| 1 | Future ABI changes |
| 2 | ================== |
| 3 | |
| 4 | This file collects items that require a libperf ABI bump. Each entry |
| 5 | should describe the current limitation, the desired end state, and the |
| 6 | scope of the change so that a future ABI revision can batch them |
| 7 | together. |
| 8 | |
| 9 | 1. Widen struct perf_cpu.cpu from int16_t to int |
| 10 | - Current limit: 32767 CPUs. No architecture exceeds this today |
| 11 | (x86_64 max is 8192, arm64 is 4096), but NR_CPUS limits keep |
| 12 | growing. perf clamps to INT16_MAX in set_max_cpu_num() as a |
| 13 | safety net. |
| 14 | - Code simplification: the int16_t forces defensive truncation |
| 15 | checks at every boundary where a wider CPU index (int from |
| 16 | sample->cpu, al->cpu, etc.) is narrowed into struct perf_cpu. |
| 17 | Without these checks, values > 32767 silently wrap to negative |
| 18 | numbers (two's complement), bypassing bounds validation. |
| 19 | Widening to int eliminates this entire class of silent |
| 20 | truncation bugs and removes the need for the INT16_MAX clamp |
| 21 | in set_max_cpu_num(). |
| 22 | - Scope: struct perf_cpu is embedded everywhere — perf_cpu_map__cpu(), |
| 23 | perf_cpu_map__min(), perf_cpu_map__max(), perf_cpu_map__has(), the |
| 24 | for_each_cpu macros, and all internal callers. The perf_cpu_map |
| 25 | internal array (RC_CHK_ACCESS(map)->map[]) stores struct perf_cpu |
| 26 | directly. Widening changes the struct layout and every function |
| 27 | that returns or accepts struct perf_cpu by value. |
| 28 | - Migration: bump LIBPERF version in libperf.map, audit all |
| 29 | sizeof(struct perf_cpu) assumptions, update perf.data |
| 30 | serialization if needed. |