Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.31 as permitted sender) client-ip=23.128.96.31;
Received-SPF: Pass (protection.outlook.com: domain of nvidia.com designates
 216.228.117.160 as permitted sender) receiver=protection.outlook.com;
 client-ip=216.228.117.160; helo=mail.nvidia.com; pr=C
Date:   Fri, 1 Dec 2023 14:12:28 -0800
From:   Nicolin Chen <nicolinc@nvidia.com>
To:     Jason Gunthorpe <jgg@nvidia.com>
CC:     "Tian, Kevin" <kevin.tian@intel.com>,
        "Liu, Yi L" <yi.l.liu@intel.com>,
        "joro@8bytes.org" <joro@8bytes.org>,
        "alex.williamson@redhat.com" <alex.williamson@redhat.com>,
        "robin.murphy@arm.com" <robin.murphy@arm.com>,
        "baolu.lu@linux.intel.com" <baolu.lu@linux.intel.com>,
        "cohuck@redhat.com" <cohuck@redhat.com>,
        "eric.auger@redhat.com" <eric.auger@redhat.com>,
        "kvm@vger.kernel.org" <kvm@vger.kernel.org>,
        "mjrosato@linux.ibm.com" <mjrosato@linux.ibm.com>,
        "chao.p.peng@linux.intel.com" <chao.p.peng@linux.intel.com>,
        "yi.y.sun@linux.intel.com" <yi.y.sun@linux.intel.com>,
        "peterx@redhat.com" <peterx@redhat.com>,
        "jasowang@redhat.com" <jasowang@redhat.com>,
        "shameerali.kolothum.thodi@huawei.com" 
        <shameerali.kolothum.thodi@huawei.com>,
        "lulu@redhat.com" <lulu@redhat.com>,
        "suravee.suthikulpanit@amd.com" <suravee.suthikulpanit@amd.com>,
        "iommu@lists.linux.dev" <iommu@lists.linux.dev>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "linux-kselftest@vger.kernel.org" <linux-kselftest@vger.kernel.org>,
        "Duan, Zhenzhong" <zhenzhong.duan@intel.com>,
        "joao.m.martins@oracle.com" <joao.m.martins@oracle.com>,
        "Zeng, Xin" <xin.zeng@intel.com>,
        "Zhao, Yan Y" <yan.y.zhao@intel.com>
Subject: Re: [PATCH v6 2/6] iommufd: Add IOMMU_HWPT_INVALIDATE
Message-ID: <ZWpaTD9dVge+suyv@Asurada-Nvidia>
References: <ZWaPM4p7yjJ0sEKk@Asurada-Nvidia>
 <20231129195804.GF436702@nvidia.com>
 <ZWe2PvatTkkyNCY5@Asurada-Nvidia>
 <20231130000816.GB1389974@nvidia.com>
 <ZWjzcEAAg8ptVH4A@Asurada-Nvidia>
 <20231201004523.GJ1389974@nvidia.com>
 <ZWlhLk3JVwX0hRt/@Asurada-Nvidia>
 <20231201125538.GK1389974@nvidia.com>
 <ZWo6z59tnmS8F2V7@Asurada-Nvidia>
 <20231201204340.GA1493156@nvidia.com>
MIME-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Content-Disposition: inline
In-Reply-To: <20231201204340.GA1493156@nvidia.com>
X-MS-Exchange-CrossTenant-OriginalArrivalTime: 01 Dec 2023 22:12:52.7264
 (UTC)
X-MS-Exchange-CrossTenant-Network-Message-Id: 7795332a-641a-45b8-230f-08dbf2baa9c7
X-MS-Exchange-CrossTenant-Id: 43083d15-7273-40c1-b7db-39efd9ccc17a
X-MS-Exchange-CrossTenant-OriginalAttributedTenantConnectingIp: TenantId=43083d15-7273-40c1-b7db-39efd9ccc17a;Ip=[216.228.117.160];Helo=[mail.nvidia.com]
X-MS-Exchange-CrossTenant-AuthSource: BL6PEPF0001AB50.namprd04.prod.outlook.com
X-MS-Exchange-CrossTenant-AuthAs: Anonymous
X-MS-Exchange-CrossTenant-FromEntityHeader: HybridOnPrem
X-MS-Exchange-Transport-CrossTenantHeadersStamped: DS7PR12MB6096
X-Spam-Status: No, score=-0.9 required=5.0 tests=DKIMWL_WL_HIGH,DKIM_SIGNED,
	DKIM_VALID,DKIM_VALID_AU,HEADER_FROM_DIFFERENT_DOMAINS,
	MAILING_LIST_MULTI,SPF_HELO_NONE,SPF_PASS,T_SCC_BODY_TEXT_LINE
	autolearn=unavailable autolearn_force=no version=3.4.6
X-Spam-Checker-Version: SpamAssassin 3.4.6 (2021-04-09) on morse.vger.email
Precedence: bulk
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org
X-Greylist: Sender passed SPF test, not delayed by milter-greylist-4.6.4 (morse.vger.email [0.0.0.0]); Fri, 01 Dec 2023 14:13:07 -0800 (PST)

