linux/arch/powerpc/kvm/book3s_64_vio.c

717 lines
17 KiB
C
Raw Normal View History

/*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License, version 2, as
* published by the Free Software Foundation.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301, USA.
*
* Copyright 2010 Paul Mackerras, IBM Corp. <paulus@au1.ibm.com>
* Copyright 2011 David Gibson, IBM Corporation <dwg@au1.ibm.com>
* Copyright 2016 Alexey Kardashevskiy, IBM Corporation <aik@au1.ibm.com>
*/
#include <linux/types.h>
#include <linux/string.h>
#include <linux/kvm.h>
#include <linux/kvm_host.h>
#include <linux/highmem.h>
#include <linux/gfp.h>
#include <linux/slab.h>
#include <linux/sched/signal.h>
#include <linux/hugetlb.h>
#include <linux/list.h>
#include <linux/anon_inodes.h>
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
#include <linux/iommu.h>
#include <linux/file.h>
#include <asm/kvm_ppc.h>
#include <asm/kvm_book3s.h>
#include <asm/book3s/64/mmu-hash.h>
#include <asm/hvcall.h>
#include <asm/synch.h>
#include <asm/ppc-opcode.h>
#include <asm/kvm_host.h>
#include <asm/udbg.h>
#include <asm/iommu.h>
#include <asm/tce.h>
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
#include <asm/mmu_context.h>
static unsigned long kvmppc_tce_pages(unsigned long iommu_pages)
{
return ALIGN(iommu_pages * sizeof(u64), PAGE_SIZE) / PAGE_SIZE;
}
static unsigned long kvmppc_stt_pages(unsigned long tce_pages)
{
unsigned long stt_bytes = sizeof(struct kvmppc_spapr_tce_table) +
(tce_pages * sizeof(struct page *));
return tce_pages + ALIGN(stt_bytes, PAGE_SIZE) / PAGE_SIZE;
}
static long kvmppc_account_memlimit(unsigned long stt_pages, bool inc)
{
long ret = 0;
if (!current || !current->mm)
return ret; /* process exited */
down_write(&current->mm->mmap_sem);
if (inc) {
unsigned long locked, lock_limit;
locked = current->mm->locked_vm + stt_pages;
lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
if (locked > lock_limit && !capable(CAP_IPC_LOCK))
ret = -ENOMEM;
else
current->mm->locked_vm += stt_pages;
} else {
if (WARN_ON_ONCE(stt_pages > current->mm->locked_vm))
stt_pages = current->mm->locked_vm;
current->mm->locked_vm -= stt_pages;
}
pr_debug("[%d] RLIMIT_MEMLOCK KVM %c%ld %ld/%ld%s\n", current->pid,
inc ? '+' : '-',
stt_pages << PAGE_SHIFT,
current->mm->locked_vm << PAGE_SHIFT,
rlimit(RLIMIT_MEMLOCK),
ret ? " - exceeded" : "");
up_write(&current->mm->mmap_sem);
return ret;
}
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
static void kvm_spapr_tce_iommu_table_free(struct rcu_head *head)
{
struct kvmppc_spapr_tce_iommu_table *stit = container_of(head,
struct kvmppc_spapr_tce_iommu_table, rcu);
iommu_tce_table_put(stit->tbl);
kfree(stit);
}
static void kvm_spapr_tce_liobn_put(struct kref *kref)
{
struct kvmppc_spapr_tce_iommu_table *stit = container_of(kref,
struct kvmppc_spapr_tce_iommu_table, kref);
list_del_rcu(&stit->next);
call_rcu(&stit->rcu, kvm_spapr_tce_iommu_table_free);
}
extern void kvm_spapr_tce_release_iommu_group(struct kvm *kvm,
struct iommu_group *grp)
{
int i;
struct kvmppc_spapr_tce_table *stt;
struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
struct iommu_table_group *table_group = NULL;
list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
table_group = iommu_group_get_iommudata(grp);
if (WARN_ON(!table_group))
continue;
list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
if (table_group->tables[i] != stit->tbl)
continue;
kref_put(&stit->kref, kvm_spapr_tce_liobn_put);
return;
}
}
}
}
extern long kvm_spapr_tce_attach_iommu_group(struct kvm *kvm, int tablefd,
struct iommu_group *grp)
{
struct kvmppc_spapr_tce_table *stt = NULL;
bool found = false;
struct iommu_table *tbl = NULL;
struct iommu_table_group *table_group;
long i;
struct kvmppc_spapr_tce_iommu_table *stit;
struct fd f;
f = fdget(tablefd);
if (!f.file)
return -EBADF;
list_for_each_entry_rcu(stt, &kvm->arch.spapr_tce_tables, list) {
if (stt == f.file->private_data) {
found = true;
break;
}
}
fdput(f);
if (!found)
return -EINVAL;
table_group = iommu_group_get_iommudata(grp);
if (WARN_ON(!table_group))
return -EFAULT;
for (i = 0; i < IOMMU_TABLE_GROUP_MAX_TABLES; ++i) {
struct iommu_table *tbltmp = table_group->tables[i];
if (!tbltmp)
continue;
/* Make sure hardware table parameters are compatible */
if ((tbltmp->it_page_shift <= stt->page_shift) &&
(tbltmp->it_offset << tbltmp->it_page_shift ==
stt->offset << stt->page_shift) &&
KVM: PPC: Book3S: Fix matching of hardware and emulated TCE tables When attaching a hardware table to LIOBN in KVM, we match table parameters such as page size, table offset and table size. However the tables are created via very different paths - VFIO and KVM - and the VFIO path goes through the platform code which has minimum TCE page size requirement (which is 4K but since we allocate memory by pages and cannot avoid alignment anyway, we align to 64k pages for powernv_defconfig). So when we match the tables, one might be bigger that the other which means the hardware table cannot get attached to LIOBN and DMA mapping fails. This removes the table size alignment from the guest visible table. This does not affect the memory allocation which is still aligned - kvmppc_tce_pages() takes care of this. This relaxes the check we do when attaching tables to allow the hardware table be bigger than the guest visible table. Ideally we want the KVM table to cover the same space as the hardware table does but since the hardware table may use multiple levels, and all levels must use the same table size (IODA2 design), the area it can actually cover might get very different from the window size which the guest requested, even though the guest won't map it all. Fixes: ca1fc489cf "KVM: PPC: Book3S: Allow backing bigger guest IOMMU pages with smaller physical pages" Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-06-20 16:42:58 +08:00
(tbltmp->it_size << tbltmp->it_page_shift >=
stt->size << stt->page_shift)) {
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
/*
* Reference the table to avoid races with
* add/remove DMA windows.
