KVM: PPC: VFIO: Add in-kernel acceleration for VFIO

This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT
and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO
without passing them to user space which saves time on switching
to user space and back.

This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM.
KVM tries to handle a TCE request in the real mode, if failed
it passes the request to the virtual mode to complete the operation.
If it a virtual mode handler fails, the request is passed to
the user space; this is not expected to happen though.

To avoid dealing with page use counters (which is tricky in real mode),
this only accelerates SPAPR TCE IOMMU v2 clients which are required
to pre-register the userspace memory. The very first TCE request will
be handled in the VFIO SPAPR TCE driver anyway as the userspace view
of the TCE table (iommu_table::it_userspace) is not allocated till
the very first mapping happens and we cannot call vmalloc in real mode.

If we fail to update a hardware IOMMU table unexpected reason, we just
clear it and move on as there is nothing really we can do about it -
for example, if we hot plug a VFIO device to a guest, existing TCE tables
will be mirrored automatically to the hardware and there is no interface
to report to the guest about possible failures.

This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to
the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd
and associates a physical IOMMU table with the SPAPR TCE table (which
is a guest view of the hardware IOMMU table). The iommu_table object
is cached and referenced so we do not have to look up for it in real mode.

This does not implement the UNSET counterpart as there is no use for it -
once the acceleration is enabled, the existing userspace won't
disable it unless a VFIO container is destroyed; this adds necessary
cleanup to the KVM_DEV_VFIO_GROUP_DEL handler.

This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user
space.

This adds real mode version of WARN_ON_ONCE() as the generic version
causes problems with rcu_sched. Since we testing what vmalloc_to_phys()
returns in the code, this also adds a check for already existing
vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect().

This finally makes use of vfio_external_user_iommu_id() which was
introduced quite some time ago and was considered for removal.

Tests show that this patch increases transmission speed from 220MB/s
to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card).
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
Acked-by: Alex Williamson <alex.williamson@redhat.com>
Reviewed-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Paul Mackerras <paulus@ozlabs.org>

Documentation/virtual/kvm/devices/vfio.txt

@@ -16,7 +16,21 @@ Groups:
 
 KVM_DEV_VFIO_GROUP attributes:
   KVM_DEV_VFIO_GROUP_ADD: Add a VFIO group to VFIO-KVM device tracking
+	kvm_device_attr.addr points to an int32_t file descriptor
+	for the VFIO group.
   KVM_DEV_VFIO_GROUP_DEL: Remove a VFIO group from VFIO-KVM device tracking
+	kvm_device_attr.addr points to an int32_t file descriptor
+	for the VFIO group.
+  KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE: attaches a guest visible TCE table
+	allocated by sPAPR KVM.
+	kvm_device_attr.addr points to a struct:
 
-For each, kvm_device_attr.addr points to an int32_t file descriptor
-for the VFIO group.
+	struct kvm_vfio_spapr_tce {
+		__s32	groupfd;
+		__s32	tablefd;
+	};
+
+	where
+	@groupfd is a file descriptor for a VFIO group;
+	@tablefd is a file descriptor for a TCE table allocated via
+		KVM_CREATE_SPAPR_TCE.

arch/powerpc/include/asm/kvm_host.h

@@ -188,6 +188,13 @@ struct kvmppc_pginfo {
 	atomic_t refcnt;
 };
 
+struct kvmppc_spapr_tce_iommu_table {
+	struct rcu_head rcu;
+	struct list_head next;
+	struct iommu_table *tbl;
+	struct kref kref;
+};
+
 struct kvmppc_spapr_tce_table {
 	struct list_head list;
 	struct kvm *kvm;
@@ -196,6 +203,7 @@ struct kvmppc_spapr_tce_table {
 	u32 page_shift;
 	u64 offset;		/* in pages */
 	u64 size;		/* window size in pages */
+	struct list_head iommu_tables;
 	struct page *pages[0];
 };
 

