iommu: Introduce per cpu flag (iommu_dont_flush_iotlb) to avoid unnecessary iotlb...

Add cpu flag that will be checked by the iommu low level code
to skip iotlb flushes. iommu_iotlb_flush shall be called explicitly.

Signed-off-by: Jean Guyader <jean.guyader@eu.citrix.com>
Committed-by: Keir Fraser <keir@xen.org>

diff --git a/xen/arch/x86/mm.c b/xen/arch/x86/mm.c
index 44a444eae960ae4bd00675f0a102975e44895642..23afda908863c3bd462af7fd4698a52298a87c82 100644 (file)
--- a/xen/arch/x86/mm.c
+++ b/xen/arch/x86/mm.c
@@ -4794,10 +4794,15 @@ static int xenmem_add_to_physmap_once(struct domain *d,
 static int xenmem_add_to_physmap(struct domain *d,
                                  struct xen_add_to_physmap *xatp)
 {
+    struct xen_add_to_physmap start_xatp;
     int rc = 0;
 
     if ( xatp->space == XENMAPSPACE_gmfn_range )
     {
+        if ( need_iommu(d) )
+            this_cpu(iommu_dont_flush_iotlb) = 1;
+
+        start_xatp = *xatp;
         while ( xatp->size > 0 )
         {
             rc = xenmem_add_to_physmap_once(d, xatp);
@@ -4816,6 +4821,13 @@ static int xenmem_add_to_physmap(struct domain *d,
             }
         }
 
+        if ( need_iommu(d) )
+        {
+            this_cpu(iommu_dont_flush_iotlb) = 0;
+            iommu_iotlb_flush(d, start_xatp.idx, start_xatp.size - xatp->size);
+            iommu_iotlb_flush(d, start_xatp.gpfn, start_xatp.size - xatp->size);
+        }
+
         return rc;
     }
 

diff --git a/xen/drivers/passthrough/iommu.c b/xen/drivers/passthrough/iommu.c
index ca7b37b6c6cb3e6db41029a720505b5ac9c1bb58..bacca11d1691b503eb072ff2337bb83309925d26 100644 (file)
--- a/xen/drivers/passthrough/iommu.c
+++ b/xen/drivers/passthrough/iommu.c
@@ -52,6 +52,8 @@ bool_t __read_mostly iommu_hap_pt_share = 1;
 bool_t __read_mostly iommu_debug;
 bool_t __read_mostly amd_iommu_perdev_intremap;
 
+DEFINE_PER_CPU(bool_t, iommu_dont_flush_iotlb);
+
 static void __init parse_iommu_param(char *s)
 {
     char *ss;
@@ -227,6 +229,7 @@ static int iommu_populate_page_table(struct domain *d)
 
     spin_lock(&d->page_alloc_lock);
 
+    this_cpu(iommu_dont_flush_iotlb) = 1;
     page_list_for_each ( page, &d->page_list )
     {
         if ( is_hvm_domain(d) ||
@@ -244,6 +247,8 @@ static int iommu_populate_page_table(struct domain *d)
             }
         }
     }
+    this_cpu(iommu_dont_flush_iotlb) = 0;
+    iommu_iotlb_flush_all(d);
     spin_unlock(&d->page_alloc_lock);
     return 0;
 }

diff --git a/xen/drivers/passthrough/vtd/iommu.c b/xen/drivers/passthrough/vtd/iommu.c
index 73ebeb57f095201d289120919e6e611cd16f4b2d..adf06cc55918b65caaef6d3cc6e127fcb11ef7ce 100644 (file)
--- a/xen/drivers/passthrough/vtd/iommu.c
+++ b/xen/drivers/passthrough/vtd/iommu.c
@@ -663,7 +663,8 @@ static void dma_pte_clear_one(struct domain *domain, u64 addr)
     spin_unlock(&hd->mapping_lock);
     iommu_flush_cache_entry(pte, sizeof(struct dma_pte));
 
-    __intel_iommu_iotlb_flush(domain, addr >> PAGE_SHIFT_4K , 0, 1);
+    if ( !this_cpu(iommu_dont_flush_iotlb) )
+        __intel_iommu_iotlb_flush(domain, addr >> PAGE_SHIFT_4K , 0, 1);
 
     unmap_vtd_domain_page(page);
 
@@ -1760,7 +1761,8 @@ static int intel_iommu_map_page(
     spin_unlock(&hd->mapping_lock);
     unmap_vtd_domain_page(page);
 
-    __intel_iommu_iotlb_flush(d, gfn, dma_pte_present(old), 1);
+    if ( !this_cpu(iommu_dont_flush_iotlb) )
+        __intel_iommu_iotlb_flush(d, gfn, dma_pte_present(old), 1);
 
     return 0;
 }

diff --git a/xen/include/xen/iommu.h b/xen/include/xen/iommu.h
index 82d4cc4605e545ed3eee88c1c0e0c8943889151a..4f21e02818e8e5815d21704e6e8f7b6ac5d420fb 100644 (file)
--- a/xen/include/xen/iommu.h
+++ b/xen/include/xen/iommu.h
@@ -164,4 +164,16 @@ int iommu_do_domctl(struct xen_domctl *, XEN_GUEST_HANDLE(xen_domctl_t));
 void iommu_iotlb_flush(struct domain *d, unsigned long gfn, unsigned int page_count);
 void iommu_iotlb_flush_all(struct domain *d);
 
+/*
+ * The purpose of the iommu_dont_flush_iotlb optional cpu flag is to
+ * avoid unecessary iotlb_flush in the low level IOMMU code.
+ *
+ * iommu_map_page/iommu_unmap_page must flush the iotlb but somethimes
+ * this operation can be really expensive. This flag will be set by the
+ * caller to notify the low level IOMMU code to avoid the iotlb flushes.
+ * iommu_iotlb_flush/iommu_iotlb_flush_all will be explicitly called by
+ * the caller.
+ */
+DECLARE_PER_CPU(bool_t, iommu_dont_flush_iotlb);
+
 #endif /* _IOMMU_H_ */