Re: [PATCH v4 14/20] gpu: nova-core: add falcon register definitions and base code

[Danilo Krummrich <dakr@kernel.org>]

On Wed, May 21, 2025 at 03:45:09PM +0900, Alexandre Courbot wrote:
> Add the common Falcon code and HAL for Ampere GPUs, and instantiate the
> GSP and SEC2 Falcons that will be required to boot the GSP.

Maybe add a few more words about the architectural approach taken here?

> +/// Valid values for the `size` field of the [`crate::regs::NV_PFALCON_FALCON_DMATRFCMD`] register.
> +#[repr(u8)]
> +#[derive(Debug, Default, Copy, Clone, PartialEq, Eq)]
> +pub(crate) enum DmaTrfCmdSize {
> +    /// 256 bytes transfer.
> +    #[default]
> +    Size256B = 0x6,

Can we use a constant from `regs` to assign this value? Or is *this* meant to be
the corresponding constant?

> +}

I wonder what's the correct thing to do for enum variants that do *not* have an
arbitrary value, but match a specific register value in general.

Should those be part of the `regs` module?

> +    /// Wait for memory scrubbing to complete.
> +    fn reset_wait_mem_scrubbing(&self, bar: &Bar0) -> Result {
> +        util::wait_on(Duration::from_millis(20), || {

I general, I think there can be quite a lot of parameters such timeouts can
depend on, e.g. chipset, firmware version, etc.

I think it could make sense to establish a rule for the project that for such
timeouts we require a dedicated `// TIMEOUT: ` comment that mentions the worst
case scenario, which we derived this timeout value from.

> +    /// Perform a DMA write according to `load_offsets` from `dma_handle` into the falcon's
> +    /// `target_mem`.
> +    ///
> +    /// `sec` is set if the loaded firmware is expected to run in secure mode.
> +    fn dma_wr(
> +        &self,
> +        bar: &Bar0,
> +        dma_handle: bindings::dma_addr_t,
> +        target_mem: FalconMem,
> +        load_offsets: FalconLoadTarget,
> +        sec: bool,
> +    ) -> Result {
> +        const DMA_LEN: u32 = 256;
> +
> +        // For IMEM, we want to use the start offset as a virtual address tag for each page, since
> +        // code addresses in the firmware (and the boot vector) are virtual.
> +        //
> +        // For DMEM we can fold the start offset into the DMA handle.
> +        let (src_start, dma_start) = match target_mem {
> +            FalconMem::Imem => (load_offsets.src_start, dma_handle),
> +            FalconMem::Dmem => (
> +                0,
> +                dma_handle + load_offsets.src_start as bindings::dma_addr_t,

We should make this a method of CoherentAllocation, such that we can get a
boundary check on the offset calculation.

For this purpose dma_rw() should also have the `F: FalconFirmware<Target = E>`
generic I think.

(No worries about the dependencies; I can create a shared tag for the DMA
patches and merge it into the nova tree, such that it doesn't block this
series.)

> +            // Wait for the transfer to complete.
> +            util::wait_on(Duration::from_millis(2000), || {

Yeah, I really think some timeout justification would be nice.

> +/// Hardware Abstraction Layer for Falcon cores.
> +///
> +/// Implements chipset-specific low-level operations. The trait is generic against [`FalconEngine`]
> +/// so its `BASE` parameter can be used in order to avoid runtime bound checks when accessing
> +/// registers.
> +pub(crate) trait FalconHal<E: FalconEngine>: Sync {
> +    // Activates the Falcon core if the engine is a risvc/falcon dual engine.
> +    fn select_core(&self, _falcon: &Falcon<E>, _bar: &Bar0) -> Result<()> {
> +        Ok(())
> +    }
> +
> +    /// Returns the fused version of the signature to use in order to run a HS firmware on this
> +    /// falcon instance. `engine_id_mask` and `ucode_id` are obtained from the firmware header.
> +    fn get_signature_reg_fuse_version(

Unless the method increases a reference count, please don't use the 'get'
prefix.

> +        &self,
> +        falcon: &Falcon<E>,
> +        bar: &Bar0,
> +        engine_id_mask: u16,
> +        ucode_id: u8,
> +    ) -> Result<u32>;
> +
> +    // Program the boot ROM registers prior to starting a secure firmware.
> +    fn program_brom(&self, falcon: &Falcon<E>, bar: &Bar0, params: &FalconBromParams)
> +        -> Result<()>;
> +}
> +
> +impl Chipset {
> +    /// Returns a boxed falcon HAL adequate for this chipset.
> +    ///
> +    /// We use a heap-allocated trait object instead of a statically defined one because the
> +    /// generic `FalconEngine` argument makes it difficult to define all the combinations
> +    /// statically.
> +    ///
> +    /// TODO: replace the return type with `KBox` once it gains the ability to host trait objects.

I think we can do this for v5. :-)