kernel: timer: fixes for log_timestamp test case failer due to timer PR regression #72352
Conversation
zephyrbot
requested review from
TaiJuWu,
andyross,
ceolin,
cfriedt,
dcpleung,
nashif,
npitre and
peter-mitsis
fixes for log_timestamp test case failer (issue zephyrproject-rtos#72344) due to timer PR regression. The regression is due to following reason: sys_clock_hw_cycles_per_sec() used in log_core_init() which invoked during cstart (EARLY init) but HPET driver init is PRE_KERNEL_2. So the z_clock_hw_cycles_per_sec need to init with some default CONFIG_SYS_CLOCK_HW_CYCLES_PER_SEC (though it is not right) or ideally need to init z_clock_hw_cycles_per_sec run time during cstart (EARLY init). Current Fixes: put back the older code for static initialization for z_clock_hw_cycles_per_sec with CONFIG_SYS_CLOCK_HW_CYCLES_PER_SEC. This will be later update with run time during HPTET driver init (PRE_KERNEL2). Signed-off-by: Najumon B.A <najumon.ba@intel.com>
add a static fallback option for timer clock frequency (HW clock cycle per sec) during early init. Later this will update once retrieve runtime clock frequency Signed-off-by: Najumon B.A <najumon.ba@intel.com>
|
|
||
| config SYS_CLOCK_HW_CYCLES_PER_SEC | ||
| int "System clock's h/w timer frequency" | ||
| depends on !TIMER_READS_ITS_FREQUENCY_AT_RUNTIME |
There was a problem hiding this comment.
This is sort of an abuse. The idea behind the dependency is to prevent apps from accidentally using the "static" clock rate on devices where it might be wrong because the actual rate is selected by the hardware.
andyross
left a comment
andyross
left a comment
There was a problem hiding this comment.
What does logging need the time for? Just a timestamp? We might fix it there by detecting a zero denominator and presenting a zero/"early" timestamp to the user. That's what Linux does for very early logs.
|
Also: how does HPET get around this? I don't think it initializes before z_cstart() either. |
|
This pull request has been marked as stale because it has been open (more than) 60 days with no activity. Remove the stale label or add a comment saying that you would like to have the label removed otherwise this pull request will automatically be closed in 14 days. Note, that you can always re-open a closed pull request at any time. |
Last activity: 18 months ago (review) Recent activities: - review: Review on 'x86: ia32: linker: move bss/noinit sections after data section' zephyrproject-rtos#73044 (2024-05-22T06:47:31Z) - pr: kernel: timer: fixes for log_timestamp test case failer due to timer PR regression zephyrproject-rtos#72352 (2024-05-06T10:47:53Z) - review: Review on 'Revert 69705 and fix zephyrproject-rtos#72344' zephyrproject-rtos#72345 (2024-05-06T09:53:31Z) Signed-off-by: Benjamin Cabé <benjamin@zephyrproject.org>
3 participants
fixes for log_timestamp test case failer (issue #72344) due to timer PR regression.
The regression is due to following reason:
sys_clock_hw_cycles_per_sec() used in log_core_init() which invoked during cstart (EARLY init) but HPET driver init is PRE_KERNEL_2. So the z_clock_hw_cycles_per_sec need to init with some default CONFIG_SYS_CLOCK_HW_CYCLES_PER_SEC (though it is not right) or ideally need to init z_clock_hw_cycles_per_sec run time during cstart (EARLY init).
Current Fixes: put back the older code for static initialization for z_clock_hw_cycles_per_sec with CONFIG_SYS_CLOCK_HW_CYCLES_PER_SEC. This will be later update with run time during HPTET driver init (PRE_KERNEL2) .