arch/powerpc/include/asm/kvm_ppc.h

@@ -173,6 +173,10 @@ extern long kvmppc_prepare_vrma(struct kvm *kvm,
 extern void kvmppc_map_vrma(struct kvm_vcpu *vcpu,
 			struct kvm_memory_slot *memslot, unsigned long porder);
 extern int kvmppc_pseries_do_hcall(struct kvm_vcpu *vcpu);
+extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
+		struct iommu_group *grp);
+extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
+		struct iommu_group *grp);
 
 extern long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 				struct kvm_create_spapr_tce_64 *args);

arch/powerpc/kvm/book3s_64_vio.c

@@ -28,6 +28,8 @@
 #include <linux/hugetlb.h>
 #include <linux/list.h>
 #include <linux/anon_inodes.h>
+#include <linux/iommu.h>
+#include <linux/file.h>
 
 #include <asm/tlbflush.h>
 #include <asm/kvm_ppc.h>
@@ -40,6 +42,7 @@
 #include <asm/udbg.h>
 #include <asm/iommu.h>
 #include <asm/tce.h>
+#include <asm/mmu_context.h>
 
 static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
 {
@@ -91,6 +94,137 @@ static long kvmppc_account_memlimit(unsigned long stt_pages, bool inc)
 	return ret;
 }
 
+static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
+{
+	struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
+			struct kvmppc_spapr_tce_iommu_table, rcu);
+
+	iommu_tce_table_put(stit->tbl);
+
+	kfree(stit);
+}
+
+static void kvm_spapr_tce_liobn_put(struct kref *kref)
+{
+	struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
+			struct kvmppc_spapr_tce_iommu_table, kref);
+
+	list_del_rcu(&stit->next);
+
+	call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
+}
+
+extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
+		struct iommu_group *grp)
+{
+	int i;
+	struct kvmppc_spapr_tce_table *stt;
+	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
+	struct iommu_table_group *table_group = NULL;
+
+	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
+
+		table_group = iommu_group_get_iommudata(grp);
+		if (WARN_ON(!table_group))
+			continue;
+
+		list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
+			for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
+				if (table_group->tables[i] != stit->tbl)
+					continue;
+
+				kref_put(&stit->kref, kvm_spapr_tce_liobn_put);
+				return;
+			}
+		}
+	}
+}
+
+extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
+		struct iommu_group *grp)
+{
+	struct kvmppc_spapr_tce_table *stt = NULL;
+	bool found = false;
+	struct iommu_table *tbl = NULL;
+	struct iommu_table_group *table_group;
+	long i;
+	struct kvmppc_spapr_tce_iommu_table *stit;
+	struct fd f;
+
+	f = fdget(tablefd);
+	if (!f.file)
+		return -EBADF;
+
+	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
+		if (stt == f.file->private_data) {
+			found = true;
+			break;
+		}
+	}
+
+	fdput(f);
+
+	if (!found)
+		return -EINVAL;
+
+	table_group = iommu_group_get_iommudata(grp);
+	if (WARN_ON(!table_group))
+		return -EFAULT;
+
+	for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
+		struct iommu_table *tbltmp = table_group->tables[i];
+
+		if (!tbltmp)
+			continue;
+		/*
+		 * Make sure hardware table parameters are exactly the same;
+		 * this is used in the TCE handlers where boundary checks
+		 * use only the first attached table.
+		 */
+		if ((tbltmp->it_page_shift == stt->page_shift) &&
+				(tbltmp->it_offset == stt->offset) &&
+				(tbltmp->it_size == stt->size)) {
+			/*
+			 * Reference the table to avoid races with
+			 * add/remove DMA windows.
+			 */
+			tbl = iommu_tce_table_get(tbltmp);
+			break;
+		}
+	}
+	if (!tbl)
+		return -EINVAL;
+
+	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
+		if (tbl != stit->tbl)
+			continue;
+
+		if (!kref_get_unless_zero(&stit->kref)) {
+			/* stit is being destroyed */
+			iommu_tce_table_put(tbl);
+			return -ENOTTY;
+		}
+		/*
+		 * The table is already known to this KVM, we just increased
+		 * its KVM reference counter and can return.
+		 */
+		return 0;
+	}
+
+	stit = kzalloc(sizeof(*stit), GFP_KERNEL);
+	if (!stit) {
+		iommu_tce_table_put(tbl);
+		return -ENOMEM;
+	}
+
+	stit->tbl = tbl;
+	kref_init(&stit->kref);
+
+	list_add_rcu(&stit->next, &stt->iommu_tables);
+
+	return 0;
+}
+
 static void release_spapr_tce_table(struct rcu_head *head)
 {
 	struct kvmppc_spapr_tce_table *stt = container_of(head,
@@ -130,9 +264,18 @@ static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
 static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
 {
 	struct kvmppc_spapr_tce_table *stt = filp->private_data;
+	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
 
