[PATCH v2 0/1] gpu: nova-core: populate GSP_VF_INFO for vGPU

[PATCH v2 0/1] gpu: nova-core: populate GSP_VF_INFO for vGPU

[Zhi Wang]

This is v2 of a follow-up to the "Boot GSP with vGPU enabled" series
[1]. It provides GSP-RM with the VF topology required for per-VF
interrupt and doorbell event routing.

When vGPU mode is enabled, GspSetSystemInfo reads the SR-IOV topology
through the typed Rust PCI capability API and fills the embedded
GSP_VF_INFO. When vGPU mode is disabled, the structure remains zeroed.

This series depends on the separately posted "[PATCH v7 0/1] Rust PCI
capability infrastructure and SR-IOV support" series [3], which provides
the capability finder and VF BAR iterator used here. The prerequisite
patch ID below identifies the exact dependency used to prepare this
version.

Changes since v1 [2]:
- Adapted to the v7 Rust PCI API by handling the optional capability
  result and consuming the logical VF BAR iterator. (Zhi)
- Made the VF information builder an associated function of
  GspSetSystemInfo. (Danilo [4])
- Removed the unrelated GpuacctPerfmonUtilSamples registration. (Zhi)
- Rebased onto the latest drm-rust-next. (Zhi)

[1] https://lore.kernel.org/nova-gpu/20260709150206.1046839-1-zhiw@nvidia.com/
[2] https://lore.kernel.org/nova-gpu/20260730181404.776028-1-zhiw@nvidia.com/
[3] https://lore.kernel.org/rust-for-linux/20260804161612.776752-1-zhiw@nvidia.com/
[4] https://lore.kernel.org/nova-gpu/DKC83N1YHAGM.3BRAAMYUVIL4Y@kernel.org/

Zhi Wang (1):
  gpu: nova-core: populate GSP_VF_INFO when vGPU is enabled

 drivers/gpu/nova-core/gsp/boot.rs        |  6 ++--
 drivers/gpu/nova-core/gsp/commands.rs    | 15 ++++++++--
 drivers/gpu/nova-core/gsp/fw/commands.rs | 37 +++++++++++++++++++++++-
 3 files changed, 52 insertions(+), 6 deletions(-)


base-commit: 44e7e7f7cffb10a93bb88e7cb59b7b8b3e2deb1c
prerequisite-patch-id: 4ff67e7d1ab069e291277346b23a5a195e751cff
-- 
2.53.0

[PATCH v2 1/1] gpu: nova-core: populate GSP_VF_INFO when vGPU is enabled

[Zhi Wang]

GSP firmware requires VF BAR layout information to correctly route
interrupts and doorbell events to individual virtual functions.
Without this data the firmware cannot set up per-VF event routing.

When vGPU mode is enabled, read the VF topology from the typed PCI
SR-IOV extended capability and populate the GSP_VF_INFO embedded in
the GSP_SYSTEM_INFO command. Keep the VF information builder associated
with GspSetSystemInfo since it is only used to construct that command.

Link: https://lore.kernel.org/nova-gpu/20260730181404.776028-2-zhiw@nvidia.com/
Signed-off-by: Zhi Wang <zhiw@nvidia.com>
---
 drivers/gpu/nova-core/gsp/boot.rs        |  6 ++--
 drivers/gpu/nova-core/gsp/commands.rs    | 15 ++++++++--
 drivers/gpu/nova-core/gsp/fw/commands.rs | 37 +++++++++++++++++++++++-
 3 files changed, 52 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/nova-core/gsp/boot.rs b/drivers/gpu/nova-core/gsp/boot.rs
index 727b8ae4bcb7..c086026dc2d8 100644
--- a/drivers/gpu/nova-core/gsp/boot.rs
+++ b/drivers/gpu/nova-core/gsp/boot.rs
@@ -86,8 +86,10 @@ pub(crate) fn boot(
 
         dev_dbg!(pdev, "RISC-V active? {}\n", gsp_falcon.is_riscv_active(),);
 
