Merge pull request NVIDIA#242 from klueska/add-imex-support

Add imex support

.gitignore

@@ -12,9 +12,12 @@ libnvidia-container.so*
 deps
 src/build.h
 
-src/driver_clt.c
-src/driver_rpc.h
-src/driver_svc.c
-src/driver_xdr.c
+src/nvc.h
+src/nvc_clt.c
+src/nvc_rpc.h
+src/nvc_svc.c
+src/nvc_xdr.c
+src/nvcgo/libnvidia-container-go.h
+
 /dist
-/.vscode
+/.vscode

mk/nvidia-modprobe.mk

@@ -6,7 +6,7 @@ include $(MAKE_DIR)/common.mk
 
 ##### Source definitions #####
 
-VERSION        := 495.44
+VERSION        := 550.54.14
 PREFIX         := nvidia-modprobe-$(VERSION)
 URL            := https://github.com/NVIDIA/nvidia-modprobe/archive/$(VERSION).tar.gz
 

mk/nvidia-modprobe.patch

@@ -1,7 +1,7 @@
 diff -ruN nvidia-modprobe-495.44/modprobe-utils/nvidia-modprobe-utils.c nvidia-modprobe-495.44-patched/modprobe-utils/nvidia-modprobe-utils.c
 --- nvidia-modprobe-495.44/modprobe-utils/nvidia-modprobe-utils.c	2021-11-13 14:36:58.096684602 +0000
 +++ nvidia-modprobe-495.44-patched/modprobe-utils/nvidia-modprobe-utils.c	2021-11-13 14:43:40.965146390 +0000
-@@ -888,10 +888,10 @@
+@@ -959,10 +959,10 @@
      return mknod_helper(major, minor_num, vgpu_dev_name, NV_PROC_REGISTRY_PATH);
  }
 
@@ -19,11 +19,13 @@ diff -ruN nvidia-modprobe-495.44/modprobe-utils/nvidia-modprobe-utils.c nvidia-m
 diff -ruN nvidia-modprobe-495.44/modprobe-utils/nvidia-modprobe-utils.h nvidia-modprobe-495.44-patched/modprobe-utils/nvidia-modprobe-utils.h
 --- nvidia-modprobe-495.44/modprobe-utils/nvidia-modprobe-utils.h	2021-11-13 14:36:58.096684602 +0000
 +++ nvidia-modprobe-495.44-patched/modprobe-utils/nvidia-modprobe-utils.h	2021-11-13 14:38:34.078700961 +0000
-@@ -81,6 +81,7 @@
+@@ -87,6 +87,7 @@
  int nvidia_nvswitch_get_file_state(int minor);
  int nvidia_cap_mknod(const char* cap_file_path, int *minor);
  int nvidia_cap_get_file_state(const char* cap_file_path);
 +int nvidia_cap_get_device_file_attrs(const char* cap_file_path, int *major, int *minor, char *name);
+ int nvidia_cap_imex_channel_mknod(int minor);
+ int nvidia_cap_imex_channel_file_state(int minor);
  int nvidia_get_chardev_major(const char *name);
  int nvidia_msr_modprobe(void);
 

pkg/deb/libnvidia-container@MAJOR@.symbols

@@ -17,6 +17,7 @@ libnvidia-container.so.1 libnvidia-container1 #MINVER#
  nvc_mig_config_global_caps_mount@NVC_1.0 @VERSION_TAG@
  nvc_mig_monitor_global_caps_mount@NVC_1.0 @VERSION_TAG@
  nvc_device_mig_caps_mount@NVC_1.0 @VERSION_TAG@
+ nvc_imex_channel_mount@NVC_1.0 @VERSION_TAG@
  nvc_driver_info_free@NVC_1.0 @VERSION_TAG@
  nvc_driver_info_new@NVC_1.0 @VERSION_TAG@
  nvc_driver_mount@NVC_1.0 @VERSION_TAG@

src/cli/cli.h

@@ -54,6 +54,7 @@ struct context {
         char *devices;
         char *mig_config;
         char *mig_monitor;
+        char *imex_channels;
 };
 
 bool matches_pci_format(const char *gpu, char *buf, size_t bufsize);
@@ -74,6 +75,11 @@ void free_devices(struct devices *d);
 int print_nvcaps_device_from_proc_file(struct nvc_context *, const char*, const char*);
 int print_all_mig_minor_devices(const struct nvc_device_node *);
 