 	list_del_rcu(&stt->list);
 
+	list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
+		WARN_ON(!kref_read(&stit->kref));
+		while (1) {
+			if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
+				break;
+		}
+	}
+
 	kvm_put_kvm(stt->kvm);
 
 	kvmppc_account_memlimit(
@@ -183,6 +326,7 @@ long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 	stt->offset = args->offset;
 	stt->size = size;
 	stt->kvm = kvm;
+	INIT_LIST_HEAD_RCU(&stt->iommu_tables);
 
 	for (i = 0; i < npages; i++) {
 		stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
@@ -211,11 +355,101 @@ long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 	return ret;
 }
 
+static void kvmppc_clear_tce(struct iommu_table *tbl, unsigned long entry)
+{
+	unsigned long hpa = 0;
+	enum dma_data_direction dir = DMA_NONE;
+
+	iommu_tce_xchg(tbl, entry, &hpa, &dir);
+}
+
+static long kvmppc_tce_iommu_mapped_dec(struct kvm *kvm,
+		struct iommu_table *tbl, unsigned long entry)
+{
+	struct mm_iommu_table_group_mem_t *mem = NULL;
+	const unsigned long pgsize = 1ULL << tbl->it_page_shift;
+	unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
+
+	if (!pua)
+		/* it_userspace allocation might be delayed */
+		return H_TOO_HARD;
+
+	mem = mm_iommu_lookup(kvm->mm, *pua, pgsize);
+	if (!mem)
+		return H_TOO_HARD;
+
+	mm_iommu_mapped_dec(mem);
+
+	*pua = 0;
+
+	return H_SUCCESS;
+}
+
+static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
+		struct iommu_table *tbl, unsigned long entry)
+{
+	enum dma_data_direction dir = DMA_NONE;
+	unsigned long hpa = 0;
+	long ret;
+
+	if (WARN_ON_ONCE(iommu_tce_xchg(tbl, entry, &hpa, &dir)))
+		return H_HARDWARE;
+
+	if (dir == DMA_NONE)
+		return H_SUCCESS;
+
+	ret = kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
+	if (ret != H_SUCCESS)
+		iommu_tce_xchg(tbl, entry, &hpa, &dir);
+
+	return ret;
+}
+
+long kvmppc_tce_iommu_map(struct kvm *kvm, struct iommu_table *tbl,
+		unsigned long entry, unsigned long ua,
+		enum dma_data_direction dir)
+{
+	long ret;
+	unsigned long hpa, *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
+	struct mm_iommu_table_group_mem_t *mem;
+
+	if (!pua)
+		/* it_userspace allocation might be delayed */
+		return H_TOO_HARD;
+
+	mem = mm_iommu_lookup(kvm->mm, ua, 1ULL << tbl->it_page_shift);
+	if (!mem)
+		/* This only handles v2 IOMMU type, v1 is handled via ioctl() */
+		return H_TOO_HARD;
+
+	if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, &hpa)))
+		return H_HARDWARE;
+
+	if (mm_iommu_mapped_inc(mem))
+		return H_CLOSED;
+
+	ret = iommu_tce_xchg(tbl, entry, &hpa, &dir);
+	if (WARN_ON_ONCE(ret)) {
+		mm_iommu_mapped_dec(mem);
+		return H_HARDWARE;
+	}
+
+	if (dir != DMA_NONE)
+		kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
+
+	*pua = ua;
+
+	return 0;
+}
+
 long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 		      unsigned long ioba, unsigned long tce)
 {
 	struct kvmppc_spapr_tce_table *stt;
-	long ret;
+	long ret, idx;
+	struct kvmppc_spapr_tce_iommu_table *stit;
+	unsigned long entry, ua = 0;
+	enum dma_data_direction dir;
 
