IOMMU: make page table deallocation preemptible

This too can take an arbitrary amount of time.

In fact, the bulk of the work is being moved to a tasklet, as handling
the necessary preemption logic in line seems close to impossible given
that the teardown may also be invoked on error paths.

Signed-off-by: Jan Beulich <jbeulich@suse.com>
Reviewed-by: Andrew Cooper <andrew.cooper3@citrix.com>
Acked-by: Xiantao Zhang <xiantao.zhang@intel.com>

diff --git a/xen/drivers/passthrough/amd/pci_amd_iommu.c b/xen/drivers/passthrough/amd/pci_amd_iommu.c
index 7241aad459e4c7143f11e024053e1c1a6a387b18..c26aabce5dd014311422a65b449867e4ec7f55db 100644 (file)
--- a/xen/drivers/passthrough/amd/pci_amd_iommu.c
+++ b/xen/drivers/passthrough/amd/pci_amd_iommu.c
@@ -405,11 +405,21 @@ static int amd_iommu_assign_device(struct domain *d, u8 devfn,
     return reassign_device(dom0, d, devfn, pdev);
 }
 
-static void deallocate_next_page_table(struct page_info* pg, int level)
+static void deallocate_next_page_table(struct page_info *pg, int level)
+{
+    PFN_ORDER(pg) = level;
+    spin_lock(&iommu_pt_cleanup_lock);
+    page_list_add_tail(pg, &iommu_pt_cleanup_list);
+    spin_unlock(&iommu_pt_cleanup_lock);
+}
+
+static void deallocate_page_table(struct page_info *pg)
 {
     void *table_vaddr, *pde;
     u64 next_table_maddr;
-    int index, next_level;
+    unsigned int index, level = PFN_ORDER(pg), next_level;
+
+    PFN_ORDER(pg) = 0;
 
     if ( level <= 1 )
     {
@@ -599,6 +609,7 @@ const struct iommu_ops amd_iommu_ops = {
     .teardown = amd_iommu_domain_destroy,
     .map_page = amd_iommu_map_page,
     .unmap_page = amd_iommu_unmap_page,
+    .free_page_table = deallocate_page_table,
     .reassign_device = reassign_device,
     .get_device_group_id = amd_iommu_group_id,
     .update_ire_from_apic = amd_iommu_ioapic_update_ire,

diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
index e3b84f4f1b56939a1bc3fed170ea5e5f37a7e0e6..19b0e23feb278f875a30a819e3ed5d30307df09a 100644 (file)
--- a/xen/drivers/passthrough/iommu.c
+++ b/xen/drivers/passthrough/iommu.c
@@ -58,6 +58,10 @@ bool_t __read_mostly amd_iommu_perdev_intremap = 1;
 
 DEFINE_PER_CPU(bool_t, iommu_dont_flush_iotlb);
 
+DEFINE_SPINLOCK(iommu_pt_cleanup_lock);
+PAGE_LIST_HEAD(iommu_pt_cleanup_list);
+static struct tasklet iommu_pt_cleanup_tasklet;
+
 static struct keyhandler iommu_p2m_table = {
     .diagnostic = 0,
     .u.fn = iommu_dump_p2m_table,
@@ -251,6 +255,15 @@ int iommu_remove_device(struct pci_dev *pdev)
     return hd->platform_ops->remove_device(pdev->devfn, pdev);
 }
 
+static void iommu_teardown(struct domain *d)
+{
+    const struct hvm_iommu *hd = domain_hvm_iommu(d);
+
+    d->need_iommu = 0;
+    hd->platform_ops->teardown(d);
+    tasklet_schedule(&iommu_pt_cleanup_tasklet);
+}
+
 /*
  * If the device isn't owned by dom0, it means it already
  * has been assigned to other domain, or it doesn't exist.
@@ -325,10 +338,7 @@ static int assign_device(struct domain *d, u16 seg, u8 bus, u8 devfn)
 
  done:
     if ( !has_arch_pdevs(d) && need_iommu(d) )
-    {
-        d->need_iommu = 0;
-        hd->platform_ops->teardown(d);
-    }
+        iommu_teardown(d);
     spin_unlock(&pcidevs_lock);
 
     return rc;
@@ -393,10 +403,7 @@ static int iommu_populate_page_table(struct domain *d)
     if ( !rc )
         iommu_iotlb_flush_all(d);
     else if ( rc != -ERESTART )
-    {
-        d->need_iommu = 0;
-        hd->platform_ops->teardown(d);
-    }
+        iommu_teardown(d);
 
     return rc;
 }
@@ -413,10 +420,7 @@ void iommu_domain_destroy(struct domain *d)
         return;
 
     if ( need_iommu(d) )
-    {
-        d->need_iommu = 0;
-        hd->platform_ops->teardown(d);
-    }
+        iommu_teardown(d);
 
     list_for_each_safe ( ioport_list, tmp, &hd->g2m_ioport_list )
     {
@@ -454,6 +458,23 @@ int iommu_unmap_page(struct domain *d, unsigned long gfn)
     return hd->platform_ops->unmap_page(d, gfn);
 }
 