+int parse_imex_info(
+    struct error *err,
+    char *chans,
+    struct nvc_imex_info *imex);
+
 int select_devices(
     struct error *err,
     char *devs,

src/cli/common.c

@@ -442,6 +442,59 @@ select_mig_monitor_devices(
         return (-1);
 }
 
+int
+parse_imex_info(
+    struct error *err,
+    char *chans,
+    struct nvc_imex_info *imex)
+{
+        // Initialize local variables.
+        char sep[2] = ",";
+        struct error ierr = {0};
+        char *chan = NULL;
+        char *ptr;
+        size_t id;
+        size_t max_chans = str_count_tokens(chans, ',');
+
+        /* Clear the imex struct */
+        memset(imex, 0, sizeof(*imex));
+
+        /* Short circuit if max_chans == 0 */
+        if (max_chans == 0)
+                return (0);
+
+        /* Allocate space for all IMEX channels */
+        if ((imex->chans = xcalloc(err, max_chans, sizeof(*imex->chans))) == NULL)
+                return (-1);
+
+        // Walk through the comma separated chans string and populate
+        // 'imex->chans' from it.
+        while ((chan = strsep(&chans, sep)) != NULL) {
+                // Allow extra commas between device strings.
+                if (*chan == '\0')
+                        continue;
+
+                // Get the IMEX channel ID.
+                // Channel IDs must fit in the minor number of a dev_t (so within 20 bits).
+                id = strtoumax(chan, &ptr, 10);
+                if (!(*ptr == '\0' && id < (1 << 20))) {
+                        error_setx(&ierr, "unsupported IMEX channel value: %s", chan);
+                        goto fail;
+                }
+
+                // Add the IMEX channel to the selected list.
+                imex->chans[imex->nchans].id = (int)id;
+                imex->nchans++;
+        }
+
+        return (0);
+
+ fail:
+        error_setx(err, "%s", ierr.msg);
+        error_reset(&ierr);
+        return (-1);
+}
+
 int
 new_devices(struct error *err, const struct nvc_device_info *dev, struct devices *d)
 {

src/cli/configure.c

@@ -30,8 +30,9 @@ const struct argp configure_usage = {
                 {"compat32", 0x80, NULL, 0, "Enable 32bits compatibility", -1},
                 {"mig-config", 0x81, "ID", 0, "Enable configuration of MIG devices", -1},
                 {"mig-monitor", 0x82, "ID", 0, "Enable monitoring of MIG devices", -1},
-                {"no-cgroups", 0x83, NULL, 0, "Don't use cgroup enforcement", -1},
-                {"no-devbind", 0x84, NULL, 0, "Don't bind mount devices", -1},
+                {"imex-channel", 0x83, "CHANNEL", 0, "IMEX channel ID(s) to inject", -1},
+                {"no-cgroups", 0x84, NULL, 0, "Don't use cgroup enforcement", -1},
+                {"no-devbind", 0x85, NULL, 0, "Don't bind mount devices", -1},
                 {0},
         },
         configure_parser,
@@ -138,10 +139,14 @@ configure_parser(int key, char *arg, struct argp_state *state)
                         goto fatal;
                 break;
         case 0x83:
-                if (str_join(&err, &ctx->container_flags, "no-cgroups", " ") < 0)
+                if (str_join(&err, &ctx->imex_channels, arg, ",") < 0)
                         goto fatal;
                 break;
         case 0x84:
+                if (str_join(&err, &ctx->container_flags, "no-cgroups", " ") < 0)
+                        goto fatal;
+                break;
+        case 0x85:
                 if (str_join(&err, &ctx->container_flags, "no-devbind", " ") < 0)
                         goto fatal;
                 break;
@@ -262,6 +267,10 @@ configure_command(const struct context *ctx)
         nvc_cfg->gid = ctx->gid;
         nvc_cfg->root = ctx->root;
         nvc_cfg->ldcache = ctx->ldcache;
+        if (parse_imex_info(&err, ctx->imex_channels, &nvc_cfg->imex) < 0) {
+                warnx("error parsing IMEX info: %s", err.msg);
+                goto fail;
+        }
         if (libnvc.init(nvc, nvc_cfg, ctx->init_flags) < 0) {
                 warnx("initialization error: %s", libnvc.error(nvc));
                 goto fail;
@@ -319,7 +328,7 @@ configure_command(const struct context *ctx)
                 goto fail;
         }
 