 	/* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
 	/* 	    liobn, ioba, tce); */
@@ -232,7 +466,35 @@ long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 	if (ret != H_SUCCESS)
 		return ret;
 
-	kvmppc_tce_put(stt, ioba >> stt->page_shift, tce);
+	dir = iommu_tce_direction(tce);
+	if ((dir != DMA_NONE) && kvmppc_gpa_to_ua(vcpu->kvm,
+			tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), &ua, NULL))
+		return H_PARAMETER;
+
+	entry = ioba >> stt->page_shift;
+
+	list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
+		if (dir == DMA_NONE) {
+			ret = kvmppc_tce_iommu_unmap(vcpu->kvm,
+					stit->tbl, entry);
+		} else {
+			idx = srcu_read_lock(&vcpu->kvm->srcu);
+			ret = kvmppc_tce_iommu_map(vcpu->kvm, stit->tbl,
+					entry, ua, dir);
+			srcu_read_unlock(&vcpu->kvm->srcu, idx);
+		}
+
+		if (ret == H_SUCCESS)
+			continue;
+
+		if (ret == H_TOO_HARD)
+			return ret;
+
+		WARN_ON_ONCE(1);
+		kvmppc_clear_tce(stit->tbl, entry);
+	}
+
+	kvmppc_tce_put(stt, entry, tce);
 
 	return H_SUCCESS;
 }
@@ -247,6 +509,7 @@ long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
 	unsigned long entry, ua = 0;
 	u64 __user *tces;
 	u64 tce;
+	struct kvmppc_spapr_tce_iommu_table *stit;
 
 	stt = kvmppc_find_table(vcpu->kvm, liobn);
 	if (!stt)
@@ -285,6 +548,26 @@ long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
 		if (ret != H_SUCCESS)
 			goto unlock_exit;
 
+		if (kvmppc_gpa_to_ua(vcpu->kvm,
+				tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
+				&ua, NULL))
+			return H_PARAMETER;
+
+		list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
+			ret = kvmppc_tce_iommu_map(vcpu->kvm,
+					stit->tbl, entry + i, ua,
+					iommu_tce_direction(tce));
+
+			if (ret == H_SUCCESS)
+				continue;
+
+			if (ret == H_TOO_HARD)
+				goto unlock_exit;
+
+			WARN_ON_ONCE(1);
+			kvmppc_clear_tce(stit->tbl, entry);
+		}
+
 		kvmppc_tce_put(stt, entry + i, tce);
 	}
 
@@ -301,6 +584,7 @@ long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
 {
 	struct kvmppc_spapr_tce_table *stt;
 	long i, ret;
+	struct kvmppc_spapr_tce_iommu_table *stit;
 
 	stt = kvmppc_find_table(vcpu->kvm, liobn);
 	if (!stt)
@@ -314,6 +598,24 @@ long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
 	if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
 		return H_PARAMETER;
 
+	list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
+		unsigned long entry = ioba >> stit->tbl->it_page_shift;
+
+		for (i = 0; i < npages; ++i) {
+			ret = kvmppc_tce_iommu_unmap(vcpu->kvm,
+					stit->tbl, entry + i);
+
+			if (ret == H_SUCCESS)
+				continue;
+
+			if (ret == H_TOO_HARD)
+				return ret;
+
+			WARN_ON_ONCE(1);
+			kvmppc_clear_tce(stit->tbl, entry);
+		}
+	}
+
 	for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
 		kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
 

