[kernel,v8,10/10] KVM: PPC: VFIO: Add in-kernel acceleration for VFIO

Message ID 20170310035337.22091-11-aik@ozlabs.ru
State Superseded
Headers show

Commit Message

Alexey Kardashevskiy March 10, 2017, 3:53 a.m.
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.

As this creates a descriptor per IOMMU table-LIOBN couple (called
kvmppc_spapr_tce_iommu_table), it is possible to have several
descriptors with the same iommu_table (hardware IOMMU table) attached
to the same LIOBN; we do not remove duplicates though as
iommu_table_ops::exchange not just update a TCE entry (which is
shared among IOMMU groups) but also invalidates the TCE cache
(one per IOMMU group).

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>
---
Changes:
v8:
* changed all (!pua) checks to return H_TOO_HARD as ioctl() is supposed
to handle them
* changed vmalloc_to_phys() callers to return H_HARDWARE
* changed real mode iommu_tce_xchg_rm() callers to return H_TOO_HARD
and added a comment about this in the code
* changed virtual mode iommu_tce_xchg() callers to return H_HARDWARE
and do WARN_ON
* added WARN_ON_ONCE_RM(!rmap) in kvmppc_rm_h_put_tce_indirect() to
have all vmalloc_to_phys() callsites covered

v7:
* added realmode-friendly WARN_ON_ONCE_RM

v6:
* changed handling of errors returned by kvmppc_(rm_)tce_iommu_(un)map()
* moved kvmppc_gpa_to_ua() to TCE validation

v5:
* changed error codes in multiple places
* added bunch of WARN_ON() in places which should not really happen
* adde a check that an iommu table is not attached already to LIOBN
* dropped explicit calls to iommu_tce_clear_param_check/
iommu_tce_put_param_check as kvmppc_tce_validate/kvmppc_ioba_validate
call them anyway (since the previous patch)
* if we fail to update a hardware IOMMU table for unexpected reason,
this just clears the entry

v4:
* added note to the commit log about allowing multiple updates of
the same IOMMU table;
* instead of checking for if any memory was preregistered, this
returns H_TOO_HARD if a specific page was not;
* fixed comments from v3 about error handling in many places;
* simplified TCE handlers and merged IOMMU parts inline - for example,
there used to be kvmppc_h_put_tce_iommu(), now it is merged into
kvmppc_h_put_tce(); this allows to check IOBA boundaries against
the first attached table only (makes the code simpler);

v3:
* simplified not to use VFIO group notifiers
* reworked cleanup, should be cleaner/simpler now

v2:
* reworked to use new VFIO notifiers
* now same iommu_table may appear in the list several times, to be fixed later
---
 Documentation/virtual/kvm/devices/vfio.txt |  22 +-
 arch/powerpc/include/asm/kvm_host.h        |   8 +
 arch/powerpc/include/asm/kvm_ppc.h         |   4 +
 include/uapi/linux/kvm.h                   |   8 +
 arch/powerpc/kvm/book3s_64_vio.c           | 323 ++++++++++++++++++++++++++++-
 arch/powerpc/kvm/book3s_64_vio_hv.c        | 201 +++++++++++++++++-
 arch/powerpc/kvm/powerpc.c                 |   2 +
 virt/kvm/vfio.c                            |  60 ++++++
 8 files changed, 623 insertions(+), 5 deletions(-)

Comments

David Gibson March 10, 2017, 4:47 a.m. | #1
On Fri, Mar 10, 2017 at 02:53:37PM +1100, Alexey Kardashevskiy wrote:
> 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.
> 
> As this creates a descriptor per IOMMU table-LIOBN couple (called
> kvmppc_spapr_tce_iommu_table), it is possible to have several
> descriptors with the same iommu_table (hardware IOMMU table) attached
> to the same LIOBN; we do not remove duplicates though as
> iommu_table_ops::exchange not just update a TCE entry (which is
> shared among IOMMU groups) but also invalidates the TCE cache
> (one per IOMMU group).
> 
> 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>

Reviewed-by: David Gibson <david@gibson.dropbear.id.au>

> ---
> Changes:
> v8:
> * changed all (!pua) checks to return H_TOO_HARD as ioctl() is supposed
> to handle them
> * changed vmalloc_to_phys() callers to return H_HARDWARE
> * changed real mode iommu_tce_xchg_rm() callers to return H_TOO_HARD
> and added a comment about this in the code
> * changed virtual mode iommu_tce_xchg() callers to return H_HARDWARE
> and do WARN_ON
> * added WARN_ON_ONCE_RM(!rmap) in kvmppc_rm_h_put_tce_indirect() to
> have all vmalloc_to_phys() callsites covered
> 
> v7:
> * added realmode-friendly WARN_ON_ONCE_RM
> 
> v6:
> * changed handling of errors returned by kvmppc_(rm_)tce_iommu_(un)map()
> * moved kvmppc_gpa_to_ua() to TCE validation
> 
> v5:
> * changed error codes in multiple places
> * added bunch of WARN_ON() in places which should not really happen
> * adde a check that an iommu table is not attached already to LIOBN
> * dropped explicit calls to iommu_tce_clear_param_check/
> iommu_tce_put_param_check as kvmppc_tce_validate/kvmppc_ioba_validate
> call them anyway (since the previous patch)
> * if we fail to update a hardware IOMMU table for unexpected reason,
> this just clears the entry
> 
> v4:
> * added note to the commit log about allowing multiple updates of
> the same IOMMU table;
> * instead of checking for if any memory was preregistered, this
> returns H_TOO_HARD if a specific page was not;
> * fixed comments from v3 about error handling in many places;
> * simplified TCE handlers and merged IOMMU parts inline - for example,
> there used to be kvmppc_h_put_tce_iommu(), now it is merged into
> kvmppc_h_put_tce(); this allows to check IOBA boundaries against
> the first attached table only (makes the code simpler);
> 
> v3:
> * simplified not to use VFIO group notifiers
> * reworked cleanup, should be cleaner/simpler now
> 
> v2:
> * reworked to use new VFIO notifiers
> * now same iommu_table may appear in the list several times, to be fixed later
> ---
>  Documentation/virtual/kvm/devices/vfio.txt |  22 +-
>  arch/powerpc/include/asm/kvm_host.h        |   8 +
>  arch/powerpc/include/asm/kvm_ppc.h         |   4 +
>  include/uapi/linux/kvm.h                   |   8 +
>  arch/powerpc/kvm/book3s_64_vio.c           | 323 ++++++++++++++++++++++++++++-
>  arch/powerpc/kvm/book3s_64_vio_hv.c        | 201 +++++++++++++++++-
>  arch/powerpc/kvm/powerpc.c                 |   2 +
>  virt/kvm/vfio.c                            |  60 ++++++
>  8 files changed, 623 insertions(+), 5 deletions(-)
> 
> diff --git a/Documentation/virtual/kvm/devices/vfio.txt b/Documentation/virtual/kvm/devices/vfio.txt
> index ef51740c67ca..f95d867168ea 100644
> --- a/Documentation/virtual/kvm/devices/vfio.txt
> +++ b/Documentation/virtual/kvm/devices/vfio.txt
> @@ -16,7 +16,25 @@ 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 {
> +		__u32	argsz;
> +		__u32	flags;
> +		__s32	groupfd;
> +		__s32	tablefd;
> +	};
> +
> +	where
> +	@argsz is the size of kvm_vfio_spapr_tce_liobn;
> +	@flags are not supported now, must be zero;
> +	@groupfd is a file descriptor for a VFIO group;
> +	@tablefd is a file descriptor for a TCE table allocated via
> +		KVM_CREATE_SPAPR_TCE.
> diff --git a/arch/powerpc/include/asm/kvm_host.h b/arch/powerpc/include/asm/kvm_host.h
> index 7bba8f415627..857ae2c6aa39 100644
> --- a/arch/powerpc/include/asm/kvm_host.h
> +++ b/arch/powerpc/include/asm/kvm_host.h
> @@ -191,6 +191,13 @@ struct kvmppc_pginfo {
>  	atomic_t refcnt;
>  };
>  
> +struct kvmppc_spapr_tce_iommu_table {
> +	struct rcu_head rcu;
> +	struct list_head next;
> +	struct vfio_group *group;
> +	struct iommu_table *tbl;
> +};
> +
>  struct kvmppc_spapr_tce_table {
>  	struct list_head list;
>  	struct kvm *kvm;
> @@ -199,6 +206,7 @@ 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];
>  };
>  
> diff --git a/arch/powerpc/include/asm/kvm_ppc.h b/arch/powerpc/include/asm/kvm_ppc.h
> index 72c2a155641f..66de7e73b3d3 100644
> --- a/arch/powerpc/include/asm/kvm_ppc.h
> +++ b/arch/powerpc/include/asm/kvm_ppc.h
> @@ -164,6 +164,10 @@ 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 vfio_group *group);
> +extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
> +		struct vfio_group *group);
>  
>  extern long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
>  				struct kvm_create_spapr_tce_64 *args);
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index f5a52ffb6b58..e743cb0d176e 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -1088,6 +1088,7 @@ 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,
> @@ -1109,6 +1110,13 @@ enum kvm_device_type {
>  	KVM_DEV_TYPE_MAX,
>  };
>  
> +struct kvm_vfio_spapr_tce {
> +	__u32	argsz;
> +	__u32	flags;
> +	__s32	groupfd;
> +	__s32	tablefd;
> +};
> +
>  /*
>   * ioctls for VM fds
>   */
> diff --git a/arch/powerpc/kvm/book3s_64_vio.c b/arch/powerpc/kvm/book3s_64_vio.c
> index e96a4590464c..be18cda01e1b 100644
> --- a/arch/powerpc/kvm/book3s_64_vio.c
> +++ b/arch/powerpc/kvm/book3s_64_vio.c
> @@ -28,6 +28,10 @@
>  #include <linux/hugetlb.h>
>  #include <linux/list.h>
>  #include <linux/anon_inodes.h>
> +#include <linux/iommu.h>
> +#include <linux/file.h>
> +#include <linux/vfio.h>
> +#include <linux/module.h>
>  
>  #include <asm/tlbflush.h>
>  #include <asm/kvm_ppc.h>
> @@ -40,6 +44,36 @@
>  #include <asm/udbg.h>
>  #include <asm/iommu.h>
>  #include <asm/tce.h>
> +#include <asm/mmu_context.h>
> +
> +static void kvm_vfio_group_put_external_user(struct vfio_group *vfio_group)
> +{
> +	void (*fn)(struct vfio_group *);
> +
> +	fn = symbol_get(vfio_group_put_external_user);
> +	if (WARN_ON(!fn))
> +		return;
> +
> +	fn(vfio_group);
> +
> +	symbol_put(vfio_group_put_external_user);
> +}
> +
> +static int kvm_vfio_external_user_iommu_id(struct vfio_group *vfio_group)
> +{
> +	int (*fn)(struct vfio_group *);
> +	int ret = -1;
> +
> +	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 unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
>  {
> @@ -91,6 +125,130 @@ 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_table_put(stit->tbl);
> +	kvm_vfio_group_put_external_user(stit->group);
> +
> +	kfree(stit);
> +}
> +
> +static void kvm_spapr_tce_liobn_release_iommu_group(
> +		struct kvmppc_spapr_tce_table *stt,
> +		struct vfio_group *group)
> +{
> +	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
> +
> +	list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
> +		if (group && (stit->group != group))
> +			continue;
> +
> +		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 vfio_group *group)
> +{
> +	struct kvmppc_spapr_tce_table *stt;
> +
> +	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list)
> +		kvm_spapr_tce_liobn_release_iommu_group(stt, group);
> +}
> +
> +extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
> +		struct vfio_group *group)
> +{
> +	struct kvmppc_spapr_tce_table *stt = NULL;
> +	bool found = false;
> +	struct iommu_table *tbl = NULL;
> +	struct iommu_table_group *table_group;
> +	long i, ret = 0;
> +	struct kvmppc_spapr_tce_iommu_table *stit;
> +	struct fd f;
> +	int group_id;
> +	struct iommu_group *grp;
> +
> +	group_id = kvm_vfio_external_user_iommu_id(group);
> +	grp = iommu_group_get_by_id(group_id);
> +	if (WARN_ON(!grp))
> +		return -EIO;
> +
> +	f = fdget(tablefd);
> +	if (!f.file) {
> +		ret = -EBADF;
> +		goto put_exit;
> +	}
> +
> +	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) {
> +		ret = -EINVAL;
> +		goto put_exit;
> +	}
> +
> +	table_group = iommu_group_get_iommudata(grp);
> +	if (WARN_ON(!table_group)) {
> +		ret = -EFAULT;
> +		goto put_exit;
> +	}
> +
> +	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)) {
> +			tbl = tbltmp;
> +			break;
> +		}
> +	}
> +	if (!tbl) {
> +		ret = -EINVAL;
> +		goto put_exit;
> +	}
> +
> +	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
> +		if ((stit->tbl == tbl) && (stit->group == group)) {
> +			ret = -EBUSY;
> +			goto put_exit;
> +		}
> +	}
> +
> +	iommu_table_get(tbl);
> +
> +	stit = kzalloc(sizeof(*stit), GFP_KERNEL);
> +	stit->tbl = tbl;
> +	stit->group = group;
> +
> +	list_add_rcu(&stit->next, &stt->iommu_tables);
> +
> +put_exit:
> +	iommu_group_put(grp);
> +
> +	return ret;
> +}
> +
>  static void release_spapr_tce_table(struct rcu_head *head)
>  {
>  	struct kvmppc_spapr_tce_table *stt = container_of(head,
> @@ -133,6 +291,8 @@ static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
>  
>  	list_del_rcu(&stt->list);
>  
> +	kvm_spapr_tce_liobn_release_iommu_group(stt, NULL /* release all */);
> +
>  	kvm_put_kvm(stt->kvm);
>  
>  	kvmppc_account_memlimit(
> @@ -183,6 +343,7 @@ 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 +372,101 @@ 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 +483,35 @@ 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 +526,7 @@ 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 +565,26 @@ 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 +601,7 @@ 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 +615,24 @@ 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);
>  
> diff --git a/arch/powerpc/kvm/book3s_64_vio_hv.c b/arch/powerpc/kvm/book3s_64_vio_hv.c
> index 440d3ab5dc32..eda0a8f6fae8 100644
> --- a/arch/powerpc/kvm/book3s_64_vio_hv.c
> +++ b/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);
>  
> diff --git a/arch/powerpc/kvm/powerpc.c b/arch/powerpc/kvm/powerpc.c
> index 95c91a9de351..62bdd6c48107 100644
> --- a/arch/powerpc/kvm/powerpc.c
> +++ b/arch/powerpc/kvm/powerpc.c
> @@ -538,6 +538,8 @@ 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:
> diff --git a/virt/kvm/vfio.c b/virt/kvm/vfio.c
> index d32f239eb471..2b7dc22265fe 100644
> --- a/virt/kvm/vfio.c
> +++ b/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;
> @@ -211,6 +215,9 @@ 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_iommu_group(dev->kvm, vfio_group);
> +#endif
>  		kvm_vfio_group_set_kvm(vfio_group, NULL);
>  
>  		kvm_vfio_group_put_external_user(vfio_group);
> @@ -218,6 +225,53 @@ 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;
> +		unsigned long minsz;
> +		struct kvm_vfio *kv = dev->private;
> +		struct vfio_group *vfio_group;
> +		struct kvm_vfio_group *kvg;
> +		struct fd f;
> +
> +		minsz = offsetofend(struct kvm_vfio_spapr_tce, tablefd);
> +
> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (param.argsz < minsz || param.flags)
> +			return -EINVAL;
> +
> +		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);
> +
> +		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, vfio_group);
> +
> +			break;
> +		}
> +
> +		mutex_unlock(&kv->lock);
> +
> +		return ret;
> +	}
> +#endif /* CONFIG_SPAPR_TCE_IOMMU */
>  	}
>  
>  	return -ENXIO;
> @@ -242,6 +296,9 @@ 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 +314,9 @@ 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_iommu_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);
Alex Williamson March 14, 2017, 9:05 p.m. | #2
On Fri, 10 Mar 2017 14:53:37 +1100
Alexey Kardashevskiy <aik@ozlabs.ru> wrote:

