Merge remote-tracking branch 'remotes/stefanha-gitlab/tags/block-pull-request' into staging

Pull request

v2:
 * Removed clang-format call from scripts/block-coroutine-wrapper.py. This
   avoids the issue with clang version incompatibility. It could be added back
   in the future but the code is readable without reformatting and it also
   makes the build less dependent on the environment.

# gpg: Signature made Mon 05 Oct 2020 16:42:28 BST
# gpg:                using RSA key 8695A8BFD3F97CDAAC35775A9CA4ABB381AB73C8
# gpg: Good signature from "Stefan Hajnoczi <stefanha@redhat.com>" [full]
# gpg:                 aka "Stefan Hajnoczi <stefanha@gmail.com>" [full]
# Primary key fingerprint: 8695 A8BF D3F9 7CDA AC35  775A 9CA4 ABB3 81AB 73C8

* remotes/stefanha-gitlab/tags/block-pull-request:
  util/vfio-helpers: Rework the IOVA allocator to avoid IOVA reserved regions
  util/vfio-helpers: Collect IOVA reserved regions
  docs: add 'io_uring' option to 'aio' param in qemu-options.hx
  include/block/block.h: drop non-ascii quotation mark
  block/io: refactor save/load vmstate
  block: drop bdrv_prwv
  block: generate coroutine-wrapper code
  scripts: add block-coroutine-wrapper.py
  block: declare some coroutine functions in block/coroutines.h
  block/io: refactor coroutine wrappers
  block: return error-code from bdrv_invalidate_cache
  block/nvme: Replace magic value by SCALE_MS definition
  block/nvme: Use register definitions from 'block/nvme.h'
  block/nvme: Drop NVMeRegs structure, directly use NvmeBar
  block/nvme: Reduce I/O registers scope
  block/nvme: Map doorbells pages write-only
  util/vfio-helpers: Pass page protections to qemu_vfio_pci_map_bar()

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>

1 files changed, 125 insertions, 8 deletions

diff --git a/util/vfio-helpers.c b/util/vfio-helpers.c
index 583bdfb36f..c469beb061 100644
--- a/util/vfio-helpers.c
+++ b/util/vfio-helpers.c
@@ -40,6 +40,11 @@ typedef struct {
     uint64_t iova;
 } IOVAMapping;
 
+struct IOVARange {
+    uint64_t start;
+    uint64_t end;
+};
+
 struct QEMUVFIOState {
     QemuMutex lock;
 
@@ -49,6 +54,8 @@ struct QEMUVFIOState {
     int device;
     RAMBlockNotifier ram_notifier;
     struct vfio_region_info config_region_info, bar_region_info[6];
+    struct IOVARange *usable_iova_ranges;
+    uint8_t nb_iova_ranges;
 
     /* These fields are protected by @lock */
     /* VFIO's IO virtual address space is managed by splitting into a few
@@ -146,13 +153,13 @@ static int qemu_vfio_pci_init_bar(QEMUVFIOState *s, int index, Error **errp)
  * Map a PCI bar area.
  */
 void *qemu_vfio_pci_map_bar(QEMUVFIOState *s, int index,
-                            uint64_t offset, uint64_t size,
+                            uint64_t offset, uint64_t size, int prot,
                             Error **errp)
 {
     void *p;
     assert_bar_index_valid(s, index);
     p = mmap(NULL, MIN(size, s->bar_region_info[index].size - offset),
-             PROT_READ | PROT_WRITE, MAP_SHARED,
+             prot, MAP_SHARED,
              s->device, s->bar_region_info[index].offset + offset);
     if (p == MAP_FAILED) {
         error_setg_errno(errp, errno, "Failed to map BAR region");
@@ -236,6 +243,35 @@ static int qemu_vfio_pci_write_config(QEMUVFIOState *s, void *buf, int size, int
     return ret == size ? 0 : -errno;
 }
 
