[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20221107213625.939660-1-nphamcs@gmail.com>
Date: Mon, 7 Nov 2022 13:36:25 -0800
From: Nhat Pham <nphamcs@...il.com>
To: senozhatsky@...omium.org
Cc: hannes@...xchg.org, linux-mm@...ck.org,
linux-kernel@...r.kernel.org, minchan@...nel.org,
ngupta@...are.org, akpm@...ux-foundation.org, sjenning@...hat.com,
ddstreet@...e.org, vitaly.wool@...sulko.com
Subject: Re: [PATCH 4/5] zsmalloc: Add ops fields to zs_pool to store evict handlers
Essentially, the zpool constructor allows us to set things up with a null
struct zpool, zpool_ops, or zpool_ops->evict, which we have to handle. A
similar null-handling pattern can be observed in zbud (mm/zbud.c) and z3fold
(mm/z3fold.c) - see zbud_zpool_evict and zbud_zpool_create for e.g.
In particular:
1. pool->zpool_ops is the ops (containing the evict handler zpool_ops->evict)
passed into the zpool constructor (zs_zpool_create)
2. pool->ops/zs_zpool_ops (struct zs_ops) is a struct wrapping zs_zpool_evict,
which itself is a wrapper for the zpool evict handler (pool->zpool_ops->evict).
zs_zpool_evict also handles the case where zpool or zpool_ops is null, or
zpool_ops->evict is not defined (i.e return -ENOENT).
FWIW, I do think this is quite convoluted. In the long run, we might want to
simplify this design, but for this patch series I think it is wise to err on
the safe side and follow the other two allocators' design for consistency.
That said, while staring at the code again, I found a bug - in the case
pool->zpool_ops is null, pool->ops is undefined garbage. The v3 patch will fix
that to follow zbud's pattern (pool->ops = NULL in this case).
Powered by blists - more mailing lists