> 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.
> 
> As this creates a descriptor per IOMMU table-LIOBN couple (called
> kvmppc_spapr_tce_iommu_table), it is possible to have several
> descriptors with the same iommu_table (hardware IOMMU table) attached
> to the same LIOBN; we do not remove duplicates though as
> iommu_table_ops::exchange not just update a TCE entry (which is
> shared among IOMMU groups) but also invalidates the TCE cache
> (one per IOMMU group).
> 
> 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>
> ---
> Changes:
> v8:
> * changed all (!pua) checks to return H_TOO_HARD as ioctl() is supposed
> to handle them
> * changed vmalloc_to_phys() callers to return H_HARDWARE
> * changed real mode iommu_tce_xchg_rm() callers to return H_TOO_HARD
> and added a comment about this in the code
> * changed virtual mode iommu_tce_xchg() callers to return H_HARDWARE
> and do WARN_ON
> * added WARN_ON_ONCE_RM(!rmap) in kvmppc_rm_h_put_tce_indirect() to
> have all vmalloc_to_phys() callsites covered
> 
> v7:
> * added realmode-friendly WARN_ON_ONCE_RM
> 
> v6:
> * changed handling of errors returned by kvmppc_(rm_)tce_iommu_(un)map()
> * moved kvmppc_gpa_to_ua() to TCE validation
> 
> v5:
> * changed error codes in multiple places
> * added bunch of WARN_ON() in places which should not really happen
> * adde a check that an iommu table is not attached already to LIOBN
> * dropped explicit calls to iommu_tce_clear_param_check/
> iommu_tce_put_param_check as kvmppc_tce_validate/kvmppc_ioba_validate
> call them anyway (since the previous patch)
> * if we fail to update a hardware IOMMU table for unexpected reason,
> this just clears the entry
> 
> v4:
> * added note to the commit log about allowing multiple updates of
> the same IOMMU table;
> * instead of checking for if any memory was preregistered, this
> returns H_TOO_HARD if a specific page was not;
> * fixed comments from v3 about error handling in many places;
> * simplified TCE handlers and merged IOMMU parts inline - for example,
> there used to be kvmppc_h_put_tce_iommu(), now it is merged into
> kvmppc_h_put_tce(); this allows to check IOBA boundaries against
> the first attached table only (makes the code simpler);
> 
> v3:
> * simplified not to use VFIO group notifiers
> * reworked cleanup, should be cleaner/simpler now
> 
> v2:
> * reworked to use new VFIO notifiers
> * now same iommu_table may appear in the list several times, to be fixed later
> ---
>  Documentation/virtual/kvm/devices/vfio.txt |  22 +-
>  arch/powerpc/include/asm/kvm_host.h        |   8 +
>  arch/powerpc/include/asm/kvm_ppc.h         |   4 +
>  include/uapi/linux/kvm.h                   |   8 +
>  arch/powerpc/kvm/book3s_64_vio.c           | 323 ++++++++++++++++++++++++++++-
>  arch/powerpc/kvm/book3s_64_vio_hv.c        | 201 +++++++++++++++++-
>  arch/powerpc/kvm/powerpc.c                 |   2 +
>  virt/kvm/vfio.c                            |  60 ++++++
>  8 files changed, 623 insertions(+), 5 deletions(-)
> 
> diff --git a/Documentation/virtual/kvm/devices/vfio.txt b/Documentation/virtual/kvm/devices/vfio.txt
> index ef51740c67ca..f95d867168ea 100644
> --- a/Documentation/virtual/kvm/devices/vfio.txt
> +++ b/Documentation/virtual/kvm/devices/vfio.txt
> @@ -16,7 +16,25 @@ 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 {
> +		__u32	argsz;
> +		__u32	flags;
> +		__s32	groupfd;
> +		__s32	tablefd;
> +	};
> +
> +	where
> +	@argsz is the size of kvm_vfio_spapr_tce_liobn;

kvm_vfio_spapr_tce_liobn does not exist.  s/_liobn//?

> +	@flags are not supported now, must be zero;

We do this argsz/flags thing on vfio ioctls because ioctls are a bit
more of a restricted resource.  We don't want to burn through them so
we make them expandable.  I don't know that we have that restriction
here and the ADD/DEL support certainly doesn't include it.  Maybe this
isn't necessary?

