Re: [RFC PATCH v13 3/4] selftests/bpf: add iocost cost model test

[Tao Cui <cui.tao@linux.dev>]

Hello,

在 2026/11/6 20:44, sashiko-bot@kernel.org 写道:
> Thank you for your contribution! Sashiko AI review found 1 potential
> issue to consider:
> - [Medium] selftests/bpf: pread() in serial_test_iocost_model_streams()
>   hits page cache
Confirmed.  The initial buffered read() loop populates the page cache,
so the later pread() calls are served without issuing bios and don't
necessarily exercise the multi-stream path in iocost_ms_calc_cost().

I'll rework the test so the interleaved reads generate actual block
I/O instead of relying on cached pages.

Thanks,
Tao
> --
> 
> commit 371817062abb0b837764b9a93458f77eb2057dd2
> Author: Tao Cui <cuitao@kylinos.cn>
> 
> selftests/bpf: add iocost cost model test
> 
> This commit adds BPF selftests and two example struct_ops cost models for
> blk-iocost using a configfs-backed null_blk device.
> 
>> diff --git a/tools/testing/selftests/bpf/prog_tests/iocost_model.c b/tools/testing/selftests/bpf/prog_tests/iocost_model.c
>> new file mode 100644
>> index 0000000000000..a714507c040e3
>> --- /dev/null
>> +++ b/tools/testing/selftests/bpf/prog_tests/iocost_model.c
> [ ... ]
>> +void serial_test_iocost_model_streams(void)
>> +{
> [ ... ]
>> +		devfd = open(path, O_RDONLY);
>> +		if (ASSERT_OK(devfd < 0 ? -errno : 0, "open_dev_ms")) {
>> +			char iobuf[4096];
>> +			unsigned long long u1 = 0, u2 = 0;
>> +
>> +			ASSERT_OK(read_cost_usage("/iocost_ms",
>> +						  dev, &u1), "cost_usage_pre");
>> +			for (i = 0; i < 500; i++)
>> +				(void)read(devfd, iobuf, sizeof(iobuf));
>> +			/*
>> +			 * Two interleaved sequential streams: exercise the
>> +			 * stream table, not just a single cursor.
>> +			 */
>> +			for (i = 0; i < 200; i++) {
>> +				(void)pread(devfd, iobuf, sizeof(iobuf),
>> +									(i + 1) * sizeof(iobuf));
>> +				(void)pread(devfd, iobuf, sizeof(iobuf),
>> +									(1 << 20) +
>> +									(i + 1) * sizeof(iobuf));
>> +			}
> 
> [Severity: Medium]
> Do these interleaved pread() calls get satisfied from the block device page
> cache without reaching iocost_ms_calc_cost()?
> 
> Because devfd is opened without O_DIRECT, the preceding 500 sequential
> 4096-byte read() calls populate the page cache for bytes [0, 2048000).
> Both pread() streams read within that cached range ([4096, 823296) and
> [1052672, 1871872)), so no interleaved bios are submitted to the block
> layer and the multi-stream slot table is not exercised while the
> subsequent assertions pass from the initial read() loop alone.
>