【devel-7.0】Add MPAM support · Pull Request !7635 · anolis/cloud-kernel

bugzilla: https://bugzilla.openanolis.cn/show_bug.cgi?id=36714

This PR backports the MPAM resctrl stack from NV-Kernels (54 patches) plus 3 upstream MPAM bug fixes. The series brings MBM counter assignment, free-running/assignable monitors, schema format refactor, MBA per-arch rounding, MB_HLIM (max_lim) support, iommu_group assignment, debugfs observability, and assorted reliability fixes for arm_mpam.

backport from: https://github.com/NVIDIA/NV-Kernels/tree/26.04_linux-nvidia-bos

## Patchset

| Patchset | Patch Subject | Resolve | Upstream / Source Link |
|---|---|---|---|
| **upstream MPAM bug fixes** | arm_mpam: Fix monitor instance selection when checking for hardware NRDY | - | https://git.kernel.org/torvalds/c/1ef2a89584b7 |
| | arm_mpam: Fix false positive assert failure during mpam_disable() | - | https://git.kernel.org/torvalds/c/f1caff3335ea |
| | arm_mpam: Check whether the config array is allocated before destroying it | - | https://git.kernel.org/torvalds/c/6ccbb613b42a |
| **MBM counter assignment & monitors** | arm_mpam: resctrl: pick classes for use as mbm counters | - | https://github.com/NVIDIA/NV-Kernels/commit/d12ec41cde40 |
| | arm_mpam: resctrl: Pre-allocate free running monitors | - | https://github.com/NVIDIA/NV-Kernels/commit/11de6cffe0ef |
| | arm_mpam: resctrl: Pre-allocate assignable monitors | - | https://github.com/NVIDIA/NV-Kernels/commit/aa7555911398 |
| | arm_mpam: resctrl: Add kunit test for ABMC/CDP interactions | - | https://github.com/NVIDIA/NV-Kernels/commit/8453b8fc97f6 |
| | arm_mpam: resctrl: Add resctrl_arch_config_cntr() for ABMC use | - | https://github.com/NVIDIA/NV-Kernels/commit/1c0f936ee009 |
| | arm_mpam: resctrl: Add resctrl_arch_rmid_read() and resctrl_arch_reset_rmid() | - | https://github.com/NVIDIA/NV-Kernels/commit/21a0edac8cc5 |
| | arm_mpam: resctrl: Add resctrl_arch_cntr_read() & resctrl_arch_reset_cntr() | - | https://github.com/NVIDIA/NV-Kernels/commit/07e445fb2c4a |
| **dom_data_exit() race fixes** | fs/resctrl: Avoid a race with dom_data_exit() and closid_num_dirty_rmid[] | - | https://github.com/NVIDIA/NV-Kernels/commit/9fc15b5f3212 |
| | fs/resctrl: Avoid a race with dom_data_exit() and rmid_ptrs[] | - | https://github.com/NVIDIA/NV-Kernels/commit/d7c2bfdd4257 |
| **arm_mpam debugfs observability** | debugfs: Add helpers for creating cpumask entries in debugfs | - | https://github.com/NVIDIA/NV-Kernels/commit/be63b37a935d |
| | arm_mpam: Add debugfs entries to show the MSC/RIS the driver discovered | - | https://github.com/NVIDIA/NV-Kernels/commit/919bf27649a5 |
| | arm_mpam: Add force-disable debugfs trigger | - | https://github.com/NVIDIA/NV-Kernels/commit/0c6a605638f6 |
| | arm_mpam: Expose the number of NRDY retries in debugfs | - | https://github.com/NVIDIA/NV-Kernels/commit/e0da6a8d52cb |
| **MBA rounding per-arch** | arm_mpam: Add resctrl_arch_round_bw() | - | https://github.com/NVIDIA/NV-Kernels/commit/a4231552d569 |
| | fs/resctrl,x86/resctrl: Factor mba rounding to be per-arch | - | https://github.com/NVIDIA/NV-Kernels/commit/efea2576c13e |
| **arm_mpam misc** | arm_mpam: Split the locking around the mon_sel registers | - | https://github.com/NVIDIA/NV-Kernels/commit/5ce93d2d841e |
| | arm_mpam: Allow the maximum partid to be overridden from the command line | - | https://github.com/NVIDIA/NV-Kernels/commit/2a3468a1b843 |
