vhost: get guest/host physical address mappings
authorYuanhan Liu <yuanhan.liu@linux.intel.com>
Sun, 9 Oct 2016 07:27:55 +0000 (15:27 +0800)
committerYuanhan Liu <yuanhan.liu@linux.intel.com>
Wed, 12 Oct 2016 07:45:09 +0000 (09:45 +0200)
So that we can convert a guest physical address to host physical
address, which will be used in later Tx zero copy implementation.

MAP_POPULATE is set while mmaping guest memory regions, to make
sure the page tables are setup and then rte_mem_virt2phy() could
yield proper physical address.

Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Tested-by: Qian Xu <qian.q.xu@intel.com>
lib/librte_vhost/vhost.h
lib/librte_vhost/vhost_user.c

index df2107b..2d52987 100644 (file)
@@ -114,6 +114,12 @@ struct vhost_virtqueue {
  #define VIRTIO_F_VERSION_1 32
 #endif
 
+struct guest_page {
+       uint64_t guest_phys_addr;
+       uint64_t host_phys_addr;
+       uint64_t size;
+};
+
 /**
  * Device structure contains all configuration information relating
  * to the device.
@@ -137,6 +143,10 @@ struct virtio_net {
        uint64_t                log_addr;
        struct ether_addr       mac;
 
+       uint32_t                nr_guest_pages;
+       uint32_t                max_guest_pages;
+       struct guest_page       *guest_pages;
+
 } __rte_cache_aligned;
 
 /**
@@ -217,6 +227,26 @@ gpa_to_vva(struct virtio_net *dev, uint64_t gpa)
        return 0;
 }
 
+/* Convert guest physical address to host physical address */
+static inline phys_addr_t __attribute__((always_inline))
+gpa_to_hpa(struct virtio_net *dev, uint64_t gpa, uint64_t size)
+{
+       uint32_t i;
+       struct guest_page *page;
+
+       for (i = 0; i < dev->nr_guest_pages; i++) {
+               page = &dev->guest_pages[i];
+
+               if (gpa >= page->guest_phys_addr &&
+                   gpa + size < page->guest_phys_addr + page->size) {
+                       return gpa - page->guest_phys_addr +
+                              page->host_phys_addr;
+               }
+       }
+
+       return 0;
+}
+
 struct virtio_net_device_ops const *notify_ops;
 struct virtio_net *get_device(int vid);
 
index 49585b8..e651912 100644 (file)
@@ -372,6 +372,91 @@ vhost_user_set_vring_base(struct virtio_net *dev,
        return 0;
 }
 
+static void
+add_one_guest_page(struct virtio_net *dev, uint64_t guest_phys_addr,
+                  uint64_t host_phys_addr, uint64_t size)
+{
+       struct guest_page *page, *last_page;
+
+       if (dev->nr_guest_pages == dev->max_guest_pages) {
+               dev->max_guest_pages *= 2;
+               dev->guest_pages = realloc(dev->guest_pages,
+                                       dev->max_guest_pages * sizeof(*page));
+       }
+
+       if (dev->nr_guest_pages > 0) {
+               last_page = &dev->guest_pages[dev->nr_guest_pages - 1];
+               /* merge if the two pages are continuous */
+               if (host_phys_addr == last_page->host_phys_addr +
+                                     last_page->size) {
+                       last_page->size += size;
+                       return;
+               }
+       }
+
+       page = &dev->guest_pages[dev->nr_guest_pages++];
+       page->guest_phys_addr = guest_phys_addr;
+       page->host_phys_addr  = host_phys_addr;
+       page->size = size;
+}
+
+static void
+add_guest_pages(struct virtio_net *dev, struct virtio_memory_region *reg,
+               uint64_t page_size)
+{
+       uint64_t reg_size = reg->size;
+       uint64_t host_user_addr  = reg->host_user_addr;
+       uint64_t guest_phys_addr = reg->guest_phys_addr;
+       uint64_t host_phys_addr;
+       uint64_t size;
+
+       host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)host_user_addr);
+       size = page_size - (guest_phys_addr & (page_size - 1));
+       size = RTE_MIN(size, reg_size);
+
+       add_one_guest_page(dev, guest_phys_addr, host_phys_addr, size);
+       host_user_addr  += size;
+       guest_phys_addr += size;
+       reg_size -= size;
+
+       while (reg_size > 0) {
+               host_phys_addr = rte_mem_virt2phy((void *)(uintptr_t)
+                                                 host_user_addr);
+               add_one_guest_page(dev, guest_phys_addr, host_phys_addr,
+                                  page_size);
+
+               host_user_addr  += page_size;
+               guest_phys_addr += page_size;
+               reg_size -= page_size;
+       }
+}
+
+#ifdef RTE_LIBRTE_VHOST_DEBUG
+/* TODO: enable it only in debug mode? */
+static void
+dump_guest_pages(struct virtio_net *dev)
+{
+       uint32_t i;
+       struct guest_page *page;
+
+       for (i = 0; i < dev->nr_guest_pages; i++) {
+               page = &dev->guest_pages[i];
+
+               RTE_LOG(INFO, VHOST_CONFIG,
+                       "guest physical page region %u\n"
+                       "\t guest_phys_addr: %" PRIx64 "\n"
+                       "\t host_phys_addr : %" PRIx64 "\n"
+                       "\t size           : %" PRIx64 "\n",
+                       i,
+                       page->guest_phys_addr,
+                       page->host_phys_addr,
+                       page->size);
+       }
+}
+#else
+#define dump_guest_pages(dev)
+#endif
+
 static int
 vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
 {
@@ -396,6 +481,13 @@ vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
                dev->mem = NULL;
        }
 
+       dev->nr_guest_pages = 0;
+       if (!dev->guest_pages) {
+               dev->max_guest_pages = 8;
+               dev->guest_pages = malloc(dev->max_guest_pages *
+                                               sizeof(struct guest_page));
+       }
+
        dev->mem = rte_zmalloc("vhost-mem-table", sizeof(struct virtio_memory) +
                sizeof(struct virtio_memory_region) * memory.nregions, 0);
        if (dev->mem == NULL) {
@@ -434,8 +526,8 @@ vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
                }
                mmap_size = RTE_ALIGN_CEIL(mmap_size, alignment);
 
-               mmap_addr = mmap(NULL, mmap_size,
-                                PROT_READ | PROT_WRITE, MAP_SHARED, fd, 0);
+               mmap_addr = mmap(NULL, mmap_size, PROT_READ | PROT_WRITE,
+                                MAP_SHARED | MAP_POPULATE, fd, 0);
 
                if (mmap_addr == MAP_FAILED) {
                        RTE_LOG(ERR, VHOST_CONFIG,
@@ -448,6 +540,8 @@ vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
                reg->host_user_addr = (uint64_t)(uintptr_t)mmap_addr +
                                      mmap_offset;
 
+               add_guest_pages(dev, reg, alignment);
+
                RTE_LOG(INFO, VHOST_CONFIG,
                        "guest memory region %u, size: 0x%" PRIx64 "\n"
                        "\t guest physical addr: 0x%" PRIx64 "\n"
@@ -467,6 +561,8 @@ vhost_user_set_mem_table(struct virtio_net *dev, struct VhostUserMsg *pmsg)
                        mmap_offset);
        }
 
+       dump_guest_pages(dev);
+
        return 0;
 
 err_mmap: