Received-SPF: pass (google.com: domain of linux-wireless+bounces-8969-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Message-ID: <6e556f7b6b769c4d70be3e248b98d8d09d51452b.camel@sipsolutions.net>
Subject: Re: iwlwifi: Regression after migrating to 6.6.32
From: Johannes Berg <johannes@sipsolutions.net>
To: Baochen Qiang <quic_bqiang@quicinc.com>, Fabio Estevam
 <festevam@gmail.com>
Cc: miriam.rachel.korenblit@intel.com, kvalo@kernel.org, Jakub Kicinski
	 <kuba@kernel.org>, linux-wireless <linux-wireless@vger.kernel.org>, 
 linux-kernel <linux-kernel@vger.kernel.org>
Date: Thu, 13 Jun 2024 12:36:59 +0200
In-Reply-To: <35750452-e362-4dfa-803a-3360a4e16cd8@quicinc.com>
References: 
	<CAOMZO5A7+nxACoBPY0k8cOpVQByZtEV_N1489MK5wETHF_RXWA@mail.gmail.com>
	 <3fbb5317d9ff33ef1b60ca8297537335ce86a79d.camel@sipsolutions.net>
	 <CAOMZO5Aufe7zAE7TFVprvRreamYd9=RHjybjaEz2O9WaPksV=Q@mail.gmail.com>
	 <95163ee547da95964c16f87a38d3326ae4da3253.camel@sipsolutions.net>
	 <CAOMZO5CYDsh70u3To7HYXVki_MzzhFyCCHkigt_Es7o_+XG3oA@mail.gmail.com>
	 <7a8e220d77d7e30a0cfaf984404ef2f57eaa785f.camel@sipsolutions.net>
	 <CAOMZO5BktgtaSPzCf3WOOnkD2n+fj3FeQEfHeT7CYFL+tCHeaw@mail.gmail.com>
	 <fb60b7f5bcf5ba47be54398225075a5bfab7c141.camel@sipsolutions.net>
	 <CAOMZO5CMX_juW4-t6CSd2xdzXkFfBiamuSTjsTB80Ly_TUsxRA@mail.gmail.com>
	 <d49fcc32-bfa8-41d2-8666-af6256b7b4b4@quicinc.com>
	 <50101085cba7fc089339c96f531f797e27c632ff.camel@sipsolutions.net>
	 <35750452-e362-4dfa-803a-3360a4e16cd8@quicinc.com>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.52.2 (3.52.2-1.fc40) 
Precedence: bulk
MIME-Version: 1.0

On Thu, 2024-06-13 at 17:51 +0800, Baochen Qiang wrote:
> >=20
> > Not sure that means anything though, I really would have expected a 4k
> > buffer to be 4k aligned - why would you even want to map it across two
> > pages?
> IMO SWIOTLB manages buffers in a unit of IO_TLB_SIZE, it does not care ab=
out whether or not we are crossing page boundaries, no?

Well arguably it should have some 'natural' alignment guarantees?

> >=20
> > If it is SWIOTLB, shouldn't we get into this condition?
> >=20
> >         /*
> >          * Historically, swiotlb allocations >=3D PAGE_SIZE were guaran=
teed to be
> >          * page-aligned in the absence of any other alignment requireme=
nts.
> >          * 'alloc_align_mask' was later introduced to specify the align=
ment
> >          * explicitly, however this is passed as zero for streaming map=
pings
> >          * and so we preserve the old behaviour there in case any drive=
rs are
> >          * relying on it.
> >          */
> >         if (!alloc_align_mask && !iotlb_align_mask && alloc_size >=3D P=
AGE_SIZE)
> >                 alloc_align_mask =3D PAGE_SIZE - 1;
> >=20
> This is not present in 6.6.32 stable kernel.
>=20

Oh. So I guess try to pick up commit 14cebf689a78 ("swiotlb: Reinstate
page-alignment for mappings >=3D PAGE_SIZE").

johannes