linux/drivers/iommu/intel-svm.c
David Woodhouse e034992160 iommu/vt-d: Fix IOTLB flushing for global pages
When flushing kernel-mode PASIDs, we need to flush global pages too.

Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
2015-10-16 19:37:04 +01:00

595 lines
16 KiB
C

/*
* Copyright © 2015 Intel Corporation.
*
* This program is free software; you can redistribute it and/or modify it
* under the terms and conditions of the GNU General Public License,
* version 2, as published by the Free Software Foundation.
*
* This program is distributed in the hope 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.
*
* Authors: David Woodhouse <dwmw2@infradead.org>
*/
#include <linux/intel-iommu.h>
#include <linux/mmu_notifier.h>
#include <linux/sched.h>
#include <linux/slab.h>
#include <linux/intel-svm.h>
#include <linux/rculist.h>
#include <linux/pci.h>
#include <linux/pci-ats.h>
#include <linux/dmar.h>
#include <linux/interrupt.h>
static irqreturn_t prq_event_thread(int irq, void *d);
struct pasid_entry {
u64 val;
};
struct pasid_state_entry {
u64 val;
};
int intel_svm_alloc_pasid_tables(struct intel_iommu *iommu)
{
struct page *pages;
int order;
order = ecap_pss(iommu->ecap) + 7 - PAGE_SHIFT;
if (order < 0)
order = 0;
pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, order);
if (!pages) {
pr_warn("IOMMU: %s: Failed to allocate PASID table\n",
iommu->name);
return -ENOMEM;
}
iommu->pasid_table = page_address(pages);
pr_info("%s: Allocated order %d PASID table.\n", iommu->name, order);
if (ecap_dis(iommu->ecap)) {
pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, order);
if (pages)
iommu->pasid_state_table = page_address(pages);
else
pr_warn("IOMMU: %s: Failed to allocate PASID state table\n",
iommu->name);
}
idr_init(&iommu->pasid_idr);
return 0;
}
int intel_svm_free_pasid_tables(struct intel_iommu *iommu)
{
int order;
order = ecap_pss(iommu->ecap) + 7 - PAGE_SHIFT;
if (order < 0)
order = 0;
if (iommu->pasid_table) {
free_pages((unsigned long)iommu->pasid_table, order);
iommu->pasid_table = NULL;
}
if (iommu->pasid_state_table) {
free_pages((unsigned long)iommu->pasid_state_table, order);
iommu->pasid_state_table = NULL;
}
idr_destroy(&iommu->pasid_idr);
return 0;
}
#define PRQ_ORDER 0
int intel_svm_enable_prq(struct intel_iommu *iommu)
{
struct page *pages;
int irq, ret;
pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, PRQ_ORDER);
if (!pages) {
pr_warn("IOMMU: %s: Failed to allocate page request queue\n",
iommu->name);
return -ENOMEM;
}
iommu->prq = page_address(pages);
irq = dmar_alloc_hwirq(DMAR_UNITS_SUPPORTED + iommu->seq_id, iommu->node, iommu);
if (irq <= 0) {
pr_err("IOMMU: %s: Failed to create IRQ vector for page request queue\n",
iommu->name);
ret = -EINVAL;
err:
free_pages((unsigned long)iommu->prq, PRQ_ORDER);
iommu->prq = NULL;
return ret;
}
iommu->pr_irq = irq;
snprintf(iommu->prq_name, sizeof(iommu->prq_name), "dmar%d-prq", iommu->seq_id);
ret = request_threaded_irq(irq, NULL, prq_event_thread, IRQF_ONESHOT,
iommu->prq_name, iommu);
if (ret) {
pr_err("IOMMU: %s: Failed to request IRQ for page request queue\n",
iommu->name);
dmar_free_hwirq(irq);
goto err;
}
