Nouveau - Dec 2025 - [PATCH 18/31] gpu: nova-core: Hopper/Blackwell: add needs_large_reserved_mem()

Hopper, Blackwell and later need to reserve more memory than previous
GPUs did.

Signed-off-by: John Hubbard <jhubbard at nvidia.com>
---
 drivers/gpu/nova-core/gpu.rs | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/drivers/gpu/nova-core/gpu.rs b/drivers/gpu/nova-core/gpu.rs
index 624bbc2a54e8..c0473ef8ac47 100644
--- a/drivers/gpu/nova-core/gpu.rs
+++ b/drivers/gpu/nova-core/gpu.rs
@@ -123,6 +123,11 @@ pub(crate) const fn arch(&self) -> Architecture {
             | Self::GB207 => Architecture::Blackwell,
         }
     }
+
+    #[expect(dead_code)]
+    pub(crate) fn needs_large_reserved_mem(&self) -> bool {
+        matches!(self.arch(), Architecture::Hopper | Architecture::Blackwell)
+    }
 }
 
 // TODO
-- 
2.52.0

On Tue, 2025-12-02 at 21:59 -0800, John Hubbard wrote:
> +
> +??? #[expect(dead_code)]
> +??? pub(crate) fn needs_large_reserved_mem(&self) -> bool {
> +??????? matches!(self.arch(), Architecture::Hopper |
Architecture::Blackwell)
> +??? }
Wouldn't it be cleaner to return the actual amount of memory needed, instead
of just saying, "Hey, I
need more, whatever more is"?

         let frts_offset = if !resume {
             let mut frts_reserved_size = if chipset.needs_large_reserved_mem()
{
-                0x220000 // heap_size_non_wpr for Hopper/Blackwell+
+                crate::fb::calc_non_wpr_heap_size(chipset)
             } else {
                 total_reserved_size
             };

Maybe have calc_non_wpr_heap_size() be a HAL function?