+static void iommu_free_pagetables(unsigned long unused)
+{
+    do {
+        struct page_info *pg;
+
+        spin_lock(&iommu_pt_cleanup_lock);
+        pg = page_list_remove_head(&iommu_pt_cleanup_list);
+        spin_unlock(&iommu_pt_cleanup_lock);
+        if ( !pg )
+            return;
+        iommu_get_ops()->free_page_table(pg);
+    } while ( !softirq_pending(smp_processor_id()) );
+
+    tasklet_schedule_on_cpu(&iommu_pt_cleanup_tasklet,
+                            cpumask_cycle(smp_processor_id(), &cpu_online_map));
+}
+
 void iommu_iotlb_flush(struct domain *d, unsigned long gfn, unsigned int page_count)
 {
     struct hvm_iommu *hd = domain_hvm_iommu(d);
@@ -516,10 +537,7 @@ int deassign_device(struct domain *d, u16 seg, u8 bus, u8 devfn)
     pdev->fault.count = 0;
 
     if ( !has_arch_pdevs(d) && need_iommu(d) )
-    {
-        d->need_iommu = 0;
-        hd->platform_ops->teardown(d);
-    }
+        iommu_teardown(d);
 
     return ret;
 }
@@ -558,6 +576,7 @@ int __init iommu_setup(void)
                iommu_passthrough ? "Passthrough" :
                iommu_dom0_strict ? "Strict" : "Relaxed");
         printk("Interrupt remapping %sabled\n", iommu_intremap ? "en" : "dis");
+        tasklet_init(&iommu_pt_cleanup_tasklet, iommu_free_pagetables, 0);
     }
 
     return rc;

diff --git a/xen/drivers/passthrough/vtd/iommu.c b/xen/drivers/passthrough/vtd/iommu.c
index b6b48e2672e4c34fa790d1f0cf72f2859f68ea67..5f10034dd8b8c852742118a30c164ba797d0446a 100644 (file)
--- a/xen/drivers/passthrough/vtd/iommu.c
+++ b/xen/drivers/passthrough/vtd/iommu.c
@@ -668,13 +668,24 @@ static void dma_pte_clear_one(struct domain *domain, u64 addr)
 
 static void iommu_free_pagetable(u64 pt_maddr, int level)
 {
-    int i;
-    struct dma_pte *pt_vaddr, *pte;
-    int next_level = level - 1;
+    struct page_info *pg = maddr_to_page(pt_maddr);
 
     if ( pt_maddr == 0 )
         return;
 
+    PFN_ORDER(pg) = level;
+    spin_lock(&iommu_pt_cleanup_lock);
+    page_list_add_tail(pg, &iommu_pt_cleanup_list);
+    spin_unlock(&iommu_pt_cleanup_lock);
+}
+
+static void iommu_free_page_table(struct page_info *pg)
+{
+    unsigned int i, next_level = PFN_ORDER(pg) - 1;
+    u64 pt_maddr = page_to_maddr(pg);
+    struct dma_pte *pt_vaddr, *pte;
+
+    PFN_ORDER(pg) = 0;
     pt_vaddr = (struct dma_pte *)map_vtd_domain_page(pt_maddr);
 
     for ( i = 0; i < PTE_NUM; i++ )
@@ -2430,6 +2441,7 @@ const struct iommu_ops intel_iommu_ops = {
     .teardown = iommu_domain_teardown,
     .map_page = intel_iommu_map_page,
     .unmap_page = intel_iommu_unmap_page,
+    .free_page_table = iommu_free_page_table,
     .reassign_device = reassign_device_ownership,
     .get_device_group_id = intel_iommu_group_id,
     .update_ire_from_apic = io_apic_write_remap_rte,

diff --git a/xen/include/xen/iommu.h b/xen/include/xen/iommu.h
index 6f0ff9d749f28f845a3fe736e40430e74da4fb18..8bb0a1d1aac307bcaf0120a587f6480883d192ab 100644 (file)
--- a/xen/include/xen/iommu.h
+++ b/xen/include/xen/iommu.h
@@ -88,6 +88,7 @@ bool_t pt_irq_need_timer(uint32_t flags);
 
 struct msi_desc;
 struct msi_msg;
+struct page_info;
 
 struct iommu_ops {
     int (*init)(struct domain *d);
@@ -100,6 +101,7 @@ struct iommu_ops {
     int (*map_page)(struct domain *d, unsigned long gfn, unsigned long mfn,
                     unsigned int flags);
     int (*unmap_page)(struct domain *d, unsigned long gfn);
+    void (*free_page_table)(struct page_info *);
     int (*reassign_device)(struct domain *s, struct domain *t,
                           u8 devfn, struct pci_dev *);
     int (*get_device_group_id)(u16 seg, u8 bus, u8 devfn);
@@ -151,4 +153,7 @@ int adjust_vtd_irq_affinities(void);
  */
 DECLARE_PER_CPU(bool_t, iommu_dont_flush_iotlb);
 
+extern struct spinlock iommu_pt_cleanup_lock;
+extern struct page_list_head iommu_pt_cleanup_list;
+
 #endif /* _IOMMU_H_ */