iommu/amd: Implement notifiers for IOMMUv2
Joerg Roedel [Thu, 24 Nov 2011 15:21:52 +0000 (16:21 +0100)]
Since pages are not pinned anymore we need notifications
when the VMM changes the page-tables. Use mmu_notifiers for
that.
Also use the task_exit notifier from the profiling subsystem
to shutdown all contexts related to this task.

Signed-off-by: Joerg Roedel <joerg.roedel@amd.com>

drivers/iommu/Kconfig
drivers/iommu/amd_iommu_v2.c

index 2dcf984..61038cf 100644 (file)
@@ -62,7 +62,8 @@ config AMD_IOMMU_STATS
 
 config AMD_IOMMU_V2
        tristate "AMD IOMMU Version 2 driver (EXPERIMENTAL)"
-       depends on AMD_IOMMU && EXPERIMENTAL
+       depends on AMD_IOMMU && PROFILING && EXPERIMENTAL
+       select MMU_NOTIFIER
        ---help---
          This option enables support for the AMD IOMMUv2 features of the IOMMU
          hardware. Select this option if you want to use devices that support
index 8804b22..abdb839 100644 (file)
  * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307 USA
  */
 
+#include <linux/mmu_notifier.h>
 #include <linux/amd-iommu.h>
 #include <linux/mm_types.h>
+#include <linux/profile.h>
 #include <linux/module.h>
 #include <linux/sched.h>
 #include <linux/iommu.h>
@@ -45,6 +47,7 @@ struct pasid_state {
        atomic_t count;                         /* Reference count */
        struct task_struct *task;               /* Task bound to this PASID */
        struct mm_struct *mm;                   /* mm_struct for the faults */
+       struct mmu_notifier mn;                 /* mmu_otifier handle */
        struct pri_queue pri[PRI_QUEUE_SIZE];   /* PRI tag states */
        struct device_state *device_state;      /* Link to our device_state */
        int pasid;                              /* PASID index */
@@ -85,8 +88,16 @@ static DEFINE_SPINLOCK(ps_lock);
 
 static struct workqueue_struct *iommu_wq;
 
+/*
+ * Empty page table - Used between
+ * mmu_notifier_invalidate_range_start and
+ * mmu_notifier_invalidate_range_end
+ */
+static u64 *empty_page_table;
+
 static void free_pasid_states(struct device_state *dev_state);
 static void unbind_pasid(struct device_state *dev_state, int pasid);
+static int task_exit(struct notifier_block *nb, unsigned long e, void *data);
 
 static u16 device_id(struct pci_dev *pdev)
 {
@@ -144,6 +155,11 @@ static void put_device_state_wait(struct device_state *dev_state)
 
        free_device_state(dev_state);
 }
+
+static struct notifier_block profile_nb = {
+       .notifier_call = task_exit,
+};
+
 static void link_pasid_state(struct pasid_state *pasid_state)
 {
        spin_lock(&ps_lock);
@@ -294,6 +310,23 @@ static void put_pasid_state_wait(struct pasid_state *pasid_state)
        free_pasid_state(pasid_state);
 }
 
+static void __unbind_pasid(struct pasid_state *pasid_state)
+{
+       struct iommu_domain *domain;
+
+       domain = pasid_state->device_state->domain;
+
+       amd_iommu_domain_clear_gcr3(domain, pasid_state->pasid);
+       clear_pasid_state(pasid_state->device_state, pasid_state->pasid);
+
+       /* Make sure no more pending faults are in the queue */
+       flush_workqueue(iommu_wq);
+
+       mmu_notifier_unregister(&pasid_state->mn, pasid_state->mm);
+
+       put_pasid_state(pasid_state); /* Reference taken in bind() function */
+}
+
 static void unbind_pasid(struct device_state *dev_state, int pasid)
 {
        struct pasid_state *pasid_state;
@@ -303,12 +336,8 @@ static void unbind_pasid(struct device_state *dev_state, int pasid)
                return;
 