arch/powerpc/kvm/book3s_64_vio_hv.c

@@ -40,6 +40,31 @@
 #include <asm/iommu.h>
 #include <asm/tce.h>
 
+#ifdef CONFIG_BUG
+
+#define WARN_ON_ONCE_RM(condition)	({			\
+	static bool __section(.data.unlikely) __warned;		\
+	int __ret_warn_once = !!(condition);			\
+								\
+	if (unlikely(__ret_warn_once && !__warned)) {		\
+		__warned = true;				\
+		pr_err("WARN_ON_ONCE_RM: (%s) at %s:%u\n",	\
+				__stringify(condition),		\
+				__func__, __LINE__);		\
+		dump_stack();					\
+	}							\
+	unlikely(__ret_warn_once);				\
+})
+
+#else
+
+#define WARN_ON_ONCE_RM(condition) ({				\
+	int __ret_warn_on = !!(condition);			\
+	unlikely(__ret_warn_on);				\
+})
+
+#endif
+
 #define TCES_PER_PAGE	(PAGE_SIZE / sizeof(u64))
 
 /*
@@ -161,11 +186,117 @@ long kvmppc_gpa_to_ua(struct kvm *kvm, unsigned long gpa,
 EXPORT_SYMBOL_GPL(kvmppc_gpa_to_ua);
 
 #ifdef CONFIG_KVM_BOOK3S_HV_POSSIBLE
+static void kvmppc_rm_clear_tce(struct iommu_table *tbl, unsigned long entry)
+{
+	unsigned long hpa = 0;
+	enum dma_data_direction dir = DMA_NONE;
+
+	iommu_tce_xchg_rm(tbl, entry, &hpa, &dir);
+}
+
+static long kvmppc_rm_tce_iommu_mapped_dec(struct kvm *kvm,
+		struct iommu_table *tbl, unsigned long entry)
+{
+	struct mm_iommu_table_group_mem_t *mem = NULL;
+	const unsigned long pgsize = 1ULL << tbl->it_page_shift;
+	unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
+
+	if (!pua)
+		/* it_userspace allocation might be delayed */
+		return H_TOO_HARD;
+
+	pua = (void *) vmalloc_to_phys(pua);
+	if (WARN_ON_ONCE_RM(!pua))
+		return H_HARDWARE;
+
+	mem = mm_iommu_lookup_rm(kvm->mm, *pua, pgsize);
+	if (!mem)
+		return H_TOO_HARD;
+
+	mm_iommu_mapped_dec(mem);
+
+	*pua = 0;
+
+	return H_SUCCESS;
+}
+
+static long kvmppc_rm_tce_iommu_unmap(struct kvm *kvm,
+		struct iommu_table *tbl, unsigned long entry)
+{
+	enum dma_data_direction dir = DMA_NONE;
+	unsigned long hpa = 0;
+	long ret;
+
+	if (iommu_tce_xchg_rm(tbl, entry, &hpa, &dir))
+		/*
+		 * real mode xchg can fail if struct page crosses
+		 * a page boundary
+		 */
+		return H_TOO_HARD;
+
+	if (dir == DMA_NONE)
+		return H_SUCCESS;
+
+	ret = kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
+	if (ret)
+		iommu_tce_xchg_rm(tbl, entry, &hpa, &dir);
+
+	return ret;
+}
+
+static long kvmppc_rm_tce_iommu_map(struct kvm *kvm, struct iommu_table *tbl,
+		unsigned long entry, unsigned long ua,
+		enum dma_data_direction dir)
+{
+	long ret;
+	unsigned long hpa = 0;
+	unsigned long *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
+	struct mm_iommu_table_group_mem_t *mem;
+
+	if (!pua)
+		/* it_userspace allocation might be delayed */
+		return H_TOO_HARD;
+
+	mem = mm_iommu_lookup_rm(kvm->mm, ua, 1ULL << tbl->it_page_shift);
+	if (!mem)
+		return H_TOO_HARD;
+
+	if (WARN_ON_ONCE_RM(mm_iommu_ua_to_hpa_rm(mem, ua, &hpa)))
+		return H_HARDWARE;
+
+	pua = (void *) vmalloc_to_phys(pua);
+	if (WARN_ON_ONCE_RM(!pua))
+		return H_HARDWARE;
+
+	if (WARN_ON_ONCE_RM(mm_iommu_mapped_inc(mem)))
+		return H_CLOSED;
+
+	ret = iommu_tce_xchg_rm(tbl, entry, &hpa, &dir);
+	if (ret) {
+		mm_iommu_mapped_dec(mem);
+		/*
+		 * real mode xchg can fail if struct page crosses
+		 * a page boundary
+		 */
+		return H_TOO_HARD;
+	}
+
+	if (dir != DMA_NONE)
+		kvmppc_rm_tce_iommu_mapped_dec(kvm, tbl, entry);
+
+	*pua = ua;
+
+	return 0;
+}
+
 long kvmppc_rm_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 		unsigned long ioba, unsigned long tce)
 {
 	struct kvmppc_spapr_tce_table *stt;
 	long ret;
+	struct kvmppc_spapr_tce_iommu_table *stit;
+	unsigned long entry, ua = 0;
+	enum dma_data_direction dir;
 
