Re: [PATCH] x86, setup: add __flush_tlb() for Intel Quark X1000

["Bryan O'Donoghue" <pure.logic@nexus-software.ie>]

On 26/09/14 16:00, Bryan O'Donoghue wrote:
>> +    /*
>> +     * Locate the page directory and flush the TLB.
>> +     *
>> +     * On Quark X1000 CPUs we still have the PGE bit incorrectly set
>> +     * due to a processor erratum, so __flush_tlb_all() is not yet
>> +     * doing what it says.  Fortunately we have a cr3 flush here,
>> +     * which is what is needed in this processor to flush TLBs, so
>> +     * there's no need to add a Quark X1000 quirk here.
>> +     *
>> +     * early_init_intel will unset the X86_FEATURE_PGE flag later
>> +     * and __flush_tlb_all() will flush via cr3
>> +     */
>> +    __flush_tlb();
>>
>> With the extra __flush_tlb(); on the end.
>
> Scatch that.
>
> ACK Ong Boon Leong's patch

Hmm.

At the risk of contradicting myself in public I had a think about this 
on the drive home and ...

As I see it - the reload @ CR3 should have flushed the TLB. That's what 
the documentation says.

Right now with the proposed patch from Ong Boon Leong the code will be

load_cr3(swapper_pg_dir);
   __flush_tlb_all();
__flush_tlb();

While there may be no *harm* in adding an extra __flush_tlb(); - there's 
not much *sense* in it either.

If however there's a strong feeling that an explicit __flush_tlb() is 
required in Quark's case then - I believe the code should revert to the 
original logic from the Quark BSP namely.

+       if (boot_cpu_data.x86 == 5 && boot_cpu_data.x86_model == 9)
+               __flush_tlb();
+       else
+               __flush_tlb_all();


This was Peter's initial suggestion in any case and as I see it the 
right-thing-to-do rather than have another reduntant __flush_tlb() for 
everybody who's not Quark.

How does that make sense ?

I can't imagine it's preferable to do a __flush_tlb(); on *x86 for the 
sake of skipping an if/else

I'll resubmit that now with Ong Boon's commentary.


Best,
BOD