Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Date:   Thu, 22 Nov 2018 12:06:42 +0100
From:   Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To:     Gao Xiang <gaoxiang25@huawei.com>
Cc:     devel@driverdev.osuosl.org, linux-erofs@lists.ozlabs.org,
        Chao Yu <yuchao0@huawei.com>,
        LKML <linux-kernel@vger.kernel.org>, weidu.du@huawei.com,
        Miao Xie <miaoxie@huawei.com>
Subject: Re: [PATCH 02/10] staging: erofs: fix race when the managed cache is
 enabled
Message-ID: <20181122110642.GE5287@kroah.com>
References: <20181120143425.43637-1-gaoxiang25@huawei.com>
 <20181120143425.43637-3-gaoxiang25@huawei.com>
 <20181122101711.GA3189@kroah.com>
 <55aed87f-196e-9048-6aae-db09bc497664@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <55aed87f-196e-9048-6aae-db09bc497664@huawei.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Thu, Nov 22, 2018 at 06:42:52PM +0800, Gao Xiang wrote:
> Hi Greg,
> 
> On 2018/11/22 18:17, Greg Kroah-Hartman wrote:
> > Any specific reason why you are not using the refcount.h api instead of
> > "doing it yourself" with atomic_inc/dec()?
> > 
> > I'm not rejecting this, just curious.
> 
> As I explained in the previous email,
> Re: [PATCH 04/10] staging: erofs: fix `erofs_workgroup_{try_to_freeze, unfreeze}'
> 
> we need such a function when the value is >= 0, it plays as a refcount,
> but when the refcount == EROFS_LOCKED_MAGIC (<0, but not 0 as refcount.h),
> and actually there is no need to introduce a seperate spinlock_t because
> we don't actually care about its performance (rarely locked). and
> the corresponding struct is too large for now, we need to decrease its size.

Why do you need to decrease the size?  How many of these structures are
created?

And you will care about the performance when a lock is being held, as is
evident by your logic to try to fix those issues in this patch series.
Using a "real" lock will solve all of that and keep you from having to
implement it all "by hand".

thanks,

greg k-h