On Sat Aug 22, 2026 at 10:54 AM JST, John Hubbard wrote:
quoted hunk ↗ jump to hunk
GSP-RM allocates the whole radix3 page table region in one block and
zeroes it before writing entries. The booter reads each level a full
page at a time when it walks the table, so every byte of a level page is
firmware-visible whether or not it holds a valid entry.
The driver allocates each level separately and writes only the valid
entries, so the last page of each level is only partly written. The
booter reads the whole page regardless, and a non-zero word in the
unwritten remainder is indistinguishable from an entry.
Allocate each level zeroed, sized to a whole number of GSP pages, and
write the entries into it.
Assisted-by: Cursor:claude-opus-5
Reviewed-by: Timur Tabi <redacted>
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
---
drivers/gpu/nova-core/firmware/radix3.rs | 50 ++++++++++++++++--------
1 file changed, 33 insertions(+), 17 deletions(-)
diff --git a/drivers/gpu/nova-core/firmware/radix3.rs b/drivers/gpu/nova-core/firmware/radix3.rs
index b60611c7bea0..b0630fd96c01 100644
--- a/drivers/gpu/nova-core/firmware/radix3.rs
+++ b/drivers/gpu/nova-core/firmware/radix3.rs
@@ -67,22 +67,14 @@ pub(crate) fn new<'a>(
Ok(try_pin_init!(Self {
data <- SGTable::new(dev, data, DataDirection::ToDevice, GFP_KERNEL),
level2 <- {
- VVec::<u8>::with_capacity(
- data.iter().count() * core::mem::size_of::<u64>(),
- GFP_KERNEL,
- )
- .map_err(|_| ENOMEM)
- .and_then(|level2| map_into_lvl(&data, level2))
- .map(|level2| SGTable::new(dev, level2, DataDirection::ToDevice, GFP_KERNEL))?
+ let level2 = build_lvl(&data)?;
+
+ SGTable::new(dev, level2, DataDirection::ToDevice, GFP_KERNEL)
Or alternatively:
level2 <- {
build_lvl(&data)
.map(|l2| SGTable::new(dev, l2, DataDirection::ToDevice, GFP_KERNEL))
},
I don't feel strongly about this, I just think it describes the flow a
bit better when there are no fleeting local variables. Feel free to keep
your version if you prefer it.
(also applicable to `level1` below)