> +	@groupfd is a file descriptor for a VFIO group;
> +	@tablefd is a file descriptor for a TCE table allocated via
> +		KVM_CREATE_SPAPR_TCE.
> diff --git a/arch/powerpc/include/asm/kvm_host.h b/arch/powerpc/include/asm/kvm_host.h
> index 7bba8f415627..857ae2c6aa39 100644
> --- a/arch/powerpc/include/asm/kvm_host.h
> +++ b/arch/powerpc/include/asm/kvm_host.h
> @@ -191,6 +191,13 @@ struct kvmppc_pginfo {
>  	atomic_t refcnt;
>  };
>  
> +struct kvmppc_spapr_tce_iommu_table {
> +	struct rcu_head rcu;
> +	struct list_head next;
> +	struct vfio_group *group;
> +	struct iommu_table *tbl;
> +};
> +
>  struct kvmppc_spapr_tce_table {
>  	struct list_head list;
>  	struct kvm *kvm;
> @@ -199,6 +206,7 @@ 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];
>  };
>  
> diff --git a/arch/powerpc/include/asm/kvm_ppc.h b/arch/powerpc/include/asm/kvm_ppc.h
> index 72c2a155641f..66de7e73b3d3 100644
> --- a/arch/powerpc/include/asm/kvm_ppc.h
> +++ b/arch/powerpc/include/asm/kvm_ppc.h
> @@ -164,6 +164,10 @@ 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 vfio_group *group);
> +extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
> +		struct vfio_group *group);
>  
>  extern long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
>  				struct kvm_create_spapr_tce_64 *args);
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index f5a52ffb6b58..e743cb0d176e 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -1088,6 +1088,7 @@ 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,
> @@ -1109,6 +1110,13 @@ enum kvm_device_type {
>  	KVM_DEV_TYPE_MAX,
>  };
>  
> +struct kvm_vfio_spapr_tce {
> +	__u32	argsz;
> +	__u32	flags;
> +	__s32	groupfd;
> +	__s32	tablefd;
> +};
> +
>  /*
>   * ioctls for VM fds
>   */
> diff --git a/arch/powerpc/kvm/book3s_64_vio.c b/arch/powerpc/kvm/book3s_64_vio.c
> index e96a4590464c..be18cda01e1b 100644
> --- a/arch/powerpc/kvm/book3s_64_vio.c
> +++ b/arch/powerpc/kvm/book3s_64_vio.c
> @@ -28,6 +28,10 @@
>  #include <linux/hugetlb.h>
>  #include <linux/list.h>
>  #include <linux/anon_inodes.h>
> +#include <linux/iommu.h>
> +#include <linux/file.h>
> +#include <linux/vfio.h>
> +#include <linux/module.h>
>  
>  #include <asm/tlbflush.h>
>  #include <asm/kvm_ppc.h>
> @@ -40,6 +44,36 @@
>  #include <asm/udbg.h>
>  #include <asm/iommu.h>
>  #include <asm/tce.h>
> +#include <asm/mmu_context.h>
> +
> +static void kvm_vfio_group_put_external_user(struct vfio_group *vfio_group)
> +{
> +	void (*fn)(struct vfio_group *);
> +
> +	fn = symbol_get(vfio_group_put_external_user);
> +	if (WARN_ON(!fn))
> +		return;
> +
> +	fn(vfio_group);
> +
> +	symbol_put(vfio_group_put_external_user);
> +}
> +
> +static int kvm_vfio_external_user_iommu_id(struct vfio_group *vfio_group)
> +{
> +	int (*fn)(struct vfio_group *);
> +	int ret = -1;
> +
> +	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;
> +}


Ugh.  This feels so wrong.  Why can't you have kvm-vfio pass the
iommu_group?  Why do you need to hold this additional vfio_group
reference?

>  
>  static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
>  {
> @@ -91,6 +125,130 @@ 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_table_put(stit->tbl);
> +	kvm_vfio_group_put_external_user(stit->group);
> +
> +	kfree(stit);
> +}
> +
> +static void kvm_spapr_tce_liobn_release_iommu_group(
> +		struct kvmppc_spapr_tce_table *stt,
> +		struct vfio_group *group)
> +{
> +	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
> +
> +	list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
> +		if (group && (stit->group != group))
> +			continue;
> +
> +		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 vfio_group *group)
> +{
> +	struct kvmppc_spapr_tce_table *stt;
> +
> +	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list)
> +		kvm_spapr_tce_liobn_release_iommu_group(stt, group);
> +}
> +
> +extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
> +		struct vfio_group *group)
> +{
> +	struct kvmppc_spapr_tce_table *stt = NULL;
> +	bool found = false;
> +	struct iommu_table *tbl = NULL;
> +	struct iommu_table_group *table_group;
> +	long i, ret = 0;
> +	struct kvmppc_spapr_tce_iommu_table *stit;
> +	struct fd f;
> +	int group_id;
> +	struct iommu_group *grp;
> +
> +	group_id = kvm_vfio_external_user_iommu_id(group);
> +	grp = iommu_group_get_by_id(group_id);
> +	if (WARN_ON(!grp))
> +		return -EIO;
> +
> +	f = fdget(tablefd);
> +	if (!f.file) {
> +		ret = -EBADF;
> +		goto put_exit;
> +	}
> +
> +	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) {
> +		ret = -EINVAL;
> +		goto put_exit;
> +	}
> +
> +	table_group = iommu_group_get_iommudata(grp);
> +	if (WARN_ON(!table_group)) {
> +		ret = -EFAULT;
> +		goto put_exit;
> +	}
> +
> +	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)) {
> +			tbl = tbltmp;
> +			break;
> +		}
> +	}
> +	if (!tbl) {
> +		ret = -EINVAL;
> +		goto put_exit;
> +	}
> +
> +	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
> +		if ((stit->tbl == tbl) && (stit->group == group)) {
> +			ret = -EBUSY;
> +			goto put_exit;
> +		}
> +	}
> +
> +	iommu_table_get(tbl);
> +
> +	stit = kzalloc(sizeof(*stit), GFP_KERNEL);
> +	stit->tbl = tbl;
> +	stit->group = group;
> +
> +	list_add_rcu(&stit->next, &stt->iommu_tables);
> +
> +put_exit:
> +	iommu_group_put(grp);
> +
> +	return ret;
> +}
> +
>  static void release_spapr_tce_table(struct rcu_head *head)
>  {
>  	struct kvmppc_spapr_tce_table *stt = container_of(head,
> @@ -133,6 +291,8 @@ static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
>  
>  	list_del_rcu(&stt->list);
>  
> +	kvm_spapr_tce_liobn_release_iommu_group(stt, NULL /* release all */);
> +
>  	kvm_put_kvm(stt->kvm);
>  
>  	kvmppc_account_memlimit(
> @@ -183,6 +343,7 @@ 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 +372,101 @@ 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 +483,35 @@ 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 +526,7 @@ 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 +565,26 @@ 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 +601,7 @@ 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 +615,24 @@ 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);
>  
> diff --git a/arch/powerpc/kvm/book3s_64_vio_hv.c b/arch/powerpc/kvm/book3s_64_vio_hv.c
> index 440d3ab5dc32..eda0a8f6fae8 100644
> --- a/arch/powerpc/kvm/book3s_64_vio_hv.c
> +++ b/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);
>  
> diff --git a/arch/powerpc/kvm/powerpc.c b/arch/powerpc/kvm/powerpc.c
> index 95c91a9de351..62bdd6c48107 100644
> --- a/arch/powerpc/kvm/powerpc.c
> +++ b/arch/powerpc/kvm/powerpc.c
> @@ -538,6 +538,8 @@ 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:
> diff --git a/virt/kvm/vfio.c b/virt/kvm/vfio.c
> index d32f239eb471..2b7dc22265fe 100644
> --- a/virt/kvm/vfio.c
> +++ b/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;
> @@ -211,6 +215,9 @@ 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_iommu_group(dev->kvm, vfio_group);
> +#endif
>  		kvm_vfio_group_set_kvm(vfio_group, NULL);
>  
>  		kvm_vfio_group_put_external_user(vfio_group);
> @@ -218,6 +225,53 @@ 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;
> +		unsigned long minsz;
> +		struct kvm_vfio *kv = dev->private;
> +		struct vfio_group *vfio_group;
> +		struct kvm_vfio_group *kvg;
> +		struct fd f;
> +
> +		minsz = offsetofend(struct kvm_vfio_spapr_tce, tablefd);
> +
> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (param.argsz < minsz || param.flags)
> +			return -EINVAL;
> +
> +		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);
> +
> +		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, vfio_group);
> +
> +			break;
> +		}
> +
> +		mutex_unlock(&kv->lock);
> +
> +		return ret;
> +	}
> +#endif /* CONFIG_SPAPR_TCE_IOMMU */


The group reference is leaked if kvm_spapr_tce_attach_iommu_group()
fails.  My preference would be to not hold that separate group
reference in the spapr code anyway, having a parallel life cycle over
there is confusing and results in ugliness like duplicating 
kvm_vfio_group_put_external_user().  Thanks,

Alex

>  	}
>  
>  	return -ENXIO;
> @@ -242,6 +296,9 @@ 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 +314,9 @@ 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_iommu_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);

--
To unsubscribe from this list: send the line "unsubscribe kvm-ppc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
David Gibson March 15, 2017, 4:40 a.m. | #3
On Tue, Mar 14, 2017 at 03:05:27PM -0600, Alex Williamson wrote:
> On Fri, 10 Mar 2017 14:53:37 +1100
> Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
> 
> > 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.
> > 
> > As this creates a descriptor per IOMMU table-LIOBN couple (called
> > kvmppc_spapr_tce_iommu_table), it is possible to have several
> > descriptors with the same iommu_table (hardware IOMMU table) attached
> > to the same LIOBN; we do not remove duplicates though as
> > iommu_table_ops::exchange not just update a TCE entry (which is
> > shared among IOMMU groups) but also invalidates the TCE cache
> > (one per IOMMU group).
> > 
> > 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>
> > ---
> > Changes:
> > v8:
> > * changed all (!pua) checks to return H_TOO_HARD as ioctl() is supposed
> > to handle them
> > * changed vmalloc_to_phys() callers to return H_HARDWARE
> > * changed real mode iommu_tce_xchg_rm() callers to return H_TOO_HARD
> > and added a comment about this in the code
> > * changed virtual mode iommu_tce_xchg() callers to return H_HARDWARE
> > and do WARN_ON
> > * added WARN_ON_ONCE_RM(!rmap) in kvmppc_rm_h_put_tce_indirect() to
> > have all vmalloc_to_phys() callsites covered
> > 
> > v7:
> > * added realmode-friendly WARN_ON_ONCE_RM
> > 
> > v6:
> > * changed handling of errors returned by kvmppc_(rm_)tce_iommu_(un)map()
> > * moved kvmppc_gpa_to_ua() to TCE validation
> > 
> > v5:
> > * changed error codes in multiple places
> > * added bunch of WARN_ON() in places which should not really happen
> > * adde a check that an iommu table is not attached already to LIOBN
> > * dropped explicit calls to iommu_tce_clear_param_check/
> > iommu_tce_put_param_check as kvmppc_tce_validate/kvmppc_ioba_validate
> > call them anyway (since the previous patch)
> > * if we fail to update a hardware IOMMU table for unexpected reason,
> > this just clears the entry
> > 
> > v4:
> > * added note to the commit log about allowing multiple updates of
> > the same IOMMU table;
> > * instead of checking for if any memory was preregistered, this
> > returns H_TOO_HARD if a specific page was not;
> > * fixed comments from v3 about error handling in many places;
> > * simplified TCE handlers and merged IOMMU parts inline - for example,
> > there used to be kvmppc_h_put_tce_iommu(), now it is merged into
> > kvmppc_h_put_tce(); this allows to check IOBA boundaries against
> > the first attached table only (makes the code simpler);
> > 
> > v3:
> > * simplified not to use VFIO group notifiers
> > * reworked cleanup, should be cleaner/simpler now
> > 
> > v2:
> > * reworked to use new VFIO notifiers
> > * now same iommu_table may appear in the list several times, to be fixed later
> > ---
> >  Documentation/virtual/kvm/devices/vfio.txt |  22 +-
> >  arch/powerpc/include/asm/kvm_host.h        |   8 +
> >  arch/powerpc/include/asm/kvm_ppc.h         |   4 +
> >  include/uapi/linux/kvm.h                   |   8 +
> >  arch/powerpc/kvm/book3s_64_vio.c           | 323 ++++++++++++++++++++++++++++-
> >  arch/powerpc/kvm/book3s_64_vio_hv.c        | 201 +++++++++++++++++-
> >  arch/powerpc/kvm/powerpc.c                 |   2 +
> >  virt/kvm/vfio.c                            |  60 ++++++
> >  8 files changed, 623 insertions(+), 5 deletions(-)
> > 
> > diff --git a/Documentation/virtual/kvm/devices/vfio.txt b/Documentation/virtual/kvm/devices/vfio.txt
> > index ef51740c67ca..f95d867168ea 100644
> > --- a/Documentation/virtual/kvm/devices/vfio.txt
> > +++ b/Documentation/virtual/kvm/devices/vfio.txt
> > @@ -16,7 +16,25 @@ 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 {
> > +		__u32	argsz;
> > +		__u32	flags;
> > +		__s32	groupfd;
> > +		__s32	tablefd;
> > +	};
> > +
> > +	where
> > +	@argsz is the size of kvm_vfio_spapr_tce_liobn;
> 
> kvm_vfio_spapr_tce_liobn does not exist.  s/_liobn//?
> 
> > +	@flags are not supported now, must be zero;
> 
> We do this argsz/flags thing on vfio ioctls because ioctls are a bit
> more of a restricted resource.  We don't want to burn through them so
> we make them expandable.  I don't know that we have that restriction
> here and the ADD/DEL support certainly doesn't include it.  Maybe this
> isn't necessary?

