airbase: fix beacon interval timing #2177
Conversation
beansrus
force-pushed
the
master
branch
3 times, most recently
from
5546137
to
f546358
Compare
|
Hi, Would you mind rebasing on top of the lastest Thanks! |
|
Hello! rebased on master. Quickly gave it a test run and it still seems to work. |
|
I am concerned that big-endian machines will fail. An example is the timestamp; this value is internally CPU native but has been encoded in fixed form dependent on native CPU endianess. |
|
Do you mean the part like ( Also, it seems to be ok, as its updating the beacon 1 byte at a time so byte ordering shouldn't come into play. |
|
@Mister-X- Is this PR ready for merge? |
|
By the way, monotonic clock is still affected by jumps of time, like If I'm understanding correctly we'd have to use the TAI clock to be unaffected, however, that's Linux only. |
|
Could you also check and fix codestyle? We have a |
beansrus
force-pushed
the
master
branch
2 times, most recently
from
5552912
to
b3028d1
Compare
The current implementation had a few problems. * the beacons timing would not scale correctly for multiple ssids. If the time was 100 ms, and 5 ssids were beaconing, each ssid would beacon every 500ms instead of every 100ms. * The beacon timing would drift after some time. * The gettimeofday() is affected by discontinuous jumps in the system time. * the rtc code seemed to over complicate the timing code. Beacon frames are usually 100ms. A timer that takes into account small amounts of jitter should be sufficient for beacon timing. Tested with 50 ssids, on a ARMv7 @ 400mhz and didn't have any problem keeping up. I saw jitter somewhere between 400us-1.5ms
|
Hmm, I updated the comment about the NTP, perhaps I must have been thinking about monotonic_raw. Also, I don't see any changes with clang 3.8, although I do with clang-format 14 (was already installed in my system). Perhaps I missed something |
Hello. This pull request tries to fix up some of the beacon timing in airbase that I came across. Please let me know if there is anything I need to fixup to be accepted.
The current implementation had a few problems.
the time was 100 ms, and 5 ssids were beaconing, each ssid would beacon
every 500ms instead of every 100ms.
time (NTP adjustments and such).
frames are usually 100ms. A timer that takes into account small
amounts of jitter should be sufficient for beacon timing.
Tested with 50 ssids, on a ARMv7 @ 400mhz and didn't have any problem
keeping up. I saw jitter somewhere between 400us-1.5ms