[PATCHv2] arm64:kexec: have own crash_smp_send_stop() for crash dump for nonpanic cores

[PATCHv2] arm64:kexec: have own crash_smp_send_stop() for crash dump for nonpanic cores

[Hoeun Ryu]

 Commit 0ee5941 : (x86/panic: replace smp_send_stop() with kdump friendly
version in panic path) introduced crash_smp_send_stop() which is a weak
function and can be overriden by architecture codes to fix the side effect
caused by commit f06e515 : (kernel/panic.c: add "crash_kexec_post_
notifiers" option).

 ARM64 architecture uses the weak version function and the problem is that
the weak function simply calls smp_send_stop() which makes other CPUs
offline and takes away the chance to save crash information for nonpanic
CPUs in machine_crash_shutdown() when crash_kexec_post_notifiers kernel
option is enabled.

 Calling smp_send_crash_stop() in machine_crash_shutdown() is useless
because all nonpanic CPUs are already offline by smp_send_stop() in this
case and smp_send_crash_stop() only works against online CPUs.

 The result is that /proc/vmcore is not available with the error messages;
"Warning: Zero PT_NOTE entries found", "Kdump: vmcore not initialized".

 crash_smp_send_stop() is implemented to fix this problem by replacing the
exising smp_send_crash_stop() and adding a check for multiple calling to
the function. The function (strong symbol version) saves crash information
for nonpanic CPUs and machine_crash_shutdown() tries to save crash
information for nonpanic CPUs only when crash_kexec_post_notifiers kernel
option is disabled.

* crash_kexec_post_notifiers : false

  panic()
    __crash_kexec()
      machine_crash_shutdown()
        crash_smp_send_stop()    <= save crash dump for nonpanic cores

* crash_kexec_post_notifiers : true

  panic()
    crash_smp_send_stop()        <= save crash dump for nonpanic cores
    __crash_kexec()
      machine_crash_shutdown()
        crash_smp_send_stop()    <= just return.

Signed-off-by: Hoeun Ryu <hoeun.ryu@gmail.com>
---
 v2:
   - replace the existing smp_send_crash_stop() with crash_smp_send_stop()
     and adding called-twice logic to it.
   - modify the commit message

 arch/arm64/include/asm/smp.h      |  2 +-
 arch/arm64/kernel/machine_kexec.c |  2 +-
 arch/arm64/kernel/smp.c           | 12 +++++++++++-
 3 files changed, 13 insertions(+), 3 deletions(-)

diff --git a/arch/arm64/include/asm/smp.h b/arch/arm64/include/asm/smp.h
index 55f08c5..f82b447 100644
--- a/arch/arm64/include/asm/smp.h
+++ b/arch/arm64/include/asm/smp.h
@@ -148,7 +148,7 @@ static inline void cpu_panic_kernel(void)
  */
 bool cpus_are_stuck_in_kernel(void);
 
-extern void smp_send_crash_stop(void);
+extern void crash_smp_send_stop(void);
 extern bool smp_crash_stop_failed(void);
 
 #endif /* ifndef __ASSEMBLY__ */
diff --git a/arch/arm64/kernel/machine_kexec.c b/arch/arm64/kernel/machine_kexec.c
index 481f54a..11121f6 100644
--- a/arch/arm64/kernel/machine_kexec.c
+++ b/arch/arm64/kernel/machine_kexec.c
@@ -252,7 +252,7 @@ void machine_crash_shutdown(struct pt_regs *regs)
 	local_irq_disable();
 
 	/* shutdown non-crashing cpus */
-	smp_send_crash_stop();
+	crash_smp_send_stop();
 
 	/* for crashing cpu */
 	crash_save_cpu(regs, smp_processor_id());
diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
index dc66e6e..73d8f5e 100644
--- a/arch/arm64/kernel/smp.c
+++ b/arch/arm64/kernel/smp.c
@@ -977,11 +977,21 @@ void smp_send_stop(void)
 }
 
 #ifdef CONFIG_KEXEC_CORE
-void smp_send_crash_stop(void)
+void crash_smp_send_stop(void)
 {
+	static int cpus_stopped;
 	cpumask_t mask;
 	unsigned long timeout;
 
+	/*
+	 * This function can be called twice in panic path, but obviously
+	 * we execute this only once.
+	 */
+	if (cpus_stopped)
+		return;
+
+	cpus_stopped = 1;
+
 	if (num_online_cpus() == 1)
 		return;
 