        unlink_pasid_state(pasid_state);
-
-       amd_iommu_domain_clear_gcr3(dev_state->domain, pasid);
-       clear_pasid_state(dev_state, pasid);
-
-       put_pasid_state(pasid_state); /* Reference taken in this function */
-       put_pasid_state_wait(pasid_state); /* Reference from bind() function */
+       __unbind_pasid(pasid_state);
+       put_pasid_state_wait(pasid_state); /* Reference taken in this function */
 }
 
 static void free_pasid_states_level1(struct pasid_state **tbl)
@@ -361,6 +390,83 @@ static void free_pasid_states(struct device_state *dev_state)
        free_page((unsigned long)dev_state->states);
 }
 
+static struct pasid_state *mn_to_state(struct mmu_notifier *mn)
+{
+       return container_of(mn, struct pasid_state, mn);
+}
+
+static void __mn_flush_page(struct mmu_notifier *mn,
+                           unsigned long address)
+{
+       struct pasid_state *pasid_state;
+       struct device_state *dev_state;
+
+       pasid_state = mn_to_state(mn);
+       dev_state   = pasid_state->device_state;
+
+       amd_iommu_flush_page(dev_state->domain, pasid_state->pasid, address);
+}
+
+static int mn_clear_flush_young(struct mmu_notifier *mn,
+                               struct mm_struct *mm,
+                               unsigned long address)
+{
+       __mn_flush_page(mn, address);
+
+       return 0;
+}
+
+static void mn_change_pte(struct mmu_notifier *mn,
+                         struct mm_struct *mm,
+                         unsigned long address,
+                         pte_t pte)
+{
+       __mn_flush_page(mn, address);
+}
+
+static void mn_invalidate_page(struct mmu_notifier *mn,
+                              struct mm_struct *mm,
+                              unsigned long address)
+{
+       __mn_flush_page(mn, address);
+}
+
+static void mn_invalidate_range_start(struct mmu_notifier *mn,
+                                     struct mm_struct *mm,
+                                     unsigned long start, unsigned long end)
+{
+       struct pasid_state *pasid_state;
+       struct device_state *dev_state;
+
+       pasid_state = mn_to_state(mn);
+       dev_state   = pasid_state->device_state;
+
+       amd_iommu_domain_set_gcr3(dev_state->domain, pasid_state->pasid,
+                                 __pa(empty_page_table));
+}
+
+static void mn_invalidate_range_end(struct mmu_notifier *mn,
+                                   struct mm_struct *mm,
+                                   unsigned long start, unsigned long end)
+{
+       struct pasid_state *pasid_state;
+       struct device_state *dev_state;
+
+       pasid_state = mn_to_state(mn);
+       dev_state   = pasid_state->device_state;
+
+       amd_iommu_domain_set_gcr3(dev_state->domain, pasid_state->pasid,
+                                 __pa(pasid_state->mm->pgd));
+}
+
+static struct mmu_notifier_ops iommu_mn = {
+       .clear_flush_young      = mn_clear_flush_young,
+       .change_pte             = mn_change_pte,
+       .invalidate_page        = mn_invalidate_page,
+       .invalidate_range_start = mn_invalidate_range_start,
+       .invalidate_range_end   = mn_invalidate_range_end,
+};
+
 static void set_pri_tag_status(struct pasid_state *pasid_state,
                               u16 tag, int status)
 {
@@ -475,6 +581,50 @@ static struct notifier_block ppr_nb = {
        .notifier_call = ppr_notifier,
 };
 
