Re: Fwd: [PATCH v4 08/18] cxl: IRQ allocation for guests

[Frederic Barrat <fbarrat@linux.vnet.ibm.com>]

Le 21/02/2016 23:30, Manoj Kumar a écrit :
>> Subject: [PATCH v4 08/18] cxl: IRQ allocation for guests
>> Date: Tue, 16 Feb 2016 22:39:01 +0100
>> From: Frederic Barrat <fbarrat@linux.vnet.ibm.com>
>> To: imunsie@au1.ibm.com, michael.neuling@au1.ibm.com,
>> mpe@ellerman.id.au, linuxppc-dev@lists.ozlabs.org
>>
>> The PSL interrupt is not going to be multiplexed in a guest, so an
>> interrupt will be allocated for it for each context.
>
> Not clear why this is the case. Why cannot the CXL later still
> multiplex this in a guest? Is this a design choice, an
> architectural issue, or the complexity of implementation did
> not warrant this? From an API perspective it would have been
> preferable to not cascade this change down to all consumers,
> and have consumers aware whether they are working in a
> bare-metal or a guest environment.


It was a design choice made by pHyp. We cannot multiplex the PSL 
interrupt with the current pHyp implementation.

But it doesn't affect the API: the behavior of the API specifying the 
number of interrupts for a context is consistent: the driver always 
expects the number of AFU interrupts on bare-metal and in a LPAR. The 
PSL interrupt is never included.

You can see a difference in the maximum number of attachable contexts 
between bare-metal and powerVM (if the limiting factor is the number of 
available interrupts). But there's no guarantee for that at the API level.

   Fred