Merge tag 'iommu-fixes-v6.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu

Pull iommu fixes from Joerg Roedel:

- Intel VT-d fixes from Lu Baolu:
- Boot kdump kernels with VT-d scalable mode on
- Calculate the right page table levels
- Fix two recursive locking issues
- Fix a lockdep splat issue

- AMD IOMMU fixes:
- Fix for completion-wait command to use full 64 bits of data
- Fix PASID related issue where GPU sound devices failed to
initialize

- Fix for Virtio-IOMMU to report correct caching behavior, needed for
use with VFIO

* tag 'iommu-fixes-v6.0-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu:
iommu: Fix false ownership failure on AMD systems with PASID activated
iommu/vt-d: Fix possible recursive locking in intel_iommu_init()
iommu/virtio: Fix interaction with VFIO
iommu/vt-d: Fix lockdep splat due to klist iteration in atomic context
iommu/vt-d: Fix recursive lock issue in iommu_flush_dev_iotlb()
iommu/vt-d: Correctly calculate sagaw value of IOMMU
iommu/vt-d: Fix kdump kernels boot failure with scalable mode
iommu/amd: use full 64-bit value in build_completion_wait()

drivers/iommu/amd/iommu.c

···
 	memset(cmd, 0, sizeof(*cmd));
 	cmd->data[0] = lower_32_bits(paddr) | CMD_COMPL_WAIT_STORE_MASK;
 	cmd->data[1] = upper_32_bits(paddr);
-	cmd->data[2] = data;
+	cmd->data[2] = lower_32_bits(data);
+	cmd->data[3] = upper_32_bits(data);
 	CMD_SET_TYPE(cmd, CMD_COMPL_WAIT);
 }
 

drivers/iommu/amd/iommu_v2.c

···
 	if (dev_state->domain == NULL)
 		goto out_free_states;
 
+	/* See iommu_is_default_domain() */
+	dev_state->domain->type = IOMMU_DOMAIN_IDENTITY;
 	amd_iommu_domain_direct_map(dev_state->domain);
 
 	ret = amd_iommu_domain_enable_v2(dev_state->domain, pasids);

drivers/iommu/intel/dmar.c

···
 	if (!dmar_in_use())
 		return 0;
 
+	/*
+	 * It's unlikely that any I/O board is hot added before the IOMMU
+	 * subsystem is initialized.
+	 */
+	if (IS_ENABLED(CONFIG_INTEL_IOMMU) && !intel_iommu_enabled)
+		return -EOPNOTSUPP;
+
 	if (dmar_detect_dsm(handle, DMAR_DSM_FUNC_DRHD)) {
 		tmp = handle;
 	} else {

drivers/iommu/intel/iommu.c

···
 	return re->hi & VTD_PAGE_MASK;
 }
 
-static inline void context_clear_pasid_enable(struct context_entry *context)
-{
-	context->lo &= ~(1ULL << 11);
-}
-
-static inline bool context_pasid_enabled(struct context_entry *context)
-{
-	return !!(context->lo & (1ULL << 11));
-}
-
-static inline void context_set_copied(struct context_entry *context)
-{
-	context->hi |= (1ull << 3);
-}
-
-static inline bool context_copied(struct context_entry *context)
-{
-	return !!(context->hi & (1ULL << 3));
-}
-
-static inline bool __context_present(struct context_entry *context)
-{
-	return (context->lo & 1);
-}
-
-bool context_present(struct context_entry *context)
-{
-	return context_pasid_enabled(context) ?
-	     __context_present(context) :
-	     __context_present(context) && !context_copied(context);
-}
-
 static inline void context_set_present(struct context_entry *context)
 {
 	context->lo |= 1;
···
 {
 	context->lo = 0;
 	context->hi = 0;
+}
+
+static inline bool context_copied(struct intel_iommu *iommu, u8 bus, u8 devfn)
+{
+	if (!iommu->copied_tables)
+		return false;
+
+	return test_bit(((long)bus << 8) | devfn, iommu->copied_tables);
+}
+
+static inline void
+set_context_copied(struct intel_iommu *iommu, u8 bus, u8 devfn)
+{
+	set_bit(((long)bus << 8) | devfn, iommu->copied_tables);
+}
+
+static inline void
+clear_context_copied(struct intel_iommu *iommu, u8 bus, u8 devfn)
+{
+	clear_bit(((long)bus << 8) | devfn, iommu->copied_tables);
 }
 