On Fri, Dec 01, 2023 at 04:43:40PM -0400, Jason Gunthorpe wrote:
> On Fri, Dec 01, 2023 at 11:58:07AM -0800, Nicolin Chen wrote:
> > > It seems there is not a simple way to realize this error back to
> > > userspace since we can't block the global command queue and we proceed
> > > to complete commands that the real HW would not have completed.
> > > 
> > > To actually emulate this the gerror handler would have to capture all
> > > the necessary registers, return them back to the thread doing
> > > invalidate_user and all of that would have to return back to userspace
> > > to go into the virtual version of all the same registers.
> > > 
> > > Yes, it can be synchronous it seems, but we don't have any
> > > infrastructure in the driver to do this.
> > >
> > > Given this is pretty niche maybe we just don't support error
> > > forwarding and simply ensure it could be added to the uapi later?
> > 
> > If arm_smmu_cmdq_issue_cmdlist in arm_smmu_cache_invalidate_user
> > fails with ETIMEOUT, we polls the CONS register to get the error
> > code. This can cover CERROR_ABT and CERROR_ATC_INV.
> 
> Why is timeout linked to these two? Or rather, it doesn't have to be
> linked like that. Any gerror is effectively synchronous because it
> halts the queue and allows SW time to inspect which command failed and
> record the gerror flags. So each and every command can get an error
> indication.
> 
> Restarting the queue is done by putting sync in there to effectively
> nop the failed command and we hope for the best and let it rip.

I see that SMMU driver only restarts the queue when dealing with
CERROR_ILL. So only CERROR_ABT or CERROR_ATC_INV would result in
-ETIMEOUT.

> > As you remarked that we can't block the global CMDQ, so we have
> > to let a real CERROR_ILL go. Yet, we can make sure commands to
> > be fully sanitized before being issued, as we should immediately
> > reject faulty commands anyway, for errors such as unsupported op
> > codes, unzero-ed reserved fields, and unlinked vSIDs. This can
> > at least largely reduce the probability of a real CERROR_ILL.
> 
> I'm more a little more concerend with ATC_INV as a malfunctioning
> device can trigger this..

How about making sure that the invalidate handler always issues
one CMD_ATC_INV at a time, so each arm_smmu_cmdq_issue_cmdlist()
call has a chance to timeout? Then, we can simply know which one
in the user array fails.

> > So, combining these two, we can still have a basic synchronous
> > way by returning an errno to the invalidate ioctl? I see Kevin
> > replied something similar too.
> 
> It isn't enough information, you don't know which gerror bits to set
> and you don't know what cons index to stick to indicate the error
> triggering command with just a simple errno.
>
> It does need to return a bunch of data to get it all right.

The array structure returns req_num to indicate the index. This
works, even if the command consumption stops in the middle:
 * @req_num: Input the number of cache invalidation requests in the array.
 *           Output the number of requests successfully handled by kernel.

So we only need an error code of CERROR_ABT/ILL/ATC_INV.

Or am I missing some point here?

> Though again, there is no driver infrastructure to do all this and it
> doesn't seem that important so maybe we can just ensure there is a
> future extension possiblity and have userspace understand an errno
> means to generate CERROR_ILL on the last command in the batch?

Yea, it certainly can be a plan.

Thanks
Nicolin