| | arm_mpam: Allow MSC to be forced to have an unknown location | - | https://github.com/NVIDIA/NV-Kernels/commit/ebc43b8c83e1 |
| **this_is_not_abi mount option** | fs/resctrl: Add this_is_not_abi mount option | Conflict in `fs/resctrl/internal.h` `struct rdt_fs_context` (HWDRC `enable_hwdrc` already present in ANCK from ANBZ #35585); resolved by keeping `enable_hwdrc` alongside the new `enable_abi_playground` field. | https://github.com/NVIDIA/NV-Kernels/commit/a4370607dbe4 |
| **iommu/kobject prereqs (for iommu_group assignment)** | kobject: Add kset_get_next_obj() to allow a kset to be walked | - | https://github.com/NVIDIA/NV-Kernels/commit/d34afaf8edef |
| | iommu: Add helpers to retrieve iommu_groups by id or kobject | - | https://github.com/NVIDIA/NV-Kernels/commit/2cf654d1bcf2 |
| | iommu: Add helper to retrieve iommu kset | - | https://github.com/NVIDIA/NV-Kernels/commit/205a3429710f |
| | iommu/arm-smmu-v3: Register SMMU capabilities with MPAM | - | https://github.com/NVIDIA/NV-Kernels/commit/bfa7580f6c10 |
| | iommu/arm-smmu-v3: Add mpam helpers to query and set state | - | https://github.com/NVIDIA/NV-Kernels/commit/5c6ee0c8f6b6 |
| | iommu: Add helpers to get and set the QoS state | - | https://github.com/NVIDIA/NV-Kernels/commit/25fb0768a14b |
| **iommu_group assignment** | arm_mpam: resctrl: Add iommu helpers to get/set the partid and pmg | - | https://github.com/NVIDIA/NV-Kernels/commit/8716fdd41bef |
| | fs/resctrl: Add support for assigning iommu_groups to resctrl groups | - | https://github.com/NVIDIA/NV-Kernels/commit/3602b35d8ea3 |
| **Overflow handler tunable** | x86/resctrl: Add stub to allow other architecture to disable monitor overflow | - | https://github.com/NVIDIA/NV-Kernels/commit/567de3911e8d |
| | arm_mpam: resctrl: Determine if any exposed counter can overflow | - | https://github.com/NVIDIA/NV-Kernels/commit/618eec86f2f5 |
| | fs/restrl: Allow the overflow handler to be disabled | - | https://github.com/NVIDIA/NV-Kernels/commit/65148a97523c |
| **Type cleanup** | fs/resctrl: Uniform data type of component_id/domid/id/cache_id | - | https://github.com/NVIDIA/NV-Kernels/commit/e5a4f2f5181b |
| **cmax/cmin + fract16** | arm_mpam: Allow cmax/cmin to be configured | - | https://github.com/NVIDIA/NV-Kernels/commit/c4f4cbd81175 |
| | arm_mpam: Rename mbw conversion to 'fract16' for code re-use | - | https://github.com/NVIDIA/NV-Kernels/commit/fe0cb8dd5f64 |
| **Schema format refactor** | fs/resctrl: Group all the MBA specific properties in a separate struct | Conflict in `include/linux/resctrl.h`; ANCK has `bool hwdrc` in `struct resctrl_membw` from ANBZ #35585 (`cd12f1b29284`). Resolved conservatively by leaving `hwdrc` in `struct resctrl_membw` (its original ANCK location); the NV source patch only moves NV-defined fields (`mba_sc`/`mb_map`/`delay_linear`/`arch_needs_linear`/`throttle_mode`) into the new `struct resctrl_mba`, so no rdtgroup.c changes are needed. | https://github.com/NVIDIA/NV-Kernels/commit/cd4001b628f0 |
| | fs/resctrl: Abstract duplicate domain test to a helper | - | https://github.com/NVIDIA/NV-Kernels/commit/7bcced8cc98a |
| | fs/resctrl: Move MBA supported check to parse_line() instead of parse_bw() | - | https://github.com/NVIDIA/NV-Kernels/commit/81444f8241e5 |
