[SOLVED] Kernel version 5.14 breaks vmix

[Nexus]

Tested with kernel versions 5.14 and 5.15. Version 5.10 works and apparently so does 5.11 since the latest patches were specifically for that version, so the culprit could be somewhere between versions 5.12-5.14. There were no packages for the EOL versions 5.12 or 5.13 on the MX repositories so those went untested.

ossinfo:

Code: Select all

Version info: OSS 4.2 (b 2021/202112231941) (0x00040100) OSS_HG
Platform: Linux/x86_64 5.15.0-2mx-amd64 #1 SMP Debian 5.15.5-1~mx19+1 (2021-11-29)

dmesg:

Code: Select all

[   13.363559] osscore: This processor architecture is not compatible with vmix (info=0) - Not enabled.

So apparently it's the floating point support check function failing in vmix_core.c:

Code: Select all

void
vmix_core_init (oss_device_t *osdev)
{
#ifdef CONFIG_OSS_VMIX_FLOAT
  int check;
/*
 * Check that the processor is compatible with vmix (has proper FP support).
 */
   if ((check = oss_fp_check ()) <= 0)
    {
      vmix_disabled = 1;
      cmn_err (CE_WARN,
          "This processor architecture is not compatible with vmix (info=%d) - Not enabled.\n",
          check);
      return;
    }
#endif
}


[EDIT]: Never mind.

Until this gets patched, using --config-vmix=FIXEDPOINT when running configure works fairly well.

[seawright]

Hi Nexus,
I think it may just be a reporting error and floating point vmix works just fine.
Could you compile the drivers without any configuration options and (as root) apply the following patch:

Code: Select all

--- /usr/lib/oss/build/osscore_c.bak   2022-01-21 17:36:44.967159829 +0000
+++ /usr/lib/oss/build/osscore.c   2022-01-21 17:42:41.068614420 +0000
@@ -1941,7 +1941,7 @@
 
   local_save_flags (flags_reg);
   if (!(flags_reg & FLAGS_ID))
-    return 0;
+    return 1;
 
 #define CPUID_FXSR   (1<<24)
 #define CPUID_SSE   (1<<25)


Followed by:
sh /usr/lib/oss/build/install.sh
soundoff
soundon

Check that the warning is no longer present in syslog/dmesg and that vmix is present and functions as normal.

[Nexus]

Thanks, Clive. That patch seems to have fixed the issue.