*/
tbl = iommu_tce_table_get(tbltmp);
break;
}
}
if (!tbl)
return -EINVAL;
list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
if (tbl != stit->tbl)
continue;
if (!kref_get_unless_zero(&stit->kref)) {
/* stit is being destroyed */
iommu_tce_table_put(tbl);
return -ENOTTY;
}
/*
* The table is already known to this KVM, we just increased
* its KVM reference counter and can return.
*/
return 0;
}
stit = kzalloc(sizeof(*stit), GFP_KERNEL);
if (!stit) {
iommu_tce_table_put(tbl);
return -ENOMEM;
}
stit->tbl = tbl;
kref_init(&stit->kref);
list_add_rcu(&stit->next, &stt->iommu_tables);
return 0;
}
static void release_spapr_tce_table(struct rcu_head *head)
{
struct kvmppc_spapr_tce_table *stt = container_of(head,
struct kvmppc_spapr_tce_table, rcu);
unsigned long i, npages = kvmppc_tce_pages(stt->size);
for (i = 0; i < npages; i++)
__free_page(stt->pages[i]);
kfree(stt);
}
static vm_fault_t kvm_spapr_tce_fault(struct vm_fault *vmf)
{
struct kvmppc_spapr_tce_table *stt = vmf->vma->vm_file->private_data;
struct page *page;
if (vmf->pgoff >= kvmppc_tce_pages(stt->size))
return VM_FAULT_SIGBUS;
page = stt->pages[vmf->pgoff];
get_page(page);
vmf->page = page;
return 0;
}
static const struct vm_operations_struct kvm_spapr_tce_vm_ops = {
.fault = kvm_spapr_tce_fault,
};
static int kvm_spapr_tce_mmap(struct file *file, struct vm_area_struct *vma)
{
vma->vm_ops = &kvm_spapr_tce_vm_ops;
return 0;
}
static int kvm_spapr_tce_release(struct inode *inode, struct file *filp)
{
struct kvmppc_spapr_tce_table *stt = filp->private_data;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
struct kvmppc_spapr_tce_iommu_table *stit, *tmp;
KVM: PPC: Book3S HV: Protect updates to spapr_tce_tables list Al Viro pointed out that while one thread of a process is executing in kvm_vm_ioctl_create_spapr_tce(), another thread could guess the file descriptor returned by anon_inode_getfd() and close() it before the first thread has added it to the kvm->arch.spapr_tce_tables list. That highlights a more general problem: there is no mutual exclusion between writers to the spapr_tce_tables list, leading to the possibility of the list becoming corrupted, which could cause a host kernel crash. To fix the mutual exclusion problem, we add a mutex_lock/unlock pair around the list_del_rce in kvm_spapr_tce_release(). Also, this moves the call to anon_inode_getfd() inside the region protected by the kvm->lock mutex, after we have done the check for a duplicate LIOBN. This means that if another thread does guess the file descriptor and closes it, its call to kvm_spapr_tce_release() will not do any harm because it will have to wait until the first thread has released kvm->lock. With this, there are no failure points in kvm_vm_ioctl_create_spapr_tce() after the call to anon_inode_getfd(). The other things that the second thread could do with the guessed file descriptor are to mmap it or to pass it as a parameter to a KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE ioctl on a KVM device fd. An mmap call won't cause any harm because kvm_spapr_tce_mmap() and kvm_spapr_tce_fault() don't access the spapr_tce_tables list or the kvmppc_spapr_tce_table.list field, and the fields that they do use have been properly initialized by the time of the anon_inode_getfd() call. The KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE ioctl calls kvm_spapr_tce_attach_iommu_group(), which scans the spapr_tce_tables list looking for the kvmppc_spapr_tce_table struct corresponding to the fd given as the parameter. Either it will find the new entry or it won't; if it doesn't, it just returns an error, and if it does, it will function normally. So, in each case there is no harmful effect. Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-08-28 12:31:24 +08:00
struct kvm *kvm = stt->kvm;
KVM: PPC: Book3S HV: Protect updates to spapr_tce_tables list Al Viro pointed out that while one thread of a process is executing in kvm_vm_ioctl_create_spapr_tce(), another thread could guess the file descriptor returned by anon_inode_getfd() and close() it before the first thread has added it to the kvm->arch.spapr_tce_tables list. That highlights a more general problem: there is no mutual exclusion between writers to the spapr_tce_tables list, leading to the possibility of the list becoming corrupted, which could cause a host kernel crash. To fix the mutual exclusion problem, we add a mutex_lock/unlock pair around the list_del_rce in kvm_spapr_tce_release(). Also, this moves the call to anon_inode_getfd() inside the region protected by the kvm->lock mutex, after we have done the check for a duplicate LIOBN. This means that if another thread does guess the file descriptor and closes it, its call to kvm_spapr_tce_release() will not do any harm because it will have to wait until the first thread has released kvm->lock. With this, there are no failure points in kvm_vm_ioctl_create_spapr_tce() after the call to anon_inode_getfd(). The other things that the second thread could do with the guessed file descriptor are to mmap it or to pass it as a parameter to a KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE ioctl on a KVM device fd. An mmap call won't cause any harm because kvm_spapr_tce_mmap() and kvm_spapr_tce_fault() don't access the spapr_tce_tables list or the kvmppc_spapr_tce_table.list field, and the fields that they do use have been properly initialized by the time of the anon_inode_getfd() call. The KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE ioctl calls kvm_spapr_tce_attach_iommu_group(), which scans the spapr_tce_tables list looking for the kvmppc_spapr_tce_table struct corresponding to the fd given as the parameter. Either it will find the new entry or it won't; if it doesn't, it just returns an error, and if it does, it will function normally. So, in each case there is no harmful effect. Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-08-28 12:31:24 +08:00
mutex_lock(&kvm->lock);
list_del_rcu(&stt->list);