-- 
2.7.4

Re: [PATCHv2] arm64:kexec: have own crash_smp_send_stop() for crash dump for nonpanic cores

[James Morse]

Hi Hoeun,

On 07/08/17 06:09, Hoeun Ryu wrote:
>  Commit 0ee5941 : (x86/panic: replace smp_send_stop() with kdump friendly
> version in panic path) introduced crash_smp_send_stop() which is a weak
> function and can be overriden by architecture codes to fix the side effect

(overridden)


> caused by commit f06e515 : (kernel/panic.c: add "crash_kexec_post_
> notifiers" option).
> 
>  ARM64 architecture uses the weak version function and the problem is that
> the weak function simply calls smp_send_stop() which makes other CPUs
> offline and takes away the chance to save crash information for nonpanic
> CPUs in machine_crash_shutdown() when crash_kexec_post_notifiers kernel
> option is enabled.
> 
>  Calling smp_send_crash_stop() in machine_crash_shutdown() is useless
> because all nonpanic CPUs are already offline by smp_send_stop() in this
> case and smp_send_crash_stop() only works against online CPUs.


>  The result is that /proc/vmcore is not available with the error messages;
> "Warning: Zero PT_NOTE entries found", "Kdump: vmcore not initialized".

When I tried this I got one of these warnings for each secondary CPU, but the
vmcore file was still available. When I ran 'crash' on the vmcore it reported:
> CPUS: 6 [OFFLINE: 5]

Did I miss as step to reproduce this? If not, can we change this paragraph to
say something like:
> The result is that secondary CPUs registers are not saved by crash_save_cpu()
> and the vmcore file misreports these CPUs as being offline.


>  crash_smp_send_stop() is implemented to fix this problem by replacing the
> exising smp_send_crash_stop() and adding a check for multiple calling to

(existing)


> the function. The function (strong symbol version) saves crash information
> for nonpanic CPUs and machine_crash_shutdown() tries to save crash
> information for nonpanic CPUs only when crash_kexec_post_notifiers kernel
> option is disabled.
> 
> * crash_kexec_post_notifiers : false
> 
>   panic()
>     __crash_kexec()
>       machine_crash_shutdown()
>         crash_smp_send_stop()    <= save crash dump for nonpanic cores
> 
> * crash_kexec_post_notifiers : true
> 
>   panic()
>     crash_smp_send_stop()        <= save crash dump for nonpanic cores
>     __crash_kexec()
>       machine_crash_shutdown()
>         crash_smp_send_stop()    <= just return.


> diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
> index dc66e6e..73d8f5e 100644
> --- a/arch/arm64/kernel/smp.c
> +++ b/arch/arm64/kernel/smp.c
> @@ -977,11 +977,21 @@ void smp_send_stop(void)
>  }
>  
>  #ifdef CONFIG_KEXEC_CORE
> -void smp_send_crash_stop(void)
> +void crash_smp_send_stop(void)
>  {
> +	static int cpus_stopped;
>  	cpumask_t mask;
>  	unsigned long timeout;
>  
> +	/*
> +	 * This function can be called twice in panic path, but obviously
> +	 * we execute this only once.
> +	 */
> +	if (cpus_stopped)
> +		return;
> +
> +	cpus_stopped = 1;
> +

This cpus_stopped=1 can't happen on multiple CPUs at the same time as any second
call is guaranteed to be on the same CPU, both are behind panic()s
'atomic_cmpxchg()'.


Other than my '/proc/vmcore is not available' question above, this looks fine to me:
Reviewed-by: James Morse <james.morse@arm.com>
Tested-by: James Morse <james.morse@arm.com>


Thanks!

James

Re: [PATCHv2] arm64:kexec: have own crash_smp_send_stop() for crash dump for nonpanic cores

[Hoeun Ryu]

Hello, James.

Thank you for the meticulous test and review.

On Fri, 2017-08-11 at 18:02 +0100, James Morse wrote:
> Hi Hoeun,
> 
> On 07/08/17 06:09, Hoeun Ryu wrote:
> > 
> >  Commit 0ee5941 : (x86/panic: replace smp_send_stop() with kdump friendly
> > version in panic path) introduced crash_smp_send_stop() which is a weak
> > function and can be overriden by architecture codes to fix the side effect
> (overridden)

It'll be fixed in the next version.