I didn't comment on this before, but I tend to agree with Alex here.

> > +	@groupfd is a file descriptor for a VFIO group;
> > +	@tablefd is a file descriptor for a TCE table allocated via
> > +		KVM_CREATE_SPAPR_TCE.
> > diff --git a/arch/powerpc/include/asm/kvm_host.h b/arch/powerpc/include/asm/kvm_host.h
> > index 7bba8f415627..857ae2c6aa39 100644
> > --- a/arch/powerpc/include/asm/kvm_host.h
> > +++ b/arch/powerpc/include/asm/kvm_host.h
> > @@ -191,6 +191,13 @@ struct kvmppc_pginfo {
> >  	atomic_t refcnt;
> >  };
> >  
> > +struct kvmppc_spapr_tce_iommu_table {
> > +	struct rcu_head rcu;
> > +	struct list_head next;
> > +	struct vfio_group *group;
> > +	struct iommu_table *tbl;
> > +};
> > +
> >  struct kvmppc_spapr_tce_table {
> >  	struct list_head list;
> >  	struct kvm *kvm;
> > @@ -199,6 +206,7 @@ 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];
> >  };
> >  
> > diff --git a/arch/powerpc/include/asm/kvm_ppc.h b/arch/powerpc/include/asm/kvm_ppc.h
> > index 72c2a155641f..66de7e73b3d3 100644
> > --- a/arch/powerpc/include/asm/kvm_ppc.h
> > +++ b/arch/powerpc/include/asm/kvm_ppc.h
> > @@ -164,6 +164,10 @@ 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 vfio_group *group);
> > +extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
> > +		struct vfio_group *group);
> >  
> >  extern long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
> >  				struct kvm_create_spapr_tce_64 *args);
> > diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> > index f5a52ffb6b58..e743cb0d176e 100644
> > --- a/include/uapi/linux/kvm.h
> > +++ b/include/uapi/linux/kvm.h
> > @@ -1088,6 +1088,7 @@ 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,
> > @@ -1109,6 +1110,13 @@ enum kvm_device_type {
> >  	KVM_DEV_TYPE_MAX,
> >  };
> >  
> > +struct kvm_vfio_spapr_tce {
> > +	__u32	argsz;
> > +	__u32	flags;
> > +	__s32	groupfd;
> > +	__s32	tablefd;
> > +};
> > +
> >  /*
> >   * ioctls for VM fds
> >   */
> > diff --git a/arch/powerpc/kvm/book3s_64_vio.c b/arch/powerpc/kvm/book3s_64_vio.c
> > index e96a4590464c..be18cda01e1b 100644
> > --- a/arch/powerpc/kvm/book3s_64_vio.c
> > +++ b/arch/powerpc/kvm/book3s_64_vio.c
> > @@ -28,6 +28,10 @@
> >  #include <linux/hugetlb.h>
> >  #include <linux/list.h>
> >  #include <linux/anon_inodes.h>
> > +#include <linux/iommu.h>
> > +#include <linux/file.h>
> > +#include <linux/vfio.h>
> > +#include <linux/module.h>
> >  
> >  #include <asm/tlbflush.h>
> >  #include <asm/kvm_ppc.h>
> > @@ -40,6 +44,36 @@
> >  #include <asm/udbg.h>
> >  #include <asm/iommu.h>
> >  #include <asm/tce.h>
> > +#include <asm/mmu_context.h>
> > +
> > +static void kvm_vfio_group_put_external_user(struct vfio_group *vfio_group)
> > +{
> > +	void (*fn)(struct vfio_group *);
> > +
> > +	fn = symbol_get(vfio_group_put_external_user);
> > +	if (WARN_ON(!fn))
> > +		return;
> > +
> > +	fn(vfio_group);
> > +
> > +	symbol_put(vfio_group_put_external_user);
> > +}
> > +
> > +static int kvm_vfio_external_user_iommu_id(struct vfio_group *vfio_group)
> > +{
> > +	int (*fn)(struct vfio_group *);
> > +	int ret = -1;
> > +
> > +	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;
> > +}
> 
> 
> Ugh.  This feels so wrong.  Why can't you have kvm-vfio pass the
> iommu_group?  Why do you need to hold this additional vfio_group
> reference?

Keeping the vfio_group reference makes sense to me, since we don't
want the vfio context for the group to go away while it's attached to
the LIOBN.

However, going via the iommu_id rather than just having an interface
to directly grab the iommu group from the vfio_group seems bizarre to
me.  I'm ok with cleaning that up later, however.

> 
> >  
> >  static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
> >  {
> > @@ -91,6 +125,130 @@ 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_table_put(stit->tbl);
> > +	kvm_vfio_group_put_external_user(stit->group);
> > +
> > +	kfree(stit);
> > +}
> > +
> > +static void kvm_spapr_tce_liobn_release_iommu_group(
> > +		struct kvmppc_spapr_tce_table *stt,
> > +		struct vfio_group *group)
> > +{
> > +	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
> > +
> > +	list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
> > +		if (group && (stit->group != group))
> > +			continue;
> > +
> > +		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 vfio_group *group)
> > +{
> > +	struct kvmppc_spapr_tce_table *stt;
> > +
> > +	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list)
> > +		kvm_spapr_tce_liobn_release_iommu_group(stt, group);
> > +}
> > +
> > +extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
> > +		struct vfio_group *group)
> > +{
> > +	struct kvmppc_spapr_tce_table *stt = NULL;
> > +	bool found = false;
> > +	struct iommu_table *tbl = NULL;
> > +	struct iommu_table_group *table_group;
> > +	long i, ret = 0;
> > +	struct kvmppc_spapr_tce_iommu_table *stit;
> > +	struct fd f;
> > +	int group_id;
> > +	struct iommu_group *grp;
> > +
> > +	group_id = kvm_vfio_external_user_iommu_id(group);
> > +	grp = iommu_group_get_by_id(group_id);
> > +	if (WARN_ON(!grp))
> > +		return -EIO;
> > +
> > +	f = fdget(tablefd);
> > +	if (!f.file) {
> > +		ret = -EBADF;
> > +		goto put_exit;
> > +	}
> > +
> > +	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) {
> > +		ret = -EINVAL;
> > +		goto put_exit;
> > +	}
> > +
> > +	table_group = iommu_group_get_iommudata(grp);
> > +	if (WARN_ON(!table_group)) {
> > +		ret = -EFAULT;
> > +		goto put_exit;
> > +	}
> > +
> > +	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)) {
> > +			tbl = tbltmp;
> > +			break;
> > +		}
> > +	}
> > +	if (!tbl) {
> > +		ret = -EINVAL;
> > +		goto put_exit;
> > +	}
> > +
> > +	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
> > +		if ((stit->tbl == tbl) && (stit->group == group)) {
> > +			ret = -EBUSY;
> > +			goto put_exit;
> > +		}
> > +	}
> > +
> > +	iommu_table_get(tbl);
> > +
> > +	stit = kzalloc(sizeof(*stit), GFP_KERNEL);
> > +	stit->tbl = tbl;
> > +	stit->group = group;
> > +
> > +	list_add_rcu(&stit->next, &stt->iommu_tables);
> > +
> > +put_exit:
> > +	iommu_group_put(grp);
> > +
> > +	return ret;
> > +}
> > +
> >  static void release_spapr_tce_table(struct rcu_head *head)
> >  {
> >  	struct kvmppc_spapr_tce_table *stt = container_of(head,
> > @@ -133,6 +291,8 @@ static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
> >  
> >  	list_del_rcu(&stt->list);
> >  
> > +	kvm_spapr_tce_liobn_release_iommu_group(stt, NULL /* release all */);
> > +
> >  	kvm_put_kvm(stt->kvm);
> >  
> >  	kvmppc_account_memlimit(
> > @@ -183,6 +343,7 @@ 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 +372,101 @@ 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 +483,35 @@ 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 +526,7 @@ 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 +565,26 @@ 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 +601,7 @@ 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 +615,24 @@ 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);
> >  
> > diff --git a/arch/powerpc/kvm/book3s_64_vio_hv.c b/arch/powerpc/kvm/book3s_64_vio_hv.c
> > index 440d3ab5dc32..eda0a8f6fae8 100644
> > --- a/arch/powerpc/kvm/book3s_64_vio_hv.c
> > +++ b/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);
> >  
> > diff --git a/arch/powerpc/kvm/powerpc.c b/arch/powerpc/kvm/powerpc.c
> > index 95c91a9de351..62bdd6c48107 100644
> > --- a/arch/powerpc/kvm/powerpc.c
> > +++ b/arch/powerpc/kvm/powerpc.c
> > @@ -538,6 +538,8 @@ 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:
> > diff --git a/virt/kvm/vfio.c b/virt/kvm/vfio.c
> > index d32f239eb471..2b7dc22265fe 100644
> > --- a/virt/kvm/vfio.c
> > +++ b/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;
> > @@ -211,6 +215,9 @@ 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_iommu_group(dev->kvm, vfio_group);
> > +#endif
> >  		kvm_vfio_group_set_kvm(vfio_group, NULL);
> >  
> >  		kvm_vfio_group_put_external_user(vfio_group);
> > @@ -218,6 +225,53 @@ 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;
> > +		unsigned long minsz;
> > +		struct kvm_vfio *kv = dev->private;
> > +		struct vfio_group *vfio_group;
> > +		struct kvm_vfio_group *kvg;
> > +		struct fd f;
> > +
> > +		minsz = offsetofend(struct kvm_vfio_spapr_tce, tablefd);
> > +
> > +		if (copy_from_user(&param, (void __user *)arg, minsz))
> > +			return -EFAULT;
> > +
> > +		if (param.argsz < minsz || param.flags)
> > +			return -EINVAL;
> > +
> > +		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);
> > +
> > +		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, vfio_group);
> > +
> > +			break;
> > +		}
> > +
> > +		mutex_unlock(&kv->lock);
> > +
> > +		return ret;
> > +	}
> > +#endif /* CONFIG_SPAPR_TCE_IOMMU */
> 
> 
> The group reference is leaked if kvm_spapr_tce_attach_iommu_group()
> fails.