dmar_writeq(iommu->reg + DMAR_PQH_REG, 0ULL);
dmar_writeq(iommu->reg + DMAR_PQT_REG, 0ULL);
dmar_writeq(iommu->reg + DMAR_PQA_REG, virt_to_phys(iommu->prq) | PRQ_ORDER);
return 0;
}
int intel_svm_finish_prq(struct intel_iommu *iommu)
{
dmar_writeq(iommu->reg + DMAR_PQH_REG, 0ULL);
dmar_writeq(iommu->reg + DMAR_PQT_REG, 0ULL);
dmar_writeq(iommu->reg + DMAR_PQA_REG, 0ULL);
free_irq(iommu->pr_irq, iommu);
dmar_free_hwirq(iommu->pr_irq);
iommu->pr_irq = 0;
free_pages((unsigned long)iommu->prq, PRQ_ORDER);
iommu->prq = NULL;
return 0;
}
static void intel_flush_svm_range_dev (struct intel_svm *svm, struct intel_svm_dev *sdev,
unsigned long address, int pages, int ih, int gl)
{
struct qi_desc desc;
int mask = ilog2(__roundup_pow_of_two(pages));
if (pages == -1 || !cap_pgsel_inv(svm->iommu->cap) ||
mask > cap_max_amask_val(svm->iommu->cap)) {
/* For global kernel pages we have to flush them in *all* PASIDs
* because that's the only option the hardware gives us. Despite
* the fact that they are actually only accessible through one. */
if (gl)
desc.low = QI_EIOTLB_PASID(svm->pasid) | QI_EIOTLB_DID(sdev->did) |
QI_EIOTLB_GRAN(QI_GRAN_ALL_ALL) | QI_EIOTLB_TYPE;
else
desc.low = QI_EIOTLB_PASID(svm->pasid) | QI_EIOTLB_DID(sdev->did) |
QI_EIOTLB_GRAN(QI_GRAN_NONG_PASID) | QI_EIOTLB_TYPE;
desc.high = 0;
} else {
desc.low = QI_EIOTLB_PASID(svm->pasid) | QI_EIOTLB_DID(sdev->did) |
QI_EIOTLB_GRAN(QI_GRAN_PSI_PASID) | QI_EIOTLB_TYPE;
desc.high = QI_EIOTLB_ADDR(address) | QI_EIOTLB_GL(gl) |
QI_EIOTLB_IH(ih) | QI_EIOTLB_AM(mask);
}
qi_submit_sync(&desc, svm->iommu);
if (sdev->dev_iotlb) {
desc.low = QI_DEV_EIOTLB_PASID(svm->pasid) | QI_DEV_EIOTLB_SID(sdev->sid) |
QI_DEV_EIOTLB_QDEP(sdev->qdep) | QI_DEIOTLB_TYPE;
if (mask) {
unsigned long adr, delta;
/* Least significant zero bits in the address indicate the
* range of the request. So mask them out according to the
* size. */
adr = address & ((1<<(VTD_PAGE_SHIFT + mask)) - 1);
/* Now ensure that we round down further if the original
* request was not aligned w.r.t. its size */
delta = address - adr;
if (delta + (pages << VTD_PAGE_SHIFT) >= (1 << (VTD_PAGE_SHIFT + mask)))
adr &= ~(1 << (VTD_PAGE_SHIFT + mask));
desc.high = QI_DEV_EIOTLB_ADDR(adr) | QI_DEV_EIOTLB_SIZE;
} else {
desc.high = QI_DEV_EIOTLB_ADDR(address);
}
qi_submit_sync(&desc, svm->iommu);
}
}
static void intel_flush_svm_range(struct intel_svm *svm, unsigned long address,
int pages, int ih, int gl)
{
struct intel_svm_dev *sdev;
/* Try deferred invalidate if available */
if (svm->iommu->pasid_state_table &&
!cmpxchg64(&svm->iommu->pasid_state_table[svm->pasid].val, 0, 1ULL << 63))
return;
rcu_read_lock();
list_for_each_entry_rcu(sdev, &svm->devs, list)
intel_flush_svm_range_dev(svm, sdev, address, pages, ih, gl);
rcu_read_unlock();
}
static void intel_change_pte(struct mmu_notifier *mn, struct mm_struct *mm,
unsigned long address, pte_t pte)
{
struct intel_svm *svm = container_of(mn, struct intel_svm, notifier);
intel_flush_svm_range(svm, address, 1, 1, 0);
}
static void intel_invalidate_page(struct mmu_notifier *mn, struct mm_struct *mm,
unsigned long address)
{
struct intel_svm *svm = container_of(mn, struct intel_svm, notifier);
