Re: kswapd0: page allocation failure: order:0, mode:0x820(GFP_ATOMIC), nodemask=(null),cpuset=/,mems_allowed=0 (Kernel v6.5.9, 32bit ppc)

From: Sergey Senozhatsky
Date: Fri Jun 07 2024 - 07:23:34 EST


On (24/06/07 10:40), Nhat Pham wrote:
> Personally, I'm not super convinced about class locks. We're
> essentially relying on the post-compression size of the data to
> load-balance the queries - I can imagine a scenario where a workload
> has a concentrated distribution of post-compression data (i.e its
> pages are compressed to similar-ish sizes), and we're once again
> contending for a (few) lock(s) again.
>
> That said, I'll let the data tell the story :) We don't need a perfect
> solution, just a good enough solution for now.

Speaking of size class locks:

One thing to mention is that zsmalloc merges size classes, we never have
documented/claimed 256 size classe, the actual number is always much
much lower. Each such "cluster" (merged size classes) holds a range of
objects' sizes (e.g. 3504-3584 bytes). The wider the cluster's size range
the more likely the (size class) lock contention is.

Setting CONFIG_ZSMALLOC_CHAIN_SIZE to 10 or higher makes zsmalloc pool
to be configured with more size class clusters (which means that clusters
hold narrower size intervals).