KVM: PPC: Book3S HV: Protect updates to spapr_tce_tables list Al Viro pointed out that while one thread of a process is executing in kvm_vm_ioctl_create_spapr_tce(), another thread could guess the file descriptor returned by anon_inode_getfd() and close() it before the first thread has added it to the kvm->arch.spapr_tce_tables list. That highlights a more general problem: there is no mutual exclusion between writers to the spapr_tce_tables list, leading to the possibility of the list becoming corrupted, which could cause a host kernel crash. To fix the mutual exclusion problem, we add a mutex_lock/unlock pair around the list_del_rce in kvm_spapr_tce_release(). Also, this moves the call to anon_inode_getfd() inside the region protected by the kvm->lock mutex, after we have done the check for a duplicate LIOBN. This means that if another thread does guess the file descriptor and closes it, its call to kvm_spapr_tce_release() will not do any harm because it will have to wait until the first thread has released kvm->lock. With this, there are no failure points in kvm_vm_ioctl_create_spapr_tce() after the call to anon_inode_getfd(). The other things that the second thread could do with the guessed file descriptor are to mmap it or to pass it as a parameter to a KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE ioctl on a KVM device fd. An mmap call won't cause any harm because kvm_spapr_tce_mmap() and kvm_spapr_tce_fault() don't access the spapr_tce_tables list or the kvmppc_spapr_tce_table.list field, and the fields that they do use have been properly initialized by the time of the anon_inode_getfd() call. The KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE ioctl calls kvm_spapr_tce_attach_iommu_group(), which scans the spapr_tce_tables list looking for the kvmppc_spapr_tce_table struct corresponding to the fd given as the parameter. Either it will find the new entry or it won't; if it doesn't, it just returns an error, and if it does, it will function normally. So, in each case there is no harmful effect. Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-08-28 12:31:24 +08:00
mutex_unlock(&kvm->lock);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
list_for_each_entry_safe(stit, tmp, &stt->iommu_tables, next) {
WARN_ON(!kref_read(&stit->kref));
while (1) {
if (kref_put(&stit->kref, kvm_spapr_tce_liobn_put))
break;
}
}
kvm_put_kvm(stt->kvm);
kvmppc_account_memlimit(
kvmppc_stt_pages(kvmppc_tce_pages(stt->size)), false);
call_rcu(&stt->rcu, release_spapr_tce_table);
return 0;
}
static const struct file_operations kvm_spapr_tce_fops = {
.mmap = kvm_spapr_tce_mmap,
.release = kvm_spapr_tce_release,
};
long kvm_vm_ioctl_create_spapr_tce(struct kvm *kvm,
struct kvm_create_spapr_tce_64 *args)
{
struct kvmppc_spapr_tce_table *stt = NULL;
struct kvmppc_spapr_tce_table *siter;
KVM: PPC: Book3S: Fix matching of hardware and emulated TCE tables When attaching a hardware table to LIOBN in KVM, we match table parameters such as page size, table offset and table size. However the tables are created via very different paths - VFIO and KVM - and the VFIO path goes through the platform code which has minimum TCE page size requirement (which is 4K but since we allocate memory by pages and cannot avoid alignment anyway, we align to 64k pages for powernv_defconfig). So when we match the tables, one might be bigger that the other which means the hardware table cannot get attached to LIOBN and DMA mapping fails. This removes the table size alignment from the guest visible table. This does not affect the memory allocation which is still aligned - kvmppc_tce_pages() takes care of this. This relaxes the check we do when attaching tables to allow the hardware table be bigger than the guest visible table. Ideally we want the KVM table to cover the same space as the hardware table does but since the hardware table may use multiple levels, and all levels must use the same table size (IODA2 design), the area it can actually cover might get very different from the window size which the guest requested, even though the guest won't map it all. Fixes: ca1fc489cf "KVM: PPC: Book3S: Allow backing bigger guest IOMMU pages with smaller physical pages" Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2018-06-20 16:42:58 +08:00
unsigned long npages, size = args->size;
int ret = -ENOMEM;
int i;
if (!args->size || args->page_shift < 12 || args->page_shift > 34 ||
(args->offset + args->size > (ULLONG_MAX >> args->page_shift)))
return -EINVAL;
npages = kvmppc_tce_pages(size);
ret = kvmppc_account_memlimit(kvmppc_stt_pages(npages), true);
if (ret)
return ret;
ret = -ENOMEM;
stt = kzalloc(sizeof(*stt) + npages * sizeof(struct page *),
GFP_KERNEL);
if (!stt)
goto fail_acct;
stt->liobn = args->liobn;
stt->page_shift = args->page_shift;
stt->offset = args->offset;
stt->size = size;
stt->kvm = kvm;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
INIT_LIST_HEAD_RCU(&stt->iommu_tables);
for (i = 0; i < npages; i++) {
stt->pages[i] = alloc_page(GFP_KERNEL | __GFP_ZERO);
if (!stt->pages[i])
goto fail;
}
mutex_lock(&kvm->lock);
/* Check this LIOBN hasn't been previously allocated */
ret = 0;
list_for_each_entry(siter, &kvm->arch.spapr_tce_tables, list) {
if (siter->liobn == args->liobn) {
ret = -EBUSY;
break;
}
}