 	/* udbg_printf("H_PUT_TCE(): liobn=0x%lx ioba=0x%lx, tce=0x%lx\n", */
 	/* 	    liobn, ioba, tce); */
@@ -182,7 +313,32 @@ long kvmppc_rm_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
 	if (ret != H_SUCCESS)
 		return ret;
 
-	kvmppc_tce_put(stt, ioba >> stt->page_shift, tce);
+	dir = iommu_tce_direction(tce);
+	if ((dir != DMA_NONE) && kvmppc_gpa_to_ua(vcpu->kvm,
+			tce & ~(TCE_PCI_READ | TCE_PCI_WRITE), &ua, NULL))
+		return H_PARAMETER;
+
+	entry = ioba >> stt->page_shift;
+
+	list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
+		if (dir == DMA_NONE)
+			ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm,
+					stit->tbl, entry);
+		else
+			ret = kvmppc_rm_tce_iommu_map(vcpu->kvm,
+					stit->tbl, entry, ua, dir);
+
+		if (ret == H_SUCCESS)
+			continue;
+
+		if (ret == H_TOO_HARD)
+			return ret;
+
+		WARN_ON_ONCE_RM(1);
+		kvmppc_rm_clear_tce(stit->tbl, entry);
+	}
+
+	kvmppc_tce_put(stt, entry, tce);
 
 	return H_SUCCESS;
 }
@@ -223,6 +379,7 @@ long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu,
 	unsigned long tces, entry, ua = 0;
 	unsigned long *rmap = NULL;
 	bool prereg = false;
+	struct kvmppc_spapr_tce_iommu_table *stit;
 
 	stt = kvmppc_find_table(vcpu->kvm, liobn);
 	if (!stt)
@@ -270,6 +427,8 @@ long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu,
 			return H_TOO_HARD;
 
 		rmap = (void *) vmalloc_to_phys(rmap);
+		if (WARN_ON_ONCE_RM(!rmap))
+			return H_HARDWARE;
 
 		/*
 		 * Synchronize with the MMU notifier callbacks in
@@ -293,6 +452,27 @@ long kvmppc_rm_h_put_tce_indirect(struct kvm_vcpu *vcpu,
 		if (ret != H_SUCCESS)
 			goto unlock_exit;
 
+		ua = 0;
+		if (kvmppc_gpa_to_ua(vcpu->kvm,
+				tce & ~(TCE_PCI_READ | TCE_PCI_WRITE),
+				&ua, NULL))
+			return H_PARAMETER;
+
+		list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
+			ret = kvmppc_rm_tce_iommu_map(vcpu->kvm,
+					stit->tbl, entry + i, ua,
+					iommu_tce_direction(tce));
+
+			if (ret == H_SUCCESS)
+				continue;
+
+			if (ret == H_TOO_HARD)
+				goto unlock_exit;
+
+			WARN_ON_ONCE_RM(1);
+			kvmppc_rm_clear_tce(stit->tbl, entry);
+		}
+
 		kvmppc_tce_put(stt, entry + i, tce);
 	}
 
