Merge branch 'for-joerg/arm-smmu/updates' of git://git.kernel.org/pub/scm/linux/kernel/git/will/linux into arm/smmu

Documentation/admin-guide/kernel-parameters.txt

@@ -1749,6 +1749,18 @@
 		nobypass	[PPC/POWERNV]
 			Disable IOMMU bypass, using IOMMU for PCI devices.
 
+	iommu.strict=	[ARM64] Configure TLB invalidation behaviour
+			Format: { "0" | "1" }
+			0 - Lazy mode.
+			  Request that DMA unmap operations use deferred
+			  invalidation of hardware TLBs, for increased
+			  throughput at the cost of reduced device isolation.
+			  Will fall back to strict mode if not supported by
+			  the relevant IOMMU driver.
+			1 - Strict mode (default).
+			  DMA unmap operations invalidate IOMMU hardware TLBs
+			  synchronously.
+
 	iommu.passthrough=
 			[ARM64] Configure DMA to bypass the IOMMU by default.
 			Format: { "0" | "1" }

drivers/iommu/arm-smmu-v3.c

@@ -567,7 +567,8 @@ struct arm_smmu_device {
 
 	int				gerr_irq;
 	int				combined_irq;
-	atomic_t			sync_nr;
+	u32				sync_nr;
+	u8				prev_cmd_opcode;
 
 	unsigned long			ias; /* IPA */
 	unsigned long			oas; /* PA */
@@ -611,6 +612,7 @@ struct arm_smmu_domain {
 	struct mutex			init_mutex; /* Protects smmu pointer */
 
 	struct io_pgtable_ops		*pgtbl_ops;
+	bool				non_strict;
 
 	enum arm_smmu_domain_stage	stage;
 	union {
@@ -708,7 +710,7 @@ static void queue_inc_prod(struct arm_smmu_queue *q)
 }
 
 /*
- * Wait for the SMMU to consume items. If drain is true, wait until the queue
+ * Wait for the SMMU to consume items. If sync is true, wait until the queue
  * is empty. Otherwise, wait until there is at least one free slot.
  */
 static int queue_poll_cons(struct arm_smmu_queue *q, bool sync, bool wfe)
@@ -901,6 +903,8 @@ static void arm_smmu_cmdq_insert_cmd(struct arm_smmu_device *smmu, u64 *cmd)
 	struct arm_smmu_queue *q = &smmu->cmdq.q;
 	bool wfe = !!(smmu->features & ARM_SMMU_FEAT_SEV);
 
+	smmu->prev_cmd_opcode = FIELD_GET(CMDQ_0_OP, cmd[0]);
+
 	while (queue_insert_raw(q, cmd) == -ENOSPC) {
 		if (queue_poll_cons(q, false, wfe))
 			dev_err_ratelimited(smmu->dev, "CMDQ timeout\n");
@@ -948,15 +952,21 @@ static int __arm_smmu_cmdq_issue_sync_msi(struct arm_smmu_device *smmu)
 	struct arm_smmu_cmdq_ent ent = {
 		.opcode = CMDQ_OP_CMD_SYNC,
 		.sync	= {
-			.msidata = atomic_inc_return_relaxed(&smmu->sync_nr),
 			.msiaddr = virt_to_phys(&smmu->sync_count),
 		},
 	};
 
-	arm_smmu_cmdq_build_cmd(cmd, &ent);
-
 	spin_lock_irqsave(&smmu->cmdq.lock, flags);
-	arm_smmu_cmdq_insert_cmd(smmu, cmd);
+
+	/* Piggy-back on the previous command if it's a SYNC */
+	if (smmu->prev_cmd_opcode == CMDQ_OP_CMD_SYNC) {
+		ent.sync.msidata = smmu->sync_nr;
+	} else {
+		ent.sync.msidata = ++smmu->sync_nr;
+		arm_smmu_cmdq_build_cmd(cmd, &ent);
+		arm_smmu_cmdq_insert_cmd(smmu, cmd);
+	}
+
 	spin_unlock_irqrestore(&smmu->cmdq.lock, flags);
 
 	return __arm_smmu_sync_poll_msi(smmu, ent.sync.msidata);
@@ -1398,6 +1408,12 @@ static void arm_smmu_tlb_inv_context(void *cookie)
 		cmd.tlbi.vmid	= smmu_domain->s2_cfg.vmid;
 	}
 