-        self.cmdq
-            .send_command_no_wait(bar, commands::SetSystemInfo::new(pdev, chipset))?;
+        self.cmdq.send_command_no_wait(
+            bar,
+            commands::SetSystemInfo::new(pdev, chipset, ctx.vgpu.state()),
+        )?;
         self.cmdq
             .send_command_no_wait(bar, commands::SetRegistry::new(ctx.vgpu.state())?)?;
 
diff --git a/drivers/gpu/nova-core/gsp/commands.rs b/drivers/gpu/nova-core/gsp/commands.rs
index ffc25fd8c47b..11d0c7ec371e 100644
--- a/drivers/gpu/nova-core/gsp/commands.rs
+++ b/drivers/gpu/nova-core/gsp/commands.rs
@@ -41,12 +41,21 @@
 pub(crate) struct SetSystemInfo<'a> {
     pdev: &'a pci::Device<device::Bound>,
     chipset: Chipset,
+    vgpu_state: VgpuState,
 }
 
 impl<'a> SetSystemInfo<'a> {
     /// Creates a new `GspSetSystemInfo` command using the parameters of `pdev`.
-    pub(crate) fn new(pdev: &'a pci::Device<device::Bound>, chipset: Chipset) -> Self {
-        Self { pdev, chipset }
+    pub(crate) fn new(
+        pdev: &'a pci::Device<device::Bound>,
+        chipset: Chipset,
+        vgpu_state: VgpuState,
+    ) -> Self {
+        Self {
+            pdev,
+            chipset,
+            vgpu_state,
+        }
     }
 }
 
@@ -57,7 +66,7 @@ impl<'a> CommandToGsp for SetSystemInfo<'a> {
     type InitError = Error;
 
     fn init(&self) -> impl Init<Self::Command, Self::InitError> {
-        Self::Command::init(self.pdev, self.chipset)
+        Self::Command::init(self.pdev, self.chipset, self.vgpu_state)
     }
 }
 
diff --git a/drivers/gpu/nova-core/gsp/fw/commands.rs b/drivers/gpu/nova-core/gsp/fw/commands.rs
index 6dc31d1bf5ae..71feda9c1e69 100644
--- a/drivers/gpu/nova-core/gsp/fw/commands.rs
+++ b/drivers/gpu/nova-core/gsp/fw/commands.rs
@@ -16,7 +16,8 @@
 use crate::{
     gpu::Chipset,
     gsp::GSP_PAGE_SIZE,
-    num::IntoSafeCast, //
+    num::IntoSafeCast,
+    vgpu::VgpuState, //
 };
 
 use super::bindings;
@@ -29,10 +30,43 @@ pub(crate) struct GspSetSystemInfo {
 static_assert!(size_of::<GspSetSystemInfo>() < GSP_PAGE_SIZE);
 
