qemu/vfio-Maintain-DMA-mapping-range-for-the-container.patch

From bd2d81775edf285149346bf793d9b71236d7cf34 Mon Sep 17 00:00:00 2001
From: Zenghui Yu <yuzenghui@huawei.com>
Date: Sat, 8 May 2021 17:31:04 +0800
Subject: [PATCH] vfio: Maintain DMA mapping range for the container

When synchronizing dirty bitmap from kernel VFIO we do it in a
per-iova-range fashion and we allocate the userspace bitmap for each of the
ioctl. This patch introduces `struct VFIODMARange` to describe a range of
the given DMA mapping with respect to a VFIO_IOMMU_MAP_DMA operation, and
make the bitmap cache of this range be persistent so that we don't need to
g_try_malloc0() every time. Note that the new structure is almost a copy of
`struct vfio_iommu_type1_dma_map` but only internally used by QEMU.

More importantly, the cached per-iova-range dirty bitmap will be further
used when we want to add support for the CLEAR_BITMAP and this cached
bitmap will be used to guarantee we don't clear any unknown dirty bits
otherwise that can be a severe data loss issue for migration code.

It's pretty intuitive to maintain a bitmap per container since we perform
log_sync at this granule. But I don't know how to deal with things like
memory hot-{un}plug, sparse DMA mappings, etc. Suggestions welcome.

* yet something to-do:
  - can't work with guest viommu
  - no locks
  - etc

[ The idea and even the commit message are largely inherited from kvm side.
  See commit 9f4bf4baa8b820c7930e23c9566c9493db7e1d25. ]

Signed-off-by: Zenghui Yu <yuzenghui@huawei.com>
Signed-off-by: Kunkun Jiang <jinagkunkun@huawei.com>
---
 hw/vfio/common.c              |  9 +++++--
 hw/vfio/container.c           | 49 +++++++++++++++++++++++++++++++++++
 include/hw/vfio/vfio-common.h | 12 +++++++++
 3 files changed, 68 insertions(+), 2 deletions(-)

diff --git a/hw/vfio/common.c b/hw/vfio/common.c
index e70fdf5e0c..564e933135 100644
--- a/hw/vfio/common.c
+++ b/hw/vfio/common.c
@@ -1156,6 +1156,7 @@ int vfio_get_dirty_bitmap(VFIOContainer *container, uint64_t iova,
         vfio_devices_all_device_dirty_tracking(container);
     uint64_t dirty_pages;
     VFIOBitmap vbmap;
+    VFIODMARange *qrange;
     int ret;
 
     if (!container->dirty_pages_supported && !all_device_dirty_tracking) {
@@ -1165,10 +1166,16 @@ int vfio_get_dirty_bitmap(VFIOContainer *container, uint64_t iova,
         return 0;
     }
 
+    qrange = vfio_lookup_match_range(container, iova, size);
+    /* the same as vfio_dma_unmap() */
+    assert(qrange);
+
     ret = vfio_bitmap_alloc(&vbmap, size);
     if (ret) {
         return ret;
     }
+    g_free(vbmap.bitmap);
+    vbmap.bitmap = qrange->bitmap;
 
     if (all_device_dirty_tracking) {
         ret = vfio_devices_query_dirty_bitmap(container, &vbmap, iova, size);
@@ -1186,8 +1193,6 @@ int vfio_get_dirty_bitmap(VFIOContainer *container, uint64_t iova,
     trace_vfio_get_dirty_bitmap(container->fd, iova, size, vbmap.size,
                                 ram_addr, dirty_pages);
 out:
-    g_free(vbmap.bitmap);
-
     return ret;
 }
 
diff --git a/hw/vfio/container.c b/hw/vfio/container.c
index 242010036a..9a176a0d33 100644
--- a/hw/vfio/container.c
+++ b/hw/vfio/container.c
@@ -112,6 +112,29 @@ unmap_exit:
     return ret;
 }
 
+VFIODMARange *vfio_lookup_match_range(VFIOContainer *container,
+        hwaddr start_addr, hwaddr size)
+{
+    VFIODMARange *qrange;
+
+    QLIST_FOREACH(qrange, &container->dma_list, next) {
+        if (qrange->iova == start_addr && qrange->size == size) {
+            return qrange;
+        }
+    }
+    return NULL;
+}
+
+void vfio_dma_range_init_dirty_bitmap(VFIODMARange *qrange)
+{
+    uint64_t pages, size;
+
+    pages = REAL_HOST_PAGE_ALIGN(qrange->size) / qemu_real_host_page_size();
+    size = ROUND_UP(pages, sizeof(__u64) * BITS_PER_BYTE) / BITS_PER_BYTE;
+
+    qrange->bitmap = g_malloc0(size);
+}
+
 /*
  * DMA - Mapping and unmapping for the "type1" IOMMU interface used on x86
  */
@@ -124,6 +147,7 @@ int vfio_dma_unmap(VFIOContainer *container, hwaddr iova,
         .iova = iova,
         .size = size,
     };
+    VFIODMARange *qrange;
     bool need_dirty_sync = false;
     int ret;
 