+	/*
+	 * NOTE: when io-pgtable is in non-strict mode, we may get here with
+	 * PTEs previously cleared by unmaps on the current CPU not yet visible
+	 * to the SMMU. We are relying on the DSB implicit in queue_inc_prod()
+	 * to guarantee those are observed before the TLBI. Do be careful, 007.
+	 */
 	arm_smmu_cmdq_issue_cmd(smmu, &cmd);
 	__arm_smmu_tlb_sync(smmu);
 }
@@ -1624,6 +1640,9 @@ static int arm_smmu_domain_finalise(struct iommu_domain *domain)
 	if (smmu->features & ARM_SMMU_FEAT_COHERENCY)
 		pgtbl_cfg.quirks = IO_PGTABLE_QUIRK_NO_DMA;
 
+	if (smmu_domain->non_strict)
+		pgtbl_cfg.quirks |= IO_PGTABLE_QUIRK_NON_STRICT;
+
 	pgtbl_ops = alloc_io_pgtable_ops(fmt, &pgtbl_cfg, smmu_domain);
 	if (!pgtbl_ops)
 		return -ENOMEM;
@@ -1772,6 +1791,14 @@ arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
 	return ops->unmap(ops, iova, size);
 }
 
+static void arm_smmu_flush_iotlb_all(struct iommu_domain *domain)
+{
+	struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
+
+	if (smmu_domain->smmu)
+		arm_smmu_tlb_inv_context(smmu_domain);
+}
+
 static void arm_smmu_iotlb_sync(struct iommu_domain *domain)
 {
 	struct arm_smmu_device *smmu = to_smmu_domain(domain)->smmu;
@@ -1917,15 +1944,27 @@ static int arm_smmu_domain_get_attr(struct iommu_domain *domain,
 {
 	struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
 
-	if (domain->type != IOMMU_DOMAIN_UNMANAGED)
-		return -EINVAL;
-
-	switch (attr) {
-	case DOMAIN_ATTR_NESTING:
-		*(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
-		return 0;
+	switch (domain->type) {
+	case IOMMU_DOMAIN_UNMANAGED:
+		switch (attr) {
+		case DOMAIN_ATTR_NESTING:
+			*(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
+			return 0;
+		default:
+			return -ENODEV;
+		}
+		break;
+	case IOMMU_DOMAIN_DMA:
+		switch (attr) {
+		case DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE:
+			*(int *)data = smmu_domain->non_strict;
+			return 0;
+		default:
+			return -ENODEV;
+		}
+		break;
 	default:
-		return -ENODEV;
+		return -EINVAL;
 	}
 }
 
@@ -1935,26 +1974,37 @@ static int arm_smmu_domain_set_attr(struct iommu_domain *domain,
 	int ret = 0;
 	struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
 
-	if (domain->type != IOMMU_DOMAIN_UNMANAGED)
-		return -EINVAL;
-
 	mutex_lock(&smmu_domain->init_mutex);
 
-	switch (attr) {
-	case DOMAIN_ATTR_NESTING:
-		if (smmu_domain->smmu) {
-			ret = -EPERM;
-			goto out_unlock;
+	switch (domain->type) {
+	case IOMMU_DOMAIN_UNMANAGED:
+		switch (attr) {
+		case DOMAIN_ATTR_NESTING:
+			if (smmu_domain->smmu) {
+				ret = -EPERM;
+				goto out_unlock;
+			}
+
+			if (*(int *)data)
+				smmu_domain->stage = ARM_SMMU_DOMAIN_NESTED;
+			else
+				smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
+			break;
+		default:
+			ret = -ENODEV;
+		}
+		break;
+	case IOMMU_DOMAIN_DMA:
+		switch(attr) {
+		case DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE:
+			smmu_domain->non_strict = *(int *)data;
+			break;
+		default:
+			ret = -ENODEV;
 		}
-
-		if (*(int *)data)
-			smmu_domain->stage = ARM_SMMU_DOMAIN_NESTED;
-		else
-			smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
-
 		break;
 	default:
-		ret = -ENODEV;
+		ret = -EINVAL;
 	}
 
 out_unlock:
@@ -1999,7 +2049,7 @@ static struct iommu_ops arm_smmu_ops = {
 	.attach_dev		= arm_smmu_attach_dev,
 	.map			= arm_smmu_map,
 	.unmap			= arm_smmu_unmap,
-	.flush_iotlb_all	= arm_smmu_iotlb_sync,
+	.flush_iotlb_all	= arm_smmu_flush_iotlb_all,
 	.iotlb_sync		= arm_smmu_iotlb_sync,
 	.iova_to_phys		= arm_smmu_iova_to_phys,
 	.add_device		= arm_smmu_add_device,
@@ -2180,7 +2230,6 @@ static int arm_smmu_init_structures(struct arm_smmu_device *smmu)
 {
 	int ret;
 
-	atomic_set(&smmu->sync_nr, 0);
 	ret = arm_smmu_init_queues(smmu);
 	if (ret)
 		return ret;
@@ -2353,8 +2402,8 @@ static int arm_smmu_setup_irqs(struct arm_smmu_device *smmu)
 	irq = smmu->combined_irq;
 	if (irq) {
 		/*
-		 * Cavium ThunderX2 implementation doesn't not support unique
-		 * irq lines. Use single irq line for all the SMMUv3 interrupts.
+		 * Cavium ThunderX2 implementation doesn't support unique irq
+		 * lines. Use a single irq line for all the SMMUv3 interrupts.
 		 */
 		ret = devm_request_threaded_irq(smmu->dev, irq,
 					arm_smmu_combined_irq_handler,

drivers/iommu/arm-smmu.c

@@ -246,6 +246,7 @@ struct arm_smmu_domain {
 	const struct iommu_gather_ops	*tlb_ops;
 	struct arm_smmu_cfg		cfg;
 	enum arm_smmu_domain_stage	stage;
+	bool				non_strict;
 	struct mutex			init_mutex; /* Protects smmu pointer */
 	spinlock_t			cb_lock; /* Serialises ATS1* ops and TLB syncs */
 	struct iommu_domain		domain;
@@ -447,7 +448,11 @@ static void arm_smmu_tlb_inv_context_s1(void *cookie)
 	struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
 	void __iomem *base = ARM_SMMU_CB(smmu_domain->smmu, cfg->cbndx);
 
-	writel_relaxed(cfg->asid, base + ARM_SMMU_CB_S1_TLBIASID);
+	/*
+	 * NOTE: this is not a relaxed write; it needs to guarantee that PTEs
+	 * cleared by the current CPU are visible to the SMMU before the TLBI.
+	 */
+	writel(cfg->asid, base + ARM_SMMU_CB_S1_TLBIASID);
 	arm_smmu_tlb_sync_context(cookie);
 }
 
@@ -457,7 +462,8 @@ static void arm_smmu_tlb_inv_context_s2(void *cookie)
 	struct arm_smmu_device *smmu = smmu_domain->smmu;
 	void __iomem *base = ARM_SMMU_GR0(smmu);
 
-	writel_relaxed(smmu_domain->cfg.vmid, base + ARM_SMMU_GR0_TLBIVMID);
+	/* NOTE: see above */
+	writel(smmu_domain->cfg.vmid, base + ARM_SMMU_GR0_TLBIVMID);
 	arm_smmu_tlb_sync_global(smmu);
 }
 
@@ -469,6 +475,9 @@ static void arm_smmu_tlb_inv_range_nosync(unsigned long iova, size_t size,
 	bool stage1 = cfg->cbar != CBAR_TYPE_S2_TRANS;
 	void __iomem *reg = ARM_SMMU_CB(smmu_domain->smmu, cfg->cbndx);
 
+	if (smmu_domain->smmu->features & ARM_SMMU_FEAT_COHERENT_WALK)
+		wmb();
+
 	if (stage1) {
 		reg += leaf ? ARM_SMMU_CB_S1_TLBIVAL : ARM_SMMU_CB_S1_TLBIVA;
 