-        /* Select the devices available for MIG monitor among the visible . */
+        /* Select the devices available for MIG monitor among the visible devices. */
         if (select_mig_monitor_devices(&err, ctx->mig_monitor, &devices, &mig_monitor_devices) < 0) {
                 warnx("mig-monitor error: %s", err.msg);
                 goto fail;
@@ -359,7 +368,7 @@ configure_command(const struct context *ctx)
                 }
         }
 
-        /* Mount the driver, visible devices, mig-configs and mig-monitors. */
+        /* Mount the driver, visible devices, mig-configs, mig-monitors, and imex-channels. */
         if (perm_set_capabilities(&err, CAP_EFFECTIVE, ecaps[NVC_MOUNT], ecaps_size(NVC_MOUNT)) < 0) {
                 warnx("permission error: %s", err.msg);
                 goto fail;
@@ -406,6 +415,12 @@ configure_command(const struct context *ctx)
                         }
                 }
         }
+        for (size_t i = 0; i < nvc_cfg->imex.nchans; ++i) {
+                if (libnvc.imex_channel_mount(nvc, cnt, &nvc_cfg->imex.chans[i]) < 0) {
+                        warnx("mount error: %s", libnvc.error(nvc));
+                        goto fail;
+                }
+        }
 
         /* Update the container ldcache. */
         if (perm_set_capabilities(&err, CAP_EFFECTIVE, ecaps[NVC_LDCACHE], ecaps_size(NVC_LDCACHE)) < 0) {
@@ -424,6 +439,7 @@ configure_command(const struct context *ctx)
         rv = EXIT_SUCCESS;
 
  fail:
+        free(nvc_cfg->imex.chans);
         free_devices(&devices);
         libnvc.shutdown(nvc);
         libnvc.container_free(cnt);

src/cli/libnvc.c

@@ -161,6 +161,7 @@ load_libnvc_v1(void)
         load_libnvc_func(mig_config_global_caps_mount);
         load_libnvc_func(mig_monitor_global_caps_mount);
         load_libnvc_func(device_mig_caps_mount);
+        load_libnvc_func(imex_channel_mount);
 
         return (0);
 }

src/cli/libnvc.h

@@ -50,6 +50,7 @@ struct libnvc {
         libnvc_entry(mig_config_global_caps_mount);
         libnvc_entry(mig_monitor_global_caps_mount);
         libnvc_entry(device_mig_caps_mount);
+        libnvc_entry(imex_channel_mount);
 };
 
 int load_libnvc(void);