| | fs/resctrl: Rename resctrl_get_default_ctrl() to include resource | Context-only shift caused by previous `resctrl_mba` split; same logical change as source. | https://github.com/NVIDIA/NV-Kernels/commit/4f4cdb0ccf90 |
| | fs/resctrl: Add a schema format to the schema, allowing it to be different | - | https://github.com/NVIDIA/NV-Kernels/commit/125c77a8cfa3 |
| | fs/resctrl: Use schema format to check the resource is a bitmap | - | https://github.com/NVIDIA/NV-Kernels/commit/c102337e914c |
| | fs/resctrl: Add specific schema types for 'range' | - | https://github.com/NVIDIA/NV-Kernels/commit/032e23347616 |
| | arm_mpam: resctrl: Convert MB resource to use percentage | - | https://github.com/NVIDIA/NV-Kernels/commit/96636b152b15 |
| | fs/resctrl: Remove 'range' schema format | - | https://github.com/NVIDIA/NV-Kernels/commit/8927d5abf172 |
| | x86/resctrl: Move over to specifying MBA control formats | - | https://github.com/NVIDIA/NV-Kernels/commit/eedef6c6a71b |
| | fs/resctrl: Add additional files for percentage and bitmap controls | - | https://github.com/NVIDIA/NV-Kernels/commit/3ea879a9537e |
| | fs/resctrl: Add fflags_from_schema() for files based on schema format | - | https://github.com/NVIDIA/NV-Kernels/commit/cbc23872375a |
| | fs/resctrl: Expose the schema format to user-space | - | https://github.com/NVIDIA/NV-Kernels/commit/6abd323412e2 |
| | fs/resctrl: Add L2 and L3 'MAX' resource schema | - | https://github.com/NVIDIA/NV-Kernels/commit/94d9955a2d3c |
| **arm_mpam glue** | arm_mpam: resctrl: Add the glue code to convert to/from cmax | - | https://github.com/NVIDIA/NV-Kernels/commit/44cd5e901a74 |
| **arm_mpam reliability** | arm_mpam: Avoid MSC teardown for the SW programming errors | - | https://github.com/NVIDIA/NV-Kernels/commit/1e4229deb565 |
| | arm_mpam: Handle CPU-less numa nodes | - | https://github.com/NVIDIA/NV-Kernels/commit/e598d1bcfe49 |
| | arm_mpam: Include all associated MSC components during domain setup | - | https://github.com/NVIDIA/NV-Kernels/commit/1f642b98973c |
| **MPAM RIS reset** | resctrl/mpam: reset RIS by applying explicit default config | - | https://github.com/NVIDIA/NV-Kernels/commit/8d09daf4ea27 |
| **MB_HLIM / MBA max_lim** | resctrl: add MB_HLIM resource ID and schema type | - | https://github.com/NVIDIA/NV-Kernels/commit/c9039b4c3c6d |
| | resctrl: wire MB_HLIM schemata parsing and group setup | - | https://github.com/NVIDIA/NV-Kernels/commit/1133f1e51768 |
| | resctrl/mpam: probe MPAMF_MBW_IDR MAX_LIM and hardlim_rw | - | https://github.com/NVIDIA/NV-Kernels/commit/ca37d5c8e00a |
| | resctrl/mpam: track MBW max hard-limit in config | - | https://github.com/NVIDIA/NV-Kernels/commit/d2acb9656c20 |
| | resctrl/mpam: bind MB_HLIM resource to MBA MPAM class | - | https://github.com/NVIDIA/NV-Kernels/commit/00999657322b |
| | resctrl: add membw.mb_max_lim and arch_has_mb_max_lim | - | https://github.com/NVIDIA/NV-Kernels/commit/bb32f97fc61e |
| | resctrl/mpam: populate MBA mb_max_lim from MPAM probe | - | https://github.com/NVIDIA/NV-Kernels/commit/01d8a8024e15 |
| | resctrl: add MBA max_lim sysfs and visibility from init | - | https://github.com/NVIDIA/NV-Kernels/commit/50a85e171736 |
| | Documentation: resctrl: document max_lim and MB_HLIM for MPAM MBA | - | https://github.com/NVIDIA/NV-Kernels/commit/816240a7fa16 |
| **L3 monitoring fix** | Fix mbm_L3_assign and mon_local_bytes | - | https://github.com/NVIDIA/NV-Kernels/commit/b35ada977bef |