+static void collect_usable_iova_ranges(QEMUVFIOState *s, void *buf)
+{
+    struct vfio_iommu_type1_info *info = (struct vfio_iommu_type1_info *)buf;
+    struct vfio_info_cap_header *cap = (void *)buf + info->cap_offset;
+    struct vfio_iommu_type1_info_cap_iova_range *cap_iova_range;
+    int i;
+
+    while (cap->id != VFIO_IOMMU_TYPE1_INFO_CAP_IOVA_RANGE) {
+        if (!cap->next) {
+            return;
+        }
+        cap = (struct vfio_info_cap_header *)(buf + cap->next);
+    }
+
+    cap_iova_range = (struct vfio_iommu_type1_info_cap_iova_range *)cap;
+
+    s->nb_iova_ranges = cap_iova_range->nr_iovas;
+    if (s->nb_iova_ranges > 1) {
+        s->usable_iova_ranges =
+            g_realloc(s->usable_iova_ranges,
+                      s->nb_iova_ranges * sizeof(struct IOVARange));
+    }
+
+    for (i = 0; i < s->nb_iova_ranges; i++) {
+        s->usable_iova_ranges[i].start = cap_iova_range->iova_ranges[i].start;
+        s->usable_iova_ranges[i].end = cap_iova_range->iova_ranges[i].end;
+    }
+}
+
 static int qemu_vfio_init_pci(QEMUVFIOState *s, const char *device,
                               Error **errp)
 {
@@ -243,10 +279,13 @@ static int qemu_vfio_init_pci(QEMUVFIOState *s, const char *device,
     int i;
     uint16_t pci_cmd;
     struct vfio_group_status group_status = { .argsz = sizeof(group_status) };
-    struct vfio_iommu_type1_info iommu_info = { .argsz = sizeof(iommu_info) };
+    struct vfio_iommu_type1_info *iommu_info = NULL;
+    size_t iommu_info_size = sizeof(*iommu_info);
     struct vfio_device_info device_info = { .argsz = sizeof(device_info) };
     char *group_file = NULL;
 
+    s->usable_iova_ranges = NULL;
+
     /* Create a new container */
     s->container = open("/dev/vfio/vfio", O_RDWR);
 
@@ -310,13 +349,35 @@ static int qemu_vfio_init_pci(QEMUVFIOState *s, const char *device,
         goto fail;
     }
 
+    iommu_info = g_malloc0(iommu_info_size);
+    iommu_info->argsz = iommu_info_size;
+
     /* Get additional IOMMU info */
-    if (ioctl(s->container, VFIO_IOMMU_GET_INFO, &iommu_info)) {
+    if (ioctl(s->container, VFIO_IOMMU_GET_INFO, iommu_info)) {
         error_setg_errno(errp, errno, "Failed to get IOMMU info");
         ret = -errno;
         goto fail;
     }
 
+    /*
+     * if the kernel does not report usable IOVA regions, choose
+     * the legacy [QEMU_VFIO_IOVA_MIN, QEMU_VFIO_IOVA_MAX -1] region
+     */
+    s->nb_iova_ranges = 1;
+    s->usable_iova_ranges = g_new0(struct IOVARange, 1);
+    s->usable_iova_ranges[0].start = QEMU_VFIO_IOVA_MIN;
+    s->usable_iova_ranges[0].end = QEMU_VFIO_IOVA_MAX - 1;
+
+    if (iommu_info->argsz > iommu_info_size) {
+        iommu_info_size = iommu_info->argsz;
+        iommu_info = g_realloc(iommu_info, iommu_info_size);
+        if (ioctl(s->container, VFIO_IOMMU_GET_INFO, iommu_info)) {
+            ret = -errno;
+            goto fail;
+        }
+        collect_usable_iova_ranges(s, iommu_info);
+    }
+
     s->device = ioctl(s->group, VFIO_GROUP_GET_DEVICE_FD, device);
 
     if (s->device < 0) {
@@ -365,8 +426,13 @@ static int qemu_vfio_init_pci(QEMUVFIOState *s, const char *device,
     if (ret) {
         goto fail;
     }
+    g_free(iommu_info);
     return 0;
 fail:
+    g_free(s->usable_iova_ranges);
+    s->usable_iova_ranges = NULL;
+    s->nb_iova_ranges = 0;
+    g_free(iommu_info);
     close(s->group);
 fail_container:
     close(s->container);
@@ -601,6 +667,50 @@ static bool qemu_vfio_verify_mappings(QEMUVFIOState *s)
     return true;
 }
 