@@ -510,6 +519,9 @@ static void arm_smmu_tlb_inv_vmid_nosync(unsigned long iova, size_t size,
 	struct arm_smmu_domain *smmu_domain = cookie;
 	void __iomem *base = ARM_SMMU_GR0(smmu_domain->smmu);
 
+	if (smmu_domain->smmu->features & ARM_SMMU_FEAT_COHERENT_WALK)
+		wmb();
+
 	writel_relaxed(smmu_domain->cfg.vmid, base + ARM_SMMU_GR0_TLBIVMID);
 }
 
@@ -863,6 +875,9 @@ static int arm_smmu_init_domain_context(struct iommu_domain *domain,
 	if (smmu->features & ARM_SMMU_FEAT_COHERENT_WALK)
 		pgtbl_cfg.quirks = IO_PGTABLE_QUIRK_NO_DMA;
 
+	if (smmu_domain->non_strict)
+		pgtbl_cfg.quirks |= IO_PGTABLE_QUIRK_NON_STRICT;
+
 	smmu_domain->smmu = smmu;
 	pgtbl_ops = alloc_io_pgtable_ops(fmt, &pgtbl_cfg, smmu_domain);
 	if (!pgtbl_ops) {
@@ -1252,6 +1267,14 @@ static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
 	return ops->unmap(ops, iova, size);
 }
 
+static void arm_smmu_flush_iotlb_all(struct iommu_domain *domain)
+{
+	struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
+
+	if (smmu_domain->tlb_ops)
+		smmu_domain->tlb_ops->tlb_flush_all(smmu_domain);
+}
+
 static void arm_smmu_iotlb_sync(struct iommu_domain *domain)
 {
 	struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
@@ -1470,15 +1493,27 @@ static int arm_smmu_domain_get_attr(struct iommu_domain *domain,
 {
 	struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
 
-	if (domain->type != IOMMU_DOMAIN_UNMANAGED)
-		return -EINVAL;
-
-	switch (attr) {
-	case DOMAIN_ATTR_NESTING:
-		*(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
-		return 0;
+	switch(domain->type) {
+	case IOMMU_DOMAIN_UNMANAGED:
+		switch (attr) {
+		case DOMAIN_ATTR_NESTING:
+			*(int *)data = (smmu_domain->stage == ARM_SMMU_DOMAIN_NESTED);
+			return 0;
+		default:
+			return -ENODEV;
+		}
+		break;
+	case IOMMU_DOMAIN_DMA:
+		switch (attr) {
+		case DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE:
+			*(int *)data = smmu_domain->non_strict;
+			return 0;
+		default:
+			return -ENODEV;
+		}
+		break;
 	default:
-		return -ENODEV;
+		return -EINVAL;
 	}
 }
 
@@ -1488,28 +1523,38 @@ static int arm_smmu_domain_set_attr(struct iommu_domain *domain,
 	int ret = 0;
 	struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
 
-	if (domain->type != IOMMU_DOMAIN_UNMANAGED)
-		return -EINVAL;
-
 	mutex_lock(&smmu_domain->init_mutex);
 