 /*
···
 	return !(addr_width < BITS_PER_LONG && pfn >> addr_width);
 }
 
+/*
+ * Calculate the Supported Adjusted Guest Address Widths of an IOMMU.
+ * Refer to 11.4.2 of the VT-d spec for the encoding of each bit of
+ * the returned SAGAW.
+ */
+static unsigned long __iommu_calculate_sagaw(struct intel_iommu *iommu)
+{
+	unsigned long fl_sagaw, sl_sagaw;
+
+	fl_sagaw = BIT(2) | (cap_fl1gp_support(iommu->cap) ? BIT(3) : 0);
+	sl_sagaw = cap_sagaw(iommu->cap);
+
+	/* Second level only. */
+	if (!sm_supported(iommu) || !ecap_flts(iommu->ecap))
+		return sl_sagaw;
+
+	/* First level only. */
+	if (!ecap_slts(iommu->ecap))
+		return fl_sagaw;
+
+	return fl_sagaw & sl_sagaw;
+}
+
 static int __iommu_calculate_agaw(struct intel_iommu *iommu, int max_gaw)
 {
 	unsigned long sagaw;
 	int agaw;
 
-	sagaw = cap_sagaw(iommu->cap);
-	for (agaw = width_to_agaw(max_gaw);
-	     agaw >= 0; agaw--) {
+	sagaw = __iommu_calculate_sagaw(iommu);
+	for (agaw = width_to_agaw(max_gaw); agaw >= 0; agaw--) {
 		if (test_bit(agaw, &sagaw))
 			break;
 	}
···
 {
 	struct device_domain_info *info;
 	int nid = NUMA_NO_NODE;
+	unsigned long flags;
 
-	spin_lock(&domain->lock);
+	spin_lock_irqsave(&domain->lock, flags);
 	list_for_each_entry(info, &domain->devices, link) {
 		/*
 		 * There could possibly be multiple device numa nodes as devices
···
 		if (nid != NUMA_NO_NODE)
 			break;
 	}
-	spin_unlock(&domain->lock);
+	spin_unlock_irqrestore(&domain->lock, flags);
 
 	return nid;
 }
···
 	struct root_entry *root = &iommu->root_entry[bus];
 	struct context_entry *context;
 	u64 *entry;
+
+	/*
+	 * Except that the caller requested to allocate a new entry,
+	 * returning a copied context entry makes no sense.
+	 */
+	if (!alloc && context_copied(iommu, bus, devfn))
+		return NULL;
 
 	entry = &root->lo;
 	if (sm_supported(iommu)) {
···
 }
 
 #ifdef CONFIG_DMAR_DEBUG
-static void pgtable_walk(struct intel_iommu *iommu, unsigned long pfn, u8 bus, u8 devfn)
+static void pgtable_walk(struct intel_iommu *iommu, unsigned long pfn,
+			 u8 bus, u8 devfn, struct dma_pte *parent, int level)
 {
-	struct device_domain_info *info;
-	struct dma_pte *parent, *pte;
-	struct dmar_domain *domain;
-	struct pci_dev *pdev;
-	int offset, level;
-
-	pdev = pci_get_domain_bus_and_slot(iommu->segment, bus, devfn);
-	if (!pdev)
-		return;
-
-	info = dev_iommu_priv_get(&pdev->dev);
-	if (!info || !info->domain) {
-		pr_info("device [%02x:%02x.%d] not probed\n",
-			bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
-		return;
-	}
-
-	domain = info->domain;
-	level = agaw_to_level(domain->agaw);
-	parent = domain->pgd;
-	if (!parent) {
-		pr_info("no page table setup\n");
-		return;
-	}
+	struct dma_pte *pte;
+	int offset;
 
 	while (1) {
 		offset = pfn_level_offset(pfn, level);
···
 	struct pasid_entry *entries, *pte;
 	struct context_entry *ctx_entry;
 	struct root_entry *rt_entry;
+	int i, dir_index, index, level;
 	u8 devfn = source_id & 0xff;
 	u8 bus = source_id >> 8;
-	int i, dir_index, index;
+	struct dma_pte *pgtable;
 
 	pr_info("Dump %s table entries for IOVA 0x%llx\n", iommu->name, addr);
 
