Re: [v3] x86/tsc: Unset TSC_KNOWN_FREQ and TSC_RELIABLE flags on Intel Bay Trail SoC

[Andy Shevchenko]

On Wed, Jan 29, 2020 at 05:59:10PM +0200, Andy Shevchenko wrote:
> On Wed, Jan 29, 2020 at 05:53:53PM +0200, Andy Shevchenko wrote:
> > On Wed, Jan 29, 2020 at 04:13:39PM +0100, Thomas Gleixner wrote:
> > > Hans de Goede <hdegoede@xxxxxxxxxx> writes:
> > > > On 29-01-2020 15:14, Andy Shevchenko wrote:
> > > >>> The only one which is possibly suspicious here is this line:
> > > >>>
> > > >>>   * 0111:   25 * 32 /  9  =  88.8889 MHz
> > > >>>
> > > >>> The SDM says 88.9 MHz for this one.
> > > 
> > > I trust math more than the SDM :)
> > > 
> > > >> Anyway it seems need to be fixed as well.
> > > >> 
> > > >> Btw, why we are mentioning 20 / 6 and 28 / 6 when arithmetically
> > > >> it's the same as 10 / 3 and 14 / 3?
> > > >
> > > > I copied the BYT values from Thomas' email and I guess he did not
> > > > get around to simplifying them, I'll use the simplified versions
> > > > for my patch.
> > > 
> > > Too tired, too lazy :)
> > > 
> > > Andy, can you please make sure that people inside Intel who can look
> > > into the secrit documentation confirm what we are aiming for?
> > > 
> > > Ideally they should provide the X-tal frequency and the mult/div pair
> > > themself :)
> > 
> > So, I don't have access to the CPU core documentation (and may be will not be
> > given), nevertheless I dug a bit to what I have for Cherrytrail. So, the XTAL
> > is 19.2MHz, which becomes 100MHz and 1600MHz by some root PLL, then, the latter
> > two frequencies are being used by another PLL to provide a reference clock (*)
> > to PLL which derives CPU clock.
> 
> > *) According to colleagues of mine it's a fixed rate source.
> 
> One more thing.
> 
> Depends on SKU it may be 400MHz, 320MHz, 200MHz or 333MHz.

Aha, found better precision for the last one, 333.33MHz.

> (I guess these values should be kinda references in the table)
> 
> > That's all what I have.

-- 
With Best Regards,
Andy Shevchenko