svn commit: r314216 - head/sys/x86/x86

[Konstantin Belousov]

On Sat, Feb 25, 2017 at 09:50:32AM +0200, Andriy Gapon wrote:
> On 24/02/2017 20:56, Jonathan T. Looney wrote:
> > Author: jtl
> > Date: Fri Feb 24 18:56:00 2017
> > New Revision: 314216
> > URL: https://svnweb.freebsd.org/changeset/base/314216
> > 
> > Log:
> >   We have seen several cases recently where we appear to get a double-fault:
> >   We have an original panic. Then, instead of writing the core to the dump
> >   device, the kernel has a second panic: "smp_targeted_tlb_shootdown:
> >   interrupts disabled". This change is an attempt to fix that second panic.
> 
> Just curious if you were able to find out what code caused those shootdowns to
> be sent.
> It's pretty unusual for the after-panic code to do things like that.
Memory allocations for busdma in coredumping path do that.

I know this from spending a lot of time making core dumps working with
DMAR busdma.
> 
> 
> >   When the other CPUs are stopped, we can't notify them of the TLB shootdown,
> >   so we skip that operation. However, when the CPUs come back up, we
> >   invalidate the TLB to ensure they correctly observe any changes to the
> >   page mappings.
> >   
> >   Reviewed by:	kib
> >   Sponsored by:	Netflix
> >   Differential Revision:	https://reviews.freebsd.org/D9786
> > 
> > Modified:
> >   head/sys/x86/x86/mp_x86.c
> > 
> > Modified: head/sys/x86/x86/mp_x86.c
> > ==============================================================================
> > --- head/sys/x86/x86/mp_x86.c	Fri Feb 24 17:36:55 2017	(r314215)
> > +++ head/sys/x86/x86/mp_x86.c	Fri Feb 24 18:56:00 2017	(r314216)
> > @@ -45,6 +45,7 @@ __FBSDID("$FreeBSD$");
> >  #ifdef GPROF 
> >  #include <sys/gmon.h>
> >  #endif
> > +#include <sys/kdb.h>
> >  #include <sys/kernel.h>
> >  #include <sys/ktr.h>
> >  #include <sys/lock.h>
> > @@ -1269,6 +1270,12 @@ cpustop_handler_post(u_int cpu)
> >  	CPU_CLR_ATOMIC(cpu, &started_cpus);
> >  	CPU_CLR_ATOMIC(cpu, &stopped_cpus);
> >  
> > +	/*
> > +	 * We don't broadcast TLB invalidations to other CPUs when they are
> > +	 * stopped. Hence, we clear the TLB before resuming.
> > +	 */
> > +	invltlb_glob();
> > +
> >  #if defined(__amd64__) && defined(DDB)
> >  	amd64_db_resume_dbreg();
> >  #endif
> > @@ -1427,6 +1434,10 @@ smp_targeted_tlb_shootdown(cpuset_t mask
> >  	uint32_t generation;
> >  	int cpu;
> >  
> > +	/* It is not necessary to signal other CPUs while in the debugger. */
> > +	if (kdb_active || panicstr != NULL)
> > +		return;
> > +
> >  	/*
> >  	 * Check for other cpus.  Return if none.
> >  	 */
> > 
> 
> 
> -- 
> Andriy Gapon