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;
MIME-Version: 1.0
References: <20230808171446.2187795-1-mshavit@google.com> <20230809011204.v5.6.Ice063dcf87d1b777a72e008d9e3406d2bcf6d876@changeid>
 <20230809135024.GD4226@willie-the-truck>
In-Reply-To: <20230809135024.GD4226@willie-the-truck>
From:   Michael Shavit <mshavit@google.com>
Date:   Thu, 10 Aug 2023 17:23:37 +0800
Message-ID: <CAKHBV24ybm1gT+mQs1fDPZY_kGzop8+dQjccP+_qOFCmx69spQ@mail.gmail.com>
Subject: Re: [PATCH v5 6/9] iommu/arm-smmu-v3: Move CD table to arm_smmu_master
To:     Will Deacon <will@kernel.org>
Cc:     iommu@lists.linux.dev, linux-arm-kernel@lists.infradead.org,
        linux-kernel@vger.kernel.org, robin.murphy@arm.com,
        nicolinc@nvidia.com, jgg@nvidia.com, jean-philippe@linaro.org
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

>
> > @@ -2465,6 +2450,22 @@ static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
> >       if (smmu_domain->stage != ARM_SMMU_DOMAIN_BYPASS)
> >               master->ats_enabled = arm_smmu_ats_supported(master);
> >
> > +     if (smmu_domain->stage == ARM_SMMU_DOMAIN_S1) {
> > +             if (!master->cd_table.cdtab) {
> > +                     ret = arm_smmu_alloc_cd_tables(master);
> > +                     if (ret) {
> > +                             master->domain = NULL;
> > +                             return ret;
> > +                     }
> > +             }
> > +
> > +             ret = arm_smmu_write_ctx_desc(master, 0, &smmu_domain->cd);
> > +             if (ret) {
> > +                     master->domain = NULL;
> > +                     return ret;
>
> Can you leak the cd tables here if you just allocated them?

The CD table is only de-allocated when the SMMU device is released, so
this isn't "leaked" anymore than on a successful attachment. In a
previous version of this patch, this CD table was even pre-allocated
at probe time but is deferred to first attach following this
discussion: https://lore.kernel.org/lkml/ZMOzs1%2FxoEPX2+vA@nvidia.com/
.

> > @@ -2472,10 +2473,7 @@ static int arm_smmu_attach_dev(struct iommu_domain *domain, struct device *dev)
> >       spin_unlock_irqrestore(&smmu_domain->devices_lock, flags);
> >
> >       arm_smmu_enable_ats(master);
> > -
> > -out_unlock:
> > -     mutex_unlock(&smmu_domain->init_mutex);
> > -     return ret;
> > +     return 0;
> >  }
> >
> >  static int arm_smmu_map_pages(struct iommu_domain *domain, unsigned long iova,
> > @@ -2719,6 +2717,8 @@ static void arm_smmu_release_device(struct device *dev)
> >       arm_smmu_detach_dev(master);
> >       arm_smmu_disable_pasid(master);
> >       arm_smmu_remove_master(master);
> > +     if (master->cd_table.cdtab_dma)
>
> Why are you checking 'cdtab_dma' here instead of just 'cdtab'?

cd_table is statically allocated as part of the arm_smmu_master
struct. I suppose it could be allocated by arm_smmu_alloc_cd_tables()
instead?