···
 		ctx_entry->hi, ctx_entry->lo);
 
 	/* legacy mode does not require PASID entries */
-	if (!sm_supported(iommu))
+	if (!sm_supported(iommu)) {
+		level = agaw_to_level(ctx_entry->hi & 7);
+		pgtable = phys_to_virt(ctx_entry->lo & VTD_PAGE_MASK);
 		goto pgtable_walk;
+	}
 
 	/* get the pointer to pasid directory entry */
 	dir = phys_to_virt(ctx_entry->lo & VTD_PAGE_MASK);
···
 	for (i = 0; i < ARRAY_SIZE(pte->val); i++)
 		pr_info("pasid table entry[%d]: 0x%016llx\n", i, pte->val[i]);
 
+	if (pasid_pte_get_pgtt(pte) == PASID_ENTRY_PGTT_FL_ONLY) {
+		level = pte->val[2] & BIT_ULL(2) ? 5 : 4;
+		pgtable = phys_to_virt(pte->val[2] & VTD_PAGE_MASK);
+	} else {
+		level = agaw_to_level((pte->val[0] >> 2) & 0x7);
+		pgtable = phys_to_virt(pte->val[0] & VTD_PAGE_MASK);
+	}
+
 pgtable_walk:
-	pgtable_walk(iommu, addr >> VTD_PAGE_SHIFT, bus, devfn);
+	pgtable_walk(iommu, addr >> VTD_PAGE_SHIFT, bus, devfn, pgtable, level);
 }
 #endif
 
···
 			u8 bus, u8 devfn)
 {
 	struct device_domain_info *info;
+	unsigned long flags;
 
 	if (!iommu->qi)
 		return NULL;
 
-	spin_lock(&domain->lock);
+	spin_lock_irqsave(&domain->lock, flags);
 	list_for_each_entry(info, &domain->devices, link) {
 		if (info->iommu == iommu && info->bus == bus &&
 		    info->devfn == devfn) {
-			spin_unlock(&domain->lock);
+			spin_unlock_irqrestore(&domain->lock, flags);
 			return info->ats_supported ? info : NULL;
 		}
 	}
-	spin_unlock(&domain->lock);
+	spin_unlock_irqrestore(&domain->lock, flags);
 
 	return NULL;
 }
···
 {
 	struct device_domain_info *info;
 	bool has_iotlb_device = false;
+	unsigned long flags;
 
-	spin_lock(&domain->lock);
+	spin_lock_irqsave(&domain->lock, flags);
 	list_for_each_entry(info, &domain->devices, link) {
 		if (info->ats_enabled) {
 			has_iotlb_device = true;
···
 		}
 	}
 	domain->has_iotlb_device = has_iotlb_device;
-	spin_unlock(&domain->lock);
+	spin_unlock_irqrestore(&domain->lock, flags);
 }
 
 static void iommu_enable_dev_iotlb(struct device_domain_info *info)
···
 				  u64 addr, unsigned mask)
 {
 	struct device_domain_info *info;
+	unsigned long flags;
 
 	if (!domain->has_iotlb_device)
 		return;
 
-	spin_lock(&domain->lock);
+	spin_lock_irqsave(&domain->lock, flags);
 	list_for_each_entry(info, &domain->devices, link)
 		__iommu_flush_dev_iotlb(info, addr, mask);
-	spin_unlock(&domain->lock);
+	spin_unlock_irqrestore(&domain->lock, flags);
 }
 
 static void iommu_flush_iotlb_psi(struct intel_iommu *iommu,
···
 	if (iommu->domain_ids) {
 		bitmap_free(iommu->domain_ids);
 		iommu->domain_ids = NULL;
+	}
+
+	if (iommu->copied_tables) {
+		bitmap_free(iommu->copied_tables);
+		iommu->copied_tables = NULL;
 	}
 
