NVIDIA
gdrcopy_sanity fails with gdrdrv loaded
We are experiencing a failure with gdrdrv across multiple systems (x86 and arm).
With gdrdrv-2.5.1-1 installed, the kernel module loads, but gdrcopy_sanity fails. The only log of any kind is in dmesg at load, [root@illyad log]# modprobe gdrdrv [691802.796281] gdrdrv:gdrdrv_init:loading gdrdrv version 2.5 built for proprietary NVIDIA driver [691802.796286] gdrdrv:gdrdrv_init:device registered with major number 490 [691802.796288] gdrdrv:gdrdrv_init:dbg traces disabled, info traces disabled [691802.796290] gdrdrv:gdrdrv_init:Persistent mapping will be used
[root@illyad log]# modinfo gdrdrv filename: /lib/modules/4.18.0-553.16.1.el8_10.x86_64/extra/gdrdrv.ko.xz version: 2.5 description: GDRCopy kernel-mode driver built for proprietary NVIDIA driver license: Dual MIT/GPL author: [email protected] rhelversion: 8.10 srcversion: 628A1EE2623290E9768AEB0 depends: nv-p2p-dummy name: gdrdrv vermagic: 4.18.0-553.16.1.el8_10.x86_64 SMP mod_unload modversions sig_id: PKCS#7 signer: DKMS module signing key sig_key: 3D:B7:A3:06:14:1E:2D:71:02:9A:5E:B0:CC:BE:08:B5:20:31:91:BA sig_hashalgo: sha256 signature: 3E:75:F0:7B:F4:19:F4:8E:70:7F:8F:83:A6:5F:F1:91:D4:25:59:06: 7E:BA:94:0E:8A:3C:8F:3E:DA:86:CE:B5:3B:CA:E2:0C:A1:A4:D4:CF: 8C:D2:20:F5:61:E2:E9:B3:38:6B:09:FB:53:91:76:B0:73:C2:13:9E: 0A:6B:21:1A:E2:84:F2:E2:5D:9F:FD:26:8C:87:54:E4:93:DA:91:7D: AB:2F:26:A9:3A:96:8D:EF:DB:2B:70:03:69:0C:49:C3:61:CE:B3:8D: B8:7D:81:11:2A:AD:04:E8:96:17:B7:DB:82:AF:82:05:69:87:7C:44: D4:0D:BE:2E:F2:D5:E7:F8:2E:7F:85:58:48:68:E5:B6:23:1C:EE:2F: D2:62:A3:0E:F2:72:D8:C5:49:11:F3:BB:B4:7D:BB:F7:46:54:F9:BF: C5:42:2E:EA:AA:3C:2D:5C:61:40:7C:39:3E:DB:FF:DC:04:0A:B2:EA: AD:DD:6D:E5:7D:42:38:B5:BE:F1:02:88:E1:EB:54:81:75:42:58:BF: 1C:CF:F5:09:32:BF:14:E7:4A:32:01:C8:C5:A3:83:70:DF:D1:98:D6: 5A:59:18:38:9A:48:A0:A2:B6:A0:9C:A6:05:EB:C0:0E:08:0E:48:89: 93:20:CE:12:4C:B3:4E:82:D2:89:C5:59:D8:AE:B0:2B parm: dbg_enabled:enable debug tracing (int) parm: info_enabled:enable info tracing (int) parm: use_persistent_mapping:use persistent mapping instead of traditional (non-persistent) mapping (int)
Any attempt to use it fails, [root@illyad log]# gdrcopy_sanity gdr_open error: Is gdrdrv driver installed and loaded? gdr_open error: Is gdrdrv driver installed and loaded?
The module is most certainly loaded but there is no /dev/gdr*, [root@illyad log]# lsmod | grep gdr gdrdrv 24576 0 nvidia 103858176 8 nvidia_uvm,gdrdrv,nvidia_modeset [root@illyad log]# ls /dev/gdr* ls: cannot access '/dev/gdr*': No such file or directory
Nothing has deigned to lower itself to writing any sort of meaningful error output anywhere (dmesg, syslog, journald, messages, etc) which leaves me few options in providing more useful information here. SELinux is not enforcing which eliminates the most common suspect when it comes to mysterious file-[allegedly]-doesn't-exist problems.
Hi @ekeever1,
How did you load the gdrdrv driver? You should use either the service script (sudo systemctl start gdrcopy) or the sudo insmod.sh script that comes with the repository.
