Re: [PATCH] x86/tsc: RFC: re-synchronize TSCs to boot cpu TSC

[Roland Scheidegger]

Am 09.12.2016 um 10:59 schrieb Thomas Gleixner:
> On Fri, 9 Dec 2016, Roland Scheidegger wrote:
>>
>> I saw some system lockups though:
>> When doing a cold boot, this kernel never managed to boot up. The last
>> message seen is:
>> x86: Booting SMP configuration:
>> .... node  #0, CPUs:        #1
> 
> Weird. That really would be interesting to figure out what goes wrong
> there. What bothers me is that we don't see something like this:
> 
>> [    0.172334] TSC ADJUST differs: Reference CPU0: -577421768610 CPU1:
>> -577423766270
> 
> Can you please apply the debug patch below and provide the output ?
Ok, this is the output (minus some typos maybe...):

x86: Booting SMP configuration:
.... node #0, CPUs:     #1
TSC ADJUST: CPU1: -2806491604
TSC source sync 0 -> 1 runs 3
TSC ADJUST differs: Reference CPU0: -2805503200 CPU1: -2806491604
TSC ADJUST synchronize: Reference CPU0: -2805503200 CPU1: -2806491604
TSC target sync skip
TSC source sync skipped

And that's it.


> 
> 
>> [    0.094492] x86: Booting SMP configuration:
>> [    0.094534] .... node  #0, CPUs:        #1
>> [    0.172334] TSC ADJUST differs: Reference CPU0: -577421768610 CPU1:
>> -577423766270
> 
> What on earth is this BIOS doing? That's a couple of minutes back in time.
Looks like that's only after a reset, after a cold boot the numbers are
significantly smaller (1 sec or so?). Though the difference between
these two is nearly a million cycles too, so might be similar between
min and max for all.


> 
> And the difference between the max and min adjust value is 2050932 cycles.
> 
>> Without the patches on cold boot it just was as expected:
>> [    0.093700] x86: Booting SMP configuration:
>> [    0.093737] .... node  #0, CPUs:        #1
>> [    0.174304] TSC synchronization [CPU#0 -> CPU#1]:
>> [    0.174375] Measured 1837188 cycles TSC warp between CPUs, turning
>> off TSC clock.
> 
> Not surprising given the above numbers.
> 

Hope that helps,

Roland