KVM: PPC: Book3S HV: Protect updates to spapr_tce_tables list Al Viro pointed out that while one thread of a process is executing in kvm_vm_ioctl_create_spapr_tce(), another thread could guess the file descriptor returned by anon_inode_getfd() and close() it before the first thread has added it to the kvm->arch.spapr_tce_tables list. That highlights a more general problem: there is no mutual exclusion between writers to the spapr_tce_tables list, leading to the possibility of the list becoming corrupted, which could cause a host kernel crash. To fix the mutual exclusion problem, we add a mutex_lock/unlock pair around the list_del_rce in kvm_spapr_tce_release(). Also, this moves the call to anon_inode_getfd() inside the region protected by the kvm->lock mutex, after we have done the check for a duplicate LIOBN. This means that if another thread does guess the file descriptor and closes it, its call to kvm_spapr_tce_release() will not do any harm because it will have to wait until the first thread has released kvm->lock. With this, there are no failure points in kvm_vm_ioctl_create_spapr_tce() after the call to anon_inode_getfd(). The other things that the second thread could do with the guessed file descriptor are to mmap it or to pass it as a parameter to a KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE ioctl on a KVM device fd. An mmap call won't cause any harm because kvm_spapr_tce_mmap() and kvm_spapr_tce_fault() don't access the spapr_tce_tables list or the kvmppc_spapr_tce_table.list field, and the fields that they do use have been properly initialized by the time of the anon_inode_getfd() call. The KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE ioctl calls kvm_spapr_tce_attach_iommu_group(), which scans the spapr_tce_tables list looking for the kvmppc_spapr_tce_table struct corresponding to the fd given as the parameter. Either it will find the new entry or it won't; if it doesn't, it just returns an error, and if it does, it will function normally. So, in each case there is no harmful effect. Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-08-28 12:31:24 +08:00
if (!ret)
ret = anon_inode_getfd("kvm-spapr-tce", &kvm_spapr_tce_fops,
stt, O_RDWR | O_CLOEXEC);
if (ret >= 0) {
list_add_rcu(&stt->list, &kvm->arch.spapr_tce_tables);
kvm_get_kvm(kvm);
}
mutex_unlock(&kvm->lock);
KVM: PPC: Book3S HV: Protect updates to spapr_tce_tables list Al Viro pointed out that while one thread of a process is executing in kvm_vm_ioctl_create_spapr_tce(), another thread could guess the file descriptor returned by anon_inode_getfd() and close() it before the first thread has added it to the kvm->arch.spapr_tce_tables list. That highlights a more general problem: there is no mutual exclusion between writers to the spapr_tce_tables list, leading to the possibility of the list becoming corrupted, which could cause a host kernel crash. To fix the mutual exclusion problem, we add a mutex_lock/unlock pair around the list_del_rce in kvm_spapr_tce_release(). Also, this moves the call to anon_inode_getfd() inside the region protected by the kvm->lock mutex, after we have done the check for a duplicate LIOBN. This means that if another thread does guess the file descriptor and closes it, its call to kvm_spapr_tce_release() will not do any harm because it will have to wait until the first thread has released kvm->lock. With this, there are no failure points in kvm_vm_ioctl_create_spapr_tce() after the call to anon_inode_getfd(). The other things that the second thread could do with the guessed file descriptor are to mmap it or to pass it as a parameter to a KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE ioctl on a KVM device fd. An mmap call won't cause any harm because kvm_spapr_tce_mmap() and kvm_spapr_tce_fault() don't access the spapr_tce_tables list or the kvmppc_spapr_tce_table.list field, and the fields that they do use have been properly initialized by the time of the anon_inode_getfd() call. The KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE ioctl calls kvm_spapr_tce_attach_iommu_group(), which scans the spapr_tce_tables list looking for the kvmppc_spapr_tce_table struct corresponding to the fd given as the parameter. Either it will find the new entry or it won't; if it doesn't, it just returns an error, and if it does, it will function normally. So, in each case there is no harmful effect. Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-08-28 12:31:24 +08:00
if (ret >= 0)
return ret;
fail:
for (i = 0; i < npages; i++)
if (stt->pages[i])
__free_page(stt->pages[i]);
kfree(stt);
fail_acct:
kvmppc_account_memlimit(kvmppc_stt_pages(npages), false);
return ret;
}
KVM: PPC: Validate TCEs against preregistered memory page sizes The userspace can request an arbitrary supported page size for a DMA window and this works fine as long as the mapped memory is backed with the pages of the same or bigger size; if this is not the case, mm_iommu_ua_to_hpa{_rm}() fail and tables do not populated with dangerously incorrect TCEs. However since it is quite easy to misconfigure the KVM and we do not do reverts to all changes made to TCE tables if an error happens in a middle, we better do the acceptable page size validation before we even touch the tables. This enhances kvmppc_tce_validate() to check the hardware IOMMU page sizes against the preregistered memory page sizes. Since the new check uses real/virtual mode helpers, this renames kvmppc_tce_validate() to kvmppc_rm_tce_validate() to handle the real mode case and mirrors it for the virtual mode under the old name. The real mode handler is not used for the virtual mode as: 1. it uses _lockless() list traversing primitives instead of RCU; 2. realmode's mm_iommu_ua_to_hpa_rm() uses vmalloc_to_phys() which virtual mode does not have to use and since on POWER9+radix only virtual mode handlers actually work, we do not want to slow down that path even a bit. This removes EXPORT_SYMBOL_GPL(kvmppc_tce_validate) as the validators are static now. From now on the attempts on mapping IOMMU pages bigger than allowed will result in KVM exit. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> [mpe: Fix KVM_HV=n build] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-09-10 16:29:10 +08:00
static long kvmppc_tce_validate(struct kvmppc_spapr_tce_table *stt,
unsigned long tce)
{
unsigned long gpa = tce & ~(TCE_PCI_READ | TCE_PCI_WRITE);
enum dma_data_direction dir = iommu_tce_direction(tce);
struct kvmppc_spapr_tce_iommu_table *stit;
unsigned long ua = 0;
/* Allow userspace to poison TCE table */
if (dir == DMA_NONE)
return H_SUCCESS;
if (iommu_tce_check_gpa(stt->page_shift, gpa))
return H_TOO_HARD;
if (kvmppc_tce_to_ua(stt->kvm, tce, &ua, NULL))
