Re: xen 4.14.3 incorrect (~3x) cpu frequency reported

[Jan Beulich <jbeulich@suse.com>]

On 10.01.2022 13:37, Roger Pau Monné wrote:
> On Fri, Jan 07, 2022 at 12:39:04PM +0100, Jan Beulich wrote:
>> @@ -415,16 +416,35 @@ static int64_t __init init_hpet(struct p
>>  
>>      pts->frequency = hpet_rate;
>>  
>> +for(i = 0; i < 16; ++i) {//temp
>>      count = hpet_read32(HPET_COUNTER);
>>      start = rdtsc_ordered();
>>      target = count + CALIBRATE_VALUE(hpet_rate);
>>      if ( target < count )
>>          while ( hpet_read32(HPET_COUNTER) >= count )
>>              continue;
>> -    while ( hpet_read32(HPET_COUNTER) < target )
>> +    while ( (count = hpet_read32(HPET_COUNTER)) < target )
>>          continue;
>>  
>> -    return (rdtsc_ordered() - start) * CALIBRATE_FRAC;
>> +    expired = rdtsc_ordered() - start;
> 
> There's also a window between the HPET read and the TSC read where an
> SMI/NMI could cause a wrong frequency detection.

It occurred to me only while preparing a draft change that this is an
issue not only for the tail pair of reads, but also for the initial pair.
Hence I'm now actually making a helper function to be used in both
places. This slightly sub-optimal as it means that it's not as easy to
make use of the last HPET read in the loop above, but I guess we can
tolerate that one extra read in preference over more complicated logic.

Jan