Fishwaldo opened a new pull request, #3706:
URL: https://github.com/apache/nuttx-apps/pull/3706
## Summary
* `testing/libc/arch_libc` covered `strcpy`. A machine directory can
override fourteen string and memory functions, and the mistakes such
overrides actually make are specific: the word loop that works aligned
and corrupts otherwise, the copy that runs past a guard byte, the overlap
`memmove` must survive in both directions, the terminator a length cap
was supposed to beat, the sign of a comparison whose bytes wrap.
* The new suite drives every function through every combination of source
and destination offset within a register, twenty one sizes chosen to
straddle every internal boundary, overlap at every distance up to a
register width each way, terminators probed for by name, caps placed
before, at and past differences, and guard bytes around every
destination. A speed report follows, one line per scenario, measured
through `CLOCK_MONOTONIC`.
* Two ways such a benchmark can quietly measure nothing are designed out,
both learned by measuring nothing: the compiler will hoist a pure call
whose arguments never change, so the pointers are laundered through an
`asm` that claims to change them without disturbing the alignment under
test; and an earlier scenario's writes can leave the compare buffers
unequal, so equality is verified before the compare timings and the one
scenario that needs shifted content prepares it explicitly after every
clean reader has run.
* The suite is a separate source file behind its own option, on by default,
and the program's exit status now says whether anything failed. The
existing `strcpy` test is untouched, including its `perf_gettime`
dependency, which configurations without hardware performance counters
will want switched off.
* It is arch-neutral: it exercises whatever the machine directory provides,
and passes identically against the generic C library when nothing is
overridden.
* **Why now:** this was written to validate a RISC-V machine-directory
optimisation series that adds ten functions and reworks two. That series
is a separate PR against `apache/nuttx`, which I will link here in a
comment as soon as it is open. This suite does not depend on it and
stands on its own; the RISC-V work depends on this suite for its
evidence,
which is why it comes first.
## Impact
* Is new feature added? Is existing feature changed? **NEW**, additive.
A new test source behind `CONFIG_TESTING_ARCH_LIBC_STRING`, default on.
The existing `strcpy` test is unchanged.
* Impact on user? **NO**, beyond a larger test binary when the option is
enabled. The program's exit status now reflects failure, which is a
change only for anyone scripting the old always-zero behaviour.
* Impact on build? **NO.** One new object when the option is on.
* Impact on hardware? **NO.** Portable C; no arch-specific code.
* Impact on documentation? **NO.**
* Impact on security? **NO.**
* Impact on compatibility? **NO.**
* Anything else? The suite is deliberately unkind. It found real faults in
the RISC-V work it was written for, and it detects deliberately injected
corruption, shown below.
## Testing
I confirm that changes are verified on local setup and works as intended:
* Build Host: macOS 26.5.1, arm64 (Apple Silicon), xPack riscv-none-elf-gcc
15.2.0
* Targets: **RISC-V rv64 on real hardware** (ESWIN EIC7700X EVB, 1.4 GHz,
downstream board port not yet upstream) and **RISC-V rv32 under QEMU**
Run against the generic C library, with no machine-directory overrides, on
rv64 hardware:
```
########## ROUND 1 final: BASELINE (EVB rv64 generic) ##########
== correctness ==
memcpy correctness: ok (0 bad)
memmove correctness: ok (0 bad)
mv-overlap correctness: ok (0 bad)
memset correctness: ok (0 bad)
memcmp correctness: ok (0 bad)
str-scan correctness: ok (0 bad)
strcmp-fam correctness: ok (0 bad)
== fails: 0 ==
```
Run against a fully overridden machine directory on the same rv64 hardware,
with the speed report and its repetition counts:
```
########## ROUND 2 (final harness): OPTIMIZED rv64 ##########
== correctness ==
memcpy correctness: ok (0 bad)
memmove correctness: ok (0 bad)
mv-overlap correctness: ok (0 bad)
memset correctness: ok (0 bad)
memcmp correctness: ok (0 bad)
str-scan correctness: ok (0 bad)
strcmp-fam correctness: ok (0 bad)
[compare buffers equal]
memcmp 64K aligned 342.6 MB/s (1376 reps, 0.251s)
memcmp 64K s+1/d+1 337.3 MB/s (1360 reps, 0.252s)
strncmp 4K aligned 244.8 MB/s (15664 reps, 0.250s)
strncmp 4K both+2 241.0 MB/s (15424 reps, 0.250s)
strlen 4K aligned 3274.3 MB/s (209552 reps, 0.250s)
strlen 4K +3 3240.5 MB/s (207392 reps, 0.250s)
memchr 4K aligned 2496.8 MB/s (159792 reps, 0.250s)
strchr 4K absent 2281.0 MB/s (145984 reps, 0.250s)
strchr 4K absent +5 2285.0 MB/s (146240 reps, 0.250s)
strrchr 4K 700.3 MB/s (44816 reps, 0.250s)
memcmp 64K s+1/d+2 41.7 MB/s (176 reps, 0.264s)
memcpy 64K aligned 3020.0 MB/s (12080 reps, 0.250s)
memcpy 64K src+1 2584.0 MB/s (10336 reps, 0.250s)
memmove-bk 64K 3464.0 MB/s (13856 reps, 0.250s)
strcpy 4K aligned 1975.3 MB/s (126416 reps, 0.250s)
strcpy 4K both+1 1822.5 MB/s (116640 reps, 0.250s)
== fails: 0 ==
```
The same suite on rv32 under QEMU, including `strlcpy`:
```
rv32 built
== string correctness ==
memcpy correctness: ok (0 bad)
memmove correctness: ok (0 bad)
mv-overlap correctness: ok (0 bad)
memset correctness: ok (0 bad)
memcmp correctness: ok (0 bad)
str-scan correctness: ok (0 bad)
strcmp-fam correctness: ok (0 bad)
strlcpy correctness: ok (0 bad)
== string fails: 0 ==
```
A passing suite proves little on its own, so the detection was verified by
injecting a single-bit fault into an optimized `memcpy` (`xori t1, t1, 1`
added to its inner loop) and confirming the suite catches it:
```
or t1, t1, t2
xori t1, t1, 1
== correctness ==
memcpy correctness: FAIL (576 bad)
memset correctness: ok (0 bad)
strcmp correctness: ok (0 bad)
== fails: 576 ==
```
576 detections from one flipped bit, with the neighbouring functions
correctly reporting clean. That last capture is from an earlier revision of
the harness, which covered `memcpy`, `memset` and `strcmp` before the suite
was widened to the full set; it is included as evidence that the checks
detect corruption rather than as a run of the final code.
## PR verification Self-Check
* [x] This PR introduces only one functional change.
* [x] I have updated all required description fields above.
* [x] My PR adheres to Contributing Guidelines and Documentation.
* [ ] My PR is still work in progress (not ready for review).
* [x] My PR is ready for review and can be safely merged into a codebase.
---
*Claude (claude-opus-5) assisted with authoring this test suite, its code
comments and this PR description. The commit carries an `Assisted-by:` tag
per
[CONTRIBUTING.md](https://github.com/apache/nuttx/blob/master/CONTRIBUTING.md)
ยง1.5.*
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