@@ -309,6 +489,7 @@ long kvmppc_rm_h_stuff_tce(struct kvm_vcpu *vcpu,
 {
 	struct kvmppc_spapr_tce_table *stt;
 	long i, ret;
+	struct kvmppc_spapr_tce_iommu_table *stit;
 
 	stt = kvmppc_find_table(vcpu->kvm, liobn);
 	if (!stt)
@@ -322,6 +503,24 @@ long kvmppc_rm_h_stuff_tce(struct kvm_vcpu *vcpu,
 	if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
 		return H_PARAMETER;
 
+	list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
+		unsigned long entry = ioba >> stit->tbl->it_page_shift;
+
+		for (i = 0; i < npages; ++i) {
+			ret = kvmppc_rm_tce_iommu_unmap(vcpu->kvm,
+					stit->tbl, entry + i);
+
+			if (ret == H_SUCCESS)
+				continue;
+
+			if (ret == H_TOO_HARD)
+				return ret;
+
+			WARN_ON_ONCE_RM(1);
+			kvmppc_rm_clear_tce(stit->tbl, entry);
+		}
+	}
+
 	for (i = 0; i < npages; ++i, ioba += (1ULL << stt->page_shift))
 		kvmppc_tce_put(stt, ioba >> stt->page_shift, tce_value);
 

arch/powerpc/kvm/powerpc.c

@@ -534,6 +534,8 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
 #ifdef CONFIG_PPC_BOOK3S_64
 	case KVM_CAP_SPAPR_TCE:
 	case KVM_CAP_SPAPR_TCE_64:
+		/* fallthrough */
+	case KVM_CAP_SPAPR_TCE_VFIO:
 	case KVM_CAP_PPC_RTAS:
 	case KVM_CAP_PPC_FIXUP_HCALL:
 	case KVM_CAP_PPC_ENABLE_HCALL:

include/uapi/linux/kvm.h

@@ -1097,6 +1097,7 @@ struct kvm_device_attr {
 #define  KVM_DEV_VFIO_GROUP			1
 #define   KVM_DEV_VFIO_GROUP_ADD			1
 #define   KVM_DEV_VFIO_GROUP_DEL			2
+#define   KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE		3
 
 enum kvm_device_type {
 	KVM_DEV_TYPE_FSL_MPIC_20	= 1,
@@ -1118,6 +1119,11 @@ enum kvm_device_type {
 	KVM_DEV_TYPE_MAX,
 };
 
+struct kvm_vfio_spapr_tce {
+	__s32	groupfd;
+	__s32	tablefd;
+};
+
 /*
  * ioctls for VM fds
  */

virt/kvm/vfio.c

@@ -20,6 +20,10 @@
 #include <linux/vfio.h>
 #include "vfio.h"
 
+#ifdef CONFIG_SPAPR_TCE_IOMMU
+#include <asm/kvm_ppc.h>
+#endif
+
 struct kvm_vfio_group {
 	struct list_head node;
 	struct vfio_group *vfio_group;
@@ -89,6 +93,47 @@ static bool kvm_vfio_group_is_coherent(struct vfio_group *vfio_group)
 	return ret > 0;
 }
 
+#ifdef CONFIG_SPAPR_TCE_IOMMU
+static int kvm_vfio_external_user_iommu_id(struct vfio_group *vfio_group)
+{
+	int (*fn)(struct vfio_group *);
+	int ret = -EINVAL;
+
+	fn = symbol_get(vfio_external_user_iommu_id);
+	if (!fn)
+		return ret;
+
+	ret = fn(vfio_group);
+
+	symbol_put(vfio_external_user_iommu_id);
+
+	return ret;
+}
+
+static struct iommu_group *kvm_vfio_group_get_iommu_group(
+		struct vfio_group *group)
+{
+	int group_id = kvm_vfio_external_user_iommu_id(group);
+
+	if (group_id < 0)
+		return NULL;
+
+	return iommu_group_get_by_id(group_id);
+}
+
+static void kvm_spapr_tce_release_vfio_group(struct kvm *kvm,
+		struct vfio_group *vfio_group)
+{
+	struct iommu_group *grp = kvm_vfio_group_get_iommu_group(vfio_group);
+
+	if (WARN_ON_ONCE(!grp))
+		return;
+
+	kvm_spapr_tce_release_iommu_group(kvm, grp);
+	iommu_group_put(grp);
+}
+#endif
+
 /*
  * Groups can use the same or different IOMMU domains.  If the same then
  * adding a new group may change the coherency of groups we've previously
@@ -211,6 +256,9 @@ static int kvm_vfio_set_group(struct kvm_device *dev, long attr, u64 arg)
 
