Merge remote-tracking branch 'origin/master'

.github/workflows/release.yml

@@ -208,7 +208,7 @@ jobs:
           VERSION=${TAGV_NAME#v}
           echo "::set-output name=tagv::${TAG_NAME}"
           echo "::set-output name=version::${VERSION}"
-      - uses: golemfactory/build-deb-action@v4
+      - uses: golemfactory/build-deb-action@v0.6
         id: deb
         with:
           debVersion: ${{ steps.version.outputs.version }}

runtime/init-container/src/init.c

@@ -2097,6 +2097,23 @@ static int find_device_major(const char *name) {
     return major;
 }
 
+static int nvidia_gpu_count() {
+    int counter;
+    int res;
+    char buf[sizeof "sys/class/drm/card000"];
+
+    for (counter = 0; counter < 256; counter++) {
+        res = snprintf(buf, sizeof buf, "sys/class/drm/card%d", counter);
+        CHECK_BOOL(res > 0);
+        CHECK_BOOL(res < (int)sizeof buf);
+        /* iterate as long as devices are there, nothing unloads/unbinds
+         * devices so there should be no holes in numbering */
+        if (faccessat(g_sysroot_fd, buf, F_OK, 0) != 0)
+            break;
+    }
+    return counter;
+}
+
 int main(int argc, char **argv) {
     CHECK_BOOL(setvbuf(stdin, NULL, _IONBF, BUFSIZ) == 0);
     CHECK_BOOL(setvbuf(stdout, NULL, _IONBF, BUFSIZ) == 0);
@@ -2232,16 +2249,21 @@ int main(int argc, char **argv) {
     }
 
     if (nvidia_loaded) {
+        char buf[sizeof "dev/nvidia000"];
         if (do_sandbox == false) {
             fprintf(stderr, "Sandboxing is disabled, refusing to enable Nvidia GPU passthrough.\n");
             fprintf(stderr, "Please re-run the container with sandboxing enabled or disable GPU passthrough.\n");
             errno = 0;
             CHECK_BOOL(0);
         }
         int nvidia_major = CHECK(find_device_major("nvidia-frontend"));
-        /* TODO: multi-card support needs more /dev/nvidia%d nodes */
-        res = mknodat(g_sysroot_fd, "dev/nvidia0", S_IFCHR | (0666 & 0777), nvidia_major << 8 | 0);
-        CHECK_BOOL(res == 0 || (res == -1 && errno == EEXIST));
+        int nvidia_count = nvidia_gpu_count();
+        for (int i = 0; i < nvidia_count; i++) {
+            res = snprintf(buf, sizeof buf, "dev/nvidia%d", i);
+            CHECK_BOOL(res >= (int)sizeof "dev/nvidia" && res < (int)sizeof buf);
+            res = mknodat(g_sysroot_fd, buf, S_IFCHR | (0666 & 0777), nvidia_major << 8 | i);
+            CHECK_BOOL(res == 0 || (res == -1 && errno == EEXIST));
+        }
         res = mknodat(g_sysroot_fd, "dev/nvidiactl", S_IFCHR | (0666 & 0777), nvidia_major << 8 | 255);
         CHECK_BOOL(res == 0 || (res == -1 && errno == EEXIST));
         nvidia_major = CHECK(find_device_major("nvidia-uvm"));

runtime/src/lib.rs

@@ -71,7 +71,7 @@ pub struct Cli {
     inet_endpoint: Option<Url>,
     /// PCI device identifier
     #[structopt(long, env = "YA_RUNTIME_VM_PCI_DEVICE")]
-    pci_device: Option<String>,
+    pci_device: Option<Vec<String>>,
     #[structopt(flatten)]
     test_config: TestConfig,
 }

runtime/src/self_test.rs

@@ -23,7 +23,7 @@ const FILE_TEST_IMAGE: &str = "self-test.gvmi";
 const FILE_TEST_EXECUTABLE: &str = "ya-self-test";
 
 pub(crate) async fn test(
-    pci_device_id: Option<String>,
+    pci_device_id: Option<Vec<String>>,
     test_config: TestConfig,
 ) -> Result<(), Error> {
     run_self_test(verify_status, pci_device_id, test_config).await;
@@ -38,7 +38,7 @@ pub(crate) fn verify_status(status: anyhow::Result<Value>) -> anyhow::Result<Str
 
 pub(crate) async fn run_self_test<HANDLER>(
     handle_result: HANDLER,
-    pci_device_id: Option<String>,
+    pci_device_id: Option<Vec<String>>,
     test_config: TestConfig,
 ) where
     HANDLER: Fn(anyhow::Result<Value>) -> anyhow::Result<String>,
@@ -108,7 +108,7 @@ pub(crate) async fn run_self_test<HANDLER>(
     .await;
 }
 
-fn self_test_runtime(deployment: Deployment, pci_device_id: Option<String>) -> Runtime {
+fn self_test_runtime(deployment: Deployment, pci_device_id: Option<Vec<String>>) -> Runtime {
     let runtime_data = RuntimeData {
         deployment: Some(deployment),
         pci_device_id,

runtime/src/vmrt.rs

@@ -30,7 +30,7 @@ pub struct RuntimeData {
     pub inet: Option<ContainerEndpoint>,
     pub deployment: Option<Deployment>,
     pub ga: Option<Arc<Mutex<GuestAgent>>>,
-    pub pci_device_id: Option<String>,
+    pub pci_device_id: Option<Vec<String>>,
 }
 
 impl RuntimeData {
@@ -109,8 +109,10 @@ pub async fn start_vmrt(
     ]);
 
     if let Some(pci_device_id) = &data.pci_device_id {
-        cmd.arg("-device");
-        cmd.arg(format!("vfio-pci,host={}", pci_device_id).as_str());
+        for device_id in pci_device_id.iter() {
+            cmd.arg("-device");
+            cmd.arg(format!("vfio-pci,host={}", device_id).as_str());
+        }
     } else {
         cmd.arg("-vga");
         cmd.arg("none");