-	switch (attr) {
-	case DOMAIN_ATTR_NESTING:
-		if (smmu_domain->smmu) {
-			ret = -EPERM;
-			goto out_unlock;
+	switch(domain->type) {
+	case IOMMU_DOMAIN_UNMANAGED:
+		switch (attr) {
+		case DOMAIN_ATTR_NESTING:
+			if (smmu_domain->smmu) {
+				ret = -EPERM;
+				goto out_unlock;
+			}
+
+			if (*(int *)data)
+				smmu_domain->stage = ARM_SMMU_DOMAIN_NESTED;
+			else
+				smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
+			break;
+		default:
+			ret = -ENODEV;
+		}
+		break;
+	case IOMMU_DOMAIN_DMA:
+		switch (attr) {
+		case DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE:
+			smmu_domain->non_strict = *(int *)data;
+			break;
+		default:
+			ret = -ENODEV;
 		}
-
-		if (*(int *)data)
-			smmu_domain->stage = ARM_SMMU_DOMAIN_NESTED;
-		else
-			smmu_domain->stage = ARM_SMMU_DOMAIN_S1;
-
 		break;
 	default:
-		ret = -ENODEV;
+		ret = -EINVAL;
 	}
-
 out_unlock:
 	mutex_unlock(&smmu_domain->init_mutex);
 	return ret;
@@ -1562,7 +1607,7 @@ static struct iommu_ops arm_smmu_ops = {
 	.attach_dev		= arm_smmu_attach_dev,
 	.map			= arm_smmu_map,
 	.unmap			= arm_smmu_unmap,
-	.flush_iotlb_all	= arm_smmu_iotlb_sync,
+	.flush_iotlb_all	= arm_smmu_flush_iotlb_all,
 	.iotlb_sync		= arm_smmu_iotlb_sync,
 	.iova_to_phys		= arm_smmu_iova_to_phys,
 	.add_device		= arm_smmu_add_device,

drivers/iommu/dma-iommu.c

@@ -55,6 +55,9 @@ struct iommu_dma_cookie {
 	};
 	struct list_head		msi_page_list;
 	spinlock_t			msi_lock;
+
+	/* Domain for flush queue callback; NULL if flush queue not in use */
+	struct iommu_domain		*fq_domain;
 };
 
 static inline size_t cookie_msi_granule(struct iommu_dma_cookie *cookie)
@@ -257,6 +260,20 @@ static int iova_reserve_iommu_regions(struct device *dev,
 	return ret;
 }
 
+static void iommu_dma_flush_iotlb_all(struct iova_domain *iovad)
+{
+	struct iommu_dma_cookie *cookie;
+	struct iommu_domain *domain;
+
+	cookie = container_of(iovad, struct iommu_dma_cookie, iovad);
+	domain = cookie->fq_domain;
+	/*
+	 * The IOMMU driver supporting DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE
+	 * implies that ops->flush_iotlb_all must be non-NULL.
+	 */
+	domain->ops->flush_iotlb_all(domain);
+}
+
 /**
  * iommu_dma_init_domain - Initialise a DMA mapping domain
  * @domain: IOMMU domain previously prepared by iommu_get_dma_cookie()
@@ -275,6 +292,7 @@ int iommu_dma_init_domain(struct iommu_domain *domain, dma_addr_t base,
 	struct iommu_dma_cookie *cookie = domain->iova_cookie;
 	struct iova_domain *iovad = &cookie->iovad;
 	unsigned long order, base_pfn, end_pfn;
+	int attr;
 
 	if (!cookie || cookie->type != IOMMU_DMA_IOVA_COOKIE)
 		return -EINVAL;
@@ -308,6 +326,13 @@ int iommu_dma_init_domain(struct iommu_domain *domain, dma_addr_t base,
 	}
 
 	init_iova_domain(iovad, 1UL << order, base_pfn);
+
+	if (!cookie->fq_domain && !iommu_domain_get_attr(domain,
+			DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE, &attr) && attr) {
+		cookie->fq_domain = domain;
+		init_iova_flush_queue(iovad, iommu_dma_flush_iotlb_all, NULL);
+	}
+
 	if (!dev)
 		return 0;
 
@@ -393,6 +418,9 @@ static void iommu_dma_free_iova(struct iommu_dma_cookie *cookie,
 	/* The MSI case is only ever cleaning up its most recent allocation */
 	if (cookie->type == IOMMU_DMA_MSI_COOKIE)
 		cookie->msi_iova -= size;
+	else if (cookie->fq_domain)	/* non-strict mode */
+		queue_iova(iovad, iova_pfn(iovad, iova),
+				size >> iova_shift(iovad), 0);
 	else
 		free_iova_fast(iovad, iova_pfn(iovad, iova),
 				size >> iova_shift(iovad));
@@ -408,7 +436,9 @@ static void __iommu_dma_unmap(struct iommu_domain *domain, dma_addr_t dma_addr,
 	dma_addr -= iova_off;
 	size = iova_align(iovad, size + iova_off);
 