src/cli/list.c

@@ -21,6 +21,7 @@ const struct argp list_usage = {
                 {"compat32", 0x80, NULL, 0, "Enable 32bits compatibility", -1},
                 {"mig-config", 0x81, "ID", 0, "MIG devices to list config capabilities files for", -1},
                 {"mig-monitor", 0x82, "ID", 0, "MIG devices to list monitor capabilities files for", -1},
+                {"imex-channel", 0x83, "CHANNEL", 0, "IMEX channel ID(s) to inject", -1},
                 {0},
         },
         list_parser,
@@ -65,8 +66,12 @@ list_parser(int key, char *arg, struct argp_state *state)
                 if (str_join(&err, &ctx->mig_monitor, arg, ",") < 0)
                         goto fatal;
                 break;
+        case 0x83:
+                if (str_join(&err, &ctx->imex_channels, arg, ",") < 0)
+                        goto fatal;
+                break;
         case ARGP_KEY_END:
-                if (state->argc == 1) {
+                if (state->argc == 1 || (state->argc == 2 && ctx->imex_channels != NULL)) {
                         if ((ctx->devices = xstrdup(&err, "all")) == NULL)
                                 goto fatal;
                         ctx->mig_config = NULL;
@@ -127,6 +132,10 @@ list_command(const struct context *ctx)
         nvc_cfg->gid = (!run_as_root && ctx->gid == (gid_t)-1) ? getegid() : ctx->gid;
         nvc_cfg->root = ctx->root;
         nvc_cfg->ldcache = ctx->ldcache;
+        if (parse_imex_info(&err, ctx->imex_channels, &nvc_cfg->imex) < 0) {
+                warnx("error parsing IMEX info: %s", err.msg);
+                goto fail;
+        }
         if (libnvc.init(nvc, nvc_cfg, ctx->init_flags) < 0) {
                 warnx("initialization error: %s", libnvc.error(nvc));
                 goto fail;
@@ -203,6 +212,13 @@ list_command(const struct context *ctx)
                 }
         }
 
+        /* List the IMEX channel devices. */
+        if (ctx->imex_channels != NULL) {
+                for (size_t i = 0; i < nvc_cfg->imex.nchans; ++i) {
+                        printf(NV_CAPS_IMEX_DEVICE_PATH"\n", nvc_cfg->imex.chans[i].id);
+                }
+        }
+
         /* List the files required for MIG configuration of the visible devices */
         if (mig_config_devices.all && mig_config_devices.ngpus) {
                 printf("%s/%s\n", NV_MIG_CAPS_PATH, NV_MIG_CONFIG_FILE);
@@ -258,6 +274,7 @@ list_command(const struct context *ctx)
         }
         rv = EXIT_SUCCESS;
  fail:
+        free(nvc_cfg->imex.chans);
         free_devices(&devices);
         libnvc.shutdown(nvc);
         libnvc.device_info_free(dev);

src/cli/main.c

@@ -146,5 +146,8 @@ main(int argc, char *argv[])
         free(ctx.devices);
         free(ctx.init_flags);
         free(ctx.container_flags);
+        free(ctx.mig_config);
+        free(ctx.mig_monitor);
+        free(ctx.imex_channels);
         return (rv);
 }

src/libnvidia-container.lds

@@ -39,6 +39,7 @@ VERSION {
             nvc_mig_config_global_caps_mount;
             nvc_mig_monitor_global_caps_mount;
             nvc_device_mig_caps_mount;
+            nvc_imex_channel_mount;
 
             __ubsan_default_options;
         local:

src/nvc.c

@@ -33,7 +33,7 @@
 #include "xfuncs.h"
 
 static int init_within_userns(struct error *);
-static int load_kernel_modules(struct error *, const char *);
+static int load_kernel_modules(struct error *, const char *, const struct nvc_imex_info *);
 static int copy_config(struct error *, struct nvc_context *, const struct nvc_config *);
 
 const char interpreter[] __attribute__((section(".interp"))) = LIB_DIR "/" LD_SO;
@@ -229,7 +229,7 @@ mig_nvcaps_mknodes(struct error *err, int num_gpus) {
 }
 
