Received-SPF: pass (google.com: domain of linux-kernel+bounces-161475-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Message-ID: <35b0ffe54f5e4077d5ebee6b82ae9ccd568dd77d.camel@redhat.com>
Subject: Re: [PATCH 1/2] drm/nouveau/firmware: Fix SG_DEBUG error with
 nvkm_firmware_ctor()
From: Lyude Paul <lyude@redhat.com>
To: Timur Tabi <ttabi@nvidia.com>, "nouveau@lists.freedesktop.org"
	 <nouveau@lists.freedesktop.org>
Cc: "bskeggs@redhat.com" <bskeggs@redhat.com>, "kherbst@redhat.com"
 <kherbst@redhat.com>, "airlied@redhat.com" <airlied@redhat.com>, 
 "justinstitt@google.com" <justinstitt@google.com>,
 "dri-devel@lists.freedesktop.org" <dri-devel@lists.freedesktop.org>,
 "keescook@chromium.org" <keescook@chromium.org>, "airlied@gmail.com"
 <airlied@gmail.com>,  "dakr@redhat.com" <dakr@redhat.com>,
 "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>, 
 "daniel@ffwll.ch" <daniel@ffwll.ch>
Date: Sun, 28 Apr 2024 11:52:47 -0400
In-Reply-To: <4294e43a0165a36a063d74c7ef62fbdeb156f8c4.camel@nvidia.com>
References: <20240426154138.64643-1-lyude@redhat.com>
	 <4294e43a0165a36a063d74c7ef62fbdeb156f8c4.camel@nvidia.com>
Organization: Red Hat Inc.
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.52.1 (3.52.1-1.fc40) 
Precedence: bulk
MIME-Version: 1.0

On Fri, 2024-04-26 at 15:47 +0000, Timur Tabi wrote:
> On Fri, 2024-04-26 at 11:41 -0400, Lyude Paul wrote:
> > We hit this because when initializing firmware of type
> > NVKM_FIRMWARE_IMG_DMA we allocate coherent memory and then attempt
> > to
> > include that coherent memory in a scatterlist.=20
>=20
> I'm sure this patch is a good one, and I will try to test it soon,
> but I am
> very curious to know why including coherent memory in a scatterlist
> is bad.

Thanks for asking this as I think you unintentionally pointed out this
explanation I gave doesn't make sense - so I looked a bit more into it.
The issue isn't coherent memory in the scatterlist, the issue is that
we're allocating with dma_alloc_coherent(). And according to the source
in dma_alloc_attrs() (which dma_alloc_coherent() is just a wrapper)
for):

   /*
    * DMA allocations can never be turned back into a page pointer, so
    * requesting compound pages doesn't make sense (and can't even be
    * supported at all by various backends).
    */
   if (WARN_ON_ONCE(flag & __GFP_COMP))
   	return NULL;

Which explains the check in sg_set_buf() that this patch stops us from
hitting:

   BUG_ON(!virt_addr_valid(buf));

Scatterlists need page pointers (we use one later down here:)

   sg_set_page(sg, virt_to_page(buf), buflen, offset_in_page(buf));

But we can't get a page pointer from an allocation made by
dma_alloc_coherent() - but we can from vmalloc(). I'll fix the patch
explanation in the next version, I have to send out another version
anyhow since I realized that patch #2 still needs one more check to
work properly
--=20
Cheers,
 Lyude Paul (she/her)
 Software Engineer at Red Hat