**Skipped upstream patches (NRDY series):** the 3-patch upstream NRDY series (`4387970bbd84`, `ccad6001be5c`, `1ef2a89584b7`) introduces a quirk for non-architectural NRDY behavior; the first two are dropped because the `mon_sel` locking model in the NV-Kernels stack already serializes monitor reads, and only the final fix (`1ef2a89584b7`) is needed and ported as patch #1.

**Total: 65 patches (3 upstream + 62 NV-Kernels), 2 conflicts resolved, 62 patch-id-identical to source.**

# MPAM resctrl Driver Test Documentation

## Overview

This document describes the test infrastructure for the ARM MPAM (Memory Partitioning and Monitoring) resctrl driver, targeting `mpam_devices.c` and `mpam_resctrl.c` in `drivers/resctrl/`.

Tests run on an ARM Fixed Virtual Platform (FVP) with GCOV instrumentation, using a custom initramfs that exercises all accessible resctrl interfaces. A **mock MSC** (Memory System Component) module extends hardware capability coverage beyond what the FVP natively provides.

## Architecture

```
┌─────────────────────────────────────────────────────────┐
│  Host (aarch64 build server)                           │
│  ┌─────────────────────────────────────────────────┐    │
│  │  FVP (Base_RevC_AEMvA)                          │    │
│  │  ┌──────────────────────────────────────────┐    │    │
│  │  │  Kernel (CONFIG_GCOV_KERNEL=y)           │    │    │
│  │  │  ├── MPAM driver (subsys_initcall)       │    │    │
│  │  │  │   ├── MSC 0: FVP L3 @ 0x2B400000     │    │    │
│  │  │  │   │   └── CPOR only, 1 RIS            │    │    │
│  │  │  │   └── MSC 1: Mock @ 0x8E000000        │    │    │
│  │  │  │       └── Full capabilities, 2 RIS     │    │    │
│  │  │  ├── mpam_mock_msc.c (early_initcall)    │    │    │
│  │  │  └── ACPI MPAM table (2 MSC nodes)       │    │    │
│  │  │                                           │    │    │
│  │  │  initramfs/init.sh (132 test cases)       │    │    │
│  │  │  └── GCOV export via serial (base64)      │    │    │
│  │  └──────────────────────────────────────────┘    │    │
│  └─────────────────────────────────────────────────┘    │
│                                                         │
│  Post-processing: extract_gcov.py → gcov → coverage     │
└─────────────────────────────────────────────────────────┘
```

### Key Components

| Component | Path | Purpose |
|-----------|------|---------|
| Mock MSC module | `drivers/resctrl/mpam_mock_msc.c` | Built-in, fills RAM at 0x8E000000 with full-capability MSC registers |
| ACPI table generator | `acpi-overlay/gen_acpi_tables.py` | Produces MPAM table with 2 MSC nodes (FVP + Mock) |
| Test script | `rootfs/initramfs/init.sh` | 132 test cases covering resctrl sysfs interface |
| GCOV extractor | `extract_gcov.py` | Extracts .gcda files from FVP serial log |
| FVP launcher | `boot_fvp.sh` | Configures and launches FVP with correct memory layout |

### Boot Sequence

1. **early_param** `mpam_mock_msc`: `memblock_mark_nomap(0x8E000000, 64K)` — excludes mock region from linear map
2. **early_initcall** `mpam_mock_msc_init()`: ioremaps the nomap region, writes MSC register defaults, iounmaps
3. **subsys_initcall** `mpam_msc_driver_init()`: registers platform driver, ACPI parses 2 MSC nodes
4. **ACPI probe**: Both MSCs are probed → `mpam_ris_hw_probe()` detects all capabilities on mock MSC
5. **cpuhp**: `mpam_enable_once()` merges features from both MSCs → resctrl registered
6. **init.sh**: Runs 132 tests against `/sys/fs/resctrl/`, exports GCOV data, calls `poweroff -f`

## Mock MSC Design

### Why a Mock MSC?

The FVP provides only 1 MSC with L3 cache + CPOR capability. This limits coverage to ~60% because paths for CCAP, MBW, PRI, MSMON (CSU/MBWU), multi-RIS, and full enable/disable cycles are unreachable.

The mock MSC simulates a **full-featured MSC** with MEMORY locator type, enabling:
- MBW (memory bandwidth) schemata and monitoring
- CCAP (cache capacity) partitioning
- PRI (priority) partitioning
- MSMON CSU (cache occupancy) and MBWU (bandwidth usage) monitors
- Multi-RIS props merge (\_\_props\_mismatch paths)
- Free-running MBWU monitors (64 monitors avoids ABMC mode)

