Waiting for bufdaemon

[Rainer Hurling]

Am 17.01.21 um 10:49 schrieb Konstantin Belousov:
> On Sun, Jan 17, 2021 at 10:37:18AM +0100, Rainer Hurling wrote:
>> Am 17.01.21 um 05:33 schrieb Konstantin Belousov:
>>> On Sat, Jan 16, 2021 at 07:41:01PM +0100, Rainer Hurling wrote:
>>>> During another shutdown after heavy usage of the box, the following
>>>> messages were also seen:
>>>>
>>>>
>>>> [...]
>>>> Syncing disks, vnodes remaining... 22 EFI rt_settime call faulted, error 14
>>>> efirtc0: CLOCK_SETTIME error 14
>>>
>>> This means that BIOS code faulted during RTC settime call.  I doubt that
>>> it is related.
>>>
>>> On the other hand, it is good that the onfault EFI RT code got tested finally.
>>>
>>
>> Thanks for clarification :)
>>
>>
>> Any chance of getting a fix for the AMD CPUs in the foreseeable future?
>>
>> Or should I revert commit 9e680e4005b7 on affected boxes until further
>> notice (as a workaround)?
> 
> I am working on it, no ETA.
> 
> Interesting point would be to check on machines of other testers,
> if the following hides the problem.
> 
> diff --git a/sys/x86/x86/tsc.c b/sys/x86/x86/tsc.c
> index 85924df98312..5700a8ca98e5 100644
> --- a/sys/x86/x86/tsc.c
> +++ b/sys/x86/x86/tsc.c
> @@ -639,7 +639,7 @@ init_TSC_tc(void)
>  	 *   on Intel, and MFENCE;RDTSC on AMD.
>  	 * - For really old CPUs, just use RDTSC.
>  	 */
> -	if ((cpu_vendor_id == CPU_VENDOR_AMD ||
> +	if (false && (cpu_vendor_id == CPU_VENDOR_AMD ||
>  	    cpu_vendor_id == CPU_VENDOR_HYGON) &&
>  	    CPUID_TO_FAMILY(cpu_id) >= 0x17) {
>  		tsc_timecounter.tc_get_timecount = shift > 0 ?
> 

I tried the above patch on a Ryzen 3950X.

After reboot (again with long waiting for bufdaemon and more) the box
had some heavy load with several jobs (llvm on Poudriere, building
qt5-webengine, libreoffice and some more in parallel). All went fine.

Afterwards I rebooted again. This time, the shutdown was fast  without
any unusual delays :)

Many thanks!