Good catch.

> My preference would be to not hold that separate group
> reference in the spapr code anyway, having a parallel life cycle over
> there is confusing and results in ugliness like duplicating 
> kvm_vfio_group_put_external_user().  Thanks,
> 
> Alex
> 
> >  	}
> >  
> >  	return -ENXIO;
> > @@ -242,6 +296,9 @@ 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 +314,9 @@ 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_iommu_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);
>
Alexey Kardashevskiy March 15, 2017, 1:21 p.m. | #4
On 15/03/17 08:05, Alex Williamson wrote:
> On Fri, 10 Mar 2017 14:53:37 +1100
> Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
> 
>> 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.
>>
>> As this creates a descriptor per IOMMU table-LIOBN couple (called
>> kvmppc_spapr_tce_iommu_table), it is possible to have several
>> descriptors with the same iommu_table (hardware IOMMU table) attached
>> to the same LIOBN; we do not remove duplicates though as
>> iommu_table_ops::exchange not just update a TCE entry (which is
>> shared among IOMMU groups) but also invalidates the TCE cache
>> (one per IOMMU group).
>>
>> 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>
>> ---
>> Changes:
>> v8:
>> * changed all (!pua) checks to return H_TOO_HARD as ioctl() is supposed
>> to handle them
>> * changed vmalloc_to_phys() callers to return H_HARDWARE
>> * changed real mode iommu_tce_xchg_rm() callers to return H_TOO_HARD
>> and added a comment about this in the code
>> * changed virtual mode iommu_tce_xchg() callers to return H_HARDWARE
>> and do WARN_ON
>> * added WARN_ON_ONCE_RM(!rmap) in kvmppc_rm_h_put_tce_indirect() to
>> have all vmalloc_to_phys() callsites covered
>>
>> v7:
>> * added realmode-friendly WARN_ON_ONCE_RM
>>
>> v6:
>> * changed handling of errors returned by kvmppc_(rm_)tce_iommu_(un)map()
>> * moved kvmppc_gpa_to_ua() to TCE validation
>>
>> v5:
>> * changed error codes in multiple places
>> * added bunch of WARN_ON() in places which should not really happen
>> * adde a check that an iommu table is not attached already to LIOBN
>> * dropped explicit calls to iommu_tce_clear_param_check/
>> iommu_tce_put_param_check as kvmppc_tce_validate/kvmppc_ioba_validate
>> call them anyway (since the previous patch)
>> * if we fail to update a hardware IOMMU table for unexpected reason,
>> this just clears the entry
>>
>> v4:
>> * added note to the commit log about allowing multiple updates of
>> the same IOMMU table;
>> * instead of checking for if any memory was preregistered, this
>> returns H_TOO_HARD if a specific page was not;
>> * fixed comments from v3 about error handling in many places;
>> * simplified TCE handlers and merged IOMMU parts inline - for example,
>> there used to be kvmppc_h_put_tce_iommu(), now it is merged into
>> kvmppc_h_put_tce(); this allows to check IOBA boundaries against
>> the first attached table only (makes the code simpler);
>>
>> v3:
>> * simplified not to use VFIO group notifiers
>> * reworked cleanup, should be cleaner/simpler now
>>
>> v2:
>> * reworked to use new VFIO notifiers
>> * now same iommu_table may appear in the list several times, to be fixed later
>> ---
>>  Documentation/virtual/kvm/devices/vfio.txt |  22 +-
>>  arch/powerpc/include/asm/kvm_host.h        |   8 +
>>  arch/powerpc/include/asm/kvm_ppc.h         |   4 +
>>  include/uapi/linux/kvm.h                   |   8 +
>>  arch/powerpc/kvm/book3s_64_vio.c           | 323 ++++++++++++++++++++++++++++-
>>  arch/powerpc/kvm/book3s_64_vio_hv.c        | 201 +++++++++++++++++-
>>  arch/powerpc/kvm/powerpc.c                 |   2 +
>>  virt/kvm/vfio.c                            |  60 ++++++
>>  8 files changed, 623 insertions(+), 5 deletions(-)
>>
>> diff --git a/Documentation/virtual/kvm/devices/vfio.txt b/Documentation/virtual/kvm/devices/vfio.txt
>> index ef51740c67ca..f95d867168ea 100644
>> --- a/Documentation/virtual/kvm/devices/vfio.txt
>> +++ b/Documentation/virtual/kvm/devices/vfio.txt
>> @@ -16,7 +16,25 @@ 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 {
>> +		__u32	argsz;
>> +		__u32	flags;
>> +		__s32	groupfd;
>> +		__s32	tablefd;
>> +	};
>> +
>> +	where
>> +	@argsz is the size of kvm_vfio_spapr_tce_liobn;
> 
> kvm_vfio_spapr_tce_liobn does not exist.  s/_liobn//?

Correct.


> 
>> +	@flags are not supported now, must be zero;
> 
> We do this argsz/flags thing on vfio ioctls because ioctls are a bit
> more of a restricted resource.  We don't want to burn through them so
> we make them expandable.  I don't know that we have that restriction
> here and the ADD/DEL support certainly doesn't include it.  Maybe this
> isn't necessary?


It is not but since I am going to have padding there, I thought I give it a
name. Totally pointless and "u8 padding[4]" is better?


>> +	@groupfd is a file descriptor for a VFIO group;
>> +	@tablefd is a file descriptor for a TCE table allocated via
>> +		KVM_CREATE_SPAPR_TCE.
>> diff --git a/arch/powerpc/include/asm/kvm_host.h b/arch/powerpc/include/asm/kvm_host.h
>> index 7bba8f415627..857ae2c6aa39 100644
>> --- a/arch/powerpc/include/asm/kvm_host.h
>> +++ b/arch/powerpc/include/asm/kvm_host.h
>> @@ -191,6 +191,13 @@ struct kvmppc_pginfo {
>>  	atomic_t refcnt;
>>  };
>>  
>> +struct kvmppc_spapr_tce_iommu_table {
>> +	struct rcu_head rcu;
>> +	struct list_head next;
>> +	struct vfio_group *group;
>> +	struct iommu_table *tbl;
>> +};
>> +
>>  struct kvmppc_spapr_tce_table {
>>  	struct list_head list;
>>  	struct kvm *kvm;
>> @@ -199,6 +206,7 @@ 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];
>>  };
>>  
>> diff --git a/arch/powerpc/include/asm/kvm_ppc.h b/arch/powerpc/include/asm/kvm_ppc.h
>> index 72c2a155641f..66de7e73b3d3 100644
>> --- a/arch/powerpc/include/asm/kvm_ppc.h
>> +++ b/arch/powerpc/include/asm/kvm_ppc.h
>> @@ -164,6 +164,10 @@ 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 vfio_group *group);
>> +extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
>> +		struct vfio_group *group);
>>  
>>  extern long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
>>  				struct kvm_create_spapr_tce_64 *args);
>> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
>> index f5a52ffb6b58..e743cb0d176e 100644
>> --- a/include/uapi/linux/kvm.h
>> +++ b/include/uapi/linux/kvm.h
>> @@ -1088,6 +1088,7 @@ 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,
>> @@ -1109,6 +1110,13 @@ enum kvm_device_type {
>>  	KVM_DEV_TYPE_MAX,
>>  };
>>  
>> +struct kvm_vfio_spapr_tce {
>> +	__u32	argsz;
>> +	__u32	flags;
>> +	__s32	groupfd;
>> +	__s32	tablefd;
>> +};
>> +
>>  /*
>>   * ioctls for VM fds
>>   */
>> diff --git a/arch/powerpc/kvm/book3s_64_vio.c b/arch/powerpc/kvm/book3s_64_vio.c
>> index e96a4590464c..be18cda01e1b 100644
>> --- a/arch/powerpc/kvm/book3s_64_vio.c
>> +++ b/arch/powerpc/kvm/book3s_64_vio.c
>> @@ -28,6 +28,10 @@
>>  #include <linux/hugetlb.h>
>>  #include <linux/list.h>
>>  #include <linux/anon_inodes.h>
>> +#include <linux/iommu.h>
>> +#include <linux/file.h>
>> +#include <linux/vfio.h>
>> +#include <linux/module.h>
>>  
>>  #include <asm/tlbflush.h>
>>  #include <asm/kvm_ppc.h>
>> @@ -40,6 +44,36 @@
>>  #include <asm/udbg.h>
>>  #include <asm/iommu.h>
>>  #include <asm/tce.h>
>> +#include <asm/mmu_context.h>
>> +
>> +static void kvm_vfio_group_put_external_user(struct vfio_group *vfio_group)
>> +{
>> +	void (*fn)(struct vfio_group *);
>> +
>> +	fn = symbol_get(vfio_group_put_external_user);
>> +	if (WARN_ON(!fn))
>> +		return;
>> +
>> +	fn(vfio_group);
>> +
>> +	symbol_put(vfio_group_put_external_user);
>> +}
>> +
>> +static int kvm_vfio_external_user_iommu_id(struct vfio_group *vfio_group)
>> +{
>> +	int (*fn)(struct vfio_group *);
>> +	int ret = -1;
>> +
>> +	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;
>> +}
> 
> 
> Ugh.  This feels so wrong.  Why can't you have kvm-vfio pass the
> iommu_group?  Why do you need to hold this additional vfio_group
> reference?


This is embarassing but now I am not even sure I really need to hold a
reference to a group in any form, I tend to think this is a leftover from
older versions when iommu_table did not have reference counting.

iommu_table struct have a reference counter now and this patch is only
concerned about iommu_table structs, and they are referenced by KVM.

iommu_group structs are referenced as long as a pnv_ioda_pe (IODA's PE ==
IOMMU group in powenrv) lives and iommu_group structs get attached/detached
to/from iommu_table structs when a DMA window is set/unset to PE - this is
controlled by vfio_iommu_spapr_tce ioctls() + rcu. If an iommu_table does
not have attached iommu_group, it is just a table without any cache
invalidation.

So in v9 I'll remove the group reference, and I wonder if I am missing
anything now?