+static int task_exit(struct notifier_block *nb, unsigned long e, void *data)
+{
+       struct pasid_state *pasid_state;
+       struct task_struct *task;
+
+       task = data;
+
+       /*
+        * Using this notifier is a hack - but there is no other choice
+        * at the moment. What I really want is a sleeping notifier that
+        * is called when an MM goes down. But such a notifier doesn't
+        * exist yet. The notifier needs to sleep because it has to make
+        * sure that the device does not use the PASID and the address
+        * space anymore before it is destroyed. This includes waiting
+        * for pending PRI requests to pass the workqueue. The
+        * MMU-Notifiers would be a good fit, but they use RCU and so
+        * they are not allowed to sleep. Lets see how we can solve this
+        * in a more intelligent way in the future.
+        */
+again:
+       spin_lock(&ps_lock);
+       list_for_each_entry(pasid_state, &pasid_state_list, list) {
+               struct device_state *dev_state;
+               int pasid;
+
+               if (pasid_state->task != task)
+                       continue;
+
+               /* Drop Lock and unbind */
+               spin_unlock(&ps_lock);
+
+               dev_state = pasid_state->device_state;
+               pasid     = pasid_state->pasid;
+
+               unbind_pasid(dev_state, pasid);
+
+               /* Task may be in the list multiple times */
+               goto again;
+       }
+       spin_unlock(&ps_lock);
+
+       return NOTIFY_OK;
+}
+
 int amd_iommu_bind_pasid(struct pci_dev *pdev, int pasid,
                         struct task_struct *task)
 {
@@ -509,13 +659,16 @@ int amd_iommu_bind_pasid(struct pci_dev *pdev, int pasid,
        pasid_state->mm           = get_task_mm(task);
        pasid_state->device_state = dev_state;
        pasid_state->pasid        = pasid;
+       pasid_state->mn.ops       = &iommu_mn;
 
        if (pasid_state->mm == NULL)
                goto out_free;
 
+       mmu_notifier_register(&pasid_state->mn, pasid_state->mm);
+
        ret = set_pasid_state(dev_state, pasid_state, pasid);
        if (ret)
-               goto out_free;
+               goto out_unregister;
 
        ret = amd_iommu_domain_set_gcr3(dev_state->domain, pasid,
                                        __pa(pasid_state->mm->pgd));
@@ -529,6 +682,9 @@ int amd_iommu_bind_pasid(struct pci_dev *pdev, int pasid,
 out_clear_state:
        clear_pasid_state(dev_state, pasid);
 
+out_unregister:
+       mmu_notifier_unregister(&pasid_state->mn, pasid_state->mm);
+
 out_free:
        free_pasid_state(pasid_state);
 
@@ -689,15 +845,22 @@ static int __init amd_iommu_v2_init(void)
 
        ret = -ENOMEM;
        iommu_wq = create_workqueue("amd_iommu_v2");
-       if (iommu_wq == NULL) {
-               ret = -ENOMEM;
+       if (iommu_wq == NULL)
                goto out_free;
-       }
+
+       ret = -ENOMEM;
+       empty_page_table = (u64 *)get_zeroed_page(GFP_KERNEL);
+       if (empty_page_table == NULL)
+               goto out_destroy_wq;
 
        amd_iommu_register_ppr_notifier(&ppr_nb);
+       profile_event_register(PROFILE_TASK_EXIT, &profile_nb);
 
        return 0;
 
+out_destroy_wq:
+       destroy_workqueue(iommu_wq);
+
 out_free:
        free_pages((unsigned long)state_table, get_order(state_table_size));
 
@@ -710,6 +873,7 @@ static void __exit amd_iommu_v2_exit(void)
        size_t state_table_size;
        int i;
 
+       profile_event_unregister(PROFILE_TASK_EXIT, &profile_nb);
        amd_iommu_unregister_ppr_notifier(&ppr_nb);
 
        flush_workqueue(iommu_wq);
@@ -734,6 +898,8 @@ static void __exit amd_iommu_v2_exit(void)
 
        state_table_size = MAX_DEVICES * sizeof(struct device_state *);
        free_pages((unsigned long)state_table, get_order(state_table_size));
+
+       free_page((unsigned long)empty_page_table);
 }
 
 module_init(amd_iommu_v2_init);