Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 29 Jun 2023 07:44:03 +0200
From:   Petr =?UTF-8?B?VGVzYcWZw61r?= <petr@tesarici.cz>
To:     Christoph Hellwig <hch@lst.de>
Cc:     Petr Tesarik <petrtesarik@huaweicloud.com>,
        Marek Szyprowski <m.szyprowski@samsung.com>,
        Robin Murphy <robin.murphy@arm.com>,
        "open list:DMA MAPPING HELPERS" <iommu@lists.linux.dev>,
        open list <linux-kernel@vger.kernel.org>,
        Roberto Sassu <roberto.sassu@huaweicloud.com>,
        Kefeng Wang <wangkefeng.wang@huawei.com>
Subject: Re: [PATCH v1 0/2] swiotlb: Fix a couple of bugs in sizing areas
Message-ID: <20230629074403.7f8ed9d6@meshulam.tesarici.cz>
In-Reply-To: <20230629051238.GB16654@lst.de>
References: <cover.1687784289.git.petr.tesarik.ext@huawei.com>
        <20230626160725.5164ca74@meshulam.tesarici.cz>
        <20230629051238.GB16654@lst.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Thu, 29 Jun 2023 07:12:38 +0200
Christoph Hellwig <hch@lst.de> wrote:

> On Mon, Jun 26, 2023 at 04:07:25PM +0200, Petr Tesa=C5=99=C3=ADk wrote:
> > As an aside (and not directly related to the bugfixes themselves), I
> > wonder why the area size cannot be always equal to IO_TLB_SEGSIZE. Of
> > course, we would (usually) end up with more areas, but that should be
> > a good thing, shouldn't it? The area structure is quite small, so it
> > cannot be because of memory consumption concerns. The overhead of
> > taking an uncontended spinlock should also be negligible. =20
>=20
> What would be the benefit of this?

For me, as a newcomer to this code, the existence of areas _and_
segments was confusing, especially since segments are not represented by
a data structure. It took me more than one day to realize that slots
are grouped into segments for allocation, but changes to the slot
metadata are protected by area spinlocks. In my case, a segment crossed
an area boundary, so the area spinlock protected only half of the
allocated slots.

I could also ask differently: What is the benefit of grouping slots
into segments, and then again grouping segments into areas?

Petr T