KVM: PPC: Validate TCEs against preregistered memory page sizes The userspace can request an arbitrary supported page size for a DMA window and this works fine as long as the mapped memory is backed with the pages of the same or bigger size; if this is not the case, mm_iommu_ua_to_hpa{_rm}() fail and tables do not populated with dangerously incorrect TCEs. However since it is quite easy to misconfigure the KVM and we do not do reverts to all changes made to TCE tables if an error happens in a middle, we better do the acceptable page size validation before we even touch the tables. This enhances kvmppc_tce_validate() to check the hardware IOMMU page sizes against the preregistered memory page sizes. Since the new check uses real/virtual mode helpers, this renames kvmppc_tce_validate() to kvmppc_rm_tce_validate() to handle the real mode case and mirrors it for the virtual mode under the old name. The real mode handler is not used for the virtual mode as: 1. it uses _lockless() list traversing primitives instead of RCU; 2. realmode's mm_iommu_ua_to_hpa_rm() uses vmalloc_to_phys() which virtual mode does not have to use and since on POWER9+radix only virtual mode handlers actually work, we do not want to slow down that path even a bit. This removes EXPORT_SYMBOL_GPL(kvmppc_tce_validate) as the validators are static now. From now on the attempts on mapping IOMMU pages bigger than allowed will result in KVM exit. Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> [mpe: Fix KVM_HV=n build] Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-09-10 16:29:10 +08:00
return H_TOO_HARD;
list_for_each_entry_rcu(stit, &stt->iommu_tables, next) {
unsigned long hpa = 0;
struct mm_iommu_table_group_mem_t *mem;
long shift = stit->tbl->it_page_shift;
mem = mm_iommu_lookup(stt->kvm->mm, ua, 1ULL << shift);
if (!mem)
return H_TOO_HARD;
if (mm_iommu_ua_to_hpa(mem, ua, shift, &hpa))
return H_TOO_HARD;
}
return H_SUCCESS;
}
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
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;
__be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY_RO(tbl, entry);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
if (!pua)
return H_SUCCESS;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
mem = mm_iommu_lookup(kvm->mm, be64_to_cpu(*pua), pgsize);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
if (!mem)
return H_TOO_HARD;
mm_iommu_mapped_dec(mem);
*pua = cpu_to_be64(0);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
return H_SUCCESS;
}
static long kvmppc_tce_iommu_do_unmap(struct kvm *kvm,
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
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_TOO_HARD;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
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;
}
static long kvmppc_tce_iommu_unmap(struct kvm *kvm,
struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
unsigned long entry)
{
unsigned long i, ret = H_SUCCESS;
unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
unsigned long io_entry = entry * subpages;
for (i = 0; i < subpages; ++i) {
ret = kvmppc_tce_iommu_do_unmap(kvm, tbl, io_entry + i);
if (ret != H_SUCCESS)
break;
}
return ret;
}
long kvmppc_tce_iommu_do_map(struct kvm *kvm, struct iommu_table *tbl,
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
unsigned long entry, unsigned long ua,
enum dma_data_direction dir)
{
long ret;
unsigned long hpa;
__be64 *pua = IOMMU_TABLE_USERSPACE_ENTRY(tbl, entry);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
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;
KVM: PPC: Check if IOMMU page is contained in the pinned physical page A VM which has: - a DMA capable device passed through to it (eg. network card); - running a malicious kernel that ignores H_PUT_TCE failure; - capability of using IOMMU pages bigger that physical pages can create an IOMMU mapping that exposes (for example) 16MB of the host physical memory to the device when only 64K was allocated to the VM. The remaining 16MB - 64K will be some other content of host memory, possibly including pages of the VM, but also pages of host kernel memory, host programs or other VMs. The attacking VM does not control the location of the page it can map, and is only allowed to map as many pages as it has pages of RAM. We already have a check in drivers/vfio/vfio_iommu_spapr_tce.c that an IOMMU page is contained in the physical page so the PCI hardware won't get access to unassigned host memory; however this check is missing in the KVM fastpath (H_PUT_TCE accelerated code). We were lucky so far and did not hit this yet as the very first time when the mapping happens we do not have tbl::it_userspace allocated yet and fall back to the userspace which in turn calls VFIO IOMMU driver, this fails and the guest does not retry, This stores the smallest preregistered page size in the preregistered region descriptor and changes the mm_iommu_xxx API to check this against the IOMMU page size. This calculates maximum page size as a minimum of the natural region alignment and compound page size. For the page shift this uses the shift returned by find_linux_pte() which indicates how the page is mapped to the current userspace - if the page is huge and this is not a zero, then it is a leaf pte and the page is mapped within the range. Fixes: 121f80ba68f1 ("KVM: PPC: VFIO: Add in-kernel acceleration for VFIO") Cc: stable@vger.kernel.org # v4.12+ Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
2018-07-17 15:19:13 +08:00
if (WARN_ON_ONCE(mm_iommu_ua_to_hpa(mem, ua, tbl->it_page_shift, &hpa)))
return H_TOO_HARD;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
if (mm_iommu_mapped_inc(mem))
return H_TOO_HARD;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
ret = iommu_tce_xchg(tbl, entry, &hpa, &dir);
if (WARN_ON_ONCE(ret)) {
mm_iommu_mapped_dec(mem);
return H_TOO_HARD;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
}
if (dir != DMA_NONE)
kvmppc_tce_iommu_mapped_dec(kvm, tbl, entry);
*pua = cpu_to_be64(ua);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
return 0;
}
static long kvmppc_tce_iommu_map(struct kvm *kvm,
struct kvmppc_spapr_tce_table *stt, struct iommu_table *tbl,
unsigned long entry, unsigned long ua,
enum dma_data_direction dir)
{