 		mutex_unlock(&kv->lock);
 
+#ifdef CONFIG_SPAPR_TCE_IOMMU
+		kvm_spapr_tce_release_vfio_group(dev->kvm, vfio_group);
+#endif
 		kvm_vfio_group_set_kvm(vfio_group, NULL);
 
 		kvm_vfio_group_put_external_user(vfio_group);
@@ -218,6 +266,57 @@ static int kvm_vfio_set_group(struct kvm_device *dev, long attr, u64 arg)
 		kvm_vfio_update_coherency(dev);
 
 		return ret;
+
+#ifdef CONFIG_SPAPR_TCE_IOMMU
+	case KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE: {
+		struct kvm_vfio_spapr_tce param;
+		struct kvm_vfio *kv = dev->private;
+		struct vfio_group *vfio_group;
+		struct kvm_vfio_group *kvg;
+		struct fd f;
+		struct iommu_group *grp;
+
+		if (copy_from_user(&param, (void __user *)arg,
+				sizeof(struct kvm_vfio_spapr_tce)))
+			return -EFAULT;
+
+		f = fdget(param.groupfd);
+		if (!f.file)
+			return -EBADF;
+
+		vfio_group = kvm_vfio_group_get_external_user(f.file);
+		fdput(f);
+
+		if (IS_ERR(vfio_group))
+			return PTR_ERR(vfio_group);
+
+		grp = kvm_vfio_group_get_iommu_group(vfio_group);
+		if (WARN_ON_ONCE(!grp)) {
+			kvm_vfio_group_put_external_user(vfio_group);
+			return -EIO;
+		}
+
+		ret = -ENOENT;
+
+		mutex_lock(&kv->lock);
+
+		list_for_each_entry(kvg, &kv->group_list, node) {
+			if (kvg->vfio_group != vfio_group)
+				continue;
+
+			ret = kvm_spapr_tce_attach_iommu_group(dev->kvm,
+					param.tablefd, grp);
+			break;
+		}
+
+		mutex_unlock(&kv->lock);
+
+		iommu_group_put(grp);
+		kvm_vfio_group_put_external_user(vfio_group);
+
+		return ret;
+	}
+#endif /* CONFIG_SPAPR_TCE_IOMMU */
 	}
 
 	return -ENXIO;
@@ -242,6 +341,9 @@ static int kvm_vfio_has_attr(struct kvm_device *dev,
 		switch (attr->attr) {
 		case KVM_DEV_VFIO_GROUP_ADD:
 		case KVM_DEV_VFIO_GROUP_DEL:
+#ifdef CONFIG_SPAPR_TCE_IOMMU
+		case KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE:
+#endif
 			return 0;
 		}
 
@@ -257,6 +359,9 @@ static void kvm_vfio_destroy(struct kvm_device *dev)
 	struct kvm_vfio_group *kvg, *tmp;
 
 	list_for_each_entry_safe(kvg, tmp, &kv->group_list, node) {
+#ifdef CONFIG_SPAPR_TCE_IOMMU
+		kvm_spapr_tce_release_vfio_group(dev->kvm, kvg->vfio_group);
+#endif
 		kvm_vfio_group_set_kvm(kvg->vfio_group, NULL);
 		kvm_vfio_group_put_external_user(kvg->vfio_group);
 		list_del(&kvg->node);