@@ -136,6 +160,22 @@ int vfio_dma_unmap(VFIOContainer *container, hwaddr iova,
         need_dirty_sync = true;
     }
 
+    /*
+     * unregister the DMA range
+     *
+     * It seems that the memory layer will give us the same section as the one
+     * used in region_add(). Otherwise it'll be complicated to manipulate the
+     * bitmap across region_{add,del}. Is there any guarantee?
+     *
+     * But there is really not such a restriction on the kernel interface
+     * (VFIO_IOMMU_DIRTY_PAGES_FLAG_{UN}MAP_DMA, etc).
+     */
+    qrange = vfio_lookup_match_range(container, iova, size);
+    assert(qrange);
+    g_free(qrange->bitmap);
+    QLIST_REMOVE(qrange, next);
+    g_free(qrange);
+
     while (ioctl(container->fd, VFIO_IOMMU_UNMAP_DMA, &unmap)) {
         /*
          * The type1 backend has an off-by-one bug in the kernel (71a7d3d78e3c
@@ -180,6 +220,14 @@ int vfio_dma_map(VFIOContainer *container, hwaddr iova,
         .iova = iova,
         .size = size,
     };
+    VFIODMARange *qrange;
+
+    qrange = g_malloc0(sizeof(*qrange));
+    qrange->iova = iova;
+    qrange->size = size;
+    QLIST_INSERT_HEAD(&container->dma_list, qrange, next);
+    /* XXX allocate the dirty bitmap on demand */
+    vfio_dma_range_init_dirty_bitmap(qrange);
 
     if (!readonly) {
         map.flags |= VFIO_DMA_MAP_FLAG_WRITE;
@@ -552,6 +600,7 @@ static int vfio_connect_container(VFIOGroup *group, AddressSpace *as,
     container->iova_ranges = NULL;
     QLIST_INIT(&container->giommu_list);
     QLIST_INIT(&container->vrdl_list);
+    QLIST_INIT(&container->dma_list);
 
     ret = vfio_init_container(container, group->fd, errp);
     if (ret) {
diff --git a/include/hw/vfio/vfio-common.h b/include/hw/vfio/vfio-common.h
index a4a22accb9..b131d04c9c 100644
--- a/include/hw/vfio/vfio-common.h
+++ b/include/hw/vfio/vfio-common.h
@@ -80,6 +80,14 @@ typedef struct VFIOAddressSpace {
 
 struct VFIOGroup;
 
+typedef struct VFIODMARange {
+    QLIST_ENTRY(VFIODMARange) next;
+    hwaddr iova;
+    size_t size;
+    void *vaddr; /* unused */
+    unsigned long *bitmap; /* dirty bitmap cache for this range */
+} VFIODMARange;
+
 typedef struct VFIOContainer {
     VFIOAddressSpace *space;
     int fd; /* /dev/vfio/vfio, empowered by the attached groups */
@@ -97,6 +105,7 @@ typedef struct VFIOContainer {
     QLIST_HEAD(, VFIOHostDMAWindow) hostwin_list;
     QLIST_HEAD(, VFIOGroup) group_list;
     QLIST_HEAD(, VFIORamDiscardListener) vrdl_list;
+    QLIST_HEAD(, VFIODMARange) dma_list;
     QLIST_ENTRY(VFIOContainer) next;
     QLIST_HEAD(, VFIODevice) device_list;
     GList *iova_ranges;
@@ -212,6 +221,9 @@ void vfio_put_address_space(VFIOAddressSpace *space);
 bool vfio_devices_all_running_and_saving(VFIOContainer *container);
 
 /* container->fd */
+VFIODMARange *vfio_lookup_match_range(VFIOContainer *container,
+		hwaddr start_addr, hwaddr size);
+void vfio_dma_range_init_dirty_bitmap(VFIODMARange *qrange);
 int vfio_dma_unmap(VFIOContainer *container, hwaddr iova,
                    ram_addr_t size, IOMMUTLBEntry *iotlb);
 int vfio_dma_map(VFIOContainer *container, hwaddr iova,
-- 
2.27.0