On Mon, Nov 12, 2018 at 03:43:37PM -0800, David Woodhouse wrote:
>
> That can't hurt. We should probably look at the time elapsed before you
> can *write* to it (when the background scan and crc checking is
> complete) rather than just reading.
>
Here are more data points. This is again with 100meg mtdram size. I made a
script which does the mount and umount, then perf ran that 100 times over and
averaged the results.
Baseline,
Performance counter stats for 'bash test.sh' (100 runs):
111.414863 task-clock # 0.637 CPUs utilized
( +- 0.07% )
41 context-switches # 0.371 K/sec
( +- 0.50% )
3 cpu-migrations # 0.023 K/sec
( +- 2.44% )
405 page-faults # 0.004 M/sec
( +- 0.05% )
147235193 cycles # 1.322 GHz
( +- 0.47% ) [53.76%]
53688988 stalled-cycles-frontend # 36.46% frontend cycles idle
( +- 2.59% ) [45.13%]
21691444 stalled-cycles-backend # 14.73% backend cycles idle
( +- 5.81% ) [68.50%]
138433181 instructions # 0.94 insns per cycle
# 0.39 stalled cycles per insn
( +- 0.88% ) [88.11%]
25882823 branches # 232.310 M/sec
( +- 1.42% ) [85.33%]
644457 branch-misses # 2.49% of all branches
( +- 5.19% ) [74.30%]
0.175012976 seconds time elapsed
( +- 0.58% )
With Nikunj's patch,
Performance counter stats for 'bash test.sh' (100 runs):
110.436715 task-clock # 0.625 CPUs utilized
( +- 0.07% )
41 context-switches # 0.373 K/sec
( +- 0.58% )
3 cpu-migrations # 0.024 K/sec
( +- 2.18% )
405 page-faults # 0.004 M/sec
( +- 0.05% )
145964351 cycles # 1.322 GHz
( +- 0.49% ) [53.68%]
47504491 stalled-cycles-frontend # 32.55% frontend cycles idle
( +- 2.96% ) [55.47%]
20481138 stalled-cycles-backend # 14.03% backend cycles idle
( +- 6.18% ) [71.19%]
134947645 instructions # 0.92 insns per cycle
# 0.35 stalled cycles per insn
( +- 1.18% ) [82.19%]
25343960 branches # 229.489 M/sec
( +- 1.65% ) [82.50%]
693642 branch-misses # 2.74% of all branches
( +- 5.29% ) [70.06%]
0.176606850 seconds time elapsed
( +- 0.50% )
This seems to show an 0.91% speed elapsed time difference. Most of the rest of
it seems very similar.
Daniel
