RE: Re: [V3 PATCH 1/2] x86/panic: Replace smp_send_stop() with kdump friendly version

["河合英宏 / KAWAI，HIDEHIRO" <hidehiro.kawai.ez@hitachi.com>]

Hi,

> From: 'Dave Young' [mailto:dyoung@redhat.com]
> Sent: Monday, July 18, 2016 6:02 PM
> On 07/15/16 at 11:50am, 河合英宏 / KAWAI，HIDEHIRO wrote:
> > Hi Dave,
> >
> > Thanks for your reply.
> >
> > > From: 'Dave Young' [mailto:dyoung@redhat.com]
> > > Sent: Wednesday, July 13, 2016 11:04 AM
> > >
> > > On 07/12/16 at 02:49am, 河合英宏 / KAWAI，HIDEHIRO wrote:
> > > > Hi Dave,
> > > >
> > > > Thanks for the comments.
> > > >
> > > > > From: Dave Young [mailto:dyoung@redhat.com]
> > > > > Sent: Monday, July 11, 2016 5:35 PM
> > > > >
> > > > > On 07/05/16 at 08:33pm, Hidehiro Kawai wrote:
[snip] 
> > > > > As for this patch I'm not sure it is safe to replace the
> > > > > smp_send_stop with the kdump friendly function. I'm also not sure if
> > > > > the kdump friendly function is safe for kdump. Will glad to hear
> > > > > opinions from other arch experts.
> > > >
> > > > This stuff depends on architectures, so I speak only about
> > > > x86 (the logic doesn't change on other architectures at this time).
> > > >
> > > > kdump path with crash_kexec_post_notifiers disabled:
> > > >  panic()
> > > >    __crash_kexec()
> > > >      crash_setup_regs()
> > > >      crash_save_vmcoreinfo()
> > > >      machine_crash_shutdown()
> > > >        native_machine_crash_shutdown()
> > > >          panic_smp_send_stop() /* mostly same as original
> > > >                                 * kdump_nmi_shootdown_cpus()
> > > >                                 */
> > > >
> > > > kdump path with crash_kexec_post_notifiers enabled:
> > > >  panic()
> > > >    panic_smp_send_stop()
> > > >    __crash_kexec()
> > > >      crash_setup_regs()
> > > >      crash_save_vmcoreinfo()
> > > >      machine_crash_shutdown()
> > > >        native_machine_crash_shutdown()
> > > >          panic_smp_send_stop() // do nothing
> > > >
> > > > The difference is that stopping other CPUs before crash_setup_regs()
> > > > and crash_save_vmcoreinfo() or not.  Since crash_setup_regs() and
> > > > crash_save_vmcoreinfo() just save information to some memory area,
> > > > they wouldn't be affected by panic_smp_send_stop().  This means
> > > > placing panic_smp_send_stop before __crash_kexec is safe.
> > > >
> > > > BTW, I noticed my patch breaks Xen kernel.  I'll fix it in the next
> > > > version.
> > >
> > > But it does breaks stuff which depends on cpu not being disabled like problem 1 you mentioned in patch log.
> >
> > As I mentioned in the description of this patch, we should stop
> > other CPUs ASAP to preserve current state either
> > crash_kexec_post_notifiers is enabled or not.
> > Then, all remaining procedures should work well
> > after stopping other CPUs (but keep the CPU map online).
> >
> > Vivek also mentioned similar things:
> > https://lkml.org/lkml/2015/7/14/433
> 
> The implementation in this patchset is different from suggestion in above link?
> 
> I think Vivek's suggestion is a good idea, to drop smp_send_stop and do below:
> 
> stop_cpus_save_register_state;
> 
> if (!crash_kexec_post_notifiers)
> 	crash_kexec()
> atomic_notifier_call_chain()
> kmsg_dump()
> 
> I'm just commenting from code flow point of view, the detail implementation
> definitely need more comments from Arch experts.
> 
> Any reason did not move the kdump friendly function to earlier point like
> before previous __crash_kexec() below?
>         if (!crash_kexec_post_notifiers) {
>                 printk_nmi_flush_on_panic();
>                 __crash_kexec(NULL);
>         }

The reason why the implementation differs from Vivek's is to keep
the current code flow if crash_kexec_post_notifiers is not specified.

If we apply Vivek's or your suggestion, it may always cause kdump
to fail on MIPS OCTEON due to Problem 1.  I don't want to make things
any worse.  I may post a patch for MIPS OCTEON, but I can't test it.
For other architectures, I'm not sure what problems there are.
So at first, I want to fix the case where crash_kexec_post_notifiers is
specified on x86.  Then, if all other architectures support
`stop other CPUs before crash_kexec', switch to your or Vivek's
suggesting code.

Is this acceptable?

Best regards,

Hidehiro Kawai
Hitachi, Ltd. Research & Development Group