Pcie transfer performance#126

Draft

amd-vserbu wants to merge 23 commits into

Xilinx:devfrom

amd-vserbu:perf/pcie-transfer-performance

Commits on Jun 17, 2026

Added slash_compat.h force include and fix from_timer backport issue for RHEL 9.8
amd-vserbu
committed
driver: large-page qdma transfers, NoC channel steering, libqdma patches
amd-vserbu
committed
libvrtd: hugepage host buffers and granule-aware partial sync
amd-vserbu
committed
smi: validate bandwidth modes with raw SLASH and stock qdma backends
amd-vserbu
committed
packaging: ship libqdma patches, add deps, harden install test, ignore /tmp
amd-vserbu
committed
driver: guard libqdma pr_fmt under force-included compat header
amd-vserbu
committed
driver: verify qdma host-profile readback and add tunable hugepage descriptor size
amd-vserbu
committed
smi: add validate buffer placement, channel allocation, and bandwidth knobs
amd-vserbu
committed
docs: document new validate placement and bandwidth options
amd-vserbu
committed
Added 4k|2M explicit page specification
amd-vserbu
committed
driver: add qdma registered-buffer abi with pinned, pre-dma-mapped transfers
amd-vserbu
committed
libslash: add qdma buffer register/unregister/transfer wrappers and mock
amd-vserbu
committed
tests: cover qdma registered-buffer kselftest abi
amd-vserbu
committed
docs: document qdma registered-buffer abi
amd-vserbu
committed
smi: add --ring-size-index and use registered buffers for raw transfers
amd-vserbu
committed
docs: document validate --ring-size-index option
amd-vserbu
committed
vrtd: plumb mm-channel selection through buffer open
amd-vserbu
committed
driver+libslash: added transfer performance hint
amd-vserbu
committed
vrt/vrtd: use new performance buffer ioctl api
amd-vserbu
committed
qdma stack: change policy from dual-channel to the more complex v80 policy
amd-vserbu
committed
driver+vrt+smi: drop libqdma sg/channel patches, make transfers 4 KiB-only
amd-vserbu
committed
Changed API to kernel allocated-buffers
amd-vserbu
committed
vrtd: track client ownership on raw buffers and close per-owner
amd-vserbu
committed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Pcie transfer performance#126

Pcie transfer performance#126
amd-vserbu wants to merge 23 commits into
Xilinx:devfrom
amd-vserbu:perf/pcie-transfer-performance

Commits on Jun 17, 2026

Added slash_compat.h force include and fix from_timer backport issue for RHEL 9.8

driver: large-page qdma transfers, NoC channel steering, libqdma patches

libvrtd: hugepage host buffers and granule-aware partial sync

smi: validate bandwidth modes with raw SLASH and stock qdma backends

packaging: ship libqdma patches, add deps, harden install test, ignore /tmp

driver: guard libqdma pr_fmt under force-included compat header

driver: verify qdma host-profile readback and add tunable hugepage descriptor size

smi: add validate buffer placement, channel allocation, and bandwidth knobs

docs: document new validate placement and bandwidth options

Added 4k|2M explicit page specification

driver: add qdma registered-buffer abi with pinned, pre-dma-mapped transfers

libslash: add qdma buffer register/unregister/transfer wrappers and mock

tests: cover qdma registered-buffer kselftest abi

docs: document qdma registered-buffer abi

smi: add --ring-size-index and use registered buffers for raw transfers

docs: document validate --ring-size-index option

vrtd: plumb mm-channel selection through buffer open

driver+libslash: added transfer performance hint

vrt/vrtd: use new performance buffer ioctl api

qdma stack: change policy from dual-channel to the more complex v80 policy

driver+vrt+smi: drop libqdma sg/channel patches, make transfers 4 KiB-only

Changed API to kernel allocated-buffers

vrtd: track client ownership on raw buffers and close per-owner