-	WARN_ON(iommu_unmap(domain, dma_addr, size) != size);
+	WARN_ON(iommu_unmap_fast(domain, dma_addr, size) != size);
+	if (!cookie->fq_domain)
+		iommu_tlb_sync(domain);
 	iommu_dma_free_iova(cookie, dma_addr, size);
 }
 

drivers/iommu/io-pgtable-arm-v7s.c

@@ -587,6 +587,7 @@ static size_t arm_v7s_split_blk_unmap(struct arm_v7s_io_pgtable *data,
 	}
 
 	io_pgtable_tlb_add_flush(&data->iop, iova, size, size, true);
+	io_pgtable_tlb_sync(&data->iop);
 	return size;
 }
 
@@ -642,6 +643,13 @@ static size_t __arm_v7s_unmap(struct arm_v7s_io_pgtable *data,
 				io_pgtable_tlb_sync(iop);
 				ptep = iopte_deref(pte[i], lvl);
 				__arm_v7s_free_table(ptep, lvl + 1, data);
+			} else if (iop->cfg.quirks & IO_PGTABLE_QUIRK_NON_STRICT) {
+				/*
+				 * Order the PTE update against queueing the IOVA, to
+				 * guarantee that a flush callback from a different CPU
+				 * has observed it before the TLBIALL can be issued.
+				 */
+				smp_wmb();
 			} else {
 				io_pgtable_tlb_add_flush(iop, iova, blk_size,
 							 blk_size, true);
@@ -712,7 +720,8 @@ static struct io_pgtable *arm_v7s_alloc_pgtable(struct io_pgtable_cfg *cfg,
 			    IO_PGTABLE_QUIRK_NO_PERMS |
 			    IO_PGTABLE_QUIRK_TLBI_ON_MAP |
 			    IO_PGTABLE_QUIRK_ARM_MTK_4GB |
-			    IO_PGTABLE_QUIRK_NO_DMA))
+			    IO_PGTABLE_QUIRK_NO_DMA |
+			    IO_PGTABLE_QUIRK_NON_STRICT))
 		return NULL;
 
 	/* If ARM_MTK_4GB is enabled, the NO_PERMS is also expected. */

drivers/iommu/io-pgtable-arm.c

@@ -574,13 +574,13 @@ static size_t arm_lpae_split_blk_unmap(struct arm_lpae_io_pgtable *data,
 			return 0;
 
 		tablep = iopte_deref(pte, data);
+	} else if (unmap_idx >= 0) {
+		io_pgtable_tlb_add_flush(&data->iop, iova, size, size, true);
+		io_pgtable_tlb_sync(&data->iop);
+		return size;
 	}
 
-	if (unmap_idx < 0)
-		return __arm_lpae_unmap(data, iova, size, lvl, tablep);
-
-	io_pgtable_tlb_add_flush(&data->iop, iova, size, size, true);
-	return size;
+	return __arm_lpae_unmap(data, iova, size, lvl, tablep);
 }
 
 static size_t __arm_lpae_unmap(struct arm_lpae_io_pgtable *data,
@@ -610,6 +610,13 @@ static size_t __arm_lpae_unmap(struct arm_lpae_io_pgtable *data,
 			io_pgtable_tlb_sync(iop);
 			ptep = iopte_deref(pte, data);
 			__arm_lpae_free_pgtable(data, lvl + 1, ptep);
+		} else if (iop->cfg.quirks & IO_PGTABLE_QUIRK_NON_STRICT) {
+			/*
+			 * Order the PTE update against queueing the IOVA, to
+			 * guarantee that a flush callback from a different CPU
+			 * has observed it before the TLBIALL can be issued.
+			 */
+			smp_wmb();
 		} else {
 			io_pgtable_tlb_add_flush(iop, iova, size, size, true);
 		}
@@ -772,7 +779,8 @@ arm_64_lpae_alloc_pgtable_s1(struct io_pgtable_cfg *cfg, void *cookie)
 	u64 reg;
 	struct arm_lpae_io_pgtable *data;
 
-	if (cfg->quirks & ~(IO_PGTABLE_QUIRK_ARM_NS | IO_PGTABLE_QUIRK_NO_DMA))
+	if (cfg->quirks & ~(IO_PGTABLE_QUIRK_ARM_NS | IO_PGTABLE_QUIRK_NO_DMA |
+			    IO_PGTABLE_QUIRK_NON_STRICT))
 		return NULL;
 
 	data = arm_lpae_alloc_pgtable(cfg);
@@ -864,7 +872,8 @@ arm_64_lpae_alloc_pgtable_s2(struct io_pgtable_cfg *cfg, void *cookie)
 	struct arm_lpae_io_pgtable *data;
 
 	/* The NS quirk doesn't apply at stage 2 */
-	if (cfg->quirks & ~IO_PGTABLE_QUIRK_NO_DMA)
+	if (cfg->quirks & ~(IO_PGTABLE_QUIRK_NO_DMA |
+			    IO_PGTABLE_QUIRK_NON_STRICT))
 		return NULL;
 
 	data = arm_lpae_alloc_pgtable(cfg);

drivers/iommu/io-pgtable.h

@@ -71,12 +71,17 @@ struct io_pgtable_cfg {
 	 *	be accessed by a fully cache-coherent IOMMU or CPU (e.g. for a
 	 *	software-emulated IOMMU), such that pagetable updates need not
 	 *	be treated as explicit DMA data.
+	 *
+	 * IO_PGTABLE_QUIRK_NON_STRICT: Skip issuing synchronous leaf TLBIs
+	 *	on unmap, for DMA domains using the flush queue mechanism for
+	 *	delayed invalidation.
 	 */
 	#define IO_PGTABLE_QUIRK_ARM_NS		BIT(0)
 	#define IO_PGTABLE_QUIRK_NO_PERMS	BIT(1)
 	#define IO_PGTABLE_QUIRK_TLBI_ON_MAP	BIT(2)
 	#define IO_PGTABLE_QUIRK_ARM_MTK_4GB	BIT(3)
 	#define IO_PGTABLE_QUIRK_NO_DMA		BIT(4)
+	#define IO_PGTABLE_QUIRK_NON_STRICT	BIT(5)
 	unsigned long			quirks;
 	unsigned long			pgsize_bitmap;
 	unsigned int			ias;

drivers/iommu/iommu.c

@@ -41,6 +41,7 @@ static unsigned int iommu_def_domain_type = IOMMU_DOMAIN_IDENTITY;
 #else
 static unsigned int iommu_def_domain_type = IOMMU_DOMAIN_DMA;
 #endif
+static bool iommu_dma_strict __read_mostly = true;
 
 struct iommu_callback_data {
 	const struct iommu_ops *ops;
@@ -131,6 +132,12 @@ static int __init iommu_set_def_domain_type(char *str)
 }
 early_param("iommu.passthrough", iommu_set_def_domain_type);
 
+static int __init iommu_dma_setup(char *str)
+{
+	return kstrtobool(str, &iommu_dma_strict);
+}
+early_param("iommu.strict", iommu_dma_setup);
+
 static ssize_t iommu_group_attr_show(struct kobject *kobj,
 				     struct attribute *__attr, char *buf)
 {
@@ -1072,6 +1079,13 @@ struct iommu_group *iommu_group_get_for_dev(struct device *dev)
 		group->default_domain = dom;
 		if (!group->domain)
 			group->domain = dom;
+
+		if (dom && !iommu_dma_strict) {
+			int attr = 1;
+			iommu_domain_set_attr(dom,
+					      DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE,
+					      &attr);
+		}
 	}
 
 	ret = iommu_group_add_device(group, dev);

include/linux/iommu.h

@@ -124,6 +124,7 @@ enum iommu_attr {
 	DOMAIN_ATTR_FSL_PAMU_ENABLE,
 	DOMAIN_ATTR_FSL_PAMUV1,
 	DOMAIN_ATTR_NESTING,	/* two stages of translation */
+	DOMAIN_ATTR_DMA_USE_FLUSH_QUEUE,
 	DOMAIN_ATTR_MAX,
 };