Different targets will have different ESI affinity For CQ due to difference in CPU cluster configurations. Hence same affinity won't give max performance for all targets. Also currently UFS MCQ ESI cpu mask are populated from device tree nodes for assigning the cpu mask. Add provision to populate and configure ESI affinity for completion queue for UFS MCQ. We affine ESI for completion queue among top 2 clusters and focus for assigning gold clusters request to Prime cluster limited to 3 queues. This helps to affine ESI for CQ queue as below - assigning ESI affinity for request on gold core to prime core - assigning ESI affinity for request on other core to gold core. This helps in UFS random IO KPI improvement of around ~15% for cliffs. Change-Id: I582babff928292e3c9d6b44431a1a6c1a7020027 Co-developed-by: Nitin Rawat <quic_nitirawa@quicinc.com> Signed-off-by: Nitin Rawat <quic_nitirawa@quicinc.com> Signed-off-by: Manish Pandey <quic_mapa@quicinc.com> |
||
---|---|---|
.. | ||
core | ||
host | ||
Kconfig | ||
Makefile |