Test: Refactored performance tests on rxtxapp

[awilczyns]

Purpose
Introduces a dual-host session capacity sweep framework for MTL ST2110-20 performance testing. Uses binary search to find the maximum number of video sessions that maintain target FPS on a single lcore (or multi-core), across resolutions (1080p/4K/8K), framerates (25/29/50/59fps), with optional ST2022-7 redundancy and DSA offloading.

File	Lines	What it does
collect_platform_info.py	373	Gathers SW/HW platform info (OS, kernel, CPU, NIC firmware, BIOS, memory) from remote hosts via SSH
generate_report.py	859	Generates self-contained HTML performance report by parsing pytest log dirs — sweep results, topology, platform tables, per-session FPS/throughput details
dsa.py	330	DSA device management — auto-discovery via lspci, NUMA-aware selection, VFIO binding with IOMMU group validation
performance_monitoring.py	451	FPS validation (warmup/cooldown filtering, 99% tolerance), frame counting, throughput extraction, CpuCoreMonitor class (background mpstat sampler)
test_vf_perf_dualhost.py	911	Main test file — 4 parametrized test functions (test_tx, test_rx, test_tx_redundant, test_rx_redundant), each with 48 combinations. Binary-search sweep with crash recovery, inter-iteration VF cleanup, and sweep summary logging

Key Modifications
• conftest.py — Major expansion: topology fixture override for extra fields (dsa_device, build_path, per-host mtl_path), nic_port_list now sets up redundant VFs on 2nd NIC, new collect_platform_config autouse fixture, log_session routes perf tests to dedicated log folder, auto-base_performance marker for 1080p/59fps tests
• rxtxapp.py — Config generation refactored for ST2022-7 redundant mode (dual interfaces, dual IP arrays), queue count overrides (rx_queues_cnt/tx_queues_cnt), netsniff capture_time fix
• application_base.py — start_process changed to background=True (was blocking), get_executable_path handles direct paths
• nicctl.py — Per-host MTL paths, new reset_vfio_bindings() (crash recovery) and ensure_vfio_bound() (inter-iteration cleanup)
• gen_config.py — New --media_path and --dsa_device CLI args, extra_info field in topology config
• execute.py — New helpers: kill_all_rxtxapp(), stop_remote_process(), read_remote_log()
• universal_params.py — New params: redundant, nic_port_r, source_ip_r, destination_ip_r, rx_queues_cnt, tx_queues_cnt, disable_migrate, tx_copy_once
• ip_pools.py — New tx_r/rx_r redundant IP address pools
pytest.ini — Registered performance and base_performance markers

[DawidWesierski4]

I would rename it to DMA, this is the general dma handling isn't it ?

[DawidWesierski4]

i don't think we wan't to do this this way there is a pytest fixture for input files ramdisk creaton,

Here you just put the files randomly outside of that framework if you do so the perf will and is limited via the disk operations wchich is suboptimal

IF there is no file we exit with error and thats fine
Please configure the ramdisk way

[DawidWesierski4]

i geniully think that the most importat would be the ramdisk fixture workaround fixes
and the genconfig -> i would try to simplify it, we don't wnat to add anything there

we should propably also add copyright everywhere i see a lot of these are missing