unsigned long i, pgoff, ret = H_SUCCESS;
unsigned long subpages = 1ULL << (stt->page_shift - tbl->it_page_shift);
unsigned long io_entry = entry * subpages;
for (i = 0, pgoff = 0; i < subpages;
++i, pgoff += IOMMU_PAGE_SIZE(tbl)) {
ret = kvmppc_tce_iommu_do_map(kvm, tbl,
io_entry + i, ua + pgoff, dir);
if (ret != H_SUCCESS)
break;
}
return ret;
}
long kvmppc_h_put_tce(struct kvm_vcpu *vcpu, unsigned long liobn,
unsigned long ioba, unsigned long tce)
{
struct kvmppc_spapr_tce_table *stt;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
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); */
stt = kvmppc_find_table(vcpu->kvm, liobn);
if (!stt)
return H_TOO_HARD;
ret = kvmppc_ioba_validate(stt, ioba, 1);
if (ret != H_SUCCESS)
return ret;
ret = kvmppc_tce_validate(stt, tce);
if (ret != H_SUCCESS)
return ret;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
dir = iommu_tce_direction(tce);
KVM: PPC: Book3S: Protect kvmppc_gpa_to_ua() with SRCU kvmppc_gpa_to_ua() accesses KVM memory slot array via srcu_dereference_check() and this produces warnings from RCU like below. This extends the existing srcu_read_lock/unlock to cover that kvmppc_gpa_to_ua() as well. We did not hit this before as this lock is not needed for the realmode handlers and hash guests would use the realmode path all the time; however the radix guests are always redirected to the virtual mode handlers and hence the warning. [ 68.253798] ./include/linux/kvm_host.h:575 suspicious rcu_dereference_check() usage! [ 68.253799] other info that might help us debug this: [ 68.253802] rcu_scheduler_active = 2, debug_locks = 1 [ 68.253804] 1 lock held by qemu-system-ppc/6413: [ 68.253806] #0: (&vcpu->mutex){+.+.}, at: [<c00800000e3c22f4>] vcpu_load+0x3c/0xc0 [kvm] [ 68.253826] stack backtrace: [ 68.253830] CPU: 92 PID: 6413 Comm: qemu-system-ppc Tainted: G W 4.14.0-rc3-00553-g432dcba58e9c-dirty #72 [ 68.253833] Call Trace: [ 68.253839] [c000000fd3d9f790] [c000000000b7fcc8] dump_stack+0xe8/0x160 (unreliable) [ 68.253845] [c000000fd3d9f7d0] [c0000000001924c0] lockdep_rcu_suspicious+0x110/0x180 [ 68.253851] [c000000fd3d9f850] [c0000000000e825c] kvmppc_gpa_to_ua+0x26c/0x2b0 [ 68.253858] [c000000fd3d9f8b0] [c00800000e3e1984] kvmppc_h_put_tce+0x12c/0x2a0 [kvm] Fixes: 121f80ba68f1 ("KVM: PPC: VFIO: Add in-kernel acceleration for VFIO") Cc: stable@vger.kernel.org # v4.12+ Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-11 13:00:34 +08:00
idx = srcu_read_lock(&vcpu->kvm->srcu);
if ((dir != DMA_NONE) && kvmppc_tce_to_ua(vcpu->kvm, tce, &ua, NULL)) {
KVM: PPC: Book3S: Protect kvmppc_gpa_to_ua() with SRCU kvmppc_gpa_to_ua() accesses KVM memory slot array via srcu_dereference_check() and this produces warnings from RCU like below. This extends the existing srcu_read_lock/unlock to cover that kvmppc_gpa_to_ua() as well. We did not hit this before as this lock is not needed for the realmode handlers and hash guests would use the realmode path all the time; however the radix guests are always redirected to the virtual mode handlers and hence the warning. [ 68.253798] ./include/linux/kvm_host.h:575 suspicious rcu_dereference_check() usage! [ 68.253799] other info that might help us debug this: [ 68.253802] rcu_scheduler_active = 2, debug_locks = 1 [ 68.253804] 1 lock held by qemu-system-ppc/6413: [ 68.253806] #0: (&vcpu->mutex){+.+.}, at: [<c00800000e3c22f4>] vcpu_load+0x3c/0xc0 [kvm] [ 68.253826] stack backtrace: [ 68.253830] CPU: 92 PID: 6413 Comm: qemu-system-ppc Tainted: G W 4.14.0-rc3-00553-g432dcba58e9c-dirty #72 [ 68.253833] Call Trace: [ 68.253839] [c000000fd3d9f790] [c000000000b7fcc8] dump_stack+0xe8/0x160 (unreliable) [ 68.253845] [c000000fd3d9f7d0] [c0000000001924c0] lockdep_rcu_suspicious+0x110/0x180 [ 68.253851] [c000000fd3d9f850] [c0000000000e825c] kvmppc_gpa_to_ua+0x26c/0x2b0 [ 68.253858] [c000000fd3d9f8b0] [c00800000e3e1984] kvmppc_h_put_tce+0x12c/0x2a0 [kvm] Fixes: 121f80ba68f1 ("KVM: PPC: VFIO: Add in-kernel acceleration for VFIO") Cc: stable@vger.kernel.org # v4.12+ Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-11 13:00:34 +08:00
ret = H_PARAMETER;
goto unlock_exit;
}
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
entry = ioba >> stt->page_shift;
list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
KVM: PPC: Book3S: Protect kvmppc_gpa_to_ua() with SRCU kvmppc_gpa_to_ua() accesses KVM memory slot array via srcu_dereference_check() and this produces warnings from RCU like below. This extends the existing srcu_read_lock/unlock to cover that kvmppc_gpa_to_ua() as well. We did not hit this before as this lock is not needed for the realmode handlers and hash guests would use the realmode path all the time; however the radix guests are always redirected to the virtual mode handlers and hence the warning. [ 68.253798] ./include/linux/kvm_host.h:575 suspicious rcu_dereference_check() usage! [ 68.253799] other info that might help us debug this: [ 68.253802] rcu_scheduler_active = 2, debug_locks = 1 [ 68.253804] 1 lock held by qemu-system-ppc/6413: [ 68.253806] #0: (&vcpu->mutex){+.+.}, at: [<c00800000e3c22f4>] vcpu_load+0x3c/0xc0 [kvm] [ 68.253826] stack backtrace: [ 68.253830] CPU: 92 PID: 6413 Comm: qemu-system-ppc Tainted: G W 4.14.0-rc3-00553-g432dcba58e9c-dirty #72 [ 68.253833] Call Trace: [ 68.253839] [c000000fd3d9f790] [c000000000b7fcc8] dump_stack+0xe8/0x160 (unreliable) [ 68.253845] [c000000fd3d9f7d0] [c0000000001924c0] lockdep_rcu_suspicious+0x110/0x180 [ 68.253851] [c000000fd3d9f850] [c0000000000e825c] kvmppc_gpa_to_ua+0x26c/0x2b0 [ 68.253858] [c000000fd3d9f8b0] [c00800000e3e1984] kvmppc_h_put_tce+0x12c/0x2a0 [kvm] Fixes: 121f80ba68f1 ("KVM: PPC: VFIO: Add in-kernel acceleration for VFIO") Cc: stable@vger.kernel.org # v4.12+ Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-11 13:00:34 +08:00
if (dir == DMA_NONE)
ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
stit->tbl, entry);