 impl GspSetSystemInfo {
+    /// Builds the VF topology passed to GSP-RM as part of the system information.
+    fn build_vf_info(
+        dev: &pci::Device<device::Bound>,
+        vgpu_state: VgpuState,
+    ) -> Result<bindings::GSP_VF_INFO> {
+        let VgpuState::Enabled { total_vfs } = vgpu_state else {
+            return Ok(Default::default());
+        };
+
+        let sriov = dev
+            .config_space_extended()?
+            .find_ext_capability::<pci::ExtSriovRegs>()?
+            .ok_or(ENODEV)?;
+
+        let mut vf_bars = sriov.vf_bars()?;
+        let bar0 = vf_bars.next().ok_or(EINVAL)?;
+        let bar1 = vf_bars.next().ok_or(EINVAL)?;
+        let bar2 = vf_bars.next().ok_or(EINVAL)?;
+
+        Ok(bindings::GSP_VF_INFO {
+            totalVFs: u32::from(total_vfs.get()),
+            firstVFOffset: u32::from(kernel::io_read!(sriov, .vf_offset)),
+            FirstVFBar0Address: bar0.address,
+            FirstVFBar1Address: bar1.address,
+            FirstVFBar2Address: bar2.address,
+            b64bitBar0: u8::from(bar0.is_64bit),
+            b64bitBar1: u8::from(bar1.is_64bit),
+            b64bitBar2: u8::from(bar2.is_64bit),
+            ..Zeroable::zeroed()
+        })
+    }
+
     /// Returns an in-place initializer for the `GspSetSystemInfo` command.
     pub(crate) fn init<'a>(
         dev: &'a pci::Device<device::Bound>,
         chipset: Chipset,
+        vgpu_state: VgpuState,
     ) -> impl Init<Self, Error> + 'a {
         type InnerGspSystemInfo = bindings::GspSystemInfo;
         let pci_config_mirror_range = chipset.pci_config_mirror_range();
@@ -54,6 +88,7 @@ pub(crate) fn init<'a>(
             PCIRevisionID: u32::from(dev.revision_id()),
             bIsPrimary: 0,
             bPreserveVideoMemoryAllocations: 0,
+            gspVFInfo: Self::build_vf_info(dev, vgpu_state)?,
             ..Zeroable::init_zeroed()
         });
 
-- 
2.53.0

Re: [PATCH v2 1/1] gpu: nova-core: populate GSP_VF_INFO when vGPU is enabled

[Alexandre Courbot]

On Wed Aug 5, 2026 at 2:00 AM JST, Zhi Wang wrote:
> GSP firmware requires VF BAR layout information to correctly route
> interrupts and doorbell events to individual virtual functions.
> Without this data the firmware cannot set up per-VF event routing.
>
> When vGPU mode is enabled, read the VF topology from the typed PCI
> SR-IOV extended capability and populate the GSP_VF_INFO embedded in
> the GSP_SYSTEM_INFO command. Keep the VF information builder associated
> with GspSetSystemInfo since it is only used to construct that command.
>
> Link: https://lore.kernel.org/nova-gpu/20260730181404.776028-2-zhiw@nvidia.com/
> Signed-off-by: Zhi Wang <zhiw@nvidia.com>

This is looking good IMHO, and actually I would suggest including it
with your PCI extended capabilities' v9: that way, the added SR-IOV
capability gets a user immediately and we can merge both patches in one
go. It is also helpful for review to see the code being used in
practice.

Just one question: we are passing `VgpuState` by value everywhere, which
is ok for now, but is it expected to grow further in the future? Its
name suggests that it could end up storing much more things.

<...>
> diff --git a/drivers/gpu/nova-core/gsp/fw/commands.rs b/drivers/gpu/nova-core/gsp/fw/commands.rs
> index 6dc31d1bf5ae..71feda9c1e69 100644
> --- a/drivers/gpu/nova-core/gsp/fw/commands.rs
> +++ b/drivers/gpu/nova-core/gsp/fw/commands.rs
> @@ -16,7 +16,8 @@
>  use crate::{
>      gpu::Chipset,
>      gsp::GSP_PAGE_SIZE,
> -    num::IntoSafeCast, //
> +    num::IntoSafeCast,
> +    vgpu::VgpuState, //
>  };
>  
>  use super::bindings;
> @@ -29,10 +30,43 @@ pub(crate) struct GspSetSystemInfo {
>  static_assert!(size_of::<GspSetSystemInfo>() < GSP_PAGE_SIZE);
>  
>  impl GspSetSystemInfo {
> +    /// Builds the VF topology passed to GSP-RM as part of the system information.
> +    fn build_vf_info(
> +        dev: &pci::Device<device::Bound>,
> +        vgpu_state: VgpuState,
> +    ) -> Result<bindings::GSP_VF_INFO> {
> +        let VgpuState::Enabled { total_vfs } = vgpu_state else {
> +            return Ok(Default::default());

Maybe `Zeroable::zeroed()` for consistency with the `Enabled` path below?