 	/* free context mapping */
···
 		goto out_unlock;
 
 	ret = 0;
-	if (context_present(context))
+	if (context_present(context) && !context_copied(iommu, bus, devfn))
 		goto out_unlock;
 
 	/*
···
 	 * in-flight DMA will exist, and we don't need to worry anymore
 	 * hereafter.
 	 */
-	if (context_copied(context)) {
+	if (context_copied(iommu, bus, devfn)) {
 		u16 did_old = context_domain_id(context);
 
 		if (did_old < cap_ndoms(iommu->cap)) {
···
 			iommu->flush.flush_iotlb(iommu, did_old, 0, 0,
 						 DMA_TLB_DSI_FLUSH);
 		}
+
+		clear_context_copied(iommu, bus, devfn);
 	}
 
 	context_clear_entry(context);
···
 {
 	struct device_domain_info *info = dev_iommu_priv_get(dev);
 	struct intel_iommu *iommu;
+	unsigned long flags;
 	u8 bus, devfn;
 	int ret;
 
···
 	if (ret)
 		return ret;
 	info->domain = domain;
-	spin_lock(&domain->lock);
+	spin_lock_irqsave(&domain->lock, flags);
 	list_add(&info->link, &domain->devices);
-	spin_unlock(&domain->lock);
+	spin_unlock_irqrestore(&domain->lock, flags);
 
 	/* PASID table is mandatory for a PCI device in scalable mode. */
 	if (sm_supported(iommu) && !dev_is_real_dma_subdevice(dev)) {
···
 		/* Now copy the context entry */
 		memcpy(&ce, old_ce + idx, sizeof(ce));
 
-		if (!__context_present(&ce))
+		if (!context_present(&ce))
 			continue;
 
 		did = context_domain_id(&ce);
 		if (did >= 0 && did < cap_ndoms(iommu->cap))
 			set_bit(did, iommu->domain_ids);
 
-		/*
-		 * We need a marker for copied context entries. This
-		 * marker needs to work for the old format as well as
-		 * for extended context entries.
-		 *
-		 * Bit 67 of the context entry is used. In the old
-		 * format this bit is available to software, in the
-		 * extended format it is the PGE bit, but PGE is ignored
-		 * by HW if PASIDs are disabled (and thus still
-		 * available).
-		 *
-		 * So disable PASIDs first and then mark the entry
-		 * copied. This means that we don't copy PASID
-		 * translations from the old kernel, but this is fine as
-		 * faults there are not fatal.
-		 */
-		context_clear_pasid_enable(&ce);
-		context_set_copied(&ce);
-
+		set_context_copied(iommu, bus, devfn);
 		new_ce[idx] = ce;
 	}
 
···
 	bool new_ext, ext;
 
 	rtaddr_reg = dmar_readq(iommu->reg + DMAR_RTADDR_REG);
-	ext        = !!(rtaddr_reg & DMA_RTADDR_RTT);
-	new_ext    = !!ecap_ecs(iommu->ecap);
+	ext        = !!(rtaddr_reg & DMA_RTADDR_SMT);
+	new_ext    = !!sm_supported(iommu);
 
 	/*
 	 * The RTT bit can only be changed when translation is disabled,
···
 	 */
 	if (new_ext != ext)
 		return -EINVAL;
+
+	iommu->copied_tables = bitmap_zalloc(BIT_ULL(16), GFP_KERNEL);
+	if (!iommu->copied_tables)
+		return -ENOMEM;
 
 	old_rt_phys = rtaddr_reg & VTD_PAGE_MASK;
 	if (!old_rt_phys)
···
 
 #ifdef CONFIG_INTEL_IOMMU_SVM
 		if (pasid_supported(iommu) && ecap_prs(iommu->ecap)) {
-			/*
-			 * Call dmar_alloc_hwirq() with dmar_global_lock held,
-			 * could cause possible lock race condition.
-			 */
-			up_write(&dmar_global_lock);
 			ret = intel_svm_enable_prq(iommu);
-			down_write(&dmar_global_lock);
 			if (ret)
 				goto free_iommu;
 		}
···
 	force_on = (!intel_iommu_tboot_noforce && tboot_force_iommu()) ||
 		    platform_optin_force_iommu();
 
