Bug #2586
closed
fix timer duration calculations
Added by msuraev over 6 years ago.
Updated almost 5 years ago.
Description
The code in gprs_ns.c
uses nsvc_timer_elapsed_ms() which relies on osmo_gettimeofday() wrapper around gettimeofday() which is affected by discontinuous jumps in the system time (e.g., if the system administrator manually changes the system time).
It's better to use monotonic clock via clock_gettime(CLOCK_MONOTONIC, ..) but this would requre replacing "struct timeval
" with "struct timespec
" for timer_started
in "struct gprs_nsvc
". I'm not aware of any external users for this field but nevertheless we should properly change API/ABI version when doing so.
On a related note: we should carefully check all the uses of osmo_gettimeofday()
for elapsed time calculations and replace it with monotonic clock.
In commit 1fa8dfbaf42a4341255b3bbef3c8a3c574e8c477 corresponding explanation have been added to osmo_gettimeofday().
- Subject changed from fix NS timer calculations to fix timer duration calculations
- Status changed from New to In Progress
The problem is more widespread than initially anticipated: BSSGP and osmo_timer_* also use non-monotonic osmo_gettimeofday() for duration calculations.
- % Done changed from 0 to 10
- Related to Bug #2407: N3101 not implemented in OsmoPCU added
- % Done changed from 10 to 20
The groundwork is in gerrit 5103, fix for NS in 5104, fix for BSSGP in 5629, cleanup in 5631.
This bigger task of using monotonic clock for osmo_timer_*()
still remains to be implemented.
- Description updated (diff)
- Status changed from In Progress to Stalled
- Assignee deleted (
msuraev)
- Related to Feature #3309: osmo_timer_* should use monotonic clock added
- Status changed from Stalled to Resolved
- % Done changed from 20 to 100
the not-yet-implemented part of this issue is described in #3309
Also available in: Atom
PDF