Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 9 Aug 2022 21:57:50 +0900
From:   Sergey Senozhatsky <senozhatsky@chromium.org>
To:     Jiri Slaby <jirislaby@kernel.org>
Cc:     Sergey Senozhatsky <senozhatsky@chromium.org>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        minchan@kernel.org, ngupta@vflare.org, Jan Kara <jack@suse.com>,
        Ted Ts'o <tytso@mit.edu>,
        Andreas Dilger <adilger.kernel@dilger.ca>,
        Ext4 Developers List <linux-ext4@vger.kernel.org>,
        avromanov@sberdevices.ru, ddrokosov@sberdevices.ru
Subject: Re: ext2/zram issue [was: Linux 5.19]
Message-ID: <YvJZztvKtedJfeK5@google.com>
References: <CAHk-=wgrz5BBk=rCz7W28Fj_o02s0Xi0OEQ3H1uQgOdFvHgx0w@mail.gmail.com>
 <702b3187-14bf-b733-263b-20272f53105d@kernel.org>
 <c7c69f77-4ea2-3375-33f3-322a3d35def5@kernel.org>
 <8710b302-9415-458d-f8a2-b78cc3a96e49@kernel.org>
 <YvIeLHuEb9XDSp6N@google.com>
 <YvIk3SdC7wP3uItR@google.com>
 <YvImxBsHJcpNzC+i@google.com>
 <9fd860a8-4e4f-6a95-5c3f-1b3c4a76cf51@kernel.org>
 <b2662600-07cd-7125-d7c3-b7bbe1bad824@kernel.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <b2662600-07cd-7125-d7c3-b7bbe1bad824@kernel.org>
Precedence: bulk

On (22/08/09 14:45), Jiri Slaby wrote:
> On 09. 08. 22, 14:35, Jiri Slaby wrote:
> > But the installer is different. It just creates memory pressure, yet,
> > reclaim works well and is able to find memory and go on. I would say
> > atomic vs non-atomic retry in the original (pre-5.19) approach makes the
> > difference.
> 
> Sorry, I meant no-direct-reclaim (5.19) vs direct-reclaim (pre-5.19).

Sure, I understood.

This somehow makes me scratch my head and ask if we really want to
continue using per-CPU steams. We previously (many years ago) had
a list of idle compression streams, which would do compression in
preemptible context and we would have only one zs_malloc handle
allocation path, which would do direct reclaim (when needed)