-	down_write(&dmar_global_lock);
 	if (dmar_table_init()) {
 		if (force_on)
 			panic("tboot: Failed to initialize DMAR table\n");
···
 			panic("tboot: Failed to initialize DMAR device scope\n");
 		goto out_free_dmar;
 	}
-
-	up_write(&dmar_global_lock);
-
-	/*
-	 * The bus notifier takes the dmar_global_lock, so lockdep will
-	 * complain later when we register it under the lock.
-	 */
-	dmar_register_bus_notifier();
-
-	down_write(&dmar_global_lock);
 
 	if (!no_iommu)
 		intel_iommu_debugfs_init();
···
 		pr_err("Initialization failed\n");
 		goto out_free_dmar;
 	}
-	up_write(&dmar_global_lock);
 
 	init_iommu_pm_ops();
 
-	down_read(&dmar_global_lock);
 	for_each_active_iommu(iommu, drhd) {
 		/*
 		 * The flush queue implementation does not perform
···
 				       "%s", iommu->name);
 		iommu_device_register(&iommu->iommu, &intel_iommu_ops, NULL);
 	}
-	up_read(&dmar_global_lock);
 
 	bus_set_iommu(&pci_bus_type, &intel_iommu_ops);
 	if (si_domain && !hw_pass_through)
 		register_memory_notifier(&intel_iommu_memory_nb);
 
-	down_read(&dmar_global_lock);
 	if (probe_acpi_namespace_devices())
 		pr_warn("ACPI name space devices didn't probe correctly\n");
 
···
 
 		iommu_disable_protect_mem_regions(iommu);
 	}
-	up_read(&dmar_global_lock);
-
-	pr_info("Intel(R) Virtualization Technology for Directed I/O\n");
 
 	intel_iommu_enabled = 1;
+	dmar_register_bus_notifier();
+	pr_info("Intel(R) Virtualization Technology for Directed I/O\n");
 
 	return 0;
 
 out_free_dmar:
 	intel_iommu_free_dmars();
-	up_write(&dmar_global_lock);
 	return ret;
 }
 
···
 	struct device_domain_info *info = dev_iommu_priv_get(dev);
 	struct dmar_domain *domain = info->domain;
 	struct intel_iommu *iommu = info->iommu;
+	unsigned long flags;
 
 	if (!dev_is_real_dma_subdevice(info->dev)) {
 		if (dev_is_pci(info->dev) && sm_supported(iommu))
···
 		intel_pasid_free_table(info->dev);
 	}
 
-	spin_lock(&domain->lock);
+	spin_lock_irqsave(&domain->lock, flags);
 	list_del(&info->link);
-	spin_unlock(&domain->lock);
+	spin_unlock_irqrestore(&domain->lock, flags);
 
 	domain_detach_iommu(domain, iommu);
 	info->domain = NULL;
···
 static bool intel_iommu_enforce_cache_coherency(struct iommu_domain *domain)
 {
 	struct dmar_domain *dmar_domain = to_dmar_domain(domain);
+	unsigned long flags;
 
 	if (dmar_domain->force_snooping)
 		return true;
 
-	spin_lock(&dmar_domain->lock);
+	spin_lock_irqsave(&dmar_domain->lock, flags);
 	if (!domain_support_force_snooping(dmar_domain)) {
-		spin_unlock(&dmar_domain->lock);
+		spin_unlock_irqrestore(&dmar_domain->lock, flags);
 		return false;
 	}
 
 	domain_set_force_snooping(dmar_domain);
 	dmar_domain->force_snooping = true;
-	spin_unlock(&dmar_domain->lock);
+	spin_unlock_irqrestore(&dmar_domain->lock, flags);
 
 	return true;
 }

drivers/iommu/intel/iommu.h

···
 #define ecap_dis(e)		(((e) >> 27) & 0x1)
 #define ecap_nest(e)		(((e) >> 26) & 0x1)
 #define ecap_mts(e)		(((e) >> 25) & 0x1)
-#define ecap_ecs(e)		(((e) >> 24) & 0x1)
 #define ecap_iotlb_offset(e) 	((((e) >> 8) & 0x3ff) * 16)
 #define ecap_max_iotlb_offset(e) (ecap_iotlb_offset(e) + 16)
 #define ecap_coherent(e)	((e) & 0x1)
···
 #define DMA_GSTS_CFIS (((u32)1) << 23)
 
