Raspberry Pi 4 boot hangs in sched_idletd.

[Robert Crowston]

I figured it out: when I patched armstub8.bin to enable PSCI, I didn't bring in a #DEFINE from upstream required to enable the GIC on the cpu. With that done, timercb fires, the callout that boot_run_interrupt_driven_config_hooks() was waiting for happens, and I get as far as the mount root prompt.

Next weekend I'll work on the SD card driver.


‐‐‐‐‐‐‐ Original Message ‐‐‐‐‐‐‐
On Sunday, 21 July 2019 17:05, Konstantin Belousov <kostikbel at gmail.com> wrote:

> On Sun, Jul 21, 2019 at 03:58:24PM +0000, Robert Crowston wrote:
>
> > > Sleeping state is terminated by wakeup which happens either by timeouts
> > > (which indeed requires working event timers) or due to explicit wakeup when
> > > the waiting wait channel is signalled (by wakeup(9)).
> >
> > I have seen threads resume through wakeup(), so I think that part is working.
>
> Wakeup(9) itself most likely work, otherwise you would not get to the
> mount root stage of boot. What did not worked, probably, is some specific
> event which occurence causes wakeup for intr_config_hook. There, only
> you can inspect the state and see why it did not happen.
>
> > > But really you should start examining the sleep chain of the thread0 and
> > > see which exact condition it waits for.
> >
> > That's a helpful insight.
> >
> > > You did not even provided the backtrace for it.
> >
> > Breakpoint 8, mi_switch (flags=260, newtd=0x0) at /skeleton/root/sandbox/src/sys/kern/kern_synch.c:402
> > 402 td = curthread; /* XXX */
> > #0 mi_switch (flags=260, newtd=0x0) at /skeleton/root/sandbox/src/sys/kern/kern_synch.c:402
> > #1 0xffff0000004492ac in sleepq_switch (wchan=0xffff00000099fba0 <intr_config_hook_list>, pri=<optimized out>) at /skeleton/root/sandbox/src/sys/kern/subr_sleepqueue.c:625
> > #2 0xffff000000449920 in sleepq_timedwait (wchan=0xffff00000099fba0 <intr_config_hook_list>, pri=0) at /skeleton/root/sandbox/src/sys/kern/subr_sleepqueue.c:739
> > #3 0xffff0000004006bc in _sleep (ident=0xffff00000099fba0 <intr_config_hook_list>, lock=0xffff000000d461e0 <intr_config_hook_lock>, priority=0, wmesg=0xffff00000080569d "conifhk",
> > sbt=257698020000, pr=0, flags=256) at /skeleton/root/sandbox/src/sys/kern/kern_synch.c:213
> > #4 0xffff00000042576c in boot_run_interrupt_driven_config_hooks (dummy=<optimized out>) at /skeleton/root/sandbox/src/sys/kern/subr_autoconf.c:162
> > #5 0xffff0000003908a8 in mi_startup () at /skeleton/root/sandbox/src/sys/kern/init_main.c:314
> > #6 0xffff000000001088 in virtdone () at /skeleton/root/sandbox/src/sys/arm64/arm64/locore.S:142
> > Backtrace stopped: previous frame identical to this frame (corrupt stack?)
> > The final time I see we are on thread0, we are running boot_run_interrupt_driven_config_hooks, which looks like it has a sixty second delay. It's not the first or the last sleepq_timedwait in the boot, but it's probably the longest. I'll dig into the event timer side and try to figure out what interrupts its expecting.
>
> It is almost certainly not the eventtimers issue.