KVM: PPC: Book3S: Protect kvmppc_gpa_to_ua() with SRCU kvmppc_gpa_to_ua() accesses KVM memory slot array via srcu_dereference_check() and this produces warnings from RCU like below. This extends the existing srcu_read_lock/unlock to cover that kvmppc_gpa_to_ua() as well. We did not hit this before as this lock is not needed for the realmode handlers and hash guests would use the realmode path all the time; however the radix guests are always redirected to the virtual mode handlers and hence the warning. [ 68.253798] ./include/linux/kvm_host.h:575 suspicious rcu_dereference_check() usage! [ 68.253799] other info that might help us debug this: [ 68.253802] rcu_scheduler_active = 2, debug_locks = 1 [ 68.253804] 1 lock held by qemu-system-ppc/6413: [ 68.253806] #0: (&vcpu->mutex){+.+.}, at: [<c00800000e3c22f4>] vcpu_load+0x3c/0xc0 [kvm] [ 68.253826] stack backtrace: [ 68.253830] CPU: 92 PID: 6413 Comm: qemu-system-ppc Tainted: G W 4.14.0-rc3-00553-g432dcba58e9c-dirty #72 [ 68.253833] Call Trace: [ 68.253839] [c000000fd3d9f790] [c000000000b7fcc8] dump_stack+0xe8/0x160 (unreliable) [ 68.253845] [c000000fd3d9f7d0] [c0000000001924c0] lockdep_rcu_suspicious+0x110/0x180 [ 68.253851] [c000000fd3d9f850] [c0000000000e825c] kvmppc_gpa_to_ua+0x26c/0x2b0 [ 68.253858] [c000000fd3d9f8b0] [c00800000e3e1984] kvmppc_h_put_tce+0x12c/0x2a0 [kvm] Fixes: 121f80ba68f1 ("KVM: PPC: VFIO: Add in-kernel acceleration for VFIO") Cc: stable@vger.kernel.org # v4.12+ Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-11 13:00:34 +08:00
else
ret = kvmppc_tce_iommu_map(vcpu->kvm, stt, stit->tbl,
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
entry, ua, dir);
if (ret != H_SUCCESS) {
kvmppc_clear_tce(stit->tbl, entry);
KVM: PPC: Book3S: Protect kvmppc_gpa_to_ua() with SRCU kvmppc_gpa_to_ua() accesses KVM memory slot array via srcu_dereference_check() and this produces warnings from RCU like below. This extends the existing srcu_read_lock/unlock to cover that kvmppc_gpa_to_ua() as well. We did not hit this before as this lock is not needed for the realmode handlers and hash guests would use the realmode path all the time; however the radix guests are always redirected to the virtual mode handlers and hence the warning. [ 68.253798] ./include/linux/kvm_host.h:575 suspicious rcu_dereference_check() usage! [ 68.253799] other info that might help us debug this: [ 68.253802] rcu_scheduler_active = 2, debug_locks = 1 [ 68.253804] 1 lock held by qemu-system-ppc/6413: [ 68.253806] #0: (&vcpu->mutex){+.+.}, at: [<c00800000e3c22f4>] vcpu_load+0x3c/0xc0 [kvm] [ 68.253826] stack backtrace: [ 68.253830] CPU: 92 PID: 6413 Comm: qemu-system-ppc Tainted: G W 4.14.0-rc3-00553-g432dcba58e9c-dirty #72 [ 68.253833] Call Trace: [ 68.253839] [c000000fd3d9f790] [c000000000b7fcc8] dump_stack+0xe8/0x160 (unreliable) [ 68.253845] [c000000fd3d9f7d0] [c0000000001924c0] lockdep_rcu_suspicious+0x110/0x180 [ 68.253851] [c000000fd3d9f850] [c0000000000e825c] kvmppc_gpa_to_ua+0x26c/0x2b0 [ 68.253858] [c000000fd3d9f8b0] [c00800000e3e1984] kvmppc_h_put_tce+0x12c/0x2a0 [kvm] Fixes: 121f80ba68f1 ("KVM: PPC: VFIO: Add in-kernel acceleration for VFIO") Cc: stable@vger.kernel.org # v4.12+ Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-11 13:00:34 +08:00
goto unlock_exit;
}
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
}
kvmppc_tce_put(stt, entry, tce);
KVM: PPC: Book3S: Protect kvmppc_gpa_to_ua() with SRCU kvmppc_gpa_to_ua() accesses KVM memory slot array via srcu_dereference_check() and this produces warnings from RCU like below. This extends the existing srcu_read_lock/unlock to cover that kvmppc_gpa_to_ua() as well. We did not hit this before as this lock is not needed for the realmode handlers and hash guests would use the realmode path all the time; however the radix guests are always redirected to the virtual mode handlers and hence the warning. [ 68.253798] ./include/linux/kvm_host.h:575 suspicious rcu_dereference_check() usage! [ 68.253799] other info that might help us debug this: [ 68.253802] rcu_scheduler_active = 2, debug_locks = 1 [ 68.253804] 1 lock held by qemu-system-ppc/6413: [ 68.253806] #0: (&vcpu->mutex){+.+.}, at: [<c00800000e3c22f4>] vcpu_load+0x3c/0xc0 [kvm] [ 68.253826] stack backtrace: [ 68.253830] CPU: 92 PID: 6413 Comm: qemu-system-ppc Tainted: G W 4.14.0-rc3-00553-g432dcba58e9c-dirty #72 [ 68.253833] Call Trace: [ 68.253839] [c000000fd3d9f790] [c000000000b7fcc8] dump_stack+0xe8/0x160 (unreliable) [ 68.253845] [c000000fd3d9f7d0] [c0000000001924c0] lockdep_rcu_suspicious+0x110/0x180 [ 68.253851] [c000000fd3d9f850] [c0000000000e825c] kvmppc_gpa_to_ua+0x26c/0x2b0 [ 68.253858] [c000000fd3d9f8b0] [c00800000e3e1984] kvmppc_h_put_tce+0x12c/0x2a0 [kvm] Fixes: 121f80ba68f1 ("KVM: PPC: VFIO: Add in-kernel acceleration for VFIO") Cc: stable@vger.kernel.org # v4.12+ Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-10-11 13:00:34 +08:00
unlock_exit:
srcu_read_unlock(&vcpu->kvm->srcu, idx);
return ret;
}
EXPORT_SYMBOL_GPL(kvmppc_h_put_tce);
long kvmppc_h_put_tce_indirect(struct kvm_vcpu *vcpu,
unsigned long liobn, unsigned long ioba,
unsigned long tce_list, unsigned long npages)
{
struct kvmppc_spapr_tce_table *stt;
long i, ret = H_SUCCESS, idx;
unsigned long entry, ua = 0;
u64 __user *tces;
u64 tce;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
struct kvmppc_spapr_tce_iommu_table *stit;
stt = kvmppc_find_table(vcpu->kvm, liobn);
if (!stt)
return H_TOO_HARD;
entry = ioba >> stt->page_shift;
/*
* SPAPR spec says that the maximum size of the list is 512 TCEs
* so the whole table fits in 4K page
*/
if (npages > 512)
return H_PARAMETER;
if (tce_list & (SZ_4K - 1))
return H_PARAMETER;
ret = kvmppc_ioba_validate(stt, ioba, npages);
if (ret != H_SUCCESS)
return ret;
idx = srcu_read_lock(&vcpu->kvm->srcu);
if (kvmppc_tce_to_ua(vcpu->kvm, tce_list, &ua, NULL)) {
ret = H_TOO_HARD;
goto unlock_exit;
}
tces = (u64 __user *) ua;
for (i = 0; i < npages; ++i) {
if (get_user(tce, tces + i)) {
ret = H_TOO_HARD;
goto unlock_exit;
}
tce = be64_to_cpu(tce);
ret = kvmppc_tce_validate(stt, tce);
if (ret != H_SUCCESS)
goto unlock_exit;
}
for (i = 0; i < npages; ++i) {
/*
* This looks unsafe, because we validate, then regrab
* the TCE from userspace which could have been changed by
* another thread.
