shim and vxdna: update to support unlimited dev no allocation

[wendyliang25]

See details description in this PR: https://github.com/amd/xdna-driver/pull/1233

[Copilot]

Pull request overview

Updates the shim and vxdna virt-accel implementation to support “unlimited” device BO allocations by enabling a dynamically expandable device heap made up of multiple 64MB chunks (up to a configured maximum).

Changes:

• Add DEV_HEAP VA-range reservation and fixed-address mappings in vxdna to place heap chunks contiguously in userspace VA.
• Add an expandable heap buffer in the shim with a max size (512MB) and retry/expand behavior on allocation retry errors.
• Adjust ccmd exception handling/logging behavior to avoid QEMU-side cascading failures.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 6 comments.

Show a summary per file

File	Description
src/vxdna/src/vaccel_amdxdna.h	Adds heap VA tracking in vxdna_context, BO type accessor, and changes ccmd error-wrap behavior.
src/vxdna/src/vaccel_amdxdna.cpp	Implements heap VA reservation + MAP_FIXED mapping for DEV_HEAP and updates error response behavior.
src/shim/kmq/pcidev.cpp	Creates an expandable DEV_HEAP BO on first open and retries DEV BO allocation by expanding heap on retry error.
src/shim/buffer.h	Introduces shared heap_page_size constant and new expandable buffer constructor signature.
src/shim/buffer.cpp	Implements expandable buffer constructor and refines expand() behavior for heap growth.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[maxzhen]

Please add comment and clarify locking schema when we need to expand heap.

[wendyliang25]

Please add comment and clarify locking schema when we need to expand heap.

Hi @maxzhen , i have added the comments about the locking:

  // we need to lock when we are trying to allocate DEV BO as
  // it is possible that the heap is being expanded by another thread.

Please check again. thanks

[Copilot]

Pull request overview

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Comments suppressed due to low confidence (1)

src/shim/kmq/pcidev.cpp:19

• Debug.num_heap_pages can be configured to 0, which makes heap_sz become 0 and results in constructing an expandable DEV_HEAP buffer with initial_size == 0 (eventually attempting to create a 0-byte BO). Consider clamping get_heap_num_pages() to at least 1 (or explicitly rejecting 0 with a clear error) so the first-open heap allocation is always valid.

unsigned int
get_heap_num_pages()
{
  static unsigned int num = 0;

  if (!num)
    num = xrt_core::config::detail::get_uint_value("Debug.num_heap_pages", 1);
  return num;

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.