On 09/04/2020 16:10, jan.som...@dlr.de wrote:
-----Ursprüngliche Nachricht-----
Von: Sommer, Jan
Gesendet: Donnerstag, 9. April 2020 15:57
An:devel@rtems.org
Cc: Sommer, Jan
Betreff: [PATCH v4 3/3] i386: Port to RTEMS
- Update imported files to compile rtems-libbsd for i386 based BSPs
---
[...]
diff --git a/freebsd/sys/kern/subr_gtaskqueue.c
b/freebsd/sys/kern/subr_gtaskqueue.c
index c061c6b0..4ef05e0a 100644
--- a/freebsd/sys/kern/subr_gtaskqueue.c
+++ b/freebsd/sys/kern/subr_gtaskqueue.c
@@ -744,7 +744,9 @@ taskqgroup_attach(struct taskqgroup *qgroup, struct
grouptask *gtask,
__func__, gtask->gt_name, error);
} else
#else /* __rtems__ */
+#ifndef __i386__
BSD_ASSERT(irq == -1);
+#endif /* __i386 */
#endif /* __rtems__ */
mtx_unlock(&qgroup->tqg_lock);
}
@@ -776,7 +778,9 @@ taskqgroup_attach_deferred(struct taskqgroup
*qgroup, struct grouptask *gtask)
}
#else /* __rtems__ */
+#ifndef __i386__
BSD_ASSERT(gtask->gt_irq == -1);
+#endif /* __i386 */
#endif /* __rtems__ */
qgroup->tqg_queue[qid].tgc_cnt++;
LIST_INSERT_HEAD(&qgroup->tqg_queue[qid].tgc_tasks, gtask,
gt_list);
For example in "iflib_legacy_setup" the function " taskqgroup_attach" is called
with an irq value other than -1, which triggers the assert.
Now, with the assert disabled the driver seems to work and we can, e.g. get an
IP with the dhcpcd0x tests.
We haven't managed to find out what the assert is for exactly, so I don't know
if that opens a can of worms.
Is this safe to do?
It looks like your upper layers request an unsupported feature. I
disabled several things related to CPU affinity in libbsd since this was
not properly supported by RTEMS before the EDF SMP scheduler was
introduced. It may be possible to enable these features now. If you want
to enable something in this area, please add test cases.
FreeBSD sometimes uses per-CPU data structures. They must be used with care.
_______________________________________________
devel mailing list
devel@rtems.org
http://lists.rtems.org/mailman/listinfo/devel
