Skip to content

Honor RMDisableNoncontigAlloc in the PMA vidmem path - #1305

Open
matte-schwartz wants to merge 1 commit into
NVIDIA:mainfrom
matte-schwartz:honor-rmdisablenoncontigalloc-pma
Open

Honor RMDisableNoncontigAlloc in the PMA vidmem path#1305
matte-schwartz wants to merge 1 commit into
NVIDIA:mainfrom
matte-schwartz:honor-rmdisablenoncontigalloc-pma

Conversation

@matte-schwartz

Copy link
Copy Markdown

The RMDisableNoncontigAlloc registry key clears
bAllowNoncontiguousAllocation, but the flag is only consulted by vidmemAllocResources() and heapAlloc() on the legacy heap path. _vidmemPmaAllocate() decides contiguity from the client's PHYSICALITY attribute alone and silently retries a failed contiguous allocation as noncontiguous, so the key has no effect on PMA-managed framebuffer heaps, which covers every modern GPU.

Consult the flag in the PMA path as well. When noncontiguous allocation is disabled, allocate contiguously regardless of the requested physicality and skip the noncontiguous retry, so the allocation fails cleanly instead of silently degrading.

The RMDisableNoncontigAlloc registry key clears
bAllowNoncontiguousAllocation, but the flag is only consulted by
vidmemAllocResources() and heapAlloc() on the legacy heap path.
_vidmemPmaAllocate() decides contiguity from the client's PHYSICALITY
attribute alone and silently retries a failed contiguous allocation as
noncontiguous, so the key has no effect on PMA-managed framebuffer
heaps, which covers every modern GPU.

Consult the flag in the PMA path as well. When noncontiguous
allocation is disabled, allocate contiguously regardless of the
requested physicality and skip the noncontiguous retry, so the
allocation fails cleanly instead of silently degrading.
@CLAassistant

CLAassistant commented Aug 18, 2026

Copy link
Copy Markdown

CLA assistant check
All committers have signed the CLA.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants