On 05.08.2020 18:25, Andrew Cooper wrote:
> On 05/08/2020 15:54, Jan Beulich wrote:
>> On 05.08.2020 16:18, Andrew Cooper wrote:
>>> --- a/xen/arch/x86/cpu/intel.c
>>> +++ b/xen/arch/x86/cpu/intel.c
>>> @@ -396,14 +396,14 @@ static void intel_log_freq(const struct cpuinfo_x86
>>> *c)
>>>
>>> val *= ebx;
>>> do_div(val, eax);
>>> - printk("CPU%u: TSC: %uMHz * %u / %u = %LuMHz\n",
>>> + printk("CPU%u: TSC: %u Hz * %u / %u = %Lu Hz\n",
>>> smp_processor_id(), ecx, ebx, eax, val);
>> For this one I wonder whether ecx wouldn't better be scaled down to
>> kHz, and val down to MHz.
>
> That depends on whether we will lose precision in the process.
I don't think losing the last three digits for the base clock and
the last six ones of the calculated value would do any harm at all.
All it would do (imo) is to make the numbers better readable (due
less counting, and hence less possible counting mistakes).
> In principle we can, given ecx's unit of Hz, so I'd be tempted to leave
> it as is.
Well, okay.
Jan