### Register Configuration

| Register | Value | Capability |
|----------|-------|-----------|
| MPAMF_IDR | HAS_CCAP \| HAS_CPOR \| HAS_MBW \| HAS_PRI \| HAS_MSMON \| HAS_ESR, PARTID_MAX=15, PMG_MAX=3, RIS_MAX=1 | All feature bits |
| MPAMF_AIDR | MAJOR=1, MINOR=1 | MPAM v1.1 |
| MPAMF_CPOR_IDR | CPBM_WD=16 | 16-bit portion bitmap |
| MPAMF_CCAP_IDR | CMAX_WD=16, HAS_SOFTLIM, HAS_CMIN, HAS_CASSOC, CASSOC_WD=4 | Full cache capacity |
| MPAMF_MBW_IDR | BWA_WD=16, HAS_MAX, HAS_MIN, HAS_PROP, HAS_PBM, BWPBM_WD=16, MAX_LIM=0 | Full bandwidth |
| MPAMF_PRI_IDR | HAS_INTPRI, HAS_DSPRI, WD=4, 0_IS_LOW | Internal + dispatch priority |
| MPAMF_MSMON_IDR | MSMON_CSU \| MSMON_MBWU | Both monitor types |
| MPAMF_CSUMON_IDR | NUM_MON=64, HAS_XCL | 64 cache monitors |
| MPAMF_MBWUMON_IDR | NUM_MON=64, HAS_LONG, HAS_RWBW | 64 bandwidth monitors, free-running |

### ACPI Table Structure

Two MSC nodes in MPAM table:

```
MSC 0 (FVP native):
  base_address = 0x2B400000, mmio_size = 0x10000
  1 resource_node: PROCESSOR_CACHE locator, ris_index=0

MSC 1 (Mock):
  base_address = 0x8E000000, mmio_size = 0x10000
  2 resource_nodes:
    RIS 0: MEMORY locator, proximity_domain=0
    RIS 1: MEMORY locator, proximity_domain=0
```

Two RIS entries on the mock MSC trigger `__vmsc_props_mismatch()` → `__props_mismatch(alias=true)` merge paths, covering ~33 additional lines in mpam_devices.c.

## Test Cases (132 total)

### Section 1: Core resctrl Operations (41 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| mount | `mount -t resctrl resctrl /r` | `mpam_resctrl_setup`, filesystem mount |
| info_L3 | Read L3 cbm_mask, num_closids, domain | `resctrl_arch_get_config`, info files |
| schemata | `grep L3: schemata` | Schema parsing |
| info_detail | Read min_cbm_bits, num_closids | Detail info files |
| mb_info | Read MB info (min_bandwidth, bandwidth_gran) | Mock MSC MB resource info |
| mkdir | `mkdir /r/g1` | Group creation, `__allocate_component_cfg` |
| cbm_f / cbm_full / cbm_min | Write various CBM values | `resctrl_arch_update_one` CPOR path |
| cfg_nochange / cfg_change | Write+read schemata | Config idempotency, update path |
| reject0 / reject_wide / reject_noncontig | Invalid CBM writes | Error handling in schemata write |
| mb_write | `echo "MB:0=50"` | `resctrl_arch_update_one` MBW_MAX path |
| iso_AB | Write 2 groups, check isolation | Multi-group schemata |
| task_g1 / task_g2 | Write PID to tasks | Task assignment, `mpam_thread_switch` |
| mode | Read mode file | Mode attribute |
| size | Read size file | Size calculation |
| cpus | Read cpus_list | CPU affinity |
| mongrp | Create mon_groups subdirectory | Monitor group creation |
| csu_ctrl / csu_default / csu_sub | Read llc_occupancy | `mpam_msmon_read` CSU path |
| mbm_read / mbm_default | Read mbm_total_bytes | `__ris_msmon_read` MBWU path |
| debugfs | Read debugfs files | `mpam_debugfs_setup` |
| rmdir_simple / rmdir_g1 / rmdir_g2 | Remove groups | `__destroy_component_cfg` |
| remount_reset | Remount, verify default schemata | Reset on mount |

### Section 2: Advanced Configuration (30 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| enospc | Exhaust CLOSIDs, expect ENOSPC | Allocation limit handling |
| reuse | mkdir after rmdir, reuse CLOSID | CLOSID recycling |
| lifecycle | Full create→config→monitor→destroy | End-to-end lifecycle |
| interleave | Interleaved config between 2 groups | Concurrent group configuration |
| mon_cfg_change | Monitor read after config change | Config→monitor interaction |
| cycle2_cfg / cycle2_mon | Second mount cycle with config | Re-initialization |
| rewrite3 | Triple schemata rewrite | `mpam_reprogram_ris_partid` repeated |
| cycle3 | Third mount cycle | Stability |
| bulk8 | Create 8 groups simultaneously | Bulk allocation |
| multi_cbm | Write different CBMs to 3 groups | Multi-group config |
| multi_mon | Read monitors from 3 groups | Multi-group monitoring |
| mon10 | 10 repeated monitor reads | Monitor stability |
| mon_interleave | Interleaved config+monitor reads | Race-free config/monitor |
| sweep16 | Sweep 16 CBM values (bit patterns) | Full CBM range |
| highbit8 / midrange5 | Targeted CBM patterns | Edge case bit patterns |
| cycle3_mon / default_schemata | Post-cycle verification | State persistence |
| multi_mongrp | 3 monitor subgroups | `mpam_resctrl_domain_insert` |
| multi_csu | CSU across 3 subgroups | Multi-subgroup monitoring |
| task_mongrp | Task in monitor subgroup | Task→subgroup binding |
| rapid_cfg_mon | Rapid config+monitor sequence | Fast path exercise |
| cycle4 | Fourth mount cycle | Long-term stability |
| debugfs_deep | Deep debugfs traversal | All debugfs entries |
| mon_types | Read all mon_data directories | Monitor type enumeration |

