lkml.org 
[lkml]   [2020]   [Mar]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC bpf-next 0/2] sharing bpf runtime stats with /dev/bpf_stats
On Fri, Mar 13, 2020 at 05:35:16PM -0700, Song Liu wrote:
> Motivation (copied from 2/2):
>
> ======================= 8< =======================
> Currently, sysctl kernel.bpf_stats_enabled controls BPF runtime stats.
> Typical userspace tools use kernel.bpf_stats_enabled as follows:
>
> 1. Enable kernel.bpf_stats_enabled;
> 2. Check program run_time_ns;
> 3. Sleep for the monitoring period;
> 4. Check program run_time_ns again, calculate the difference;
> 5. Disable kernel.bpf_stats_enabled.
>
> The problem with this approach is that only one userspace tool can toggle
> this sysctl. If multiple tools toggle the sysctl at the same time, the
> measurement may be inaccurate.
>
> To fix this problem while keep backward compatibility, introduce
> /dev/bpf_stats. sysctl kernel.bpf_stats_enabled will only change the
> lowest bit of the static key. /dev/bpf_stats, on the other hand, adds 2
> to the static key for each open fd. The runtime stats is enabled when
> kernel.bpf_stats_enabled == 1 or there is open fd to /dev/bpf_stats.
>
> With /dev/bpf_stats, user space tool would have the following flow:
>
> 1. Open a fd to /dev/bpf_stats;
> 2. Check program run_time_ns;
> 3. Sleep for the monitoring period;
> 4. Check program run_time_ns again, calculate the difference;
> 5. Close the fd.
> ======================= 8< =======================
>
> 1/2 adds a few new API to jump_label.
> 2/2 adds the /dev/bpf_stats and adjust kernel.bpf_stats_enabled handler.
>
> Please share your comments.

Conceptually makes sense to me. Few comments:
1. I don't understand why +2 logic is necessary.
Just do +1 for every FD and change proc_do_static_key() from doing
explicit enable/disable to do +1/-1 as well on transition from 0->1 and 1->0.
The handler would need to check that 1->1 and 0->0 is a nop.

2. /dev is kinda awkward. May be introduce a new bpf command that returns fd?

3. Instead of 1 and 2 tweak sysctl to do ++/-- unconditionally?
Like repeated sysctl kernel.bpf_stats_enabled=1 will keep incrementing it
and would need equal amount of sysctl kernel.bpf_stats_enabled=0 to get
it back to zero where it will stay zero even if users keep spamming
sysctl kernel.bpf_stats_enabled=0.
This way current services that use sysctl will keep working as-is.
Multiple services that currently collide on sysctl will magically start
working without any changes to them. It is still backwards compatible.

\
 
 \ /
  Last update: 2020-03-14 03:44    [W:0.055 / U:1.532 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site