>>  
>>  static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
>>  {
>> @@ -91,6 +125,130 @@ 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_table_put(stit->tbl);
>> +	kvm_vfio_group_put_external_user(stit->group);
>> +
>> +	kfree(stit);
>> +}
>> +
>> +static void kvm_spapr_tce_liobn_release_iommu_group(
>> +		struct kvmppc_spapr_tce_table *stt,
>> +		struct vfio_group *group)
>> +{
>> +	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
>> +
>> +	list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
>> +		if (group && (stit->group != group))
>> +			continue;
>> +
>> +		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 vfio_group *group)
>> +{
>> +	struct kvmppc_spapr_tce_table *stt;
>> +
>> +	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list)
>> +		kvm_spapr_tce_liobn_release_iommu_group(stt, group);
>> +}
>> +
>> +extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
>> +		struct vfio_group *group)
>> +{
>> +	struct kvmppc_spapr_tce_table *stt = NULL;
>> +	bool found = false;
>> +	struct iommu_table *tbl = NULL;
>> +	struct iommu_table_group *table_group;
>> +	long i, ret = 0;
>> +	struct kvmppc_spapr_tce_iommu_table *stit;
>> +	struct fd f;
>> +	int group_id;
>> +	struct iommu_group *grp;
>> +
>> +	group_id = kvm_vfio_external_user_iommu_id(group);
>> +	grp = iommu_group_get_by_id(group_id);
>> +	if (WARN_ON(!grp))
>> +		return -EIO;
>> +
>> +	f = fdget(tablefd);
>> +	if (!f.file) {
>> +		ret = -EBADF;
>> +		goto put_exit;
>> +	}
>> +
>> +	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) {
>> +		ret = -EINVAL;
>> +		goto put_exit;
>> +	}
>> +
>> +	table_group = iommu_group_get_iommudata(grp);
>> +	if (WARN_ON(!table_group)) {
>> +		ret = -EFAULT;
>> +		goto put_exit;
>> +	}
>> +
>> +	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)) {
>> +			tbl = tbltmp;
>> +			break;
>> +		}
>> +	}
>> +	if (!tbl) {
>> +		ret = -EINVAL;
>> +		goto put_exit;
>> +	}
>> +
>> +	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
>> +		if ((stit->tbl == tbl) && (stit->group == group)) {
>> +			ret = -EBUSY;
>> +			goto put_exit;
>> +		}
>> +	}
>> +
>> +	iommu_table_get(tbl);
>> +
>> +	stit = kzalloc(sizeof(*stit), GFP_KERNEL);
>> +	stit->tbl = tbl;
>> +	stit->group = group;
>> +
>> +	list_add_rcu(&stit->next, &stt->iommu_tables);
>> +
>> +put_exit:
>> +	iommu_group_put(grp);
>> +
>> +	return ret;
>> +}
>> +
>>  static void release_spapr_tce_table(struct rcu_head *head)
>>  {
>>  	struct kvmppc_spapr_tce_table *stt = container_of(head,
>> @@ -133,6 +291,8 @@ static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
>>  
>>  	list_del_rcu(&stt->list);
>>  
>> +	kvm_spapr_tce_liobn_release_iommu_group(stt, NULL /* release all */);
>> +
>>  	kvm_put_kvm(stt->kvm);
>>  
>>  	kvmppc_account_memlimit(
>> @@ -183,6 +343,7 @@ 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 +372,101 @@ 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 +483,35 @@ 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 +526,7 @@ 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 +565,26 @@ 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 +601,7 @@ 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 +615,24 @@ 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);
>>  
>> diff --git a/arch/powerpc/kvm/book3s_64_vio_hv.c b/arch/powerpc/kvm/book3s_64_vio_hv.c
>> index 440d3ab5dc32..eda0a8f6fae8 100644
>> --- a/arch/powerpc/kvm/book3s_64_vio_hv.c
>> +++ b/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);
>>  
>> diff --git a/arch/powerpc/kvm/powerpc.c b/arch/powerpc/kvm/powerpc.c
>> index 95c91a9de351..62bdd6c48107 100644
>> --- a/arch/powerpc/kvm/powerpc.c
>> +++ b/arch/powerpc/kvm/powerpc.c
>> @@ -538,6 +538,8 @@ 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:
>> diff --git a/virt/kvm/vfio.c b/virt/kvm/vfio.c
>> index d32f239eb471..2b7dc22265fe 100644
>> --- a/virt/kvm/vfio.c
>> +++ b/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;
>> @@ -211,6 +215,9 @@ 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_iommu_group(dev->kvm, vfio_group);
>> +#endif
>>  		kvm_vfio_group_set_kvm(vfio_group, NULL);
>>  
>>  		kvm_vfio_group_put_external_user(vfio_group);
>> @@ -218,6 +225,53 @@ 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;
>> +		unsigned long minsz;
>> +		struct kvm_vfio *kv = dev->private;
>> +		struct vfio_group *vfio_group;
>> +		struct kvm_vfio_group *kvg;
>> +		struct fd f;
>> +
>> +		minsz = offsetofend(struct kvm_vfio_spapr_tce, tablefd);
>> +
>> +		if (copy_from_user(&param, (void __user *)arg, minsz))
>> +			return -EFAULT;
>> +
>> +		if (param.argsz < minsz || param.flags)
>> +			return -EINVAL;
>> +
>> +		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);
>> +
>> +		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, vfio_group);
>> +
>> +			break;
>> +		}
>> +
>> +		mutex_unlock(&kv->lock);
>> +
>> +		return ret;
>> +	}
>> +#endif /* CONFIG_SPAPR_TCE_IOMMU */
> 
> 
> The group reference is leaked if kvm_spapr_tce_attach_iommu_group()
> fails.  My preference would be to not hold that separate group
> reference in the spapr code anyway, having a parallel life cycle over
> there is confusing and results in ugliness like duplicating 
> kvm_vfio_group_put_external_user().  Thanks,
> 
> Alex
> 
>>  	}
>>  
>>  	return -ENXIO;
>> @@ -242,6 +296,9 @@ 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 +314,9 @@ 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_iommu_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);
>
Alex Williamson March 15, 2017, 4:18 p.m. | #5
On Wed, 15 Mar 2017 15:40:14 +1100
David Gibson <david@gibson.dropbear.id.au> wrote:
> > > diff --git a/arch/powerpc/kvm/book3s_64_vio.c b/arch/powerpc/kvm/book3s_64_vio.c
> > > index e96a4590464c..be18cda01e1b 100644
> > > --- a/arch/powerpc/kvm/book3s_64_vio.c
> > > +++ b/arch/powerpc/kvm/book3s_64_vio.c
> > > @@ -28,6 +28,10 @@
> > >  #include <linux/hugetlb.h>
> > >  #include <linux/list.h>
> > >  #include <linux/anon_inodes.h>
> > > +#include <linux/iommu.h>
> > > +#include <linux/file.h>
> > > +#include <linux/vfio.h>
> > > +#include <linux/module.h>
> > >  
> > >  #include <asm/tlbflush.h>
> > >  #include <asm/kvm_ppc.h>
> > > @@ -40,6 +44,36 @@
> > >  #include <asm/udbg.h>
> > >  #include <asm/iommu.h>
> > >  #include <asm/tce.h>
> > > +#include <asm/mmu_context.h>
> > > +
> > > +static void kvm_vfio_group_put_external_user(struct vfio_group *vfio_group)
> > > +{
> > > +	void (*fn)(struct vfio_group *);
> > > +
> > > +	fn = symbol_get(vfio_group_put_external_user);
> > > +	if (WARN_ON(!fn))
> > > +		return;
> > > +
> > > +	fn(vfio_group);
> > > +
> > > +	symbol_put(vfio_group_put_external_user);
> > > +}
> > > +
> > > +static int kvm_vfio_external_user_iommu_id(struct vfio_group *vfio_group)
> > > +{
> > > +	int (*fn)(struct vfio_group *);
> > > +	int ret = -1;
> > > +
> > > +	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;
> > > +}  
> > 
> > 
> > Ugh.  This feels so wrong.  Why can't you have kvm-vfio pass the
> > iommu_group?  Why do you need to hold this additional vfio_group
> > reference?  
> 
> Keeping the vfio_group reference makes sense to me, since we don't
> want the vfio context for the group to go away while it's attached to
> the LIOBN.

But there's already a reference for that, it's taken by
KVM_DEV_VFIO_GROUP_ADD and held until KVM_DEV_VFIO_GROUP_DEL.  Both the
DEL path and the cleanup path call kvm_spapr_tce_release_iommu_group()
before releasing that reference, so it seems entirely redundant.

> However, going via the iommu_id rather than just having an interface
> to directly grab the iommu group from the vfio_group seems bizarre to
> me.  I'm ok with cleaning that up later, however.

We have kvm_spapr_tce_attach_iommu_group() and
kvm_spapr_tce_release_iommu_group(), but both take a vfio_group, not an
iommu_group as a parameter.  I don't particularly have a problem with
the vfio_group -> iommu ID -> iommu_group, but if we drop the extra
vfio_group reference and pass the iommu_group itself to these functions
then we can keep all the symbol reference stuff in the kvm-vfio glue
layer.  Thanks,

Alex
--
To unsubscribe from this list: send the line "unsubscribe kvm-ppc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Alex Williamson March 15, 2017, 4:39 p.m. | #6
On Thu, 16 Mar 2017 00:21:07 +1100
Alexey Kardashevskiy <aik@ozlabs.ru> wrote:

> On 15/03/17 08:05, Alex Williamson wrote:
> > On Fri, 10 Mar 2017 14:53:37 +1100
> > Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
> >   
> >> 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.
> >>
> >> As this creates a descriptor per IOMMU table-LIOBN couple (called
> >> kvmppc_spapr_tce_iommu_table), it is possible to have several
> >> descriptors with the same iommu_table (hardware IOMMU table) attached
> >> to the same LIOBN; we do not remove duplicates though as
> >> iommu_table_ops::exchange not just update a TCE entry (which is
> >> shared among IOMMU groups) but also invalidates the TCE cache
> >> (one per IOMMU group).
> >>
> >> 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>
> >> ---
> >> Changes:
> >> v8:
> >> * changed all (!pua) checks to return H_TOO_HARD as ioctl() is supposed
> >> to handle them
> >> * changed vmalloc_to_phys() callers to return H_HARDWARE
> >> * changed real mode iommu_tce_xchg_rm() callers to return H_TOO_HARD
> >> and added a comment about this in the code
> >> * changed virtual mode iommu_tce_xchg() callers to return H_HARDWARE
> >> and do WARN_ON
> >> * added WARN_ON_ONCE_RM(!rmap) in kvmppc_rm_h_put_tce_indirect() to
> >> have all vmalloc_to_phys() callsites covered
> >>
> >> v7:
> >> * added realmode-friendly WARN_ON_ONCE_RM
> >>
> >> v6:
> >> * changed handling of errors returned by kvmppc_(rm_)tce_iommu_(un)map()
> >> * moved kvmppc_gpa_to_ua() to TCE validation
> >>
> >> v5:
> >> * changed error codes in multiple places
> >> * added bunch of WARN_ON() in places which should not really happen
> >> * adde a check that an iommu table is not attached already to LIOBN
> >> * dropped explicit calls to iommu_tce_clear_param_check/
> >> iommu_tce_put_param_check as kvmppc_tce_validate/kvmppc_ioba_validate
> >> call them anyway (since the previous patch)
> >> * if we fail to update a hardware IOMMU table for unexpected reason,
> >> this just clears the entry
> >>
> >> v4:
> >> * added note to the commit log about allowing multiple updates of
> >> the same IOMMU table;
> >> * instead of checking for if any memory was preregistered, this
> >> returns H_TOO_HARD if a specific page was not;
> >> * fixed comments from v3 about error handling in many places;
> >> * simplified TCE handlers and merged IOMMU parts inline - for example,
> >> there used to be kvmppc_h_put_tce_iommu(), now it is merged into
> >> kvmppc_h_put_tce(); this allows to check IOBA boundaries against
> >> the first attached table only (makes the code simpler);
> >>
> >> v3:
> >> * simplified not to use VFIO group notifiers
> >> * reworked cleanup, should be cleaner/simpler now
> >>
> >> v2:
> >> * reworked to use new VFIO notifiers
> >> * now same iommu_table may appear in the list several times, to be fixed later
> >> ---
> >>  Documentation/virtual/kvm/devices/vfio.txt |  22 +-
> >>  arch/powerpc/include/asm/kvm_host.h        |   8 +
> >>  arch/powerpc/include/asm/kvm_ppc.h         |   4 +
> >>  include/uapi/linux/kvm.h                   |   8 +
> >>  arch/powerpc/kvm/book3s_64_vio.c           | 323 ++++++++++++++++++++++++++++-
> >>  arch/powerpc/kvm/book3s_64_vio_hv.c        | 201 +++++++++++++++++-
> >>  arch/powerpc/kvm/powerpc.c                 |   2 +
> >>  virt/kvm/vfio.c                            |  60 ++++++
> >>  8 files changed, 623 insertions(+), 5 deletions(-)
> >>
> >> diff --git a/Documentation/virtual/kvm/devices/vfio.txt b/Documentation/virtual/kvm/devices/vfio.txt
> >> index ef51740c67ca..f95d867168ea 100644
> >> --- a/Documentation/virtual/kvm/devices/vfio.txt
> >> +++ b/Documentation/virtual/kvm/devices/vfio.txt
> >> @@ -16,7 +16,25 @@ 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 {
> >> +		__u32	argsz;
> >> +		__u32	flags;
> >> +		__s32	groupfd;
> >> +		__s32	tablefd;
> >> +	};
> >> +
> >> +	where
> >> +	@argsz is the size of kvm_vfio_spapr_tce_liobn;  
> > 
> > kvm_vfio_spapr_tce_liobn does not exist.  s/_liobn//?  
> 
> Correct.
> 
> 
> >   
> >> +	@flags are not supported now, must be zero;  
> > 
> > We do this argsz/flags thing on vfio ioctls because ioctls are a bit
> > more of a restricted resource.  We don't want to burn through them so
> > we make them expandable.  I don't know that we have that restriction
> > here and the ADD/DEL support certainly doesn't include it.  Maybe this
> > isn't necessary?  
> 
> 
> It is not but since I am going to have padding there, I thought I give it a
> name. Totally pointless and "u8 padding[4]" is better?

Or since these are not ioctls, we simply consider them to have a very
specific and limited purpose.  If we have a need to expand on that,
define a new one that either builds on the existing ones, as
GROUP_SET_SPAPR_TCE builds on GROUP_ADD, or supersedes them.  For
ioctls, it's more costly if we abandon one, here it doesn't seem like a
big deal.  Thanks,

Alex
--
To unsubscribe from this list: send the line "unsubscribe kvm-ppc" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Alexey Kardashevskiy March 15, 2017, 11:39 p.m. | #7
On 16/03/17 03:39, Alex Williamson wrote:
> On Thu, 16 Mar 2017 00:21:07 +1100
> Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
> 
>> On 15/03/17 08:05, Alex Williamson wrote:
>>> On Fri, 10 Mar 2017 14:53:37 +1100
>>> Alexey Kardashevskiy <aik@ozlabs.ru> wrote:
>>>   
>>>> 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.
>>>>
>>>> As this creates a descriptor per IOMMU table-LIOBN couple (called
>>>> kvmppc_spapr_tce_iommu_table), it is possible to have several
>>>> descriptors with the same iommu_table (hardware IOMMU table) attached
>>>> to the same LIOBN; we do not remove duplicates though as
>>>> iommu_table_ops::exchange not just update a TCE entry (which is
>>>> shared among IOMMU groups) but also invalidates the TCE cache
>>>> (one per IOMMU group).
>>>>
>>>> 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>
>>>> ---
>>>> Changes:
>>>> v8:
>>>> * changed all (!pua) checks to return H_TOO_HARD as ioctl() is supposed
>>>> to handle them
>>>> * changed vmalloc_to_phys() callers to return H_HARDWARE
>>>> * changed real mode iommu_tce_xchg_rm() callers to return H_TOO_HARD
>>>> and added a comment about this in the code
>>>> * changed virtual mode iommu_tce_xchg() callers to return H_HARDWARE
>>>> and do WARN_ON
>>>> * added WARN_ON_ONCE_RM(!rmap) in kvmppc_rm_h_put_tce_indirect() to
>>>> have all vmalloc_to_phys() callsites covered
>>>>
>>>> v7:
>>>> * added realmode-friendly WARN_ON_ONCE_RM
>>>>
>>>> v6:
>>>> * changed handling of errors returned by kvmppc_(rm_)tce_iommu_(un)map()
>>>> * moved kvmppc_gpa_to_ua() to TCE validation
>>>>
>>>> v5:
>>>> * changed error codes in multiple places
>>>> * added bunch of WARN_ON() in places which should not really happen
>>>> * adde a check that an iommu table is not attached already to LIOBN
>>>> * dropped explicit calls to iommu_tce_clear_param_check/
>>>> iommu_tce_put_param_check as kvmppc_tce_validate/kvmppc_ioba_validate
>>>> call them anyway (since the previous patch)
>>>> * if we fail to update a hardware IOMMU table for unexpected reason,
>>>> this just clears the entry
>>>>
>>>> v4:
>>>> * added note to the commit log about allowing multiple updates of
>>>> the same IOMMU table;
>>>> * instead of checking for if any memory was preregistered, this
>>>> returns H_TOO_HARD if a specific page was not;
>>>> * fixed comments from v3 about error handling in many places;
>>>> * simplified TCE handlers and merged IOMMU parts inline - for example,
>>>> there used to be kvmppc_h_put_tce_iommu(), now it is merged into
>>>> kvmppc_h_put_tce(); this allows to check IOBA boundaries against
>>>> the first attached table only (makes the code simpler);
>>>>
>>>> v3:
>>>> * simplified not to use VFIO group notifiers
>>>> * reworked cleanup, should be cleaner/simpler now
>>>>
>>>> v2:
>>>> * reworked to use new VFIO notifiers
>>>> * now same iommu_table may appear in the list several times, to be fixed later
>>>> ---
>>>>  Documentation/virtual/kvm/devices/vfio.txt |  22 +-
>>>>  arch/powerpc/include/asm/kvm_host.h        |   8 +
>>>>  arch/powerpc/include/asm/kvm_ppc.h         |   4 +
>>>>  include/uapi/linux/kvm.h                   |   8 +
>>>>  arch/powerpc/kvm/book3s_64_vio.c           | 323 ++++++++++++++++++++++++++++-
>>>>  arch/powerpc/kvm/book3s_64_vio_hv.c        | 201 +++++++++++++++++-
>>>>  arch/powerpc/kvm/powerpc.c                 |   2 +
>>>>  virt/kvm/vfio.c                            |  60 ++++++
>>>>  8 files changed, 623 insertions(+), 5 deletions(-)
>>>>
>>>> diff --git a/Documentation/virtual/kvm/devices/vfio.txt b/Documentation/virtual/kvm/devices/vfio.txt
>>>> index ef51740c67ca..f95d867168ea 100644
>>>> --- a/Documentation/virtual/kvm/devices/vfio.txt
>>>> +++ b/Documentation/virtual/kvm/devices/vfio.txt
>>>> @@ -16,7 +16,25 @@ 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 {
>>>> +		__u32	argsz;
>>>> +		__u32	flags;
>>>> +		__s32	groupfd;
>>>> +		__s32	tablefd;
>>>> +	};
>>>> +
>>>> +	where
>>>> +	@argsz is the size of kvm_vfio_spapr_tce_liobn;  
>>>
>>> kvm_vfio_spapr_tce_liobn does not exist.  s/_liobn//?  
>>
>> Correct.
>>
>>
>>>   
>>>> +	@flags are not supported now, must be zero;  
>>>
>>> We do this argsz/flags thing on vfio ioctls because ioctls are a bit
>>> more of a restricted resource.  We don't want to burn through them so
>>> we make them expandable.  I don't know that we have that restriction
>>> here and the ADD/DEL support certainly doesn't include it.  Maybe this
>>> isn't necessary?  
>>
>>
>> It is not but since I am going to have padding there, I thought I give it a
>> name. Totally pointless and "u8 padding[4]" is better?
> 
> Or since these are not ioctls, we simply consider them to have a very
> specific and limited purpose.  If we have a need to expand on that,
> define a new one that either builds on the existing ones, as
> GROUP_SET_SPAPR_TCE builds on GROUP_ADD, or supersedes them.  For
> ioctls, it's more costly if we abandon one, here it doesn't seem like a
> big deal.  Thanks,

Did you just suggest ditching the extra 4 bytes now? Sorry, I am not
following you here. I am just saying that having 3*u32 while the actual
size is 4*u32 is not good for binary interface.

Also, you have not noticed my other comment about ditching the group
reference or decided to leave it for David to comment? :) Thanks.
David Gibson March 16, 2017, 3:42 a.m. | #8
On Wed, Mar 15, 2017 at 10:18:18AM -0600, Alex Williamson wrote:
> On Wed, 15 Mar 2017 15:40:14 +1100
> David Gibson <david@gibson.dropbear.id.au> wrote:
> > > > diff --git a/arch/powerpc/kvm/book3s_64_vio.c b/arch/powerpc/kvm/book3s_64_vio.c
> > > > index e96a4590464c..be18cda01e1b 100644
> > > > --- a/arch/powerpc/kvm/book3s_64_vio.c
> > > > +++ b/arch/powerpc/kvm/book3s_64_vio.c
> > > > @@ -28,6 +28,10 @@
> > > >  #include <linux/hugetlb.h>
> > > >  #include <linux/list.h>
> > > >  #include <linux/anon_inodes.h>
> > > > +#include <linux/iommu.h>
> > > > +#include <linux/file.h>
> > > > +#include <linux/vfio.h>
> > > > +#include <linux/module.h>
> > > >  
> > > >  #include <asm/tlbflush.h>
> > > >  #include <asm/kvm_ppc.h>
> > > > @@ -40,6 +44,36 @@
> > > >  #include <asm/udbg.h>
> > > >  #include <asm/iommu.h>
> > > >  #include <asm/tce.h>
> > > > +#include <asm/mmu_context.h>
> > > > +
> > > > +static void kvm_vfio_group_put_external_user(struct vfio_group *vfio_group)
> > > > +{
> > > > +	void (*fn)(struct vfio_group *);
> > > > +
> > > > +	fn = symbol_get(vfio_group_put_external_user);
> > > > +	if (WARN_ON(!fn))
> > > > +		return;
> > > > +
> > > > +	fn(vfio_group);
> > > > +
> > > > +	symbol_put(vfio_group_put_external_user);
> > > > +}
> > > > +
> > > > +static int kvm_vfio_external_user_iommu_id(struct vfio_group *vfio_group)
> > > > +{
> > > > +	int (*fn)(struct vfio_group *);
> > > > +	int ret = -1;
> > > > +
> > > > +	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;
> > > > +}  
> > > 
> > > 
> > > Ugh.  This feels so wrong.  Why can't you have kvm-vfio pass the
> > > iommu_group?  Why do you need to hold this additional vfio_group
> > > reference?  
> > 
> > Keeping the vfio_group reference makes sense to me, since we don't
> > want the vfio context for the group to go away while it's attached to
> > the LIOBN.
> 
> But there's already a reference for that, it's taken by
> KVM_DEV_VFIO_GROUP_ADD and held until KVM_DEV_VFIO_GROUP_DEL.  Both the
> DEL path and the cleanup path call kvm_spapr_tce_release_iommu_group()
> before releasing that reference, so it seems entirely redundant.