*
* But it actually is safe, because the relevant checks will be
* re-executed in the following code. If userspace tries to
* change this dodgily it will result in a messier failure mode
* but won't threaten the host.
*/
if (get_user(tce, tces + i)) {
ret = H_TOO_HARD;
goto unlock_exit;
}
tce = be64_to_cpu(tce);
if (kvmppc_tce_to_ua(vcpu->kvm, tce, &ua, NULL))
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
return H_PARAMETER;
list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
ret = kvmppc_tce_iommu_map(vcpu->kvm, stt,
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
stit->tbl, entry + i, ua,
iommu_tce_direction(tce));
if (ret != H_SUCCESS) {
kvmppc_clear_tce(stit->tbl, entry);
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
goto unlock_exit;
}
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
}
kvmppc_tce_put(stt, entry + i, tce);
}
unlock_exit:
srcu_read_unlock(&vcpu->kvm->srcu, idx);
return ret;
}
EXPORT_SYMBOL_GPL(kvmppc_h_put_tce_indirect);
long kvmppc_h_stuff_tce(struct kvm_vcpu *vcpu,
unsigned long liobn, unsigned long ioba,
unsigned long tce_value, unsigned long npages)
{
struct kvmppc_spapr_tce_table *stt;
long i, ret;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
struct kvmppc_spapr_tce_iommu_table *stit;
stt = kvmppc_find_table(vcpu->kvm, liobn);
if (!stt)
return H_TOO_HARD;
ret = kvmppc_ioba_validate(stt, ioba, npages);
if (ret != H_SUCCESS)
return ret;
/* Check permission bits only to allow userspace poison TCE for debug */
if (tce_value & (TCE_PCI_WRITE | TCE_PCI_READ))
return H_PARAMETER;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
list_for_each_entry_lockless(stit, &stt->iommu_tables, next) {
unsigned long entry = ioba >> stt->page_shift;
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
for (i = 0; i < npages; ++i) {
ret = kvmppc_tce_iommu_unmap(vcpu->kvm, stt,
KVM: PPC: VFIO: Add in-kernel acceleration for VFIO This allows the host kernel to handle H_PUT_TCE, H_PUT_TCE_INDIRECT and H_STUFF_TCE requests targeted an IOMMU TCE table used for VFIO without passing them to user space which saves time on switching to user space and back. This adds H_PUT_TCE/H_PUT_TCE_INDIRECT/H_STUFF_TCE handlers to KVM. KVM tries to handle a TCE request in the real mode, if failed it passes the request to the virtual mode to complete the operation. If it a virtual mode handler fails, the request is passed to the user space; this is not expected to happen though. To avoid dealing with page use counters (which is tricky in real mode), this only accelerates SPAPR TCE IOMMU v2 clients which are required to pre-register the userspace memory. The very first TCE request will be handled in the VFIO SPAPR TCE driver anyway as the userspace view of the TCE table (iommu_table::it_userspace) is not allocated till the very first mapping happens and we cannot call vmalloc in real mode. If we fail to update a hardware IOMMU table unexpected reason, we just clear it and move on as there is nothing really we can do about it - for example, if we hot plug a VFIO device to a guest, existing TCE tables will be mirrored automatically to the hardware and there is no interface to report to the guest about possible failures. This adds new attribute - KVM_DEV_VFIO_GROUP_SET_SPAPR_TCE - to the VFIO KVM device. It takes a VFIO group fd and SPAPR TCE table fd and associates a physical IOMMU table with the SPAPR TCE table (which is a guest view of the hardware IOMMU table). The iommu_table object is cached and referenced so we do not have to look up for it in real mode. This does not implement the UNSET counterpart as there is no use for it - once the acceleration is enabled, the existing userspace won't disable it unless a VFIO container is destroyed; this adds necessary cleanup to the KVM_DEV_VFIO_GROUP_DEL handler. This advertises the new KVM_CAP_SPAPR_TCE_VFIO capability to the user space. This adds real mode version of WARN_ON_ONCE() as the generic version causes problems with rcu_sched. Since we testing what vmalloc_to_phys() returns in the code, this also adds a check for already existing vmalloc_to_phys() call in kvmppc_rm_h_put_tce_indirect(). This finally makes use of vfio_external_user_iommu_id() which was introduced quite some time ago and was considered for removal. Tests show that this patch increases transmission speed from 220MB/s to 750..1020MB/s on 10Gb network (Chelsea CXGB3 10Gb ethernet card). Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru> Acked-by: Alex Williamson <alex.williamson@redhat.com> Reviewed-by: David Gibson <david@gibson.dropbear.id.au> Signed-off-by: Paul Mackerras <paulus@ozlabs.org>
2017-03-22 12:21:56 +08:00
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);
return H_SUCCESS;
}
EXPORT_SYMBOL_GPL(kvmppc_h_stuff_tce);