intel_flush_svm_range(svm, address, 1, 1, 0);
}
/* Pages have been freed at this point */
static void intel_invalidate_range(struct mmu_notifier *mn,
struct mm_struct *mm,
unsigned long start, unsigned long end)
{
struct intel_svm *svm = container_of(mn, struct intel_svm, notifier);
intel_flush_svm_range(svm, start,
(end - start + PAGE_SIZE - 1) >> VTD_PAGE_SHIFT, 0, 0);
}
static void intel_flush_pasid_dev(struct intel_svm *svm, struct intel_svm_dev *sdev)
{
struct qi_desc desc;
desc.high = 0;
desc.low = QI_PC_TYPE | QI_PC_DID(sdev->did) | QI_PC_PASID_SEL | QI_PC_PASID(svm->pasid);
qi_submit_sync(&desc, svm->iommu);
}
static void intel_mm_release(struct mmu_notifier *mn, struct mm_struct *mm)
{
struct intel_svm *svm = container_of(mn, struct intel_svm, notifier);
svm->iommu->pasid_table[svm->pasid].val = 0;
/* There's no need to do any flush because we can't get here if there
* are any devices left anyway. */
WARN_ON(!list_empty(&svm->devs));
}
static const struct mmu_notifier_ops intel_mmuops = {
.release = intel_mm_release,
.change_pte = intel_change_pte,
.invalidate_page = intel_invalidate_page,
.invalidate_range = intel_invalidate_range,
};
static DEFINE_MUTEX(pasid_mutex);
int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_ops *ops)
{
struct intel_iommu *iommu = intel_svm_device_to_iommu(dev);
struct intel_svm_dev *sdev;
struct intel_svm *svm = NULL;
struct mm_struct *mm = NULL;
int pasid_max;
int ret;
if (WARN_ON(!iommu))
return -EINVAL;
if (dev_is_pci(dev)) {
pasid_max = pci_max_pasids(to_pci_dev(dev));
if (pasid_max < 0)
return -EINVAL;
} else
pasid_max = 1 << 20;
if ((flags & SVM_FLAG_SUPERVISOR_MODE)) {
if (!ecap_srs(iommu->ecap))
return -EINVAL;
} else if (pasid) {
mm = get_task_mm(current);
BUG_ON(!mm);
}
mutex_lock(&pasid_mutex);
if (pasid && !(flags & SVM_FLAG_PRIVATE_PASID)) {
int i;
idr_for_each_entry(&iommu->pasid_idr, svm, i) {
if (svm->mm != mm ||
(svm->flags & SVM_FLAG_PRIVATE_PASID))
continue;
if (svm->pasid >= pasid_max) {
dev_warn(dev,
"Limited PASID width. Cannot use existing PASID %d\n",
svm->pasid);
ret = -ENOSPC;
goto out;
}
list_for_each_entry(sdev, &svm->devs, list) {
if (dev == sdev->dev) {
if (sdev->ops != ops) {
ret = -EBUSY;
goto out;
}
sdev->users++;
goto success;
}
}
break;
}
}
sdev = kzalloc(sizeof(*sdev), GFP_KERNEL);
if (!sdev) {
ret = -ENOMEM;
goto out;
}
sdev->dev = dev;
ret = intel_iommu_enable_pasid(iommu, sdev);
if (ret || !pasid) {
/* If they don't actually want to assign a PASID, this is
* just an enabling check/preparation. */
kfree(sdev);
goto out;
}
/* Finish the setup now we know we're keeping it */
sdev->users = 1;
sdev->ops = ops;
init_rcu_head(&sdev->rcu);
if (!svm) {
svm = kzalloc(sizeof(*svm), GFP_KERNEL);
if (!svm) {
ret = -ENOMEM;
kfree(sdev);
goto out;
}
svm->iommu = iommu;
if (pasid_max > 2 << ecap_pss(iommu->ecap))
pasid_max = 2 << ecap_pss(iommu->ecap);
ret = idr_alloc(&iommu->pasid_idr, svm, 0, pasid_max - 1,
GFP_KERNEL);
if (ret < 0) {
kfree(svm);
goto out;
}
svm->pasid = ret;
svm->notifier.ops = &intel_mmuops;
svm->mm = mm;
svm->flags = flags;
INIT_LIST_HEAD_RCU(&svm->devs);
ret = -ENOMEM;
if (mm) {
ret = mmu_notifier_register(&svm->notifier, mm);