 /* DMA_RTADDR_REG */
-#define DMA_RTADDR_RTT (((u64)1) << 11)
 #define DMA_RTADDR_SMT (((u64)1) << 10)
 
 /* CCMD_REG */
···
 
 #ifdef CONFIG_INTEL_IOMMU
 	unsigned long 	*domain_ids; /* bitmap of domains */
+	unsigned long	*copied_tables; /* bitmap of copied tables */
 	spinlock_t	lock; /* protect context, domain ids */
 	struct root_entry *root_entry; /* virtual address */
 
···
 		(struct dma_pte *)ALIGN((unsigned long)pte, VTD_PAGE_SIZE) - pte;
 }
 
+static inline bool context_present(struct context_entry *context)
+{
+	return (context->lo & 1);
+}
+
 extern struct dmar_drhd_unit * dmar_find_matched_drhd_unit(struct pci_dev *dev);
 
 extern int dmar_enable_qi(struct intel_iommu *iommu);
···
 #endif /* CONFIG_INTEL_IOMMU_DEBUGFS */
 
 extern const struct attribute_group *intel_iommu_groups[];
-bool context_present(struct context_entry *context);
 struct context_entry *iommu_context_addr(struct intel_iommu *iommu, u8 bus,
 					 u8 devfn, int alloc);
 

drivers/iommu/iommu.c

···
 	return ret;
 }
 
+static bool iommu_is_default_domain(struct iommu_group *group)
+{
+	if (group->domain == group->default_domain)
+		return true;
+
+	/*
+	 * If the default domain was set to identity and it is still an identity
+	 * domain then we consider this a pass. This happens because of
+	 * amd_iommu_init_device() replacing the default idenytity domain with an
+	 * identity domain that has a different configuration for AMDGPU.
+	 */
+	if (group->default_domain &&
+	    group->default_domain->type == IOMMU_DOMAIN_IDENTITY &&
+	    group->domain && group->domain->type == IOMMU_DOMAIN_IDENTITY)
+		return true;
+	return false;
+}
+
 /**
  * iommu_device_use_default_domain() - Device driver wants to handle device
  *                                     DMA through the kernel DMA API.
···
 
 	mutex_lock(&group->mutex);
 	if (group->owner_cnt) {
-		if (group->domain != group->default_domain ||
-		    group->owner) {
+		if (group->owner || !iommu_is_default_domain(group)) {
 			ret = -EBUSY;
 			goto unlock_out;
 		}

drivers/iommu/virtio-iommu.c

···
 	return iommu_fwspec_add_ids(dev, args->args, 1);
 }
 
+static bool viommu_capable(enum iommu_cap cap)
+{
+	switch (cap) {
+	case IOMMU_CAP_CACHE_COHERENCY:
+		return true;
+	default:
+		return false;
+	}
+}
+
 static struct iommu_ops viommu_ops = {
+	.capable		= viommu_capable,
 	.domain_alloc		= viommu_domain_alloc,
 	.probe_device		= viommu_probe_device,
 	.probe_finalize		= viommu_probe_finalize,

include/linux/dmar.h

···
 
 extern struct rw_semaphore dmar_global_lock;
 extern struct list_head dmar_drhd_units;
+extern int intel_iommu_enabled;
 
 #define for_each_drhd_unit(drhd)					\
 	list_for_each_entry_rcu(drhd, &dmar_drhd_units, list,		\
···
 static inline bool dmar_rcu_check(void)
 {
 	return rwsem_is_locked(&dmar_global_lock) ||
-	       system_state == SYSTEM_BOOTING;
+	       system_state == SYSTEM_BOOTING ||
+	       (IS_ENABLED(CONFIG_INTEL_IOMMU) && !intel_iommu_enabled);
 }
 
 #define	dmar_rcu_dereference(p)	rcu_dereference_check((p), dmar_rcu_check())