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:   Wed, 12 Jul 2023 10:45:22 +0100
From:   Jean-Philippe Brucker <jean-philippe@linaro.org>
To:     Baolu Lu <baolu.lu@linux.intel.com>
Cc:     "Tian, Kevin" <kevin.tian@intel.com>,
        Joerg Roedel <joro@8bytes.org>, Will Deacon <will@kernel.org>,
        Robin Murphy <robin.murphy@arm.com>,
        Jason Gunthorpe <jgg@ziepe.ca>,
        Nicolin Chen <nicolinc@nvidia.com>,
        "Liu, Yi L" <yi.l.liu@intel.com>,
        Jacob Pan <jacob.jun.pan@linux.intel.com>,
        "iommu@lists.linux.dev" <iommu@lists.linux.dev>,
        "kvm@vger.kernel.org" <kvm@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH 3/9] iommu: Add common code to handle IO page faults
Message-ID: <20230712094522.GB507884@myrica>
References: <20230711010642.19707-1-baolu.lu@linux.intel.com>
 <20230711010642.19707-4-baolu.lu@linux.intel.com>
 <BN9PR11MB52761F71BA509501C1766E9A8C31A@BN9PR11MB5276.namprd11.prod.outlook.com>
 <cbbe1175-40f3-805e-02c2-f887b3289f04@linux.intel.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <cbbe1175-40f3-805e-02c2-f887b3289f04@linux.intel.com>
Precedence: bulk

On Wed, Jul 12, 2023 at 10:32:13AM +0800, Baolu Lu wrote:
> > btw is there value of moving the group handling logic from
> > iommu_queue_iopf() to this common function?
> > 
> > I wonder whether there is any correctness issue if not forwarding
> > partial request to iommufd. If not this can also help reduce
> > notifications to the user until the group is ready.
> 
> I don't think there's any correctness issue. But it should be better if
> we can inject the page faults to vm guests as soon as possible. There's
> no requirement to put page requests to vIOMMU's hardware page request
> queue at the granularity of a fault group. Thoughts?

Not sure I understand you correctly, but we can't inject partial fault
groups: if the HW PRI queue overflows, the last fault in a group may be
lost, so the non-last faults in that group already injected won't be
completed (until PRGI reuse), leaking PRI request credits and guest
resources.

Thanks,
Jean