Re: [PATCH] viridian: Fix bank count counting

["Teddy Astie" <teddy.astie@xxxxxxxxxx>]

Le 01/12/2025 à 18:48, Andrew Cooper a écrit :
> On 01/12/2025 4:29 pm, Teddy Astie wrote:
>> hv_vpset_nr_banks() incorrectly counts the number of bank by using 
>> hweight64()
>> instead of flsl(). This for instance problematic in case only the second bank
>> is selected (i.e >64 vCPUs, where here hweight64 gives 1), causing only the 
>> first
>> bank to be checked (non-valid) and the second (meaningful) one to be skipped.
>>
>> Fixes: b4124682db6e ("viridian: add ExProcessorMasks variants of the flush 
>> hypercalls")
>> Signed-off-by: Teddy Astie <teddy.astie@xxxxxxxxxx>
>> ---
>>   xen/arch/x86/hvm/viridian/viridian.c | 2 +-
>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/xen/arch/x86/hvm/viridian/viridian.c 
>> b/xen/arch/x86/hvm/viridian/viridian.c
>> index 90e749ceb5..f5e526241d 100644
>> --- a/xen/arch/x86/hvm/viridian/viridian.c
>> +++ b/xen/arch/x86/hvm/viridian/viridian.c
>> @@ -601,7 +601,7 @@ static DEFINE_PER_CPU(union hypercall_vpset, 
>> hypercall_vpset);
>>
>>   static unsigned int hv_vpset_nr_banks(struct hv_vpset *vpset)
>>   {
>> -    return hweight64(vpset->valid_bank_mask);
>> +    return fls64(vpset->valid_bank_mask);
>>   }
>>
>>   static int hv_vpset_to_vpmask(const struct hv_vpset *in, paddr_t bank_gpa,
>
> https://learn.microsoft.com/en-us/virtualization/hyper-v-on-windows/tlfs/datatypes/hv_vp_set#processor-set-example
>
> This example is poor (it has several errors in the text), but the final
> statement says fairly clearly that the banks are tightly packed and not
> contiguous.
>
> Therefore, hweight64() the correct calculation, and AFAICT, the single
> use of this helper in Xen correct.
>

Ok, I see what you mean, such that only set bit of the mask imply a bank
in the "BankContents". Which means that 2 set bits (regardless of their
position) means 2 banks (regardless of which one).

But that looks contradictory with the second sentence
> The total set of virtual processors is split up into chunks of 64, known as a 
> “bank”. For example,
> processors 0-63 are in bank 0, 64-127 are in bank 1, and so on.

Even though, all the banks and "BankContents" are different things.

Interestingly, all open-source users of this hypercall I found fills the
mask with 1s (up to the maximum vCPU they want to target). So in such
case the layout is fully contiguous at the end.

> ~Andrew
>



--
Teddy Astie | Vates XCP-ng Developer

XCP-ng & Xen Orchestra - Vates solutions

web: https://vates.tech