Oh, good point.  And we already verify that the group has been ADDed
before setting the LIOBN association.

> > However, going via the iommu_id rather than just having an interface
> > to directly grab the iommu group from the vfio_group seems bizarre to
> > me.  I'm ok with cleaning that up later, however.
> 
> We have kvm_spapr_tce_attach_iommu_group() and
> kvm_spapr_tce_release_iommu_group(), but both take a vfio_group, not an
> iommu_group as a parameter.  I don't particularly have a problem with
> the vfio_group -> iommu ID -> iommu_group, but if we drop the extra
> vfio_group reference and pass the iommu_group itself to these functions
> then we can keep all the symbol reference stuff in the kvm-vfio glue
> layer.  Thanks,

Makes sense.

Patch

diff --git a/Documentation/virtual/kvm/devices/vfio.txt b/Documentation/virtual/kvm/devices/vfio.txt
index ef51740c67ca..f95d867168ea 100644
--- a/Documentation/virtual/kvm/devices/vfio.txt
+++ b/Documentation/virtual/kvm/devices/vfio.txt
@@ -16,7 +16,25 @@  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 {
+		__u32	argsz;
+		__u32	flags;
+		__s32	groupfd;
+		__s32	tablefd;
+	};
+
+	where
+	@argsz is the size of kvm_vfio_spapr_tce_liobn;
+	@flags are not supported now, must be zero;
+	@groupfd is a file descriptor for a VFIO group;
+	@tablefd is a file descriptor for a TCE table allocated via
+		KVM_CREATE_SPAPR_TCE.
diff --git a/arch/powerpc/include/asm/kvm_host.h b/arch/powerpc/include/asm/kvm_host.h
index 7bba8f415627..857ae2c6aa39 100644
--- a/arch/powerpc/include/asm/kvm_host.h
+++ b/arch/powerpc/include/asm/kvm_host.h
@@ -191,6 +191,13 @@  struct kvmppc_pginfo {
 	atomic_t refcnt;
 };
 
+struct kvmppc_spapr_tce_iommu_table {
+	struct rcu_head rcu;
+	struct list_head next;
+	struct vfio_group *group;
+	struct iommu_table *tbl;
+};
+
 struct kvmppc_spapr_tce_table {
 	struct list_head list;
 	struct kvm *kvm;
@@ -199,6 +206,7 @@  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];
 };
 
diff --git a/arch/powerpc/include/asm/kvm_ppc.h b/arch/powerpc/include/asm/kvm_ppc.h
index 72c2a155641f..66de7e73b3d3 100644
--- a/arch/powerpc/include/asm/kvm_ppc.h
+++ b/arch/powerpc/include/asm/kvm_ppc.h
@@ -164,6 +164,10 @@  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 vfio_group *group);
+extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
+		struct vfio_group *group);
 
 extern long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
 				struct kvm_create_spapr_tce_64 *args);
diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
index f5a52ffb6b58..e743cb0d176e 100644
--- a/include/uapi/linux/kvm.h
+++ b/include/uapi/linux/kvm.h
@@ -1088,6 +1088,7 @@  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,
@@ -1109,6 +1110,13 @@  enum kvm_device_type {
 	KVM_DEV_TYPE_MAX,
 };
 
+struct kvm_vfio_spapr_tce {
+	__u32	argsz;
+	__u32	flags;
+	__s32	groupfd;
+	__s32	tablefd;
+};
+
 /*
  * ioctls for VM fds
  */
diff --git a/arch/powerpc/kvm/book3s_64_vio.c b/arch/powerpc/kvm/book3s_64_vio.c
index e96a4590464c..be18cda01e1b 100644
--- a/arch/powerpc/kvm/book3s_64_vio.c
+++ b/arch/powerpc/kvm/book3s_64_vio.c
@@ -28,6 +28,10 @@ 
 #include <linux/hugetlb.h>
 #include <linux/list.h>
 #include <linux/anon_inodes.h>
+#include <linux/iommu.h>
+#include <linux/file.h>
+#include <linux/vfio.h>
+#include <linux/module.h>
 
 #include <asm/tlbflush.h>
 #include <asm/kvm_ppc.h>
@@ -40,6 +44,36 @@ 
 #include <asm/udbg.h>
 #include <asm/iommu.h>
 #include <asm/tce.h>
+#include <asm/mmu_context.h>
+
+static void kvm_vfio_group_put_external_user(struct vfio_group *vfio_group)
+{
+	void (*fn)(struct vfio_group *);
+
+	fn = symbol_get(vfio_group_put_external_user);
+	if (WARN_ON(!fn))
+		return;
+
+	fn(vfio_group);
+
+	symbol_put(vfio_group_put_external_user);
+}
+
+static int kvm_vfio_external_user_iommu_id(struct vfio_group *vfio_group)
+{
+	int (*fn)(struct vfio_group *);
+	int ret = -1;
+
+	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 unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
 {
@@ -91,6 +125,130 @@  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_table_put(stit->tbl);
+	kvm_vfio_group_put_external_user(stit->group);
+
+	kfree(stit);
+}
+
+static void kvm_spapr_tce_liobn_release_iommu_group(
+		struct kvmppc_spapr_tce_table *stt,
+		struct vfio_group *group)
+{
+	struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
+
+	list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
+		if (group && (stit->group != group))
+			continue;
+
+		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 vfio_group *group)
+{
+	struct kvmppc_spapr_tce_table *stt;
+
+	list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list)
+		kvm_spapr_tce_liobn_release_iommu_group(stt, group);
+}
+
+extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
+		struct vfio_group *group)
+{
+	struct kvmppc_spapr_tce_table *stt = NULL;
+	bool found = false;
+	struct iommu_table *tbl = NULL;
+	struct iommu_table_group *table_group;
+	long i, ret = 0;
+	struct kvmppc_spapr_tce_iommu_table *stit;
+	struct fd f;
+	int group_id;
+	struct iommu_group *grp;
+
+	group_id = kvm_vfio_external_user_iommu_id(group);
+	grp = iommu_group_get_by_id(group_id);
+	if (WARN_ON(!grp))
+		return -EIO;
+
+	f = fdget(tablefd);
+	if (!f.file) {
+		ret = -EBADF;
+		goto put_exit;
+	}
+
+	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) {
+		ret = -EINVAL;
+		goto put_exit;
+	}
+
+	table_group = iommu_group_get_iommudata(grp);
+	if (WARN_ON(!table_group)) {
+		ret = -EFAULT;
+		goto put_exit;
+	}
+
+	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)) {
+			tbl = tbltmp;
+			break;
+		}
+	}
+	if (!tbl) {
+		ret = -EINVAL;
+		goto put_exit;
+	}
+
+	list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
+		if ((stit->tbl == tbl) && (stit->group == group)) {
+			ret = -EBUSY;
+			goto put_exit;
+		}
+	}
+
+	iommu_table_get(tbl);
+
+	stit = kzalloc(sizeof(*stit), GFP_KERNEL);
+	stit->tbl = tbl;
+	stit->group = group;
+
+	list_add_rcu(&stit->next, &stt->iommu_tables);
+
+put_exit:
+	iommu_group_put(grp);
+
+	return ret;
+}
+
 static void release_spapr_tce_table(struct rcu_head *head)
 {
 	struct kvmppc_spapr_tce_table *stt = container_of(head,
@@ -133,6 +291,8 @@  static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
 
 	list_del_rcu(&stt->list);
 
+	kvm_spapr_tce_liobn_release_iommu_group(stt, NULL /* release all */);
+
 	kvm_put_kvm(stt->kvm);
 
 	kvmppc_account_memlimit(
@@ -183,6 +343,7 @@  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 +372,101 @@  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 +483,35 @@  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 +526,7 @@  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 +565,26 @@  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 +601,7 @@  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 +615,24 @@  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);
 
diff --git a/arch/powerpc/kvm/book3s_64_vio_hv.c b/arch/powerpc/kvm/book3s_64_vio_hv.c
index 440d3ab5dc32..eda0a8f6fae8 100644
--- a/arch/powerpc/kvm/book3s_64_vio_hv.c
+++ b/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);
 
diff --git a/arch/powerpc/kvm/powerpc.c b/arch/powerpc/kvm/powerpc.c
index 95c91a9de351..62bdd6c48107 100644
--- a/arch/powerpc/kvm/powerpc.c
+++ b/arch/powerpc/kvm/powerpc.c
@@ -538,6 +538,8 @@  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:
diff --git a/virt/kvm/vfio.c b/virt/kvm/vfio.c
index d32f239eb471..2b7dc22265fe 100644
--- a/virt/kvm/vfio.c
+++ b/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;
@@ -211,6 +215,9 @@  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_iommu_group(dev->kvm, vfio_group);
+#endif
 		kvm_vfio_group_set_kvm(vfio_group, NULL);
 
 		kvm_vfio_group_put_external_user(vfio_group);
@@ -218,6 +225,53 @@  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;
+		unsigned long minsz;
+		struct kvm_vfio *kv = dev->private;
+		struct vfio_group *vfio_group;
+		struct kvm_vfio_group *kvg;
+		struct fd f;
+
+		minsz = offsetofend(struct kvm_vfio_spapr_tce, tablefd);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz || param.flags)
+			return -EINVAL;
+
+		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);
+
+		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, vfio_group);
+
+			break;
+		}
+
+		mutex_unlock(&kv->lock);
+
+		return ret;
+	}
+#endif /* CONFIG_SPAPR_TCE_IOMMU */
 	}
 
 	return -ENXIO;
@@ -242,6 +296,9 @@  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 +314,9 @@  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_iommu_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);