pool when the user-defined limit is > > hit. This means that if we set the limit too high, cold data that are > > unlikely to be used again will reside in the pool, wasting precious > > memory. It is hard to predict how much zswap space will be needed ahead > > of time, as this depends on the workload (specifically, on factors such > > as memory access patterns and compressibility of the memory pages). > > > > This patch implements a memcg- and NUMA-aware shrinker for zswap, that > > is initiated when there is memory pressure. The shrinker does not > > have any parameter that must be tuned by the user, and can be opted in > > or out on a per-memcg basis. > > > > Furthermore, to make it more robust for many workloads and prevent > > overshrinking (i.e evicting warm pages that might be refaulted into > > memory), we build in the following heuristics: > > > > * Estimate the number of warm pages residing in zswap, and attempt to > > protect this region of the zswap LRU. > > * Scale the number of freeable objects by an estimate of the memory > > saving factor. The better zswap compresses the data, the fewer pages > > we will evict to swap (as we will otherwise incur IO for relatively > > small memory saving). > > * During reclaim, if the shrinker encounters a page that is also being > > brought into memory, the shrinker will cautiously terminate its > > shrinking action, as this is a sign that it is touching the warmer > > region of the zswap LRU. > > > > As a proof of concept, we ran the following synthetic benchmark: > > build the linux kernel in a memory-limited cgroup, and allocate some > > cold data in tmpfs to see if the shrinker could write them out and > > improved the overall performance. Depending on the amount of cold data > > generated, we observe from 14% to 35% reduction in kernel CPU time used > > in the kernel builds. > > > > ... > > > > --- a/include/linux/mmzone.h > > +++ b/include/linux/mmzone.h > > @@ -22,6 +22,7 @@ > > #include <linux/mm_types.h> > > #include <linux/page-flags.h> > > #include <linux/local_lock.h> > > +#include <linux/zswap.h> > > #include <asm/page.h> > > > > /* Free memory management - zoned buddy allocator. */ > > @@ -641,6 +642,7 @@ struct lruvec { > > #ifdef CONFIG_MEMCG > > struct pglist_data *pgdat; > > #endif > > + struct zswap_lruvec_state zswap_lruvec_state; > > Normally we'd put this in #ifdef CONFIG_ZSWAP. > > > --- a/include/linux/zswap.h > > +++ b/include/linux/zswap.h > > @@ -5,20 +5,40 @@ > > #include <linux/types.h> > > #include <linux/mm_types.h> > > > > +struct lruvec; > > + > > extern u64 zswap_pool_total_size; > > extern atomic_t zswap_stored_pages; > > > > #ifdef CONFIG_ZSWAP > > > > +struct zswap_lruvec_state { > > + /* > > + * Number of pages in zswap that should be protected from the shrinker. > > + * This number is an estimate of the following counts: > > + * > > + * a) Recent page faults. > > + * b) Recent insertion to the zswap LRU. This includes new zswap stores, > > + * as well as recent zswap LRU rotations. > > + * > > + * These pages are likely to be warm, and might incur IO if the are written > > + * to swap. > > + */ > > + atomic_long_t nr_zswap_protected; > > +}; > > + > > bool zswap_store(struct folio *folio); > > bool zswap_load(struct folio *folio); > > void zswap_invalidate(int type, pgoff_t offset); > > void zswap_swapon(int type); > > void zswap_swapoff(int type); > > void zswap_memcg_offline_cleanup(struct mem_cgroup *memcg); > > - > > +void zswap_lruvec_state_init(struct lruvec *lruvec); > > +void zswap_lruvec_swapin(struct page *page); > > #else > > > > +struct zswap_lruvec_state {}; > > But instead you made it an empty struct in this case. > > That's a bit funky, but I guess OK. It does send a careful reader of > struct lruvec over to look at the zswap_lruvec_state definition to > understand what's going on. I agree. Originally, I included the fields in struct lruvec itself, guarded by a #ifdef CONFIG_ZSWAP as you pointed out here. Yosry gave me this suggestion to hide the zswap-specific states and details from ordinary lruvec user, and direct people who care and/or need to know about details towards the zswap codebase, which is good IMHO. It is a bit weird I admit, but in this case I think it works out. > > > static inline bool zswap_store(struct folio *folio) > > { > > return false; > > @@ -33,7 +53,8 @@ static inline void zswap_invalidate(int type, pgoff_t offset) {} > > static inline void zswap_swapon(int type) {} > > static inline void zswap_swapoff(int type) {} > > static inline void zswap_memcg_offline_cleanup(struct mem_cgroup *memcg) {} > > - > > +static inline void zswap_lruvec_init(struct lruvec *lruvec) {} > > +static inline void zswap_lruvec_swapin(struct page *page) {} > > Needed this build fix: > > --- a/include/linux/zswap.h~zswap-shrinks-zswap-pool-based-on-memory-pressure-fix > +++ a/include/linux/zswap.h > @@ -54,6 +54,7 @@ static inline void zswap_swapon(int type > static inline void zswap_swapoff(int type) {} > static inline void zswap_memcg_offline_cleanup(struct mem_cgroup *memcg) {} > static inline void zswap_lruvec_init(struct lruvec *lruvec) {} > +static inline void zswap_lruvec_state_init(struct lruvec *lruvec) {} > static inline void zswap_lruvec_swapin(struct page *page) {} > #endif > > _ > Yeah that looks like a typo on my part. My bad. v7 includes this fix.