> 
> 
> > 
> > caused by commit f06e515 : (kernel/panic.c: add "crash_kexec_post_
> > notifiers" option).
> > 
> >  ARM64 architecture uses the weak version function and the problem is that
> > the weak function simply calls smp_send_stop() which makes other CPUs
> > offline and takes away the chance to save crash information for nonpanic
> > CPUs in machine_crash_shutdown() when crash_kexec_post_notifiers kernel
> > option is enabled.
> > 
> >  Calling smp_send_crash_stop() in machine_crash_shutdown() is useless
> > because all nonpanic CPUs are already offline by smp_send_stop() in this
> > case and smp_send_crash_stop() only works against online CPUs.
> 
> > 
> >  The result is that /proc/vmcore is not available with the error messages;
> > "Warning: Zero PT_NOTE entries found", "Kdump: vmcore not initialized".
> When I tried this I got one of these warnings for each secondary CPU, but the
> vmcore file was still available. When I ran 'crash' on the vmcore it reported:
> > 
> > CPUS: 6 [OFFLINE: 5]
> Did I miss as step to reproduce this? If not, can we change this paragraph to
> say something like:
> > 
> > The result is that secondary CPUs registers are not saved by crash_save_cpu()
> > and the vmcore file misreports these CPUs as being offline.

Actually the commit log comes from the patch to fix a similar issue in arm port.
I'll change the commit log with yours.

> 
> > 
> >  crash_smp_send_stop() is implemented to fix this problem by replacing the
> > exising smp_send_crash_stop() and adding a check for multiple calling to
> (existing)

It'll be fixed in the next version.

> 
> 
> > 
> > the function. The function (strong symbol version) saves crash information
> > for nonpanic CPUs and machine_crash_shutdown() tries to save crash
> > information for nonpanic CPUs only when crash_kexec_post_notifiers kernel
> > option is disabled.
> > 
> > * crash_kexec_post_notifiers : false
> > 
> >   panic()
> >     __crash_kexec()
> >       machine_crash_shutdown()
> >         crash_smp_send_stop()    <= save crash dump for nonpanic cores
> > 
> > * crash_kexec_post_notifiers : true
> > 
> >   panic()
> >     crash_smp_send_stop()        <= save crash dump for nonpanic cores
> >     __crash_kexec()
> >       machine_crash_shutdown()
> >         crash_smp_send_stop()    <= just return.
> 
> > 
> > diff --git a/arch/arm64/kernel/smp.c b/arch/arm64/kernel/smp.c
> > index dc66e6e..73d8f5e 100644
> > --- a/arch/arm64/kernel/smp.c
> > +++ b/arch/arm64/kernel/smp.c
> > @@ -977,11 +977,21 @@ void smp_send_stop(void)
> >  }
> >  
> >  #ifdef CONFIG_KEXEC_CORE
> > -void smp_send_crash_stop(void)
> > +void crash_smp_send_stop(void)
> >  {
> > +	static int cpus_stopped;
> >  	cpumask_t mask;
> >  	unsigned long timeout;
> >  
> > +	/*
> > +	 * This function can be called twice in panic path, but obviously
> > +	 * we execute this only once.
> > +	 */
> > +	if (cpus_stopped)
> > +		return;
> > +
> > +	cpus_stopped = 1;
> > +
> This cpus_stopped=1 can't happen on multiple CPUs at the same time as any second
> call is guaranteed to be on the same CPU, both are behind panic()s
> 'atomic_cmpxchg()'.

'cpu_stopped' variable is not for the race of multi CPUs.
This variable is simply to prevent from calling
'smp_cross_call(&mask, IPI_CPU_CRASH_STOP)' twice in the machine_crash_shutdown().
Please look at following call path.
 
* crash_kexec_post_notifiers : true
 panic()
     crash_smp_send_stop() {
         ...
         cpu_stopped = 1              <= make it '1'
         smp_cross_call()             <= save crash dump for nonpanic cores
     }
     __crash_kexec()
         machine_crash_shutdown()
             crash_smp_send_stop() { 
                 if (cpu_stopped)
                     return           <= just return.
             }
> 
> 
> Other than my '/proc/vmcore is not available' question above, this looks fine to me:
> Reviewed-by: James Morse <james.morse@arm.com>
> Tested-by: James Morse <james.morse@arm.com>
> 
> 
> Thanks!
> 
> James
> 
> 
>