Thank you for the review. Replies are inline below. > Is the microsecond description accurate for all values the kernel can > report here? pie_dump_stats() [...] truncates the nanosecond value to > u32 before dividing [...] a qdelay above roughly 4.295 s wraps modulo > 2^32 ns [...] Would it make sense to convert sch_pie.c to div_u64() in > the same series so the two ABI producers agree with the comment being > added here?
Agreed, this wrap is real, and microseconds is the correct unit pie_dump_stats() is supposed to produce. But this patch is meant to be comment-only. We'd rather fix pie_dump_stats() with div_u64() in its own separate series right after this series, instead of mixing such a fix into this series on fq_pie. > Does the bytes/second wording hold on 32-bit kernels? pie_dump_stats() > does the scaling without widening [...] On ILP32 both operands stay > 32-bit, so the product wraps once avg_dq_rate exceeds [...] roughly > 16.8 MB/s. [...] Could the same (u64) cast be added to > pie_dump_stats()? Same answer as above: this is a real bug too, and we'll fix it in a separate follow-up series.