 static int
-load_kernel_modules(struct error *err, const char *root)
+load_kernel_modules(struct error *err, const char *root, const struct nvc_imex_info *imex)
 {
         int userns;
         pid_t pid;
@@ -290,6 +290,11 @@ load_kernel_modules(struct error *err, const char *root)
                         log_info("running mknod for all nvcaps in " NV_CAPS_DEVICE_DIR);
                         if (mig_nvcaps_mknodes(err, devs.num_matches) < 0)
                                 log_errf("could not create kernel module device nodes: %s", err->msg);
+                        for (int i = 0; i < (int)imex->nchans; ++i) {
+                                log_infof("running mknod for " NV_CAPS_IMEX_DEVICE_PATH, imex->chans[i].id);
+                                if (nvidia_cap_imex_channel_mknod(imex->chans[i].id) == 0)
+                                    log_errf("could not mknod for IMEX channel %d", imex->chans[i].id);
+                        }
                         error_reset(err);
                 }
 
@@ -347,9 +352,21 @@ copy_config(struct error *err, struct nvc_context *ctx, const struct nvc_config
                 ctx->cfg.gid = (gid_t)gid;
         }
 
+        if (cfg->imex.nchans > 0) {
+                if ((ctx->cfg.imex.chans = xcalloc(err, cfg->imex.nchans, sizeof(*ctx->cfg.imex.chans))) == NULL)
+                        return (-1);
+        }
+        for (size_t i = 0; i < cfg->imex.nchans; ++i) {
+                ctx->cfg.imex.chans[i] = cfg->imex.chans[i];
+        }
+        ctx->cfg.imex.nchans = cfg->imex.nchans;
+
         log_infof("using root %s", ctx->cfg.root);
         log_infof("using ldcache %s", ctx->cfg.ldcache);
         log_infof("using unprivileged user %"PRIu32":%"PRIu32, (uint32_t)ctx->cfg.uid, (uint32_t)ctx->cfg.gid);
+        for (size_t i = 0; i < ctx->cfg.imex.nchans; ++i) {
+            log_infof("using IMEX channel %d", ctx->cfg.imex.chans[i].id);
+        }
         return (0);
 }
 
@@ -364,7 +381,7 @@ nvc_init(struct nvc_context *ctx, const struct nvc_config *cfg, const char *opts
         if (ctx->initialized)
                 return (0);
         if (cfg == NULL)
-                cfg = &(struct nvc_config){NULL, NULL, (uid_t)-1, (gid_t)-1};
+                cfg = &(struct nvc_config){NULL, NULL, (uid_t)-1, (gid_t)-1, {0}};
         if (validate_args(ctx, !str_empty(cfg->ldcache) && !str_empty(cfg->root)) < 0)
                 return (-1);
         if (opts == NULL)
@@ -403,7 +420,7 @@ nvc_init(struct nvc_context *ctx, const struct nvc_config *cfg, const char *opts
         if (flags & OPT_LOAD_KMODS) {
                 if (ctx->dxcore.initialized)
                         log_warn("skipping kernel modules load on WSL");
-                else if (load_kernel_modules(&ctx->err, ctx->cfg.root) < 0)
+                else if (load_kernel_modules(&ctx->err, ctx->cfg.root, &ctx->cfg.imex) < 0)
                         goto fail;
         }
 
@@ -421,6 +438,7 @@ nvc_init(struct nvc_context *ctx, const struct nvc_config *cfg, const char *opts
  fail:
         free(ctx->cfg.root);
         free(ctx->cfg.ldcache);
+        free(ctx->cfg.imex.chans);
         xclose(ctx->mnt_ns);
         return (-1);
 }
@@ -453,6 +471,7 @@ nvc_shutdown(struct nvc_context *ctx)
 
         free(ctx->cfg.root);
         free(ctx->cfg.ldcache);
+        free(ctx->cfg.imex.chans);
         xclose(ctx->mnt_ns);
 
         memset(&ctx->cfg, 0, sizeof(ctx->cfg));