if (ret) {
idr_remove(&svm->iommu->pasid_idr, svm->pasid);
kfree(svm);
kfree(sdev);
goto out;
}
iommu->pasid_table[svm->pasid].val = (u64)__pa(mm->pgd) | 1;
mm = NULL;
} else
iommu->pasid_table[svm->pasid].val = (u64)__pa(init_mm.pgd) | 1 | (1ULL << 11);
wmb();
}
list_add_rcu(&sdev->list, &svm->devs);
success:
*pasid = svm->pasid;
ret = 0;
out:
mutex_unlock(&pasid_mutex);
if (mm)
mmput(mm);
return ret;
}
EXPORT_SYMBOL_GPL(intel_svm_bind_mm);
int intel_svm_unbind_mm(struct device *dev, int pasid)
{
struct intel_svm_dev *sdev;
struct intel_iommu *iommu;
struct intel_svm *svm;
int ret = -EINVAL;
mutex_lock(&pasid_mutex);
iommu = intel_svm_device_to_iommu(dev);
if (!iommu || !iommu->pasid_table)
goto out;
svm = idr_find(&iommu->pasid_idr, pasid);
if (!svm)
goto out;
list_for_each_entry(sdev, &svm->devs, list) {
if (dev == sdev->dev) {
ret = 0;
sdev->users--;
if (!sdev->users) {
list_del_rcu(&sdev->list);
/* Flush the PASID cache and IOTLB for this device.
* Note that we do depend on the hardware *not* using
* the PASID any more. Just as we depend on other
* devices never using PASIDs that they have no right
* to use. We have a *shared* PASID table, because it's
* large and has to be physically contiguous. So it's
* hard to be as defensive as we might like. */
intel_flush_pasid_dev(svm, sdev);
intel_flush_svm_range_dev(svm, sdev, 0, -1, 0, !svm->mm);
kfree_rcu(sdev, rcu);
if (list_empty(&svm->devs)) {
mmu_notifier_unregister(&svm->notifier, svm->mm);
idr_remove(&svm->iommu->pasid_idr, svm->pasid);
if (svm->mm)
mmput(svm->mm);
/* We mandate that no page faults may be outstanding
* for the PASID when intel_svm_unbind_mm() is called.
* If that is not obeyed, subtle errors will happen.
* Let's make them less subtle... */
memset(svm, 0x6b, sizeof(*svm));
kfree(svm);
}
}
break;
}
}
out:
mutex_unlock(&pasid_mutex);
return ret;
}
EXPORT_SYMBOL_GPL(intel_svm_unbind_mm);
/* Page request queue descriptor */
struct page_req_dsc {
u64 srr:1;
u64 bof:1;
u64 pasid_present:1;
u64 lpig:1;
u64 pasid:20;
u64 bus:8;
u64 private:23;
u64 prg_index:9;
u64 rd_req:1;
u64 wr_req:1;
u64 exe_req:1;
u64 priv_req:1;
u64 devfn:8;
u64 addr:52;
};
#define PRQ_RING_MASK ((0x1000 << PRQ_ORDER) - 0x10)
static irqreturn_t prq_event_thread(int irq, void *d)
{
struct intel_iommu *iommu = d;
struct intel_svm *svm = NULL;
int head, tail, handled = 0;
tail = dmar_readq(iommu->reg + DMAR_PQT_REG) & PRQ_RING_MASK;
head = dmar_readq(iommu->reg + DMAR_PQH_REG) & PRQ_RING_MASK;
while (head != tail) {
struct intel_svm_dev *sdev;
struct vm_area_struct *vma;
struct page_req_dsc *req;
struct qi_desc resp;
int ret, result;
u64 address;
handled = 1;
req = &iommu->prq[head / sizeof(*req)];
result = QI_RESP_FAILURE;
address = (u64)req->addr << VTD_PAGE_SHIFT;
if (!req->pasid_present) {
pr_err("%s: Page request without PASID: %08llx %08llx\n",
iommu->name, ((unsigned long long *)req)[0],
((unsigned long long *)req)[1]);
goto bad_req;
}
if (!svm || svm->pasid != req->pasid) {
rcu_read_lock();
svm = idr_find(&iommu->pasid_idr, req->pasid);
/* It *can't* go away, because the driver is not permitted
* to unbind the mm while any page faults are outstanding.
