tegrakernel/kernel/nvidia/Documentation/devicetree/bindings/platform/tegra/nvidia,carmel-pmu.txt

47 lines
1.3 KiB
Plaintext

NVIDIA Carmel Uncore PMU
========================
This driver supports hardware perfmon event counting for the 4 L2 clusters
and for the L3. The supported events are standard ARM L2/L3 event numbering,
but shifted and masked with a unit number for the sake of granular unit
tracking. The event format is EEU, where 'EE' is the standard ARM event number
and U is the unit (0-3 for the L2s, 4 for the L3).
These standard ARM events are supported:
Event ID Type
L2D_CACHE 0x16 L2
L2D_CACHE_REFILL 0x17 L2
L2D_CACHE_WB 0x18 L2
BUS_ACCESS 0x19 L3
BUS_CYCLES 0x1D L3
L3D_CACHE_ALLOCATE 0x29 L3
L3D_CACHE_REFILL 0x2A L3
L3D_CACHE 0x2B L3
L3D_CACHE_WB 0x2C L3
L2D_CACHE_LD 0x50 L2
L2D_CACHE_ST 0x51 L2
L2D_CACHE_REFILL_LD 0x52 L2
L2D_CACHE_REFILL_ST 0x53 L2
L2D_CACHE_REFILL_VICTIM 0x56 L2
For example, to track L2 cache events (0x16) on clusters 0 and 2 as well as L3
events (0x2b):
perf stat -a -e r160,r162,r2b4
Required properties:
- compatible : "nvidia,carmel-pmu"
- interrupts : single hardware-specified interrupt
- interrupt-affinity : this driver must be pinned to Carmel core 0 (cpu@0)
in order to correctly field PMU interrupts from the L2/L3 units.
Example:
carmel-pmu {
compatible = "nvidia,carmel-pmu";
interrupts = <0 365 0x4>;
interrupt-affinity = <&{/cpus/cpu@0}>;
};