### Section 3: Mock MSC Verification (14 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| mock_probe | Check debugfs msc.1 exists | Mock MSC probe success |
| mock_mb_visible | `grep MB: schemata` | MB resource registered |
| mock_mb_info | Read MB info files | MB info attributes |
| mock_mb_write | Write+read MB% value | `resctrl_arch_update_one` MBW path |
| mock_mb_boundary | Write MB=1, MB=100 | MBW boundary values |
| mock_mb_rewrite | Triple MB% rewrite | MBW reprogram path |
| mock_mbwu | Read mbm_total_bytes from MB domain | MBWU monitor read |
| mock_mbm_local | Read mbm_local_bytes | MBWU local counter |
| mock_mon_types | Enumerate MB mon_data dirs | MB monitor enumeration |
| mock_debugfs | Deep read mock MSC debugfs | Mock MSC debugfs entries |

### Section 4: Additional Coverage (22 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| mb_hlim_write/zero/max | MB hardlim schemata writes | `mpam_read_mbw_max_hardlim` |
| mode_exclusive/shareable | Set ctrl_hw mode | Mode attribute writes |
| last_cmd_status | Read last_cmd_status | Error reporting |
| thread_throttle | Read thread_throttle_mode | Thread throttle attribute |
| info_all | Read all info/* files | Info file enumeration |
| cpu_assign/range/clear | Write cpus, cpus_list | CPU affinity modification |
| ext_mongrp_create/mon/task | External monitor group ops | Monitor group lifecycle |
| ext_mongrp_cleanup | Cleanup monitor groups | Group cleanup |
| err_bad_domain/resource/empty | Invalid schemata writes | Error path handling |
| combined_write | L3+MB combined schemata | Multi-resource write |
| comprehensive_mon | All monitor types read | Comprehensive monitoring |
| abmc_* (6 tests) | ABMC counter assign/read/unassign | ABMC paths |
| mb_sweep | Sweep MB% values 10-100 | MBW range exercise |
| size_detail / def_mon_sweep | Size + default monitor reads | Detail file exercise |

### Section 5: CDP Tests (8 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| cdp_mount | `mount -o cdp` | `resctrl_arch_set_cdp_enabled` |
| cdp_schemata | L3CODE/L3DATA in schemata | CDP schema split |
| cdp_info_code/data | CDP info files | CDP info attributes |
| cdp_write | Write L3CODE/L3DATA values | CDP config write |
| cdp_rapid | Rapid CDP reconfig | CDP fast path |
| cdp_monitor | Read CSU under CDP | `read_mon_cdp_safe` |
| cdp_size | Read size under CDP | CDP size calculation |

### Section 6: Module Parameter Tests (4 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| param_partid_read | Read cmdline_partid_max | Parameter sysfs read |
| param_force_read | Read force_unknown_msc | Parameter sysfs read |
| param_partid_write | Write cmdline_partid_max | `mpam_cmdline_partid_max_set` |
| param_effect | Verify CLOSID change after partid_max | Parameter effect on allocation |

### Section 7: Mount Cycle Tests (3 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| rapid_cycle | 5x mount/mkdir/write/umount | Mount cycle stability |
| cleanup_umount | Final unmount | Clean unmount |
| cleanup_verify | Verify no stale groups | State cleanup verification |

### Section 8: force_unknown_msc Tests (3 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| force_unknown_write | Write to force_unknown_msc param | `mpam_force_unknown_msc_set` |
| force_unknown_read | Read back param value | `mpam_force_unknown_msc_get` |
| force_unknown_multi | Write comma-separated values | Multi-value parameter parse |

