[ECOS] cyg_scheduler_read_lock() seems wrong in the synthetic environment
Tue Dec 9 00:15:00 GMT 2003
>>>>> "Dan" == Dan Jakubiec <firstname.lastname@example.org> writes:
Dan> We are running eCos on an ARM7 platform and shadowing this
Dan> development in a Linux synthetic environment. I noticed that
Dan> in the synthetic environment the scheduler lock count seems
Dan> to be "off-by-one". When running in "thread context",
Dan> cyg_scheduler_read_lock() always reports 1 (instead of 0) and
Dan> DSR report >= 2. The lock count never drops below 1.
Dan> On our real ARM hardware, the lock counts look right: 0 for
Dan> thread context, >= 1 for DSRs. However, both environments
Dan> seem to be working fine.
Dan> Is this expected behavior or is something wrong? Can someone
Dan> explain what is/should be going on here?
It is not expected behaviour. As far as the kernel is concerned the
synthetic target HAL should behave the same as any other HAL. In fact
there is only one place in the HAL that even knows about the scheduler
lock: the interrupt handler needs to increment the lock early on.
A lot of eCos will actually continue running happily even if the
scheduler is permanently locked. However there will be no preemption
of threads and no timeslicing, because the interrupt handling code
will always think that the foreground thread was in a critical section
and must be allowed to complete.
As an experiment I just tried a clean build of the synthetic target in
the default configuration. Then I modified the kernel's timeslice.c
test, adding some diag_printf()'s and calls to
cyg_scheduler_read_lock(). The test passes and the reported scheduler
lock value is always 0, as it should be. This implies the problem is
not with the synthetic target HAL, instead something else is leaving
the scheduler locked.
Bart Veer eCos Configuration Architect
http://www.ecoscentric.com/ The eCos and RedBoot experts
Before posting, please read the FAQ: http://sources.redhat.com/fom/ecos
and search the list archive: http://sources.redhat.com/ml/ecos-discuss
More information about the Ecos-discuss