On 08/02/2017 03:20 PM, Rik van Riel wrote:
> On Wed, 2017-08-02 at 14:59 -0400, Nitesh Narayan Lal wrote:
>>>> -struct hypervisor_pages hypervisor_pagelist[MAX_FGPT_ENTRIES];
>>>> +struct hypervisor_pages hypervisor_pagelist[MAX_FGPT_ENTRIES -
>>>> 1];
>>>> +
>>>> +static void empty_hyperlist(void)
>>>> +{
>>>> + int i = 0;
>>>> +
>>>> + while (i < MAX_FGPT_ENTRIES - 1) {
>>> MAX_FGPT_ENTRIES in-place of 'MAX_FGPT_ENTRIES - 1' here
>>> and at similar other places?
>> This is because CPU local list has a total of 1000 entries
>> (MAX_FGPT_ENTRIES) where as CPU global list has 999 entries. If you
>> see
>> the arch_free_page_slowpath() and consider a situation where there
>> are
>> 1000 entries of singly allocated free pages in cpu-local list i.e.,
>> none
>> of them are re-allocated. While adding them to the cpu global list
>> when
>> cpu local list index reaches to 1000 the outer loop will terminate
>> due
>> to which cpu global list index will never reach to 1000 and
>> compress_hyperlist()/make_hypercall() will never be called.
> Can you explain why the hypervisor_pagelist
> is smaller than the cpu local list?
>
> This makes no sense to me. Why are they not
> the same size?
>
> That would certainly make the code easier to read.
I am sorry. There is no point of complicating things when it could be
done in a simpler way.
I will make the required changes in my next patch-set.
--
Regards
Nitesh
Attachment:
signature.asc
Description: OpenPGP digital signature