* So we only need RCU to protect the internal idr code. */
rcu_read_unlock();
if (!svm) {
pr_err("%s: Page request for invalid PASID %d: %08llx %08llx\n",
iommu->name, req->pasid, ((unsigned long long *)req)[0],
((unsigned long long *)req)[1]);
goto no_pasid;
}
}
result = QI_RESP_INVALID;
/* Since we're using init_mm.pgd directly, we should never take
* any faults on kernel addresses. */
if (!svm->mm)
goto bad_req;
down_read(&svm->mm->mmap_sem);
vma = find_extend_vma(svm->mm, address);
if (!vma || address < vma->vm_start)
goto invalid;
ret = handle_mm_fault(svm->mm, vma, address,
req->wr_req ? FAULT_FLAG_WRITE : 0);
if (ret & VM_FAULT_ERROR)
goto invalid;
result = QI_RESP_SUCCESS;
invalid:
up_read(&svm->mm->mmap_sem);
bad_req:
/* Accounting for major/minor faults? */
rcu_read_lock();
list_for_each_entry_rcu(sdev, &svm->devs, list) {
if (sdev->sid == PCI_DEVID(req->bus, req->devfn));
break;
}
/* Other devices can go away, but the drivers are not permitted
* to unbind while any page faults might be in flight. So it's
* OK to drop the 'lock' here now we have it. */
rcu_read_unlock();
if (WARN_ON(&sdev->list == &svm->devs))
sdev = NULL;
if (sdev && sdev->ops && sdev->ops->fault_cb) {
int rwxp = (req->rd_req << 3) | (req->wr_req << 2) |
(req->wr_req << 1) | (req->exe_req);
sdev->ops->fault_cb(sdev->dev, req->pasid, req->addr, req->private, rwxp, result);
}
/* We get here in the error case where the PASID lookup failed,
and these can be NULL. Do not use them below this point! */
sdev = NULL;
svm = NULL;
no_pasid:
if (req->lpig) {
/* Page Group Response */
resp.low = QI_PGRP_PASID(req->pasid) |
QI_PGRP_DID((req->bus << 8) | req->devfn) |
QI_PGRP_PASID_P(req->pasid_present) |
QI_PGRP_RESP_TYPE;
resp.high = QI_PGRP_IDX(req->prg_index) |
QI_PGRP_PRIV(req->private) | QI_PGRP_RESP_CODE(result);
qi_submit_sync(&resp, iommu);
} else if (req->srr) {
/* Page Stream Response */
resp.low = QI_PSTRM_IDX(req->prg_index) |
QI_PSTRM_PRIV(req->private) | QI_PSTRM_BUS(req->bus) |
QI_PSTRM_PASID(req->pasid) | QI_PSTRM_RESP_TYPE;
resp.high = QI_PSTRM_ADDR(address) | QI_PSTRM_DEVFN(req->devfn) |
QI_PSTRM_RESP_CODE(result);
qi_submit_sync(&resp, iommu);
}
head = (head + sizeof(*req)) & PRQ_RING_MASK;
}
dmar_writeq(iommu->reg + DMAR_PQH_REG, tail);
return IRQ_RETVAL(handled);
}