Re: [PATCH 5/7] gpu: nova-core: Add support for managing GSP falcon interrupts

[Joel Fernandes <joelagnelf@nvidia.com>]

On 10/22/2025 2:47 AM, Alexandre Courbot wrote:
> On Tue Oct 21, 2025 at 3:55 AM JST, Joel Fernandes wrote:
>> Add support for managing GSP falcon interrupts. These are required for
>> GSP message queue interrupt handling.
>>
>> Also rename clear_swgen0_intr() to enable_msq_interrupt() for
>> readability.
> 
> Let's make this "also" its own patch as it is a different thing.

Sure, will do. To clarify, my intention was to make this a "prerequisite" patch
series that's why it has all the goodies in a single series. However, it is
probably less confusing to have them independently sent as you pointed out.

>> Signed-off-by: Joel Fernandes <joelagnelf@nvidia.com>
>> ---
>>  drivers/gpu/nova-core/falcon/gsp.rs | 26 +++++++++++++++++++++++---
>>  drivers/gpu/nova-core/gpu.rs        |  2 +-
>>  drivers/gpu/nova-core/regs.rs       | 10 ++++++++++
>>  3 files changed, 34 insertions(+), 4 deletions(-)
>>
>> diff --git a/drivers/gpu/nova-core/falcon/gsp.rs b/drivers/gpu/nova-core/falcon/gsp.rs
>> index f17599cb49fa..6da63823996b 100644
>> --- a/drivers/gpu/nova-core/falcon/gsp.rs
>> +++ b/drivers/gpu/nova-core/falcon/gsp.rs
>> @@ -22,11 +22,31 @@ impl FalconEngine for Gsp {
>>  }
>>  
>>  impl Falcon<Gsp> {
>> -    /// Clears the SWGEN0 bit in the Falcon's IRQ status clear register to
>> -    /// allow GSP to signal CPU for processing new messages in message queue.
>> -    pub(crate) fn clear_swgen0_intr(&self, bar: &Bar0) {
>> +    /// Enable the GSP Falcon message queue interrupt (SWGEN0 interrupt).
>> +    #[expect(dead_code)]
>> +    pub(crate) fn enable_msgq_interrupt(&self, bar: &Bar0) {
>> +        regs::NV_PFALCON_FALCON_IRQMASK::alter(bar, &Gsp::ID, |r| r.set_swgen0(true));
>> +    }
>> +
>> +    /// Check if the message queue interrupt is pending.
>> +    #[expect(dead_code)]
>> +    pub(crate) fn has_msgq_interrupt(&self, bar: &Bar0) -> bool {
>> +        regs::NV_PFALCON_FALCON_IRQSTAT::read(bar, &Gsp::ID).swgen0()
>> +    }
>> +
>> +    /// Clears the message queue interrupt to allow GSP to signal CPU
>> +    /// for processing new messages.
>> +    pub(crate) fn clear_msgq_interrupt(&self, bar: &Bar0) {
>>          regs::NV_PFALCON_FALCON_IRQSCLR::default()
>>              .set_swgen0(true)
>>              .write(bar, &Gsp::ID);
>>      }
>> +
>> +    /// Acknowledge all pending GSP interrupts.
>> +    #[expect(dead_code)]
>> +    pub(crate) fn ack_all_interrupts(&self, bar: &Bar0) {
>> +        // Read status and write the raw value to IRQSCLR to clear all pending interrupts.
>> +        let status = regs::NV_PFALCON_FALCON_IRQSTAT::read(bar, &Gsp::ID);
>> +        regs::NV_PFALCON_FALCON_IRQSCLR::from(u32::from(status)).write(bar, &Gsp::ID);
>> +    }
> 
> I think this can be a bit more generic than that: all interrupts for all
> falcons are handled the same way, so we shouldn't need to write
> `enable`, `clear`, and other methods for each.
> 
> So the first step should probably be a generic `impl<E> Falcon<E>` block
> that provides base methods for specialized blocks to reuse. It could
> work with just the index of the bit corresponding to the interrupt to
> enable/clear, but if we want to involve the type system we could also
> define a `FalconInterrupt` trait with an associated type for the engine
> on which this interrupt is valid, and its bit index as an associated
> const.
> 
> But I suspect that the set of interrupts is going to be pretty standard,
> so maybe we can use the standard nomenclature for the generic methods
> (i.e. SWGEN0), and have dedicated methods for specialized units where
> relevant.

Good point. I'll start with the `impl<E> Falcon<E>` block as it is simple.
I also suspect as you do that the layout should be standard across falcons.
Thanks for the suggestion and reviews.
thanks,

 - Joel