+static int
+qemu_vfio_find_fixed_iova(QEMUVFIOState *s, size_t size, uint64_t *iova)
+{
+    int i;
+
+    for (i = 0; i < s->nb_iova_ranges; i++) {
+        if (s->usable_iova_ranges[i].end < s->low_water_mark) {
+            continue;
+        }
+        s->low_water_mark =
+            MAX(s->low_water_mark, s->usable_iova_ranges[i].start);
+
+        if (s->usable_iova_ranges[i].end - s->low_water_mark + 1 >= size ||
+            s->usable_iova_ranges[i].end - s->low_water_mark + 1 == 0) {
+            *iova = s->low_water_mark;
+            s->low_water_mark += size;
+            return 0;
+        }
+    }
+    return -ENOMEM;
+}
+
+static int
+qemu_vfio_find_temp_iova(QEMUVFIOState *s, size_t size, uint64_t *iova)
+{
+    int i;
+
+    for (i = s->nb_iova_ranges - 1; i >= 0; i--) {
+        if (s->usable_iova_ranges[i].start > s->high_water_mark) {
+            continue;
+        }
+        s->high_water_mark =
+            MIN(s->high_water_mark, s->usable_iova_ranges[i].end + 1);
+
+        if (s->high_water_mark - s->usable_iova_ranges[i].start + 1 >= size ||
+            s->high_water_mark - s->usable_iova_ranges[i].start + 1 == 0) {
+            *iova = s->high_water_mark - size;
+            s->high_water_mark = *iova;
+            return 0;
+        }
+    }
+    return -ENOMEM;
+}
+
 /* Map [host, host + size) area into a contiguous IOVA address space, and store
  * the result in @iova if not NULL. The caller need to make sure the area is
  * aligned to page size, and mustn't overlap with existing mapping areas (split
@@ -627,7 +737,11 @@ int qemu_vfio_dma_map(QEMUVFIOState *s, void *host, size_t size,
             goto out;
         }
         if (!temporary) {
-            iova0 = s->low_water_mark;
+            if (qemu_vfio_find_fixed_iova(s, size, &iova0)) {
+                ret = -ENOMEM;
+                goto out;
+            }
+
             mapping = qemu_vfio_add_mapping(s, host, size, index + 1, iova0);
             if (!mapping) {
                 ret = -ENOMEM;
@@ -639,15 +753,16 @@ int qemu_vfio_dma_map(QEMUVFIOState *s, void *host, size_t size,
                 qemu_vfio_undo_mapping(s, mapping, NULL);
                 goto out;
             }
-            s->low_water_mark += size;
             qemu_vfio_dump_mappings(s);
         } else {
-            iova0 = s->high_water_mark - size;
+            if (qemu_vfio_find_temp_iova(s, size, &iova0)) {
+                ret = -ENOMEM;
+                goto out;
+            }
             ret = qemu_vfio_do_mapping(s, host, size, iova0);
             if (ret) {
                 goto out;
             }
-            s->high_water_mark -= size;
         }
     }
     if (iova) {
@@ -716,6 +831,8 @@ void qemu_vfio_close(QEMUVFIOState *s)
         qemu_vfio_undo_mapping(s, &s->mappings[i], NULL);
     }
     ram_block_notifier_remove(&s->ram_notifier);
+    g_free(s->usable_iova_ranges);
+    s->nb_iova_ranges = 0;
     qemu_vfio_reset(s);
     close(s->device);
     close(s->group);