### Section 9: CDP Enhanced Tests (4 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| cdp2_mon_csu | CSU read under CDP with task | CDP monitor + task interaction |
| cdp2_mon_mbm | MBM read under CDP | CDP + MBM cross-path |
| cdp2_submon | Monitor subgroup under CDP | CDP + subgroup monitoring |
| cdp2_reconfig | L3CODE/L3DATA rapid reconfig | CDP reprogram path |

### Section 10: CPU Hotplug Tests (6 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| cpu_hotplug | Offline/online CPU 1 | `mpam_cpu_offline/online` |
| hp_mon_offline/online | Monitor read during hotplug | Monitor stability during hotplug |
| hp_mbm_hotplug | MBM read during CPU 2 hotplug | MBWU + hotplug interaction |
| hp2_multioff | Offline CPUs 3,5,7 simultaneously | Multi-CPU offline |
| hp2_multionline | Online all CPUs back | Multi-CPU online |
| hp2_rapid_cycle | Rapid off/on cycle for CPU 4 | Hotplug stress |

### Section 11: Enhanced MBM Tests (4 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| mbm_freerun | Read default MBM (free-running) | Free-running MBWU path |
| mbm_repeat | 5x repeated MBM reads | MBWU read stability |
| mbm_subgrp | MBM from monitor subgroup | Subgroup MBWU |
| mbm_interleave | MBM read between config changes | Config→MBWU interaction |

### Section 12: Mount Option Tests (2 tests)

| ID | Test | What it exercises |
|----|------|-------------------|
| mount_cdpl2 | Mount with -o cdpl2 | L2 CDP attempt |
| closid_halving | Compare CLOSIDs normal vs CDP | CDP CLOSID halving |

### Section 13: force_disable Test (1 test)

| ID | Test | What it exercises |
|----|------|-------------------|
| force_disable | `echo 1 > force_disable` | `mpam_disable`, `mpam_resctrl_exit`, full teardown |

## Coverage Results

### v12 (current)

| File | Lines | Covered | Percentage | Functions |
|------|-------|---------|------------|-----------|
| mpam_devices.c | 1634 | 1264 | **77.36%** | 96/118 (81%) |
| mpam_resctrl.c | 1008 | 742 | **73.61%** | 58/75 (77%) |
| **Combined** | **2642** | **2006** | **75.93%** | **154/193 (80%)** |

### Coverage History

| Version | mpam_devices.c | mpam_resctrl.c | Tests | Key Changes |
|---------|---------------|----------------|-------|-------------|
| v9 | ~60% | ~65% | 78 | Baseline with FVP-only MSC |
| v11 | 75.70% | 73.51% | 120 | Added mock MSC, MB tests, debugfs |
| v12 | 77.36% | 73.61% | 132 | Added 2nd RIS, CDP enhanced, multi-hotplug, force_unknown_msc |

### Structurally Uncoverable Code (~200 lines)

These paths cannot be reached in our test environment due to hardware/architectural constraints:

#### mpam_devices.c (~160 lines)

| Category | Functions | Lines | Reason |
|----------|-----------|-------|--------|
| NVIDIA T241 quirks | `mpam_apply_t241_erratum`, `mpam_enable_quirk_nvidia_t241_1`, `mpam_wa_t241_calc_min_from_max` | ~50 | Requires NVIDIA T241 SoC ID |
| IRQ handlers | `__mpam_irq_handler`, `__setup_ppi`, `mpam_spi/ppi_handler`, `mpam_enable/disable_msc_ecr` | ~55 | Mock MSC has no real IRQ wiring |
| Last-CPU-offline | `mpam_save_mbwu_state` | 35 | Requires all CPUs on an MSC to go offline |
| Driver remove | `mpam_msc_drv_remove` | 9 | Built-in driver, never unloaded |
| Compiler artifacts | `readq_relaxed` (unused inline) | 11 | GCC generates unreachable paths |

#### mpam_resctrl.c (~40 lines)

| Category | Functions | Lines | Reason |
|----------|-----------|-------|--------|
| IOMMU integration | `resctrl_arch_match_iommu_*`, `resctrl_arch_set_iommu_*` | ~24 | No IOMMU + MPAM wiring on FVP |
| L3 topology matching | `topology_matches_l3`, `traffic_matches_l3` | ~52 | FVP topology doesn't trigger these paths |
| ABMC counter config | `__config_cntr`, `resctrl_arch_config_cntr`, `resctrl_arch_cntr_read` | ~27 | ABMC assign path requires specific firmware support |

### Theoretical Maximum Coverage

```
mpam_devices.c: (1634 - 160) / 1634 = ~90%  (current: 77%)
mpam_resctrl.c: (1008 - 40) / 1008  = ~96%  (current: 74%)
```

Remaining gap (~13% for mpam_devices.c, ~22% for mpam_resctrl.c) comes from:
- Error paths (allocation failures, invalid states) that cannot be triggered through sysfs
- `__props_mismatch` mismatch branches — both mock RIS have identical capabilities
- Partially covered functions with unreachable branches (e.g., PCC interface, SMMU locator)
- PARTID narrowing (NRW) paths
- ABMC full lifecycle paths

## Running Tests

### Prerequisites

- ARM FVP (Base_RevC_AEMvA) installed
- TF-A (Trusted Firmware-A) built for FVP
- Kernel built with: `CONFIG_MPAM_MOCK_MSC=y`, `CONFIG_GCOV_KERNEL=y`, `CONFIG_GCOV_PROFILE_RESCTRL=y`
- Kernel cmdline includes: `mpam_mock_msc`

### Build

```bash
# Kernel
cd /disk7/cjk/fvp/build/kernel-fvp
make ARCH=arm64 O=. -j$(nproc) Image

# Initramfs
cd /disk7/cjk/fvp/rootfs
find . | cpio -o -H newc > ../initramfs.cpio

# ACPI tables (if modified)
cd /disk7/cjk/fvp/acpi-overlay
python3 gen_acpi_tables.py

# TF-A (if initrd size changes)
cd /disk7/cjk/fvp/tf-a
make PLAT=fvp BL33=../build/kernel-fvp/arch/arm64/boot/Image ...
```

### Execute

```bash
cd /disk7/cjk/fvp
timeout 1800 ./boot_fvp.sh --no-regen 2>&1 | tee fvp_v12.log
```

### Extract Coverage

```bash
# Extract .gcda files from serial log
python3 extract_gcov.py fvp_v12.log gcov_v12/

# Copy to build directory
cp gcov_v12/.../mpam_devices.gcda build/kernel-fvp/drivers/resctrl/
cp gcov_v12/.../mpam_resctrl.gcda build/kernel-fvp/drivers/resctrl/

# Generate reports
cd build/kernel-fvp/drivers/resctrl
gcov -f -o . ../../../../build/kernel-fvp/source/drivers/resctrl/mpam_devices.c
gcov -f -o . ../../../../build/kernel-fvp/source/drivers/resctrl/mpam_resctrl.c
```

## Known Limitations

1. **Mock MSC is RAM-backed**: All registers are simple read/write. Hardware behaviors like NRdy polling, actual cache partitioning effects, and real bandwidth throttling are not simulated.

2. **Single ACPI table layout**: The mock MSC uses MEMORY locator type. PROCESSOR_CACHE locator paths (PPTT cache_id lookup) are only covered by the real FVP MSC.

3. **No real IRQ delivery**: The mock MSC declares no interrupts. Error IRQ and overflow IRQ handler paths remain uncovered.

4. **No PCC (Platform Communication Channel)**: PCC-based MSC access paths are entirely uncovered. Would require PCCT ACPI table.

5. **Monitor values are static**: Mock MSC monitors return fixed values (0x100/0x200/0x300). Dynamic monitoring behavior (counter increment, overflow, long-counter rollover) is not tested.

6. **FVP timing**: Full test suite takes ~15 minutes on FVP. Timeout must be ≥1800s.

## KUNIT test
  mpam_devices_test_suite:  pass:3   fail:0  skip:0  total:3
  mpam_resctrl_test_suite:  pass:7   fail:0  skip:0  total:7
  Totals:                   pass:119 fail:0  skip:0  total:119

Revert NVIDIA 补丁后，所有 119 个 kunit 测试用例（含 42 个 fract16_to_percent、42 个 percent_to_mbw_max、16 个 mbw_max_to_percent_limits、16 个 roundtrip_stability 等子测试）全部 PASS，0 fail，0 skip。

## Attention
 NVIDIA: SAUCE: resctrl/mpam: reset RIS by applying explicit default config   revert this patch which can lead to kunit test failed;

类型	指派人员	状态
审查	josephhz xunlei JeffleXu 半叶枫尘 Artie Ding winterddd xueshuai 冯光辉 Guixin Liu kun-llfl Xuchun Shang XiangZao	已完成 (0/0人)
测试	小龙已测试通过	已完成 (1/1人)

GVPanolis/cloud-kernel

搜索帮助

GVP anolis/cloud-kernel