On 07/30/2018 07:28 PM, AKASHI Takahiro wrote:

On Mon, Jul 30, 2018 at 04:36:28PM -0500, Dave Kleikamp wrote:

quoted

On 07/30/2018 11:57 AM, Will Deacon wrote:

quoted

On Mon, Jul 30, 2018 at 11:46:24AM -0500, Dave Kleikamp wrote:

quoted

On 07/30/2018 11:22 AM, Will Deacon wrote:

quoted

On Mon, Jul 30, 2018 at 05:16:42PM +0100, Catalin Marinas wrote:

quoted

On Mon, Jul 30, 2018 at 10:29:21AM -0500, Dave Kleikamp wrote:

quoted

machine_kexec flushes the reboot_code_buffer from the icache
after stopping the other cpus.

Commit 3b8c9f1cdfc5 ("arm64: IPI each CPU after invalidating the I-cache
for kernel mappings") added an IPI call to flush_icache_range, which
causes a hang here, so replace the call with __flush_icache_range

While machine_kexec() may be called with interrupts disabled (IIUC) and
we shouldn't IPI other CPUs, I don't understand why it hangs here. Are
there any other CPUs online at this point?

The BUG_ON and WARN_ON at the start of machine_kexec() suggest to me that
this should only happen if we're kexec'ing a crash kernel and
smp_crash_stop_failed(). Is that something we need to care about?

I observed the hang trying to kexec a crash kernel and I did not see the
warning that smp_crash_stop_failed(). I'm not exactly sure why
flush_icache_range() hung (but it did), but I think that
__flush_icache_range() makes more sense here anyway.

Yeah, I'll pick the patch up, but it would be nice to understand the
failure case you observed.

I see why it failed. ipi_cpu_crash_stop() does not call
set_cpu_online(cpu, false) the way ipi_cpu_stop() does. So
cpu_online_mask is still populated with the stopped cpus.

Any reason why it isn't called there?

Because I wanted that saved cpu-related state be as close to as it was
at panic.
If cpus go offline, the core dump would show that all the cores but
a panicked one be offline whether or not they actually were.

That makes sense.

Thanks,
Dave

Thanks,
-Takahiro AKASHI

quoted

Thanks,
Dave

quoted

Will